arX

ivs

ubm

it00

8995

3 [

gr-q

c] 1

1 A

ug 2

010

Discontinuous Galerkin method for the spherically reduced

BSSN system with second-order operators

Scott E Field1lowast Jan S Hesthaven2dagger Stephen R Lau3Dagger and Abdul H Mroue4sect

1Department of Physics Brown University Providence RI 02912 USA2Division of Applied Mathematics Brown University Providence RI 02912 USA

3Mathematics and Statistics University of New Mexico Albuquerque NM 87131 USA4Canadian Institute for Theoretical Astrophysics

University of Toronto Toronto Ontario M5S 3H8 Canada

We present a high-order accurate discontinuous Galerkin method for evolving

the spherically-reduced Baumgarte-Shapiro-Shibata-Nakamura (BSSN) system ex-

pressed in terms of second-order spatial operators Our multi-domain method

achieves global spectral accuracy and long-time stability on short computational

domains We discuss in detail both our scheme for the BSSN system and its im-

plementation After a theoretical and computational verification of the proposed

scheme we conclude with a brief discussion of issues likely to arise when one consid-

ers the full BSSN system

PACS numbers 0425Dm (Numerical Relativity) 0270Hm (Spectral Methods) 0270Jn

(Collocation methods) AMS numbers 65M70 (Spectral collocation and related methods)

83-08 (Relativity and gravitational theory Computational methods) 83C57 (General rela-

tivity Black holes)

I INTRODUCTION

Breakthroughs in numerical relativity during this decade have made it possible to sim-ulate via evolution of the full 3D Einstein equations binary black hole dynamics throughinspiral merger and ringdown of the remnant single black hole [1ndash15] (see e g recentreviews [16 17]) Inspiraling binaries are among the most promising sources of gravi-tational waves for the network of laser interferometric detectors such as LIGO [18] andVIRGO [19 20] Through the construction of templates for matched filtering waveformsextracted from numerical-relativity simulations are expected to facilitate the detection ofgenuine gravitational waveforms by interferometric detectors

Early attempts to evolve the Einstein equations relied on the Arnowitt-Deser-Misner(ADM) decomposition [21 22] The resulting ADM system proved only weakly hyperbolicwhen expressed in first-order form a fact partly accounting for difficulties associated withits numerical evolution [23 24] Difficulties in evolving black-hole solutions to the Einsteinequations also stem from singularities gauge conditions within the computational domainand unstable constraint violation For over ten years the goal of accurate and stable numeri-cal integration of the Einstein equations has continuously spurred the interest of numericistsand theorists alike leading to a wealth of new formalisms [25ndash52] (this list is not exhaustive)

lowast Scott Fieldbrownedu dagger Jan Hesthavenbrownedu Dagger srlaumathunmedusect mrouecitautorontoca

2

To evolve binary black holes numerical relativists currently use one of the followingversions of the Einstein equations the generalized harmonic (GH) system [48 49 51 53]or the Baumgarte-Shapiro-Shibata-Nakamura (BSSN) system [26 36 50] Using a finite-difference approach with adaptive mesh refinement Pretorius [1 2 49] used a constraint-suppressing second-order form of the GH system (suggested by Gundlach et al [48]) to evolvea binary through inspiral merger and ringdown Lindblom et al [51] recast the second-order GH system into a first-order symmetric-hyperbolic evolution system with constraintsuppression comparable to that of the second-order system This first-order GH systemhas been used to successfully simulate binary black holes evolution with nodal spectral(pseudospectral) methods [14 15 54] More recently Ref [55] has introduced a new penaltymethod for nodal spectral evolutions of spatially second-order wave equations This workprovides a foundation for solution of the second-order GH system via spectral methodsand has been used to evolve the Kerr solution [56] and the inspiral of binaries Typicallywritten in a spatially second-order form the BSSN system [36] has seen widespread useby numerical relativity groups that employ finite-difference techniques to evolve binariesRef [57] presented a nodal spectral code to evolve the BSSN system in second-order formThe system proved unstable when tested on a single black hole In more recent work [58]longer evolutions were obtained through the adoption of better gauge conditions filteringmethods and more distant outer boundaries The BSSN system has also been evolvedin a first-order strongly-hyperbolic formulation for a single black hole with nodal spectralmethods [59] Such evolutions of a single black hole exhibited instabilities similar to thosereported in Ref [58]

Corresponding to the two versions of the Einstein equations discussed in the last para-graph are two distinct techniques for the treatment of singularities in numerical relativityEvolutions based on the GH system have used black-hole excision whereby the interior ofan apparent horizon is removed (excised) from the computational domain This techniquerelies on horizon-tracking and gauge conditions which ensure that inner boundaries of thecomputational domain are pure out-flow whence no inner boundary conditions are neededEvolutions based on the BSSN system have relied on the moving-punctures technique [3 4]also coined ldquonatural excisionrdquo Technically much easier to implement than excision thistechnique features mild central singularities which evolve freely in the computational do-main (initially these puncture points may represent either asymptotically flat regions orldquotrumpetsrdquo [60ndash64])

Relative to the alternative systems previously discussed the BSSN system in second or-der form affords an easier treatment of singularities and features a relatively small numberof geometric variables directly related to the foliation of spacetime into spacelike hypersur-faces However to date spectral methods for black-hole binaries have been successfullyimplemented only for the first-order GH system The binary black hole problem is essen-tially a smooth one (singularities reside on sets of measure zero censored by horizons) andspectral methods exhibit well-established advantages over finite-difference methods for long-time simulation of such problems [65] Therefore the development and analysis of a stablespectral implementation of the full BSSN system is a worthwhile goal in numerical relativityand the motivation behind the pioneering investigations reported in Refs [57ndash59]

In Refs [50 60] Brown introduced a spherically reduced version of the BSSN system as atest bed for tractable examination of theoretical and computational issues involved in solvingthis system Indeed appealing to the simplicity of this system he offered geometrical andphysical insights into the nature of the moving-puncture technique and its finite-difference

3

implementation [60 63 64] (see also [61 62]) Here we exploit this system to a similar endusing it as a simplified setting in which to develop spectral methods for the stable integrationof the BSSN system Precisely we develop and test a nodal discontinuous Galerkin method(dG) [66] for integration of the spherically reduced BSSN system While Brownrsquos chief focuslay with moving punctures for further simplicity we adopt the excision technique Clearlythe problem we consider is not as daunting as the one confronted by both Tichy and Mroue[57ndash59] Nevertheless our method is robustly stable and therefore might serve as a steppingstone toward a stable dG-based formulation for the full BSSN system The conclusion offersfurther comments toward this end

Nodal dG schemes are both well-suited and well-developed for hyperbolic problems [66]Although mostly used for hyperbolic problems expressed as first-order systems dG methodshave also been applied to systems involving second-order spatial operators typically via dGinterior penalty (IP) methods [67ndash72] (Refs [73ndash75] discuss the concept of hyperbolicity[76] in the context of such systems) Penalty methods of a different type were exploited inRef [55] for the wave equation written in second order form Local discontinuous Galerkin

(LDG) schemes developed initially by Shu and coworkers [77ndash79] constitute an alternateapproach for integration of spatially second-order systems LDG schemes feature essentiallythe same auxiliary variables as those appearing in traditional first-order reductions howeverin LDG schemes such variables are not evolved and arise only as local variables The basicdifference between dGndashIP and LDGmethods is the manner in which subdomains are coupledThe method we described for the spherically reduced BSSN system is essentially an LDGscheme

This paper is organized as follows Section II collects the relevant equations from Brownrsquospresentation and develops some further notation useful for expressing the spherically re-duced BSSN system in various abstract forms Section III presents our nodal dG schemein detail and Section IV documents the results of several numerical simulations testing ourscheme Our conclusion discusses possible generalization of our method to the full BSSNsystem Several appendices collect further technical details In particular Appendix Cconsiders a simple system which models the spherically reduced BSSN system giving ananalytical proof that the model system is L2 stable in the semi-discrete sense

II SPHERICALLY SYMMETRIC (GENERALIZED) BSSN EQUATIONS

As shown by Brown [50] the BSSN system can be generalized to allow for a conformalmetric without unit determinant and this paper focuses on the spherical reduction of thissystem also considered by Brown in [60] In fact this spherical reduction relies on freedompresent in the generalized BSSN system since spherical-polar coordinates should not beassociated with a unit-determinant conformal metric Although we work with the sphericallyreduced generalized BSSN system (subject to Brownrsquos Lagrangian condition to be precise)we will nevertheless describe it as the spherically reduced BSSN system

A Basic variables and spherically reduced system

The conformal-traceless decomposition of the geometry associated with a spacelike 3-surface is

gab = χminus1gab Kab = χminus1(

Aab +1

3gabK

)

(1)

4

where gab is the physical 3-metric and Kab is the physical extrinsic curvature tensor TheBSSN variables are the conformal metric gab the conformal factor χ the trace-free extrinsiccurvature Aab the trace K = gabKab and the conformal connection Γa equiv minusgminus12partb(g

12gab)where g is the determinant of the metric The BSSN system also includes the lapse α shiftvector βa and an auxiliary vector field Ba used to define the ldquoΓ-driverrdquo for the shift

Following Brown we adopt a spherically symmetric line element

ds2 = minusα2dt2 + χminus1grr(dr + βrdt)2 + χminus1gθθ(dθ2 + sin2 θdφ2) (2)

along with the spherically symmetric Ansatz

Γa =

Γr

minus cos θ(gθθ sin θ)0

Aab = Arr

1 0 00 minusgθθ(2grr) 00 0 minusgθθ sin2 θ(2grr)

(3a)

Subject to the assumption of spherical symmetry the basic variables are χ grr gθθ Arr KΓr α βr Br All are functions of t and r and satisfy the following spherically symmetric(generalized Lagrangian-form) BSSN system1

parttα = βrαprime minus 2αK minus (parttα)0 (4a)

parttβr = βrβrprime +

3

4Br minus (parttβ

r)0 (4b)

parttBr = βrBr prime + λ(parttΓ

r minus βrΓr prime)minus ηBr minus (parttBr)0 (4c)

parttχ = βrχprime +2

3Kαχminus βrgprimerrχ

3grrminus 2βrgprimeθθχ

3gθθminus 2

3βr primeχ (4d)

parttgrr =2

3βrgprimerr +

4

3grrβ

r prime minus 2Arrαminus 2grrβrgprimeθθ

3gθθ(4e)

parttgθθ =1

3βrgprimeθθ +

Arrgθθα

grrminus gθθβ

rgprimerr3grr

minus 2

3gθθβ

r prime (4f)

parttArr = βrAprimerr +

4

3Arrβ

rprime minus βrgprimerrArr

3grrminus 2βrgprimeθθArr

3gθθ+

2αχ(gprimerr)2

3g2rrminus αχ(gprimeθθ)

2

3g2θθminus α(χprime)2

6χ

+2

3grrαχΓ

r prime minus αχgprimerrgprimeθθ

2grrgθθ+χgprimerrα

prime

3grr+χgprimeθθα

prime

3gθθminus αgprimerrχ

prime

6grrminus αgprimeθθχ

prime

6gθθminus 2

3αprimeχprime +

αχprimeprime

3

minus 2

3χαprimeprime minus αχgprimeprimerr

3grr+αχgprimeprimeθθ3gθθ

minus 2αA2rr

grr+KαArr minus

2grrαχ

3gθθ(4g)

parttK = βrK prime +χgprimerrα

prime

2g2rrminus χgprimeθθα

prime

grrgθθ+αprimeχprime

2grrminus χαprimeprime

grr+

3αA2rr

2g2rr+

1

3αK2 (4h)

parttΓr = βrΓr prime +

Arrαgprimeθθ

g2rrgθθ+

2βrprimegprimeθθ3grrgθθ

+Arrαg

primerr

g3rrminus 4αK prime

3grrminus 2Arrα

prime

g2rrminus 3Arrαχ

prime

g2rrχ

+4βr primeprime

3grrminus βr(gprimeθθ)

2

grr(gθθ)2+

βrgprimeprimerr6(grr)2

+βrgprimeprimeθθ3gθθgrr

(4i)

where the prime stands for partial r-differentiation Eqs (4d-i) are Brownrsquos Eqs (9a-f)listed in [60] subject to his Lagrangian condition (corresponding to v = 1 in Brownrsquos equa-tions) The first three equations (4a-c) comprise the gauge sector and these are essentially

1 For this system the determinant g = grr(gθθ)2 sin4 θ is not unity

5

spherically symmetric versions of the standard ldquo1+logrdquo and ldquoΓ-driverrdquo conditions listed inEqs (1) and (2) of [60] However we have introduced the following minor modificationsFirst (parttα)0 designates a constant term which ensures that the right-hand side of the αevolution equation (4a) vanishes at the initial time This source term as well as the analo-gous terms appearing in the evolution equations (4bc) for βr and Br are needed to enablea static evolution of the Schwarzschild solution in Kerr-Schild coordinates Second the pa-rameter λ (perhaps with functional dependence) modifies the hyperbolicity of the first-ordersystem [80] The damping parameter η typically appears in standard versions of these gaugeevolution equations (See Sections IIC and IVB for further discussions) For this BSSNsystem we have three constraints the Hamiltonian constraint H the momentum constraintMr and the constraint Gr resulting from the definition of the conformal connection Γr Inspherical symmetry these constraints are written as follows

H = minus3A2rr

2g2rr+

2K2

3minus 5(χprime)2

2χgrr+

2χprimeprime

grr+

2χ

gθθminus 2χgprimeprimeθθgrrgθθ

+2χprimegprimeθθgrrgθθ

+χgprimerrg

primeθθ

g2rrgθθminus χprimegprimerr

g2rr+χ(gprimeθθ)

2

2grrg2θθ

(5a)

Mr =Aprime

rr

grrminus 2K prime

3minus 3Arrχ

prime

2χgrr+

3Arrgprimeθθ

2grrgθθminus Arrg

primerr

g2rr(5b)

Gr = minus gprimerr2g2rr

+gprimeθθgrrgθθ

+ Γr (5c)

These expressions are the ones listed by Brown in [60] Eqs (4ef) also ensure that thedeterminant factor g sin4 θ = grr(gθθ)

2 remains fixed throughout an evolution

B Abstract expressions of the system

We define the following vectors built with system variables

u =

χgrrgθθαβr

v =

Br

Arr

KΓr

Q =

χprime

gprimerrgprimeθθαprime

βrprime

(6)

Introduction of Qmight seem unnecessary at this stage but proves useful in the constructionof our discontinuous Galerkin scheme In terms of the vectors u v and Q we further define

Wuv =

(

uv

)

WvQ =

(

vQ

)

W =WuQ =

uvQ

(7)

Here we have introduced ldquocolon notationrdquo [81] to represent (sub)vectors and (sub)matricesalthough we employ the notation over block rather than individual elements In the first-order version of the system (4) the components of Q are promoted to independent fields in

6

which case the corresponding principal part features

parttBr = βrBr prime minus 4λα

3grrK prime +

4λ

3grrQprime

βr +λβr

6(grr)2Qprime

grr +λβr

3gθθgrrQprime

gθθ(8a)

parttArr = βrAprimerr +

2

3grrαχΓ

r prime +1

3αQprime

χ minus2

3χQprime

α minus αχ

3grrQprime

grr +αχ

3gθθQprime

gθθ(8b)

parttK = βrK prime minus χ

grrQprime

α (8c)

parttΓr = βrΓr prime minus 4αK prime

3grr+

4

3grrQprime

βr +βr

6(grr)2Qprime

grr +βr

3gθθgrrQprime

gθθ(8d)

parttQχ = βrQprimeχ +

2

3αχK prime minus βrχ

3grrQprime

grr minus2βrχ

3gθθQprime

gθθminus 2

3χQprime

βr (8e)

parttQgrr =2

3βrQprime

grr +4

3grrQ

primeβr minus 2αAprime

rr minus2grrβ

r

3gθθQprime

gθθ(8f)

parttQgθθ =1

3βrQprime

gθθ+gθθα

grrAprime

rr minusgθθβ

r

3grrQprime

grr minus2

3gθθQ

primeβr (8g)

parttQα = βrQprimeα minus 2αK prime (8h)

parttQβr = βrQprimeβr +

3

4Br prime (8i)

where all lower-order terms on the right-hand side have been dropped This sector of prin-cipal parts of the first-order system has the form

parttWvQ + A(u)W primevQ = 0 (9)

where (minus) the explicit form of the 9-by-9 matrix A(u) is given below in (A1) Thefirst-order version of (4) takes the nonconservative form

parttW +A(u)W prime = S(W ) A(u) =

(

05times5 05times9

09times5 A(u)

)

(10)

where S(W ) is a vector of lower order terms built with all components of W Partition ofA(u) = A(u)vQvQ into blocks corresponding to the v and Q sectors yields

A(u) =

(

A(u)vv A(u)vQ

A(u)Qv A(u)QQ

)

(11)

Using these blocks we then define the 9-by-9 matrix

A(u) = A(u)uvvQ =

(

05times4 05times5

A(u)vv A(u)vQ

)

(12)

and express (4) as

parttWuv + A(u)W primevQ = S(W ) (13a)

Q = uprime (13b)

where S(W ) = S(W )uv

7

field speed

X1 micro1 = 0

X23 micro23 = minusβr

Xplusmn4 microplusmn

4 = minusβr plusmnradic

2αχgrr

Xplusmn5 microplusmn

5 = minusβr plusmn αradic

χgrr

Xplusmn6 microplusmn

6 = minusβr plusmnradic

λgrr

TABLE I Characteristic speeds These speeds are the eigenvalues listed in (A2)

C Hyperbolicity and characteristic fields

Although our numerical scheme deals directly with the second-order spatial operatorsappearing in (4) we first consider the hyperbolicity of the corresponding first-order system(10) The characteristic fields and their speeds are found by instantaneously ldquofreezingrdquothe fields u in A(u) to some value u0 corresponding to a linearization around a uniformstate Below we continue to write u for simplicity with the understanding that u is reallythe background solution u0 Of primary interest is the range of u0 for which the system isstrongly hyperbolic [73ndash76]

Appendix A shows that the characteristic fields corresponding to (4) are as follows (i)all components of u (each with speed 0) and (ii) the fields

X1 = gθθQgrr + 2grrQgθθ (14a)

X2 = grrΓr +

2

χQχ minus

1

2grrQgrr minus

1

gθθQgθθ (14b)

X3 =grrλBr +

2

χQχ minus

1

2grrQgrr minus

1

gθθQgθθ (14c)

Xplusmn4 = plusmn

radic

2αgrrχ

K +Qα (14d)

Xplusmn5 = ∓ 3radic

grrχArr plusmn 2

radic

grrχK + 2grrΓ

r +1

χQχ minus

1

grrQgrr +

1

gθθQgθθ (14e)

Xplusmn6 = minus3

4

grrλBr plusmn α

radicλgrr

(2αχminus λ)K minus βr

8(βrgrr ∓radicλgrr)

Qgrr

minus βrgrr

4gθθ(βrgrr ∓radicλgrr)

Qgθθ +αχ

(2αχminus λ)Qα plusmn

radic

grrλQβr (14f)

with the speeds listed in Table I To ensure strong hyperbolicity we must necessarily require

λ gt 0 (βr)2grr minus λ 6= 0 2αχminus λ 6= 0 (15)

as shown in in Appendix A where further conditions are also given When λ = 1 thehyperbolicity condition of Ref [60] is recovered In fact the system could be recast assymmetric hyperbolic Indeed as it involves one spatial dimension the relevant symmetrizercan be constructed via polar decomposition of the diagonalizing similiarity transformationHowever we will not exploit this possibility

8

This system admits an inner excision boundary provided

βr ge max

(

radic

2αχ

grr

radic

α2χ

grr

radic

λ

grr

)

(16)

holds at the inner boundary This condition ensures each characteristic field has a nonposi-tive speed at the inner boundary and therefore the inner boundary is an excision boundaryat which no boundary conditions are needed The extra flexibility afforded by the param-eter λ could be used to maintain rigorous hyperbolicity by moving the points at whichthe conditions in (15) are violated outside of the computational domain Furthermore forλ = 1 Eq (16) conceivably fails or is only satisfied close to r = 0 where field gradients are

prohibitively large The troublesome X+6 gauge mode has a positive speed minusβr +

radic

λgrrIndeed for the conformally flat Kerr-Schild system considered in section IVB an inner ex-cision boundary is only possible provided λ is small enough

The transformation (14) can be inverted in order to express the fundamental fields interms of the characteristic fields

Br = minus1

6

λ

grrgθθ

[

(βr)2

(βr)2grr minus λ

]

X1 +2

3

λαχ

grr(2αχminus λ)(X+

4 +Xminus4 )minus

2

3

λ

grr(X+

6 +Xminus6 )

(17a)

Arr =1

3

radic

grrχ

2α(X+

4 minusXminus4 )minus

radicgrrχ

6(X+

5 minusXminus5 ) (17b)

K =

radic

χ

8αgrr(X+

4 minusXminus4 ) (17c)

Γr = minus1

6

1

grrgθθ

[

(βr)2

(βr)2grr minus λ

]

X1 +1

grr(X2 minusX3) +

2

3

αχ

grr(2αχminus λ)(X+

4 +Xminus4 )

minus 2

3

1

grr(X+

6 +Xminus6 ) (17d)

Qχ =1

12

χ

grrgθθ

[

4(βr)2grr minus 3λ

(βr)2grr minus λ

]

X1 +χ

2X3 minus

1

3

αχ2

(2αχminus λ)(X+

4 +Xminus4 ) (17e)

+χ

3(X+

6 +Xminus6 ) (17f)

Qgrr =2(βr)2grr minus 3λ

6gθθ((βr)2grr minus λ)X1 +

4

3grrX2 minus grrX3 +

2

3

αχgrr(2αχminus λ)

(X+4 +Xminus

4 )

minus 1

3grr(X

+5 +Xminus

5 )minus2

3grr(X

+6 +Xminus

6 ) (17g)

Qgθθ =

[

1

4grr+

(βr)2

12((βr)2grr minus λ)

]

X1 minus2

3gθθX2 +

1

2gθθX3 minus

1

3

αχgθθ(2αχminus λ)

(X+4 +Xminus

4 )

+1

6gθθ(X

+5 +Xminus

5 ) +1

3gθθ(X

+6 +Xminus

6 ) (17h)

Qα =1

2(X+

4 +Xminus4 ) (17i)

Qβr =βrλ

8grrgθθ((βr)2grr minus λ)X1 minus

λ

(2αχminus λ)

radic

αχ

8grr(X+

4 minusXminus4 ) +

1

2

radic

λ

grr(X+

6 minusXminus6 )

(17j)

9

We will refer to this inverse transformation when discussing outer boundary conditions forour numerical simulations in Sec IVB

III DISCONTINUOUS GALERKIN METHOD

This section describes the nodal discontinuous Galerkin method used to numerically solve(4) We adopt a method-of-lines strategy and here describe the relevant semi-discrete schemewhile leaving the temporal dimension continuous To approximate (4) we follow the generalprocedure first introduced in Ref [82] Our approach defines local auxiliary variables Q = uprimeand rewrites the spatially second-order system (4) as the first-order system (13a) Once weuse (13b) to eliminate Q from (13a) we recover the primal equations (4) The auxiliaryvariable approach was later generalized and coined the local discontinuous Galkerin (LDG)method in Ref [77] We may refer to our particular scheme as an LDG method but notethat many variations exist in the literature We stress that in LDG methods Q is not evolvedand is introduced primarily to assist in the construction of a stable scheme

Equations (12) and (13a) imply that the physical flux function is

F (W ) =

(

Fu(W )

Fv(W )

)

equiv A(u)WvQ =

(

05times1

f(W )

)

f =

fBr

fArr

fKfΓ

(18)

Only the evolution equations for Br Arr K and Γr give rise to non-zero components in F and we have collected these non-zero components into a smaller vector f = Fv Inspectionof (8) determines these components For example from (8c) we find

fK = minusβrK +χ

grrQα (19)

A Local approximation of the system (13)

Our treatment closely follows [83] but with the equations and notations relevant forthis paper Our computational domain Ω is the closed r-interval [a b] We cover Ω withkmax gt 1 non-overlapping intervals Dk = [ak bk] where a = a1 b = bkmax and bkminus1 = ak fork = 2 middot middot middot kmax

On each interval Dk we approximate each component of the system vector W by a localinterpolating polynomial of degree N For example

χkh(t r) =

Nsum

j=0

χ(t rkj )ℓkj (r) (20)

approximates χ(t r) Throughout this section approximations are denoted by a subscripth (see [66] for the notation) For example Wh and fh are approximations of W and f Although Q = uprime Qh and uprimeh are not necessarily the same In (20) ℓkj (r) is the jth Lagrange

polynomial belonging to Dk

ℓkj (r) =Nprod

i=0i 6=j

r minus rkirkj minus rki

(21)

10

Evidently the polynomial χkh interpolates χ at rkj To define the nodes rkj consider the

mapping from the unit interval [minus1 1] to Dk

rk(u) = ak + 12(1 + u)(bk minus ak) (22)

and the N+1 Legendre-Gauss-Lobatto (LGL) nodes uj The uj are the roots of the equation

(1minus u2)P primeN(u) = 0 (23)

where PN(u) is the Nth degree Legendre polynomial and the physical nodes are simplyrkj = rk(uj) In vector notation the approximation (20) takes the form

χkh(t r) = χk(t)Tℓk(r) (24)

in terms of the column vectors

χk(t) =[

χ(t rk0) middot middot middot χ(t rkN)]T ℓk(r) =

[

ℓk0(r) middot middot middot ℓkN(r)]T (25)

On each open interval (ak bk) sub Dk and for each component of the equations in (13) we

define local residuals measuring the extent to which our approximations satisfy the originalcontinuum system Dropping the subdomain label k on the polynomials and focusing onthe K equation as a representative example the local residual corresponding to (4h) is

minus(RK)kh equivminus parttKh + (βrK prime)h minus

(

χQprimeα

grr

)

h

+

(

χQgrrQα

2g2rr

)

h

minus(

χQgθθQα

grrgθθ

)

h

+

(

QαQχ

2grr

)

h

+

(

3αA2rr

2g2rr

)

h

+

(

1

3αK2

)

h

(26)

Here for example the expressions 2

(βrK prime)h = βrhK

primeh

(

QαQχ

2grr

)

h

=QαhQχh

2grrh (27)

We similarly construct the remaining eight residuals eg (Rgrr)h and (RΓr)h as well as fiveresiduals corresponding to (13b) For example one of these remaining five is

(RQα)kh equiv minusQαh + αprime

h (28)

Let the kth inner product be defined as

(

u v)

Dk equivint bk

akdru(r)v(r) (29)

and consider the expression (ℓkj (RK)kh)Dk We call the requirement that this inner product

vanish forallj the kth Galerkin condition For each component of the system and for each k

2 At this stage the first expression is generically a polynomial of degree 2N minus 1 and the latter is not a

polynomial The conventions adopted in Eq (27) prove useful while working with the residual However

later on in Sec III C to obtain the final form (47) of the numerical approximation corresponding to (26)

we will replace nonlinear terms with degree-N polynomials

11

there is a corresponding Galerkin condition in total 9kmax(N + 1) equations for (13a) and5kmax(N+1) for (13b) Enforcement of the Galerkin conditions on each D

k will not recover ameaningful global solution since they provide no mechanism for coupling the local solutionson the different intervals Borrowing from the finite volume toolbox we achieve couplingthrough integration by parts on r and introduction of the numerical flux f lowast at the interfacebetween subdomains

In (26) we only need to consider (βrK prime)h and (χQprimeαgrr)h as the other terms comprise a

component of the source vector Sh Using integration by parts we write

(

ℓkj (βrK prime)h

)

Dk =minusint bk

akdr[

(

ℓkjβrh

)primeKh

]

+ (βrhKh) ℓ

kj

∣

∣

∣

bk

ak (30a)

(

ℓkj (χQprimeαgrr)h

)

Dk =minusint bk

akdr

[(

ℓkjχh

grrh

)prime

Qαh

]

+

(

χhQαh

grrh

)

ℓkj

∣

∣

∣

bk

ak (30b)

In these formulas we have retained the domain index k on ℓkj while continuing to suppress iton Kh grrh etc Moreover we have suppressed the r-dependence in all terms on the right-hand side Addition of these formulas along with the definition fKh = minus(βrK)h+(χQαgrr)hgives

(

ℓkj (βrK prime)h minus (χQprime

αgrr)h)

Dk = minusint bk

akdr

[

(

ℓkjβrh

)primeKh minus

(

ℓkjχh

grrh

)primeQαh

]

minus fKhℓkj

∣

∣

∣

bk

ak (31)

In lieu of (31) we will instead work with the replacement

(

ℓkj (βrK prime)h minus (χQprime

αgrr)h)

Dk rarr minusint bk

akdr

[

(

ℓkjβrh

)primeKh minus

(

ℓkjχh

grrh

)primeQαh

]

minus f lowastKℓ

kj

∣

∣

∣

bk

ak (32)

This replacement features a component f lowastK of the numerical flux rather than a component

fKh of the boundary flux The numerical flux is determined by (as yet not chosen) functions3

f lowast = f lowast(W+Wminus) (33)

where for example Wminus is an interior boundary value [either W kh (t a

k) or W kh (t b

k)] of theapproximation defined on D

k andW+ is an exterior boundary value [either W kminus1h (t bkminus1) or

W k+1h (t ak+1)] of the approximation defined on either Dkminus1 or Dk+1 We discuss our choice

of numerical flux in the next subsection We now employ additional integration by parts towrite the above replacement as

(

ℓkj (βrK prime)h minus (χQprime

αgrr)h)

Dk rarrint bk

akdrℓkj

(

βrK prime minus χQprimeα

grr

)

h

+ (fKh minus f lowastK) ℓ

kj

∣

∣

∣

bk

ak (34)

Rather than the exact kth Galerkin condition(

ℓkj (RK)kh

)

Dk = 0 forallj for the K component

of (13) on Dk we will instead strive to enforce

(

ℓkj (RK)kh

)

Dk = (fKh minus f lowastK) ℓ

kj

∣

∣

∣

bk

ak forallj (35)

3 In the context of the dG method here + and minus denote ldquoexteriorrdquo and ldquointeriorrdquo and have no relation to

the plusmn using to denote the characteristic fields and speeds in Table I For characteristic fields and speeds

+ and minus mean ldquoright-movingrdquo and ldquoleft-movingrdquo

12

although our treatment of nonlinear terms will lead to a slight modification of these equations(we return to this issue shortly) The other components of (13a) are treated similarly asare the components of (13b) Recall that for example Qα = αprime Formally using the samedG method to solve for Qα we arrive at the replacement

(

ℓkj (RQα)kh

)

Dk rarrint bk

akdrℓkj (minusQαh + αprime

h)minus (αh minus αlowast) ℓkj

∣

∣

∣

bk

ak (36)

which again features a component αlowast of the numerical flux The auxiliary variables areconstructed and used at each stage of temporal integration We then have

(

ℓkj (RQα)kh

)

Dk = (αh minus αlowast) ℓkj

∣

∣

∣

bk

ak forallj (37)

as the corresponding enforced kth Galerkin condition

B Numerical Flux

To further complete our dG scheme we must specify functional forms for the compo-nents of the numerical flux introduced in the previous section We distinguish between thephysical fluxes (components of f) and the auxiliary fluxes (components of u) arising fromthe definition of the auxiliary variables These choices are not independent as the resultingscheme must be stable and consistent Our choice follows [84] which considered diffusionproblems Additional analysis of this flux choice appears in [66 85]

Let us first consider the numerical fluxes corresponding to the physical fluxes and of theform (33) The numerical flux vector is a function of the system and auxiliary variablesinterior and exterior to a subdomain A common choice for f lowast is

f lowast = fh+τ

2

[[

vh]]

K-component of f lowast f lowastK = fKh+

τ

2

[[

Kh

]]

(38)

where as an example we have also shown the component of f lowast corresponding to the analysisabove Respectively the average and jump across the interface are

fh =1

2

(

f+ + fminus) [[

vh]]

= nminusvminus + n+v+ (39)

Here τ is a position dependent penalty parameter (fixed below) and nminus(n+) is the localoutward pointing normal to the interior (exterior) subdomain The role of τ is to ldquopenalizerdquo(i e yield a negative contribution to the L2 energy norm) jumps across an interface Anappropriate choice of τ will ensure stability and we now provide some motivation for thechoice (41) of τ we make below

Were we treating the fully first-order system (10) the local Lax-Friedrichs flux wouldoften be a preferred choice due to its simplicity [66] In this case the constant ω in thenumerical flux formula Flowast = Fh + 1

2ω[[

Wh

]]

obeys ω ge max∣

∣micro(nablaWF(W ))∣

∣ HereF(W ) = A(u)W the notation micro(middot) indicates the spectral radius of the matrix within andthe max is taken over interior Wminus and exterior W+ states Motivated by (9) we adopt asimilar but simpler prescription substituting the field gradient

nablaWvQA(u)WvQ = A(u) (40)

13

for nablaWF(W ) Precisely we assume the scaling

τ(bk) = τ(ak+1) = τk+12 equiv C middotmax∣

∣micro(

A(u))∣

∣ (41)

where C = O(1) is a constant chosen for stability Larger values of C will result in schemeswith better stability properties whereas too large a value will impact the CFL conditionAt the interface point Ik+12 equiv D

k cap Dk+1 the vector uh has two representations uminus at bk

and u+ at ak+1 The max in (41) is taken over the corresponding two sets of field speedsMore precisely the speeds in Table I are computed for both uminus and u+ and the maximumtaken over all resulting speeds For the auxiliary variables a penalized central flux is usedThe definition with one representative component is

ulowast = uh minus1

2

[[

uh]]

α-component of ulowast αlowast = αh minus1

2

[[

αh

]]

(42)

with similar expressions for the remaining componentsWe stress the following point Since the interior coupling between subdomains is achieved

through the numerical flux forms (41) and (42) the inverse transformation (17) expressingthe fundamental fields in terms of the characteristic fields is not required to achieve thiscoupling On the other hand imposition of physical boundary conditions may still rely on(17) since this transformation allows one to fix only incoming characteristic modes

C Matrix form of the semi-discrete equations

Let us introduce the kth mass and stiffness matrices

Mkij =

int bk

akdrℓki (r)ℓ

kj (r) Sk

ij =

int bk

akdrℓki (r)ℓ

kjprime(r) (43)

These matrices belong to Dk and the corresponding matrices defined on the reference interval[minus1 1] are

Mij =

int 1

minus1

duℓi(u)ℓj(u) Sij =

int 1

minus1

duℓi(u)ℓprimej(u) (44)

where ℓj(u) is the jth Lagrange polynomial determined by the LGL nodes uj on [minus1 1]These matrices are related by Mk

ij =12(bk minus ak)Mij and S

kij = Sij whence only the reference

matrices require computation and storageWe will use the matrices Mk and Sk in obtaining an ODE system from (26) and (35)

Towards this end we first approximate the nonlinear terms (products and quotients) in(26) by degree-N interpolating polynomials Such approximations are achieved throughpointwise representations For example (QαQχgrr)h appears in (26) and is expressed inthe following way [cf footnote 2]

(

QαQχ

grr

)

h

(t r) =

(

QαhQχh

grrh

)

(t r) rarrNsum

j=0

Qαh(t rkj )Qχh(t r

kj )

grrh(t rkj )

ℓkj (r) (45)

Note that the expressions on the right and left are not equivalent due to aliasing error [65]Our vector notation for this replacement will be

(

QαQχ

grr

)

h

(t r) rarr(

QαQχ

grr

)

(t)Tℓk(r) (46)

14

Operations among bold variables are always performed pointwise Making similar replace-ments for all terms in (26) and then carrying out the integrations in (35) which bring inMk and Sk we arrive at

parttK = βrDK minus χDQα

grr

+1

2

χQgrrQα

g2rr

minusχQgθθ

Qα

grrgθθ

+1

2

QαQχ

grr

+3

2

αA2rr

g2rr

+1

3αK2 +Mminus1ℓk (fKh minus f lowast

K)∣

∣

∣

bk

ak (47)

where we have again suppressed the superscript k on all terms except ℓk(r) and the sub-script h is dropped on all boldfaced variables As described in [66] the spectral collocationderivative matrix

(Dk)ij =dℓkjdr

∣

∣

∣

∣

∣

r=rki

(48)

can also be expressed as Dk = (Mk)minus1Sk which appears above Eight other semi-discreteevolution equations are similarly obtained with nine in total (one for each component ofWuv) Additionally we have

Qα = Dα+Mminus1ℓk(αlowast minusαh)∣

∣

∣

bk

ak (49)

as one of the auxiliary equations with five in total (one for each component of Q = WQQ)

D Filtering

Like other nodal (pseudospectral) methods our scheme may suffer from instabilitiesdriven by aliasing error [65] Filtering is a simple yet robust remedy To filter a solu-tion component such as χ we use the modal (as opposed to nodal) representation of thesolution

χkh(t r) =

Nsum

j=0

χ(t rkj )ℓkj (r) =

Nsum

j=0

χkj (t)Pj(r) (50)

where Pj(r) is the jth Legendre polynomial Let ηj = jN and define the filter function

σ(ηj) =

1 for 0 le ηj le NcN

exp(

minus ǫ(

ηjminusNcN

1minusNcN

)2s)

for NcN le ηj le 1(51)

At each timestep we modify our solution component according to

χkh rarr

(

χkh

)filtered=

Nsum

j=0

σ(ηj)χkj (t)Pj(r) (52)

Evidently the modification only affects the top N minus Nc modes and is sufficient to controlthe type of weak instability driven by aliasing [66] The numerical parameters Nc and ǫ areproblem dependent For our simulations we have taken ǫ ≃ minuslog(εmach) = 36 where εmach

is machine accuracy in double precision

15

E Model system

To better illustrate the basic properties of our method we consider a toy model Namelythe following spatially second-order system

parttu = uprime + av minus u3 + g(t x) (53a)

parttv = uprimeprime + vprime minus (u+ v)(uprime)2 + v2u2 + h(t x) (53b)

where a ge 1 is constant and g and h are analytic source terms to be specified In contrastto (6) here u v and Q = uprime are scalars rather than vectors System (53) admits a first-order reduction in which uprime is defined as an extra variable Since this first-order reductionis strongly hyperbolic the spatially second-order system (53) is also strongly hyperbolic byone of the definitions considered in [75] The characteristic fields Xplusmn and speeds microplusmn are

X+ =radicav minus uprime micro+ =

radicaminus 1 Xminus =

radicav + uprime microminus = minus(

radica+ 1) (54)

To construct a local dG scheme for this system we first rewrite it as

parttu = Q+ av minus u3 + g(t x) (55a)

parttv = Qprime + vprime minus (u+ v)Q2 + v2u2 + h(t x) (55b)

Q = uprime (55c)

Evidently f = minus(Q + v) is the v-component of the physical flux vector

F (vQ) equiv(

Fu

Fv

)

=

(

0

f

)

(56)

Note that F has the same structure as (u v)T Borrowing from the presentation for theBSSN system we write the analogous semidiscrete scheme on each subdomain D

k for themodel system

parttu = Q+ av minus u3 + g(t) (57a)

parttv = DQ+Dv minus (u+ v)Q2 + v2u2 + h(t) +Mminus1ℓk(fh minus f lowast)∣

∣

∣

bk

ak (57b)

Q = Du+Mminus1ℓk(ulowast minus uh)∣

∣

∣

bk

ak (57c)

Here we have suppressed the subinterval label k from all variables except for the vectorℓk of Lagrange polynomial values Moreover following the guidelines discussed above thenumerical fluxes are given by

f lowast = fh+1 +

radica

2

[[

vh]]

ulowast = uh minus1

2

[[

uh]]

(58)

Appendix C analyzes the stability of our scheme for a more general numerical flux choiceas applied to (53) with the nonlinear and source terms dropped

IV RESULTS FROM NUMERICAL SIMULATIONS

This section presents results found by numerically solving both the model system (53)and BSSN system (4) with the dG scheme presented in Sec III

16

10minus10

10minus6

10minus2

∆uinfin

10minus10

10minus6

10minus2

∆vinfin

0 500 1000 1500 2000 2500 3000 3500 400010

minus10

10minus6

10minus2

∆(partxu)infin

t

N = 3N = 6N = 9N = 12

FIG 1 Spectral convergence of fields for model PDE Respectively for N = 3 6 9 12

a timestep of ∆t = 00578 00178 00084 00049 has been chosen for stability and accuracy In the

title headings for example ∆u equiv unumer minus uexact

A Simulations of the model system

The semi-discrete scheme (57) has been integrated with the classical fourth-order Runge-Kutta method When integrating this system we have first constructed Q at each Runge-Kutta stage and then substituted into the evolution equations (57ab) for u and v Theproblem has been solved on a computational domain [0 4π] comprised of two subdomainswith a timestep chosen small enough for stability The initial data has been taken from thefollowing exact solution to (53)

uprimeexact(t x) =1

2

[

sin(xminus microminust)minus sin(xminus micro+t)]

(59a)

vexact(t x) =1

2radica

[

sin(xminus microminust) + sin(xminus micro+t)]

(59b)

g(t x) = u3exact (59c)

h(t x) = (uexact + vexact)(uprimeexact)

2 minus v2exactu2exact (59d)

where the speeds microplusmn are found in (54) Specification of the boundary condition at a physicalendpoint amounts to choosing the external state for at the endpoint We have considered

17

8 16 32 64 128

10minus3

10minus2

10minus1

N

∆t

max stable ∆t8N2

FIG 2 Scaling of maximum stable ∆t with N for model PDE

two possibilities (i) the analytic state (Q+ v+) = (Qexact vexact) and (ii) an upwind stateFor example at x = 4π the upwind state is4

Q+ = Qupwind =1

2

[

(Xminus)exact minus (X+)numer

]

v+ = vupwind =1

2radica

[

(Xminus)exact + (X+)numer

]

(60)Either choice of (Q+ v+) leads to similar results and the plots here correspond to theanalytic state Figure 1 clearly shows spectral convergence with increasing polynomial orderN across all fields for the case a = 2 Other values of a including a = 1 for which X+

is a static characteristic field have also been considered with similar results Appendix Cdemonstrates that our proposed scheme for the system (57) with nonlinear and source termsdropped is stable in a semi-discrete sense Nevertheless the fully discrete scheme obtainedvia temporal discretization by the fourth-order Runge-Kutta method is still subject to thestandard absolute stability requirement Namely if microh is any eigenvalue corresponding tothe (linearized) discrete spatial operator then a necessary condition for stability is thatmicroh∆t lies in absolute stability region for Runge-Kutta 4 We here show empirically that theassociated timestep restriction scales like Nminus2 ie ∆t = O(Nminus2) for stability We note thatsuch scaling is welcome in light of the second-order spatial operators which appear in thesystem and suggest a possible worse scaling like Nminus4 Fig 2 plots the maximum stabletimestep for a range of N demonstrating the Nminus2 scaling in line with behavior known fromanalysis of first-order systems [66] This scaling also holds for the BSSN system

4 We remind the reader that unfortunately the plusmn on Xplusmn means something different than the plusmn indicating

exteriorinterior dG states [cf footnote 3]

18

B Simulations of the BSSN system

This subsection documents results for simulations of the unit-mass-parameter (M = 1)Schwarzschild solution (B9) expressed in terms of ingoing Kerr-Schild coordinates Sincethe solution is stationary temporal integration of the semi-discrete scheme has been carriedout with the forward Euler method The r-coordinate domain [04 34] has been split into3 equally spaced subdomains and we have set η = 10 λ = 01 and C = 2 [cf Eq (41)]For all simulations ∆t has been chosen for stability With the chosen λ the inner physicalboundary rmin = 04 is an excision surface At each timestep we have applied an (order2s = 20) exponential filter on the top two-thirds of the modal coefficient set for all fieldsexcept for grr and gθθ For stability we have empirically observed that grr and gθθ must not

be filtered A detailed understanding of this is still lackingIssues related to physical boundary conditions are similar to the one encountered in

Sec IVA for the model problem Similar to before we have retained Eqs (3842) as thechoice of numerical flux even at the endpoints Therefore at an endpoint the specificationof the boundary condition amounts to the choice W+ of external state We have typicallychosen the inner boundary of the radial domain as an excision boundary and in this caseW+ = Wminus is enforced at the inner physical boundary At the outer physical boundaryfor W+ we have again considered two choices (i) Wexact and (ii) Wupwind To enforcechoice (ii) the inverse transformation (17) must be used with incoming characteristic fieldsfixed to their exact values similar to (60) We have tried various versions of choice (ii)and in all cases the resulting simulations have been unstable We therefore present resultscorresponding to choice (i) Although the choice of an analytical external state Wexact

at the outer boundary is stable for our problem such a boundary condition is unlikelyto generalize to more complicated scenarios involving dynamical fields Indeed the issueof outer boundary conditions for the BSSN system is an active area of research with aproper treatment requiring fixation of incoming radiation control of the constraints andspecification of gauge (see Ref [86] for a recent analysis)

For BSSN simulations our main diagnostic is to monitor the Hamiltonian momentumand conformal connection constraints Figure 3 depicts long-time histories of constraintviolations whereas Figs 4 and 5 depict long-time error histories for the individual BSSNfield components From the middle plot in Fig 5 we infer that up to the indicated numericalerror the factor g sin4 θ = grr(gθθ)

2 remains at its initial fixed profile r4 throughout theevolution These figures indicate that the proposed scheme is stable for long times andexhibits spectral converge with increased polynomial order N Similar results are recoveredfromM = 0 Minkowski initial data The stability documented in these plots does not appearto rely on inordinate parameter tuning For example with the fixed parameters describedabove we obtain similar plots if we individually vary (i) rmin over 0325 035 04 0475(values still corresponding to an excision surface for the given choice of λ) (ii) η over1 3 7 10 (iii) s over 8 9 10 With the polynomial order N ranging over 23 26 29 31both stability and qualitatively similar exponential convergence is achieved with a singlesubdomain Likewise adoption of a larger coordinate domain with more subdomains doesnot impact our results Finally we have considered the addition of random noise to all fieldcomponents at the initial time Precisely with the system component χ as an example wehave set

χ(0) 7rarr χ(0) + δχ(0) (61)

where each component (nodal value) of δχ(0) is 10minus5 times a random variable drawn from a

19

10minus9

10minus7

10minus5

10minus3

Hinfin

10minus9

10minus7

10minus5

10minus3

Mrinfin

0 2000 4000 6000 8000 10000

10minus9

10minus7

10minus5

10minus3

Grinfin

t

N = 11N = 14N = 17N = 20

FIG 3 Spectral convergence of constraint violations for M = 1 Kerr-Schild ini-

tial data Respectively for N = 11 14 17 19 a timestep of ∆t ≃ 00041 00026 00018 00013

has been chosen for stability and accuracy

standard normal distribution Such perturbed initial data also gives rise to stable evolutions

V CONCLUSION

We have introduced a discontinuous Galerkin method for solving the spherically reducedBSSN system with second-order spatial operators Our scheme shares similarities withother discontinuous Galerkin methods that use local auxiliary variables to handle high-orderspatial derivatives [66 77ndash79 82 84 85] and which have typically been applied to eitherelliptic parabolic or mixed type problems The key ingredient of a stable dG scheme is anappropriate choice of numerical flux and our particular choice has been motivated by theanalysis presented in Appendix C When used to evolve the Schwarzschild solution in Kerr-Schild coordinates our numerical implementation of the BSSN system (4) is robustly stableand converges to the analytic solution exponentially with increased polynomial order Byapproximating the spatially second-order form of the BSSN system we have not introducedextra fields which are evolved Evolved auxiliary fields result in new constraints which mayspoil stability Our main goal has been stable evolution of the spherically reduced BSSNsystem as a first step towards understanding how a discontinuous Galerkin method might be

20

10minus9

10minus7

10minus5

∆Arrinfin

10minus9

10minus7

10minus5

∆Kinfin

0 2000 4000 6000 8000 10000

10minus9

10minus7

10minus5

max(∆Brinfin ∆Γrinfin)

t

N = 11N = 14N = 17N = 20

FIG 4 Spectral convergence of solution for M = 1 Kerr-Schild initial data

Timestep choices are described in the caption for Fig 3 In the title headings for example

∆Arr equiv (Arr)numer minus (Arr)exact

applied to the full BSSN system Towards that goal we now discuss treatment of singularitiesand generalization of the described dG method to higher space dimension

To deal with the fixed Schwarzschild singularity we have used excision which is easy inthe context of the spherically reduced BSSN system However excision for the binary blackhole problem in full general relativity requires attention to the technical challenge of hori-zon tracking State-of-the-art BSSN codes avoid such complication relying instead on themoving-puncture technique While the moving-puncture technique does involve mild centralsingularities it may still prove amenable to spectral methods Indeed spectral methods fornon-smooth problems is well-developed in both theory and for complex applications Sincethe moving-puncture technique can be performed in spherical symmetry [60] a first-steptoward a spectral moving-puncture code would be to implement a moving puncture withthe nodal dG method described here Such an implementation may adopt Legendre-Gauss-Radau nodes on the innermost subdomain thereby ensuring that the physical singularitydoes not lie on a nodal point (in much the same way finite difference codes use a staggeredgrid) Beyond traditional excision and moving punctures one might also retain a smoothproblem for which spectral methods are especially well-suited via the turducken approach[87 88] to singularities

21

10minus11

10minus9

10minus7

10minus5

∆χinfin

10minus11

10minus9

10minus7

10minus5

max(∆grrinfin ∆gθθinfin)

0 2000 4000 6000 8000 10000

10minus11

10minus9

10minus7

10minus5

max(∆αinfin ∆βrinfin)

t

N = 11N = 14N = 17N = 20

FIG 5 Spectral convergence of solution violations for M = 1 Kerr-Schild initial

data See the caption of Fig 4 for details

Discontinuous Galerkin methods for hyperbolic problems in two and three space dimen-sions are well-developed A generalization of the method described here to three-dimensionsand the full BSSN system would likely rely on an unstructured mesh Appropriate local poly-nomial expansions for the subelements are well-understood as are choices for the numericalfluxes which would now live on two-dimensional faces rather than single points Whether ornot it would ultimately prove successful generalization of our dG method to a higher dimen-sion would rely on an established conceptual framework Further computational advancesof relevance to a generalization of our dG method to the full BSSN system (possibly includ-ing matter) may include mesh hp-adaptivity local timestepping shock capturing and slopelimiting techniques [66] Moreover recent work [89] indicates that enhanced performancewould be expected were our scheme implemented on graphics processor units

VI ACKNOWLEDGMENTS

We thank Nick Taylor for discussions about treating second-order operators with spectralmethods Benjamin Stamm for helping polish up a few parts of Appendix C Khosro Shah-bazi for discussions on LDG and IP methods We also acknowledge helpful conversationswith David Brown and Manuel Tiglio about the BSSN system and previous work on its nu-

22

merical implementation We gratefully acknowledge funding through grants DMS 0554377and DARPAAFOSR FA9550-05-1-0108 to Brown University and NSF grant PHY 0855678to the University of New Mexico

Appendix A Hyperbolicity of the first-order system

This appendix analyzes the matrix A(u) appearing in (10) in order to construct thecharacteristic fields (14) In matrix form the sector (8) of the principal part of (10) reads asfollows

partt

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

=

βr 0 minus 4λα3grr

0 0 λβr

6(grr)2λβr

3gθθgrr0 4λ

3grr

0 βr 0 23grrαχ

13α minus αχ

3grr

αχ3gθθ

minus23χ 0

0 0 βr 0 0 0 0 minus χgrr

0

0 0 minus 4α3grr

βr 0 βr

6(grr)2βr

3gθθgrr0 4

3grr

0 0 23αχ 0 βr minus βrχ

3grrminus2βrχ

3gθθ0 minus2

3χ

0 minus2α 0 0 0 23βr minus2grrβr

3gθθ0 4

3grr

0 gθθαgrr

0 0 0 minusgθθβr

3grr13βr 0 minus2

3gθθ

0 0 minus2α 0 0 0 0 βr 0

34

0 0 0 0 0 0 0 βr

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

prime

(A1)which defines the matrix A(u) appearing in (9) and so also the matrix A(u) in (10) Notethat in the last equation the matrix within the square brackets is minusA(u) For certainconfigurations of u and λ the system (10) is strongly hyperbolic [76] that is A(u) hasa complete set of eigenvectors and real eigenvalues Indeed five eigenpairs of A(u) aretrivially recovered upon inspection of A(u)rsquos leading 5times5 diagonal block These correspondto eigenvalue 0 and the left eigenspace ξj = eTj 1 le j le 5 where ej are the canonical

basis vectors Since each component of u arises as eTj W each is also a characteristic field

The remaining nine eigenpairs are determined by A(u) The eigenvalues of A(u) are

micro1 = 0 micro23 = minusβr microplusmn4 = minusβr plusmn

radic

2αχ

grr microplusmn

5 = minusβr plusmn α

radic

χ

grr microplusmn

6 = minusβr plusmnradic

λ

grr

(A2)

23

and the corresponding left eigenvectors are

x1 = (0 0 0 0 0 gθθ 2grr 0 0) (A3a)

x2 =

(

0 0 0 grr2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3b)

x3 =

(

grrλ 0 0 0

2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3c)

xplusmn4 =

(

0 0plusmnradic

2αgrrχ

0 0 0 0 1 0

)

(A3d)

xplusmn5 =

(

0∓ 3radicgrrχ

plusmn2

radic

grrχ 2grr

1

χminus 1

grr1

gθθ 0 0

)

(A3e)

xplusmn6 =

(

minus3

4

grrλ 0plusmn α

radicλgrr

(2αχminus λ) 0 0minus βr

8(βrgrr ∓radicλgrr)

minus βrgrr

4gθθ(βrgrr ∓radicλgrr)

αχ

(2αχminus λ)plusmnradic

grrλ

)

(A3f)

where for example xplusmn5 A(u) = microplusmn5 x

plusmn5 Assuming that grr gθθ χ and α are everywhere strictly

positive the eigenvalues are real and the eigenvectors are linearly independent providedthat (15) holds These eigenvectors are easily extended to eigenvectors of A(u) e g asxplusmn6 rarr (01times5 x

plusmn6 ) Then for example the characteristic field

Xplusmn6 equiv (01times5 x

plusmn6 )W = xplusmn6WvQ (A4)

and similarly Xplusmnj = xplusmnj WvQ for j = 4 5 and Xk = xkWvQ for k = 1 2 3 The characteristic

speeds for these fields are microk and microplusmnj With this convention the speeds listed in Table I

correspond to the Xk and Xplusmnj in (14)

Appendix B Schwarzschild solution in conformal Kerr-Schild coordinates

In Kerr-Schild coordinates here the system directly related to incoming Eddington-Finkelstein null coordinates the line element for the Schwarzschild solution reads

ds2 = minusα2dt2 + (1 + 2MR)(dR + βRdt)2 +R2dθ2 +R2 sin2 θdφ2 (B1)

where R is the area radius α = (1 + 2MR)minus12 is the lapse and βR = 2M(R + 2M) isthe shift vector The physical spatial metric gab is the spatial part of this line element

To define the corresponding solution to the BSSN system we use equation gab = χgab todefine the following relationship between line elements

dr2 + r2(dθ2 + sin2 θdφ2) = χ[(1 + 2MR)dR2 +R2dθ2 +R2 sin2 θdφ2] (B2)

so that

χ

(

1 +2M

R

)(

dR

dr

)2

= 1 χR2 = r2 (B3)

Then we have(

1 +2M

R

)12dR

R=dr

r (B4)

24

with integration yielding

r =R

4

(

1 +

radic

1 +2M

R

)2

e2minus2radic

1+2MR (B5)

where the constant of integration has been chosen so that the R r rarr infin limits are consistentThe second relation in (B3) shows that

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR =rArr χminus4 =2eradic

1+2MRminus1

1 +radic

1 + 2MR (B6)

The extrinsic curvature tensor is specified by the expression forK given in (B9h) the identityK = KR

R + 2Kθθ and

Kθθ =

(

1 +2M

R

)minus122M

R2 (B7)

Since KRR = Kr

r we compute that

Krr = K minus 2Kθ

θ = minus(

1 +2M

R

)minus12(R +M

R + 2M

)

2M

R2 (B8)

Next since Krr = grrKrr = χminus1Kr

r we have Krr = Arr+

13grrK This implies Arr = Kr

r minus 13K

from which we get (B9g) In all we have

α =

(

1 +2M

R

)minus12

(B9a)

βr = βR dr

dR= χ12

(

1 +2M

R

)minus122M

R(B9b)

grr = 1 (B9c)

gθθ = r2 = χR2 (B9d)

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR (B9e)

Br = 0 (B9f)

Arr = minus(

1 +2M

R

)minus124M

3R2

(

2R + 3M

R + 2M

)

(B9g)

K =

(

1 +2M

R

)minus32(

1 +3M

R

)

2M

R2(B9h)

Γr = minus2

r= minus 2

χ12R (B9i)

To differentiate these expressions with respect to r we use the identity

dR

dr= χminus12

(

1 +2M

R

)minus12

(B10)

along with the chain rule

25

Appendix C Stability of the model system

The following stability analysis for the model system (53) has been inspired by [78 79]After dropping all nonlinear source terms the system (53) becomes

parttu = uprime + av (C1a)

parttv = uprimeprime + vprime (C1b)

This section analyzes the stability of (C1) considering both the continuum system itself aswell as its semi-discrete dG approximation The latter analysis offers some insight into theempirically observed stability of our dG scheme for the spherically reduced BSSN equations

1 Analysis for a single interval

Throughout we work with the L2 inner product and norm

(f g)D =

int

D

fg fD =radic

(f f)D (C2)

where D is the spatial coordinate interval (here D may represent a subdomain Dk or the

whole domain Ω) and we have suppressed all integration measures For the continuummodel we will establish the following estimate

uprime(T middot)2D+ av(T middot)2

Dle C(T )

(

uprime(0 middot)2D+ av(0 middot)2

D

)

(C3)

where the time-dependent constant C(T ) is determined solely by the choice of boundaryconditions To show (C3) we first change variables with v =

radicav thereby rewriting (C1)

in the following symmetric form

parttu = uprime +radicav (C4a)

parttv =radicauprimeprime + vprime (C4b)

Equations (C4ab) then imply

1

2partt

int

D

(uprime)2 =

int

D

uprime(uprimeprime +radicavprime) =

int

D

radicauprimevprime +

1

2

int

partD

(uprime)2 (C5a)

1

2partt

int

D

(v)2 =

int

D

v(radicauprimeprime + vprime) = minus

int

D

radicauprimevprime +

1

2

int

partD

(v2 + 2radicauprimev) (C5b)

Here vvprime and uprimeuprimeprime have been expressed as exact derivatives and then integrated to boundaryterms the second equation employs an extra integration by parts and with only one spacedimension

int

partDdenotes a difference of endpoint evaluations Addition of Eqs (C5ab) gives

1

2partt

int

D

[

v2 + (uprime)2]

=1

2

int

partD

[

v2 + (uprime)2 + 2radicauprimev

]

(C6)

Substitutions with the identities

[

v2 + (uprime)2]

=1

2

[

(v + uprime)2 + (v minus uprime)2]

2uprimev =1

2

[

(v + uprime)2 minus (v minus uprime)2]

(C7)

26

and replacements to recover the original variable v = vradica yield

1

2partt

int

D

[

av2 + (uprime)2]

=1 +

radica

4

int

partD

(radicav + uprime)2 +

1minusradica

4

int

partD

(radicav minus uprime)2 (C8)

From (C8) we deduce that the time-dependent constant C(T ) in (C3) must satisfy

∣

∣

∣

∣

∣

∣

1 +

int T

0

[

1+radica

2

int

partD(radicav + uprime)2 + 1minus

radica

2

int

partD(radicav minus uprime)2

]

dt

uprime(0 middot)2 + av(0 middot)2

∣

∣

∣

∣

∣

∣

le C(T ) (C9)

For periodic boundary conditions we may choose C(T ) = 1 Moreover if a ge 1 anduprime = minusradic

av is specified at partD+ then uprime(t middot)2 + av(t middot)2 decaysStill working on a single interval (subdomain) we now consider the semi-discrete scheme

for (C4) i e (57) with all nonlinear source terms dropped and with v replaced by vradica

Derivation of a formula analogous to (C8) is our first step toward establishing L2 stabilityof the semi-discrete scheme While (57) features vectors for example u(t) taking values atthe Legendre-Gauss-Lobatto nodal points here we work with the numerical solution as apolynomial for example uh(t x) These two representations are related by the Lagrange in-terpolating polynomials for the nodal set here taken to span both the space of test functionsand the space of basis functions Our scheme is

int

Dk

ψparttuh =

int

Dk

ψ(Qh +radicavh) (C10a)

int

Dk

ξparttvh = minusint

Dk

ξprime(radicaQh + vh) +

int

partDk

ξ(radicaQlowast + vlowast) (C10b)

int

Dk

ϕQh =

int

Dk

ϕuprimeh +

int

partDk

ϕ (ulowast minus uh) (C10c)

where ψ ξ and ϕ are polynomial test functions These test functions are arbitrary exceptthat they must be degree-N polynomials In (C10) the variables uh vh and Qh should alsocarry a superscript k but we have suppressed this Derivation of a formula analogous to(C8) is complicated by the fact that Qh is not evolved Nevertheless at a given instant t wecan assemble Qh from (C10c)

Mimicking the calculation (C5b) from the continuum case we first use (C10b) with ξ = vhto write

1

2partt

int

Dk

v2h = minusint

Dk

(radicaQh + vh)v

primeh +

int

partDk

(radicaQlowast + vlowast)vh

= minusint

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2(radicaQlowast + vlowast)vh minus v2h

]

(C11)

The right-hand side of (C5a) is analogous to

1

2partt

int

Dk

Q2h =

int

Dk

QhparttQh (C12)

However since Qh is not evolved the term parttQh must be given a suitable interpretationOn the right hand side of (C10c) only uh u

primeh and u

lowast necessarily depend on time since the

27

test function ϕ need not be time-dependent Furthermore ulowast is explicitly given as a linearcombination of uh as seen in Eq (C21c) below Choosing ϕ = ℓj taking the time derivativeof (C10c) and appealing to the commutivity of mixed partial derivatives we therefore arriveat

int

Dk

ℓjparttQh =

int

Dk

ℓj(parttuh)prime +

int

partDk

ℓj(

(parttu)lowast minus parttuh

)

(C13)

where (parttu)lowast depends on parttuh in precisely the same way that ulowast depends on uh We have

written ℓj rather than ϕ in the last equation to emphasize that the result also holds for anylinear combination of ℓj (for example ϕ) and even for time-dependent combinations SinceQh is itself such a combination we obtain

1

2partt

int

Dk

Q2h =

int

Dk

Qh(parttuh)prime +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

Qh(Qprimeh +

radicavprimeh) +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2((parttu)lowast minus parttuh)Qh +Q2

h

]

(C14)

Addition of (C11) and (C14) gives

1

2partt

int

Dk

(Q2h + v2h) =

1

2

int

partDk

[

Q2h minus v2h + 2(

radicaQlowast + vlowast)vh + 2((parttu)

lowast minus parttuh)Qh

]

(C15)

the aforementioned analog of (C8) This formula holds on a single subdomain Dk and we

now combine multiple copies of it one for each value of k

2 Analysis for multiple intervals

To facilitate combination of (C15) over all k we change notation At every subdomaininterface Ik+12 equiv partDk cap partDk+1 let the superscripts L and R denote field values respectivelytaken from the left and right Then the fields evaluated at Ik+12 which belong to D

k will beuLk+12 v

Lk+12 and Q

Lk+12 while those belonging to D

k+1 will be uRk+12 vRk+12 and Q

Rk+12

However at Ikminus12 the values taken from D

k are uRkminus12 vRkminus12 and QR

kminus12 Note that we

have also replaced the subscript h denoting a numerical solution with k plusmn 12 denotingthe location of the endpoint value of the numerical solution With this notation we define

∆Lα =

1

2

[

(QLα)

2 minus (vLα)2]

+(radic

aQlowastα + vlowastα

)

vLα +[

(parttuα)lowast minus parttu

Lα

]

QLα (C16)

and similarly for ∆Rα The same numerical fluxes appear in both ∆L

α and ∆Rα (ie each

numerical flux takes the same value on either side of an interface) whence fluxes like Qlowastα do

not carry an L or R superscript In terms of these definitions (C15) becomes

1

2partt

int

Dk

(Q2h + v2h) = ∆L

k+12 minus∆Rkminus12 (C17)

28

Summation over all Dk yields

1

2partt

kmaxsum

k=1

int

Dk

(Q2h + v2h) =

kmaxminus1sum

k=1

(

∆Lk+12 minus∆R

k+12

)

+∆Lkmax+12 minus∆R

12

=

kmaxminus1sum

k=1

(

∆Lh minus∆R

h

)∣

∣

Ik+12 +∆Lkmax+12 minus∆R

12 (C18)

We have reverted to h-notation denoting the numerical solution since the LR superscriptsindicate unambiguously the relevant domain used for evaluation at Ik+12

We again seek an estimate of the form

kmaxsum

k=1

(

Qh(T middot)2Dk + avh(T middot)2Dk

)

le C(T )kmaxsum

k=1

(

Qh(0 middot)2Dk + avh(0 middot)2Dk

)

(C19)

that is essentially the same as the one (C3) considered in the continuum case Assume thatthe chosen boundary conditions ensure ∆L

kmax+12 minus ∆R12 is bounded by a time-dependent

constant which does not depend on the numerical parameters N and h (subdomain width)Establishment of stability then amounts to showing that the remaining sum over interfaceterms in (C18) is non-positive whence this remaining sum is consistent with C(T ) le 1although the boundary conditions may give rise to a different bound In fact we will choosethe numerical fluxes such that each individual interface term is non-positive At interfaceIk+12 and in LR notation the jump and average of vh for example are

1

2

(

v+ + vminus)

equiv vh =1

2

(

vLk+12 + vRk+12

)

(C20a)

nminusvminus + n+v+ equiv[[

vh]]

= vLk+12 minus vRk+12 (C20b)

Consider numerical fluxes of the form

Qlowast = Qh minusτQ2

[[

Qh

]]

(C21a)

vlowast = vh minusτv2

[[

vh]]

(C21b)

ulowast = uh minusτu2

[[

uh]]

(C21c)

(parttu)lowast = parttuh minus

τu2

[[

parttuh]]

(C21d)

where (C21c) induces (C21d) and where the penalty parameters τu τv and τQ are realnumbers The fluxes defined in (58) correspond to τu = 1 τv = 1 +

radica and τQ = 0 In

terms of these quantities the kth interface contribution in (C18) is

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

parttuh]]

minus τu2

[[

parttuh]][[

Qh

]]

(C22)

where we have suppressed the k dependence of the right-hand side Now consider the term[[

parttuh]]

Because parttuh and Qh+radicavh are both polynomials of degree N Eq (C10a) implies

29

τu

τ Q

minus1 0 1minus1

minus05

0

05

1

τu

minus1 0 1minus1

minus05

0

05

1

FIG 6 Stable evolutions for the model system For fixed τv = 10minus6 and τv = 1 +radic2

respectively the left and right plots depict stable choices (determined empirically) of τu and τQfor the linear model system (C1) The stable regions are colored black but the jagged edges result

from the discretization of the (τu τQ)-plane

the vector equation parttu = Q +radicav that is pointwise equivalence on the nodal points of

Dk which in turn implies

[[

parttuh]]

=[[

Qh+radicavh]]

Upon substituting this identity into thelast equation we arrive at an expression which features only vh and Qh

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

Qh +radicavh]]

minus τu2

[[

Qh +radicavh]][[

Qh

]]

(C23)

The identities vh[[

vh]]

= 12

[[

v2h]]

and[[

Qh +radicavh]]

=[[

Qh

]]

+radica[[

vh]]

then simplify(C23) to

(∆Lh minus∆R

h )∣

∣

Ik+12 = minusτv2

[[

vh]]2 minus

radica(τu + τQ)

2

[[

Qh

]][[

vh]]

minus τu2

[[

Qh

]]2 (C24)

The role of a penalty parameter is now clear Positive values of τv penalize jumps in vhthrough a negative contribution to the energy Likewise positive values of τu penalizejumps in Qh through a negative contribution to the energy However because the sign of[[

Qh

]][[

vh]]

can be positive or negative only the choice τQ = minusτu yields an expression for(∆L

h minus ∆Rh )|Ik+12 which is manifestly negative for τu ge 0 and τv ge 0 A simple estimate

based on Youngrsquos inequality with ε (that is 2αβ le εminus1α2 + εβ2 where α β ge 0 and ε gt 0)shows that for τQ = 0 the choice τv ge aτu4 also yields stability

Figure 6 depicts certain choices of stable penalty parameters for the linear model systemevolved to tfinal = 1000 (with a = 2 N = 10 and ∆t ≃ 00553) as determined empiricallywith simulations similar to those described in Sec IVA The left plot corresponds to asmall τv = 10minus6 for which the choice τu = 1 τQ = 0 is not stable as expected from thetheoretical analysis However the right plot corresponds to τv = 1 +

radica for which τu = 1

τQ = 0 is stable Motivated by the numerical flux choices (3842) used for the BSSN system

30

(4) we have (as mentioned above) set τu = 1 τv = 1 +radica and τQ = 0 in simulations of

the nonlinear model (53) For the nonlinear model system (53) the theoretically motivatedchoice τQ = minusτu also yields numerically stable evolutions when τu ge 0 and τv ge 0

For the nonlinear systems (4) and (53) we do not attempt a formal stability proofNevertheless the results of this appendix have served as a guide for our choices of penaltyparameters For the BSSN system (4) u v and Q are block indices [cf Eq (6)] Similarto the model problem we have penalized Q with τu = 1 with τv chosen large enough toheuristically overcome the cross-terms of indefinite size that arise from τQ = 0 (we interpretequations like τu = 1 componentwise) An analogous choice ldquoτQ = minusτurdquo for the BSSNsystem might be possible but would be considerably more complicated Indeed such achoice likely entails a matrix of penalty parameters but we do not give the details here

[1] F Pretorius Evolution of Binary Black-Hole Spacetimes Phys Rev Lett 95 121101 (2005)

(4 pages)

[2] F Pretorius Simulation of binary black hole spacetimes with a harmonic evolution scheme

Class Quantum Grav 23 S529-S552 (2006)

[3] M Campanelli C O Lousto P Marronetti and Y Zlochower Accurate Evolutions of Or-

biting Black-Hole Binaries without Excision Phys Rev Lett 96 111101 (2006) (4 pages)

[4] J G Baker J Centrella D-I Choi M Koppitz and J van Meter Gravitational-Wave

Extraction from an Inspiraling Configuration of Merging Black Holes Phys Rev Lett 96

111102 (2006) (4 pages)

[5] M Campanelli C O Lousto and Y Zlochower Last orbit of binary black holes Phys Rev D

73 061501(R) (2006) (5 pages)

[6] F Herrmann I Hinder D Shoemaker and P Laguna Unequal mass binary black hole plunges

and gravitational recoil Class Quantum Grav 24 S33-S42 (2007)

[7] P Diener F Herrmann D Pollney E Schnetter E Seidel R Takahashi J Thornburg and

J Ventrella Accurate Evolution of Orbiting Binary Black Holes Phys Rev Lett 96 121101

(2006) (4 pages)

[8] M A Scheel H P Pfeiffer L Lindblom L E Kidder O Rinne and S A Teukolsky Solving

Einsteinrsquos Equations With Dual Coordinate Frames Phys Rev D 74 104006 (2006)

[9] U Sperhake Binary black-hole evolutions of excision and puncture data Phys Rev D 76

104015 (2007) (20 pages)

[10] B Brugmann J A Gonzalez M Hannam S Husa U Sperhake and W Tichy Calibration

of moving puncture simulations Phys Rev D 77 024027 (2008) (25 pages)

[11] P Marronetti W Tichy B Brugmann J Gonzalez M Hannam S Husa and U Sperhake

Binary black holes on a budget simulations using workstations Class Quantum Grav 24

S43-S58 (2007)

[12] Z B Etienne J A Faber Y T Liu S L Shapiro and T W Baumgarte Filling the holes

Evolving excised binary black hole initial data with puncture techniques Phys Rev D 76

101503(R) (2007) (5 pages)

[13] B Szilagyi D Pollney L Rezzolla J Thornburg and J Winicour An explicit harmonic code

for black-hole evolution using excision Class Quantum Grav 24 S275-S293 (2007)

[14] L Boyle M Kesden and S Nissanke Binary-Black-Hole Merger Symmetry and the Spin

Expansion Phys Rev Lett 100 151101 (2008) (4 pages)

31

[15] M A Scheel M Boyle T Chu L E Kidder K D Matthews and H P Pfeiffer High-

accuracy waveforms for binary black hole inspiral merger and ringdown Phys Rev D 79

024003 (2009) (14 pages)

[16] M Hannam Status of black-hole-binary simulations for gravitational-wave detection

Class Quantum Grav 26 114001 (2009) (21 pages)

[17] I Hinder The current status of binary black hole simulations in numerical relativity

Class Quantum Grav 27 114004 (2010) (19 pages)

[18] B P Abbott et al LIGO The Laser Interferometer Gravitational-Wave Observatory

Rep Prog Phys 72 076901 (2009) (25 pages)

[19] F Acernese et al The status of VIRGO Class Quantum Grav 23 S63-S69 (2006)

[20] F Acernese et al Virgo status Class Quantum Grav 25 184001 (2008) (9 pages)

[21] R Arnowitt S Deser and C W Misner The Dynamics of General Relativity in Gravitation

an introduction to current research edited by L Witten (Wiley New York 1962) Chapter 7

227-265 Also available as gr-qc0405109

[22] J W York Jr Kinematics and Dynamics of General Relativity in Sources of Gravitational

Radiation edited by L L Smarr (Cambridge University Press Cambridge 1979)

[23] L E Kidder M A Scheel and S A Teukolsky Extending the lifetime of 3D black hole

computations with a new hyperbolic system of evolution equations Phys Rev D 64 064017

(2001) (13 pages)

[24] G Calabrese J Pullin O Sarbach and M Tiglio Convergence and stability in numerical

relativity Phys Rev D 66 041501(R) (2002) (4 pages)

[25] S Frittelli and O Reula On the Newtonian Limit of General Relativity Commun Math

Phys 166 221-235 (1994)

[26] M Shibata and T Nakamura Evolution of three-dimensional gravitational waves Harmonic

slicing case Phys Rev D 52 5428-5444 (1995)

[27] Y Choquet-Bruhat and J W York Geometrical well posed systems for the Einstein equations

C R Acad Sci Paris t 321 Serie I 1089-1095 (1995)

[28] A Abrahams A Anderson Y Choquet-Bruhat and J W York Jr Einstein and Yang-Mills

Theories in Hyperbolic Form without Gauge Fixing Phys Rev Lett 75 3377-3381 (1995)

[29] C Bona J Masso E Seidel and J Stela New Formalism for Numerical Relativity

Phys Rev Lett 75 600-603 (1995)

[30] M H P M van Putten and D M Eardley Nonlinear wave equations for relativity

Phys Rev D 53 3056-3063 (1996)

[31] S Frittelli and O A Reula First-Order Symmetric Hyperbolic Einstein Equations with Ar-

bitrary Fixed Gauge Phys Rev Lett 76 4667-4670 (1996)

[32] H Friedrich Hyperbolic reductions for Einsteinrsquos equations Class Quantum Grav 13 1451-

1469 (1996)

[33] F B Estabrook R S Robinson H D Wahlquist Hyperbolic equations for vacuum gravity

using special orthonormal frames Class Quantum Grav 14 1237-1247 (1997)

[34] M S Iriondo E O Leguizamon and O A Reula Einsteinrsquos Equations in Ashtekarrsquos Vari-

ables Constitute a Symmetric Hyperbolic System Phys Rev Lett 79 4732-4735 (1997)

[35] A Anderson Y Choquet-Bruhat and J W York Einstein-Bianchi hyperbolic system for

general relativity Topol Meth Nonlinear Anal 10 353-373 (1997)

[36] T W Baumgarte and S L Shapiro Numerical integration of Einsteinrsquos field equations

Phys Rev D 59 024007 (1998) (7 pages)

[37] M A G Bonilla Symmetric hyperbolic systems for Bianchi equations Class Quantum

3

implementation [60 63 64] (see also [61 62]) Here we exploit this system to a similar endusing it as a simplified setting in which to develop spectral methods for the stable integrationof the BSSN system Precisely we develop and test a nodal discontinuous Galerkin method(dG) [66] for integration of the spherically reduced BSSN system While Brownrsquos chief focuslay with moving punctures for further simplicity we adopt the excision technique Clearlythe problem we consider is not as daunting as the one confronted by both Tichy and Mroue[57ndash59] Nevertheless our method is robustly stable and therefore might serve as a steppingstone toward a stable dG-based formulation for the full BSSN system The conclusion offersfurther comments toward this end

Nodal dG schemes are both well-suited and well-developed for hyperbolic problems [66]Although mostly used for hyperbolic problems expressed as first-order systems dG methodshave also been applied to systems involving second-order spatial operators typically via dGinterior penalty (IP) methods [67ndash72] (Refs [73ndash75] discuss the concept of hyperbolicity[76] in the context of such systems) Penalty methods of a different type were exploited inRef [55] for the wave equation written in second order form Local discontinuous Galerkin

(LDG) schemes developed initially by Shu and coworkers [77ndash79] constitute an alternateapproach for integration of spatially second-order systems LDG schemes feature essentiallythe same auxiliary variables as those appearing in traditional first-order reductions howeverin LDG schemes such variables are not evolved and arise only as local variables The basicdifference between dGndashIP and LDGmethods is the manner in which subdomains are coupledThe method we described for the spherically reduced BSSN system is essentially an LDGscheme

This paper is organized as follows Section II collects the relevant equations from Brownrsquospresentation and develops some further notation useful for expressing the spherically re-duced BSSN system in various abstract forms Section III presents our nodal dG schemein detail and Section IV documents the results of several numerical simulations testing ourscheme Our conclusion discusses possible generalization of our method to the full BSSNsystem Several appendices collect further technical details In particular Appendix Cconsiders a simple system which models the spherically reduced BSSN system giving ananalytical proof that the model system is L2 stable in the semi-discrete sense

II SPHERICALLY SYMMETRIC (GENERALIZED) BSSN EQUATIONS

As shown by Brown [50] the BSSN system can be generalized to allow for a conformalmetric without unit determinant and this paper focuses on the spherical reduction of thissystem also considered by Brown in [60] In fact this spherical reduction relies on freedompresent in the generalized BSSN system since spherical-polar coordinates should not beassociated with a unit-determinant conformal metric Although we work with the sphericallyreduced generalized BSSN system (subject to Brownrsquos Lagrangian condition to be precise)we will nevertheless describe it as the spherically reduced BSSN system

A Basic variables and spherically reduced system

The conformal-traceless decomposition of the geometry associated with a spacelike 3-surface is

gab = χminus1gab Kab = χminus1(

Aab +1

3gabK

)

(1)

4

where gab is the physical 3-metric and Kab is the physical extrinsic curvature tensor TheBSSN variables are the conformal metric gab the conformal factor χ the trace-free extrinsiccurvature Aab the trace K = gabKab and the conformal connection Γa equiv minusgminus12partb(g

12gab)where g is the determinant of the metric The BSSN system also includes the lapse α shiftvector βa and an auxiliary vector field Ba used to define the ldquoΓ-driverrdquo for the shift

Following Brown we adopt a spherically symmetric line element

ds2 = minusα2dt2 + χminus1grr(dr + βrdt)2 + χminus1gθθ(dθ2 + sin2 θdφ2) (2)

along with the spherically symmetric Ansatz

Γa =

Γr

minus cos θ(gθθ sin θ)0

Aab = Arr

1 0 00 minusgθθ(2grr) 00 0 minusgθθ sin2 θ(2grr)

(3a)

Subject to the assumption of spherical symmetry the basic variables are χ grr gθθ Arr KΓr α βr Br All are functions of t and r and satisfy the following spherically symmetric(generalized Lagrangian-form) BSSN system1

parttα = βrαprime minus 2αK minus (parttα)0 (4a)

parttβr = βrβrprime +

3

4Br minus (parttβ

r)0 (4b)

parttBr = βrBr prime + λ(parttΓ

r minus βrΓr prime)minus ηBr minus (parttBr)0 (4c)

parttχ = βrχprime +2

3Kαχminus βrgprimerrχ

3grrminus 2βrgprimeθθχ

3gθθminus 2

3βr primeχ (4d)

parttgrr =2

3βrgprimerr +

4

3grrβ

r prime minus 2Arrαminus 2grrβrgprimeθθ

3gθθ(4e)

parttgθθ =1

3βrgprimeθθ +

Arrgθθα

grrminus gθθβ

rgprimerr3grr

minus 2

3gθθβ

r prime (4f)

parttArr = βrAprimerr +

4

3Arrβ

rprime minus βrgprimerrArr

3grrminus 2βrgprimeθθArr

3gθθ+

2αχ(gprimerr)2

3g2rrminus αχ(gprimeθθ)

2

3g2θθminus α(χprime)2

6χ

+2

3grrαχΓ

r prime minus αχgprimerrgprimeθθ

2grrgθθ+χgprimerrα

prime

3grr+χgprimeθθα

prime

3gθθminus αgprimerrχ

prime

6grrminus αgprimeθθχ

prime

6gθθminus 2

3αprimeχprime +

αχprimeprime

3

minus 2

3χαprimeprime minus αχgprimeprimerr

3grr+αχgprimeprimeθθ3gθθ

minus 2αA2rr

grr+KαArr minus

2grrαχ

3gθθ(4g)

parttK = βrK prime +χgprimerrα

prime

2g2rrminus χgprimeθθα

prime

grrgθθ+αprimeχprime

2grrminus χαprimeprime

grr+

3αA2rr

2g2rr+

1

3αK2 (4h)

parttΓr = βrΓr prime +

Arrαgprimeθθ

g2rrgθθ+

2βrprimegprimeθθ3grrgθθ

+Arrαg

primerr

g3rrminus 4αK prime

3grrminus 2Arrα

prime

g2rrminus 3Arrαχ

prime

g2rrχ

+4βr primeprime

3grrminus βr(gprimeθθ)

2

grr(gθθ)2+

βrgprimeprimerr6(grr)2

+βrgprimeprimeθθ3gθθgrr

(4i)

where the prime stands for partial r-differentiation Eqs (4d-i) are Brownrsquos Eqs (9a-f)listed in [60] subject to his Lagrangian condition (corresponding to v = 1 in Brownrsquos equa-tions) The first three equations (4a-c) comprise the gauge sector and these are essentially

1 For this system the determinant g = grr(gθθ)2 sin4 θ is not unity

5

spherically symmetric versions of the standard ldquo1+logrdquo and ldquoΓ-driverrdquo conditions listed inEqs (1) and (2) of [60] However we have introduced the following minor modificationsFirst (parttα)0 designates a constant term which ensures that the right-hand side of the αevolution equation (4a) vanishes at the initial time This source term as well as the analo-gous terms appearing in the evolution equations (4bc) for βr and Br are needed to enablea static evolution of the Schwarzschild solution in Kerr-Schild coordinates Second the pa-rameter λ (perhaps with functional dependence) modifies the hyperbolicity of the first-ordersystem [80] The damping parameter η typically appears in standard versions of these gaugeevolution equations (See Sections IIC and IVB for further discussions) For this BSSNsystem we have three constraints the Hamiltonian constraint H the momentum constraintMr and the constraint Gr resulting from the definition of the conformal connection Γr Inspherical symmetry these constraints are written as follows

H = minus3A2rr

2g2rr+

2K2

3minus 5(χprime)2

2χgrr+

2χprimeprime

grr+

2χ

gθθminus 2χgprimeprimeθθgrrgθθ

+2χprimegprimeθθgrrgθθ

+χgprimerrg

primeθθ

g2rrgθθminus χprimegprimerr

g2rr+χ(gprimeθθ)

2

2grrg2θθ

(5a)

Mr =Aprime

rr

grrminus 2K prime

3minus 3Arrχ

prime

2χgrr+

3Arrgprimeθθ

2grrgθθminus Arrg

primerr

g2rr(5b)

Gr = minus gprimerr2g2rr

+gprimeθθgrrgθθ

+ Γr (5c)

These expressions are the ones listed by Brown in [60] Eqs (4ef) also ensure that thedeterminant factor g sin4 θ = grr(gθθ)

2 remains fixed throughout an evolution

B Abstract expressions of the system

We define the following vectors built with system variables

u =

χgrrgθθαβr

v =

Br

Arr

KΓr

Q =

χprime

gprimerrgprimeθθαprime

βrprime

(6)

Introduction of Qmight seem unnecessary at this stage but proves useful in the constructionof our discontinuous Galerkin scheme In terms of the vectors u v and Q we further define

Wuv =

(

uv

)

WvQ =

(

vQ

)

W =WuQ =

uvQ

(7)

Here we have introduced ldquocolon notationrdquo [81] to represent (sub)vectors and (sub)matricesalthough we employ the notation over block rather than individual elements In the first-order version of the system (4) the components of Q are promoted to independent fields in

6

which case the corresponding principal part features

parttBr = βrBr prime minus 4λα

3grrK prime +

4λ

3grrQprime

βr +λβr

6(grr)2Qprime

grr +λβr

3gθθgrrQprime

gθθ(8a)

parttArr = βrAprimerr +

2

3grrαχΓ

r prime +1

3αQprime

χ minus2

3χQprime

α minus αχ

3grrQprime

grr +αχ

3gθθQprime

gθθ(8b)

parttK = βrK prime minus χ

grrQprime

α (8c)

parttΓr = βrΓr prime minus 4αK prime

3grr+

4

3grrQprime

βr +βr

6(grr)2Qprime

grr +βr

3gθθgrrQprime

gθθ(8d)

parttQχ = βrQprimeχ +

2

3αχK prime minus βrχ

3grrQprime

grr minus2βrχ

3gθθQprime

gθθminus 2

3χQprime

βr (8e)

parttQgrr =2

3βrQprime

grr +4

3grrQ

primeβr minus 2αAprime

rr minus2grrβ

r

3gθθQprime

gθθ(8f)

parttQgθθ =1

3βrQprime

gθθ+gθθα

grrAprime

rr minusgθθβ

r

3grrQprime

grr minus2

3gθθQ

primeβr (8g)

parttQα = βrQprimeα minus 2αK prime (8h)

parttQβr = βrQprimeβr +

3

4Br prime (8i)

where all lower-order terms on the right-hand side have been dropped This sector of prin-cipal parts of the first-order system has the form

parttWvQ + A(u)W primevQ = 0 (9)

where (minus) the explicit form of the 9-by-9 matrix A(u) is given below in (A1) Thefirst-order version of (4) takes the nonconservative form

parttW +A(u)W prime = S(W ) A(u) =

(

05times5 05times9

09times5 A(u)

)

(10)

where S(W ) is a vector of lower order terms built with all components of W Partition ofA(u) = A(u)vQvQ into blocks corresponding to the v and Q sectors yields

A(u) =

(

A(u)vv A(u)vQ

A(u)Qv A(u)QQ

)

(11)

Using these blocks we then define the 9-by-9 matrix

A(u) = A(u)uvvQ =

(

05times4 05times5

A(u)vv A(u)vQ

)

(12)

and express (4) as

parttWuv + A(u)W primevQ = S(W ) (13a)

Q = uprime (13b)

where S(W ) = S(W )uv

7

field speed

X1 micro1 = 0

X23 micro23 = minusβr

Xplusmn4 microplusmn

4 = minusβr plusmnradic

2αχgrr

Xplusmn5 microplusmn

5 = minusβr plusmn αradic

χgrr

Xplusmn6 microplusmn

6 = minusβr plusmnradic

λgrr

TABLE I Characteristic speeds These speeds are the eigenvalues listed in (A2)

C Hyperbolicity and characteristic fields

Although our numerical scheme deals directly with the second-order spatial operatorsappearing in (4) we first consider the hyperbolicity of the corresponding first-order system(10) The characteristic fields and their speeds are found by instantaneously ldquofreezingrdquothe fields u in A(u) to some value u0 corresponding to a linearization around a uniformstate Below we continue to write u for simplicity with the understanding that u is reallythe background solution u0 Of primary interest is the range of u0 for which the system isstrongly hyperbolic [73ndash76]

Appendix A shows that the characteristic fields corresponding to (4) are as follows (i)all components of u (each with speed 0) and (ii) the fields

X1 = gθθQgrr + 2grrQgθθ (14a)

X2 = grrΓr +

2

χQχ minus

1

2grrQgrr minus

1

gθθQgθθ (14b)

X3 =grrλBr +

2

χQχ minus

1

2grrQgrr minus

1

gθθQgθθ (14c)

Xplusmn4 = plusmn

radic

2αgrrχ

K +Qα (14d)

Xplusmn5 = ∓ 3radic

grrχArr plusmn 2

radic

grrχK + 2grrΓ

r +1

χQχ minus

1

grrQgrr +

1

gθθQgθθ (14e)

Xplusmn6 = minus3

4

grrλBr plusmn α

radicλgrr

(2αχminus λ)K minus βr

8(βrgrr ∓radicλgrr)

Qgrr

minus βrgrr

4gθθ(βrgrr ∓radicλgrr)

Qgθθ +αχ

(2αχminus λ)Qα plusmn

radic

grrλQβr (14f)

with the speeds listed in Table I To ensure strong hyperbolicity we must necessarily require

λ gt 0 (βr)2grr minus λ 6= 0 2αχminus λ 6= 0 (15)

as shown in in Appendix A where further conditions are also given When λ = 1 thehyperbolicity condition of Ref [60] is recovered In fact the system could be recast assymmetric hyperbolic Indeed as it involves one spatial dimension the relevant symmetrizercan be constructed via polar decomposition of the diagonalizing similiarity transformationHowever we will not exploit this possibility

8

This system admits an inner excision boundary provided

βr ge max

(

radic

2αχ

grr

radic

α2χ

grr

radic

λ

grr

)

(16)

holds at the inner boundary This condition ensures each characteristic field has a nonposi-tive speed at the inner boundary and therefore the inner boundary is an excision boundaryat which no boundary conditions are needed The extra flexibility afforded by the param-eter λ could be used to maintain rigorous hyperbolicity by moving the points at whichthe conditions in (15) are violated outside of the computational domain Furthermore forλ = 1 Eq (16) conceivably fails or is only satisfied close to r = 0 where field gradients are

prohibitively large The troublesome X+6 gauge mode has a positive speed minusβr +

radic

λgrrIndeed for the conformally flat Kerr-Schild system considered in section IVB an inner ex-cision boundary is only possible provided λ is small enough

The transformation (14) can be inverted in order to express the fundamental fields interms of the characteristic fields

Br = minus1

6

λ

grrgθθ

[

(βr)2

(βr)2grr minus λ

]

X1 +2

3

λαχ

grr(2αχminus λ)(X+

4 +Xminus4 )minus

2

3

λ

grr(X+

6 +Xminus6 )

(17a)

Arr =1

3

radic

grrχ

2α(X+

4 minusXminus4 )minus

radicgrrχ

6(X+

5 minusXminus5 ) (17b)

K =

radic

χ

8αgrr(X+

4 minusXminus4 ) (17c)

Γr = minus1

6

1

grrgθθ

[

(βr)2

(βr)2grr minus λ

]

X1 +1

grr(X2 minusX3) +

2

3

αχ

grr(2αχminus λ)(X+

4 +Xminus4 )

minus 2

3

1

grr(X+

6 +Xminus6 ) (17d)

Qχ =1

12

χ

grrgθθ

[

4(βr)2grr minus 3λ

(βr)2grr minus λ

]

X1 +χ

2X3 minus

1

3

αχ2

(2αχminus λ)(X+

4 +Xminus4 ) (17e)

+χ

3(X+

6 +Xminus6 ) (17f)

Qgrr =2(βr)2grr minus 3λ

6gθθ((βr)2grr minus λ)X1 +

4

3grrX2 minus grrX3 +

2

3

αχgrr(2αχminus λ)

(X+4 +Xminus

4 )

minus 1

3grr(X

+5 +Xminus

5 )minus2

3grr(X

+6 +Xminus

6 ) (17g)

Qgθθ =

[

1

4grr+

(βr)2

12((βr)2grr minus λ)

]

X1 minus2

3gθθX2 +

1

2gθθX3 minus

1

3

αχgθθ(2αχminus λ)

(X+4 +Xminus

4 )

+1

6gθθ(X

+5 +Xminus

5 ) +1

3gθθ(X

+6 +Xminus

6 ) (17h)

Qα =1

2(X+

4 +Xminus4 ) (17i)

Qβr =βrλ

8grrgθθ((βr)2grr minus λ)X1 minus

λ

(2αχminus λ)

radic

αχ

8grr(X+

4 minusXminus4 ) +

1

2

radic

λ

grr(X+

6 minusXminus6 )

(17j)

9

We will refer to this inverse transformation when discussing outer boundary conditions forour numerical simulations in Sec IVB

III DISCONTINUOUS GALERKIN METHOD

This section describes the nodal discontinuous Galerkin method used to numerically solve(4) We adopt a method-of-lines strategy and here describe the relevant semi-discrete schemewhile leaving the temporal dimension continuous To approximate (4) we follow the generalprocedure first introduced in Ref [82] Our approach defines local auxiliary variables Q = uprimeand rewrites the spatially second-order system (4) as the first-order system (13a) Once weuse (13b) to eliminate Q from (13a) we recover the primal equations (4) The auxiliaryvariable approach was later generalized and coined the local discontinuous Galkerin (LDG)method in Ref [77] We may refer to our particular scheme as an LDG method but notethat many variations exist in the literature We stress that in LDG methods Q is not evolvedand is introduced primarily to assist in the construction of a stable scheme

Equations (12) and (13a) imply that the physical flux function is

F (W ) =

(

Fu(W )

Fv(W )

)

equiv A(u)WvQ =

(

05times1

f(W )

)

f =

fBr

fArr

fKfΓ

(18)

Only the evolution equations for Br Arr K and Γr give rise to non-zero components in F and we have collected these non-zero components into a smaller vector f = Fv Inspectionof (8) determines these components For example from (8c) we find

fK = minusβrK +χ

grrQα (19)

A Local approximation of the system (13)

Our treatment closely follows [83] but with the equations and notations relevant forthis paper Our computational domain Ω is the closed r-interval [a b] We cover Ω withkmax gt 1 non-overlapping intervals Dk = [ak bk] where a = a1 b = bkmax and bkminus1 = ak fork = 2 middot middot middot kmax

On each interval Dk we approximate each component of the system vector W by a localinterpolating polynomial of degree N For example

χkh(t r) =

Nsum

j=0

χ(t rkj )ℓkj (r) (20)

approximates χ(t r) Throughout this section approximations are denoted by a subscripth (see [66] for the notation) For example Wh and fh are approximations of W and f Although Q = uprime Qh and uprimeh are not necessarily the same In (20) ℓkj (r) is the jth Lagrange

polynomial belonging to Dk

ℓkj (r) =Nprod

i=0i 6=j

r minus rkirkj minus rki

(21)

10

Evidently the polynomial χkh interpolates χ at rkj To define the nodes rkj consider the

mapping from the unit interval [minus1 1] to Dk

rk(u) = ak + 12(1 + u)(bk minus ak) (22)

and the N+1 Legendre-Gauss-Lobatto (LGL) nodes uj The uj are the roots of the equation

(1minus u2)P primeN(u) = 0 (23)

where PN(u) is the Nth degree Legendre polynomial and the physical nodes are simplyrkj = rk(uj) In vector notation the approximation (20) takes the form

χkh(t r) = χk(t)Tℓk(r) (24)

in terms of the column vectors

χk(t) =[

χ(t rk0) middot middot middot χ(t rkN)]T ℓk(r) =

[

ℓk0(r) middot middot middot ℓkN(r)]T (25)

On each open interval (ak bk) sub Dk and for each component of the equations in (13) we

define local residuals measuring the extent to which our approximations satisfy the originalcontinuum system Dropping the subdomain label k on the polynomials and focusing onthe K equation as a representative example the local residual corresponding to (4h) is

minus(RK)kh equivminus parttKh + (βrK prime)h minus

(

χQprimeα

grr

)

h

+

(

χQgrrQα

2g2rr

)

h

minus(

χQgθθQα

grrgθθ

)

h

+

(

QαQχ

2grr

)

h

+

(

3αA2rr

2g2rr

)

h

+

(

1

3αK2

)

h

(26)

Here for example the expressions 2

(βrK prime)h = βrhK

primeh

(

QαQχ

2grr

)

h

=QαhQχh

2grrh (27)

We similarly construct the remaining eight residuals eg (Rgrr)h and (RΓr)h as well as fiveresiduals corresponding to (13b) For example one of these remaining five is

(RQα)kh equiv minusQαh + αprime

h (28)

Let the kth inner product be defined as

(

u v)

Dk equivint bk

akdru(r)v(r) (29)

and consider the expression (ℓkj (RK)kh)Dk We call the requirement that this inner product

vanish forallj the kth Galerkin condition For each component of the system and for each k

2 At this stage the first expression is generically a polynomial of degree 2N minus 1 and the latter is not a

polynomial The conventions adopted in Eq (27) prove useful while working with the residual However

later on in Sec III C to obtain the final form (47) of the numerical approximation corresponding to (26)

we will replace nonlinear terms with degree-N polynomials

11

there is a corresponding Galerkin condition in total 9kmax(N + 1) equations for (13a) and5kmax(N+1) for (13b) Enforcement of the Galerkin conditions on each D

k will not recover ameaningful global solution since they provide no mechanism for coupling the local solutionson the different intervals Borrowing from the finite volume toolbox we achieve couplingthrough integration by parts on r and introduction of the numerical flux f lowast at the interfacebetween subdomains

In (26) we only need to consider (βrK prime)h and (χQprimeαgrr)h as the other terms comprise a

component of the source vector Sh Using integration by parts we write

(

ℓkj (βrK prime)h

)

Dk =minusint bk

akdr[

(

ℓkjβrh

)primeKh

]

+ (βrhKh) ℓ

kj

∣

∣

∣

bk

ak (30a)

(

ℓkj (χQprimeαgrr)h

)

Dk =minusint bk

akdr

[(

ℓkjχh

grrh

)prime

Qαh

]

+

(

χhQαh

grrh

)

ℓkj

∣

∣

∣

bk

ak (30b)

In these formulas we have retained the domain index k on ℓkj while continuing to suppress iton Kh grrh etc Moreover we have suppressed the r-dependence in all terms on the right-hand side Addition of these formulas along with the definition fKh = minus(βrK)h+(χQαgrr)hgives

(

ℓkj (βrK prime)h minus (χQprime

αgrr)h)

Dk = minusint bk

akdr

[

(

ℓkjβrh

)primeKh minus

(

ℓkjχh

grrh

)primeQαh

]

minus fKhℓkj

∣

∣

∣

bk

ak (31)

In lieu of (31) we will instead work with the replacement

(

ℓkj (βrK prime)h minus (χQprime

αgrr)h)

Dk rarr minusint bk

akdr

[

(

ℓkjβrh

)primeKh minus

(

ℓkjχh

grrh

)primeQαh

]

minus f lowastKℓ

kj

∣

∣

∣

bk

ak (32)

This replacement features a component f lowastK of the numerical flux rather than a component

fKh of the boundary flux The numerical flux is determined by (as yet not chosen) functions3

f lowast = f lowast(W+Wminus) (33)

where for example Wminus is an interior boundary value [either W kh (t a

k) or W kh (t b

k)] of theapproximation defined on D

k andW+ is an exterior boundary value [either W kminus1h (t bkminus1) or

W k+1h (t ak+1)] of the approximation defined on either Dkminus1 or Dk+1 We discuss our choice

of numerical flux in the next subsection We now employ additional integration by parts towrite the above replacement as

(

ℓkj (βrK prime)h minus (χQprime

αgrr)h)

Dk rarrint bk

akdrℓkj

(

βrK prime minus χQprimeα

grr

)

h

+ (fKh minus f lowastK) ℓ

kj

∣

∣

∣

bk

ak (34)

Rather than the exact kth Galerkin condition(

ℓkj (RK)kh

)

Dk = 0 forallj for the K component

of (13) on Dk we will instead strive to enforce

(

ℓkj (RK)kh

)

Dk = (fKh minus f lowastK) ℓ

kj

∣

∣

∣

bk

ak forallj (35)

3 In the context of the dG method here + and minus denote ldquoexteriorrdquo and ldquointeriorrdquo and have no relation to

the plusmn using to denote the characteristic fields and speeds in Table I For characteristic fields and speeds

+ and minus mean ldquoright-movingrdquo and ldquoleft-movingrdquo

12

although our treatment of nonlinear terms will lead to a slight modification of these equations(we return to this issue shortly) The other components of (13a) are treated similarly asare the components of (13b) Recall that for example Qα = αprime Formally using the samedG method to solve for Qα we arrive at the replacement

(

ℓkj (RQα)kh

)

Dk rarrint bk

akdrℓkj (minusQαh + αprime

h)minus (αh minus αlowast) ℓkj

∣

∣

∣

bk

ak (36)

which again features a component αlowast of the numerical flux The auxiliary variables areconstructed and used at each stage of temporal integration We then have

(

ℓkj (RQα)kh

)

Dk = (αh minus αlowast) ℓkj

∣

∣

∣

bk

ak forallj (37)

as the corresponding enforced kth Galerkin condition

B Numerical Flux

To further complete our dG scheme we must specify functional forms for the compo-nents of the numerical flux introduced in the previous section We distinguish between thephysical fluxes (components of f) and the auxiliary fluxes (components of u) arising fromthe definition of the auxiliary variables These choices are not independent as the resultingscheme must be stable and consistent Our choice follows [84] which considered diffusionproblems Additional analysis of this flux choice appears in [66 85]

Let us first consider the numerical fluxes corresponding to the physical fluxes and of theform (33) The numerical flux vector is a function of the system and auxiliary variablesinterior and exterior to a subdomain A common choice for f lowast is

f lowast = fh+τ

2

[[

vh]]

K-component of f lowast f lowastK = fKh+

τ

2

[[

Kh

]]

(38)

where as an example we have also shown the component of f lowast corresponding to the analysisabove Respectively the average and jump across the interface are

fh =1

2

(

f+ + fminus) [[

vh]]

= nminusvminus + n+v+ (39)

Here τ is a position dependent penalty parameter (fixed below) and nminus(n+) is the localoutward pointing normal to the interior (exterior) subdomain The role of τ is to ldquopenalizerdquo(i e yield a negative contribution to the L2 energy norm) jumps across an interface Anappropriate choice of τ will ensure stability and we now provide some motivation for thechoice (41) of τ we make below

Were we treating the fully first-order system (10) the local Lax-Friedrichs flux wouldoften be a preferred choice due to its simplicity [66] In this case the constant ω in thenumerical flux formula Flowast = Fh + 1

2ω[[

Wh

]]

obeys ω ge max∣

∣micro(nablaWF(W ))∣

∣ HereF(W ) = A(u)W the notation micro(middot) indicates the spectral radius of the matrix within andthe max is taken over interior Wminus and exterior W+ states Motivated by (9) we adopt asimilar but simpler prescription substituting the field gradient

nablaWvQA(u)WvQ = A(u) (40)

13

for nablaWF(W ) Precisely we assume the scaling

τ(bk) = τ(ak+1) = τk+12 equiv C middotmax∣

∣micro(

A(u))∣

∣ (41)

where C = O(1) is a constant chosen for stability Larger values of C will result in schemeswith better stability properties whereas too large a value will impact the CFL conditionAt the interface point Ik+12 equiv D

k cap Dk+1 the vector uh has two representations uminus at bk

and u+ at ak+1 The max in (41) is taken over the corresponding two sets of field speedsMore precisely the speeds in Table I are computed for both uminus and u+ and the maximumtaken over all resulting speeds For the auxiliary variables a penalized central flux is usedThe definition with one representative component is

ulowast = uh minus1

2

[[

uh]]

α-component of ulowast αlowast = αh minus1

2

[[

αh

]]

(42)

with similar expressions for the remaining componentsWe stress the following point Since the interior coupling between subdomains is achieved

through the numerical flux forms (41) and (42) the inverse transformation (17) expressingthe fundamental fields in terms of the characteristic fields is not required to achieve thiscoupling On the other hand imposition of physical boundary conditions may still rely on(17) since this transformation allows one to fix only incoming characteristic modes

C Matrix form of the semi-discrete equations

Let us introduce the kth mass and stiffness matrices

Mkij =

int bk

akdrℓki (r)ℓ

kj (r) Sk

ij =

int bk

akdrℓki (r)ℓ

kjprime(r) (43)

These matrices belong to Dk and the corresponding matrices defined on the reference interval[minus1 1] are

Mij =

int 1

minus1

duℓi(u)ℓj(u) Sij =

int 1

minus1

duℓi(u)ℓprimej(u) (44)

where ℓj(u) is the jth Lagrange polynomial determined by the LGL nodes uj on [minus1 1]These matrices are related by Mk

ij =12(bk minus ak)Mij and S

kij = Sij whence only the reference

matrices require computation and storageWe will use the matrices Mk and Sk in obtaining an ODE system from (26) and (35)

Towards this end we first approximate the nonlinear terms (products and quotients) in(26) by degree-N interpolating polynomials Such approximations are achieved throughpointwise representations For example (QαQχgrr)h appears in (26) and is expressed inthe following way [cf footnote 2]

(

QαQχ

grr

)

h

(t r) =

(

QαhQχh

grrh

)

(t r) rarrNsum

j=0

Qαh(t rkj )Qχh(t r

kj )

grrh(t rkj )

ℓkj (r) (45)

Note that the expressions on the right and left are not equivalent due to aliasing error [65]Our vector notation for this replacement will be

(

QαQχ

grr

)

h

(t r) rarr(

QαQχ

grr

)

(t)Tℓk(r) (46)

14

Operations among bold variables are always performed pointwise Making similar replace-ments for all terms in (26) and then carrying out the integrations in (35) which bring inMk and Sk we arrive at

parttK = βrDK minus χDQα

grr

+1

2

χQgrrQα

g2rr

minusχQgθθ

Qα

grrgθθ

+1

2

QαQχ

grr

+3

2

αA2rr

g2rr

+1

3αK2 +Mminus1ℓk (fKh minus f lowast

K)∣

∣

∣

bk

ak (47)

where we have again suppressed the superscript k on all terms except ℓk(r) and the sub-script h is dropped on all boldfaced variables As described in [66] the spectral collocationderivative matrix

(Dk)ij =dℓkjdr

∣

∣

∣

∣

∣

r=rki

(48)

can also be expressed as Dk = (Mk)minus1Sk which appears above Eight other semi-discreteevolution equations are similarly obtained with nine in total (one for each component ofWuv) Additionally we have

Qα = Dα+Mminus1ℓk(αlowast minusαh)∣

∣

∣

bk

ak (49)

as one of the auxiliary equations with five in total (one for each component of Q = WQQ)

D Filtering

Like other nodal (pseudospectral) methods our scheme may suffer from instabilitiesdriven by aliasing error [65] Filtering is a simple yet robust remedy To filter a solu-tion component such as χ we use the modal (as opposed to nodal) representation of thesolution

χkh(t r) =

Nsum

j=0

χ(t rkj )ℓkj (r) =

Nsum

j=0

χkj (t)Pj(r) (50)

where Pj(r) is the jth Legendre polynomial Let ηj = jN and define the filter function

σ(ηj) =

1 for 0 le ηj le NcN

exp(

minus ǫ(

ηjminusNcN

1minusNcN

)2s)

for NcN le ηj le 1(51)

At each timestep we modify our solution component according to

χkh rarr

(

χkh

)filtered=

Nsum

j=0

σ(ηj)χkj (t)Pj(r) (52)

Evidently the modification only affects the top N minus Nc modes and is sufficient to controlthe type of weak instability driven by aliasing [66] The numerical parameters Nc and ǫ areproblem dependent For our simulations we have taken ǫ ≃ minuslog(εmach) = 36 where εmach

is machine accuracy in double precision

15

E Model system

To better illustrate the basic properties of our method we consider a toy model Namelythe following spatially second-order system

parttu = uprime + av minus u3 + g(t x) (53a)

parttv = uprimeprime + vprime minus (u+ v)(uprime)2 + v2u2 + h(t x) (53b)

where a ge 1 is constant and g and h are analytic source terms to be specified In contrastto (6) here u v and Q = uprime are scalars rather than vectors System (53) admits a first-order reduction in which uprime is defined as an extra variable Since this first-order reductionis strongly hyperbolic the spatially second-order system (53) is also strongly hyperbolic byone of the definitions considered in [75] The characteristic fields Xplusmn and speeds microplusmn are

X+ =radicav minus uprime micro+ =

radicaminus 1 Xminus =

radicav + uprime microminus = minus(

radica+ 1) (54)

To construct a local dG scheme for this system we first rewrite it as

parttu = Q+ av minus u3 + g(t x) (55a)

parttv = Qprime + vprime minus (u+ v)Q2 + v2u2 + h(t x) (55b)

Q = uprime (55c)

Evidently f = minus(Q + v) is the v-component of the physical flux vector

F (vQ) equiv(

Fu

Fv

)

=

(

0

f

)

(56)

Note that F has the same structure as (u v)T Borrowing from the presentation for theBSSN system we write the analogous semidiscrete scheme on each subdomain D

k for themodel system

parttu = Q+ av minus u3 + g(t) (57a)

parttv = DQ+Dv minus (u+ v)Q2 + v2u2 + h(t) +Mminus1ℓk(fh minus f lowast)∣

∣

∣

bk

ak (57b)

Q = Du+Mminus1ℓk(ulowast minus uh)∣

∣

∣

bk

ak (57c)

Here we have suppressed the subinterval label k from all variables except for the vectorℓk of Lagrange polynomial values Moreover following the guidelines discussed above thenumerical fluxes are given by

f lowast = fh+1 +

radica

2

[[

vh]]

ulowast = uh minus1

2

[[

uh]]

(58)

Appendix C analyzes the stability of our scheme for a more general numerical flux choiceas applied to (53) with the nonlinear and source terms dropped

IV RESULTS FROM NUMERICAL SIMULATIONS

This section presents results found by numerically solving both the model system (53)and BSSN system (4) with the dG scheme presented in Sec III

16

10minus10

10minus6

10minus2

∆uinfin

10minus10

10minus6

10minus2

∆vinfin

0 500 1000 1500 2000 2500 3000 3500 400010

minus10

10minus6

10minus2

∆(partxu)infin

t

N = 3N = 6N = 9N = 12

FIG 1 Spectral convergence of fields for model PDE Respectively for N = 3 6 9 12

a timestep of ∆t = 00578 00178 00084 00049 has been chosen for stability and accuracy In the

title headings for example ∆u equiv unumer minus uexact

A Simulations of the model system

The semi-discrete scheme (57) has been integrated with the classical fourth-order Runge-Kutta method When integrating this system we have first constructed Q at each Runge-Kutta stage and then substituted into the evolution equations (57ab) for u and v Theproblem has been solved on a computational domain [0 4π] comprised of two subdomainswith a timestep chosen small enough for stability The initial data has been taken from thefollowing exact solution to (53)

uprimeexact(t x) =1

2

[

sin(xminus microminust)minus sin(xminus micro+t)]

(59a)

vexact(t x) =1

2radica

[

sin(xminus microminust) + sin(xminus micro+t)]

(59b)

g(t x) = u3exact (59c)

h(t x) = (uexact + vexact)(uprimeexact)

2 minus v2exactu2exact (59d)

where the speeds microplusmn are found in (54) Specification of the boundary condition at a physicalendpoint amounts to choosing the external state for at the endpoint We have considered

17

8 16 32 64 128

10minus3

10minus2

10minus1

N

∆t

max stable ∆t8N2

FIG 2 Scaling of maximum stable ∆t with N for model PDE

two possibilities (i) the analytic state (Q+ v+) = (Qexact vexact) and (ii) an upwind stateFor example at x = 4π the upwind state is4

Q+ = Qupwind =1

2

[

(Xminus)exact minus (X+)numer

]

v+ = vupwind =1

2radica

[

(Xminus)exact + (X+)numer

]

(60)Either choice of (Q+ v+) leads to similar results and the plots here correspond to theanalytic state Figure 1 clearly shows spectral convergence with increasing polynomial orderN across all fields for the case a = 2 Other values of a including a = 1 for which X+

is a static characteristic field have also been considered with similar results Appendix Cdemonstrates that our proposed scheme for the system (57) with nonlinear and source termsdropped is stable in a semi-discrete sense Nevertheless the fully discrete scheme obtainedvia temporal discretization by the fourth-order Runge-Kutta method is still subject to thestandard absolute stability requirement Namely if microh is any eigenvalue corresponding tothe (linearized) discrete spatial operator then a necessary condition for stability is thatmicroh∆t lies in absolute stability region for Runge-Kutta 4 We here show empirically that theassociated timestep restriction scales like Nminus2 ie ∆t = O(Nminus2) for stability We note thatsuch scaling is welcome in light of the second-order spatial operators which appear in thesystem and suggest a possible worse scaling like Nminus4 Fig 2 plots the maximum stabletimestep for a range of N demonstrating the Nminus2 scaling in line with behavior known fromanalysis of first-order systems [66] This scaling also holds for the BSSN system

4 We remind the reader that unfortunately the plusmn on Xplusmn means something different than the plusmn indicating

exteriorinterior dG states [cf footnote 3]

18

B Simulations of the BSSN system

This subsection documents results for simulations of the unit-mass-parameter (M = 1)Schwarzschild solution (B9) expressed in terms of ingoing Kerr-Schild coordinates Sincethe solution is stationary temporal integration of the semi-discrete scheme has been carriedout with the forward Euler method The r-coordinate domain [04 34] has been split into3 equally spaced subdomains and we have set η = 10 λ = 01 and C = 2 [cf Eq (41)]For all simulations ∆t has been chosen for stability With the chosen λ the inner physicalboundary rmin = 04 is an excision surface At each timestep we have applied an (order2s = 20) exponential filter on the top two-thirds of the modal coefficient set for all fieldsexcept for grr and gθθ For stability we have empirically observed that grr and gθθ must not

be filtered A detailed understanding of this is still lackingIssues related to physical boundary conditions are similar to the one encountered in

Sec IVA for the model problem Similar to before we have retained Eqs (3842) as thechoice of numerical flux even at the endpoints Therefore at an endpoint the specificationof the boundary condition amounts to the choice W+ of external state We have typicallychosen the inner boundary of the radial domain as an excision boundary and in this caseW+ = Wminus is enforced at the inner physical boundary At the outer physical boundaryfor W+ we have again considered two choices (i) Wexact and (ii) Wupwind To enforcechoice (ii) the inverse transformation (17) must be used with incoming characteristic fieldsfixed to their exact values similar to (60) We have tried various versions of choice (ii)and in all cases the resulting simulations have been unstable We therefore present resultscorresponding to choice (i) Although the choice of an analytical external state Wexact

at the outer boundary is stable for our problem such a boundary condition is unlikelyto generalize to more complicated scenarios involving dynamical fields Indeed the issueof outer boundary conditions for the BSSN system is an active area of research with aproper treatment requiring fixation of incoming radiation control of the constraints andspecification of gauge (see Ref [86] for a recent analysis)

For BSSN simulations our main diagnostic is to monitor the Hamiltonian momentumand conformal connection constraints Figure 3 depicts long-time histories of constraintviolations whereas Figs 4 and 5 depict long-time error histories for the individual BSSNfield components From the middle plot in Fig 5 we infer that up to the indicated numericalerror the factor g sin4 θ = grr(gθθ)

2 remains at its initial fixed profile r4 throughout theevolution These figures indicate that the proposed scheme is stable for long times andexhibits spectral converge with increased polynomial order N Similar results are recoveredfromM = 0 Minkowski initial data The stability documented in these plots does not appearto rely on inordinate parameter tuning For example with the fixed parameters describedabove we obtain similar plots if we individually vary (i) rmin over 0325 035 04 0475(values still corresponding to an excision surface for the given choice of λ) (ii) η over1 3 7 10 (iii) s over 8 9 10 With the polynomial order N ranging over 23 26 29 31both stability and qualitatively similar exponential convergence is achieved with a singlesubdomain Likewise adoption of a larger coordinate domain with more subdomains doesnot impact our results Finally we have considered the addition of random noise to all fieldcomponents at the initial time Precisely with the system component χ as an example wehave set

χ(0) 7rarr χ(0) + δχ(0) (61)

where each component (nodal value) of δχ(0) is 10minus5 times a random variable drawn from a

19

10minus9

10minus7

10minus5

10minus3

Hinfin

10minus9

10minus7

10minus5

10minus3

Mrinfin

0 2000 4000 6000 8000 10000

10minus9

10minus7

10minus5

10minus3

Grinfin

t

N = 11N = 14N = 17N = 20

FIG 3 Spectral convergence of constraint violations for M = 1 Kerr-Schild ini-

tial data Respectively for N = 11 14 17 19 a timestep of ∆t ≃ 00041 00026 00018 00013

has been chosen for stability and accuracy

standard normal distribution Such perturbed initial data also gives rise to stable evolutions

V CONCLUSION

We have introduced a discontinuous Galerkin method for solving the spherically reducedBSSN system with second-order spatial operators Our scheme shares similarities withother discontinuous Galerkin methods that use local auxiliary variables to handle high-orderspatial derivatives [66 77ndash79 82 84 85] and which have typically been applied to eitherelliptic parabolic or mixed type problems The key ingredient of a stable dG scheme is anappropriate choice of numerical flux and our particular choice has been motivated by theanalysis presented in Appendix C When used to evolve the Schwarzschild solution in Kerr-Schild coordinates our numerical implementation of the BSSN system (4) is robustly stableand converges to the analytic solution exponentially with increased polynomial order Byapproximating the spatially second-order form of the BSSN system we have not introducedextra fields which are evolved Evolved auxiliary fields result in new constraints which mayspoil stability Our main goal has been stable evolution of the spherically reduced BSSNsystem as a first step towards understanding how a discontinuous Galerkin method might be

20

10minus9

10minus7

10minus5

∆Arrinfin

10minus9

10minus7

10minus5

∆Kinfin

0 2000 4000 6000 8000 10000

10minus9

10minus7

10minus5

max(∆Brinfin ∆Γrinfin)

t

N = 11N = 14N = 17N = 20

FIG 4 Spectral convergence of solution for M = 1 Kerr-Schild initial data

Timestep choices are described in the caption for Fig 3 In the title headings for example

∆Arr equiv (Arr)numer minus (Arr)exact

applied to the full BSSN system Towards that goal we now discuss treatment of singularitiesand generalization of the described dG method to higher space dimension

To deal with the fixed Schwarzschild singularity we have used excision which is easy inthe context of the spherically reduced BSSN system However excision for the binary blackhole problem in full general relativity requires attention to the technical challenge of hori-zon tracking State-of-the-art BSSN codes avoid such complication relying instead on themoving-puncture technique While the moving-puncture technique does involve mild centralsingularities it may still prove amenable to spectral methods Indeed spectral methods fornon-smooth problems is well-developed in both theory and for complex applications Sincethe moving-puncture technique can be performed in spherical symmetry [60] a first-steptoward a spectral moving-puncture code would be to implement a moving puncture withthe nodal dG method described here Such an implementation may adopt Legendre-Gauss-Radau nodes on the innermost subdomain thereby ensuring that the physical singularitydoes not lie on a nodal point (in much the same way finite difference codes use a staggeredgrid) Beyond traditional excision and moving punctures one might also retain a smoothproblem for which spectral methods are especially well-suited via the turducken approach[87 88] to singularities

21

10minus11

10minus9

10minus7

10minus5

∆χinfin

10minus11

10minus9

10minus7

10minus5

max(∆grrinfin ∆gθθinfin)

0 2000 4000 6000 8000 10000

10minus11

10minus9

10minus7

10minus5

max(∆αinfin ∆βrinfin)

t

N = 11N = 14N = 17N = 20

FIG 5 Spectral convergence of solution violations for M = 1 Kerr-Schild initial

data See the caption of Fig 4 for details

Discontinuous Galerkin methods for hyperbolic problems in two and three space dimen-sions are well-developed A generalization of the method described here to three-dimensionsand the full BSSN system would likely rely on an unstructured mesh Appropriate local poly-nomial expansions for the subelements are well-understood as are choices for the numericalfluxes which would now live on two-dimensional faces rather than single points Whether ornot it would ultimately prove successful generalization of our dG method to a higher dimen-sion would rely on an established conceptual framework Further computational advancesof relevance to a generalization of our dG method to the full BSSN system (possibly includ-ing matter) may include mesh hp-adaptivity local timestepping shock capturing and slopelimiting techniques [66] Moreover recent work [89] indicates that enhanced performancewould be expected were our scheme implemented on graphics processor units

VI ACKNOWLEDGMENTS

We thank Nick Taylor for discussions about treating second-order operators with spectralmethods Benjamin Stamm for helping polish up a few parts of Appendix C Khosro Shah-bazi for discussions on LDG and IP methods We also acknowledge helpful conversationswith David Brown and Manuel Tiglio about the BSSN system and previous work on its nu-

22

merical implementation We gratefully acknowledge funding through grants DMS 0554377and DARPAAFOSR FA9550-05-1-0108 to Brown University and NSF grant PHY 0855678to the University of New Mexico

Appendix A Hyperbolicity of the first-order system

This appendix analyzes the matrix A(u) appearing in (10) in order to construct thecharacteristic fields (14) In matrix form the sector (8) of the principal part of (10) reads asfollows

partt

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

=

βr 0 minus 4λα3grr

0 0 λβr

6(grr)2λβr

3gθθgrr0 4λ

3grr

0 βr 0 23grrαχ

13α minus αχ

3grr

αχ3gθθ

minus23χ 0

0 0 βr 0 0 0 0 minus χgrr

0

0 0 minus 4α3grr

βr 0 βr

6(grr)2βr

3gθθgrr0 4

3grr

0 0 23αχ 0 βr minus βrχ

3grrminus2βrχ

3gθθ0 minus2

3χ

0 minus2α 0 0 0 23βr minus2grrβr

3gθθ0 4

3grr

0 gθθαgrr

0 0 0 minusgθθβr

3grr13βr 0 minus2

3gθθ

0 0 minus2α 0 0 0 0 βr 0

34

0 0 0 0 0 0 0 βr

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

prime

(A1)which defines the matrix A(u) appearing in (9) and so also the matrix A(u) in (10) Notethat in the last equation the matrix within the square brackets is minusA(u) For certainconfigurations of u and λ the system (10) is strongly hyperbolic [76] that is A(u) hasa complete set of eigenvectors and real eigenvalues Indeed five eigenpairs of A(u) aretrivially recovered upon inspection of A(u)rsquos leading 5times5 diagonal block These correspondto eigenvalue 0 and the left eigenspace ξj = eTj 1 le j le 5 where ej are the canonical

basis vectors Since each component of u arises as eTj W each is also a characteristic field

The remaining nine eigenpairs are determined by A(u) The eigenvalues of A(u) are

micro1 = 0 micro23 = minusβr microplusmn4 = minusβr plusmn

radic

2αχ

grr microplusmn

5 = minusβr plusmn α

radic

χ

grr microplusmn

6 = minusβr plusmnradic

λ

grr

(A2)

23

and the corresponding left eigenvectors are

x1 = (0 0 0 0 0 gθθ 2grr 0 0) (A3a)

x2 =

(

0 0 0 grr2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3b)

x3 =

(

grrλ 0 0 0

2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3c)

xplusmn4 =

(

0 0plusmnradic

2αgrrχ

0 0 0 0 1 0

)

(A3d)

xplusmn5 =

(

0∓ 3radicgrrχ

plusmn2

radic

grrχ 2grr

1

χminus 1

grr1

gθθ 0 0

)

(A3e)

xplusmn6 =

(

minus3

4

grrλ 0plusmn α

radicλgrr

(2αχminus λ) 0 0minus βr

8(βrgrr ∓radicλgrr)

minus βrgrr

4gθθ(βrgrr ∓radicλgrr)

αχ

(2αχminus λ)plusmnradic

grrλ

)

(A3f)

where for example xplusmn5 A(u) = microplusmn5 x

plusmn5 Assuming that grr gθθ χ and α are everywhere strictly

positive the eigenvalues are real and the eigenvectors are linearly independent providedthat (15) holds These eigenvectors are easily extended to eigenvectors of A(u) e g asxplusmn6 rarr (01times5 x

plusmn6 ) Then for example the characteristic field

Xplusmn6 equiv (01times5 x

plusmn6 )W = xplusmn6WvQ (A4)

and similarly Xplusmnj = xplusmnj WvQ for j = 4 5 and Xk = xkWvQ for k = 1 2 3 The characteristic

speeds for these fields are microk and microplusmnj With this convention the speeds listed in Table I

correspond to the Xk and Xplusmnj in (14)

Appendix B Schwarzschild solution in conformal Kerr-Schild coordinates

In Kerr-Schild coordinates here the system directly related to incoming Eddington-Finkelstein null coordinates the line element for the Schwarzschild solution reads

ds2 = minusα2dt2 + (1 + 2MR)(dR + βRdt)2 +R2dθ2 +R2 sin2 θdφ2 (B1)

where R is the area radius α = (1 + 2MR)minus12 is the lapse and βR = 2M(R + 2M) isthe shift vector The physical spatial metric gab is the spatial part of this line element

To define the corresponding solution to the BSSN system we use equation gab = χgab todefine the following relationship between line elements

dr2 + r2(dθ2 + sin2 θdφ2) = χ[(1 + 2MR)dR2 +R2dθ2 +R2 sin2 θdφ2] (B2)

so that

χ

(

1 +2M

R

)(

dR

dr

)2

= 1 χR2 = r2 (B3)

Then we have(

1 +2M

R

)12dR

R=dr

r (B4)

24

with integration yielding

r =R

4

(

1 +

radic

1 +2M

R

)2

e2minus2radic

1+2MR (B5)

where the constant of integration has been chosen so that the R r rarr infin limits are consistentThe second relation in (B3) shows that

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR =rArr χminus4 =2eradic

1+2MRminus1

1 +radic

1 + 2MR (B6)

The extrinsic curvature tensor is specified by the expression forK given in (B9h) the identityK = KR

R + 2Kθθ and

Kθθ =

(

1 +2M

R

)minus122M

R2 (B7)

Since KRR = Kr

r we compute that

Krr = K minus 2Kθ

θ = minus(

1 +2M

R

)minus12(R +M

R + 2M

)

2M

R2 (B8)

Next since Krr = grrKrr = χminus1Kr

r we have Krr = Arr+

13grrK This implies Arr = Kr

r minus 13K

from which we get (B9g) In all we have

α =

(

1 +2M

R

)minus12

(B9a)

βr = βR dr

dR= χ12

(

1 +2M

R

)minus122M

R(B9b)

grr = 1 (B9c)

gθθ = r2 = χR2 (B9d)

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR (B9e)

Br = 0 (B9f)

Arr = minus(

1 +2M

R

)minus124M

3R2

(

2R + 3M

R + 2M

)

(B9g)

K =

(

1 +2M

R

)minus32(

1 +3M

R

)

2M

R2(B9h)

Γr = minus2

r= minus 2

χ12R (B9i)

To differentiate these expressions with respect to r we use the identity

dR

dr= χminus12

(

1 +2M

R

)minus12

(B10)

along with the chain rule

25

Appendix C Stability of the model system

The following stability analysis for the model system (53) has been inspired by [78 79]After dropping all nonlinear source terms the system (53) becomes

parttu = uprime + av (C1a)

parttv = uprimeprime + vprime (C1b)

This section analyzes the stability of (C1) considering both the continuum system itself aswell as its semi-discrete dG approximation The latter analysis offers some insight into theempirically observed stability of our dG scheme for the spherically reduced BSSN equations

1 Analysis for a single interval

Throughout we work with the L2 inner product and norm

(f g)D =

int

D

fg fD =radic

(f f)D (C2)

where D is the spatial coordinate interval (here D may represent a subdomain Dk or the

whole domain Ω) and we have suppressed all integration measures For the continuummodel we will establish the following estimate

uprime(T middot)2D+ av(T middot)2

Dle C(T )

(

uprime(0 middot)2D+ av(0 middot)2

D

)

(C3)

where the time-dependent constant C(T ) is determined solely by the choice of boundaryconditions To show (C3) we first change variables with v =

radicav thereby rewriting (C1)

in the following symmetric form

parttu = uprime +radicav (C4a)

parttv =radicauprimeprime + vprime (C4b)

Equations (C4ab) then imply

1

2partt

int

D

(uprime)2 =

int

D

uprime(uprimeprime +radicavprime) =

int

D

radicauprimevprime +

1

2

int

partD

(uprime)2 (C5a)

1

2partt

int

D

(v)2 =

int

D

v(radicauprimeprime + vprime) = minus

int

D

radicauprimevprime +

1

2

int

partD

(v2 + 2radicauprimev) (C5b)

Here vvprime and uprimeuprimeprime have been expressed as exact derivatives and then integrated to boundaryterms the second equation employs an extra integration by parts and with only one spacedimension

int

partDdenotes a difference of endpoint evaluations Addition of Eqs (C5ab) gives

1

2partt

int

D

[

v2 + (uprime)2]

=1

2

int

partD

[

v2 + (uprime)2 + 2radicauprimev

]

(C6)

Substitutions with the identities

[

v2 + (uprime)2]

=1

2

[

(v + uprime)2 + (v minus uprime)2]

2uprimev =1

2

[

(v + uprime)2 minus (v minus uprime)2]

(C7)

26

and replacements to recover the original variable v = vradica yield

1

2partt

int

D

[

av2 + (uprime)2]

=1 +

radica

4

int

partD

(radicav + uprime)2 +

1minusradica

4

int

partD

(radicav minus uprime)2 (C8)

From (C8) we deduce that the time-dependent constant C(T ) in (C3) must satisfy

∣

∣

∣

∣

∣

∣

1 +

int T

0

[

1+radica

2

int

partD(radicav + uprime)2 + 1minus

radica

2

int

partD(radicav minus uprime)2

]

dt

uprime(0 middot)2 + av(0 middot)2

∣

∣

∣

∣

∣

∣

le C(T ) (C9)

For periodic boundary conditions we may choose C(T ) = 1 Moreover if a ge 1 anduprime = minusradic

av is specified at partD+ then uprime(t middot)2 + av(t middot)2 decaysStill working on a single interval (subdomain) we now consider the semi-discrete scheme

for (C4) i e (57) with all nonlinear source terms dropped and with v replaced by vradica

Derivation of a formula analogous to (C8) is our first step toward establishing L2 stabilityof the semi-discrete scheme While (57) features vectors for example u(t) taking values atthe Legendre-Gauss-Lobatto nodal points here we work with the numerical solution as apolynomial for example uh(t x) These two representations are related by the Lagrange in-terpolating polynomials for the nodal set here taken to span both the space of test functionsand the space of basis functions Our scheme is

int

Dk

ψparttuh =

int

Dk

ψ(Qh +radicavh) (C10a)

int

Dk

ξparttvh = minusint

Dk

ξprime(radicaQh + vh) +

int

partDk

ξ(radicaQlowast + vlowast) (C10b)

int

Dk

ϕQh =

int

Dk

ϕuprimeh +

int

partDk

ϕ (ulowast minus uh) (C10c)

where ψ ξ and ϕ are polynomial test functions These test functions are arbitrary exceptthat they must be degree-N polynomials In (C10) the variables uh vh and Qh should alsocarry a superscript k but we have suppressed this Derivation of a formula analogous to(C8) is complicated by the fact that Qh is not evolved Nevertheless at a given instant t wecan assemble Qh from (C10c)

Mimicking the calculation (C5b) from the continuum case we first use (C10b) with ξ = vhto write

1

2partt

int

Dk

v2h = minusint

Dk

(radicaQh + vh)v

primeh +

int

partDk

(radicaQlowast + vlowast)vh

= minusint

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2(radicaQlowast + vlowast)vh minus v2h

]

(C11)

The right-hand side of (C5a) is analogous to

1

2partt

int

Dk

Q2h =

int

Dk

QhparttQh (C12)

However since Qh is not evolved the term parttQh must be given a suitable interpretationOn the right hand side of (C10c) only uh u

primeh and u

lowast necessarily depend on time since the

27

test function ϕ need not be time-dependent Furthermore ulowast is explicitly given as a linearcombination of uh as seen in Eq (C21c) below Choosing ϕ = ℓj taking the time derivativeof (C10c) and appealing to the commutivity of mixed partial derivatives we therefore arriveat

int

Dk

ℓjparttQh =

int

Dk

ℓj(parttuh)prime +

int

partDk

ℓj(

(parttu)lowast minus parttuh

)

(C13)

where (parttu)lowast depends on parttuh in precisely the same way that ulowast depends on uh We have

written ℓj rather than ϕ in the last equation to emphasize that the result also holds for anylinear combination of ℓj (for example ϕ) and even for time-dependent combinations SinceQh is itself such a combination we obtain

1

2partt

int

Dk

Q2h =

int

Dk

Qh(parttuh)prime +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

Qh(Qprimeh +

radicavprimeh) +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2((parttu)lowast minus parttuh)Qh +Q2

h

]

(C14)

Addition of (C11) and (C14) gives

1

2partt

int

Dk

(Q2h + v2h) =

1

2

int

partDk

[

Q2h minus v2h + 2(

radicaQlowast + vlowast)vh + 2((parttu)

lowast minus parttuh)Qh

]

(C15)

the aforementioned analog of (C8) This formula holds on a single subdomain Dk and we

now combine multiple copies of it one for each value of k

2 Analysis for multiple intervals

To facilitate combination of (C15) over all k we change notation At every subdomaininterface Ik+12 equiv partDk cap partDk+1 let the superscripts L and R denote field values respectivelytaken from the left and right Then the fields evaluated at Ik+12 which belong to D

k will beuLk+12 v

Lk+12 and Q

Lk+12 while those belonging to D

k+1 will be uRk+12 vRk+12 and Q

Rk+12

However at Ikminus12 the values taken from D

k are uRkminus12 vRkminus12 and QR

kminus12 Note that we

have also replaced the subscript h denoting a numerical solution with k plusmn 12 denotingthe location of the endpoint value of the numerical solution With this notation we define

∆Lα =

1

2

[

(QLα)

2 minus (vLα)2]

+(radic

aQlowastα + vlowastα

)

vLα +[

(parttuα)lowast minus parttu

Lα

]

QLα (C16)

and similarly for ∆Rα The same numerical fluxes appear in both ∆L

α and ∆Rα (ie each

numerical flux takes the same value on either side of an interface) whence fluxes like Qlowastα do

not carry an L or R superscript In terms of these definitions (C15) becomes

1

2partt

int

Dk

(Q2h + v2h) = ∆L

k+12 minus∆Rkminus12 (C17)

28

Summation over all Dk yields

1

2partt

kmaxsum

k=1

int

Dk

(Q2h + v2h) =

kmaxminus1sum

k=1

(

∆Lk+12 minus∆R

k+12

)

+∆Lkmax+12 minus∆R

12

=

kmaxminus1sum

k=1

(

∆Lh minus∆R

h

)∣

∣

Ik+12 +∆Lkmax+12 minus∆R

12 (C18)

We have reverted to h-notation denoting the numerical solution since the LR superscriptsindicate unambiguously the relevant domain used for evaluation at Ik+12

We again seek an estimate of the form

kmaxsum

k=1

(

Qh(T middot)2Dk + avh(T middot)2Dk

)

le C(T )kmaxsum

k=1

(

Qh(0 middot)2Dk + avh(0 middot)2Dk

)

(C19)

that is essentially the same as the one (C3) considered in the continuum case Assume thatthe chosen boundary conditions ensure ∆L

kmax+12 minus ∆R12 is bounded by a time-dependent

constant which does not depend on the numerical parameters N and h (subdomain width)Establishment of stability then amounts to showing that the remaining sum over interfaceterms in (C18) is non-positive whence this remaining sum is consistent with C(T ) le 1although the boundary conditions may give rise to a different bound In fact we will choosethe numerical fluxes such that each individual interface term is non-positive At interfaceIk+12 and in LR notation the jump and average of vh for example are

1

2

(

v+ + vminus)

equiv vh =1

2

(

vLk+12 + vRk+12

)

(C20a)

nminusvminus + n+v+ equiv[[

vh]]

= vLk+12 minus vRk+12 (C20b)

Consider numerical fluxes of the form

Qlowast = Qh minusτQ2

[[

Qh

]]

(C21a)

vlowast = vh minusτv2

[[

vh]]

(C21b)

ulowast = uh minusτu2

[[

uh]]

(C21c)

(parttu)lowast = parttuh minus

τu2

[[

parttuh]]

(C21d)

where (C21c) induces (C21d) and where the penalty parameters τu τv and τQ are realnumbers The fluxes defined in (58) correspond to τu = 1 τv = 1 +

radica and τQ = 0 In

terms of these quantities the kth interface contribution in (C18) is

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

parttuh]]

minus τu2

[[

parttuh]][[

Qh

]]

(C22)

where we have suppressed the k dependence of the right-hand side Now consider the term[[

parttuh]]

Because parttuh and Qh+radicavh are both polynomials of degree N Eq (C10a) implies

29

τu

τ Q

minus1 0 1minus1

minus05

0

05

1

τu

minus1 0 1minus1

minus05

0

05

1

FIG 6 Stable evolutions for the model system For fixed τv = 10minus6 and τv = 1 +radic2

respectively the left and right plots depict stable choices (determined empirically) of τu and τQfor the linear model system (C1) The stable regions are colored black but the jagged edges result

from the discretization of the (τu τQ)-plane

the vector equation parttu = Q +radicav that is pointwise equivalence on the nodal points of

Dk which in turn implies

[[

parttuh]]

=[[

Qh+radicavh]]

Upon substituting this identity into thelast equation we arrive at an expression which features only vh and Qh

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

Qh +radicavh]]

minus τu2

[[

Qh +radicavh]][[

Qh

]]

(C23)

The identities vh[[

vh]]

= 12

[[

v2h]]

and[[

Qh +radicavh]]

=[[

Qh

]]

+radica[[

vh]]

then simplify(C23) to

(∆Lh minus∆R

h )∣

∣

Ik+12 = minusτv2

[[

vh]]2 minus

radica(τu + τQ)

2

[[

Qh

]][[

vh]]

minus τu2

[[

Qh

]]2 (C24)

The role of a penalty parameter is now clear Positive values of τv penalize jumps in vhthrough a negative contribution to the energy Likewise positive values of τu penalizejumps in Qh through a negative contribution to the energy However because the sign of[[

Qh

]][[

vh]]

can be positive or negative only the choice τQ = minusτu yields an expression for(∆L

h minus ∆Rh )|Ik+12 which is manifestly negative for τu ge 0 and τv ge 0 A simple estimate

based on Youngrsquos inequality with ε (that is 2αβ le εminus1α2 + εβ2 where α β ge 0 and ε gt 0)shows that for τQ = 0 the choice τv ge aτu4 also yields stability

Figure 6 depicts certain choices of stable penalty parameters for the linear model systemevolved to tfinal = 1000 (with a = 2 N = 10 and ∆t ≃ 00553) as determined empiricallywith simulations similar to those described in Sec IVA The left plot corresponds to asmall τv = 10minus6 for which the choice τu = 1 τQ = 0 is not stable as expected from thetheoretical analysis However the right plot corresponds to τv = 1 +

radica for which τu = 1

τQ = 0 is stable Motivated by the numerical flux choices (3842) used for the BSSN system

30

(4) we have (as mentioned above) set τu = 1 τv = 1 +radica and τQ = 0 in simulations of

the nonlinear model (53) For the nonlinear model system (53) the theoretically motivatedchoice τQ = minusτu also yields numerically stable evolutions when τu ge 0 and τv ge 0

For the nonlinear systems (4) and (53) we do not attempt a formal stability proofNevertheless the results of this appendix have served as a guide for our choices of penaltyparameters For the BSSN system (4) u v and Q are block indices [cf Eq (6)] Similarto the model problem we have penalized Q with τu = 1 with τv chosen large enough toheuristically overcome the cross-terms of indefinite size that arise from τQ = 0 (we interpretequations like τu = 1 componentwise) An analogous choice ldquoτQ = minusτurdquo for the BSSNsystem might be possible but would be considerably more complicated Indeed such achoice likely entails a matrix of penalty parameters but we do not give the details here

[1] F Pretorius Evolution of Binary Black-Hole Spacetimes Phys Rev Lett 95 121101 (2005)

(4 pages)

[2] F Pretorius Simulation of binary black hole spacetimes with a harmonic evolution scheme

Class Quantum Grav 23 S529-S552 (2006)

[3] M Campanelli C O Lousto P Marronetti and Y Zlochower Accurate Evolutions of Or-

biting Black-Hole Binaries without Excision Phys Rev Lett 96 111101 (2006) (4 pages)

[4] J G Baker J Centrella D-I Choi M Koppitz and J van Meter Gravitational-Wave

Extraction from an Inspiraling Configuration of Merging Black Holes Phys Rev Lett 96

111102 (2006) (4 pages)

[5] M Campanelli C O Lousto and Y Zlochower Last orbit of binary black holes Phys Rev D

73 061501(R) (2006) (5 pages)

[6] F Herrmann I Hinder D Shoemaker and P Laguna Unequal mass binary black hole plunges

and gravitational recoil Class Quantum Grav 24 S33-S42 (2007)

[7] P Diener F Herrmann D Pollney E Schnetter E Seidel R Takahashi J Thornburg and

J Ventrella Accurate Evolution of Orbiting Binary Black Holes Phys Rev Lett 96 121101

(2006) (4 pages)

[8] M A Scheel H P Pfeiffer L Lindblom L E Kidder O Rinne and S A Teukolsky Solving

Einsteinrsquos Equations With Dual Coordinate Frames Phys Rev D 74 104006 (2006)

[9] U Sperhake Binary black-hole evolutions of excision and puncture data Phys Rev D 76

104015 (2007) (20 pages)

[10] B Brugmann J A Gonzalez M Hannam S Husa U Sperhake and W Tichy Calibration

of moving puncture simulations Phys Rev D 77 024027 (2008) (25 pages)

[11] P Marronetti W Tichy B Brugmann J Gonzalez M Hannam S Husa and U Sperhake

Binary black holes on a budget simulations using workstations Class Quantum Grav 24

S43-S58 (2007)

[12] Z B Etienne J A Faber Y T Liu S L Shapiro and T W Baumgarte Filling the holes

Evolving excised binary black hole initial data with puncture techniques Phys Rev D 76

101503(R) (2007) (5 pages)

[13] B Szilagyi D Pollney L Rezzolla J Thornburg and J Winicour An explicit harmonic code

for black-hole evolution using excision Class Quantum Grav 24 S275-S293 (2007)

[14] L Boyle M Kesden and S Nissanke Binary-Black-Hole Merger Symmetry and the Spin

Expansion Phys Rev Lett 100 151101 (2008) (4 pages)

31

[15] M A Scheel M Boyle T Chu L E Kidder K D Matthews and H P Pfeiffer High-

accuracy waveforms for binary black hole inspiral merger and ringdown Phys Rev D 79

024003 (2009) (14 pages)

[16] M Hannam Status of black-hole-binary simulations for gravitational-wave detection

Class Quantum Grav 26 114001 (2009) (21 pages)

[17] I Hinder The current status of binary black hole simulations in numerical relativity

Class Quantum Grav 27 114004 (2010) (19 pages)

[18] B P Abbott et al LIGO The Laser Interferometer Gravitational-Wave Observatory

Rep Prog Phys 72 076901 (2009) (25 pages)

[19] F Acernese et al The status of VIRGO Class Quantum Grav 23 S63-S69 (2006)

[20] F Acernese et al Virgo status Class Quantum Grav 25 184001 (2008) (9 pages)

[21] R Arnowitt S Deser and C W Misner The Dynamics of General Relativity in Gravitation

an introduction to current research edited by L Witten (Wiley New York 1962) Chapter 7

227-265 Also available as gr-qc0405109

[22] J W York Jr Kinematics and Dynamics of General Relativity in Sources of Gravitational

Radiation edited by L L Smarr (Cambridge University Press Cambridge 1979)

[23] L E Kidder M A Scheel and S A Teukolsky Extending the lifetime of 3D black hole

computations with a new hyperbolic system of evolution equations Phys Rev D 64 064017

(2001) (13 pages)

[24] G Calabrese J Pullin O Sarbach and M Tiglio Convergence and stability in numerical

relativity Phys Rev D 66 041501(R) (2002) (4 pages)

[25] S Frittelli and O Reula On the Newtonian Limit of General Relativity Commun Math

Phys 166 221-235 (1994)

[26] M Shibata and T Nakamura Evolution of three-dimensional gravitational waves Harmonic

slicing case Phys Rev D 52 5428-5444 (1995)

[27] Y Choquet-Bruhat and J W York Geometrical well posed systems for the Einstein equations

C R Acad Sci Paris t 321 Serie I 1089-1095 (1995)

[28] A Abrahams A Anderson Y Choquet-Bruhat and J W York Jr Einstein and Yang-Mills

Theories in Hyperbolic Form without Gauge Fixing Phys Rev Lett 75 3377-3381 (1995)

[29] C Bona J Masso E Seidel and J Stela New Formalism for Numerical Relativity

Phys Rev Lett 75 600-603 (1995)

[30] M H P M van Putten and D M Eardley Nonlinear wave equations for relativity

Phys Rev D 53 3056-3063 (1996)

[31] S Frittelli and O A Reula First-Order Symmetric Hyperbolic Einstein Equations with Ar-

bitrary Fixed Gauge Phys Rev Lett 76 4667-4670 (1996)

[32] H Friedrich Hyperbolic reductions for Einsteinrsquos equations Class Quantum Grav 13 1451-

1469 (1996)

[33] F B Estabrook R S Robinson H D Wahlquist Hyperbolic equations for vacuum gravity

using special orthonormal frames Class Quantum Grav 14 1237-1247 (1997)

[34] M S Iriondo E O Leguizamon and O A Reula Einsteinrsquos Equations in Ashtekarrsquos Vari-

ables Constitute a Symmetric Hyperbolic System Phys Rev Lett 79 4732-4735 (1997)

[35] A Anderson Y Choquet-Bruhat and J W York Einstein-Bianchi hyperbolic system for

general relativity Topol Meth Nonlinear Anal 10 353-373 (1997)

[36] T W Baumgarte and S L Shapiro Numerical integration of Einsteinrsquos field equations

Phys Rev D 59 024007 (1998) (7 pages)

[37] M A G Bonilla Symmetric hyperbolic systems for Bianchi equations Class Quantum

4

where gab is the physical 3-metric and Kab is the physical extrinsic curvature tensor TheBSSN variables are the conformal metric gab the conformal factor χ the trace-free extrinsiccurvature Aab the trace K = gabKab and the conformal connection Γa equiv minusgminus12partb(g

12gab)where g is the determinant of the metric The BSSN system also includes the lapse α shiftvector βa and an auxiliary vector field Ba used to define the ldquoΓ-driverrdquo for the shift

Following Brown we adopt a spherically symmetric line element

ds2 = minusα2dt2 + χminus1grr(dr + βrdt)2 + χminus1gθθ(dθ2 + sin2 θdφ2) (2)

along with the spherically symmetric Ansatz

Γa =

Γr

minus cos θ(gθθ sin θ)0

Aab = Arr

1 0 00 minusgθθ(2grr) 00 0 minusgθθ sin2 θ(2grr)

(3a)

Subject to the assumption of spherical symmetry the basic variables are χ grr gθθ Arr KΓr α βr Br All are functions of t and r and satisfy the following spherically symmetric(generalized Lagrangian-form) BSSN system1

parttα = βrαprime minus 2αK minus (parttα)0 (4a)

parttβr = βrβrprime +

3

4Br minus (parttβ

r)0 (4b)

parttBr = βrBr prime + λ(parttΓ

r minus βrΓr prime)minus ηBr minus (parttBr)0 (4c)

parttχ = βrχprime +2

3Kαχminus βrgprimerrχ

3grrminus 2βrgprimeθθχ

3gθθminus 2

3βr primeχ (4d)

parttgrr =2

3βrgprimerr +

4

3grrβ

r prime minus 2Arrαminus 2grrβrgprimeθθ

3gθθ(4e)

parttgθθ =1

3βrgprimeθθ +

Arrgθθα

grrminus gθθβ

rgprimerr3grr

minus 2

3gθθβ

r prime (4f)

parttArr = βrAprimerr +

4

3Arrβ

rprime minus βrgprimerrArr

3grrminus 2βrgprimeθθArr

3gθθ+

2αχ(gprimerr)2

3g2rrminus αχ(gprimeθθ)

2

3g2θθminus α(χprime)2

6χ

+2

3grrαχΓ

r prime minus αχgprimerrgprimeθθ

2grrgθθ+χgprimerrα

prime

3grr+χgprimeθθα

prime

3gθθminus αgprimerrχ

prime

6grrminus αgprimeθθχ

prime

6gθθminus 2

3αprimeχprime +

αχprimeprime

3

minus 2

3χαprimeprime minus αχgprimeprimerr

3grr+αχgprimeprimeθθ3gθθ

minus 2αA2rr

grr+KαArr minus

2grrαχ

3gθθ(4g)

parttK = βrK prime +χgprimerrα

prime

2g2rrminus χgprimeθθα

prime

grrgθθ+αprimeχprime

2grrminus χαprimeprime

grr+

3αA2rr

2g2rr+

1

3αK2 (4h)

parttΓr = βrΓr prime +

Arrαgprimeθθ

g2rrgθθ+

2βrprimegprimeθθ3grrgθθ

+Arrαg

primerr

g3rrminus 4αK prime

3grrminus 2Arrα

prime

g2rrminus 3Arrαχ

prime

g2rrχ

+4βr primeprime

3grrminus βr(gprimeθθ)

2

grr(gθθ)2+

βrgprimeprimerr6(grr)2

+βrgprimeprimeθθ3gθθgrr

(4i)

where the prime stands for partial r-differentiation Eqs (4d-i) are Brownrsquos Eqs (9a-f)listed in [60] subject to his Lagrangian condition (corresponding to v = 1 in Brownrsquos equa-tions) The first three equations (4a-c) comprise the gauge sector and these are essentially

1 For this system the determinant g = grr(gθθ)2 sin4 θ is not unity

5

spherically symmetric versions of the standard ldquo1+logrdquo and ldquoΓ-driverrdquo conditions listed inEqs (1) and (2) of [60] However we have introduced the following minor modificationsFirst (parttα)0 designates a constant term which ensures that the right-hand side of the αevolution equation (4a) vanishes at the initial time This source term as well as the analo-gous terms appearing in the evolution equations (4bc) for βr and Br are needed to enablea static evolution of the Schwarzschild solution in Kerr-Schild coordinates Second the pa-rameter λ (perhaps with functional dependence) modifies the hyperbolicity of the first-ordersystem [80] The damping parameter η typically appears in standard versions of these gaugeevolution equations (See Sections IIC and IVB for further discussions) For this BSSNsystem we have three constraints the Hamiltonian constraint H the momentum constraintMr and the constraint Gr resulting from the definition of the conformal connection Γr Inspherical symmetry these constraints are written as follows

H = minus3A2rr

2g2rr+

2K2

3minus 5(χprime)2

2χgrr+

2χprimeprime

grr+

2χ

gθθminus 2χgprimeprimeθθgrrgθθ

+2χprimegprimeθθgrrgθθ

+χgprimerrg

primeθθ

g2rrgθθminus χprimegprimerr

g2rr+χ(gprimeθθ)

2

2grrg2θθ

(5a)

Mr =Aprime

rr

grrminus 2K prime

3minus 3Arrχ

prime

2χgrr+

3Arrgprimeθθ

2grrgθθminus Arrg

primerr

g2rr(5b)

Gr = minus gprimerr2g2rr

+gprimeθθgrrgθθ

+ Γr (5c)

These expressions are the ones listed by Brown in [60] Eqs (4ef) also ensure that thedeterminant factor g sin4 θ = grr(gθθ)

2 remains fixed throughout an evolution

B Abstract expressions of the system

We define the following vectors built with system variables

u =

χgrrgθθαβr

v =

Br

Arr

KΓr

Q =

χprime

gprimerrgprimeθθαprime

βrprime

(6)

Introduction of Qmight seem unnecessary at this stage but proves useful in the constructionof our discontinuous Galerkin scheme In terms of the vectors u v and Q we further define

Wuv =

(

uv

)

WvQ =

(

vQ

)

W =WuQ =

uvQ

(7)

Here we have introduced ldquocolon notationrdquo [81] to represent (sub)vectors and (sub)matricesalthough we employ the notation over block rather than individual elements In the first-order version of the system (4) the components of Q are promoted to independent fields in

6

which case the corresponding principal part features

parttBr = βrBr prime minus 4λα

3grrK prime +

4λ

3grrQprime

βr +λβr

6(grr)2Qprime

grr +λβr

3gθθgrrQprime

gθθ(8a)

parttArr = βrAprimerr +

2

3grrαχΓ

r prime +1

3αQprime

χ minus2

3χQprime

α minus αχ

3grrQprime

grr +αχ

3gθθQprime

gθθ(8b)

parttK = βrK prime minus χ

grrQprime

α (8c)

parttΓr = βrΓr prime minus 4αK prime

3grr+

4

3grrQprime

βr +βr

6(grr)2Qprime

grr +βr

3gθθgrrQprime

gθθ(8d)

parttQχ = βrQprimeχ +

2

3αχK prime minus βrχ

3grrQprime

grr minus2βrχ

3gθθQprime

gθθminus 2

3χQprime

βr (8e)

parttQgrr =2

3βrQprime

grr +4

3grrQ

primeβr minus 2αAprime

rr minus2grrβ

r

3gθθQprime

gθθ(8f)

parttQgθθ =1

3βrQprime

gθθ+gθθα

grrAprime

rr minusgθθβ

r

3grrQprime

grr minus2

3gθθQ

primeβr (8g)

parttQα = βrQprimeα minus 2αK prime (8h)

parttQβr = βrQprimeβr +

3

4Br prime (8i)

where all lower-order terms on the right-hand side have been dropped This sector of prin-cipal parts of the first-order system has the form

parttWvQ + A(u)W primevQ = 0 (9)

where (minus) the explicit form of the 9-by-9 matrix A(u) is given below in (A1) Thefirst-order version of (4) takes the nonconservative form

parttW +A(u)W prime = S(W ) A(u) =

(

05times5 05times9

09times5 A(u)

)

(10)

where S(W ) is a vector of lower order terms built with all components of W Partition ofA(u) = A(u)vQvQ into blocks corresponding to the v and Q sectors yields

A(u) =

(

A(u)vv A(u)vQ

A(u)Qv A(u)QQ

)

(11)

Using these blocks we then define the 9-by-9 matrix

A(u) = A(u)uvvQ =

(

05times4 05times5

A(u)vv A(u)vQ

)

(12)

and express (4) as

parttWuv + A(u)W primevQ = S(W ) (13a)

Q = uprime (13b)

where S(W ) = S(W )uv

7

field speed

X1 micro1 = 0

X23 micro23 = minusβr

Xplusmn4 microplusmn

4 = minusβr plusmnradic

2αχgrr

Xplusmn5 microplusmn

5 = minusβr plusmn αradic

χgrr

Xplusmn6 microplusmn

6 = minusβr plusmnradic

λgrr

TABLE I Characteristic speeds These speeds are the eigenvalues listed in (A2)

C Hyperbolicity and characteristic fields

Although our numerical scheme deals directly with the second-order spatial operatorsappearing in (4) we first consider the hyperbolicity of the corresponding first-order system(10) The characteristic fields and their speeds are found by instantaneously ldquofreezingrdquothe fields u in A(u) to some value u0 corresponding to a linearization around a uniformstate Below we continue to write u for simplicity with the understanding that u is reallythe background solution u0 Of primary interest is the range of u0 for which the system isstrongly hyperbolic [73ndash76]

Appendix A shows that the characteristic fields corresponding to (4) are as follows (i)all components of u (each with speed 0) and (ii) the fields

X1 = gθθQgrr + 2grrQgθθ (14a)

X2 = grrΓr +

2

χQχ minus

1

2grrQgrr minus

1

gθθQgθθ (14b)

X3 =grrλBr +

2

χQχ minus

1

2grrQgrr minus

1

gθθQgθθ (14c)

Xplusmn4 = plusmn

radic

2αgrrχ

K +Qα (14d)

Xplusmn5 = ∓ 3radic

grrχArr plusmn 2

radic

grrχK + 2grrΓ

r +1

χQχ minus

1

grrQgrr +

1

gθθQgθθ (14e)

Xplusmn6 = minus3

4

grrλBr plusmn α

radicλgrr

(2αχminus λ)K minus βr

8(βrgrr ∓radicλgrr)

Qgrr

minus βrgrr

4gθθ(βrgrr ∓radicλgrr)

Qgθθ +αχ

(2αχminus λ)Qα plusmn

radic

grrλQβr (14f)

with the speeds listed in Table I To ensure strong hyperbolicity we must necessarily require

λ gt 0 (βr)2grr minus λ 6= 0 2αχminus λ 6= 0 (15)

as shown in in Appendix A where further conditions are also given When λ = 1 thehyperbolicity condition of Ref [60] is recovered In fact the system could be recast assymmetric hyperbolic Indeed as it involves one spatial dimension the relevant symmetrizercan be constructed via polar decomposition of the diagonalizing similiarity transformationHowever we will not exploit this possibility

8

This system admits an inner excision boundary provided

βr ge max

(

radic

2αχ

grr

radic

α2χ

grr

radic

λ

grr

)

(16)

holds at the inner boundary This condition ensures each characteristic field has a nonposi-tive speed at the inner boundary and therefore the inner boundary is an excision boundaryat which no boundary conditions are needed The extra flexibility afforded by the param-eter λ could be used to maintain rigorous hyperbolicity by moving the points at whichthe conditions in (15) are violated outside of the computational domain Furthermore forλ = 1 Eq (16) conceivably fails or is only satisfied close to r = 0 where field gradients are

prohibitively large The troublesome X+6 gauge mode has a positive speed minusβr +

radic

λgrrIndeed for the conformally flat Kerr-Schild system considered in section IVB an inner ex-cision boundary is only possible provided λ is small enough

The transformation (14) can be inverted in order to express the fundamental fields interms of the characteristic fields

Br = minus1

6

λ

grrgθθ

[

(βr)2

(βr)2grr minus λ

]

X1 +2

3

λαχ

grr(2αχminus λ)(X+

4 +Xminus4 )minus

2

3

λ

grr(X+

6 +Xminus6 )

(17a)

Arr =1

3

radic

grrχ

2α(X+

4 minusXminus4 )minus

radicgrrχ

6(X+

5 minusXminus5 ) (17b)

K =

radic

χ

8αgrr(X+

4 minusXminus4 ) (17c)

Γr = minus1

6

1

grrgθθ

[

(βr)2

(βr)2grr minus λ

]

X1 +1

grr(X2 minusX3) +

2

3

αχ

grr(2αχminus λ)(X+

4 +Xminus4 )

minus 2

3

1

grr(X+

6 +Xminus6 ) (17d)

Qχ =1

12

χ

grrgθθ

[

4(βr)2grr minus 3λ

(βr)2grr minus λ

]

X1 +χ

2X3 minus

1

3

αχ2

(2αχminus λ)(X+

4 +Xminus4 ) (17e)

+χ

3(X+

6 +Xminus6 ) (17f)

Qgrr =2(βr)2grr minus 3λ

6gθθ((βr)2grr minus λ)X1 +

4

3grrX2 minus grrX3 +

2

3

αχgrr(2αχminus λ)

(X+4 +Xminus

4 )

minus 1

3grr(X

+5 +Xminus

5 )minus2

3grr(X

+6 +Xminus

6 ) (17g)

Qgθθ =

[

1

4grr+

(βr)2

12((βr)2grr minus λ)

]

X1 minus2

3gθθX2 +

1

2gθθX3 minus

1

3

αχgθθ(2αχminus λ)

(X+4 +Xminus

4 )

+1

6gθθ(X

+5 +Xminus

5 ) +1

3gθθ(X

+6 +Xminus

6 ) (17h)

Qα =1

2(X+

4 +Xminus4 ) (17i)

Qβr =βrλ

8grrgθθ((βr)2grr minus λ)X1 minus

λ

(2αχminus λ)

radic

αχ

8grr(X+

4 minusXminus4 ) +

1

2

radic

λ

grr(X+

6 minusXminus6 )

(17j)

9

We will refer to this inverse transformation when discussing outer boundary conditions forour numerical simulations in Sec IVB

III DISCONTINUOUS GALERKIN METHOD

This section describes the nodal discontinuous Galerkin method used to numerically solve(4) We adopt a method-of-lines strategy and here describe the relevant semi-discrete schemewhile leaving the temporal dimension continuous To approximate (4) we follow the generalprocedure first introduced in Ref [82] Our approach defines local auxiliary variables Q = uprimeand rewrites the spatially second-order system (4) as the first-order system (13a) Once weuse (13b) to eliminate Q from (13a) we recover the primal equations (4) The auxiliaryvariable approach was later generalized and coined the local discontinuous Galkerin (LDG)method in Ref [77] We may refer to our particular scheme as an LDG method but notethat many variations exist in the literature We stress that in LDG methods Q is not evolvedand is introduced primarily to assist in the construction of a stable scheme

Equations (12) and (13a) imply that the physical flux function is

F (W ) =

(

Fu(W )

Fv(W )

)

equiv A(u)WvQ =

(

05times1

f(W )

)

f =

fBr

fArr

fKfΓ

(18)

Only the evolution equations for Br Arr K and Γr give rise to non-zero components in F and we have collected these non-zero components into a smaller vector f = Fv Inspectionof (8) determines these components For example from (8c) we find

fK = minusβrK +χ

grrQα (19)

A Local approximation of the system (13)

Our treatment closely follows [83] but with the equations and notations relevant forthis paper Our computational domain Ω is the closed r-interval [a b] We cover Ω withkmax gt 1 non-overlapping intervals Dk = [ak bk] where a = a1 b = bkmax and bkminus1 = ak fork = 2 middot middot middot kmax

On each interval Dk we approximate each component of the system vector W by a localinterpolating polynomial of degree N For example

χkh(t r) =

Nsum

j=0

χ(t rkj )ℓkj (r) (20)

approximates χ(t r) Throughout this section approximations are denoted by a subscripth (see [66] for the notation) For example Wh and fh are approximations of W and f Although Q = uprime Qh and uprimeh are not necessarily the same In (20) ℓkj (r) is the jth Lagrange

polynomial belonging to Dk

ℓkj (r) =Nprod

i=0i 6=j

r minus rkirkj minus rki

(21)

10

Evidently the polynomial χkh interpolates χ at rkj To define the nodes rkj consider the

mapping from the unit interval [minus1 1] to Dk

rk(u) = ak + 12(1 + u)(bk minus ak) (22)

and the N+1 Legendre-Gauss-Lobatto (LGL) nodes uj The uj are the roots of the equation

(1minus u2)P primeN(u) = 0 (23)

where PN(u) is the Nth degree Legendre polynomial and the physical nodes are simplyrkj = rk(uj) In vector notation the approximation (20) takes the form

χkh(t r) = χk(t)Tℓk(r) (24)

in terms of the column vectors

χk(t) =[

χ(t rk0) middot middot middot χ(t rkN)]T ℓk(r) =

[

ℓk0(r) middot middot middot ℓkN(r)]T (25)

On each open interval (ak bk) sub Dk and for each component of the equations in (13) we

define local residuals measuring the extent to which our approximations satisfy the originalcontinuum system Dropping the subdomain label k on the polynomials and focusing onthe K equation as a representative example the local residual corresponding to (4h) is

minus(RK)kh equivminus parttKh + (βrK prime)h minus

(

χQprimeα

grr

)

h

+

(

χQgrrQα

2g2rr

)

h

minus(

χQgθθQα

grrgθθ

)

h

+

(

QαQχ

2grr

)

h

+

(

3αA2rr

2g2rr

)

h

+

(

1

3αK2

)

h

(26)

Here for example the expressions 2

(βrK prime)h = βrhK

primeh

(

QαQχ

2grr

)

h

=QαhQχh

2grrh (27)

We similarly construct the remaining eight residuals eg (Rgrr)h and (RΓr)h as well as fiveresiduals corresponding to (13b) For example one of these remaining five is

(RQα)kh equiv minusQαh + αprime

h (28)

Let the kth inner product be defined as

(

u v)

Dk equivint bk

akdru(r)v(r) (29)

and consider the expression (ℓkj (RK)kh)Dk We call the requirement that this inner product

vanish forallj the kth Galerkin condition For each component of the system and for each k

2 At this stage the first expression is generically a polynomial of degree 2N minus 1 and the latter is not a

polynomial The conventions adopted in Eq (27) prove useful while working with the residual However

later on in Sec III C to obtain the final form (47) of the numerical approximation corresponding to (26)

we will replace nonlinear terms with degree-N polynomials

11

there is a corresponding Galerkin condition in total 9kmax(N + 1) equations for (13a) and5kmax(N+1) for (13b) Enforcement of the Galerkin conditions on each D

k will not recover ameaningful global solution since they provide no mechanism for coupling the local solutionson the different intervals Borrowing from the finite volume toolbox we achieve couplingthrough integration by parts on r and introduction of the numerical flux f lowast at the interfacebetween subdomains

In (26) we only need to consider (βrK prime)h and (χQprimeαgrr)h as the other terms comprise a

component of the source vector Sh Using integration by parts we write

(

ℓkj (βrK prime)h

)

Dk =minusint bk

akdr[

(

ℓkjβrh

)primeKh

]

+ (βrhKh) ℓ

kj

∣

∣

∣

bk

ak (30a)

(

ℓkj (χQprimeαgrr)h

)

Dk =minusint bk

akdr

[(

ℓkjχh

grrh

)prime

Qαh

]

+

(

χhQαh

grrh

)

ℓkj

∣

∣

∣

bk

ak (30b)

In these formulas we have retained the domain index k on ℓkj while continuing to suppress iton Kh grrh etc Moreover we have suppressed the r-dependence in all terms on the right-hand side Addition of these formulas along with the definition fKh = minus(βrK)h+(χQαgrr)hgives

(

ℓkj (βrK prime)h minus (χQprime

αgrr)h)

Dk = minusint bk

akdr

[

(

ℓkjβrh

)primeKh minus

(

ℓkjχh

grrh

)primeQαh

]

minus fKhℓkj

∣

∣

∣

bk

ak (31)

In lieu of (31) we will instead work with the replacement

(

ℓkj (βrK prime)h minus (χQprime

αgrr)h)

Dk rarr minusint bk

akdr

[

(

ℓkjβrh

)primeKh minus

(

ℓkjχh

grrh

)primeQαh

]

minus f lowastKℓ

kj

∣

∣

∣

bk

ak (32)

This replacement features a component f lowastK of the numerical flux rather than a component

fKh of the boundary flux The numerical flux is determined by (as yet not chosen) functions3

f lowast = f lowast(W+Wminus) (33)

where for example Wminus is an interior boundary value [either W kh (t a

k) or W kh (t b

k)] of theapproximation defined on D

k andW+ is an exterior boundary value [either W kminus1h (t bkminus1) or

W k+1h (t ak+1)] of the approximation defined on either Dkminus1 or Dk+1 We discuss our choice

of numerical flux in the next subsection We now employ additional integration by parts towrite the above replacement as

(

ℓkj (βrK prime)h minus (χQprime

αgrr)h)

Dk rarrint bk

akdrℓkj

(

βrK prime minus χQprimeα

grr

)

h

+ (fKh minus f lowastK) ℓ

kj

∣

∣

∣

bk

ak (34)

Rather than the exact kth Galerkin condition(

ℓkj (RK)kh

)

Dk = 0 forallj for the K component

of (13) on Dk we will instead strive to enforce

(

ℓkj (RK)kh

)

Dk = (fKh minus f lowastK) ℓ

kj

∣

∣

∣

bk

ak forallj (35)

3 In the context of the dG method here + and minus denote ldquoexteriorrdquo and ldquointeriorrdquo and have no relation to

the plusmn using to denote the characteristic fields and speeds in Table I For characteristic fields and speeds

+ and minus mean ldquoright-movingrdquo and ldquoleft-movingrdquo

12

although our treatment of nonlinear terms will lead to a slight modification of these equations(we return to this issue shortly) The other components of (13a) are treated similarly asare the components of (13b) Recall that for example Qα = αprime Formally using the samedG method to solve for Qα we arrive at the replacement

(

ℓkj (RQα)kh

)

Dk rarrint bk

akdrℓkj (minusQαh + αprime

h)minus (αh minus αlowast) ℓkj

∣

∣

∣

bk

ak (36)

which again features a component αlowast of the numerical flux The auxiliary variables areconstructed and used at each stage of temporal integration We then have

(

ℓkj (RQα)kh

)

Dk = (αh minus αlowast) ℓkj

∣

∣

∣

bk

ak forallj (37)

as the corresponding enforced kth Galerkin condition

B Numerical Flux

To further complete our dG scheme we must specify functional forms for the compo-nents of the numerical flux introduced in the previous section We distinguish between thephysical fluxes (components of f) and the auxiliary fluxes (components of u) arising fromthe definition of the auxiliary variables These choices are not independent as the resultingscheme must be stable and consistent Our choice follows [84] which considered diffusionproblems Additional analysis of this flux choice appears in [66 85]

Let us first consider the numerical fluxes corresponding to the physical fluxes and of theform (33) The numerical flux vector is a function of the system and auxiliary variablesinterior and exterior to a subdomain A common choice for f lowast is

f lowast = fh+τ

2

[[

vh]]

K-component of f lowast f lowastK = fKh+

τ

2

[[

Kh

]]

(38)

where as an example we have also shown the component of f lowast corresponding to the analysisabove Respectively the average and jump across the interface are

fh =1

2

(

f+ + fminus) [[

vh]]

= nminusvminus + n+v+ (39)

Here τ is a position dependent penalty parameter (fixed below) and nminus(n+) is the localoutward pointing normal to the interior (exterior) subdomain The role of τ is to ldquopenalizerdquo(i e yield a negative contribution to the L2 energy norm) jumps across an interface Anappropriate choice of τ will ensure stability and we now provide some motivation for thechoice (41) of τ we make below

Were we treating the fully first-order system (10) the local Lax-Friedrichs flux wouldoften be a preferred choice due to its simplicity [66] In this case the constant ω in thenumerical flux formula Flowast = Fh + 1

2ω[[

Wh

]]

obeys ω ge max∣

∣micro(nablaWF(W ))∣

∣ HereF(W ) = A(u)W the notation micro(middot) indicates the spectral radius of the matrix within andthe max is taken over interior Wminus and exterior W+ states Motivated by (9) we adopt asimilar but simpler prescription substituting the field gradient

nablaWvQA(u)WvQ = A(u) (40)

13

for nablaWF(W ) Precisely we assume the scaling

τ(bk) = τ(ak+1) = τk+12 equiv C middotmax∣

∣micro(

A(u))∣

∣ (41)

where C = O(1) is a constant chosen for stability Larger values of C will result in schemeswith better stability properties whereas too large a value will impact the CFL conditionAt the interface point Ik+12 equiv D

k cap Dk+1 the vector uh has two representations uminus at bk

and u+ at ak+1 The max in (41) is taken over the corresponding two sets of field speedsMore precisely the speeds in Table I are computed for both uminus and u+ and the maximumtaken over all resulting speeds For the auxiliary variables a penalized central flux is usedThe definition with one representative component is

ulowast = uh minus1

2

[[

uh]]

α-component of ulowast αlowast = αh minus1

2

[[

αh

]]

(42)

with similar expressions for the remaining componentsWe stress the following point Since the interior coupling between subdomains is achieved

through the numerical flux forms (41) and (42) the inverse transformation (17) expressingthe fundamental fields in terms of the characteristic fields is not required to achieve thiscoupling On the other hand imposition of physical boundary conditions may still rely on(17) since this transformation allows one to fix only incoming characteristic modes

C Matrix form of the semi-discrete equations

Let us introduce the kth mass and stiffness matrices

Mkij =

int bk

akdrℓki (r)ℓ

kj (r) Sk

ij =

int bk

akdrℓki (r)ℓ

kjprime(r) (43)

These matrices belong to Dk and the corresponding matrices defined on the reference interval[minus1 1] are

Mij =

int 1

minus1

duℓi(u)ℓj(u) Sij =

int 1

minus1

duℓi(u)ℓprimej(u) (44)

where ℓj(u) is the jth Lagrange polynomial determined by the LGL nodes uj on [minus1 1]These matrices are related by Mk

ij =12(bk minus ak)Mij and S

kij = Sij whence only the reference

matrices require computation and storageWe will use the matrices Mk and Sk in obtaining an ODE system from (26) and (35)

Towards this end we first approximate the nonlinear terms (products and quotients) in(26) by degree-N interpolating polynomials Such approximations are achieved throughpointwise representations For example (QαQχgrr)h appears in (26) and is expressed inthe following way [cf footnote 2]

(

QαQχ

grr

)

h

(t r) =

(

QαhQχh

grrh

)

(t r) rarrNsum

j=0

Qαh(t rkj )Qχh(t r

kj )

grrh(t rkj )

ℓkj (r) (45)

Note that the expressions on the right and left are not equivalent due to aliasing error [65]Our vector notation for this replacement will be

(

QαQχ

grr

)

h

(t r) rarr(

QαQχ

grr

)

(t)Tℓk(r) (46)

14

Operations among bold variables are always performed pointwise Making similar replace-ments for all terms in (26) and then carrying out the integrations in (35) which bring inMk and Sk we arrive at

parttK = βrDK minus χDQα

grr

+1

2

χQgrrQα

g2rr

minusχQgθθ

Qα

grrgθθ

+1

2

QαQχ

grr

+3

2

αA2rr

g2rr

+1

3αK2 +Mminus1ℓk (fKh minus f lowast

K)∣

∣

∣

bk

ak (47)

where we have again suppressed the superscript k on all terms except ℓk(r) and the sub-script h is dropped on all boldfaced variables As described in [66] the spectral collocationderivative matrix

(Dk)ij =dℓkjdr

∣

∣

∣

∣

∣

r=rki

(48)

can also be expressed as Dk = (Mk)minus1Sk which appears above Eight other semi-discreteevolution equations are similarly obtained with nine in total (one for each component ofWuv) Additionally we have

Qα = Dα+Mminus1ℓk(αlowast minusαh)∣

∣

∣

bk

ak (49)

as one of the auxiliary equations with five in total (one for each component of Q = WQQ)

D Filtering

Like other nodal (pseudospectral) methods our scheme may suffer from instabilitiesdriven by aliasing error [65] Filtering is a simple yet robust remedy To filter a solu-tion component such as χ we use the modal (as opposed to nodal) representation of thesolution

χkh(t r) =

Nsum

j=0

χ(t rkj )ℓkj (r) =

Nsum

j=0

χkj (t)Pj(r) (50)

where Pj(r) is the jth Legendre polynomial Let ηj = jN and define the filter function

σ(ηj) =

1 for 0 le ηj le NcN

exp(

minus ǫ(

ηjminusNcN

1minusNcN

)2s)

for NcN le ηj le 1(51)

At each timestep we modify our solution component according to

χkh rarr

(

χkh

)filtered=

Nsum

j=0

σ(ηj)χkj (t)Pj(r) (52)

Evidently the modification only affects the top N minus Nc modes and is sufficient to controlthe type of weak instability driven by aliasing [66] The numerical parameters Nc and ǫ areproblem dependent For our simulations we have taken ǫ ≃ minuslog(εmach) = 36 where εmach

is machine accuracy in double precision

15

E Model system

To better illustrate the basic properties of our method we consider a toy model Namelythe following spatially second-order system

parttu = uprime + av minus u3 + g(t x) (53a)

parttv = uprimeprime + vprime minus (u+ v)(uprime)2 + v2u2 + h(t x) (53b)

where a ge 1 is constant and g and h are analytic source terms to be specified In contrastto (6) here u v and Q = uprime are scalars rather than vectors System (53) admits a first-order reduction in which uprime is defined as an extra variable Since this first-order reductionis strongly hyperbolic the spatially second-order system (53) is also strongly hyperbolic byone of the definitions considered in [75] The characteristic fields Xplusmn and speeds microplusmn are

X+ =radicav minus uprime micro+ =

radicaminus 1 Xminus =

radicav + uprime microminus = minus(

radica+ 1) (54)

To construct a local dG scheme for this system we first rewrite it as

parttu = Q+ av minus u3 + g(t x) (55a)

parttv = Qprime + vprime minus (u+ v)Q2 + v2u2 + h(t x) (55b)

Q = uprime (55c)

Evidently f = minus(Q + v) is the v-component of the physical flux vector

F (vQ) equiv(

Fu

Fv

)

=

(

0

f

)

(56)

Note that F has the same structure as (u v)T Borrowing from the presentation for theBSSN system we write the analogous semidiscrete scheme on each subdomain D

k for themodel system

parttu = Q+ av minus u3 + g(t) (57a)

parttv = DQ+Dv minus (u+ v)Q2 + v2u2 + h(t) +Mminus1ℓk(fh minus f lowast)∣

∣

∣

bk

ak (57b)

Q = Du+Mminus1ℓk(ulowast minus uh)∣

∣

∣

bk

ak (57c)

Here we have suppressed the subinterval label k from all variables except for the vectorℓk of Lagrange polynomial values Moreover following the guidelines discussed above thenumerical fluxes are given by

f lowast = fh+1 +

radica

2

[[

vh]]

ulowast = uh minus1

2

[[

uh]]

(58)

Appendix C analyzes the stability of our scheme for a more general numerical flux choiceas applied to (53) with the nonlinear and source terms dropped

IV RESULTS FROM NUMERICAL SIMULATIONS

This section presents results found by numerically solving both the model system (53)and BSSN system (4) with the dG scheme presented in Sec III

16

10minus10

10minus6

10minus2

∆uinfin

10minus10

10minus6

10minus2

∆vinfin

0 500 1000 1500 2000 2500 3000 3500 400010

minus10

10minus6

10minus2

∆(partxu)infin

t

N = 3N = 6N = 9N = 12

FIG 1 Spectral convergence of fields for model PDE Respectively for N = 3 6 9 12

a timestep of ∆t = 00578 00178 00084 00049 has been chosen for stability and accuracy In the

title headings for example ∆u equiv unumer minus uexact

A Simulations of the model system

The semi-discrete scheme (57) has been integrated with the classical fourth-order Runge-Kutta method When integrating this system we have first constructed Q at each Runge-Kutta stage and then substituted into the evolution equations (57ab) for u and v Theproblem has been solved on a computational domain [0 4π] comprised of two subdomainswith a timestep chosen small enough for stability The initial data has been taken from thefollowing exact solution to (53)

uprimeexact(t x) =1

2

[

sin(xminus microminust)minus sin(xminus micro+t)]

(59a)

vexact(t x) =1

2radica

[

sin(xminus microminust) + sin(xminus micro+t)]

(59b)

g(t x) = u3exact (59c)

h(t x) = (uexact + vexact)(uprimeexact)

2 minus v2exactu2exact (59d)

where the speeds microplusmn are found in (54) Specification of the boundary condition at a physicalendpoint amounts to choosing the external state for at the endpoint We have considered

17

8 16 32 64 128

10minus3

10minus2

10minus1

N

∆t

max stable ∆t8N2

FIG 2 Scaling of maximum stable ∆t with N for model PDE

two possibilities (i) the analytic state (Q+ v+) = (Qexact vexact) and (ii) an upwind stateFor example at x = 4π the upwind state is4

Q+ = Qupwind =1

2

[

(Xminus)exact minus (X+)numer

]

v+ = vupwind =1

2radica

[

(Xminus)exact + (X+)numer

]

(60)Either choice of (Q+ v+) leads to similar results and the plots here correspond to theanalytic state Figure 1 clearly shows spectral convergence with increasing polynomial orderN across all fields for the case a = 2 Other values of a including a = 1 for which X+

is a static characteristic field have also been considered with similar results Appendix Cdemonstrates that our proposed scheme for the system (57) with nonlinear and source termsdropped is stable in a semi-discrete sense Nevertheless the fully discrete scheme obtainedvia temporal discretization by the fourth-order Runge-Kutta method is still subject to thestandard absolute stability requirement Namely if microh is any eigenvalue corresponding tothe (linearized) discrete spatial operator then a necessary condition for stability is thatmicroh∆t lies in absolute stability region for Runge-Kutta 4 We here show empirically that theassociated timestep restriction scales like Nminus2 ie ∆t = O(Nminus2) for stability We note thatsuch scaling is welcome in light of the second-order spatial operators which appear in thesystem and suggest a possible worse scaling like Nminus4 Fig 2 plots the maximum stabletimestep for a range of N demonstrating the Nminus2 scaling in line with behavior known fromanalysis of first-order systems [66] This scaling also holds for the BSSN system

4 We remind the reader that unfortunately the plusmn on Xplusmn means something different than the plusmn indicating

exteriorinterior dG states [cf footnote 3]

18

B Simulations of the BSSN system

This subsection documents results for simulations of the unit-mass-parameter (M = 1)Schwarzschild solution (B9) expressed in terms of ingoing Kerr-Schild coordinates Sincethe solution is stationary temporal integration of the semi-discrete scheme has been carriedout with the forward Euler method The r-coordinate domain [04 34] has been split into3 equally spaced subdomains and we have set η = 10 λ = 01 and C = 2 [cf Eq (41)]For all simulations ∆t has been chosen for stability With the chosen λ the inner physicalboundary rmin = 04 is an excision surface At each timestep we have applied an (order2s = 20) exponential filter on the top two-thirds of the modal coefficient set for all fieldsexcept for grr and gθθ For stability we have empirically observed that grr and gθθ must not

be filtered A detailed understanding of this is still lackingIssues related to physical boundary conditions are similar to the one encountered in

Sec IVA for the model problem Similar to before we have retained Eqs (3842) as thechoice of numerical flux even at the endpoints Therefore at an endpoint the specificationof the boundary condition amounts to the choice W+ of external state We have typicallychosen the inner boundary of the radial domain as an excision boundary and in this caseW+ = Wminus is enforced at the inner physical boundary At the outer physical boundaryfor W+ we have again considered two choices (i) Wexact and (ii) Wupwind To enforcechoice (ii) the inverse transformation (17) must be used with incoming characteristic fieldsfixed to their exact values similar to (60) We have tried various versions of choice (ii)and in all cases the resulting simulations have been unstable We therefore present resultscorresponding to choice (i) Although the choice of an analytical external state Wexact

at the outer boundary is stable for our problem such a boundary condition is unlikelyto generalize to more complicated scenarios involving dynamical fields Indeed the issueof outer boundary conditions for the BSSN system is an active area of research with aproper treatment requiring fixation of incoming radiation control of the constraints andspecification of gauge (see Ref [86] for a recent analysis)

For BSSN simulations our main diagnostic is to monitor the Hamiltonian momentumand conformal connection constraints Figure 3 depicts long-time histories of constraintviolations whereas Figs 4 and 5 depict long-time error histories for the individual BSSNfield components From the middle plot in Fig 5 we infer that up to the indicated numericalerror the factor g sin4 θ = grr(gθθ)

2 remains at its initial fixed profile r4 throughout theevolution These figures indicate that the proposed scheme is stable for long times andexhibits spectral converge with increased polynomial order N Similar results are recoveredfromM = 0 Minkowski initial data The stability documented in these plots does not appearto rely on inordinate parameter tuning For example with the fixed parameters describedabove we obtain similar plots if we individually vary (i) rmin over 0325 035 04 0475(values still corresponding to an excision surface for the given choice of λ) (ii) η over1 3 7 10 (iii) s over 8 9 10 With the polynomial order N ranging over 23 26 29 31both stability and qualitatively similar exponential convergence is achieved with a singlesubdomain Likewise adoption of a larger coordinate domain with more subdomains doesnot impact our results Finally we have considered the addition of random noise to all fieldcomponents at the initial time Precisely with the system component χ as an example wehave set

χ(0) 7rarr χ(0) + δχ(0) (61)

where each component (nodal value) of δχ(0) is 10minus5 times a random variable drawn from a

19

10minus9

10minus7

10minus5

10minus3

Hinfin

10minus9

10minus7

10minus5

10minus3

Mrinfin

0 2000 4000 6000 8000 10000

10minus9

10minus7

10minus5

10minus3

Grinfin

t

N = 11N = 14N = 17N = 20

FIG 3 Spectral convergence of constraint violations for M = 1 Kerr-Schild ini-

tial data Respectively for N = 11 14 17 19 a timestep of ∆t ≃ 00041 00026 00018 00013

has been chosen for stability and accuracy

standard normal distribution Such perturbed initial data also gives rise to stable evolutions

V CONCLUSION

We have introduced a discontinuous Galerkin method for solving the spherically reducedBSSN system with second-order spatial operators Our scheme shares similarities withother discontinuous Galerkin methods that use local auxiliary variables to handle high-orderspatial derivatives [66 77ndash79 82 84 85] and which have typically been applied to eitherelliptic parabolic or mixed type problems The key ingredient of a stable dG scheme is anappropriate choice of numerical flux and our particular choice has been motivated by theanalysis presented in Appendix C When used to evolve the Schwarzschild solution in Kerr-Schild coordinates our numerical implementation of the BSSN system (4) is robustly stableand converges to the analytic solution exponentially with increased polynomial order Byapproximating the spatially second-order form of the BSSN system we have not introducedextra fields which are evolved Evolved auxiliary fields result in new constraints which mayspoil stability Our main goal has been stable evolution of the spherically reduced BSSNsystem as a first step towards understanding how a discontinuous Galerkin method might be

20

10minus9

10minus7

10minus5

∆Arrinfin

10minus9

10minus7

10minus5

∆Kinfin

0 2000 4000 6000 8000 10000

10minus9

10minus7

10minus5

max(∆Brinfin ∆Γrinfin)

t

N = 11N = 14N = 17N = 20

FIG 4 Spectral convergence of solution for M = 1 Kerr-Schild initial data

Timestep choices are described in the caption for Fig 3 In the title headings for example

∆Arr equiv (Arr)numer minus (Arr)exact

applied to the full BSSN system Towards that goal we now discuss treatment of singularitiesand generalization of the described dG method to higher space dimension

To deal with the fixed Schwarzschild singularity we have used excision which is easy inthe context of the spherically reduced BSSN system However excision for the binary blackhole problem in full general relativity requires attention to the technical challenge of hori-zon tracking State-of-the-art BSSN codes avoid such complication relying instead on themoving-puncture technique While the moving-puncture technique does involve mild centralsingularities it may still prove amenable to spectral methods Indeed spectral methods fornon-smooth problems is well-developed in both theory and for complex applications Sincethe moving-puncture technique can be performed in spherical symmetry [60] a first-steptoward a spectral moving-puncture code would be to implement a moving puncture withthe nodal dG method described here Such an implementation may adopt Legendre-Gauss-Radau nodes on the innermost subdomain thereby ensuring that the physical singularitydoes not lie on a nodal point (in much the same way finite difference codes use a staggeredgrid) Beyond traditional excision and moving punctures one might also retain a smoothproblem for which spectral methods are especially well-suited via the turducken approach[87 88] to singularities

21

10minus11

10minus9

10minus7

10minus5

∆χinfin

10minus11

10minus9

10minus7

10minus5

max(∆grrinfin ∆gθθinfin)

0 2000 4000 6000 8000 10000

10minus11

10minus9

10minus7

10minus5

max(∆αinfin ∆βrinfin)

t

N = 11N = 14N = 17N = 20

FIG 5 Spectral convergence of solution violations for M = 1 Kerr-Schild initial

data See the caption of Fig 4 for details

Discontinuous Galerkin methods for hyperbolic problems in two and three space dimen-sions are well-developed A generalization of the method described here to three-dimensionsand the full BSSN system would likely rely on an unstructured mesh Appropriate local poly-nomial expansions for the subelements are well-understood as are choices for the numericalfluxes which would now live on two-dimensional faces rather than single points Whether ornot it would ultimately prove successful generalization of our dG method to a higher dimen-sion would rely on an established conceptual framework Further computational advancesof relevance to a generalization of our dG method to the full BSSN system (possibly includ-ing matter) may include mesh hp-adaptivity local timestepping shock capturing and slopelimiting techniques [66] Moreover recent work [89] indicates that enhanced performancewould be expected were our scheme implemented on graphics processor units

VI ACKNOWLEDGMENTS

We thank Nick Taylor for discussions about treating second-order operators with spectralmethods Benjamin Stamm for helping polish up a few parts of Appendix C Khosro Shah-bazi for discussions on LDG and IP methods We also acknowledge helpful conversationswith David Brown and Manuel Tiglio about the BSSN system and previous work on its nu-

22

merical implementation We gratefully acknowledge funding through grants DMS 0554377and DARPAAFOSR FA9550-05-1-0108 to Brown University and NSF grant PHY 0855678to the University of New Mexico

Appendix A Hyperbolicity of the first-order system

This appendix analyzes the matrix A(u) appearing in (10) in order to construct thecharacteristic fields (14) In matrix form the sector (8) of the principal part of (10) reads asfollows

partt

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

=

βr 0 minus 4λα3grr

0 0 λβr

6(grr)2λβr

3gθθgrr0 4λ

3grr

0 βr 0 23grrαχ

13α minus αχ

3grr

αχ3gθθ

minus23χ 0

0 0 βr 0 0 0 0 minus χgrr

0

0 0 minus 4α3grr

βr 0 βr

6(grr)2βr

3gθθgrr0 4

3grr

0 0 23αχ 0 βr minus βrχ

3grrminus2βrχ

3gθθ0 minus2

3χ

0 minus2α 0 0 0 23βr minus2grrβr

3gθθ0 4

3grr

0 gθθαgrr

0 0 0 minusgθθβr

3grr13βr 0 minus2

3gθθ

0 0 minus2α 0 0 0 0 βr 0

34

0 0 0 0 0 0 0 βr

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

prime

(A1)which defines the matrix A(u) appearing in (9) and so also the matrix A(u) in (10) Notethat in the last equation the matrix within the square brackets is minusA(u) For certainconfigurations of u and λ the system (10) is strongly hyperbolic [76] that is A(u) hasa complete set of eigenvectors and real eigenvalues Indeed five eigenpairs of A(u) aretrivially recovered upon inspection of A(u)rsquos leading 5times5 diagonal block These correspondto eigenvalue 0 and the left eigenspace ξj = eTj 1 le j le 5 where ej are the canonical

basis vectors Since each component of u arises as eTj W each is also a characteristic field

The remaining nine eigenpairs are determined by A(u) The eigenvalues of A(u) are

micro1 = 0 micro23 = minusβr microplusmn4 = minusβr plusmn

radic

2αχ

grr microplusmn

5 = minusβr plusmn α

radic

χ

grr microplusmn

6 = minusβr plusmnradic

λ

grr

(A2)

23

and the corresponding left eigenvectors are

x1 = (0 0 0 0 0 gθθ 2grr 0 0) (A3a)

x2 =

(

0 0 0 grr2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3b)

x3 =

(

grrλ 0 0 0

2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3c)

xplusmn4 =

(

0 0plusmnradic

2αgrrχ

0 0 0 0 1 0

)

(A3d)

xplusmn5 =

(

0∓ 3radicgrrχ

plusmn2

radic

grrχ 2grr

1

χminus 1

grr1

gθθ 0 0

)

(A3e)

xplusmn6 =

(

minus3

4

grrλ 0plusmn α

radicλgrr

(2αχminus λ) 0 0minus βr

8(βrgrr ∓radicλgrr)

minus βrgrr

4gθθ(βrgrr ∓radicλgrr)

αχ

(2αχminus λ)plusmnradic

grrλ

)

(A3f)

where for example xplusmn5 A(u) = microplusmn5 x

plusmn5 Assuming that grr gθθ χ and α are everywhere strictly

positive the eigenvalues are real and the eigenvectors are linearly independent providedthat (15) holds These eigenvectors are easily extended to eigenvectors of A(u) e g asxplusmn6 rarr (01times5 x

plusmn6 ) Then for example the characteristic field

Xplusmn6 equiv (01times5 x

plusmn6 )W = xplusmn6WvQ (A4)

and similarly Xplusmnj = xplusmnj WvQ for j = 4 5 and Xk = xkWvQ for k = 1 2 3 The characteristic

speeds for these fields are microk and microplusmnj With this convention the speeds listed in Table I

correspond to the Xk and Xplusmnj in (14)

Appendix B Schwarzschild solution in conformal Kerr-Schild coordinates

In Kerr-Schild coordinates here the system directly related to incoming Eddington-Finkelstein null coordinates the line element for the Schwarzschild solution reads

ds2 = minusα2dt2 + (1 + 2MR)(dR + βRdt)2 +R2dθ2 +R2 sin2 θdφ2 (B1)

where R is the area radius α = (1 + 2MR)minus12 is the lapse and βR = 2M(R + 2M) isthe shift vector The physical spatial metric gab is the spatial part of this line element

To define the corresponding solution to the BSSN system we use equation gab = χgab todefine the following relationship between line elements

dr2 + r2(dθ2 + sin2 θdφ2) = χ[(1 + 2MR)dR2 +R2dθ2 +R2 sin2 θdφ2] (B2)

so that

χ

(

1 +2M

R

)(

dR

dr

)2

= 1 χR2 = r2 (B3)

Then we have(

1 +2M

R

)12dR

R=dr

r (B4)

24

with integration yielding

r =R

4

(

1 +

radic

1 +2M

R

)2

e2minus2radic

1+2MR (B5)

where the constant of integration has been chosen so that the R r rarr infin limits are consistentThe second relation in (B3) shows that

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR =rArr χminus4 =2eradic

1+2MRminus1

1 +radic

1 + 2MR (B6)

The extrinsic curvature tensor is specified by the expression forK given in (B9h) the identityK = KR

R + 2Kθθ and

Kθθ =

(

1 +2M

R

)minus122M

R2 (B7)

Since KRR = Kr

r we compute that

Krr = K minus 2Kθ

θ = minus(

1 +2M

R

)minus12(R +M

R + 2M

)

2M

R2 (B8)

Next since Krr = grrKrr = χminus1Kr

r we have Krr = Arr+

13grrK This implies Arr = Kr

r minus 13K

from which we get (B9g) In all we have

α =

(

1 +2M

R

)minus12

(B9a)

βr = βR dr

dR= χ12

(

1 +2M

R

)minus122M

R(B9b)

grr = 1 (B9c)

gθθ = r2 = χR2 (B9d)

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR (B9e)

Br = 0 (B9f)

Arr = minus(

1 +2M

R

)minus124M

3R2

(

2R + 3M

R + 2M

)

(B9g)

K =

(

1 +2M

R

)minus32(

1 +3M

R

)

2M

R2(B9h)

Γr = minus2

r= minus 2

χ12R (B9i)

To differentiate these expressions with respect to r we use the identity

dR

dr= χminus12

(

1 +2M

R

)minus12

(B10)

along with the chain rule

25

Appendix C Stability of the model system

The following stability analysis for the model system (53) has been inspired by [78 79]After dropping all nonlinear source terms the system (53) becomes

parttu = uprime + av (C1a)

parttv = uprimeprime + vprime (C1b)

This section analyzes the stability of (C1) considering both the continuum system itself aswell as its semi-discrete dG approximation The latter analysis offers some insight into theempirically observed stability of our dG scheme for the spherically reduced BSSN equations

1 Analysis for a single interval

Throughout we work with the L2 inner product and norm

(f g)D =

int

D

fg fD =radic

(f f)D (C2)

where D is the spatial coordinate interval (here D may represent a subdomain Dk or the

whole domain Ω) and we have suppressed all integration measures For the continuummodel we will establish the following estimate

uprime(T middot)2D+ av(T middot)2

Dle C(T )

(

uprime(0 middot)2D+ av(0 middot)2

D

)

(C3)

where the time-dependent constant C(T ) is determined solely by the choice of boundaryconditions To show (C3) we first change variables with v =

radicav thereby rewriting (C1)

in the following symmetric form

parttu = uprime +radicav (C4a)

parttv =radicauprimeprime + vprime (C4b)

Equations (C4ab) then imply

1

2partt

int

D

(uprime)2 =

int

D

uprime(uprimeprime +radicavprime) =

int

D

radicauprimevprime +

1

2

int

partD

(uprime)2 (C5a)

1

2partt

int

D

(v)2 =

int

D

v(radicauprimeprime + vprime) = minus

int

D

radicauprimevprime +

1

2

int

partD

(v2 + 2radicauprimev) (C5b)

Here vvprime and uprimeuprimeprime have been expressed as exact derivatives and then integrated to boundaryterms the second equation employs an extra integration by parts and with only one spacedimension

int

partDdenotes a difference of endpoint evaluations Addition of Eqs (C5ab) gives

1

2partt

int

D

[

v2 + (uprime)2]

=1

2

int

partD

[

v2 + (uprime)2 + 2radicauprimev

]

(C6)

Substitutions with the identities

[

v2 + (uprime)2]

=1

2

[

(v + uprime)2 + (v minus uprime)2]

2uprimev =1

2

[

(v + uprime)2 minus (v minus uprime)2]

(C7)

26

and replacements to recover the original variable v = vradica yield

1

2partt

int

D

[

av2 + (uprime)2]

=1 +

radica

4

int

partD

(radicav + uprime)2 +

1minusradica

4

int

partD

(radicav minus uprime)2 (C8)

From (C8) we deduce that the time-dependent constant C(T ) in (C3) must satisfy

∣

∣

∣

∣

∣

∣

1 +

int T

0

[

1+radica

2

int

partD(radicav + uprime)2 + 1minus

radica

2

int

partD(radicav minus uprime)2

]

dt

uprime(0 middot)2 + av(0 middot)2

∣

∣

∣

∣

∣

∣

le C(T ) (C9)

For periodic boundary conditions we may choose C(T ) = 1 Moreover if a ge 1 anduprime = minusradic

av is specified at partD+ then uprime(t middot)2 + av(t middot)2 decaysStill working on a single interval (subdomain) we now consider the semi-discrete scheme

for (C4) i e (57) with all nonlinear source terms dropped and with v replaced by vradica

Derivation of a formula analogous to (C8) is our first step toward establishing L2 stabilityof the semi-discrete scheme While (57) features vectors for example u(t) taking values atthe Legendre-Gauss-Lobatto nodal points here we work with the numerical solution as apolynomial for example uh(t x) These two representations are related by the Lagrange in-terpolating polynomials for the nodal set here taken to span both the space of test functionsand the space of basis functions Our scheme is

int

Dk

ψparttuh =

int

Dk

ψ(Qh +radicavh) (C10a)

int

Dk

ξparttvh = minusint

Dk

ξprime(radicaQh + vh) +

int

partDk

ξ(radicaQlowast + vlowast) (C10b)

int

Dk

ϕQh =

int

Dk

ϕuprimeh +

int

partDk

ϕ (ulowast minus uh) (C10c)

where ψ ξ and ϕ are polynomial test functions These test functions are arbitrary exceptthat they must be degree-N polynomials In (C10) the variables uh vh and Qh should alsocarry a superscript k but we have suppressed this Derivation of a formula analogous to(C8) is complicated by the fact that Qh is not evolved Nevertheless at a given instant t wecan assemble Qh from (C10c)

Mimicking the calculation (C5b) from the continuum case we first use (C10b) with ξ = vhto write

1

2partt

int

Dk

v2h = minusint

Dk

(radicaQh + vh)v

primeh +

int

partDk

(radicaQlowast + vlowast)vh

= minusint

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2(radicaQlowast + vlowast)vh minus v2h

]

(C11)

The right-hand side of (C5a) is analogous to

1

2partt

int

Dk

Q2h =

int

Dk

QhparttQh (C12)

However since Qh is not evolved the term parttQh must be given a suitable interpretationOn the right hand side of (C10c) only uh u

primeh and u

lowast necessarily depend on time since the

27

test function ϕ need not be time-dependent Furthermore ulowast is explicitly given as a linearcombination of uh as seen in Eq (C21c) below Choosing ϕ = ℓj taking the time derivativeof (C10c) and appealing to the commutivity of mixed partial derivatives we therefore arriveat

int

Dk

ℓjparttQh =

int

Dk

ℓj(parttuh)prime +

int

partDk

ℓj(

(parttu)lowast minus parttuh

)

(C13)

where (parttu)lowast depends on parttuh in precisely the same way that ulowast depends on uh We have

written ℓj rather than ϕ in the last equation to emphasize that the result also holds for anylinear combination of ℓj (for example ϕ) and even for time-dependent combinations SinceQh is itself such a combination we obtain

1

2partt

int

Dk

Q2h =

int

Dk

Qh(parttuh)prime +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

Qh(Qprimeh +

radicavprimeh) +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2((parttu)lowast minus parttuh)Qh +Q2

h

]

(C14)

Addition of (C11) and (C14) gives

1

2partt

int

Dk

(Q2h + v2h) =

1

2

int

partDk

[

Q2h minus v2h + 2(

radicaQlowast + vlowast)vh + 2((parttu)

lowast minus parttuh)Qh

]

(C15)

the aforementioned analog of (C8) This formula holds on a single subdomain Dk and we

now combine multiple copies of it one for each value of k

2 Analysis for multiple intervals

To facilitate combination of (C15) over all k we change notation At every subdomaininterface Ik+12 equiv partDk cap partDk+1 let the superscripts L and R denote field values respectivelytaken from the left and right Then the fields evaluated at Ik+12 which belong to D

k will beuLk+12 v

Lk+12 and Q

Lk+12 while those belonging to D

k+1 will be uRk+12 vRk+12 and Q

Rk+12

However at Ikminus12 the values taken from D

k are uRkminus12 vRkminus12 and QR

kminus12 Note that we

have also replaced the subscript h denoting a numerical solution with k plusmn 12 denotingthe location of the endpoint value of the numerical solution With this notation we define

∆Lα =

1

2

[

(QLα)

2 minus (vLα)2]

+(radic

aQlowastα + vlowastα

)

vLα +[

(parttuα)lowast minus parttu

Lα

]

QLα (C16)

and similarly for ∆Rα The same numerical fluxes appear in both ∆L

α and ∆Rα (ie each

numerical flux takes the same value on either side of an interface) whence fluxes like Qlowastα do

not carry an L or R superscript In terms of these definitions (C15) becomes

1

2partt

int

Dk

(Q2h + v2h) = ∆L

k+12 minus∆Rkminus12 (C17)

28

Summation over all Dk yields

1

2partt

kmaxsum

k=1

int

Dk

(Q2h + v2h) =

kmaxminus1sum

k=1

(

∆Lk+12 minus∆R

k+12

)

+∆Lkmax+12 minus∆R

12

=

kmaxminus1sum

k=1

(

∆Lh minus∆R

h

)∣

∣

Ik+12 +∆Lkmax+12 minus∆R

12 (C18)

We have reverted to h-notation denoting the numerical solution since the LR superscriptsindicate unambiguously the relevant domain used for evaluation at Ik+12

We again seek an estimate of the form

kmaxsum

k=1

(

Qh(T middot)2Dk + avh(T middot)2Dk

)

le C(T )kmaxsum

k=1

(

Qh(0 middot)2Dk + avh(0 middot)2Dk

)

(C19)

that is essentially the same as the one (C3) considered in the continuum case Assume thatthe chosen boundary conditions ensure ∆L

kmax+12 minus ∆R12 is bounded by a time-dependent

constant which does not depend on the numerical parameters N and h (subdomain width)Establishment of stability then amounts to showing that the remaining sum over interfaceterms in (C18) is non-positive whence this remaining sum is consistent with C(T ) le 1although the boundary conditions may give rise to a different bound In fact we will choosethe numerical fluxes such that each individual interface term is non-positive At interfaceIk+12 and in LR notation the jump and average of vh for example are

1

2

(

v+ + vminus)

equiv vh =1

2

(

vLk+12 + vRk+12

)

(C20a)

nminusvminus + n+v+ equiv[[

vh]]

= vLk+12 minus vRk+12 (C20b)

Consider numerical fluxes of the form

Qlowast = Qh minusτQ2

[[

Qh

]]

(C21a)

vlowast = vh minusτv2

[[

vh]]

(C21b)

ulowast = uh minusτu2

[[

uh]]

(C21c)

(parttu)lowast = parttuh minus

τu2

[[

parttuh]]

(C21d)

where (C21c) induces (C21d) and where the penalty parameters τu τv and τQ are realnumbers The fluxes defined in (58) correspond to τu = 1 τv = 1 +

radica and τQ = 0 In

terms of these quantities the kth interface contribution in (C18) is

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

parttuh]]

minus τu2

[[

parttuh]][[

Qh

]]

(C22)

where we have suppressed the k dependence of the right-hand side Now consider the term[[

parttuh]]

Because parttuh and Qh+radicavh are both polynomials of degree N Eq (C10a) implies

29

τu

τ Q

minus1 0 1minus1

minus05

0

05

1

τu

minus1 0 1minus1

minus05

0

05

1

FIG 6 Stable evolutions for the model system For fixed τv = 10minus6 and τv = 1 +radic2

respectively the left and right plots depict stable choices (determined empirically) of τu and τQfor the linear model system (C1) The stable regions are colored black but the jagged edges result

from the discretization of the (τu τQ)-plane

the vector equation parttu = Q +radicav that is pointwise equivalence on the nodal points of

Dk which in turn implies

[[

parttuh]]

=[[

Qh+radicavh]]

Upon substituting this identity into thelast equation we arrive at an expression which features only vh and Qh

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

Qh +radicavh]]

minus τu2

[[

Qh +radicavh]][[

Qh

]]

(C23)

The identities vh[[

vh]]

= 12

[[

v2h]]

and[[

Qh +radicavh]]

=[[

Qh

]]

+radica[[

vh]]

then simplify(C23) to

(∆Lh minus∆R

h )∣

∣

Ik+12 = minusτv2

[[

vh]]2 minus

radica(τu + τQ)

2

[[

Qh

]][[

vh]]

minus τu2

[[

Qh

]]2 (C24)

The role of a penalty parameter is now clear Positive values of τv penalize jumps in vhthrough a negative contribution to the energy Likewise positive values of τu penalizejumps in Qh through a negative contribution to the energy However because the sign of[[

Qh

]][[

vh]]

can be positive or negative only the choice τQ = minusτu yields an expression for(∆L

h minus ∆Rh )|Ik+12 which is manifestly negative for τu ge 0 and τv ge 0 A simple estimate

based on Youngrsquos inequality with ε (that is 2αβ le εminus1α2 + εβ2 where α β ge 0 and ε gt 0)shows that for τQ = 0 the choice τv ge aτu4 also yields stability

Figure 6 depicts certain choices of stable penalty parameters for the linear model systemevolved to tfinal = 1000 (with a = 2 N = 10 and ∆t ≃ 00553) as determined empiricallywith simulations similar to those described in Sec IVA The left plot corresponds to asmall τv = 10minus6 for which the choice τu = 1 τQ = 0 is not stable as expected from thetheoretical analysis However the right plot corresponds to τv = 1 +

radica for which τu = 1

τQ = 0 is stable Motivated by the numerical flux choices (3842) used for the BSSN system

30

(4) we have (as mentioned above) set τu = 1 τv = 1 +radica and τQ = 0 in simulations of

the nonlinear model (53) For the nonlinear model system (53) the theoretically motivatedchoice τQ = minusτu also yields numerically stable evolutions when τu ge 0 and τv ge 0

For the nonlinear systems (4) and (53) we do not attempt a formal stability proofNevertheless the results of this appendix have served as a guide for our choices of penaltyparameters For the BSSN system (4) u v and Q are block indices [cf Eq (6)] Similarto the model problem we have penalized Q with τu = 1 with τv chosen large enough toheuristically overcome the cross-terms of indefinite size that arise from τQ = 0 (we interpretequations like τu = 1 componentwise) An analogous choice ldquoτQ = minusτurdquo for the BSSNsystem might be possible but would be considerably more complicated Indeed such achoice likely entails a matrix of penalty parameters but we do not give the details here

[1] F Pretorius Evolution of Binary Black-Hole Spacetimes Phys Rev Lett 95 121101 (2005)

(4 pages)

[2] F Pretorius Simulation of binary black hole spacetimes with a harmonic evolution scheme

Class Quantum Grav 23 S529-S552 (2006)

[3] M Campanelli C O Lousto P Marronetti and Y Zlochower Accurate Evolutions of Or-

biting Black-Hole Binaries without Excision Phys Rev Lett 96 111101 (2006) (4 pages)

[4] J G Baker J Centrella D-I Choi M Koppitz and J van Meter Gravitational-Wave

Extraction from an Inspiraling Configuration of Merging Black Holes Phys Rev Lett 96

111102 (2006) (4 pages)

[5] M Campanelli C O Lousto and Y Zlochower Last orbit of binary black holes Phys Rev D

73 061501(R) (2006) (5 pages)

[6] F Herrmann I Hinder D Shoemaker and P Laguna Unequal mass binary black hole plunges

and gravitational recoil Class Quantum Grav 24 S33-S42 (2007)

[7] P Diener F Herrmann D Pollney E Schnetter E Seidel R Takahashi J Thornburg and

J Ventrella Accurate Evolution of Orbiting Binary Black Holes Phys Rev Lett 96 121101

(2006) (4 pages)

[8] M A Scheel H P Pfeiffer L Lindblom L E Kidder O Rinne and S A Teukolsky Solving

Einsteinrsquos Equations With Dual Coordinate Frames Phys Rev D 74 104006 (2006)

[9] U Sperhake Binary black-hole evolutions of excision and puncture data Phys Rev D 76

104015 (2007) (20 pages)

[10] B Brugmann J A Gonzalez M Hannam S Husa U Sperhake and W Tichy Calibration

of moving puncture simulations Phys Rev D 77 024027 (2008) (25 pages)

[11] P Marronetti W Tichy B Brugmann J Gonzalez M Hannam S Husa and U Sperhake

Binary black holes on a budget simulations using workstations Class Quantum Grav 24

S43-S58 (2007)

[12] Z B Etienne J A Faber Y T Liu S L Shapiro and T W Baumgarte Filling the holes

Evolving excised binary black hole initial data with puncture techniques Phys Rev D 76

101503(R) (2007) (5 pages)

[13] B Szilagyi D Pollney L Rezzolla J Thornburg and J Winicour An explicit harmonic code

for black-hole evolution using excision Class Quantum Grav 24 S275-S293 (2007)

[14] L Boyle M Kesden and S Nissanke Binary-Black-Hole Merger Symmetry and the Spin

Expansion Phys Rev Lett 100 151101 (2008) (4 pages)

31

[15] M A Scheel M Boyle T Chu L E Kidder K D Matthews and H P Pfeiffer High-

accuracy waveforms for binary black hole inspiral merger and ringdown Phys Rev D 79

024003 (2009) (14 pages)

[16] M Hannam Status of black-hole-binary simulations for gravitational-wave detection

Class Quantum Grav 26 114001 (2009) (21 pages)

[17] I Hinder The current status of binary black hole simulations in numerical relativity

Class Quantum Grav 27 114004 (2010) (19 pages)

[18] B P Abbott et al LIGO The Laser Interferometer Gravitational-Wave Observatory

Rep Prog Phys 72 076901 (2009) (25 pages)

[19] F Acernese et al The status of VIRGO Class Quantum Grav 23 S63-S69 (2006)

[20] F Acernese et al Virgo status Class Quantum Grav 25 184001 (2008) (9 pages)

[21] R Arnowitt S Deser and C W Misner The Dynamics of General Relativity in Gravitation

an introduction to current research edited by L Witten (Wiley New York 1962) Chapter 7

227-265 Also available as gr-qc0405109

[22] J W York Jr Kinematics and Dynamics of General Relativity in Sources of Gravitational

Radiation edited by L L Smarr (Cambridge University Press Cambridge 1979)

[23] L E Kidder M A Scheel and S A Teukolsky Extending the lifetime of 3D black hole

computations with a new hyperbolic system of evolution equations Phys Rev D 64 064017

(2001) (13 pages)

[24] G Calabrese J Pullin O Sarbach and M Tiglio Convergence and stability in numerical

relativity Phys Rev D 66 041501(R) (2002) (4 pages)

[25] S Frittelli and O Reula On the Newtonian Limit of General Relativity Commun Math

Phys 166 221-235 (1994)

[26] M Shibata and T Nakamura Evolution of three-dimensional gravitational waves Harmonic

slicing case Phys Rev D 52 5428-5444 (1995)

[27] Y Choquet-Bruhat and J W York Geometrical well posed systems for the Einstein equations

C R Acad Sci Paris t 321 Serie I 1089-1095 (1995)

[28] A Abrahams A Anderson Y Choquet-Bruhat and J W York Jr Einstein and Yang-Mills

Theories in Hyperbolic Form without Gauge Fixing Phys Rev Lett 75 3377-3381 (1995)

[29] C Bona J Masso E Seidel and J Stela New Formalism for Numerical Relativity

Phys Rev Lett 75 600-603 (1995)

[30] M H P M van Putten and D M Eardley Nonlinear wave equations for relativity

Phys Rev D 53 3056-3063 (1996)

[31] S Frittelli and O A Reula First-Order Symmetric Hyperbolic Einstein Equations with Ar-

bitrary Fixed Gauge Phys Rev Lett 76 4667-4670 (1996)

[32] H Friedrich Hyperbolic reductions for Einsteinrsquos equations Class Quantum Grav 13 1451-

1469 (1996)

[33] F B Estabrook R S Robinson H D Wahlquist Hyperbolic equations for vacuum gravity

using special orthonormal frames Class Quantum Grav 14 1237-1247 (1997)

[34] M S Iriondo E O Leguizamon and O A Reula Einsteinrsquos Equations in Ashtekarrsquos Vari-

ables Constitute a Symmetric Hyperbolic System Phys Rev Lett 79 4732-4735 (1997)

[35] A Anderson Y Choquet-Bruhat and J W York Einstein-Bianchi hyperbolic system for

general relativity Topol Meth Nonlinear Anal 10 353-373 (1997)

[36] T W Baumgarte and S L Shapiro Numerical integration of Einsteinrsquos field equations

Phys Rev D 59 024007 (1998) (7 pages)

[37] M A G Bonilla Symmetric hyperbolic systems for Bianchi equations Class Quantum

5

spherically symmetric versions of the standard ldquo1+logrdquo and ldquoΓ-driverrdquo conditions listed inEqs (1) and (2) of [60] However we have introduced the following minor modificationsFirst (parttα)0 designates a constant term which ensures that the right-hand side of the αevolution equation (4a) vanishes at the initial time This source term as well as the analo-gous terms appearing in the evolution equations (4bc) for βr and Br are needed to enablea static evolution of the Schwarzschild solution in Kerr-Schild coordinates Second the pa-rameter λ (perhaps with functional dependence) modifies the hyperbolicity of the first-ordersystem [80] The damping parameter η typically appears in standard versions of these gaugeevolution equations (See Sections IIC and IVB for further discussions) For this BSSNsystem we have three constraints the Hamiltonian constraint H the momentum constraintMr and the constraint Gr resulting from the definition of the conformal connection Γr Inspherical symmetry these constraints are written as follows

H = minus3A2rr

2g2rr+

2K2

3minus 5(χprime)2

2χgrr+

2χprimeprime

grr+

2χ

gθθminus 2χgprimeprimeθθgrrgθθ

+2χprimegprimeθθgrrgθθ

+χgprimerrg

primeθθ

g2rrgθθminus χprimegprimerr

g2rr+χ(gprimeθθ)

2

2grrg2θθ

(5a)

Mr =Aprime

rr

grrminus 2K prime

3minus 3Arrχ

prime

2χgrr+

3Arrgprimeθθ

2grrgθθminus Arrg

primerr

g2rr(5b)

Gr = minus gprimerr2g2rr

+gprimeθθgrrgθθ

+ Γr (5c)

These expressions are the ones listed by Brown in [60] Eqs (4ef) also ensure that thedeterminant factor g sin4 θ = grr(gθθ)

2 remains fixed throughout an evolution

B Abstract expressions of the system

We define the following vectors built with system variables

u =

χgrrgθθαβr

v =

Br

Arr

KΓr

Q =

χprime

gprimerrgprimeθθαprime

βrprime

(6)

Introduction of Qmight seem unnecessary at this stage but proves useful in the constructionof our discontinuous Galerkin scheme In terms of the vectors u v and Q we further define

Wuv =

(

uv

)

WvQ =

(

vQ

)

W =WuQ =

uvQ

(7)

Here we have introduced ldquocolon notationrdquo [81] to represent (sub)vectors and (sub)matricesalthough we employ the notation over block rather than individual elements In the first-order version of the system (4) the components of Q are promoted to independent fields in

6

which case the corresponding principal part features

parttBr = βrBr prime minus 4λα

3grrK prime +

4λ

3grrQprime

βr +λβr

6(grr)2Qprime

grr +λβr

3gθθgrrQprime

gθθ(8a)

parttArr = βrAprimerr +

2

3grrαχΓ

r prime +1

3αQprime

χ minus2

3χQprime

α minus αχ

3grrQprime

grr +αχ

3gθθQprime

gθθ(8b)

parttK = βrK prime minus χ

grrQprime

α (8c)

parttΓr = βrΓr prime minus 4αK prime

3grr+

4

3grrQprime

βr +βr

6(grr)2Qprime

grr +βr

3gθθgrrQprime

gθθ(8d)

parttQχ = βrQprimeχ +

2

3αχK prime minus βrχ

3grrQprime

grr minus2βrχ

3gθθQprime

gθθminus 2

3χQprime

βr (8e)

parttQgrr =2

3βrQprime

grr +4

3grrQ

primeβr minus 2αAprime

rr minus2grrβ

r

3gθθQprime

gθθ(8f)

parttQgθθ =1

3βrQprime

gθθ+gθθα

grrAprime

rr minusgθθβ

r

3grrQprime

grr minus2

3gθθQ

primeβr (8g)

parttQα = βrQprimeα minus 2αK prime (8h)

parttQβr = βrQprimeβr +

3

4Br prime (8i)

where all lower-order terms on the right-hand side have been dropped This sector of prin-cipal parts of the first-order system has the form

parttWvQ + A(u)W primevQ = 0 (9)

where (minus) the explicit form of the 9-by-9 matrix A(u) is given below in (A1) Thefirst-order version of (4) takes the nonconservative form

parttW +A(u)W prime = S(W ) A(u) =

(

05times5 05times9

09times5 A(u)

)

(10)

where S(W ) is a vector of lower order terms built with all components of W Partition ofA(u) = A(u)vQvQ into blocks corresponding to the v and Q sectors yields

A(u) =

(

A(u)vv A(u)vQ

A(u)Qv A(u)QQ

)

(11)

Using these blocks we then define the 9-by-9 matrix

A(u) = A(u)uvvQ =

(

05times4 05times5

A(u)vv A(u)vQ

)

(12)

and express (4) as

parttWuv + A(u)W primevQ = S(W ) (13a)

Q = uprime (13b)

where S(W ) = S(W )uv

7

field speed

X1 micro1 = 0

X23 micro23 = minusβr

Xplusmn4 microplusmn

4 = minusβr plusmnradic

2αχgrr

Xplusmn5 microplusmn

5 = minusβr plusmn αradic

χgrr

Xplusmn6 microplusmn

6 = minusβr plusmnradic

λgrr

TABLE I Characteristic speeds These speeds are the eigenvalues listed in (A2)

C Hyperbolicity and characteristic fields

Although our numerical scheme deals directly with the second-order spatial operatorsappearing in (4) we first consider the hyperbolicity of the corresponding first-order system(10) The characteristic fields and their speeds are found by instantaneously ldquofreezingrdquothe fields u in A(u) to some value u0 corresponding to a linearization around a uniformstate Below we continue to write u for simplicity with the understanding that u is reallythe background solution u0 Of primary interest is the range of u0 for which the system isstrongly hyperbolic [73ndash76]

Appendix A shows that the characteristic fields corresponding to (4) are as follows (i)all components of u (each with speed 0) and (ii) the fields

X1 = gθθQgrr + 2grrQgθθ (14a)

X2 = grrΓr +

2

χQχ minus

1

2grrQgrr minus

1

gθθQgθθ (14b)

X3 =grrλBr +

2

χQχ minus

1

2grrQgrr minus

1

gθθQgθθ (14c)

Xplusmn4 = plusmn

radic

2αgrrχ

K +Qα (14d)

Xplusmn5 = ∓ 3radic

grrχArr plusmn 2

radic

grrχK + 2grrΓ

r +1

χQχ minus

1

grrQgrr +

1

gθθQgθθ (14e)

Xplusmn6 = minus3

4

grrλBr plusmn α

radicλgrr

(2αχminus λ)K minus βr

8(βrgrr ∓radicλgrr)

Qgrr

minus βrgrr

4gθθ(βrgrr ∓radicλgrr)

Qgθθ +αχ

(2αχminus λ)Qα plusmn

radic

grrλQβr (14f)

with the speeds listed in Table I To ensure strong hyperbolicity we must necessarily require

λ gt 0 (βr)2grr minus λ 6= 0 2αχminus λ 6= 0 (15)

as shown in in Appendix A where further conditions are also given When λ = 1 thehyperbolicity condition of Ref [60] is recovered In fact the system could be recast assymmetric hyperbolic Indeed as it involves one spatial dimension the relevant symmetrizercan be constructed via polar decomposition of the diagonalizing similiarity transformationHowever we will not exploit this possibility

8

This system admits an inner excision boundary provided

βr ge max

(

radic

2αχ

grr

radic

α2χ

grr

radic

λ

grr

)

(16)

holds at the inner boundary This condition ensures each characteristic field has a nonposi-tive speed at the inner boundary and therefore the inner boundary is an excision boundaryat which no boundary conditions are needed The extra flexibility afforded by the param-eter λ could be used to maintain rigorous hyperbolicity by moving the points at whichthe conditions in (15) are violated outside of the computational domain Furthermore forλ = 1 Eq (16) conceivably fails or is only satisfied close to r = 0 where field gradients are

prohibitively large The troublesome X+6 gauge mode has a positive speed minusβr +

radic

λgrrIndeed for the conformally flat Kerr-Schild system considered in section IVB an inner ex-cision boundary is only possible provided λ is small enough

The transformation (14) can be inverted in order to express the fundamental fields interms of the characteristic fields

Br = minus1

6

λ

grrgθθ

[

(βr)2

(βr)2grr minus λ

]

X1 +2

3

λαχ

grr(2αχminus λ)(X+

4 +Xminus4 )minus

2

3

λ

grr(X+

6 +Xminus6 )

(17a)

Arr =1

3

radic

grrχ

2α(X+

4 minusXminus4 )minus

radicgrrχ

6(X+

5 minusXminus5 ) (17b)

K =

radic

χ

8αgrr(X+

4 minusXminus4 ) (17c)

Γr = minus1

6

1

grrgθθ

[

(βr)2

(βr)2grr minus λ

]

X1 +1

grr(X2 minusX3) +

2

3

αχ

grr(2αχminus λ)(X+

4 +Xminus4 )

minus 2

3

1

grr(X+

6 +Xminus6 ) (17d)

Qχ =1

12

χ

grrgθθ

[

4(βr)2grr minus 3λ

(βr)2grr minus λ

]

X1 +χ

2X3 minus

1

3

αχ2

(2αχminus λ)(X+

4 +Xminus4 ) (17e)

+χ

3(X+

6 +Xminus6 ) (17f)

Qgrr =2(βr)2grr minus 3λ

6gθθ((βr)2grr minus λ)X1 +

4

3grrX2 minus grrX3 +

2

3

αχgrr(2αχminus λ)

(X+4 +Xminus

4 )

minus 1

3grr(X

+5 +Xminus

5 )minus2

3grr(X

+6 +Xminus

6 ) (17g)

Qgθθ =

[

1

4grr+

(βr)2

12((βr)2grr minus λ)

]

X1 minus2

3gθθX2 +

1

2gθθX3 minus

1

3

αχgθθ(2αχminus λ)

(X+4 +Xminus

4 )

+1

6gθθ(X

+5 +Xminus

5 ) +1

3gθθ(X

+6 +Xminus

6 ) (17h)

Qα =1

2(X+

4 +Xminus4 ) (17i)

Qβr =βrλ

8grrgθθ((βr)2grr minus λ)X1 minus

λ

(2αχminus λ)

radic

αχ

8grr(X+

4 minusXminus4 ) +

1

2

radic

λ

grr(X+

6 minusXminus6 )

(17j)

9

We will refer to this inverse transformation when discussing outer boundary conditions forour numerical simulations in Sec IVB

III DISCONTINUOUS GALERKIN METHOD

This section describes the nodal discontinuous Galerkin method used to numerically solve(4) We adopt a method-of-lines strategy and here describe the relevant semi-discrete schemewhile leaving the temporal dimension continuous To approximate (4) we follow the generalprocedure first introduced in Ref [82] Our approach defines local auxiliary variables Q = uprimeand rewrites the spatially second-order system (4) as the first-order system (13a) Once weuse (13b) to eliminate Q from (13a) we recover the primal equations (4) The auxiliaryvariable approach was later generalized and coined the local discontinuous Galkerin (LDG)method in Ref [77] We may refer to our particular scheme as an LDG method but notethat many variations exist in the literature We stress that in LDG methods Q is not evolvedand is introduced primarily to assist in the construction of a stable scheme

Equations (12) and (13a) imply that the physical flux function is

F (W ) =

(

Fu(W )

Fv(W )

)

equiv A(u)WvQ =

(

05times1

f(W )

)

f =

fBr

fArr

fKfΓ

(18)

Only the evolution equations for Br Arr K and Γr give rise to non-zero components in F and we have collected these non-zero components into a smaller vector f = Fv Inspectionof (8) determines these components For example from (8c) we find

fK = minusβrK +χ

grrQα (19)

A Local approximation of the system (13)

Our treatment closely follows [83] but with the equations and notations relevant forthis paper Our computational domain Ω is the closed r-interval [a b] We cover Ω withkmax gt 1 non-overlapping intervals Dk = [ak bk] where a = a1 b = bkmax and bkminus1 = ak fork = 2 middot middot middot kmax

On each interval Dk we approximate each component of the system vector W by a localinterpolating polynomial of degree N For example

χkh(t r) =

Nsum

j=0

χ(t rkj )ℓkj (r) (20)

approximates χ(t r) Throughout this section approximations are denoted by a subscripth (see [66] for the notation) For example Wh and fh are approximations of W and f Although Q = uprime Qh and uprimeh are not necessarily the same In (20) ℓkj (r) is the jth Lagrange

polynomial belonging to Dk

ℓkj (r) =Nprod

i=0i 6=j

r minus rkirkj minus rki

(21)

10

Evidently the polynomial χkh interpolates χ at rkj To define the nodes rkj consider the

mapping from the unit interval [minus1 1] to Dk

rk(u) = ak + 12(1 + u)(bk minus ak) (22)

and the N+1 Legendre-Gauss-Lobatto (LGL) nodes uj The uj are the roots of the equation

(1minus u2)P primeN(u) = 0 (23)

where PN(u) is the Nth degree Legendre polynomial and the physical nodes are simplyrkj = rk(uj) In vector notation the approximation (20) takes the form

χkh(t r) = χk(t)Tℓk(r) (24)

in terms of the column vectors

χk(t) =[

χ(t rk0) middot middot middot χ(t rkN)]T ℓk(r) =

[

ℓk0(r) middot middot middot ℓkN(r)]T (25)

On each open interval (ak bk) sub Dk and for each component of the equations in (13) we

define local residuals measuring the extent to which our approximations satisfy the originalcontinuum system Dropping the subdomain label k on the polynomials and focusing onthe K equation as a representative example the local residual corresponding to (4h) is

minus(RK)kh equivminus parttKh + (βrK prime)h minus

(

χQprimeα

grr

)

h

+

(

χQgrrQα

2g2rr

)

h

minus(

χQgθθQα

grrgθθ

)

h

+

(

QαQχ

2grr

)

h

+

(

3αA2rr

2g2rr

)

h

+

(

1

3αK2

)

h

(26)

Here for example the expressions 2

(βrK prime)h = βrhK

primeh

(

QαQχ

2grr

)

h

=QαhQχh

2grrh (27)

We similarly construct the remaining eight residuals eg (Rgrr)h and (RΓr)h as well as fiveresiduals corresponding to (13b) For example one of these remaining five is

(RQα)kh equiv minusQαh + αprime

h (28)

Let the kth inner product be defined as

(

u v)

Dk equivint bk

akdru(r)v(r) (29)

and consider the expression (ℓkj (RK)kh)Dk We call the requirement that this inner product

vanish forallj the kth Galerkin condition For each component of the system and for each k

2 At this stage the first expression is generically a polynomial of degree 2N minus 1 and the latter is not a

polynomial The conventions adopted in Eq (27) prove useful while working with the residual However

later on in Sec III C to obtain the final form (47) of the numerical approximation corresponding to (26)

we will replace nonlinear terms with degree-N polynomials

11

there is a corresponding Galerkin condition in total 9kmax(N + 1) equations for (13a) and5kmax(N+1) for (13b) Enforcement of the Galerkin conditions on each D

k will not recover ameaningful global solution since they provide no mechanism for coupling the local solutionson the different intervals Borrowing from the finite volume toolbox we achieve couplingthrough integration by parts on r and introduction of the numerical flux f lowast at the interfacebetween subdomains

In (26) we only need to consider (βrK prime)h and (χQprimeαgrr)h as the other terms comprise a

component of the source vector Sh Using integration by parts we write

(

ℓkj (βrK prime)h

)

Dk =minusint bk

akdr[

(

ℓkjβrh

)primeKh

]

+ (βrhKh) ℓ

kj

∣

∣

∣

bk

ak (30a)

(

ℓkj (χQprimeαgrr)h

)

Dk =minusint bk

akdr

[(

ℓkjχh

grrh

)prime

Qαh

]

+

(

χhQαh

grrh

)

ℓkj

∣

∣

∣

bk

ak (30b)

In these formulas we have retained the domain index k on ℓkj while continuing to suppress iton Kh grrh etc Moreover we have suppressed the r-dependence in all terms on the right-hand side Addition of these formulas along with the definition fKh = minus(βrK)h+(χQαgrr)hgives

(

ℓkj (βrK prime)h minus (χQprime

αgrr)h)

Dk = minusint bk

akdr

[

(

ℓkjβrh

)primeKh minus

(

ℓkjχh

grrh

)primeQαh

]

minus fKhℓkj

∣

∣

∣

bk

ak (31)

In lieu of (31) we will instead work with the replacement

(

ℓkj (βrK prime)h minus (χQprime

αgrr)h)

Dk rarr minusint bk

akdr

[

(

ℓkjβrh

)primeKh minus

(

ℓkjχh

grrh

)primeQαh

]

minus f lowastKℓ

kj

∣

∣

∣

bk

ak (32)

This replacement features a component f lowastK of the numerical flux rather than a component

fKh of the boundary flux The numerical flux is determined by (as yet not chosen) functions3

f lowast = f lowast(W+Wminus) (33)

where for example Wminus is an interior boundary value [either W kh (t a

k) or W kh (t b

k)] of theapproximation defined on D

k andW+ is an exterior boundary value [either W kminus1h (t bkminus1) or

W k+1h (t ak+1)] of the approximation defined on either Dkminus1 or Dk+1 We discuss our choice

of numerical flux in the next subsection We now employ additional integration by parts towrite the above replacement as

(

ℓkj (βrK prime)h minus (χQprime

αgrr)h)

Dk rarrint bk

akdrℓkj

(

βrK prime minus χQprimeα

grr

)

h

+ (fKh minus f lowastK) ℓ

kj

∣

∣

∣

bk

ak (34)

Rather than the exact kth Galerkin condition(

ℓkj (RK)kh

)

Dk = 0 forallj for the K component

of (13) on Dk we will instead strive to enforce

(

ℓkj (RK)kh

)

Dk = (fKh minus f lowastK) ℓ

kj

∣

∣

∣

bk

ak forallj (35)

3 In the context of the dG method here + and minus denote ldquoexteriorrdquo and ldquointeriorrdquo and have no relation to

the plusmn using to denote the characteristic fields and speeds in Table I For characteristic fields and speeds

+ and minus mean ldquoright-movingrdquo and ldquoleft-movingrdquo

12

although our treatment of nonlinear terms will lead to a slight modification of these equations(we return to this issue shortly) The other components of (13a) are treated similarly asare the components of (13b) Recall that for example Qα = αprime Formally using the samedG method to solve for Qα we arrive at the replacement

(

ℓkj (RQα)kh

)

Dk rarrint bk

akdrℓkj (minusQαh + αprime

h)minus (αh minus αlowast) ℓkj

∣

∣

∣

bk

ak (36)

which again features a component αlowast of the numerical flux The auxiliary variables areconstructed and used at each stage of temporal integration We then have

(

ℓkj (RQα)kh

)

Dk = (αh minus αlowast) ℓkj

∣

∣

∣

bk

ak forallj (37)

as the corresponding enforced kth Galerkin condition

B Numerical Flux

To further complete our dG scheme we must specify functional forms for the compo-nents of the numerical flux introduced in the previous section We distinguish between thephysical fluxes (components of f) and the auxiliary fluxes (components of u) arising fromthe definition of the auxiliary variables These choices are not independent as the resultingscheme must be stable and consistent Our choice follows [84] which considered diffusionproblems Additional analysis of this flux choice appears in [66 85]

Let us first consider the numerical fluxes corresponding to the physical fluxes and of theform (33) The numerical flux vector is a function of the system and auxiliary variablesinterior and exterior to a subdomain A common choice for f lowast is

f lowast = fh+τ

2

[[

vh]]

K-component of f lowast f lowastK = fKh+

τ

2

[[

Kh

]]

(38)

where as an example we have also shown the component of f lowast corresponding to the analysisabove Respectively the average and jump across the interface are

fh =1

2

(

f+ + fminus) [[

vh]]

= nminusvminus + n+v+ (39)

Here τ is a position dependent penalty parameter (fixed below) and nminus(n+) is the localoutward pointing normal to the interior (exterior) subdomain The role of τ is to ldquopenalizerdquo(i e yield a negative contribution to the L2 energy norm) jumps across an interface Anappropriate choice of τ will ensure stability and we now provide some motivation for thechoice (41) of τ we make below

Were we treating the fully first-order system (10) the local Lax-Friedrichs flux wouldoften be a preferred choice due to its simplicity [66] In this case the constant ω in thenumerical flux formula Flowast = Fh + 1

2ω[[

Wh

]]

obeys ω ge max∣

∣micro(nablaWF(W ))∣

∣ HereF(W ) = A(u)W the notation micro(middot) indicates the spectral radius of the matrix within andthe max is taken over interior Wminus and exterior W+ states Motivated by (9) we adopt asimilar but simpler prescription substituting the field gradient

nablaWvQA(u)WvQ = A(u) (40)

13

for nablaWF(W ) Precisely we assume the scaling

τ(bk) = τ(ak+1) = τk+12 equiv C middotmax∣

∣micro(

A(u))∣

∣ (41)

where C = O(1) is a constant chosen for stability Larger values of C will result in schemeswith better stability properties whereas too large a value will impact the CFL conditionAt the interface point Ik+12 equiv D

k cap Dk+1 the vector uh has two representations uminus at bk

and u+ at ak+1 The max in (41) is taken over the corresponding two sets of field speedsMore precisely the speeds in Table I are computed for both uminus and u+ and the maximumtaken over all resulting speeds For the auxiliary variables a penalized central flux is usedThe definition with one representative component is

ulowast = uh minus1

2

[[

uh]]

α-component of ulowast αlowast = αh minus1

2

[[

αh

]]

(42)

with similar expressions for the remaining componentsWe stress the following point Since the interior coupling between subdomains is achieved

through the numerical flux forms (41) and (42) the inverse transformation (17) expressingthe fundamental fields in terms of the characteristic fields is not required to achieve thiscoupling On the other hand imposition of physical boundary conditions may still rely on(17) since this transformation allows one to fix only incoming characteristic modes

C Matrix form of the semi-discrete equations

Let us introduce the kth mass and stiffness matrices

Mkij =

int bk

akdrℓki (r)ℓ

kj (r) Sk

ij =

int bk

akdrℓki (r)ℓ

kjprime(r) (43)

These matrices belong to Dk and the corresponding matrices defined on the reference interval[minus1 1] are

Mij =

int 1

minus1

duℓi(u)ℓj(u) Sij =

int 1

minus1

duℓi(u)ℓprimej(u) (44)

where ℓj(u) is the jth Lagrange polynomial determined by the LGL nodes uj on [minus1 1]These matrices are related by Mk

ij =12(bk minus ak)Mij and S

kij = Sij whence only the reference

matrices require computation and storageWe will use the matrices Mk and Sk in obtaining an ODE system from (26) and (35)

Towards this end we first approximate the nonlinear terms (products and quotients) in(26) by degree-N interpolating polynomials Such approximations are achieved throughpointwise representations For example (QαQχgrr)h appears in (26) and is expressed inthe following way [cf footnote 2]

(

QαQχ

grr

)

h

(t r) =

(

QαhQχh

grrh

)

(t r) rarrNsum

j=0

Qαh(t rkj )Qχh(t r

kj )

grrh(t rkj )

ℓkj (r) (45)

Note that the expressions on the right and left are not equivalent due to aliasing error [65]Our vector notation for this replacement will be

(

QαQχ

grr

)

h

(t r) rarr(

QαQχ

grr

)

(t)Tℓk(r) (46)

14

Operations among bold variables are always performed pointwise Making similar replace-ments for all terms in (26) and then carrying out the integrations in (35) which bring inMk and Sk we arrive at

parttK = βrDK minus χDQα

grr

+1

2

χQgrrQα

g2rr

minusχQgθθ

Qα

grrgθθ

+1

2

QαQχ

grr

+3

2

αA2rr

g2rr

+1

3αK2 +Mminus1ℓk (fKh minus f lowast

K)∣

∣

∣

bk

ak (47)

where we have again suppressed the superscript k on all terms except ℓk(r) and the sub-script h is dropped on all boldfaced variables As described in [66] the spectral collocationderivative matrix

(Dk)ij =dℓkjdr

∣

∣

∣

∣

∣

r=rki

(48)

can also be expressed as Dk = (Mk)minus1Sk which appears above Eight other semi-discreteevolution equations are similarly obtained with nine in total (one for each component ofWuv) Additionally we have

Qα = Dα+Mminus1ℓk(αlowast minusαh)∣

∣

∣

bk

ak (49)

as one of the auxiliary equations with five in total (one for each component of Q = WQQ)

D Filtering

Like other nodal (pseudospectral) methods our scheme may suffer from instabilitiesdriven by aliasing error [65] Filtering is a simple yet robust remedy To filter a solu-tion component such as χ we use the modal (as opposed to nodal) representation of thesolution

χkh(t r) =

Nsum

j=0

χ(t rkj )ℓkj (r) =

Nsum

j=0

χkj (t)Pj(r) (50)

where Pj(r) is the jth Legendre polynomial Let ηj = jN and define the filter function

σ(ηj) =

1 for 0 le ηj le NcN

exp(

minus ǫ(

ηjminusNcN

1minusNcN

)2s)

for NcN le ηj le 1(51)

At each timestep we modify our solution component according to

χkh rarr

(

χkh

)filtered=

Nsum

j=0

σ(ηj)χkj (t)Pj(r) (52)

Evidently the modification only affects the top N minus Nc modes and is sufficient to controlthe type of weak instability driven by aliasing [66] The numerical parameters Nc and ǫ areproblem dependent For our simulations we have taken ǫ ≃ minuslog(εmach) = 36 where εmach

is machine accuracy in double precision

15

E Model system

To better illustrate the basic properties of our method we consider a toy model Namelythe following spatially second-order system

parttu = uprime + av minus u3 + g(t x) (53a)

parttv = uprimeprime + vprime minus (u+ v)(uprime)2 + v2u2 + h(t x) (53b)

where a ge 1 is constant and g and h are analytic source terms to be specified In contrastto (6) here u v and Q = uprime are scalars rather than vectors System (53) admits a first-order reduction in which uprime is defined as an extra variable Since this first-order reductionis strongly hyperbolic the spatially second-order system (53) is also strongly hyperbolic byone of the definitions considered in [75] The characteristic fields Xplusmn and speeds microplusmn are

X+ =radicav minus uprime micro+ =

radicaminus 1 Xminus =

radicav + uprime microminus = minus(

radica+ 1) (54)

To construct a local dG scheme for this system we first rewrite it as

parttu = Q+ av minus u3 + g(t x) (55a)

parttv = Qprime + vprime minus (u+ v)Q2 + v2u2 + h(t x) (55b)

Q = uprime (55c)

Evidently f = minus(Q + v) is the v-component of the physical flux vector

F (vQ) equiv(

Fu

Fv

)

=

(

0

f

)

(56)

Note that F has the same structure as (u v)T Borrowing from the presentation for theBSSN system we write the analogous semidiscrete scheme on each subdomain D

k for themodel system

parttu = Q+ av minus u3 + g(t) (57a)

parttv = DQ+Dv minus (u+ v)Q2 + v2u2 + h(t) +Mminus1ℓk(fh minus f lowast)∣

∣

∣

bk

ak (57b)

Q = Du+Mminus1ℓk(ulowast minus uh)∣

∣

∣

bk

ak (57c)

Here we have suppressed the subinterval label k from all variables except for the vectorℓk of Lagrange polynomial values Moreover following the guidelines discussed above thenumerical fluxes are given by

f lowast = fh+1 +

radica

2

[[

vh]]

ulowast = uh minus1

2

[[

uh]]

(58)

Appendix C analyzes the stability of our scheme for a more general numerical flux choiceas applied to (53) with the nonlinear and source terms dropped

IV RESULTS FROM NUMERICAL SIMULATIONS

This section presents results found by numerically solving both the model system (53)and BSSN system (4) with the dG scheme presented in Sec III

16

10minus10

10minus6

10minus2

∆uinfin

10minus10

10minus6

10minus2

∆vinfin

0 500 1000 1500 2000 2500 3000 3500 400010

minus10

10minus6

10minus2

∆(partxu)infin

t

N = 3N = 6N = 9N = 12

FIG 1 Spectral convergence of fields for model PDE Respectively for N = 3 6 9 12

a timestep of ∆t = 00578 00178 00084 00049 has been chosen for stability and accuracy In the

title headings for example ∆u equiv unumer minus uexact

A Simulations of the model system

The semi-discrete scheme (57) has been integrated with the classical fourth-order Runge-Kutta method When integrating this system we have first constructed Q at each Runge-Kutta stage and then substituted into the evolution equations (57ab) for u and v Theproblem has been solved on a computational domain [0 4π] comprised of two subdomainswith a timestep chosen small enough for stability The initial data has been taken from thefollowing exact solution to (53)

uprimeexact(t x) =1

2

[

sin(xminus microminust)minus sin(xminus micro+t)]

(59a)

vexact(t x) =1

2radica

[

sin(xminus microminust) + sin(xminus micro+t)]

(59b)

g(t x) = u3exact (59c)

h(t x) = (uexact + vexact)(uprimeexact)

2 minus v2exactu2exact (59d)

where the speeds microplusmn are found in (54) Specification of the boundary condition at a physicalendpoint amounts to choosing the external state for at the endpoint We have considered

17

8 16 32 64 128

10minus3

10minus2

10minus1

N

∆t

max stable ∆t8N2

FIG 2 Scaling of maximum stable ∆t with N for model PDE

two possibilities (i) the analytic state (Q+ v+) = (Qexact vexact) and (ii) an upwind stateFor example at x = 4π the upwind state is4

Q+ = Qupwind =1

2

[

(Xminus)exact minus (X+)numer

]

v+ = vupwind =1

2radica

[

(Xminus)exact + (X+)numer

]

(60)Either choice of (Q+ v+) leads to similar results and the plots here correspond to theanalytic state Figure 1 clearly shows spectral convergence with increasing polynomial orderN across all fields for the case a = 2 Other values of a including a = 1 for which X+

is a static characteristic field have also been considered with similar results Appendix Cdemonstrates that our proposed scheme for the system (57) with nonlinear and source termsdropped is stable in a semi-discrete sense Nevertheless the fully discrete scheme obtainedvia temporal discretization by the fourth-order Runge-Kutta method is still subject to thestandard absolute stability requirement Namely if microh is any eigenvalue corresponding tothe (linearized) discrete spatial operator then a necessary condition for stability is thatmicroh∆t lies in absolute stability region for Runge-Kutta 4 We here show empirically that theassociated timestep restriction scales like Nminus2 ie ∆t = O(Nminus2) for stability We note thatsuch scaling is welcome in light of the second-order spatial operators which appear in thesystem and suggest a possible worse scaling like Nminus4 Fig 2 plots the maximum stabletimestep for a range of N demonstrating the Nminus2 scaling in line with behavior known fromanalysis of first-order systems [66] This scaling also holds for the BSSN system

4 We remind the reader that unfortunately the plusmn on Xplusmn means something different than the plusmn indicating

exteriorinterior dG states [cf footnote 3]

18

B Simulations of the BSSN system

This subsection documents results for simulations of the unit-mass-parameter (M = 1)Schwarzschild solution (B9) expressed in terms of ingoing Kerr-Schild coordinates Sincethe solution is stationary temporal integration of the semi-discrete scheme has been carriedout with the forward Euler method The r-coordinate domain [04 34] has been split into3 equally spaced subdomains and we have set η = 10 λ = 01 and C = 2 [cf Eq (41)]For all simulations ∆t has been chosen for stability With the chosen λ the inner physicalboundary rmin = 04 is an excision surface At each timestep we have applied an (order2s = 20) exponential filter on the top two-thirds of the modal coefficient set for all fieldsexcept for grr and gθθ For stability we have empirically observed that grr and gθθ must not

be filtered A detailed understanding of this is still lackingIssues related to physical boundary conditions are similar to the one encountered in

Sec IVA for the model problem Similar to before we have retained Eqs (3842) as thechoice of numerical flux even at the endpoints Therefore at an endpoint the specificationof the boundary condition amounts to the choice W+ of external state We have typicallychosen the inner boundary of the radial domain as an excision boundary and in this caseW+ = Wminus is enforced at the inner physical boundary At the outer physical boundaryfor W+ we have again considered two choices (i) Wexact and (ii) Wupwind To enforcechoice (ii) the inverse transformation (17) must be used with incoming characteristic fieldsfixed to their exact values similar to (60) We have tried various versions of choice (ii)and in all cases the resulting simulations have been unstable We therefore present resultscorresponding to choice (i) Although the choice of an analytical external state Wexact

at the outer boundary is stable for our problem such a boundary condition is unlikelyto generalize to more complicated scenarios involving dynamical fields Indeed the issueof outer boundary conditions for the BSSN system is an active area of research with aproper treatment requiring fixation of incoming radiation control of the constraints andspecification of gauge (see Ref [86] for a recent analysis)

For BSSN simulations our main diagnostic is to monitor the Hamiltonian momentumand conformal connection constraints Figure 3 depicts long-time histories of constraintviolations whereas Figs 4 and 5 depict long-time error histories for the individual BSSNfield components From the middle plot in Fig 5 we infer that up to the indicated numericalerror the factor g sin4 θ = grr(gθθ)

2 remains at its initial fixed profile r4 throughout theevolution These figures indicate that the proposed scheme is stable for long times andexhibits spectral converge with increased polynomial order N Similar results are recoveredfromM = 0 Minkowski initial data The stability documented in these plots does not appearto rely on inordinate parameter tuning For example with the fixed parameters describedabove we obtain similar plots if we individually vary (i) rmin over 0325 035 04 0475(values still corresponding to an excision surface for the given choice of λ) (ii) η over1 3 7 10 (iii) s over 8 9 10 With the polynomial order N ranging over 23 26 29 31both stability and qualitatively similar exponential convergence is achieved with a singlesubdomain Likewise adoption of a larger coordinate domain with more subdomains doesnot impact our results Finally we have considered the addition of random noise to all fieldcomponents at the initial time Precisely with the system component χ as an example wehave set

χ(0) 7rarr χ(0) + δχ(0) (61)

where each component (nodal value) of δχ(0) is 10minus5 times a random variable drawn from a

19

10minus9

10minus7

10minus5

10minus3

Hinfin

10minus9

10minus7

10minus5

10minus3

Mrinfin

0 2000 4000 6000 8000 10000

10minus9

10minus7

10minus5

10minus3

Grinfin

t

N = 11N = 14N = 17N = 20

FIG 3 Spectral convergence of constraint violations for M = 1 Kerr-Schild ini-

tial data Respectively for N = 11 14 17 19 a timestep of ∆t ≃ 00041 00026 00018 00013

has been chosen for stability and accuracy

standard normal distribution Such perturbed initial data also gives rise to stable evolutions

V CONCLUSION

We have introduced a discontinuous Galerkin method for solving the spherically reducedBSSN system with second-order spatial operators Our scheme shares similarities withother discontinuous Galerkin methods that use local auxiliary variables to handle high-orderspatial derivatives [66 77ndash79 82 84 85] and which have typically been applied to eitherelliptic parabolic or mixed type problems The key ingredient of a stable dG scheme is anappropriate choice of numerical flux and our particular choice has been motivated by theanalysis presented in Appendix C When used to evolve the Schwarzschild solution in Kerr-Schild coordinates our numerical implementation of the BSSN system (4) is robustly stableand converges to the analytic solution exponentially with increased polynomial order Byapproximating the spatially second-order form of the BSSN system we have not introducedextra fields which are evolved Evolved auxiliary fields result in new constraints which mayspoil stability Our main goal has been stable evolution of the spherically reduced BSSNsystem as a first step towards understanding how a discontinuous Galerkin method might be

20

10minus9

10minus7

10minus5

∆Arrinfin

10minus9

10minus7

10minus5

∆Kinfin

0 2000 4000 6000 8000 10000

10minus9

10minus7

10minus5

max(∆Brinfin ∆Γrinfin)

t

N = 11N = 14N = 17N = 20

FIG 4 Spectral convergence of solution for M = 1 Kerr-Schild initial data

Timestep choices are described in the caption for Fig 3 In the title headings for example

∆Arr equiv (Arr)numer minus (Arr)exact

applied to the full BSSN system Towards that goal we now discuss treatment of singularitiesand generalization of the described dG method to higher space dimension

To deal with the fixed Schwarzschild singularity we have used excision which is easy inthe context of the spherically reduced BSSN system However excision for the binary blackhole problem in full general relativity requires attention to the technical challenge of hori-zon tracking State-of-the-art BSSN codes avoid such complication relying instead on themoving-puncture technique While the moving-puncture technique does involve mild centralsingularities it may still prove amenable to spectral methods Indeed spectral methods fornon-smooth problems is well-developed in both theory and for complex applications Sincethe moving-puncture technique can be performed in spherical symmetry [60] a first-steptoward a spectral moving-puncture code would be to implement a moving puncture withthe nodal dG method described here Such an implementation may adopt Legendre-Gauss-Radau nodes on the innermost subdomain thereby ensuring that the physical singularitydoes not lie on a nodal point (in much the same way finite difference codes use a staggeredgrid) Beyond traditional excision and moving punctures one might also retain a smoothproblem for which spectral methods are especially well-suited via the turducken approach[87 88] to singularities

21

10minus11

10minus9

10minus7

10minus5

∆χinfin

10minus11

10minus9

10minus7

10minus5

max(∆grrinfin ∆gθθinfin)

0 2000 4000 6000 8000 10000

10minus11

10minus9

10minus7

10minus5

max(∆αinfin ∆βrinfin)

t

N = 11N = 14N = 17N = 20

FIG 5 Spectral convergence of solution violations for M = 1 Kerr-Schild initial

data See the caption of Fig 4 for details

Discontinuous Galerkin methods for hyperbolic problems in two and three space dimen-sions are well-developed A generalization of the method described here to three-dimensionsand the full BSSN system would likely rely on an unstructured mesh Appropriate local poly-nomial expansions for the subelements are well-understood as are choices for the numericalfluxes which would now live on two-dimensional faces rather than single points Whether ornot it would ultimately prove successful generalization of our dG method to a higher dimen-sion would rely on an established conceptual framework Further computational advancesof relevance to a generalization of our dG method to the full BSSN system (possibly includ-ing matter) may include mesh hp-adaptivity local timestepping shock capturing and slopelimiting techniques [66] Moreover recent work [89] indicates that enhanced performancewould be expected were our scheme implemented on graphics processor units

VI ACKNOWLEDGMENTS

We thank Nick Taylor for discussions about treating second-order operators with spectralmethods Benjamin Stamm for helping polish up a few parts of Appendix C Khosro Shah-bazi for discussions on LDG and IP methods We also acknowledge helpful conversationswith David Brown and Manuel Tiglio about the BSSN system and previous work on its nu-

22

merical implementation We gratefully acknowledge funding through grants DMS 0554377and DARPAAFOSR FA9550-05-1-0108 to Brown University and NSF grant PHY 0855678to the University of New Mexico

Appendix A Hyperbolicity of the first-order system

This appendix analyzes the matrix A(u) appearing in (10) in order to construct thecharacteristic fields (14) In matrix form the sector (8) of the principal part of (10) reads asfollows

partt

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

=

βr 0 minus 4λα3grr

0 0 λβr

6(grr)2λβr

3gθθgrr0 4λ

3grr

0 βr 0 23grrαχ

13α minus αχ

3grr

αχ3gθθ

minus23χ 0

0 0 βr 0 0 0 0 minus χgrr

0

0 0 minus 4α3grr

βr 0 βr

6(grr)2βr

3gθθgrr0 4

3grr

0 0 23αχ 0 βr minus βrχ

3grrminus2βrχ

3gθθ0 minus2

3χ

0 minus2α 0 0 0 23βr minus2grrβr

3gθθ0 4

3grr

0 gθθαgrr

0 0 0 minusgθθβr

3grr13βr 0 minus2

3gθθ

0 0 minus2α 0 0 0 0 βr 0

34

0 0 0 0 0 0 0 βr

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

prime

(A1)which defines the matrix A(u) appearing in (9) and so also the matrix A(u) in (10) Notethat in the last equation the matrix within the square brackets is minusA(u) For certainconfigurations of u and λ the system (10) is strongly hyperbolic [76] that is A(u) hasa complete set of eigenvectors and real eigenvalues Indeed five eigenpairs of A(u) aretrivially recovered upon inspection of A(u)rsquos leading 5times5 diagonal block These correspondto eigenvalue 0 and the left eigenspace ξj = eTj 1 le j le 5 where ej are the canonical

basis vectors Since each component of u arises as eTj W each is also a characteristic field

The remaining nine eigenpairs are determined by A(u) The eigenvalues of A(u) are

micro1 = 0 micro23 = minusβr microplusmn4 = minusβr plusmn

radic

2αχ

grr microplusmn

5 = minusβr plusmn α

radic

χ

grr microplusmn

6 = minusβr plusmnradic

λ

grr

(A2)

23

and the corresponding left eigenvectors are

x1 = (0 0 0 0 0 gθθ 2grr 0 0) (A3a)

x2 =

(

0 0 0 grr2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3b)

x3 =

(

grrλ 0 0 0

2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3c)

xplusmn4 =

(

0 0plusmnradic

2αgrrχ

0 0 0 0 1 0

)

(A3d)

xplusmn5 =

(

0∓ 3radicgrrχ

plusmn2

radic

grrχ 2grr

1

χminus 1

grr1

gθθ 0 0

)

(A3e)

xplusmn6 =

(

minus3

4

grrλ 0plusmn α

radicλgrr

(2αχminus λ) 0 0minus βr

8(βrgrr ∓radicλgrr)

minus βrgrr

4gθθ(βrgrr ∓radicλgrr)

αχ

(2αχminus λ)plusmnradic

grrλ

)

(A3f)

where for example xplusmn5 A(u) = microplusmn5 x

plusmn5 Assuming that grr gθθ χ and α are everywhere strictly

positive the eigenvalues are real and the eigenvectors are linearly independent providedthat (15) holds These eigenvectors are easily extended to eigenvectors of A(u) e g asxplusmn6 rarr (01times5 x

plusmn6 ) Then for example the characteristic field

Xplusmn6 equiv (01times5 x

plusmn6 )W = xplusmn6WvQ (A4)

and similarly Xplusmnj = xplusmnj WvQ for j = 4 5 and Xk = xkWvQ for k = 1 2 3 The characteristic

speeds for these fields are microk and microplusmnj With this convention the speeds listed in Table I

correspond to the Xk and Xplusmnj in (14)

Appendix B Schwarzschild solution in conformal Kerr-Schild coordinates

In Kerr-Schild coordinates here the system directly related to incoming Eddington-Finkelstein null coordinates the line element for the Schwarzschild solution reads

ds2 = minusα2dt2 + (1 + 2MR)(dR + βRdt)2 +R2dθ2 +R2 sin2 θdφ2 (B1)

where R is the area radius α = (1 + 2MR)minus12 is the lapse and βR = 2M(R + 2M) isthe shift vector The physical spatial metric gab is the spatial part of this line element

To define the corresponding solution to the BSSN system we use equation gab = χgab todefine the following relationship between line elements

dr2 + r2(dθ2 + sin2 θdφ2) = χ[(1 + 2MR)dR2 +R2dθ2 +R2 sin2 θdφ2] (B2)

so that

χ

(

1 +2M

R

)(

dR

dr

)2

= 1 χR2 = r2 (B3)

Then we have(

1 +2M

R

)12dR

R=dr

r (B4)

24

with integration yielding

r =R

4

(

1 +

radic

1 +2M

R

)2

e2minus2radic

1+2MR (B5)

where the constant of integration has been chosen so that the R r rarr infin limits are consistentThe second relation in (B3) shows that

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR =rArr χminus4 =2eradic

1+2MRminus1

1 +radic

1 + 2MR (B6)

The extrinsic curvature tensor is specified by the expression forK given in (B9h) the identityK = KR

R + 2Kθθ and

Kθθ =

(

1 +2M

R

)minus122M

R2 (B7)

Since KRR = Kr

r we compute that

Krr = K minus 2Kθ

θ = minus(

1 +2M

R

)minus12(R +M

R + 2M

)

2M

R2 (B8)

Next since Krr = grrKrr = χminus1Kr

r we have Krr = Arr+

13grrK This implies Arr = Kr

r minus 13K

from which we get (B9g) In all we have

α =

(

1 +2M

R

)minus12

(B9a)

βr = βR dr

dR= χ12

(

1 +2M

R

)minus122M

R(B9b)

grr = 1 (B9c)

gθθ = r2 = χR2 (B9d)

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR (B9e)

Br = 0 (B9f)

Arr = minus(

1 +2M

R

)minus124M

3R2

(

2R + 3M

R + 2M

)

(B9g)

K =

(

1 +2M

R

)minus32(

1 +3M

R

)

2M

R2(B9h)

Γr = minus2

r= minus 2

χ12R (B9i)

To differentiate these expressions with respect to r we use the identity

dR

dr= χminus12

(

1 +2M

R

)minus12

(B10)

along with the chain rule

25

Appendix C Stability of the model system

The following stability analysis for the model system (53) has been inspired by [78 79]After dropping all nonlinear source terms the system (53) becomes

parttu = uprime + av (C1a)

parttv = uprimeprime + vprime (C1b)

This section analyzes the stability of (C1) considering both the continuum system itself aswell as its semi-discrete dG approximation The latter analysis offers some insight into theempirically observed stability of our dG scheme for the spherically reduced BSSN equations

1 Analysis for a single interval

Throughout we work with the L2 inner product and norm

(f g)D =

int

D

fg fD =radic

(f f)D (C2)

where D is the spatial coordinate interval (here D may represent a subdomain Dk or the

whole domain Ω) and we have suppressed all integration measures For the continuummodel we will establish the following estimate

uprime(T middot)2D+ av(T middot)2

Dle C(T )

(

uprime(0 middot)2D+ av(0 middot)2

D

)

(C3)

where the time-dependent constant C(T ) is determined solely by the choice of boundaryconditions To show (C3) we first change variables with v =

radicav thereby rewriting (C1)

in the following symmetric form

parttu = uprime +radicav (C4a)

parttv =radicauprimeprime + vprime (C4b)

Equations (C4ab) then imply

1

2partt

int

D

(uprime)2 =

int

D

uprime(uprimeprime +radicavprime) =

int

D

radicauprimevprime +

1

2

int

partD

(uprime)2 (C5a)

1

2partt

int

D

(v)2 =

int

D

v(radicauprimeprime + vprime) = minus

int

D

radicauprimevprime +

1

2

int

partD

(v2 + 2radicauprimev) (C5b)

Here vvprime and uprimeuprimeprime have been expressed as exact derivatives and then integrated to boundaryterms the second equation employs an extra integration by parts and with only one spacedimension

int

partDdenotes a difference of endpoint evaluations Addition of Eqs (C5ab) gives

1

2partt

int

D

[

v2 + (uprime)2]

=1

2

int

partD

[

v2 + (uprime)2 + 2radicauprimev

]

(C6)

Substitutions with the identities

[

v2 + (uprime)2]

=1

2

[

(v + uprime)2 + (v minus uprime)2]

2uprimev =1

2

[

(v + uprime)2 minus (v minus uprime)2]

(C7)

26

and replacements to recover the original variable v = vradica yield

1

2partt

int

D

[

av2 + (uprime)2]

=1 +

radica

4

int

partD

(radicav + uprime)2 +

1minusradica

4

int

partD

(radicav minus uprime)2 (C8)

From (C8) we deduce that the time-dependent constant C(T ) in (C3) must satisfy

∣

∣

∣

∣

∣

∣

1 +

int T

0

[

1+radica

2

int

partD(radicav + uprime)2 + 1minus

radica

2

int

partD(radicav minus uprime)2

]

dt

uprime(0 middot)2 + av(0 middot)2

∣

∣

∣

∣

∣

∣

le C(T ) (C9)

For periodic boundary conditions we may choose C(T ) = 1 Moreover if a ge 1 anduprime = minusradic

av is specified at partD+ then uprime(t middot)2 + av(t middot)2 decaysStill working on a single interval (subdomain) we now consider the semi-discrete scheme

for (C4) i e (57) with all nonlinear source terms dropped and with v replaced by vradica

Derivation of a formula analogous to (C8) is our first step toward establishing L2 stabilityof the semi-discrete scheme While (57) features vectors for example u(t) taking values atthe Legendre-Gauss-Lobatto nodal points here we work with the numerical solution as apolynomial for example uh(t x) These two representations are related by the Lagrange in-terpolating polynomials for the nodal set here taken to span both the space of test functionsand the space of basis functions Our scheme is

int

Dk

ψparttuh =

int

Dk

ψ(Qh +radicavh) (C10a)

int

Dk

ξparttvh = minusint

Dk

ξprime(radicaQh + vh) +

int

partDk

ξ(radicaQlowast + vlowast) (C10b)

int

Dk

ϕQh =

int

Dk

ϕuprimeh +

int

partDk

ϕ (ulowast minus uh) (C10c)

where ψ ξ and ϕ are polynomial test functions These test functions are arbitrary exceptthat they must be degree-N polynomials In (C10) the variables uh vh and Qh should alsocarry a superscript k but we have suppressed this Derivation of a formula analogous to(C8) is complicated by the fact that Qh is not evolved Nevertheless at a given instant t wecan assemble Qh from (C10c)

Mimicking the calculation (C5b) from the continuum case we first use (C10b) with ξ = vhto write

1

2partt

int

Dk

v2h = minusint

Dk

(radicaQh + vh)v

primeh +

int

partDk

(radicaQlowast + vlowast)vh

= minusint

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2(radicaQlowast + vlowast)vh minus v2h

]

(C11)

The right-hand side of (C5a) is analogous to

1

2partt

int

Dk

Q2h =

int

Dk

QhparttQh (C12)

However since Qh is not evolved the term parttQh must be given a suitable interpretationOn the right hand side of (C10c) only uh u

primeh and u

lowast necessarily depend on time since the

27

test function ϕ need not be time-dependent Furthermore ulowast is explicitly given as a linearcombination of uh as seen in Eq (C21c) below Choosing ϕ = ℓj taking the time derivativeof (C10c) and appealing to the commutivity of mixed partial derivatives we therefore arriveat

int

Dk

ℓjparttQh =

int

Dk

ℓj(parttuh)prime +

int

partDk

ℓj(

(parttu)lowast minus parttuh

)

(C13)

where (parttu)lowast depends on parttuh in precisely the same way that ulowast depends on uh We have

written ℓj rather than ϕ in the last equation to emphasize that the result also holds for anylinear combination of ℓj (for example ϕ) and even for time-dependent combinations SinceQh is itself such a combination we obtain

1

2partt

int

Dk

Q2h =

int

Dk

Qh(parttuh)prime +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

Qh(Qprimeh +

radicavprimeh) +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2((parttu)lowast minus parttuh)Qh +Q2

h

]

(C14)

Addition of (C11) and (C14) gives

1

2partt

int

Dk

(Q2h + v2h) =

1

2

int

partDk

[

Q2h minus v2h + 2(

radicaQlowast + vlowast)vh + 2((parttu)

lowast minus parttuh)Qh

]

(C15)

the aforementioned analog of (C8) This formula holds on a single subdomain Dk and we

now combine multiple copies of it one for each value of k

2 Analysis for multiple intervals

To facilitate combination of (C15) over all k we change notation At every subdomaininterface Ik+12 equiv partDk cap partDk+1 let the superscripts L and R denote field values respectivelytaken from the left and right Then the fields evaluated at Ik+12 which belong to D

k will beuLk+12 v

Lk+12 and Q

Lk+12 while those belonging to D

k+1 will be uRk+12 vRk+12 and Q

Rk+12

However at Ikminus12 the values taken from D

k are uRkminus12 vRkminus12 and QR

kminus12 Note that we

have also replaced the subscript h denoting a numerical solution with k plusmn 12 denotingthe location of the endpoint value of the numerical solution With this notation we define

∆Lα =

1

2

[

(QLα)

2 minus (vLα)2]

+(radic

aQlowastα + vlowastα

)

vLα +[

(parttuα)lowast minus parttu

Lα

]

QLα (C16)

and similarly for ∆Rα The same numerical fluxes appear in both ∆L

α and ∆Rα (ie each

numerical flux takes the same value on either side of an interface) whence fluxes like Qlowastα do

not carry an L or R superscript In terms of these definitions (C15) becomes

1

2partt

int

Dk

(Q2h + v2h) = ∆L

k+12 minus∆Rkminus12 (C17)

28

Summation over all Dk yields

1

2partt

kmaxsum

k=1

int

Dk

(Q2h + v2h) =

kmaxminus1sum

k=1

(

∆Lk+12 minus∆R

k+12

)

+∆Lkmax+12 minus∆R

12

=

kmaxminus1sum

k=1

(

∆Lh minus∆R

h

)∣

∣

Ik+12 +∆Lkmax+12 minus∆R

12 (C18)

We have reverted to h-notation denoting the numerical solution since the LR superscriptsindicate unambiguously the relevant domain used for evaluation at Ik+12

We again seek an estimate of the form

kmaxsum

k=1

(

Qh(T middot)2Dk + avh(T middot)2Dk

)

le C(T )kmaxsum

k=1

(

Qh(0 middot)2Dk + avh(0 middot)2Dk

)

(C19)

that is essentially the same as the one (C3) considered in the continuum case Assume thatthe chosen boundary conditions ensure ∆L

kmax+12 minus ∆R12 is bounded by a time-dependent

constant which does not depend on the numerical parameters N and h (subdomain width)Establishment of stability then amounts to showing that the remaining sum over interfaceterms in (C18) is non-positive whence this remaining sum is consistent with C(T ) le 1although the boundary conditions may give rise to a different bound In fact we will choosethe numerical fluxes such that each individual interface term is non-positive At interfaceIk+12 and in LR notation the jump and average of vh for example are

1

2

(

v+ + vminus)

equiv vh =1

2

(

vLk+12 + vRk+12

)

(C20a)

nminusvminus + n+v+ equiv[[

vh]]

= vLk+12 minus vRk+12 (C20b)

Consider numerical fluxes of the form

Qlowast = Qh minusτQ2

[[

Qh

]]

(C21a)

vlowast = vh minusτv2

[[

vh]]

(C21b)

ulowast = uh minusτu2

[[

uh]]

(C21c)

(parttu)lowast = parttuh minus

τu2

[[

parttuh]]

(C21d)

where (C21c) induces (C21d) and where the penalty parameters τu τv and τQ are realnumbers The fluxes defined in (58) correspond to τu = 1 τv = 1 +

radica and τQ = 0 In

terms of these quantities the kth interface contribution in (C18) is

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

parttuh]]

minus τu2

[[

parttuh]][[

Qh

]]

(C22)

where we have suppressed the k dependence of the right-hand side Now consider the term[[

parttuh]]

Because parttuh and Qh+radicavh are both polynomials of degree N Eq (C10a) implies

29

τu

τ Q

minus1 0 1minus1

minus05

0

05

1

τu

minus1 0 1minus1

minus05

0

05

1

FIG 6 Stable evolutions for the model system For fixed τv = 10minus6 and τv = 1 +radic2

respectively the left and right plots depict stable choices (determined empirically) of τu and τQfor the linear model system (C1) The stable regions are colored black but the jagged edges result

from the discretization of the (τu τQ)-plane

the vector equation parttu = Q +radicav that is pointwise equivalence on the nodal points of

Dk which in turn implies

[[

parttuh]]

=[[

Qh+radicavh]]

Upon substituting this identity into thelast equation we arrive at an expression which features only vh and Qh

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

Qh +radicavh]]

minus τu2

[[

Qh +radicavh]][[

Qh

]]

(C23)

The identities vh[[

vh]]

= 12

[[

v2h]]

and[[

Qh +radicavh]]

=[[

Qh

]]

+radica[[

vh]]

then simplify(C23) to

(∆Lh minus∆R

h )∣

∣

Ik+12 = minusτv2

[[

vh]]2 minus

radica(τu + τQ)

2

[[

Qh

]][[

vh]]

minus τu2

[[

Qh

]]2 (C24)

The role of a penalty parameter is now clear Positive values of τv penalize jumps in vhthrough a negative contribution to the energy Likewise positive values of τu penalizejumps in Qh through a negative contribution to the energy However because the sign of[[

Qh

]][[

vh]]

can be positive or negative only the choice τQ = minusτu yields an expression for(∆L

h minus ∆Rh )|Ik+12 which is manifestly negative for τu ge 0 and τv ge 0 A simple estimate

based on Youngrsquos inequality with ε (that is 2αβ le εminus1α2 + εβ2 where α β ge 0 and ε gt 0)shows that for τQ = 0 the choice τv ge aτu4 also yields stability

Figure 6 depicts certain choices of stable penalty parameters for the linear model systemevolved to tfinal = 1000 (with a = 2 N = 10 and ∆t ≃ 00553) as determined empiricallywith simulations similar to those described in Sec IVA The left plot corresponds to asmall τv = 10minus6 for which the choice τu = 1 τQ = 0 is not stable as expected from thetheoretical analysis However the right plot corresponds to τv = 1 +

radica for which τu = 1

τQ = 0 is stable Motivated by the numerical flux choices (3842) used for the BSSN system

30

(4) we have (as mentioned above) set τu = 1 τv = 1 +radica and τQ = 0 in simulations of

the nonlinear model (53) For the nonlinear model system (53) the theoretically motivatedchoice τQ = minusτu also yields numerically stable evolutions when τu ge 0 and τv ge 0

For the nonlinear systems (4) and (53) we do not attempt a formal stability proofNevertheless the results of this appendix have served as a guide for our choices of penaltyparameters For the BSSN system (4) u v and Q are block indices [cf Eq (6)] Similarto the model problem we have penalized Q with τu = 1 with τv chosen large enough toheuristically overcome the cross-terms of indefinite size that arise from τQ = 0 (we interpretequations like τu = 1 componentwise) An analogous choice ldquoτQ = minusτurdquo for the BSSNsystem might be possible but would be considerably more complicated Indeed such achoice likely entails a matrix of penalty parameters but we do not give the details here

[1] F Pretorius Evolution of Binary Black-Hole Spacetimes Phys Rev Lett 95 121101 (2005)

(4 pages)

[2] F Pretorius Simulation of binary black hole spacetimes with a harmonic evolution scheme

Class Quantum Grav 23 S529-S552 (2006)

[3] M Campanelli C O Lousto P Marronetti and Y Zlochower Accurate Evolutions of Or-

biting Black-Hole Binaries without Excision Phys Rev Lett 96 111101 (2006) (4 pages)

[4] J G Baker J Centrella D-I Choi M Koppitz and J van Meter Gravitational-Wave

Extraction from an Inspiraling Configuration of Merging Black Holes Phys Rev Lett 96

111102 (2006) (4 pages)

[5] M Campanelli C O Lousto and Y Zlochower Last orbit of binary black holes Phys Rev D

73 061501(R) (2006) (5 pages)

[6] F Herrmann I Hinder D Shoemaker and P Laguna Unequal mass binary black hole plunges

and gravitational recoil Class Quantum Grav 24 S33-S42 (2007)

[7] P Diener F Herrmann D Pollney E Schnetter E Seidel R Takahashi J Thornburg and

J Ventrella Accurate Evolution of Orbiting Binary Black Holes Phys Rev Lett 96 121101

(2006) (4 pages)

[8] M A Scheel H P Pfeiffer L Lindblom L E Kidder O Rinne and S A Teukolsky Solving

Einsteinrsquos Equations With Dual Coordinate Frames Phys Rev D 74 104006 (2006)

[9] U Sperhake Binary black-hole evolutions of excision and puncture data Phys Rev D 76

104015 (2007) (20 pages)

[10] B Brugmann J A Gonzalez M Hannam S Husa U Sperhake and W Tichy Calibration

of moving puncture simulations Phys Rev D 77 024027 (2008) (25 pages)

[11] P Marronetti W Tichy B Brugmann J Gonzalez M Hannam S Husa and U Sperhake

Binary black holes on a budget simulations using workstations Class Quantum Grav 24

S43-S58 (2007)

[12] Z B Etienne J A Faber Y T Liu S L Shapiro and T W Baumgarte Filling the holes

Evolving excised binary black hole initial data with puncture techniques Phys Rev D 76

101503(R) (2007) (5 pages)

[13] B Szilagyi D Pollney L Rezzolla J Thornburg and J Winicour An explicit harmonic code

for black-hole evolution using excision Class Quantum Grav 24 S275-S293 (2007)

[14] L Boyle M Kesden and S Nissanke Binary-Black-Hole Merger Symmetry and the Spin

Expansion Phys Rev Lett 100 151101 (2008) (4 pages)

31

[15] M A Scheel M Boyle T Chu L E Kidder K D Matthews and H P Pfeiffer High-

accuracy waveforms for binary black hole inspiral merger and ringdown Phys Rev D 79

024003 (2009) (14 pages)

[16] M Hannam Status of black-hole-binary simulations for gravitational-wave detection

Class Quantum Grav 26 114001 (2009) (21 pages)

[17] I Hinder The current status of binary black hole simulations in numerical relativity

Class Quantum Grav 27 114004 (2010) (19 pages)

[18] B P Abbott et al LIGO The Laser Interferometer Gravitational-Wave Observatory

Rep Prog Phys 72 076901 (2009) (25 pages)

[19] F Acernese et al The status of VIRGO Class Quantum Grav 23 S63-S69 (2006)

[20] F Acernese et al Virgo status Class Quantum Grav 25 184001 (2008) (9 pages)

[21] R Arnowitt S Deser and C W Misner The Dynamics of General Relativity in Gravitation

an introduction to current research edited by L Witten (Wiley New York 1962) Chapter 7

227-265 Also available as gr-qc0405109

[22] J W York Jr Kinematics and Dynamics of General Relativity in Sources of Gravitational

Radiation edited by L L Smarr (Cambridge University Press Cambridge 1979)

[23] L E Kidder M A Scheel and S A Teukolsky Extending the lifetime of 3D black hole

computations with a new hyperbolic system of evolution equations Phys Rev D 64 064017

(2001) (13 pages)

[24] G Calabrese J Pullin O Sarbach and M Tiglio Convergence and stability in numerical

relativity Phys Rev D 66 041501(R) (2002) (4 pages)

[25] S Frittelli and O Reula On the Newtonian Limit of General Relativity Commun Math

Phys 166 221-235 (1994)

[26] M Shibata and T Nakamura Evolution of three-dimensional gravitational waves Harmonic

slicing case Phys Rev D 52 5428-5444 (1995)

[27] Y Choquet-Bruhat and J W York Geometrical well posed systems for the Einstein equations

C R Acad Sci Paris t 321 Serie I 1089-1095 (1995)

[28] A Abrahams A Anderson Y Choquet-Bruhat and J W York Jr Einstein and Yang-Mills

Theories in Hyperbolic Form without Gauge Fixing Phys Rev Lett 75 3377-3381 (1995)

[29] C Bona J Masso E Seidel and J Stela New Formalism for Numerical Relativity

Phys Rev Lett 75 600-603 (1995)

[30] M H P M van Putten and D M Eardley Nonlinear wave equations for relativity

Phys Rev D 53 3056-3063 (1996)

[31] S Frittelli and O A Reula First-Order Symmetric Hyperbolic Einstein Equations with Ar-

bitrary Fixed Gauge Phys Rev Lett 76 4667-4670 (1996)

[32] H Friedrich Hyperbolic reductions for Einsteinrsquos equations Class Quantum Grav 13 1451-

1469 (1996)

[33] F B Estabrook R S Robinson H D Wahlquist Hyperbolic equations for vacuum gravity

using special orthonormal frames Class Quantum Grav 14 1237-1247 (1997)

[34] M S Iriondo E O Leguizamon and O A Reula Einsteinrsquos Equations in Ashtekarrsquos Vari-

ables Constitute a Symmetric Hyperbolic System Phys Rev Lett 79 4732-4735 (1997)

[35] A Anderson Y Choquet-Bruhat and J W York Einstein-Bianchi hyperbolic system for

general relativity Topol Meth Nonlinear Anal 10 353-373 (1997)

[36] T W Baumgarte and S L Shapiro Numerical integration of Einsteinrsquos field equations

Phys Rev D 59 024007 (1998) (7 pages)

[37] M A G Bonilla Symmetric hyperbolic systems for Bianchi equations Class Quantum

6

which case the corresponding principal part features

parttBr = βrBr prime minus 4λα

3grrK prime +

4λ

3grrQprime

βr +λβr

6(grr)2Qprime

grr +λβr

3gθθgrrQprime

gθθ(8a)

parttArr = βrAprimerr +

2

3grrαχΓ

r prime +1

3αQprime

χ minus2

3χQprime

α minus αχ

3grrQprime

grr +αχ

3gθθQprime

gθθ(8b)

parttK = βrK prime minus χ

grrQprime

α (8c)

parttΓr = βrΓr prime minus 4αK prime

3grr+

4

3grrQprime

βr +βr

6(grr)2Qprime

grr +βr

3gθθgrrQprime

gθθ(8d)

parttQχ = βrQprimeχ +

2

3αχK prime minus βrχ

3grrQprime

grr minus2βrχ

3gθθQprime

gθθminus 2

3χQprime

βr (8e)

parttQgrr =2

3βrQprime

grr +4

3grrQ

primeβr minus 2αAprime

rr minus2grrβ

r

3gθθQprime

gθθ(8f)

parttQgθθ =1

3βrQprime

gθθ+gθθα

grrAprime

rr minusgθθβ

r

3grrQprime

grr minus2

3gθθQ

primeβr (8g)

parttQα = βrQprimeα minus 2αK prime (8h)

parttQβr = βrQprimeβr +

3

4Br prime (8i)

where all lower-order terms on the right-hand side have been dropped This sector of prin-cipal parts of the first-order system has the form

parttWvQ + A(u)W primevQ = 0 (9)

where (minus) the explicit form of the 9-by-9 matrix A(u) is given below in (A1) Thefirst-order version of (4) takes the nonconservative form

parttW +A(u)W prime = S(W ) A(u) =

(

05times5 05times9

09times5 A(u)

)

(10)

where S(W ) is a vector of lower order terms built with all components of W Partition ofA(u) = A(u)vQvQ into blocks corresponding to the v and Q sectors yields

A(u) =

(

A(u)vv A(u)vQ

A(u)Qv A(u)QQ

)

(11)

Using these blocks we then define the 9-by-9 matrix

A(u) = A(u)uvvQ =

(

05times4 05times5

A(u)vv A(u)vQ

)

(12)

and express (4) as

parttWuv + A(u)W primevQ = S(W ) (13a)

Q = uprime (13b)

where S(W ) = S(W )uv

7

field speed

X1 micro1 = 0

X23 micro23 = minusβr

Xplusmn4 microplusmn

4 = minusβr plusmnradic

2αχgrr

Xplusmn5 microplusmn

5 = minusβr plusmn αradic

χgrr

Xplusmn6 microplusmn

6 = minusβr plusmnradic

λgrr

TABLE I Characteristic speeds These speeds are the eigenvalues listed in (A2)

C Hyperbolicity and characteristic fields

Although our numerical scheme deals directly with the second-order spatial operatorsappearing in (4) we first consider the hyperbolicity of the corresponding first-order system(10) The characteristic fields and their speeds are found by instantaneously ldquofreezingrdquothe fields u in A(u) to some value u0 corresponding to a linearization around a uniformstate Below we continue to write u for simplicity with the understanding that u is reallythe background solution u0 Of primary interest is the range of u0 for which the system isstrongly hyperbolic [73ndash76]

Appendix A shows that the characteristic fields corresponding to (4) are as follows (i)all components of u (each with speed 0) and (ii) the fields

X1 = gθθQgrr + 2grrQgθθ (14a)

X2 = grrΓr +

2

χQχ minus

1

2grrQgrr minus

1

gθθQgθθ (14b)

X3 =grrλBr +

2

χQχ minus

1

2grrQgrr minus

1

gθθQgθθ (14c)

Xplusmn4 = plusmn

radic

2αgrrχ

K +Qα (14d)

Xplusmn5 = ∓ 3radic

grrχArr plusmn 2

radic

grrχK + 2grrΓ

r +1

χQχ minus

1

grrQgrr +

1

gθθQgθθ (14e)

Xplusmn6 = minus3

4

grrλBr plusmn α

radicλgrr

(2αχminus λ)K minus βr

8(βrgrr ∓radicλgrr)

Qgrr

minus βrgrr

4gθθ(βrgrr ∓radicλgrr)

Qgθθ +αχ

(2αχminus λ)Qα plusmn

radic

grrλQβr (14f)

with the speeds listed in Table I To ensure strong hyperbolicity we must necessarily require

λ gt 0 (βr)2grr minus λ 6= 0 2αχminus λ 6= 0 (15)

as shown in in Appendix A where further conditions are also given When λ = 1 thehyperbolicity condition of Ref [60] is recovered In fact the system could be recast assymmetric hyperbolic Indeed as it involves one spatial dimension the relevant symmetrizercan be constructed via polar decomposition of the diagonalizing similiarity transformationHowever we will not exploit this possibility

8

This system admits an inner excision boundary provided

βr ge max

(

radic

2αχ

grr

radic

α2χ

grr

radic

λ

grr

)

(16)

holds at the inner boundary This condition ensures each characteristic field has a nonposi-tive speed at the inner boundary and therefore the inner boundary is an excision boundaryat which no boundary conditions are needed The extra flexibility afforded by the param-eter λ could be used to maintain rigorous hyperbolicity by moving the points at whichthe conditions in (15) are violated outside of the computational domain Furthermore forλ = 1 Eq (16) conceivably fails or is only satisfied close to r = 0 where field gradients are

prohibitively large The troublesome X+6 gauge mode has a positive speed minusβr +

radic

λgrrIndeed for the conformally flat Kerr-Schild system considered in section IVB an inner ex-cision boundary is only possible provided λ is small enough

The transformation (14) can be inverted in order to express the fundamental fields interms of the characteristic fields

Br = minus1

6

λ

grrgθθ

[

(βr)2

(βr)2grr minus λ

]

X1 +2

3

λαχ

grr(2αχminus λ)(X+

4 +Xminus4 )minus

2

3

λ

grr(X+

6 +Xminus6 )

(17a)

Arr =1

3

radic

grrχ

2α(X+

4 minusXminus4 )minus

radicgrrχ

6(X+

5 minusXminus5 ) (17b)

K =

radic

χ

8αgrr(X+

4 minusXminus4 ) (17c)

Γr = minus1

6

1

grrgθθ

[

(βr)2

(βr)2grr minus λ

]

X1 +1

grr(X2 minusX3) +

2

3

αχ

grr(2αχminus λ)(X+

4 +Xminus4 )

minus 2

3

1

grr(X+

6 +Xminus6 ) (17d)

Qχ =1

12

χ

grrgθθ

[

4(βr)2grr minus 3λ

(βr)2grr minus λ

]

X1 +χ

2X3 minus

1

3

αχ2

(2αχminus λ)(X+

4 +Xminus4 ) (17e)

+χ

3(X+

6 +Xminus6 ) (17f)

Qgrr =2(βr)2grr minus 3λ

6gθθ((βr)2grr minus λ)X1 +

4

3grrX2 minus grrX3 +

2

3

αχgrr(2αχminus λ)

(X+4 +Xminus

4 )

minus 1

3grr(X

+5 +Xminus

5 )minus2

3grr(X

+6 +Xminus

6 ) (17g)

Qgθθ =

[

1

4grr+

(βr)2

12((βr)2grr minus λ)

]

X1 minus2

3gθθX2 +

1

2gθθX3 minus

1

3

αχgθθ(2αχminus λ)

(X+4 +Xminus

4 )

+1

6gθθ(X

+5 +Xminus

5 ) +1

3gθθ(X

+6 +Xminus

6 ) (17h)

Qα =1

2(X+

4 +Xminus4 ) (17i)

Qβr =βrλ

8grrgθθ((βr)2grr minus λ)X1 minus

λ

(2αχminus λ)

radic

αχ

8grr(X+

4 minusXminus4 ) +

1

2

radic

λ

grr(X+

6 minusXminus6 )

(17j)

9

We will refer to this inverse transformation when discussing outer boundary conditions forour numerical simulations in Sec IVB

III DISCONTINUOUS GALERKIN METHOD

This section describes the nodal discontinuous Galerkin method used to numerically solve(4) We adopt a method-of-lines strategy and here describe the relevant semi-discrete schemewhile leaving the temporal dimension continuous To approximate (4) we follow the generalprocedure first introduced in Ref [82] Our approach defines local auxiliary variables Q = uprimeand rewrites the spatially second-order system (4) as the first-order system (13a) Once weuse (13b) to eliminate Q from (13a) we recover the primal equations (4) The auxiliaryvariable approach was later generalized and coined the local discontinuous Galkerin (LDG)method in Ref [77] We may refer to our particular scheme as an LDG method but notethat many variations exist in the literature We stress that in LDG methods Q is not evolvedand is introduced primarily to assist in the construction of a stable scheme

Equations (12) and (13a) imply that the physical flux function is

F (W ) =

(

Fu(W )

Fv(W )

)

equiv A(u)WvQ =

(

05times1

f(W )

)

f =

fBr

fArr

fKfΓ

(18)

Only the evolution equations for Br Arr K and Γr give rise to non-zero components in F and we have collected these non-zero components into a smaller vector f = Fv Inspectionof (8) determines these components For example from (8c) we find

fK = minusβrK +χ

grrQα (19)

A Local approximation of the system (13)

Our treatment closely follows [83] but with the equations and notations relevant forthis paper Our computational domain Ω is the closed r-interval [a b] We cover Ω withkmax gt 1 non-overlapping intervals Dk = [ak bk] where a = a1 b = bkmax and bkminus1 = ak fork = 2 middot middot middot kmax

On each interval Dk we approximate each component of the system vector W by a localinterpolating polynomial of degree N For example

χkh(t r) =

Nsum

j=0

χ(t rkj )ℓkj (r) (20)

approximates χ(t r) Throughout this section approximations are denoted by a subscripth (see [66] for the notation) For example Wh and fh are approximations of W and f Although Q = uprime Qh and uprimeh are not necessarily the same In (20) ℓkj (r) is the jth Lagrange

polynomial belonging to Dk

ℓkj (r) =Nprod

i=0i 6=j

r minus rkirkj minus rki

(21)

10

Evidently the polynomial χkh interpolates χ at rkj To define the nodes rkj consider the

mapping from the unit interval [minus1 1] to Dk

rk(u) = ak + 12(1 + u)(bk minus ak) (22)

and the N+1 Legendre-Gauss-Lobatto (LGL) nodes uj The uj are the roots of the equation

(1minus u2)P primeN(u) = 0 (23)

where PN(u) is the Nth degree Legendre polynomial and the physical nodes are simplyrkj = rk(uj) In vector notation the approximation (20) takes the form

χkh(t r) = χk(t)Tℓk(r) (24)

in terms of the column vectors

χk(t) =[

χ(t rk0) middot middot middot χ(t rkN)]T ℓk(r) =

[

ℓk0(r) middot middot middot ℓkN(r)]T (25)

On each open interval (ak bk) sub Dk and for each component of the equations in (13) we

define local residuals measuring the extent to which our approximations satisfy the originalcontinuum system Dropping the subdomain label k on the polynomials and focusing onthe K equation as a representative example the local residual corresponding to (4h) is

minus(RK)kh equivminus parttKh + (βrK prime)h minus

(

χQprimeα

grr

)

h

+

(

χQgrrQα

2g2rr

)

h

minus(

χQgθθQα

grrgθθ

)

h

+

(

QαQχ

2grr

)

h

+

(

3αA2rr

2g2rr

)

h

+

(

1

3αK2

)

h

(26)

Here for example the expressions 2

(βrK prime)h = βrhK

primeh

(

QαQχ

2grr

)

h

=QαhQχh

2grrh (27)

We similarly construct the remaining eight residuals eg (Rgrr)h and (RΓr)h as well as fiveresiduals corresponding to (13b) For example one of these remaining five is

(RQα)kh equiv minusQαh + αprime

h (28)

Let the kth inner product be defined as

(

u v)

Dk equivint bk

akdru(r)v(r) (29)

and consider the expression (ℓkj (RK)kh)Dk We call the requirement that this inner product

vanish forallj the kth Galerkin condition For each component of the system and for each k

2 At this stage the first expression is generically a polynomial of degree 2N minus 1 and the latter is not a

polynomial The conventions adopted in Eq (27) prove useful while working with the residual However

later on in Sec III C to obtain the final form (47) of the numerical approximation corresponding to (26)

we will replace nonlinear terms with degree-N polynomials

11

there is a corresponding Galerkin condition in total 9kmax(N + 1) equations for (13a) and5kmax(N+1) for (13b) Enforcement of the Galerkin conditions on each D

k will not recover ameaningful global solution since they provide no mechanism for coupling the local solutionson the different intervals Borrowing from the finite volume toolbox we achieve couplingthrough integration by parts on r and introduction of the numerical flux f lowast at the interfacebetween subdomains

In (26) we only need to consider (βrK prime)h and (χQprimeαgrr)h as the other terms comprise a

component of the source vector Sh Using integration by parts we write

(

ℓkj (βrK prime)h

)

Dk =minusint bk

akdr[

(

ℓkjβrh

)primeKh

]

+ (βrhKh) ℓ

kj

∣

∣

∣

bk

ak (30a)

(

ℓkj (χQprimeαgrr)h

)

Dk =minusint bk

akdr

[(

ℓkjχh

grrh

)prime

Qαh

]

+

(

χhQαh

grrh

)

ℓkj

∣

∣

∣

bk

ak (30b)

In these formulas we have retained the domain index k on ℓkj while continuing to suppress iton Kh grrh etc Moreover we have suppressed the r-dependence in all terms on the right-hand side Addition of these formulas along with the definition fKh = minus(βrK)h+(χQαgrr)hgives

(

ℓkj (βrK prime)h minus (χQprime

αgrr)h)

Dk = minusint bk

akdr

[

(

ℓkjβrh

)primeKh minus

(

ℓkjχh

grrh

)primeQαh

]

minus fKhℓkj

∣

∣

∣

bk

ak (31)

In lieu of (31) we will instead work with the replacement

(

ℓkj (βrK prime)h minus (χQprime

αgrr)h)

Dk rarr minusint bk

akdr

[

(

ℓkjβrh

)primeKh minus

(

ℓkjχh

grrh

)primeQαh

]

minus f lowastKℓ

kj

∣

∣

∣

bk

ak (32)

This replacement features a component f lowastK of the numerical flux rather than a component

fKh of the boundary flux The numerical flux is determined by (as yet not chosen) functions3

f lowast = f lowast(W+Wminus) (33)

where for example Wminus is an interior boundary value [either W kh (t a

k) or W kh (t b

k)] of theapproximation defined on D

k andW+ is an exterior boundary value [either W kminus1h (t bkminus1) or

W k+1h (t ak+1)] of the approximation defined on either Dkminus1 or Dk+1 We discuss our choice

of numerical flux in the next subsection We now employ additional integration by parts towrite the above replacement as

(

ℓkj (βrK prime)h minus (χQprime

αgrr)h)

Dk rarrint bk

akdrℓkj

(

βrK prime minus χQprimeα

grr

)

h

+ (fKh minus f lowastK) ℓ

kj

∣

∣

∣

bk

ak (34)

Rather than the exact kth Galerkin condition(

ℓkj (RK)kh

)

Dk = 0 forallj for the K component

of (13) on Dk we will instead strive to enforce

(

ℓkj (RK)kh

)

Dk = (fKh minus f lowastK) ℓ

kj

∣

∣

∣

bk

ak forallj (35)

3 In the context of the dG method here + and minus denote ldquoexteriorrdquo and ldquointeriorrdquo and have no relation to

the plusmn using to denote the characteristic fields and speeds in Table I For characteristic fields and speeds

+ and minus mean ldquoright-movingrdquo and ldquoleft-movingrdquo

12

although our treatment of nonlinear terms will lead to a slight modification of these equations(we return to this issue shortly) The other components of (13a) are treated similarly asare the components of (13b) Recall that for example Qα = αprime Formally using the samedG method to solve for Qα we arrive at the replacement

(

ℓkj (RQα)kh

)

Dk rarrint bk

akdrℓkj (minusQαh + αprime

h)minus (αh minus αlowast) ℓkj

∣

∣

∣

bk

ak (36)

which again features a component αlowast of the numerical flux The auxiliary variables areconstructed and used at each stage of temporal integration We then have

(

ℓkj (RQα)kh

)

Dk = (αh minus αlowast) ℓkj

∣

∣

∣

bk

ak forallj (37)

as the corresponding enforced kth Galerkin condition

B Numerical Flux

To further complete our dG scheme we must specify functional forms for the compo-nents of the numerical flux introduced in the previous section We distinguish between thephysical fluxes (components of f) and the auxiliary fluxes (components of u) arising fromthe definition of the auxiliary variables These choices are not independent as the resultingscheme must be stable and consistent Our choice follows [84] which considered diffusionproblems Additional analysis of this flux choice appears in [66 85]

Let us first consider the numerical fluxes corresponding to the physical fluxes and of theform (33) The numerical flux vector is a function of the system and auxiliary variablesinterior and exterior to a subdomain A common choice for f lowast is

f lowast = fh+τ

2

[[

vh]]

K-component of f lowast f lowastK = fKh+

τ

2

[[

Kh

]]

(38)

where as an example we have also shown the component of f lowast corresponding to the analysisabove Respectively the average and jump across the interface are

fh =1

2

(

f+ + fminus) [[

vh]]

= nminusvminus + n+v+ (39)

Here τ is a position dependent penalty parameter (fixed below) and nminus(n+) is the localoutward pointing normal to the interior (exterior) subdomain The role of τ is to ldquopenalizerdquo(i e yield a negative contribution to the L2 energy norm) jumps across an interface Anappropriate choice of τ will ensure stability and we now provide some motivation for thechoice (41) of τ we make below

Were we treating the fully first-order system (10) the local Lax-Friedrichs flux wouldoften be a preferred choice due to its simplicity [66] In this case the constant ω in thenumerical flux formula Flowast = Fh + 1

2ω[[

Wh

]]

obeys ω ge max∣

∣micro(nablaWF(W ))∣

∣ HereF(W ) = A(u)W the notation micro(middot) indicates the spectral radius of the matrix within andthe max is taken over interior Wminus and exterior W+ states Motivated by (9) we adopt asimilar but simpler prescription substituting the field gradient

nablaWvQA(u)WvQ = A(u) (40)

13

for nablaWF(W ) Precisely we assume the scaling

τ(bk) = τ(ak+1) = τk+12 equiv C middotmax∣

∣micro(

A(u))∣

∣ (41)

where C = O(1) is a constant chosen for stability Larger values of C will result in schemeswith better stability properties whereas too large a value will impact the CFL conditionAt the interface point Ik+12 equiv D

k cap Dk+1 the vector uh has two representations uminus at bk

and u+ at ak+1 The max in (41) is taken over the corresponding two sets of field speedsMore precisely the speeds in Table I are computed for both uminus and u+ and the maximumtaken over all resulting speeds For the auxiliary variables a penalized central flux is usedThe definition with one representative component is

ulowast = uh minus1

2

[[

uh]]

α-component of ulowast αlowast = αh minus1

2

[[

αh

]]

(42)

with similar expressions for the remaining componentsWe stress the following point Since the interior coupling between subdomains is achieved

through the numerical flux forms (41) and (42) the inverse transformation (17) expressingthe fundamental fields in terms of the characteristic fields is not required to achieve thiscoupling On the other hand imposition of physical boundary conditions may still rely on(17) since this transformation allows one to fix only incoming characteristic modes

C Matrix form of the semi-discrete equations

Let us introduce the kth mass and stiffness matrices

Mkij =

int bk

akdrℓki (r)ℓ

kj (r) Sk

ij =

int bk

akdrℓki (r)ℓ

kjprime(r) (43)

These matrices belong to Dk and the corresponding matrices defined on the reference interval[minus1 1] are

Mij =

int 1

minus1

duℓi(u)ℓj(u) Sij =

int 1

minus1

duℓi(u)ℓprimej(u) (44)

where ℓj(u) is the jth Lagrange polynomial determined by the LGL nodes uj on [minus1 1]These matrices are related by Mk

ij =12(bk minus ak)Mij and S

kij = Sij whence only the reference

matrices require computation and storageWe will use the matrices Mk and Sk in obtaining an ODE system from (26) and (35)

Towards this end we first approximate the nonlinear terms (products and quotients) in(26) by degree-N interpolating polynomials Such approximations are achieved throughpointwise representations For example (QαQχgrr)h appears in (26) and is expressed inthe following way [cf footnote 2]

(

QαQχ

grr

)

h

(t r) =

(

QαhQχh

grrh

)

(t r) rarrNsum

j=0

Qαh(t rkj )Qχh(t r

kj )

grrh(t rkj )

ℓkj (r) (45)

Note that the expressions on the right and left are not equivalent due to aliasing error [65]Our vector notation for this replacement will be

(

QαQχ

grr

)

h

(t r) rarr(

QαQχ

grr

)

(t)Tℓk(r) (46)

14

Operations among bold variables are always performed pointwise Making similar replace-ments for all terms in (26) and then carrying out the integrations in (35) which bring inMk and Sk we arrive at

parttK = βrDK minus χDQα

grr

+1

2

χQgrrQα

g2rr

minusχQgθθ

Qα

grrgθθ

+1

2

QαQχ

grr

+3

2

αA2rr

g2rr

+1

3αK2 +Mminus1ℓk (fKh minus f lowast

K)∣

∣

∣

bk

ak (47)

where we have again suppressed the superscript k on all terms except ℓk(r) and the sub-script h is dropped on all boldfaced variables As described in [66] the spectral collocationderivative matrix

(Dk)ij =dℓkjdr

∣

∣

∣

∣

∣

r=rki

(48)

can also be expressed as Dk = (Mk)minus1Sk which appears above Eight other semi-discreteevolution equations are similarly obtained with nine in total (one for each component ofWuv) Additionally we have

Qα = Dα+Mminus1ℓk(αlowast minusαh)∣

∣

∣

bk

ak (49)

as one of the auxiliary equations with five in total (one for each component of Q = WQQ)

D Filtering

Like other nodal (pseudospectral) methods our scheme may suffer from instabilitiesdriven by aliasing error [65] Filtering is a simple yet robust remedy To filter a solu-tion component such as χ we use the modal (as opposed to nodal) representation of thesolution

χkh(t r) =

Nsum

j=0

χ(t rkj )ℓkj (r) =

Nsum

j=0

χkj (t)Pj(r) (50)

where Pj(r) is the jth Legendre polynomial Let ηj = jN and define the filter function

σ(ηj) =

1 for 0 le ηj le NcN

exp(

minus ǫ(

ηjminusNcN

1minusNcN

)2s)

for NcN le ηj le 1(51)

At each timestep we modify our solution component according to

χkh rarr

(

χkh

)filtered=

Nsum

j=0

σ(ηj)χkj (t)Pj(r) (52)

Evidently the modification only affects the top N minus Nc modes and is sufficient to controlthe type of weak instability driven by aliasing [66] The numerical parameters Nc and ǫ areproblem dependent For our simulations we have taken ǫ ≃ minuslog(εmach) = 36 where εmach

is machine accuracy in double precision

15

E Model system

To better illustrate the basic properties of our method we consider a toy model Namelythe following spatially second-order system

parttu = uprime + av minus u3 + g(t x) (53a)

parttv = uprimeprime + vprime minus (u+ v)(uprime)2 + v2u2 + h(t x) (53b)

where a ge 1 is constant and g and h are analytic source terms to be specified In contrastto (6) here u v and Q = uprime are scalars rather than vectors System (53) admits a first-order reduction in which uprime is defined as an extra variable Since this first-order reductionis strongly hyperbolic the spatially second-order system (53) is also strongly hyperbolic byone of the definitions considered in [75] The characteristic fields Xplusmn and speeds microplusmn are

X+ =radicav minus uprime micro+ =

radicaminus 1 Xminus =

radicav + uprime microminus = minus(

radica+ 1) (54)

To construct a local dG scheme for this system we first rewrite it as

parttu = Q+ av minus u3 + g(t x) (55a)

parttv = Qprime + vprime minus (u+ v)Q2 + v2u2 + h(t x) (55b)

Q = uprime (55c)

Evidently f = minus(Q + v) is the v-component of the physical flux vector

F (vQ) equiv(

Fu

Fv

)

=

(

0

f

)

(56)

Note that F has the same structure as (u v)T Borrowing from the presentation for theBSSN system we write the analogous semidiscrete scheme on each subdomain D

k for themodel system

parttu = Q+ av minus u3 + g(t) (57a)

parttv = DQ+Dv minus (u+ v)Q2 + v2u2 + h(t) +Mminus1ℓk(fh minus f lowast)∣

∣

∣

bk

ak (57b)

Q = Du+Mminus1ℓk(ulowast minus uh)∣

∣

∣

bk

ak (57c)

Here we have suppressed the subinterval label k from all variables except for the vectorℓk of Lagrange polynomial values Moreover following the guidelines discussed above thenumerical fluxes are given by

f lowast = fh+1 +

radica

2

[[

vh]]

ulowast = uh minus1

2

[[

uh]]

(58)

Appendix C analyzes the stability of our scheme for a more general numerical flux choiceas applied to (53) with the nonlinear and source terms dropped

IV RESULTS FROM NUMERICAL SIMULATIONS

This section presents results found by numerically solving both the model system (53)and BSSN system (4) with the dG scheme presented in Sec III

16

10minus10

10minus6

10minus2

∆uinfin

10minus10

10minus6

10minus2

∆vinfin

0 500 1000 1500 2000 2500 3000 3500 400010

minus10

10minus6

10minus2

∆(partxu)infin

t

N = 3N = 6N = 9N = 12

FIG 1 Spectral convergence of fields for model PDE Respectively for N = 3 6 9 12

a timestep of ∆t = 00578 00178 00084 00049 has been chosen for stability and accuracy In the

title headings for example ∆u equiv unumer minus uexact

A Simulations of the model system

The semi-discrete scheme (57) has been integrated with the classical fourth-order Runge-Kutta method When integrating this system we have first constructed Q at each Runge-Kutta stage and then substituted into the evolution equations (57ab) for u and v Theproblem has been solved on a computational domain [0 4π] comprised of two subdomainswith a timestep chosen small enough for stability The initial data has been taken from thefollowing exact solution to (53)

uprimeexact(t x) =1

2

[

sin(xminus microminust)minus sin(xminus micro+t)]

(59a)

vexact(t x) =1

2radica

[

sin(xminus microminust) + sin(xminus micro+t)]

(59b)

g(t x) = u3exact (59c)

h(t x) = (uexact + vexact)(uprimeexact)

2 minus v2exactu2exact (59d)

where the speeds microplusmn are found in (54) Specification of the boundary condition at a physicalendpoint amounts to choosing the external state for at the endpoint We have considered

17

8 16 32 64 128

10minus3

10minus2

10minus1

N

∆t

max stable ∆t8N2

FIG 2 Scaling of maximum stable ∆t with N for model PDE

two possibilities (i) the analytic state (Q+ v+) = (Qexact vexact) and (ii) an upwind stateFor example at x = 4π the upwind state is4

Q+ = Qupwind =1

2

[

(Xminus)exact minus (X+)numer

]

v+ = vupwind =1

2radica

[

(Xminus)exact + (X+)numer

]

(60)Either choice of (Q+ v+) leads to similar results and the plots here correspond to theanalytic state Figure 1 clearly shows spectral convergence with increasing polynomial orderN across all fields for the case a = 2 Other values of a including a = 1 for which X+

is a static characteristic field have also been considered with similar results Appendix Cdemonstrates that our proposed scheme for the system (57) with nonlinear and source termsdropped is stable in a semi-discrete sense Nevertheless the fully discrete scheme obtainedvia temporal discretization by the fourth-order Runge-Kutta method is still subject to thestandard absolute stability requirement Namely if microh is any eigenvalue corresponding tothe (linearized) discrete spatial operator then a necessary condition for stability is thatmicroh∆t lies in absolute stability region for Runge-Kutta 4 We here show empirically that theassociated timestep restriction scales like Nminus2 ie ∆t = O(Nminus2) for stability We note thatsuch scaling is welcome in light of the second-order spatial operators which appear in thesystem and suggest a possible worse scaling like Nminus4 Fig 2 plots the maximum stabletimestep for a range of N demonstrating the Nminus2 scaling in line with behavior known fromanalysis of first-order systems [66] This scaling also holds for the BSSN system

4 We remind the reader that unfortunately the plusmn on Xplusmn means something different than the plusmn indicating

exteriorinterior dG states [cf footnote 3]

18

B Simulations of the BSSN system

This subsection documents results for simulations of the unit-mass-parameter (M = 1)Schwarzschild solution (B9) expressed in terms of ingoing Kerr-Schild coordinates Sincethe solution is stationary temporal integration of the semi-discrete scheme has been carriedout with the forward Euler method The r-coordinate domain [04 34] has been split into3 equally spaced subdomains and we have set η = 10 λ = 01 and C = 2 [cf Eq (41)]For all simulations ∆t has been chosen for stability With the chosen λ the inner physicalboundary rmin = 04 is an excision surface At each timestep we have applied an (order2s = 20) exponential filter on the top two-thirds of the modal coefficient set for all fieldsexcept for grr and gθθ For stability we have empirically observed that grr and gθθ must not

be filtered A detailed understanding of this is still lackingIssues related to physical boundary conditions are similar to the one encountered in

Sec IVA for the model problem Similar to before we have retained Eqs (3842) as thechoice of numerical flux even at the endpoints Therefore at an endpoint the specificationof the boundary condition amounts to the choice W+ of external state We have typicallychosen the inner boundary of the radial domain as an excision boundary and in this caseW+ = Wminus is enforced at the inner physical boundary At the outer physical boundaryfor W+ we have again considered two choices (i) Wexact and (ii) Wupwind To enforcechoice (ii) the inverse transformation (17) must be used with incoming characteristic fieldsfixed to their exact values similar to (60) We have tried various versions of choice (ii)and in all cases the resulting simulations have been unstable We therefore present resultscorresponding to choice (i) Although the choice of an analytical external state Wexact

at the outer boundary is stable for our problem such a boundary condition is unlikelyto generalize to more complicated scenarios involving dynamical fields Indeed the issueof outer boundary conditions for the BSSN system is an active area of research with aproper treatment requiring fixation of incoming radiation control of the constraints andspecification of gauge (see Ref [86] for a recent analysis)

For BSSN simulations our main diagnostic is to monitor the Hamiltonian momentumand conformal connection constraints Figure 3 depicts long-time histories of constraintviolations whereas Figs 4 and 5 depict long-time error histories for the individual BSSNfield components From the middle plot in Fig 5 we infer that up to the indicated numericalerror the factor g sin4 θ = grr(gθθ)

2 remains at its initial fixed profile r4 throughout theevolution These figures indicate that the proposed scheme is stable for long times andexhibits spectral converge with increased polynomial order N Similar results are recoveredfromM = 0 Minkowski initial data The stability documented in these plots does not appearto rely on inordinate parameter tuning For example with the fixed parameters describedabove we obtain similar plots if we individually vary (i) rmin over 0325 035 04 0475(values still corresponding to an excision surface for the given choice of λ) (ii) η over1 3 7 10 (iii) s over 8 9 10 With the polynomial order N ranging over 23 26 29 31both stability and qualitatively similar exponential convergence is achieved with a singlesubdomain Likewise adoption of a larger coordinate domain with more subdomains doesnot impact our results Finally we have considered the addition of random noise to all fieldcomponents at the initial time Precisely with the system component χ as an example wehave set

χ(0) 7rarr χ(0) + δχ(0) (61)

where each component (nodal value) of δχ(0) is 10minus5 times a random variable drawn from a

19

10minus9

10minus7

10minus5

10minus3

Hinfin

10minus9

10minus7

10minus5

10minus3

Mrinfin

0 2000 4000 6000 8000 10000

10minus9

10minus7

10minus5

10minus3

Grinfin

t

N = 11N = 14N = 17N = 20

FIG 3 Spectral convergence of constraint violations for M = 1 Kerr-Schild ini-

tial data Respectively for N = 11 14 17 19 a timestep of ∆t ≃ 00041 00026 00018 00013

has been chosen for stability and accuracy

standard normal distribution Such perturbed initial data also gives rise to stable evolutions

V CONCLUSION

We have introduced a discontinuous Galerkin method for solving the spherically reducedBSSN system with second-order spatial operators Our scheme shares similarities withother discontinuous Galerkin methods that use local auxiliary variables to handle high-orderspatial derivatives [66 77ndash79 82 84 85] and which have typically been applied to eitherelliptic parabolic or mixed type problems The key ingredient of a stable dG scheme is anappropriate choice of numerical flux and our particular choice has been motivated by theanalysis presented in Appendix C When used to evolve the Schwarzschild solution in Kerr-Schild coordinates our numerical implementation of the BSSN system (4) is robustly stableand converges to the analytic solution exponentially with increased polynomial order Byapproximating the spatially second-order form of the BSSN system we have not introducedextra fields which are evolved Evolved auxiliary fields result in new constraints which mayspoil stability Our main goal has been stable evolution of the spherically reduced BSSNsystem as a first step towards understanding how a discontinuous Galerkin method might be

20

10minus9

10minus7

10minus5

∆Arrinfin

10minus9

10minus7

10minus5

∆Kinfin

0 2000 4000 6000 8000 10000

10minus9

10minus7

10minus5

max(∆Brinfin ∆Γrinfin)

t

N = 11N = 14N = 17N = 20

FIG 4 Spectral convergence of solution for M = 1 Kerr-Schild initial data

Timestep choices are described in the caption for Fig 3 In the title headings for example

∆Arr equiv (Arr)numer minus (Arr)exact

applied to the full BSSN system Towards that goal we now discuss treatment of singularitiesand generalization of the described dG method to higher space dimension

To deal with the fixed Schwarzschild singularity we have used excision which is easy inthe context of the spherically reduced BSSN system However excision for the binary blackhole problem in full general relativity requires attention to the technical challenge of hori-zon tracking State-of-the-art BSSN codes avoid such complication relying instead on themoving-puncture technique While the moving-puncture technique does involve mild centralsingularities it may still prove amenable to spectral methods Indeed spectral methods fornon-smooth problems is well-developed in both theory and for complex applications Sincethe moving-puncture technique can be performed in spherical symmetry [60] a first-steptoward a spectral moving-puncture code would be to implement a moving puncture withthe nodal dG method described here Such an implementation may adopt Legendre-Gauss-Radau nodes on the innermost subdomain thereby ensuring that the physical singularitydoes not lie on a nodal point (in much the same way finite difference codes use a staggeredgrid) Beyond traditional excision and moving punctures one might also retain a smoothproblem for which spectral methods are especially well-suited via the turducken approach[87 88] to singularities

21

10minus11

10minus9

10minus7

10minus5

∆χinfin

10minus11

10minus9

10minus7

10minus5

max(∆grrinfin ∆gθθinfin)

0 2000 4000 6000 8000 10000

10minus11

10minus9

10minus7

10minus5

max(∆αinfin ∆βrinfin)

t

N = 11N = 14N = 17N = 20

FIG 5 Spectral convergence of solution violations for M = 1 Kerr-Schild initial

data See the caption of Fig 4 for details

Discontinuous Galerkin methods for hyperbolic problems in two and three space dimen-sions are well-developed A generalization of the method described here to three-dimensionsand the full BSSN system would likely rely on an unstructured mesh Appropriate local poly-nomial expansions for the subelements are well-understood as are choices for the numericalfluxes which would now live on two-dimensional faces rather than single points Whether ornot it would ultimately prove successful generalization of our dG method to a higher dimen-sion would rely on an established conceptual framework Further computational advancesof relevance to a generalization of our dG method to the full BSSN system (possibly includ-ing matter) may include mesh hp-adaptivity local timestepping shock capturing and slopelimiting techniques [66] Moreover recent work [89] indicates that enhanced performancewould be expected were our scheme implemented on graphics processor units

VI ACKNOWLEDGMENTS

We thank Nick Taylor for discussions about treating second-order operators with spectralmethods Benjamin Stamm for helping polish up a few parts of Appendix C Khosro Shah-bazi for discussions on LDG and IP methods We also acknowledge helpful conversationswith David Brown and Manuel Tiglio about the BSSN system and previous work on its nu-

22

merical implementation We gratefully acknowledge funding through grants DMS 0554377and DARPAAFOSR FA9550-05-1-0108 to Brown University and NSF grant PHY 0855678to the University of New Mexico

Appendix A Hyperbolicity of the first-order system

This appendix analyzes the matrix A(u) appearing in (10) in order to construct thecharacteristic fields (14) In matrix form the sector (8) of the principal part of (10) reads asfollows

partt

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

=

βr 0 minus 4λα3grr

0 0 λβr

6(grr)2λβr

3gθθgrr0 4λ

3grr

0 βr 0 23grrαχ

13α minus αχ

3grr

αχ3gθθ

minus23χ 0

0 0 βr 0 0 0 0 minus χgrr

0

0 0 minus 4α3grr

βr 0 βr

6(grr)2βr

3gθθgrr0 4

3grr

0 0 23αχ 0 βr minus βrχ

3grrminus2βrχ

3gθθ0 minus2

3χ

0 minus2α 0 0 0 23βr minus2grrβr

3gθθ0 4

3grr

0 gθθαgrr

0 0 0 minusgθθβr

3grr13βr 0 minus2

3gθθ

0 0 minus2α 0 0 0 0 βr 0

34

0 0 0 0 0 0 0 βr

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

prime

(A1)which defines the matrix A(u) appearing in (9) and so also the matrix A(u) in (10) Notethat in the last equation the matrix within the square brackets is minusA(u) For certainconfigurations of u and λ the system (10) is strongly hyperbolic [76] that is A(u) hasa complete set of eigenvectors and real eigenvalues Indeed five eigenpairs of A(u) aretrivially recovered upon inspection of A(u)rsquos leading 5times5 diagonal block These correspondto eigenvalue 0 and the left eigenspace ξj = eTj 1 le j le 5 where ej are the canonical

basis vectors Since each component of u arises as eTj W each is also a characteristic field

The remaining nine eigenpairs are determined by A(u) The eigenvalues of A(u) are

micro1 = 0 micro23 = minusβr microplusmn4 = minusβr plusmn

radic

2αχ

grr microplusmn

5 = minusβr plusmn α

radic

χ

grr microplusmn

6 = minusβr plusmnradic

λ

grr

(A2)

23

and the corresponding left eigenvectors are

x1 = (0 0 0 0 0 gθθ 2grr 0 0) (A3a)

x2 =

(

0 0 0 grr2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3b)

x3 =

(

grrλ 0 0 0

2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3c)

xplusmn4 =

(

0 0plusmnradic

2αgrrχ

0 0 0 0 1 0

)

(A3d)

xplusmn5 =

(

0∓ 3radicgrrχ

plusmn2

radic

grrχ 2grr

1

χminus 1

grr1

gθθ 0 0

)

(A3e)

xplusmn6 =

(

minus3

4

grrλ 0plusmn α

radicλgrr

(2αχminus λ) 0 0minus βr

8(βrgrr ∓radicλgrr)

minus βrgrr

4gθθ(βrgrr ∓radicλgrr)

αχ

(2αχminus λ)plusmnradic

grrλ

)

(A3f)

where for example xplusmn5 A(u) = microplusmn5 x

plusmn5 Assuming that grr gθθ χ and α are everywhere strictly

positive the eigenvalues are real and the eigenvectors are linearly independent providedthat (15) holds These eigenvectors are easily extended to eigenvectors of A(u) e g asxplusmn6 rarr (01times5 x

plusmn6 ) Then for example the characteristic field

Xplusmn6 equiv (01times5 x

plusmn6 )W = xplusmn6WvQ (A4)

and similarly Xplusmnj = xplusmnj WvQ for j = 4 5 and Xk = xkWvQ for k = 1 2 3 The characteristic

speeds for these fields are microk and microplusmnj With this convention the speeds listed in Table I

correspond to the Xk and Xplusmnj in (14)

Appendix B Schwarzschild solution in conformal Kerr-Schild coordinates

In Kerr-Schild coordinates here the system directly related to incoming Eddington-Finkelstein null coordinates the line element for the Schwarzschild solution reads

ds2 = minusα2dt2 + (1 + 2MR)(dR + βRdt)2 +R2dθ2 +R2 sin2 θdφ2 (B1)

where R is the area radius α = (1 + 2MR)minus12 is the lapse and βR = 2M(R + 2M) isthe shift vector The physical spatial metric gab is the spatial part of this line element

To define the corresponding solution to the BSSN system we use equation gab = χgab todefine the following relationship between line elements

dr2 + r2(dθ2 + sin2 θdφ2) = χ[(1 + 2MR)dR2 +R2dθ2 +R2 sin2 θdφ2] (B2)

so that

χ

(

1 +2M

R

)(

dR

dr

)2

= 1 χR2 = r2 (B3)

Then we have(

1 +2M

R

)12dR

R=dr

r (B4)

24

with integration yielding

r =R

4

(

1 +

radic

1 +2M

R

)2

e2minus2radic

1+2MR (B5)

where the constant of integration has been chosen so that the R r rarr infin limits are consistentThe second relation in (B3) shows that

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR =rArr χminus4 =2eradic

1+2MRminus1

1 +radic

1 + 2MR (B6)

The extrinsic curvature tensor is specified by the expression forK given in (B9h) the identityK = KR

R + 2Kθθ and

Kθθ =

(

1 +2M

R

)minus122M

R2 (B7)

Since KRR = Kr

r we compute that

Krr = K minus 2Kθ

θ = minus(

1 +2M

R

)minus12(R +M

R + 2M

)

2M

R2 (B8)

Next since Krr = grrKrr = χminus1Kr

r we have Krr = Arr+

13grrK This implies Arr = Kr

r minus 13K

from which we get (B9g) In all we have

α =

(

1 +2M

R

)minus12

(B9a)

βr = βR dr

dR= χ12

(

1 +2M

R

)minus122M

R(B9b)

grr = 1 (B9c)

gθθ = r2 = χR2 (B9d)

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR (B9e)

Br = 0 (B9f)

Arr = minus(

1 +2M

R

)minus124M

3R2

(

2R + 3M

R + 2M

)

(B9g)

K =

(

1 +2M

R

)minus32(

1 +3M

R

)

2M

R2(B9h)

Γr = minus2

r= minus 2

χ12R (B9i)

To differentiate these expressions with respect to r we use the identity

dR

dr= χminus12

(

1 +2M

R

)minus12

(B10)

along with the chain rule

25

Appendix C Stability of the model system

The following stability analysis for the model system (53) has been inspired by [78 79]After dropping all nonlinear source terms the system (53) becomes

parttu = uprime + av (C1a)

parttv = uprimeprime + vprime (C1b)

This section analyzes the stability of (C1) considering both the continuum system itself aswell as its semi-discrete dG approximation The latter analysis offers some insight into theempirically observed stability of our dG scheme for the spherically reduced BSSN equations

1 Analysis for a single interval

Throughout we work with the L2 inner product and norm

(f g)D =

int

D

fg fD =radic

(f f)D (C2)

where D is the spatial coordinate interval (here D may represent a subdomain Dk or the

whole domain Ω) and we have suppressed all integration measures For the continuummodel we will establish the following estimate

uprime(T middot)2D+ av(T middot)2

Dle C(T )

(

uprime(0 middot)2D+ av(0 middot)2

D

)

(C3)

where the time-dependent constant C(T ) is determined solely by the choice of boundaryconditions To show (C3) we first change variables with v =

radicav thereby rewriting (C1)

in the following symmetric form

parttu = uprime +radicav (C4a)

parttv =radicauprimeprime + vprime (C4b)

Equations (C4ab) then imply

1

2partt

int

D

(uprime)2 =

int

D

uprime(uprimeprime +radicavprime) =

int

D

radicauprimevprime +

1

2

int

partD

(uprime)2 (C5a)

1

2partt

int

D

(v)2 =

int

D

v(radicauprimeprime + vprime) = minus

int

D

radicauprimevprime +

1

2

int

partD

(v2 + 2radicauprimev) (C5b)

Here vvprime and uprimeuprimeprime have been expressed as exact derivatives and then integrated to boundaryterms the second equation employs an extra integration by parts and with only one spacedimension

int

partDdenotes a difference of endpoint evaluations Addition of Eqs (C5ab) gives

1

2partt

int

D

[

v2 + (uprime)2]

=1

2

int

partD

[

v2 + (uprime)2 + 2radicauprimev

]

(C6)

Substitutions with the identities

[

v2 + (uprime)2]

=1

2

[

(v + uprime)2 + (v minus uprime)2]

2uprimev =1

2

[

(v + uprime)2 minus (v minus uprime)2]

(C7)

26

and replacements to recover the original variable v = vradica yield

1

2partt

int

D

[

av2 + (uprime)2]

=1 +

radica

4

int

partD

(radicav + uprime)2 +

1minusradica

4

int

partD

(radicav minus uprime)2 (C8)

From (C8) we deduce that the time-dependent constant C(T ) in (C3) must satisfy

∣

∣

∣

∣

∣

∣

1 +

int T

0

[

1+radica

2

int

partD(radicav + uprime)2 + 1minus

radica

2

int

partD(radicav minus uprime)2

]

dt

uprime(0 middot)2 + av(0 middot)2

∣

∣

∣

∣

∣

∣

le C(T ) (C9)

For periodic boundary conditions we may choose C(T ) = 1 Moreover if a ge 1 anduprime = minusradic

av is specified at partD+ then uprime(t middot)2 + av(t middot)2 decaysStill working on a single interval (subdomain) we now consider the semi-discrete scheme

for (C4) i e (57) with all nonlinear source terms dropped and with v replaced by vradica

Derivation of a formula analogous to (C8) is our first step toward establishing L2 stabilityof the semi-discrete scheme While (57) features vectors for example u(t) taking values atthe Legendre-Gauss-Lobatto nodal points here we work with the numerical solution as apolynomial for example uh(t x) These two representations are related by the Lagrange in-terpolating polynomials for the nodal set here taken to span both the space of test functionsand the space of basis functions Our scheme is

int

Dk

ψparttuh =

int

Dk

ψ(Qh +radicavh) (C10a)

int

Dk

ξparttvh = minusint

Dk

ξprime(radicaQh + vh) +

int

partDk

ξ(radicaQlowast + vlowast) (C10b)

int

Dk

ϕQh =

int

Dk

ϕuprimeh +

int

partDk

ϕ (ulowast minus uh) (C10c)

where ψ ξ and ϕ are polynomial test functions These test functions are arbitrary exceptthat they must be degree-N polynomials In (C10) the variables uh vh and Qh should alsocarry a superscript k but we have suppressed this Derivation of a formula analogous to(C8) is complicated by the fact that Qh is not evolved Nevertheless at a given instant t wecan assemble Qh from (C10c)

Mimicking the calculation (C5b) from the continuum case we first use (C10b) with ξ = vhto write

1

2partt

int

Dk

v2h = minusint

Dk

(radicaQh + vh)v

primeh +

int

partDk

(radicaQlowast + vlowast)vh

= minusint

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2(radicaQlowast + vlowast)vh minus v2h

]

(C11)

The right-hand side of (C5a) is analogous to

1

2partt

int

Dk

Q2h =

int

Dk

QhparttQh (C12)

However since Qh is not evolved the term parttQh must be given a suitable interpretationOn the right hand side of (C10c) only uh u

primeh and u

lowast necessarily depend on time since the

27

test function ϕ need not be time-dependent Furthermore ulowast is explicitly given as a linearcombination of uh as seen in Eq (C21c) below Choosing ϕ = ℓj taking the time derivativeof (C10c) and appealing to the commutivity of mixed partial derivatives we therefore arriveat

int

Dk

ℓjparttQh =

int

Dk

ℓj(parttuh)prime +

int

partDk

ℓj(

(parttu)lowast minus parttuh

)

(C13)

where (parttu)lowast depends on parttuh in precisely the same way that ulowast depends on uh We have

written ℓj rather than ϕ in the last equation to emphasize that the result also holds for anylinear combination of ℓj (for example ϕ) and even for time-dependent combinations SinceQh is itself such a combination we obtain

1

2partt

int

Dk

Q2h =

int

Dk

Qh(parttuh)prime +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

Qh(Qprimeh +

radicavprimeh) +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2((parttu)lowast minus parttuh)Qh +Q2

h

]

(C14)

Addition of (C11) and (C14) gives

1

2partt

int

Dk

(Q2h + v2h) =

1

2

int

partDk

[

Q2h minus v2h + 2(

radicaQlowast + vlowast)vh + 2((parttu)

lowast minus parttuh)Qh

]

(C15)

the aforementioned analog of (C8) This formula holds on a single subdomain Dk and we

now combine multiple copies of it one for each value of k

2 Analysis for multiple intervals

To facilitate combination of (C15) over all k we change notation At every subdomaininterface Ik+12 equiv partDk cap partDk+1 let the superscripts L and R denote field values respectivelytaken from the left and right Then the fields evaluated at Ik+12 which belong to D

k will beuLk+12 v

Lk+12 and Q

Lk+12 while those belonging to D

k+1 will be uRk+12 vRk+12 and Q

Rk+12

However at Ikminus12 the values taken from D

k are uRkminus12 vRkminus12 and QR

kminus12 Note that we

have also replaced the subscript h denoting a numerical solution with k plusmn 12 denotingthe location of the endpoint value of the numerical solution With this notation we define

∆Lα =

1

2

[

(QLα)

2 minus (vLα)2]

+(radic

aQlowastα + vlowastα

)

vLα +[

(parttuα)lowast minus parttu

Lα

]

QLα (C16)

and similarly for ∆Rα The same numerical fluxes appear in both ∆L

α and ∆Rα (ie each

numerical flux takes the same value on either side of an interface) whence fluxes like Qlowastα do

not carry an L or R superscript In terms of these definitions (C15) becomes

1

2partt

int

Dk

(Q2h + v2h) = ∆L

k+12 minus∆Rkminus12 (C17)

28

Summation over all Dk yields

1

2partt

kmaxsum

k=1

int

Dk

(Q2h + v2h) =

kmaxminus1sum

k=1

(

∆Lk+12 minus∆R

k+12

)

+∆Lkmax+12 minus∆R

12

=

kmaxminus1sum

k=1

(

∆Lh minus∆R

h

)∣

∣

Ik+12 +∆Lkmax+12 minus∆R

12 (C18)

We have reverted to h-notation denoting the numerical solution since the LR superscriptsindicate unambiguously the relevant domain used for evaluation at Ik+12

We again seek an estimate of the form

kmaxsum

k=1

(

Qh(T middot)2Dk + avh(T middot)2Dk

)

le C(T )kmaxsum

k=1

(

Qh(0 middot)2Dk + avh(0 middot)2Dk

)

(C19)

that is essentially the same as the one (C3) considered in the continuum case Assume thatthe chosen boundary conditions ensure ∆L

kmax+12 minus ∆R12 is bounded by a time-dependent

constant which does not depend on the numerical parameters N and h (subdomain width)Establishment of stability then amounts to showing that the remaining sum over interfaceterms in (C18) is non-positive whence this remaining sum is consistent with C(T ) le 1although the boundary conditions may give rise to a different bound In fact we will choosethe numerical fluxes such that each individual interface term is non-positive At interfaceIk+12 and in LR notation the jump and average of vh for example are

1

2

(

v+ + vminus)

equiv vh =1

2

(

vLk+12 + vRk+12

)

(C20a)

nminusvminus + n+v+ equiv[[

vh]]

= vLk+12 minus vRk+12 (C20b)

Consider numerical fluxes of the form

Qlowast = Qh minusτQ2

[[

Qh

]]

(C21a)

vlowast = vh minusτv2

[[

vh]]

(C21b)

ulowast = uh minusτu2

[[

uh]]

(C21c)

(parttu)lowast = parttuh minus

τu2

[[

parttuh]]

(C21d)

where (C21c) induces (C21d) and where the penalty parameters τu τv and τQ are realnumbers The fluxes defined in (58) correspond to τu = 1 τv = 1 +

radica and τQ = 0 In

terms of these quantities the kth interface contribution in (C18) is

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

parttuh]]

minus τu2

[[

parttuh]][[

Qh

]]

(C22)

where we have suppressed the k dependence of the right-hand side Now consider the term[[

parttuh]]

Because parttuh and Qh+radicavh are both polynomials of degree N Eq (C10a) implies

29

τu

τ Q

minus1 0 1minus1

minus05

0

05

1

τu

minus1 0 1minus1

minus05

0

05

1

FIG 6 Stable evolutions for the model system For fixed τv = 10minus6 and τv = 1 +radic2

respectively the left and right plots depict stable choices (determined empirically) of τu and τQfor the linear model system (C1) The stable regions are colored black but the jagged edges result

from the discretization of the (τu τQ)-plane

the vector equation parttu = Q +radicav that is pointwise equivalence on the nodal points of

Dk which in turn implies

[[

parttuh]]

=[[

Qh+radicavh]]

Upon substituting this identity into thelast equation we arrive at an expression which features only vh and Qh

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

Qh +radicavh]]

minus τu2

[[

Qh +radicavh]][[

Qh

]]

(C23)

The identities vh[[

vh]]

= 12

[[

v2h]]

and[[

Qh +radicavh]]

=[[

Qh

]]

+radica[[

vh]]

then simplify(C23) to

(∆Lh minus∆R

h )∣

∣

Ik+12 = minusτv2

[[

vh]]2 minus

radica(τu + τQ)

2

[[

Qh

]][[

vh]]

minus τu2

[[

Qh

]]2 (C24)

The role of a penalty parameter is now clear Positive values of τv penalize jumps in vhthrough a negative contribution to the energy Likewise positive values of τu penalizejumps in Qh through a negative contribution to the energy However because the sign of[[

Qh

]][[

vh]]

can be positive or negative only the choice τQ = minusτu yields an expression for(∆L

h minus ∆Rh )|Ik+12 which is manifestly negative for τu ge 0 and τv ge 0 A simple estimate

based on Youngrsquos inequality with ε (that is 2αβ le εminus1α2 + εβ2 where α β ge 0 and ε gt 0)shows that for τQ = 0 the choice τv ge aτu4 also yields stability

Figure 6 depicts certain choices of stable penalty parameters for the linear model systemevolved to tfinal = 1000 (with a = 2 N = 10 and ∆t ≃ 00553) as determined empiricallywith simulations similar to those described in Sec IVA The left plot corresponds to asmall τv = 10minus6 for which the choice τu = 1 τQ = 0 is not stable as expected from thetheoretical analysis However the right plot corresponds to τv = 1 +

radica for which τu = 1

τQ = 0 is stable Motivated by the numerical flux choices (3842) used for the BSSN system

30

(4) we have (as mentioned above) set τu = 1 τv = 1 +radica and τQ = 0 in simulations of

the nonlinear model (53) For the nonlinear model system (53) the theoretically motivatedchoice τQ = minusτu also yields numerically stable evolutions when τu ge 0 and τv ge 0

For the nonlinear systems (4) and (53) we do not attempt a formal stability proofNevertheless the results of this appendix have served as a guide for our choices of penaltyparameters For the BSSN system (4) u v and Q are block indices [cf Eq (6)] Similarto the model problem we have penalized Q with τu = 1 with τv chosen large enough toheuristically overcome the cross-terms of indefinite size that arise from τQ = 0 (we interpretequations like τu = 1 componentwise) An analogous choice ldquoτQ = minusτurdquo for the BSSNsystem might be possible but would be considerably more complicated Indeed such achoice likely entails a matrix of penalty parameters but we do not give the details here

[1] F Pretorius Evolution of Binary Black-Hole Spacetimes Phys Rev Lett 95 121101 (2005)

(4 pages)

[2] F Pretorius Simulation of binary black hole spacetimes with a harmonic evolution scheme

Class Quantum Grav 23 S529-S552 (2006)

[3] M Campanelli C O Lousto P Marronetti and Y Zlochower Accurate Evolutions of Or-

biting Black-Hole Binaries without Excision Phys Rev Lett 96 111101 (2006) (4 pages)

[4] J G Baker J Centrella D-I Choi M Koppitz and J van Meter Gravitational-Wave

Extraction from an Inspiraling Configuration of Merging Black Holes Phys Rev Lett 96

111102 (2006) (4 pages)

[5] M Campanelli C O Lousto and Y Zlochower Last orbit of binary black holes Phys Rev D

73 061501(R) (2006) (5 pages)

[6] F Herrmann I Hinder D Shoemaker and P Laguna Unequal mass binary black hole plunges

and gravitational recoil Class Quantum Grav 24 S33-S42 (2007)

[7] P Diener F Herrmann D Pollney E Schnetter E Seidel R Takahashi J Thornburg and

J Ventrella Accurate Evolution of Orbiting Binary Black Holes Phys Rev Lett 96 121101

(2006) (4 pages)

[8] M A Scheel H P Pfeiffer L Lindblom L E Kidder O Rinne and S A Teukolsky Solving

Einsteinrsquos Equations With Dual Coordinate Frames Phys Rev D 74 104006 (2006)

[9] U Sperhake Binary black-hole evolutions of excision and puncture data Phys Rev D 76

104015 (2007) (20 pages)

[10] B Brugmann J A Gonzalez M Hannam S Husa U Sperhake and W Tichy Calibration

of moving puncture simulations Phys Rev D 77 024027 (2008) (25 pages)

[11] P Marronetti W Tichy B Brugmann J Gonzalez M Hannam S Husa and U Sperhake

Binary black holes on a budget simulations using workstations Class Quantum Grav 24

S43-S58 (2007)

[12] Z B Etienne J A Faber Y T Liu S L Shapiro and T W Baumgarte Filling the holes

Evolving excised binary black hole initial data with puncture techniques Phys Rev D 76

101503(R) (2007) (5 pages)

[13] B Szilagyi D Pollney L Rezzolla J Thornburg and J Winicour An explicit harmonic code

for black-hole evolution using excision Class Quantum Grav 24 S275-S293 (2007)

[14] L Boyle M Kesden and S Nissanke Binary-Black-Hole Merger Symmetry and the Spin

Expansion Phys Rev Lett 100 151101 (2008) (4 pages)

31

[15] M A Scheel M Boyle T Chu L E Kidder K D Matthews and H P Pfeiffer High-

accuracy waveforms for binary black hole inspiral merger and ringdown Phys Rev D 79

024003 (2009) (14 pages)

[16] M Hannam Status of black-hole-binary simulations for gravitational-wave detection

Class Quantum Grav 26 114001 (2009) (21 pages)

[17] I Hinder The current status of binary black hole simulations in numerical relativity

Class Quantum Grav 27 114004 (2010) (19 pages)

[18] B P Abbott et al LIGO The Laser Interferometer Gravitational-Wave Observatory

Rep Prog Phys 72 076901 (2009) (25 pages)

[19] F Acernese et al The status of VIRGO Class Quantum Grav 23 S63-S69 (2006)

[20] F Acernese et al Virgo status Class Quantum Grav 25 184001 (2008) (9 pages)

[21] R Arnowitt S Deser and C W Misner The Dynamics of General Relativity in Gravitation

an introduction to current research edited by L Witten (Wiley New York 1962) Chapter 7

227-265 Also available as gr-qc0405109

[22] J W York Jr Kinematics and Dynamics of General Relativity in Sources of Gravitational

Radiation edited by L L Smarr (Cambridge University Press Cambridge 1979)

[23] L E Kidder M A Scheel and S A Teukolsky Extending the lifetime of 3D black hole

computations with a new hyperbolic system of evolution equations Phys Rev D 64 064017

(2001) (13 pages)

[24] G Calabrese J Pullin O Sarbach and M Tiglio Convergence and stability in numerical

relativity Phys Rev D 66 041501(R) (2002) (4 pages)

[25] S Frittelli and O Reula On the Newtonian Limit of General Relativity Commun Math

Phys 166 221-235 (1994)

[26] M Shibata and T Nakamura Evolution of three-dimensional gravitational waves Harmonic

slicing case Phys Rev D 52 5428-5444 (1995)

[27] Y Choquet-Bruhat and J W York Geometrical well posed systems for the Einstein equations

C R Acad Sci Paris t 321 Serie I 1089-1095 (1995)

[28] A Abrahams A Anderson Y Choquet-Bruhat and J W York Jr Einstein and Yang-Mills

Theories in Hyperbolic Form without Gauge Fixing Phys Rev Lett 75 3377-3381 (1995)

[29] C Bona J Masso E Seidel and J Stela New Formalism for Numerical Relativity

Phys Rev Lett 75 600-603 (1995)

[30] M H P M van Putten and D M Eardley Nonlinear wave equations for relativity

Phys Rev D 53 3056-3063 (1996)

[31] S Frittelli and O A Reula First-Order Symmetric Hyperbolic Einstein Equations with Ar-

bitrary Fixed Gauge Phys Rev Lett 76 4667-4670 (1996)

[32] H Friedrich Hyperbolic reductions for Einsteinrsquos equations Class Quantum Grav 13 1451-

1469 (1996)

[33] F B Estabrook R S Robinson H D Wahlquist Hyperbolic equations for vacuum gravity

using special orthonormal frames Class Quantum Grav 14 1237-1247 (1997)

[34] M S Iriondo E O Leguizamon and O A Reula Einsteinrsquos Equations in Ashtekarrsquos Vari-

ables Constitute a Symmetric Hyperbolic System Phys Rev Lett 79 4732-4735 (1997)

[35] A Anderson Y Choquet-Bruhat and J W York Einstein-Bianchi hyperbolic system for

general relativity Topol Meth Nonlinear Anal 10 353-373 (1997)

[36] T W Baumgarte and S L Shapiro Numerical integration of Einsteinrsquos field equations

Phys Rev D 59 024007 (1998) (7 pages)

[37] M A G Bonilla Symmetric hyperbolic systems for Bianchi equations Class Quantum

7

field speed

X1 micro1 = 0

X23 micro23 = minusβr

Xplusmn4 microplusmn

4 = minusβr plusmnradic

2αχgrr

Xplusmn5 microplusmn

5 = minusβr plusmn αradic

χgrr

Xplusmn6 microplusmn

6 = minusβr plusmnradic

λgrr

TABLE I Characteristic speeds These speeds are the eigenvalues listed in (A2)

C Hyperbolicity and characteristic fields

Although our numerical scheme deals directly with the second-order spatial operatorsappearing in (4) we first consider the hyperbolicity of the corresponding first-order system(10) The characteristic fields and their speeds are found by instantaneously ldquofreezingrdquothe fields u in A(u) to some value u0 corresponding to a linearization around a uniformstate Below we continue to write u for simplicity with the understanding that u is reallythe background solution u0 Of primary interest is the range of u0 for which the system isstrongly hyperbolic [73ndash76]

Appendix A shows that the characteristic fields corresponding to (4) are as follows (i)all components of u (each with speed 0) and (ii) the fields

X1 = gθθQgrr + 2grrQgθθ (14a)

X2 = grrΓr +

2

χQχ minus

1

2grrQgrr minus

1

gθθQgθθ (14b)

X3 =grrλBr +

2

χQχ minus

1

2grrQgrr minus

1

gθθQgθθ (14c)

Xplusmn4 = plusmn

radic

2αgrrχ

K +Qα (14d)

Xplusmn5 = ∓ 3radic

grrχArr plusmn 2

radic

grrχK + 2grrΓ

r +1

χQχ minus

1

grrQgrr +

1

gθθQgθθ (14e)

Xplusmn6 = minus3

4

grrλBr plusmn α

radicλgrr

(2αχminus λ)K minus βr

8(βrgrr ∓radicλgrr)

Qgrr

minus βrgrr

4gθθ(βrgrr ∓radicλgrr)

Qgθθ +αχ

(2αχminus λ)Qα plusmn

radic

grrλQβr (14f)

with the speeds listed in Table I To ensure strong hyperbolicity we must necessarily require

λ gt 0 (βr)2grr minus λ 6= 0 2αχminus λ 6= 0 (15)

as shown in in Appendix A where further conditions are also given When λ = 1 thehyperbolicity condition of Ref [60] is recovered In fact the system could be recast assymmetric hyperbolic Indeed as it involves one spatial dimension the relevant symmetrizercan be constructed via polar decomposition of the diagonalizing similiarity transformationHowever we will not exploit this possibility

8

This system admits an inner excision boundary provided

βr ge max

(

radic

2αχ

grr

radic

α2χ

grr

radic

λ

grr

)

(16)

holds at the inner boundary This condition ensures each characteristic field has a nonposi-tive speed at the inner boundary and therefore the inner boundary is an excision boundaryat which no boundary conditions are needed The extra flexibility afforded by the param-eter λ could be used to maintain rigorous hyperbolicity by moving the points at whichthe conditions in (15) are violated outside of the computational domain Furthermore forλ = 1 Eq (16) conceivably fails or is only satisfied close to r = 0 where field gradients are

prohibitively large The troublesome X+6 gauge mode has a positive speed minusβr +

radic

λgrrIndeed for the conformally flat Kerr-Schild system considered in section IVB an inner ex-cision boundary is only possible provided λ is small enough

The transformation (14) can be inverted in order to express the fundamental fields interms of the characteristic fields

Br = minus1

6

λ

grrgθθ

[

(βr)2

(βr)2grr minus λ

]

X1 +2

3

λαχ

grr(2αχminus λ)(X+

4 +Xminus4 )minus

2

3

λ

grr(X+

6 +Xminus6 )

(17a)

Arr =1

3

radic

grrχ

2α(X+

4 minusXminus4 )minus

radicgrrχ

6(X+

5 minusXminus5 ) (17b)

K =

radic

χ

8αgrr(X+

4 minusXminus4 ) (17c)

Γr = minus1

6

1

grrgθθ

[

(βr)2

(βr)2grr minus λ

]

X1 +1

grr(X2 minusX3) +

2

3

αχ

grr(2αχminus λ)(X+

4 +Xminus4 )

minus 2

3

1

grr(X+

6 +Xminus6 ) (17d)

Qχ =1

12

χ

grrgθθ

[

4(βr)2grr minus 3λ

(βr)2grr minus λ

]

X1 +χ

2X3 minus

1

3

αχ2

(2αχminus λ)(X+

4 +Xminus4 ) (17e)

+χ

3(X+

6 +Xminus6 ) (17f)

Qgrr =2(βr)2grr minus 3λ

6gθθ((βr)2grr minus λ)X1 +

4

3grrX2 minus grrX3 +

2

3

αχgrr(2αχminus λ)

(X+4 +Xminus

4 )

minus 1

3grr(X

+5 +Xminus

5 )minus2

3grr(X

+6 +Xminus

6 ) (17g)

Qgθθ =

[

1

4grr+

(βr)2

12((βr)2grr minus λ)

]

X1 minus2

3gθθX2 +

1

2gθθX3 minus

1

3

αχgθθ(2αχminus λ)

(X+4 +Xminus

4 )

+1

6gθθ(X

+5 +Xminus

5 ) +1

3gθθ(X

+6 +Xminus

6 ) (17h)

Qα =1

2(X+

4 +Xminus4 ) (17i)

Qβr =βrλ

8grrgθθ((βr)2grr minus λ)X1 minus

λ

(2αχminus λ)

radic

αχ

8grr(X+

4 minusXminus4 ) +

1

2

radic

λ

grr(X+

6 minusXminus6 )

(17j)

9

We will refer to this inverse transformation when discussing outer boundary conditions forour numerical simulations in Sec IVB

III DISCONTINUOUS GALERKIN METHOD

This section describes the nodal discontinuous Galerkin method used to numerically solve(4) We adopt a method-of-lines strategy and here describe the relevant semi-discrete schemewhile leaving the temporal dimension continuous To approximate (4) we follow the generalprocedure first introduced in Ref [82] Our approach defines local auxiliary variables Q = uprimeand rewrites the spatially second-order system (4) as the first-order system (13a) Once weuse (13b) to eliminate Q from (13a) we recover the primal equations (4) The auxiliaryvariable approach was later generalized and coined the local discontinuous Galkerin (LDG)method in Ref [77] We may refer to our particular scheme as an LDG method but notethat many variations exist in the literature We stress that in LDG methods Q is not evolvedand is introduced primarily to assist in the construction of a stable scheme

Equations (12) and (13a) imply that the physical flux function is

F (W ) =

(

Fu(W )

Fv(W )

)

equiv A(u)WvQ =

(

05times1

f(W )

)

f =

fBr

fArr

fKfΓ

(18)

Only the evolution equations for Br Arr K and Γr give rise to non-zero components in F and we have collected these non-zero components into a smaller vector f = Fv Inspectionof (8) determines these components For example from (8c) we find

fK = minusβrK +χ

grrQα (19)

A Local approximation of the system (13)

Our treatment closely follows [83] but with the equations and notations relevant forthis paper Our computational domain Ω is the closed r-interval [a b] We cover Ω withkmax gt 1 non-overlapping intervals Dk = [ak bk] where a = a1 b = bkmax and bkminus1 = ak fork = 2 middot middot middot kmax

On each interval Dk we approximate each component of the system vector W by a localinterpolating polynomial of degree N For example

χkh(t r) =

Nsum

j=0

χ(t rkj )ℓkj (r) (20)

approximates χ(t r) Throughout this section approximations are denoted by a subscripth (see [66] for the notation) For example Wh and fh are approximations of W and f Although Q = uprime Qh and uprimeh are not necessarily the same In (20) ℓkj (r) is the jth Lagrange

polynomial belonging to Dk

ℓkj (r) =Nprod

i=0i 6=j

r minus rkirkj minus rki

(21)

10

Evidently the polynomial χkh interpolates χ at rkj To define the nodes rkj consider the

mapping from the unit interval [minus1 1] to Dk

rk(u) = ak + 12(1 + u)(bk minus ak) (22)

and the N+1 Legendre-Gauss-Lobatto (LGL) nodes uj The uj are the roots of the equation

(1minus u2)P primeN(u) = 0 (23)

where PN(u) is the Nth degree Legendre polynomial and the physical nodes are simplyrkj = rk(uj) In vector notation the approximation (20) takes the form

χkh(t r) = χk(t)Tℓk(r) (24)

in terms of the column vectors

χk(t) =[

χ(t rk0) middot middot middot χ(t rkN)]T ℓk(r) =

[

ℓk0(r) middot middot middot ℓkN(r)]T (25)

On each open interval (ak bk) sub Dk and for each component of the equations in (13) we

define local residuals measuring the extent to which our approximations satisfy the originalcontinuum system Dropping the subdomain label k on the polynomials and focusing onthe K equation as a representative example the local residual corresponding to (4h) is

minus(RK)kh equivminus parttKh + (βrK prime)h minus

(

χQprimeα

grr

)

h

+

(

χQgrrQα

2g2rr

)

h

minus(

χQgθθQα

grrgθθ

)

h

+

(

QαQχ

2grr

)

h

+

(

3αA2rr

2g2rr

)

h

+

(

1

3αK2

)

h

(26)

Here for example the expressions 2

(βrK prime)h = βrhK

primeh

(

QαQχ

2grr

)

h

=QαhQχh

2grrh (27)

We similarly construct the remaining eight residuals eg (Rgrr)h and (RΓr)h as well as fiveresiduals corresponding to (13b) For example one of these remaining five is

(RQα)kh equiv minusQαh + αprime

h (28)

Let the kth inner product be defined as

(

u v)

Dk equivint bk

akdru(r)v(r) (29)

and consider the expression (ℓkj (RK)kh)Dk We call the requirement that this inner product

vanish forallj the kth Galerkin condition For each component of the system and for each k

2 At this stage the first expression is generically a polynomial of degree 2N minus 1 and the latter is not a

polynomial The conventions adopted in Eq (27) prove useful while working with the residual However

later on in Sec III C to obtain the final form (47) of the numerical approximation corresponding to (26)

we will replace nonlinear terms with degree-N polynomials

11

there is a corresponding Galerkin condition in total 9kmax(N + 1) equations for (13a) and5kmax(N+1) for (13b) Enforcement of the Galerkin conditions on each D

k will not recover ameaningful global solution since they provide no mechanism for coupling the local solutionson the different intervals Borrowing from the finite volume toolbox we achieve couplingthrough integration by parts on r and introduction of the numerical flux f lowast at the interfacebetween subdomains

In (26) we only need to consider (βrK prime)h and (χQprimeαgrr)h as the other terms comprise a

component of the source vector Sh Using integration by parts we write

(

ℓkj (βrK prime)h

)

Dk =minusint bk

akdr[

(

ℓkjβrh

)primeKh

]

+ (βrhKh) ℓ

kj

∣

∣

∣

bk

ak (30a)

(

ℓkj (χQprimeαgrr)h

)

Dk =minusint bk

akdr

[(

ℓkjχh

grrh

)prime

Qαh

]

+

(

χhQαh

grrh

)

ℓkj

∣

∣

∣

bk

ak (30b)

In these formulas we have retained the domain index k on ℓkj while continuing to suppress iton Kh grrh etc Moreover we have suppressed the r-dependence in all terms on the right-hand side Addition of these formulas along with the definition fKh = minus(βrK)h+(χQαgrr)hgives

(

ℓkj (βrK prime)h minus (χQprime

αgrr)h)

Dk = minusint bk

akdr

[

(

ℓkjβrh

)primeKh minus

(

ℓkjχh

grrh

)primeQαh

]

minus fKhℓkj

∣

∣

∣

bk

ak (31)

In lieu of (31) we will instead work with the replacement

(

ℓkj (βrK prime)h minus (χQprime

αgrr)h)

Dk rarr minusint bk

akdr

[

(

ℓkjβrh

)primeKh minus

(

ℓkjχh

grrh

)primeQαh

]

minus f lowastKℓ

kj

∣

∣

∣

bk

ak (32)

This replacement features a component f lowastK of the numerical flux rather than a component

fKh of the boundary flux The numerical flux is determined by (as yet not chosen) functions3

f lowast = f lowast(W+Wminus) (33)

where for example Wminus is an interior boundary value [either W kh (t a

k) or W kh (t b

k)] of theapproximation defined on D

k andW+ is an exterior boundary value [either W kminus1h (t bkminus1) or

W k+1h (t ak+1)] of the approximation defined on either Dkminus1 or Dk+1 We discuss our choice

of numerical flux in the next subsection We now employ additional integration by parts towrite the above replacement as

(

ℓkj (βrK prime)h minus (χQprime

αgrr)h)

Dk rarrint bk

akdrℓkj

(

βrK prime minus χQprimeα

grr

)

h

+ (fKh minus f lowastK) ℓ

kj

∣

∣

∣

bk

ak (34)

Rather than the exact kth Galerkin condition(

ℓkj (RK)kh

)

Dk = 0 forallj for the K component

of (13) on Dk we will instead strive to enforce

(

ℓkj (RK)kh

)

Dk = (fKh minus f lowastK) ℓ

kj

∣

∣

∣

bk

ak forallj (35)

3 In the context of the dG method here + and minus denote ldquoexteriorrdquo and ldquointeriorrdquo and have no relation to

the plusmn using to denote the characteristic fields and speeds in Table I For characteristic fields and speeds

+ and minus mean ldquoright-movingrdquo and ldquoleft-movingrdquo

12

although our treatment of nonlinear terms will lead to a slight modification of these equations(we return to this issue shortly) The other components of (13a) are treated similarly asare the components of (13b) Recall that for example Qα = αprime Formally using the samedG method to solve for Qα we arrive at the replacement

(

ℓkj (RQα)kh

)

Dk rarrint bk

akdrℓkj (minusQαh + αprime

h)minus (αh minus αlowast) ℓkj

∣

∣

∣

bk

ak (36)

which again features a component αlowast of the numerical flux The auxiliary variables areconstructed and used at each stage of temporal integration We then have

(

ℓkj (RQα)kh

)

Dk = (αh minus αlowast) ℓkj

∣

∣

∣

bk

ak forallj (37)

as the corresponding enforced kth Galerkin condition

B Numerical Flux

To further complete our dG scheme we must specify functional forms for the compo-nents of the numerical flux introduced in the previous section We distinguish between thephysical fluxes (components of f) and the auxiliary fluxes (components of u) arising fromthe definition of the auxiliary variables These choices are not independent as the resultingscheme must be stable and consistent Our choice follows [84] which considered diffusionproblems Additional analysis of this flux choice appears in [66 85]

Let us first consider the numerical fluxes corresponding to the physical fluxes and of theform (33) The numerical flux vector is a function of the system and auxiliary variablesinterior and exterior to a subdomain A common choice for f lowast is

f lowast = fh+τ

2

[[

vh]]

K-component of f lowast f lowastK = fKh+

τ

2

[[

Kh

]]

(38)

where as an example we have also shown the component of f lowast corresponding to the analysisabove Respectively the average and jump across the interface are

fh =1

2

(

f+ + fminus) [[

vh]]

= nminusvminus + n+v+ (39)

Here τ is a position dependent penalty parameter (fixed below) and nminus(n+) is the localoutward pointing normal to the interior (exterior) subdomain The role of τ is to ldquopenalizerdquo(i e yield a negative contribution to the L2 energy norm) jumps across an interface Anappropriate choice of τ will ensure stability and we now provide some motivation for thechoice (41) of τ we make below

Were we treating the fully first-order system (10) the local Lax-Friedrichs flux wouldoften be a preferred choice due to its simplicity [66] In this case the constant ω in thenumerical flux formula Flowast = Fh + 1

2ω[[

Wh

]]

obeys ω ge max∣

∣micro(nablaWF(W ))∣

∣ HereF(W ) = A(u)W the notation micro(middot) indicates the spectral radius of the matrix within andthe max is taken over interior Wminus and exterior W+ states Motivated by (9) we adopt asimilar but simpler prescription substituting the field gradient

nablaWvQA(u)WvQ = A(u) (40)

13

for nablaWF(W ) Precisely we assume the scaling

τ(bk) = τ(ak+1) = τk+12 equiv C middotmax∣

∣micro(

A(u))∣

∣ (41)

where C = O(1) is a constant chosen for stability Larger values of C will result in schemeswith better stability properties whereas too large a value will impact the CFL conditionAt the interface point Ik+12 equiv D

k cap Dk+1 the vector uh has two representations uminus at bk

and u+ at ak+1 The max in (41) is taken over the corresponding two sets of field speedsMore precisely the speeds in Table I are computed for both uminus and u+ and the maximumtaken over all resulting speeds For the auxiliary variables a penalized central flux is usedThe definition with one representative component is

ulowast = uh minus1

2

[[

uh]]

α-component of ulowast αlowast = αh minus1

2

[[

αh

]]

(42)

with similar expressions for the remaining componentsWe stress the following point Since the interior coupling between subdomains is achieved

through the numerical flux forms (41) and (42) the inverse transformation (17) expressingthe fundamental fields in terms of the characteristic fields is not required to achieve thiscoupling On the other hand imposition of physical boundary conditions may still rely on(17) since this transformation allows one to fix only incoming characteristic modes

C Matrix form of the semi-discrete equations

Let us introduce the kth mass and stiffness matrices

Mkij =

int bk

akdrℓki (r)ℓ

kj (r) Sk

ij =

int bk

akdrℓki (r)ℓ

kjprime(r) (43)

These matrices belong to Dk and the corresponding matrices defined on the reference interval[minus1 1] are

Mij =

int 1

minus1

duℓi(u)ℓj(u) Sij =

int 1

minus1

duℓi(u)ℓprimej(u) (44)

where ℓj(u) is the jth Lagrange polynomial determined by the LGL nodes uj on [minus1 1]These matrices are related by Mk

ij =12(bk minus ak)Mij and S

kij = Sij whence only the reference

matrices require computation and storageWe will use the matrices Mk and Sk in obtaining an ODE system from (26) and (35)

Towards this end we first approximate the nonlinear terms (products and quotients) in(26) by degree-N interpolating polynomials Such approximations are achieved throughpointwise representations For example (QαQχgrr)h appears in (26) and is expressed inthe following way [cf footnote 2]

(

QαQχ

grr

)

h

(t r) =

(

QαhQχh

grrh

)

(t r) rarrNsum

j=0

Qαh(t rkj )Qχh(t r

kj )

grrh(t rkj )

ℓkj (r) (45)

Note that the expressions on the right and left are not equivalent due to aliasing error [65]Our vector notation for this replacement will be

(

QαQχ

grr

)

h

(t r) rarr(

QαQχ

grr

)

(t)Tℓk(r) (46)

14

Operations among bold variables are always performed pointwise Making similar replace-ments for all terms in (26) and then carrying out the integrations in (35) which bring inMk and Sk we arrive at

parttK = βrDK minus χDQα

grr

+1

2

χQgrrQα

g2rr

minusχQgθθ

Qα

grrgθθ

+1

2

QαQχ

grr

+3

2

αA2rr

g2rr

+1

3αK2 +Mminus1ℓk (fKh minus f lowast

K)∣

∣

∣

bk

ak (47)

where we have again suppressed the superscript k on all terms except ℓk(r) and the sub-script h is dropped on all boldfaced variables As described in [66] the spectral collocationderivative matrix

(Dk)ij =dℓkjdr

∣

∣

∣

∣

∣

r=rki

(48)

can also be expressed as Dk = (Mk)minus1Sk which appears above Eight other semi-discreteevolution equations are similarly obtained with nine in total (one for each component ofWuv) Additionally we have

Qα = Dα+Mminus1ℓk(αlowast minusαh)∣

∣

∣

bk

ak (49)

as one of the auxiliary equations with five in total (one for each component of Q = WQQ)

D Filtering

Like other nodal (pseudospectral) methods our scheme may suffer from instabilitiesdriven by aliasing error [65] Filtering is a simple yet robust remedy To filter a solu-tion component such as χ we use the modal (as opposed to nodal) representation of thesolution

χkh(t r) =

Nsum

j=0

χ(t rkj )ℓkj (r) =

Nsum

j=0

χkj (t)Pj(r) (50)

where Pj(r) is the jth Legendre polynomial Let ηj = jN and define the filter function

σ(ηj) =

1 for 0 le ηj le NcN

exp(

minus ǫ(

ηjminusNcN

1minusNcN

)2s)

for NcN le ηj le 1(51)

At each timestep we modify our solution component according to

χkh rarr

(

χkh

)filtered=

Nsum

j=0

σ(ηj)χkj (t)Pj(r) (52)

Evidently the modification only affects the top N minus Nc modes and is sufficient to controlthe type of weak instability driven by aliasing [66] The numerical parameters Nc and ǫ areproblem dependent For our simulations we have taken ǫ ≃ minuslog(εmach) = 36 where εmach

is machine accuracy in double precision

15

E Model system

To better illustrate the basic properties of our method we consider a toy model Namelythe following spatially second-order system

parttu = uprime + av minus u3 + g(t x) (53a)

parttv = uprimeprime + vprime minus (u+ v)(uprime)2 + v2u2 + h(t x) (53b)

where a ge 1 is constant and g and h are analytic source terms to be specified In contrastto (6) here u v and Q = uprime are scalars rather than vectors System (53) admits a first-order reduction in which uprime is defined as an extra variable Since this first-order reductionis strongly hyperbolic the spatially second-order system (53) is also strongly hyperbolic byone of the definitions considered in [75] The characteristic fields Xplusmn and speeds microplusmn are

X+ =radicav minus uprime micro+ =

radicaminus 1 Xminus =

radicav + uprime microminus = minus(

radica+ 1) (54)

To construct a local dG scheme for this system we first rewrite it as

parttu = Q+ av minus u3 + g(t x) (55a)

parttv = Qprime + vprime minus (u+ v)Q2 + v2u2 + h(t x) (55b)

Q = uprime (55c)

Evidently f = minus(Q + v) is the v-component of the physical flux vector

F (vQ) equiv(

Fu

Fv

)

=

(

0

f

)

(56)

Note that F has the same structure as (u v)T Borrowing from the presentation for theBSSN system we write the analogous semidiscrete scheme on each subdomain D

k for themodel system

parttu = Q+ av minus u3 + g(t) (57a)

parttv = DQ+Dv minus (u+ v)Q2 + v2u2 + h(t) +Mminus1ℓk(fh minus f lowast)∣

∣

∣

bk

ak (57b)

Q = Du+Mminus1ℓk(ulowast minus uh)∣

∣

∣

bk

ak (57c)

Here we have suppressed the subinterval label k from all variables except for the vectorℓk of Lagrange polynomial values Moreover following the guidelines discussed above thenumerical fluxes are given by

f lowast = fh+1 +

radica

2

[[

vh]]

ulowast = uh minus1

2

[[

uh]]

(58)

Appendix C analyzes the stability of our scheme for a more general numerical flux choiceas applied to (53) with the nonlinear and source terms dropped

IV RESULTS FROM NUMERICAL SIMULATIONS

This section presents results found by numerically solving both the model system (53)and BSSN system (4) with the dG scheme presented in Sec III

16

10minus10

10minus6

10minus2

∆uinfin

10minus10

10minus6

10minus2

∆vinfin

0 500 1000 1500 2000 2500 3000 3500 400010

minus10

10minus6

10minus2

∆(partxu)infin

t

N = 3N = 6N = 9N = 12

FIG 1 Spectral convergence of fields for model PDE Respectively for N = 3 6 9 12

a timestep of ∆t = 00578 00178 00084 00049 has been chosen for stability and accuracy In the

title headings for example ∆u equiv unumer minus uexact

A Simulations of the model system

The semi-discrete scheme (57) has been integrated with the classical fourth-order Runge-Kutta method When integrating this system we have first constructed Q at each Runge-Kutta stage and then substituted into the evolution equations (57ab) for u and v Theproblem has been solved on a computational domain [0 4π] comprised of two subdomainswith a timestep chosen small enough for stability The initial data has been taken from thefollowing exact solution to (53)

uprimeexact(t x) =1

2

[

sin(xminus microminust)minus sin(xminus micro+t)]

(59a)

vexact(t x) =1

2radica

[

sin(xminus microminust) + sin(xminus micro+t)]

(59b)

g(t x) = u3exact (59c)

h(t x) = (uexact + vexact)(uprimeexact)

2 minus v2exactu2exact (59d)

where the speeds microplusmn are found in (54) Specification of the boundary condition at a physicalendpoint amounts to choosing the external state for at the endpoint We have considered

17

8 16 32 64 128

10minus3

10minus2

10minus1

N

∆t

max stable ∆t8N2

FIG 2 Scaling of maximum stable ∆t with N for model PDE

two possibilities (i) the analytic state (Q+ v+) = (Qexact vexact) and (ii) an upwind stateFor example at x = 4π the upwind state is4

Q+ = Qupwind =1

2

[

(Xminus)exact minus (X+)numer

]

v+ = vupwind =1

2radica

[

(Xminus)exact + (X+)numer

]

(60)Either choice of (Q+ v+) leads to similar results and the plots here correspond to theanalytic state Figure 1 clearly shows spectral convergence with increasing polynomial orderN across all fields for the case a = 2 Other values of a including a = 1 for which X+

is a static characteristic field have also been considered with similar results Appendix Cdemonstrates that our proposed scheme for the system (57) with nonlinear and source termsdropped is stable in a semi-discrete sense Nevertheless the fully discrete scheme obtainedvia temporal discretization by the fourth-order Runge-Kutta method is still subject to thestandard absolute stability requirement Namely if microh is any eigenvalue corresponding tothe (linearized) discrete spatial operator then a necessary condition for stability is thatmicroh∆t lies in absolute stability region for Runge-Kutta 4 We here show empirically that theassociated timestep restriction scales like Nminus2 ie ∆t = O(Nminus2) for stability We note thatsuch scaling is welcome in light of the second-order spatial operators which appear in thesystem and suggest a possible worse scaling like Nminus4 Fig 2 plots the maximum stabletimestep for a range of N demonstrating the Nminus2 scaling in line with behavior known fromanalysis of first-order systems [66] This scaling also holds for the BSSN system

4 We remind the reader that unfortunately the plusmn on Xplusmn means something different than the plusmn indicating

exteriorinterior dG states [cf footnote 3]

18

B Simulations of the BSSN system

This subsection documents results for simulations of the unit-mass-parameter (M = 1)Schwarzschild solution (B9) expressed in terms of ingoing Kerr-Schild coordinates Sincethe solution is stationary temporal integration of the semi-discrete scheme has been carriedout with the forward Euler method The r-coordinate domain [04 34] has been split into3 equally spaced subdomains and we have set η = 10 λ = 01 and C = 2 [cf Eq (41)]For all simulations ∆t has been chosen for stability With the chosen λ the inner physicalboundary rmin = 04 is an excision surface At each timestep we have applied an (order2s = 20) exponential filter on the top two-thirds of the modal coefficient set for all fieldsexcept for grr and gθθ For stability we have empirically observed that grr and gθθ must not

be filtered A detailed understanding of this is still lackingIssues related to physical boundary conditions are similar to the one encountered in

Sec IVA for the model problem Similar to before we have retained Eqs (3842) as thechoice of numerical flux even at the endpoints Therefore at an endpoint the specificationof the boundary condition amounts to the choice W+ of external state We have typicallychosen the inner boundary of the radial domain as an excision boundary and in this caseW+ = Wminus is enforced at the inner physical boundary At the outer physical boundaryfor W+ we have again considered two choices (i) Wexact and (ii) Wupwind To enforcechoice (ii) the inverse transformation (17) must be used with incoming characteristic fieldsfixed to their exact values similar to (60) We have tried various versions of choice (ii)and in all cases the resulting simulations have been unstable We therefore present resultscorresponding to choice (i) Although the choice of an analytical external state Wexact

at the outer boundary is stable for our problem such a boundary condition is unlikelyto generalize to more complicated scenarios involving dynamical fields Indeed the issueof outer boundary conditions for the BSSN system is an active area of research with aproper treatment requiring fixation of incoming radiation control of the constraints andspecification of gauge (see Ref [86] for a recent analysis)

For BSSN simulations our main diagnostic is to monitor the Hamiltonian momentumand conformal connection constraints Figure 3 depicts long-time histories of constraintviolations whereas Figs 4 and 5 depict long-time error histories for the individual BSSNfield components From the middle plot in Fig 5 we infer that up to the indicated numericalerror the factor g sin4 θ = grr(gθθ)

2 remains at its initial fixed profile r4 throughout theevolution These figures indicate that the proposed scheme is stable for long times andexhibits spectral converge with increased polynomial order N Similar results are recoveredfromM = 0 Minkowski initial data The stability documented in these plots does not appearto rely on inordinate parameter tuning For example with the fixed parameters describedabove we obtain similar plots if we individually vary (i) rmin over 0325 035 04 0475(values still corresponding to an excision surface for the given choice of λ) (ii) η over1 3 7 10 (iii) s over 8 9 10 With the polynomial order N ranging over 23 26 29 31both stability and qualitatively similar exponential convergence is achieved with a singlesubdomain Likewise adoption of a larger coordinate domain with more subdomains doesnot impact our results Finally we have considered the addition of random noise to all fieldcomponents at the initial time Precisely with the system component χ as an example wehave set

χ(0) 7rarr χ(0) + δχ(0) (61)

where each component (nodal value) of δχ(0) is 10minus5 times a random variable drawn from a

19

10minus9

10minus7

10minus5

10minus3

Hinfin

10minus9

10minus7

10minus5

10minus3

Mrinfin

0 2000 4000 6000 8000 10000

10minus9

10minus7

10minus5

10minus3

Grinfin

t

N = 11N = 14N = 17N = 20

FIG 3 Spectral convergence of constraint violations for M = 1 Kerr-Schild ini-

tial data Respectively for N = 11 14 17 19 a timestep of ∆t ≃ 00041 00026 00018 00013

has been chosen for stability and accuracy

standard normal distribution Such perturbed initial data also gives rise to stable evolutions

V CONCLUSION

We have introduced a discontinuous Galerkin method for solving the spherically reducedBSSN system with second-order spatial operators Our scheme shares similarities withother discontinuous Galerkin methods that use local auxiliary variables to handle high-orderspatial derivatives [66 77ndash79 82 84 85] and which have typically been applied to eitherelliptic parabolic or mixed type problems The key ingredient of a stable dG scheme is anappropriate choice of numerical flux and our particular choice has been motivated by theanalysis presented in Appendix C When used to evolve the Schwarzschild solution in Kerr-Schild coordinates our numerical implementation of the BSSN system (4) is robustly stableand converges to the analytic solution exponentially with increased polynomial order Byapproximating the spatially second-order form of the BSSN system we have not introducedextra fields which are evolved Evolved auxiliary fields result in new constraints which mayspoil stability Our main goal has been stable evolution of the spherically reduced BSSNsystem as a first step towards understanding how a discontinuous Galerkin method might be

20

10minus9

10minus7

10minus5

∆Arrinfin

10minus9

10minus7

10minus5

∆Kinfin

0 2000 4000 6000 8000 10000

10minus9

10minus7

10minus5

max(∆Brinfin ∆Γrinfin)

t

N = 11N = 14N = 17N = 20

FIG 4 Spectral convergence of solution for M = 1 Kerr-Schild initial data

Timestep choices are described in the caption for Fig 3 In the title headings for example

∆Arr equiv (Arr)numer minus (Arr)exact

applied to the full BSSN system Towards that goal we now discuss treatment of singularitiesand generalization of the described dG method to higher space dimension

To deal with the fixed Schwarzschild singularity we have used excision which is easy inthe context of the spherically reduced BSSN system However excision for the binary blackhole problem in full general relativity requires attention to the technical challenge of hori-zon tracking State-of-the-art BSSN codes avoid such complication relying instead on themoving-puncture technique While the moving-puncture technique does involve mild centralsingularities it may still prove amenable to spectral methods Indeed spectral methods fornon-smooth problems is well-developed in both theory and for complex applications Sincethe moving-puncture technique can be performed in spherical symmetry [60] a first-steptoward a spectral moving-puncture code would be to implement a moving puncture withthe nodal dG method described here Such an implementation may adopt Legendre-Gauss-Radau nodes on the innermost subdomain thereby ensuring that the physical singularitydoes not lie on a nodal point (in much the same way finite difference codes use a staggeredgrid) Beyond traditional excision and moving punctures one might also retain a smoothproblem for which spectral methods are especially well-suited via the turducken approach[87 88] to singularities

21

10minus11

10minus9

10minus7

10minus5

∆χinfin

10minus11

10minus9

10minus7

10minus5

max(∆grrinfin ∆gθθinfin)

0 2000 4000 6000 8000 10000

10minus11

10minus9

10minus7

10minus5

max(∆αinfin ∆βrinfin)

t

N = 11N = 14N = 17N = 20

FIG 5 Spectral convergence of solution violations for M = 1 Kerr-Schild initial

data See the caption of Fig 4 for details

Discontinuous Galerkin methods for hyperbolic problems in two and three space dimen-sions are well-developed A generalization of the method described here to three-dimensionsand the full BSSN system would likely rely on an unstructured mesh Appropriate local poly-nomial expansions for the subelements are well-understood as are choices for the numericalfluxes which would now live on two-dimensional faces rather than single points Whether ornot it would ultimately prove successful generalization of our dG method to a higher dimen-sion would rely on an established conceptual framework Further computational advancesof relevance to a generalization of our dG method to the full BSSN system (possibly includ-ing matter) may include mesh hp-adaptivity local timestepping shock capturing and slopelimiting techniques [66] Moreover recent work [89] indicates that enhanced performancewould be expected were our scheme implemented on graphics processor units

VI ACKNOWLEDGMENTS

We thank Nick Taylor for discussions about treating second-order operators with spectralmethods Benjamin Stamm for helping polish up a few parts of Appendix C Khosro Shah-bazi for discussions on LDG and IP methods We also acknowledge helpful conversationswith David Brown and Manuel Tiglio about the BSSN system and previous work on its nu-

22

merical implementation We gratefully acknowledge funding through grants DMS 0554377and DARPAAFOSR FA9550-05-1-0108 to Brown University and NSF grant PHY 0855678to the University of New Mexico

Appendix A Hyperbolicity of the first-order system

This appendix analyzes the matrix A(u) appearing in (10) in order to construct thecharacteristic fields (14) In matrix form the sector (8) of the principal part of (10) reads asfollows

partt

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

=

βr 0 minus 4λα3grr

0 0 λβr

6(grr)2λβr

3gθθgrr0 4λ

3grr

0 βr 0 23grrαχ

13α minus αχ

3grr

αχ3gθθ

minus23χ 0

0 0 βr 0 0 0 0 minus χgrr

0

0 0 minus 4α3grr

βr 0 βr

6(grr)2βr

3gθθgrr0 4

3grr

0 0 23αχ 0 βr minus βrχ

3grrminus2βrχ

3gθθ0 minus2

3χ

0 minus2α 0 0 0 23βr minus2grrβr

3gθθ0 4

3grr

0 gθθαgrr

0 0 0 minusgθθβr

3grr13βr 0 minus2

3gθθ

0 0 minus2α 0 0 0 0 βr 0

34

0 0 0 0 0 0 0 βr

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

prime

(A1)which defines the matrix A(u) appearing in (9) and so also the matrix A(u) in (10) Notethat in the last equation the matrix within the square brackets is minusA(u) For certainconfigurations of u and λ the system (10) is strongly hyperbolic [76] that is A(u) hasa complete set of eigenvectors and real eigenvalues Indeed five eigenpairs of A(u) aretrivially recovered upon inspection of A(u)rsquos leading 5times5 diagonal block These correspondto eigenvalue 0 and the left eigenspace ξj = eTj 1 le j le 5 where ej are the canonical

basis vectors Since each component of u arises as eTj W each is also a characteristic field

The remaining nine eigenpairs are determined by A(u) The eigenvalues of A(u) are

micro1 = 0 micro23 = minusβr microplusmn4 = minusβr plusmn

radic

2αχ

grr microplusmn

5 = minusβr plusmn α

radic

χ

grr microplusmn

6 = minusβr plusmnradic

λ

grr

(A2)

23

and the corresponding left eigenvectors are

x1 = (0 0 0 0 0 gθθ 2grr 0 0) (A3a)

x2 =

(

0 0 0 grr2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3b)

x3 =

(

grrλ 0 0 0

2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3c)

xplusmn4 =

(

0 0plusmnradic

2αgrrχ

0 0 0 0 1 0

)

(A3d)

xplusmn5 =

(

0∓ 3radicgrrχ

plusmn2

radic

grrχ 2grr

1

χminus 1

grr1

gθθ 0 0

)

(A3e)

xplusmn6 =

(

minus3

4

grrλ 0plusmn α

radicλgrr

(2αχminus λ) 0 0minus βr

8(βrgrr ∓radicλgrr)

minus βrgrr

4gθθ(βrgrr ∓radicλgrr)

αχ

(2αχminus λ)plusmnradic

grrλ

)

(A3f)

where for example xplusmn5 A(u) = microplusmn5 x

plusmn5 Assuming that grr gθθ χ and α are everywhere strictly

positive the eigenvalues are real and the eigenvectors are linearly independent providedthat (15) holds These eigenvectors are easily extended to eigenvectors of A(u) e g asxplusmn6 rarr (01times5 x

plusmn6 ) Then for example the characteristic field

Xplusmn6 equiv (01times5 x

plusmn6 )W = xplusmn6WvQ (A4)

and similarly Xplusmnj = xplusmnj WvQ for j = 4 5 and Xk = xkWvQ for k = 1 2 3 The characteristic

speeds for these fields are microk and microplusmnj With this convention the speeds listed in Table I

correspond to the Xk and Xplusmnj in (14)

Appendix B Schwarzschild solution in conformal Kerr-Schild coordinates

In Kerr-Schild coordinates here the system directly related to incoming Eddington-Finkelstein null coordinates the line element for the Schwarzschild solution reads

ds2 = minusα2dt2 + (1 + 2MR)(dR + βRdt)2 +R2dθ2 +R2 sin2 θdφ2 (B1)

where R is the area radius α = (1 + 2MR)minus12 is the lapse and βR = 2M(R + 2M) isthe shift vector The physical spatial metric gab is the spatial part of this line element

To define the corresponding solution to the BSSN system we use equation gab = χgab todefine the following relationship between line elements

dr2 + r2(dθ2 + sin2 θdφ2) = χ[(1 + 2MR)dR2 +R2dθ2 +R2 sin2 θdφ2] (B2)

so that

χ

(

1 +2M

R

)(

dR

dr

)2

= 1 χR2 = r2 (B3)

Then we have(

1 +2M

R

)12dR

R=dr

r (B4)

24

with integration yielding

r =R

4

(

1 +

radic

1 +2M

R

)2

e2minus2radic

1+2MR (B5)

where the constant of integration has been chosen so that the R r rarr infin limits are consistentThe second relation in (B3) shows that

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR =rArr χminus4 =2eradic

1+2MRminus1

1 +radic

1 + 2MR (B6)

The extrinsic curvature tensor is specified by the expression forK given in (B9h) the identityK = KR

R + 2Kθθ and

Kθθ =

(

1 +2M

R

)minus122M

R2 (B7)

Since KRR = Kr

r we compute that

Krr = K minus 2Kθ

θ = minus(

1 +2M

R

)minus12(R +M

R + 2M

)

2M

R2 (B8)

Next since Krr = grrKrr = χminus1Kr

r we have Krr = Arr+

13grrK This implies Arr = Kr

r minus 13K

from which we get (B9g) In all we have

α =

(

1 +2M

R

)minus12

(B9a)

βr = βR dr

dR= χ12

(

1 +2M

R

)minus122M

R(B9b)

grr = 1 (B9c)

gθθ = r2 = χR2 (B9d)

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR (B9e)

Br = 0 (B9f)

Arr = minus(

1 +2M

R

)minus124M

3R2

(

2R + 3M

R + 2M

)

(B9g)

K =

(

1 +2M

R

)minus32(

1 +3M

R

)

2M

R2(B9h)

Γr = minus2

r= minus 2

χ12R (B9i)

To differentiate these expressions with respect to r we use the identity

dR

dr= χminus12

(

1 +2M

R

)minus12

(B10)

along with the chain rule

25

Appendix C Stability of the model system

The following stability analysis for the model system (53) has been inspired by [78 79]After dropping all nonlinear source terms the system (53) becomes

parttu = uprime + av (C1a)

parttv = uprimeprime + vprime (C1b)

This section analyzes the stability of (C1) considering both the continuum system itself aswell as its semi-discrete dG approximation The latter analysis offers some insight into theempirically observed stability of our dG scheme for the spherically reduced BSSN equations

1 Analysis for a single interval

Throughout we work with the L2 inner product and norm

(f g)D =

int

D

fg fD =radic

(f f)D (C2)

where D is the spatial coordinate interval (here D may represent a subdomain Dk or the

whole domain Ω) and we have suppressed all integration measures For the continuummodel we will establish the following estimate

uprime(T middot)2D+ av(T middot)2

Dle C(T )

(

uprime(0 middot)2D+ av(0 middot)2

D

)

(C3)

where the time-dependent constant C(T ) is determined solely by the choice of boundaryconditions To show (C3) we first change variables with v =

radicav thereby rewriting (C1)

in the following symmetric form

parttu = uprime +radicav (C4a)

parttv =radicauprimeprime + vprime (C4b)

Equations (C4ab) then imply

1

2partt

int

D

(uprime)2 =

int

D

uprime(uprimeprime +radicavprime) =

int

D

radicauprimevprime +

1

2

int

partD

(uprime)2 (C5a)

1

2partt

int

D

(v)2 =

int

D

v(radicauprimeprime + vprime) = minus

int

D

radicauprimevprime +

1

2

int

partD

(v2 + 2radicauprimev) (C5b)

Here vvprime and uprimeuprimeprime have been expressed as exact derivatives and then integrated to boundaryterms the second equation employs an extra integration by parts and with only one spacedimension

int

partDdenotes a difference of endpoint evaluations Addition of Eqs (C5ab) gives

1

2partt

int

D

[

v2 + (uprime)2]

=1

2

int

partD

[

v2 + (uprime)2 + 2radicauprimev

]

(C6)

Substitutions with the identities

[

v2 + (uprime)2]

=1

2

[

(v + uprime)2 + (v minus uprime)2]

2uprimev =1

2

[

(v + uprime)2 minus (v minus uprime)2]

(C7)

26

and replacements to recover the original variable v = vradica yield

1

2partt

int

D

[

av2 + (uprime)2]

=1 +

radica

4

int

partD

(radicav + uprime)2 +

1minusradica

4

int

partD

(radicav minus uprime)2 (C8)

From (C8) we deduce that the time-dependent constant C(T ) in (C3) must satisfy

∣

∣

∣

∣

∣

∣

1 +

int T

0

[

1+radica

2

int

partD(radicav + uprime)2 + 1minus

radica

2

int

partD(radicav minus uprime)2

]

dt

uprime(0 middot)2 + av(0 middot)2

∣

∣

∣

∣

∣

∣

le C(T ) (C9)

For periodic boundary conditions we may choose C(T ) = 1 Moreover if a ge 1 anduprime = minusradic

av is specified at partD+ then uprime(t middot)2 + av(t middot)2 decaysStill working on a single interval (subdomain) we now consider the semi-discrete scheme

for (C4) i e (57) with all nonlinear source terms dropped and with v replaced by vradica

Derivation of a formula analogous to (C8) is our first step toward establishing L2 stabilityof the semi-discrete scheme While (57) features vectors for example u(t) taking values atthe Legendre-Gauss-Lobatto nodal points here we work with the numerical solution as apolynomial for example uh(t x) These two representations are related by the Lagrange in-terpolating polynomials for the nodal set here taken to span both the space of test functionsand the space of basis functions Our scheme is

int

Dk

ψparttuh =

int

Dk

ψ(Qh +radicavh) (C10a)

int

Dk

ξparttvh = minusint

Dk

ξprime(radicaQh + vh) +

int

partDk

ξ(radicaQlowast + vlowast) (C10b)

int

Dk

ϕQh =

int

Dk

ϕuprimeh +

int

partDk

ϕ (ulowast minus uh) (C10c)

where ψ ξ and ϕ are polynomial test functions These test functions are arbitrary exceptthat they must be degree-N polynomials In (C10) the variables uh vh and Qh should alsocarry a superscript k but we have suppressed this Derivation of a formula analogous to(C8) is complicated by the fact that Qh is not evolved Nevertheless at a given instant t wecan assemble Qh from (C10c)

Mimicking the calculation (C5b) from the continuum case we first use (C10b) with ξ = vhto write

1

2partt

int

Dk

v2h = minusint

Dk

(radicaQh + vh)v

primeh +

int

partDk

(radicaQlowast + vlowast)vh

= minusint

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2(radicaQlowast + vlowast)vh minus v2h

]

(C11)

The right-hand side of (C5a) is analogous to

1

2partt

int

Dk

Q2h =

int

Dk

QhparttQh (C12)

However since Qh is not evolved the term parttQh must be given a suitable interpretationOn the right hand side of (C10c) only uh u

primeh and u

lowast necessarily depend on time since the

27

test function ϕ need not be time-dependent Furthermore ulowast is explicitly given as a linearcombination of uh as seen in Eq (C21c) below Choosing ϕ = ℓj taking the time derivativeof (C10c) and appealing to the commutivity of mixed partial derivatives we therefore arriveat

int

Dk

ℓjparttQh =

int

Dk

ℓj(parttuh)prime +

int

partDk

ℓj(

(parttu)lowast minus parttuh

)

(C13)

where (parttu)lowast depends on parttuh in precisely the same way that ulowast depends on uh We have

written ℓj rather than ϕ in the last equation to emphasize that the result also holds for anylinear combination of ℓj (for example ϕ) and even for time-dependent combinations SinceQh is itself such a combination we obtain

1

2partt

int

Dk

Q2h =

int

Dk

Qh(parttuh)prime +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

Qh(Qprimeh +

radicavprimeh) +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2((parttu)lowast minus parttuh)Qh +Q2

h

]

(C14)

Addition of (C11) and (C14) gives

1

2partt

int

Dk

(Q2h + v2h) =

1

2

int

partDk

[

Q2h minus v2h + 2(

radicaQlowast + vlowast)vh + 2((parttu)

lowast minus parttuh)Qh

]

(C15)

the aforementioned analog of (C8) This formula holds on a single subdomain Dk and we

now combine multiple copies of it one for each value of k

2 Analysis for multiple intervals

To facilitate combination of (C15) over all k we change notation At every subdomaininterface Ik+12 equiv partDk cap partDk+1 let the superscripts L and R denote field values respectivelytaken from the left and right Then the fields evaluated at Ik+12 which belong to D

k will beuLk+12 v

Lk+12 and Q

Lk+12 while those belonging to D

k+1 will be uRk+12 vRk+12 and Q

Rk+12

However at Ikminus12 the values taken from D

k are uRkminus12 vRkminus12 and QR

kminus12 Note that we

have also replaced the subscript h denoting a numerical solution with k plusmn 12 denotingthe location of the endpoint value of the numerical solution With this notation we define

∆Lα =

1

2

[

(QLα)

2 minus (vLα)2]

+(radic

aQlowastα + vlowastα

)

vLα +[

(parttuα)lowast minus parttu

Lα

]

QLα (C16)

and similarly for ∆Rα The same numerical fluxes appear in both ∆L

α and ∆Rα (ie each

numerical flux takes the same value on either side of an interface) whence fluxes like Qlowastα do

not carry an L or R superscript In terms of these definitions (C15) becomes

1

2partt

int

Dk

(Q2h + v2h) = ∆L

k+12 minus∆Rkminus12 (C17)

28

Summation over all Dk yields

1

2partt

kmaxsum

k=1

int

Dk

(Q2h + v2h) =

kmaxminus1sum

k=1

(

∆Lk+12 minus∆R

k+12

)

+∆Lkmax+12 minus∆R

12

=

kmaxminus1sum

k=1

(

∆Lh minus∆R

h

)∣

∣

Ik+12 +∆Lkmax+12 minus∆R

12 (C18)

We have reverted to h-notation denoting the numerical solution since the LR superscriptsindicate unambiguously the relevant domain used for evaluation at Ik+12

We again seek an estimate of the form

kmaxsum

k=1

(

Qh(T middot)2Dk + avh(T middot)2Dk

)

le C(T )kmaxsum

k=1

(

Qh(0 middot)2Dk + avh(0 middot)2Dk

)

(C19)

that is essentially the same as the one (C3) considered in the continuum case Assume thatthe chosen boundary conditions ensure ∆L

kmax+12 minus ∆R12 is bounded by a time-dependent

constant which does not depend on the numerical parameters N and h (subdomain width)Establishment of stability then amounts to showing that the remaining sum over interfaceterms in (C18) is non-positive whence this remaining sum is consistent with C(T ) le 1although the boundary conditions may give rise to a different bound In fact we will choosethe numerical fluxes such that each individual interface term is non-positive At interfaceIk+12 and in LR notation the jump and average of vh for example are

1

2

(

v+ + vminus)

equiv vh =1

2

(

vLk+12 + vRk+12

)

(C20a)

nminusvminus + n+v+ equiv[[

vh]]

= vLk+12 minus vRk+12 (C20b)

Consider numerical fluxes of the form

Qlowast = Qh minusτQ2

[[

Qh

]]

(C21a)

vlowast = vh minusτv2

[[

vh]]

(C21b)

ulowast = uh minusτu2

[[

uh]]

(C21c)

(parttu)lowast = parttuh minus

τu2

[[

parttuh]]

(C21d)

where (C21c) induces (C21d) and where the penalty parameters τu τv and τQ are realnumbers The fluxes defined in (58) correspond to τu = 1 τv = 1 +

radica and τQ = 0 In

terms of these quantities the kth interface contribution in (C18) is

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

parttuh]]

minus τu2

[[

parttuh]][[

Qh

]]

(C22)

where we have suppressed the k dependence of the right-hand side Now consider the term[[

parttuh]]

Because parttuh and Qh+radicavh are both polynomials of degree N Eq (C10a) implies

29

τu

τ Q

minus1 0 1minus1

minus05

0

05

1

τu

minus1 0 1minus1

minus05

0

05

1

FIG 6 Stable evolutions for the model system For fixed τv = 10minus6 and τv = 1 +radic2

respectively the left and right plots depict stable choices (determined empirically) of τu and τQfor the linear model system (C1) The stable regions are colored black but the jagged edges result

from the discretization of the (τu τQ)-plane

the vector equation parttu = Q +radicav that is pointwise equivalence on the nodal points of

Dk which in turn implies

[[

parttuh]]

=[[

Qh+radicavh]]

Upon substituting this identity into thelast equation we arrive at an expression which features only vh and Qh

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

Qh +radicavh]]

minus τu2

[[

Qh +radicavh]][[

Qh

]]

(C23)

The identities vh[[

vh]]

= 12

[[

v2h]]

and[[

Qh +radicavh]]

=[[

Qh

]]

+radica[[

vh]]

then simplify(C23) to

(∆Lh minus∆R

h )∣

∣

Ik+12 = minusτv2

[[

vh]]2 minus

radica(τu + τQ)

2

[[

Qh

]][[

vh]]

minus τu2

[[

Qh

]]2 (C24)

The role of a penalty parameter is now clear Positive values of τv penalize jumps in vhthrough a negative contribution to the energy Likewise positive values of τu penalizejumps in Qh through a negative contribution to the energy However because the sign of[[

Qh

]][[

vh]]

can be positive or negative only the choice τQ = minusτu yields an expression for(∆L

h minus ∆Rh )|Ik+12 which is manifestly negative for τu ge 0 and τv ge 0 A simple estimate

based on Youngrsquos inequality with ε (that is 2αβ le εminus1α2 + εβ2 where α β ge 0 and ε gt 0)shows that for τQ = 0 the choice τv ge aτu4 also yields stability

Figure 6 depicts certain choices of stable penalty parameters for the linear model systemevolved to tfinal = 1000 (with a = 2 N = 10 and ∆t ≃ 00553) as determined empiricallywith simulations similar to those described in Sec IVA The left plot corresponds to asmall τv = 10minus6 for which the choice τu = 1 τQ = 0 is not stable as expected from thetheoretical analysis However the right plot corresponds to τv = 1 +

radica for which τu = 1

τQ = 0 is stable Motivated by the numerical flux choices (3842) used for the BSSN system

30

(4) we have (as mentioned above) set τu = 1 τv = 1 +radica and τQ = 0 in simulations of

the nonlinear model (53) For the nonlinear model system (53) the theoretically motivatedchoice τQ = minusτu also yields numerically stable evolutions when τu ge 0 and τv ge 0

For the nonlinear systems (4) and (53) we do not attempt a formal stability proofNevertheless the results of this appendix have served as a guide for our choices of penaltyparameters For the BSSN system (4) u v and Q are block indices [cf Eq (6)] Similarto the model problem we have penalized Q with τu = 1 with τv chosen large enough toheuristically overcome the cross-terms of indefinite size that arise from τQ = 0 (we interpretequations like τu = 1 componentwise) An analogous choice ldquoτQ = minusτurdquo for the BSSNsystem might be possible but would be considerably more complicated Indeed such achoice likely entails a matrix of penalty parameters but we do not give the details here

[1] F Pretorius Evolution of Binary Black-Hole Spacetimes Phys Rev Lett 95 121101 (2005)

(4 pages)

[2] F Pretorius Simulation of binary black hole spacetimes with a harmonic evolution scheme

Class Quantum Grav 23 S529-S552 (2006)

[3] M Campanelli C O Lousto P Marronetti and Y Zlochower Accurate Evolutions of Or-

biting Black-Hole Binaries without Excision Phys Rev Lett 96 111101 (2006) (4 pages)

[4] J G Baker J Centrella D-I Choi M Koppitz and J van Meter Gravitational-Wave

Extraction from an Inspiraling Configuration of Merging Black Holes Phys Rev Lett 96

111102 (2006) (4 pages)

[5] M Campanelli C O Lousto and Y Zlochower Last orbit of binary black holes Phys Rev D

73 061501(R) (2006) (5 pages)

[6] F Herrmann I Hinder D Shoemaker and P Laguna Unequal mass binary black hole plunges

and gravitational recoil Class Quantum Grav 24 S33-S42 (2007)

[7] P Diener F Herrmann D Pollney E Schnetter E Seidel R Takahashi J Thornburg and

J Ventrella Accurate Evolution of Orbiting Binary Black Holes Phys Rev Lett 96 121101

(2006) (4 pages)

[8] M A Scheel H P Pfeiffer L Lindblom L E Kidder O Rinne and S A Teukolsky Solving

Einsteinrsquos Equations With Dual Coordinate Frames Phys Rev D 74 104006 (2006)

[9] U Sperhake Binary black-hole evolutions of excision and puncture data Phys Rev D 76

104015 (2007) (20 pages)

[10] B Brugmann J A Gonzalez M Hannam S Husa U Sperhake and W Tichy Calibration

of moving puncture simulations Phys Rev D 77 024027 (2008) (25 pages)

[11] P Marronetti W Tichy B Brugmann J Gonzalez M Hannam S Husa and U Sperhake

Binary black holes on a budget simulations using workstations Class Quantum Grav 24

S43-S58 (2007)

[12] Z B Etienne J A Faber Y T Liu S L Shapiro and T W Baumgarte Filling the holes

Evolving excised binary black hole initial data with puncture techniques Phys Rev D 76

101503(R) (2007) (5 pages)

[13] B Szilagyi D Pollney L Rezzolla J Thornburg and J Winicour An explicit harmonic code

for black-hole evolution using excision Class Quantum Grav 24 S275-S293 (2007)

[14] L Boyle M Kesden and S Nissanke Binary-Black-Hole Merger Symmetry and the Spin

Expansion Phys Rev Lett 100 151101 (2008) (4 pages)

31

[15] M A Scheel M Boyle T Chu L E Kidder K D Matthews and H P Pfeiffer High-

accuracy waveforms for binary black hole inspiral merger and ringdown Phys Rev D 79

024003 (2009) (14 pages)

[16] M Hannam Status of black-hole-binary simulations for gravitational-wave detection

Class Quantum Grav 26 114001 (2009) (21 pages)

[17] I Hinder The current status of binary black hole simulations in numerical relativity

Class Quantum Grav 27 114004 (2010) (19 pages)

[18] B P Abbott et al LIGO The Laser Interferometer Gravitational-Wave Observatory

Rep Prog Phys 72 076901 (2009) (25 pages)

[19] F Acernese et al The status of VIRGO Class Quantum Grav 23 S63-S69 (2006)

[20] F Acernese et al Virgo status Class Quantum Grav 25 184001 (2008) (9 pages)

[21] R Arnowitt S Deser and C W Misner The Dynamics of General Relativity in Gravitation

an introduction to current research edited by L Witten (Wiley New York 1962) Chapter 7

227-265 Also available as gr-qc0405109

[22] J W York Jr Kinematics and Dynamics of General Relativity in Sources of Gravitational

Radiation edited by L L Smarr (Cambridge University Press Cambridge 1979)

[23] L E Kidder M A Scheel and S A Teukolsky Extending the lifetime of 3D black hole

computations with a new hyperbolic system of evolution equations Phys Rev D 64 064017

(2001) (13 pages)

[24] G Calabrese J Pullin O Sarbach and M Tiglio Convergence and stability in numerical

relativity Phys Rev D 66 041501(R) (2002) (4 pages)

[25] S Frittelli and O Reula On the Newtonian Limit of General Relativity Commun Math

Phys 166 221-235 (1994)

[26] M Shibata and T Nakamura Evolution of three-dimensional gravitational waves Harmonic

slicing case Phys Rev D 52 5428-5444 (1995)

[27] Y Choquet-Bruhat and J W York Geometrical well posed systems for the Einstein equations

C R Acad Sci Paris t 321 Serie I 1089-1095 (1995)

[28] A Abrahams A Anderson Y Choquet-Bruhat and J W York Jr Einstein and Yang-Mills

Theories in Hyperbolic Form without Gauge Fixing Phys Rev Lett 75 3377-3381 (1995)

[29] C Bona J Masso E Seidel and J Stela New Formalism for Numerical Relativity

Phys Rev Lett 75 600-603 (1995)

[30] M H P M van Putten and D M Eardley Nonlinear wave equations for relativity

Phys Rev D 53 3056-3063 (1996)

[31] S Frittelli and O A Reula First-Order Symmetric Hyperbolic Einstein Equations with Ar-

bitrary Fixed Gauge Phys Rev Lett 76 4667-4670 (1996)

[32] H Friedrich Hyperbolic reductions for Einsteinrsquos equations Class Quantum Grav 13 1451-

1469 (1996)

[33] F B Estabrook R S Robinson H D Wahlquist Hyperbolic equations for vacuum gravity

using special orthonormal frames Class Quantum Grav 14 1237-1247 (1997)

[34] M S Iriondo E O Leguizamon and O A Reula Einsteinrsquos Equations in Ashtekarrsquos Vari-

ables Constitute a Symmetric Hyperbolic System Phys Rev Lett 79 4732-4735 (1997)

[35] A Anderson Y Choquet-Bruhat and J W York Einstein-Bianchi hyperbolic system for

general relativity Topol Meth Nonlinear Anal 10 353-373 (1997)

[36] T W Baumgarte and S L Shapiro Numerical integration of Einsteinrsquos field equations

Phys Rev D 59 024007 (1998) (7 pages)

[37] M A G Bonilla Symmetric hyperbolic systems for Bianchi equations Class Quantum

8

This system admits an inner excision boundary provided

βr ge max

(

radic

2αχ

grr

radic

α2χ

grr

radic

λ

grr

)

(16)

holds at the inner boundary This condition ensures each characteristic field has a nonposi-tive speed at the inner boundary and therefore the inner boundary is an excision boundaryat which no boundary conditions are needed The extra flexibility afforded by the param-eter λ could be used to maintain rigorous hyperbolicity by moving the points at whichthe conditions in (15) are violated outside of the computational domain Furthermore forλ = 1 Eq (16) conceivably fails or is only satisfied close to r = 0 where field gradients are

prohibitively large The troublesome X+6 gauge mode has a positive speed minusβr +

radic

λgrrIndeed for the conformally flat Kerr-Schild system considered in section IVB an inner ex-cision boundary is only possible provided λ is small enough

The transformation (14) can be inverted in order to express the fundamental fields interms of the characteristic fields

Br = minus1

6

λ

grrgθθ

[

(βr)2

(βr)2grr minus λ

]

X1 +2

3

λαχ

grr(2αχminus λ)(X+

4 +Xminus4 )minus

2

3

λ

grr(X+

6 +Xminus6 )

(17a)

Arr =1

3

radic

grrχ

2α(X+

4 minusXminus4 )minus

radicgrrχ

6(X+

5 minusXminus5 ) (17b)

K =

radic

χ

8αgrr(X+

4 minusXminus4 ) (17c)

Γr = minus1

6

1

grrgθθ

[

(βr)2

(βr)2grr minus λ

]

X1 +1

grr(X2 minusX3) +

2

3

αχ

grr(2αχminus λ)(X+

4 +Xminus4 )

minus 2

3

1

grr(X+

6 +Xminus6 ) (17d)

Qχ =1

12

χ

grrgθθ

[

4(βr)2grr minus 3λ

(βr)2grr minus λ

]

X1 +χ

2X3 minus

1

3

αχ2

(2αχminus λ)(X+

4 +Xminus4 ) (17e)

+χ

3(X+

6 +Xminus6 ) (17f)

Qgrr =2(βr)2grr minus 3λ

6gθθ((βr)2grr minus λ)X1 +

4

3grrX2 minus grrX3 +

2

3

αχgrr(2αχminus λ)

(X+4 +Xminus

4 )

minus 1

3grr(X

+5 +Xminus

5 )minus2

3grr(X

+6 +Xminus

6 ) (17g)

Qgθθ =

[

1

4grr+

(βr)2

12((βr)2grr minus λ)

]

X1 minus2

3gθθX2 +

1

2gθθX3 minus

1

3

αχgθθ(2αχminus λ)

(X+4 +Xminus

4 )

+1

6gθθ(X

+5 +Xminus

5 ) +1

3gθθ(X

+6 +Xminus

6 ) (17h)

Qα =1

2(X+

4 +Xminus4 ) (17i)

Qβr =βrλ

8grrgθθ((βr)2grr minus λ)X1 minus

λ

(2αχminus λ)

radic

αχ

8grr(X+

4 minusXminus4 ) +

1

2

radic

λ

grr(X+

6 minusXminus6 )

(17j)

9

We will refer to this inverse transformation when discussing outer boundary conditions forour numerical simulations in Sec IVB

III DISCONTINUOUS GALERKIN METHOD

This section describes the nodal discontinuous Galerkin method used to numerically solve(4) We adopt a method-of-lines strategy and here describe the relevant semi-discrete schemewhile leaving the temporal dimension continuous To approximate (4) we follow the generalprocedure first introduced in Ref [82] Our approach defines local auxiliary variables Q = uprimeand rewrites the spatially second-order system (4) as the first-order system (13a) Once weuse (13b) to eliminate Q from (13a) we recover the primal equations (4) The auxiliaryvariable approach was later generalized and coined the local discontinuous Galkerin (LDG)method in Ref [77] We may refer to our particular scheme as an LDG method but notethat many variations exist in the literature We stress that in LDG methods Q is not evolvedand is introduced primarily to assist in the construction of a stable scheme

Equations (12) and (13a) imply that the physical flux function is

F (W ) =

(

Fu(W )

Fv(W )

)

equiv A(u)WvQ =

(

05times1

f(W )

)

f =

fBr

fArr

fKfΓ

(18)

Only the evolution equations for Br Arr K and Γr give rise to non-zero components in F and we have collected these non-zero components into a smaller vector f = Fv Inspectionof (8) determines these components For example from (8c) we find

fK = minusβrK +χ

grrQα (19)

A Local approximation of the system (13)

Our treatment closely follows [83] but with the equations and notations relevant forthis paper Our computational domain Ω is the closed r-interval [a b] We cover Ω withkmax gt 1 non-overlapping intervals Dk = [ak bk] where a = a1 b = bkmax and bkminus1 = ak fork = 2 middot middot middot kmax

On each interval Dk we approximate each component of the system vector W by a localinterpolating polynomial of degree N For example

χkh(t r) =

Nsum

j=0

χ(t rkj )ℓkj (r) (20)

approximates χ(t r) Throughout this section approximations are denoted by a subscripth (see [66] for the notation) For example Wh and fh are approximations of W and f Although Q = uprime Qh and uprimeh are not necessarily the same In (20) ℓkj (r) is the jth Lagrange

polynomial belonging to Dk

ℓkj (r) =Nprod

i=0i 6=j

r minus rkirkj minus rki

(21)

10

Evidently the polynomial χkh interpolates χ at rkj To define the nodes rkj consider the

mapping from the unit interval [minus1 1] to Dk

rk(u) = ak + 12(1 + u)(bk minus ak) (22)

and the N+1 Legendre-Gauss-Lobatto (LGL) nodes uj The uj are the roots of the equation

(1minus u2)P primeN(u) = 0 (23)

where PN(u) is the Nth degree Legendre polynomial and the physical nodes are simplyrkj = rk(uj) In vector notation the approximation (20) takes the form

χkh(t r) = χk(t)Tℓk(r) (24)

in terms of the column vectors

χk(t) =[

χ(t rk0) middot middot middot χ(t rkN)]T ℓk(r) =

[

ℓk0(r) middot middot middot ℓkN(r)]T (25)

On each open interval (ak bk) sub Dk and for each component of the equations in (13) we

define local residuals measuring the extent to which our approximations satisfy the originalcontinuum system Dropping the subdomain label k on the polynomials and focusing onthe K equation as a representative example the local residual corresponding to (4h) is

minus(RK)kh equivminus parttKh + (βrK prime)h minus

(

χQprimeα

grr

)

h

+

(

χQgrrQα

2g2rr

)

h

minus(

χQgθθQα

grrgθθ

)

h

+

(

QαQχ

2grr

)

h

+

(

3αA2rr

2g2rr

)

h

+

(

1

3αK2

)

h

(26)

Here for example the expressions 2

(βrK prime)h = βrhK

primeh

(

QαQχ

2grr

)

h

=QαhQχh

2grrh (27)

We similarly construct the remaining eight residuals eg (Rgrr)h and (RΓr)h as well as fiveresiduals corresponding to (13b) For example one of these remaining five is

(RQα)kh equiv minusQαh + αprime

h (28)

Let the kth inner product be defined as

(

u v)

Dk equivint bk

akdru(r)v(r) (29)

and consider the expression (ℓkj (RK)kh)Dk We call the requirement that this inner product

vanish forallj the kth Galerkin condition For each component of the system and for each k

2 At this stage the first expression is generically a polynomial of degree 2N minus 1 and the latter is not a

polynomial The conventions adopted in Eq (27) prove useful while working with the residual However

later on in Sec III C to obtain the final form (47) of the numerical approximation corresponding to (26)

we will replace nonlinear terms with degree-N polynomials

11

there is a corresponding Galerkin condition in total 9kmax(N + 1) equations for (13a) and5kmax(N+1) for (13b) Enforcement of the Galerkin conditions on each D

k will not recover ameaningful global solution since they provide no mechanism for coupling the local solutionson the different intervals Borrowing from the finite volume toolbox we achieve couplingthrough integration by parts on r and introduction of the numerical flux f lowast at the interfacebetween subdomains

In (26) we only need to consider (βrK prime)h and (χQprimeαgrr)h as the other terms comprise a

component of the source vector Sh Using integration by parts we write

(

ℓkj (βrK prime)h

)

Dk =minusint bk

akdr[

(

ℓkjβrh

)primeKh

]

+ (βrhKh) ℓ

kj

∣

∣

∣

bk

ak (30a)

(

ℓkj (χQprimeαgrr)h

)

Dk =minusint bk

akdr

[(

ℓkjχh

grrh

)prime

Qαh

]

+

(

χhQαh

grrh

)

ℓkj

∣

∣

∣

bk

ak (30b)

In these formulas we have retained the domain index k on ℓkj while continuing to suppress iton Kh grrh etc Moreover we have suppressed the r-dependence in all terms on the right-hand side Addition of these formulas along with the definition fKh = minus(βrK)h+(χQαgrr)hgives

(

ℓkj (βrK prime)h minus (χQprime

αgrr)h)

Dk = minusint bk

akdr

[

(

ℓkjβrh

)primeKh minus

(

ℓkjχh

grrh

)primeQαh

]

minus fKhℓkj

∣

∣

∣

bk

ak (31)

In lieu of (31) we will instead work with the replacement

(

ℓkj (βrK prime)h minus (χQprime

αgrr)h)

Dk rarr minusint bk

akdr

[

(

ℓkjβrh

)primeKh minus

(

ℓkjχh

grrh

)primeQαh

]

minus f lowastKℓ

kj

∣

∣

∣

bk

ak (32)

This replacement features a component f lowastK of the numerical flux rather than a component

fKh of the boundary flux The numerical flux is determined by (as yet not chosen) functions3

f lowast = f lowast(W+Wminus) (33)

where for example Wminus is an interior boundary value [either W kh (t a

k) or W kh (t b

k)] of theapproximation defined on D

k andW+ is an exterior boundary value [either W kminus1h (t bkminus1) or

W k+1h (t ak+1)] of the approximation defined on either Dkminus1 or Dk+1 We discuss our choice

of numerical flux in the next subsection We now employ additional integration by parts towrite the above replacement as

(

ℓkj (βrK prime)h minus (χQprime

αgrr)h)

Dk rarrint bk

akdrℓkj

(

βrK prime minus χQprimeα

grr

)

h

+ (fKh minus f lowastK) ℓ

kj

∣

∣

∣

bk

ak (34)

Rather than the exact kth Galerkin condition(

ℓkj (RK)kh

)

Dk = 0 forallj for the K component

of (13) on Dk we will instead strive to enforce

(

ℓkj (RK)kh

)

Dk = (fKh minus f lowastK) ℓ

kj

∣

∣

∣

bk

ak forallj (35)

3 In the context of the dG method here + and minus denote ldquoexteriorrdquo and ldquointeriorrdquo and have no relation to

the plusmn using to denote the characteristic fields and speeds in Table I For characteristic fields and speeds

+ and minus mean ldquoright-movingrdquo and ldquoleft-movingrdquo

12

although our treatment of nonlinear terms will lead to a slight modification of these equations(we return to this issue shortly) The other components of (13a) are treated similarly asare the components of (13b) Recall that for example Qα = αprime Formally using the samedG method to solve for Qα we arrive at the replacement

(

ℓkj (RQα)kh

)

Dk rarrint bk

akdrℓkj (minusQαh + αprime

h)minus (αh minus αlowast) ℓkj

∣

∣

∣

bk

ak (36)

which again features a component αlowast of the numerical flux The auxiliary variables areconstructed and used at each stage of temporal integration We then have

(

ℓkj (RQα)kh

)

Dk = (αh minus αlowast) ℓkj

∣

∣

∣

bk

ak forallj (37)

as the corresponding enforced kth Galerkin condition

B Numerical Flux

To further complete our dG scheme we must specify functional forms for the compo-nents of the numerical flux introduced in the previous section We distinguish between thephysical fluxes (components of f) and the auxiliary fluxes (components of u) arising fromthe definition of the auxiliary variables These choices are not independent as the resultingscheme must be stable and consistent Our choice follows [84] which considered diffusionproblems Additional analysis of this flux choice appears in [66 85]

Let us first consider the numerical fluxes corresponding to the physical fluxes and of theform (33) The numerical flux vector is a function of the system and auxiliary variablesinterior and exterior to a subdomain A common choice for f lowast is

f lowast = fh+τ

2

[[

vh]]

K-component of f lowast f lowastK = fKh+

τ

2

[[

Kh

]]

(38)

where as an example we have also shown the component of f lowast corresponding to the analysisabove Respectively the average and jump across the interface are

fh =1

2

(

f+ + fminus) [[

vh]]

= nminusvminus + n+v+ (39)

Here τ is a position dependent penalty parameter (fixed below) and nminus(n+) is the localoutward pointing normal to the interior (exterior) subdomain The role of τ is to ldquopenalizerdquo(i e yield a negative contribution to the L2 energy norm) jumps across an interface Anappropriate choice of τ will ensure stability and we now provide some motivation for thechoice (41) of τ we make below

Were we treating the fully first-order system (10) the local Lax-Friedrichs flux wouldoften be a preferred choice due to its simplicity [66] In this case the constant ω in thenumerical flux formula Flowast = Fh + 1

2ω[[

Wh

]]

obeys ω ge max∣

∣micro(nablaWF(W ))∣

∣ HereF(W ) = A(u)W the notation micro(middot) indicates the spectral radius of the matrix within andthe max is taken over interior Wminus and exterior W+ states Motivated by (9) we adopt asimilar but simpler prescription substituting the field gradient

nablaWvQA(u)WvQ = A(u) (40)

13

for nablaWF(W ) Precisely we assume the scaling

τ(bk) = τ(ak+1) = τk+12 equiv C middotmax∣

∣micro(

A(u))∣

∣ (41)

where C = O(1) is a constant chosen for stability Larger values of C will result in schemeswith better stability properties whereas too large a value will impact the CFL conditionAt the interface point Ik+12 equiv D

k cap Dk+1 the vector uh has two representations uminus at bk

and u+ at ak+1 The max in (41) is taken over the corresponding two sets of field speedsMore precisely the speeds in Table I are computed for both uminus and u+ and the maximumtaken over all resulting speeds For the auxiliary variables a penalized central flux is usedThe definition with one representative component is

ulowast = uh minus1

2

[[

uh]]

α-component of ulowast αlowast = αh minus1

2

[[

αh

]]

(42)

with similar expressions for the remaining componentsWe stress the following point Since the interior coupling between subdomains is achieved

through the numerical flux forms (41) and (42) the inverse transformation (17) expressingthe fundamental fields in terms of the characteristic fields is not required to achieve thiscoupling On the other hand imposition of physical boundary conditions may still rely on(17) since this transformation allows one to fix only incoming characteristic modes

C Matrix form of the semi-discrete equations

Let us introduce the kth mass and stiffness matrices

Mkij =

int bk

akdrℓki (r)ℓ

kj (r) Sk

ij =

int bk

akdrℓki (r)ℓ

kjprime(r) (43)

These matrices belong to Dk and the corresponding matrices defined on the reference interval[minus1 1] are

Mij =

int 1

minus1

duℓi(u)ℓj(u) Sij =

int 1

minus1

duℓi(u)ℓprimej(u) (44)

where ℓj(u) is the jth Lagrange polynomial determined by the LGL nodes uj on [minus1 1]These matrices are related by Mk

ij =12(bk minus ak)Mij and S

kij = Sij whence only the reference

matrices require computation and storageWe will use the matrices Mk and Sk in obtaining an ODE system from (26) and (35)

Towards this end we first approximate the nonlinear terms (products and quotients) in(26) by degree-N interpolating polynomials Such approximations are achieved throughpointwise representations For example (QαQχgrr)h appears in (26) and is expressed inthe following way [cf footnote 2]

(

QαQχ

grr

)

h

(t r) =

(

QαhQχh

grrh

)

(t r) rarrNsum

j=0

Qαh(t rkj )Qχh(t r

kj )

grrh(t rkj )

ℓkj (r) (45)

Note that the expressions on the right and left are not equivalent due to aliasing error [65]Our vector notation for this replacement will be

(

QαQχ

grr

)

h

(t r) rarr(

QαQχ

grr

)

(t)Tℓk(r) (46)

14

Operations among bold variables are always performed pointwise Making similar replace-ments for all terms in (26) and then carrying out the integrations in (35) which bring inMk and Sk we arrive at

parttK = βrDK minus χDQα

grr

+1

2

χQgrrQα

g2rr

minusχQgθθ

Qα

grrgθθ

+1

2

QαQχ

grr

+3

2

αA2rr

g2rr

+1

3αK2 +Mminus1ℓk (fKh minus f lowast

K)∣

∣

∣

bk

ak (47)

where we have again suppressed the superscript k on all terms except ℓk(r) and the sub-script h is dropped on all boldfaced variables As described in [66] the spectral collocationderivative matrix

(Dk)ij =dℓkjdr

∣

∣

∣

∣

∣

r=rki

(48)

can also be expressed as Dk = (Mk)minus1Sk which appears above Eight other semi-discreteevolution equations are similarly obtained with nine in total (one for each component ofWuv) Additionally we have

Qα = Dα+Mminus1ℓk(αlowast minusαh)∣

∣

∣

bk

ak (49)

as one of the auxiliary equations with five in total (one for each component of Q = WQQ)

D Filtering

Like other nodal (pseudospectral) methods our scheme may suffer from instabilitiesdriven by aliasing error [65] Filtering is a simple yet robust remedy To filter a solu-tion component such as χ we use the modal (as opposed to nodal) representation of thesolution

χkh(t r) =

Nsum

j=0

χ(t rkj )ℓkj (r) =

Nsum

j=0

χkj (t)Pj(r) (50)

where Pj(r) is the jth Legendre polynomial Let ηj = jN and define the filter function

σ(ηj) =

1 for 0 le ηj le NcN

exp(

minus ǫ(

ηjminusNcN

1minusNcN

)2s)

for NcN le ηj le 1(51)

At each timestep we modify our solution component according to

χkh rarr

(

χkh

)filtered=

Nsum

j=0

σ(ηj)χkj (t)Pj(r) (52)

Evidently the modification only affects the top N minus Nc modes and is sufficient to controlthe type of weak instability driven by aliasing [66] The numerical parameters Nc and ǫ areproblem dependent For our simulations we have taken ǫ ≃ minuslog(εmach) = 36 where εmach

is machine accuracy in double precision

15

E Model system

To better illustrate the basic properties of our method we consider a toy model Namelythe following spatially second-order system

parttu = uprime + av minus u3 + g(t x) (53a)

parttv = uprimeprime + vprime minus (u+ v)(uprime)2 + v2u2 + h(t x) (53b)

where a ge 1 is constant and g and h are analytic source terms to be specified In contrastto (6) here u v and Q = uprime are scalars rather than vectors System (53) admits a first-order reduction in which uprime is defined as an extra variable Since this first-order reductionis strongly hyperbolic the spatially second-order system (53) is also strongly hyperbolic byone of the definitions considered in [75] The characteristic fields Xplusmn and speeds microplusmn are

X+ =radicav minus uprime micro+ =

radicaminus 1 Xminus =

radicav + uprime microminus = minus(

radica+ 1) (54)

To construct a local dG scheme for this system we first rewrite it as

parttu = Q+ av minus u3 + g(t x) (55a)

parttv = Qprime + vprime minus (u+ v)Q2 + v2u2 + h(t x) (55b)

Q = uprime (55c)

Evidently f = minus(Q + v) is the v-component of the physical flux vector

F (vQ) equiv(

Fu

Fv

)

=

(

0

f

)

(56)

Note that F has the same structure as (u v)T Borrowing from the presentation for theBSSN system we write the analogous semidiscrete scheme on each subdomain D

k for themodel system

parttu = Q+ av minus u3 + g(t) (57a)

parttv = DQ+Dv minus (u+ v)Q2 + v2u2 + h(t) +Mminus1ℓk(fh minus f lowast)∣

∣

∣

bk

ak (57b)

Q = Du+Mminus1ℓk(ulowast minus uh)∣

∣

∣

bk

ak (57c)

Here we have suppressed the subinterval label k from all variables except for the vectorℓk of Lagrange polynomial values Moreover following the guidelines discussed above thenumerical fluxes are given by

f lowast = fh+1 +

radica

2

[[

vh]]

ulowast = uh minus1

2

[[

uh]]

(58)

Appendix C analyzes the stability of our scheme for a more general numerical flux choiceas applied to (53) with the nonlinear and source terms dropped

IV RESULTS FROM NUMERICAL SIMULATIONS

This section presents results found by numerically solving both the model system (53)and BSSN system (4) with the dG scheme presented in Sec III

16

10minus10

10minus6

10minus2

∆uinfin

10minus10

10minus6

10minus2

∆vinfin

0 500 1000 1500 2000 2500 3000 3500 400010

minus10

10minus6

10minus2

∆(partxu)infin

t

N = 3N = 6N = 9N = 12

FIG 1 Spectral convergence of fields for model PDE Respectively for N = 3 6 9 12

a timestep of ∆t = 00578 00178 00084 00049 has been chosen for stability and accuracy In the

title headings for example ∆u equiv unumer minus uexact

A Simulations of the model system

The semi-discrete scheme (57) has been integrated with the classical fourth-order Runge-Kutta method When integrating this system we have first constructed Q at each Runge-Kutta stage and then substituted into the evolution equations (57ab) for u and v Theproblem has been solved on a computational domain [0 4π] comprised of two subdomainswith a timestep chosen small enough for stability The initial data has been taken from thefollowing exact solution to (53)

uprimeexact(t x) =1

2

[

sin(xminus microminust)minus sin(xminus micro+t)]

(59a)

vexact(t x) =1

2radica

[

sin(xminus microminust) + sin(xminus micro+t)]

(59b)

g(t x) = u3exact (59c)

h(t x) = (uexact + vexact)(uprimeexact)

2 minus v2exactu2exact (59d)

where the speeds microplusmn are found in (54) Specification of the boundary condition at a physicalendpoint amounts to choosing the external state for at the endpoint We have considered

17

8 16 32 64 128

10minus3

10minus2

10minus1

N

∆t

max stable ∆t8N2

FIG 2 Scaling of maximum stable ∆t with N for model PDE

two possibilities (i) the analytic state (Q+ v+) = (Qexact vexact) and (ii) an upwind stateFor example at x = 4π the upwind state is4

Q+ = Qupwind =1

2

[

(Xminus)exact minus (X+)numer

]

v+ = vupwind =1

2radica

[

(Xminus)exact + (X+)numer

]

(60)Either choice of (Q+ v+) leads to similar results and the plots here correspond to theanalytic state Figure 1 clearly shows spectral convergence with increasing polynomial orderN across all fields for the case a = 2 Other values of a including a = 1 for which X+

is a static characteristic field have also been considered with similar results Appendix Cdemonstrates that our proposed scheme for the system (57) with nonlinear and source termsdropped is stable in a semi-discrete sense Nevertheless the fully discrete scheme obtainedvia temporal discretization by the fourth-order Runge-Kutta method is still subject to thestandard absolute stability requirement Namely if microh is any eigenvalue corresponding tothe (linearized) discrete spatial operator then a necessary condition for stability is thatmicroh∆t lies in absolute stability region for Runge-Kutta 4 We here show empirically that theassociated timestep restriction scales like Nminus2 ie ∆t = O(Nminus2) for stability We note thatsuch scaling is welcome in light of the second-order spatial operators which appear in thesystem and suggest a possible worse scaling like Nminus4 Fig 2 plots the maximum stabletimestep for a range of N demonstrating the Nminus2 scaling in line with behavior known fromanalysis of first-order systems [66] This scaling also holds for the BSSN system

4 We remind the reader that unfortunately the plusmn on Xplusmn means something different than the plusmn indicating

exteriorinterior dG states [cf footnote 3]

18

B Simulations of the BSSN system

This subsection documents results for simulations of the unit-mass-parameter (M = 1)Schwarzschild solution (B9) expressed in terms of ingoing Kerr-Schild coordinates Sincethe solution is stationary temporal integration of the semi-discrete scheme has been carriedout with the forward Euler method The r-coordinate domain [04 34] has been split into3 equally spaced subdomains and we have set η = 10 λ = 01 and C = 2 [cf Eq (41)]For all simulations ∆t has been chosen for stability With the chosen λ the inner physicalboundary rmin = 04 is an excision surface At each timestep we have applied an (order2s = 20) exponential filter on the top two-thirds of the modal coefficient set for all fieldsexcept for grr and gθθ For stability we have empirically observed that grr and gθθ must not

be filtered A detailed understanding of this is still lackingIssues related to physical boundary conditions are similar to the one encountered in

Sec IVA for the model problem Similar to before we have retained Eqs (3842) as thechoice of numerical flux even at the endpoints Therefore at an endpoint the specificationof the boundary condition amounts to the choice W+ of external state We have typicallychosen the inner boundary of the radial domain as an excision boundary and in this caseW+ = Wminus is enforced at the inner physical boundary At the outer physical boundaryfor W+ we have again considered two choices (i) Wexact and (ii) Wupwind To enforcechoice (ii) the inverse transformation (17) must be used with incoming characteristic fieldsfixed to their exact values similar to (60) We have tried various versions of choice (ii)and in all cases the resulting simulations have been unstable We therefore present resultscorresponding to choice (i) Although the choice of an analytical external state Wexact

at the outer boundary is stable for our problem such a boundary condition is unlikelyto generalize to more complicated scenarios involving dynamical fields Indeed the issueof outer boundary conditions for the BSSN system is an active area of research with aproper treatment requiring fixation of incoming radiation control of the constraints andspecification of gauge (see Ref [86] for a recent analysis)

For BSSN simulations our main diagnostic is to monitor the Hamiltonian momentumand conformal connection constraints Figure 3 depicts long-time histories of constraintviolations whereas Figs 4 and 5 depict long-time error histories for the individual BSSNfield components From the middle plot in Fig 5 we infer that up to the indicated numericalerror the factor g sin4 θ = grr(gθθ)

2 remains at its initial fixed profile r4 throughout theevolution These figures indicate that the proposed scheme is stable for long times andexhibits spectral converge with increased polynomial order N Similar results are recoveredfromM = 0 Minkowski initial data The stability documented in these plots does not appearto rely on inordinate parameter tuning For example with the fixed parameters describedabove we obtain similar plots if we individually vary (i) rmin over 0325 035 04 0475(values still corresponding to an excision surface for the given choice of λ) (ii) η over1 3 7 10 (iii) s over 8 9 10 With the polynomial order N ranging over 23 26 29 31both stability and qualitatively similar exponential convergence is achieved with a singlesubdomain Likewise adoption of a larger coordinate domain with more subdomains doesnot impact our results Finally we have considered the addition of random noise to all fieldcomponents at the initial time Precisely with the system component χ as an example wehave set

χ(0) 7rarr χ(0) + δχ(0) (61)

where each component (nodal value) of δχ(0) is 10minus5 times a random variable drawn from a

19

10minus9

10minus7

10minus5

10minus3

Hinfin

10minus9

10minus7

10minus5

10minus3

Mrinfin

0 2000 4000 6000 8000 10000

10minus9

10minus7

10minus5

10minus3

Grinfin

t

N = 11N = 14N = 17N = 20

FIG 3 Spectral convergence of constraint violations for M = 1 Kerr-Schild ini-

tial data Respectively for N = 11 14 17 19 a timestep of ∆t ≃ 00041 00026 00018 00013

has been chosen for stability and accuracy

standard normal distribution Such perturbed initial data also gives rise to stable evolutions

V CONCLUSION

We have introduced a discontinuous Galerkin method for solving the spherically reducedBSSN system with second-order spatial operators Our scheme shares similarities withother discontinuous Galerkin methods that use local auxiliary variables to handle high-orderspatial derivatives [66 77ndash79 82 84 85] and which have typically been applied to eitherelliptic parabolic or mixed type problems The key ingredient of a stable dG scheme is anappropriate choice of numerical flux and our particular choice has been motivated by theanalysis presented in Appendix C When used to evolve the Schwarzschild solution in Kerr-Schild coordinates our numerical implementation of the BSSN system (4) is robustly stableand converges to the analytic solution exponentially with increased polynomial order Byapproximating the spatially second-order form of the BSSN system we have not introducedextra fields which are evolved Evolved auxiliary fields result in new constraints which mayspoil stability Our main goal has been stable evolution of the spherically reduced BSSNsystem as a first step towards understanding how a discontinuous Galerkin method might be

20

10minus9

10minus7

10minus5

∆Arrinfin

10minus9

10minus7

10minus5

∆Kinfin

0 2000 4000 6000 8000 10000

10minus9

10minus7

10minus5

max(∆Brinfin ∆Γrinfin)

t

N = 11N = 14N = 17N = 20

FIG 4 Spectral convergence of solution for M = 1 Kerr-Schild initial data

Timestep choices are described in the caption for Fig 3 In the title headings for example

∆Arr equiv (Arr)numer minus (Arr)exact

applied to the full BSSN system Towards that goal we now discuss treatment of singularitiesand generalization of the described dG method to higher space dimension

To deal with the fixed Schwarzschild singularity we have used excision which is easy inthe context of the spherically reduced BSSN system However excision for the binary blackhole problem in full general relativity requires attention to the technical challenge of hori-zon tracking State-of-the-art BSSN codes avoid such complication relying instead on themoving-puncture technique While the moving-puncture technique does involve mild centralsingularities it may still prove amenable to spectral methods Indeed spectral methods fornon-smooth problems is well-developed in both theory and for complex applications Sincethe moving-puncture technique can be performed in spherical symmetry [60] a first-steptoward a spectral moving-puncture code would be to implement a moving puncture withthe nodal dG method described here Such an implementation may adopt Legendre-Gauss-Radau nodes on the innermost subdomain thereby ensuring that the physical singularitydoes not lie on a nodal point (in much the same way finite difference codes use a staggeredgrid) Beyond traditional excision and moving punctures one might also retain a smoothproblem for which spectral methods are especially well-suited via the turducken approach[87 88] to singularities

21

10minus11

10minus9

10minus7

10minus5

∆χinfin

10minus11

10minus9

10minus7

10minus5

max(∆grrinfin ∆gθθinfin)

0 2000 4000 6000 8000 10000

10minus11

10minus9

10minus7

10minus5

max(∆αinfin ∆βrinfin)

t

N = 11N = 14N = 17N = 20

FIG 5 Spectral convergence of solution violations for M = 1 Kerr-Schild initial

data See the caption of Fig 4 for details

Discontinuous Galerkin methods for hyperbolic problems in two and three space dimen-sions are well-developed A generalization of the method described here to three-dimensionsand the full BSSN system would likely rely on an unstructured mesh Appropriate local poly-nomial expansions for the subelements are well-understood as are choices for the numericalfluxes which would now live on two-dimensional faces rather than single points Whether ornot it would ultimately prove successful generalization of our dG method to a higher dimen-sion would rely on an established conceptual framework Further computational advancesof relevance to a generalization of our dG method to the full BSSN system (possibly includ-ing matter) may include mesh hp-adaptivity local timestepping shock capturing and slopelimiting techniques [66] Moreover recent work [89] indicates that enhanced performancewould be expected were our scheme implemented on graphics processor units

VI ACKNOWLEDGMENTS

We thank Nick Taylor for discussions about treating second-order operators with spectralmethods Benjamin Stamm for helping polish up a few parts of Appendix C Khosro Shah-bazi for discussions on LDG and IP methods We also acknowledge helpful conversationswith David Brown and Manuel Tiglio about the BSSN system and previous work on its nu-

22

merical implementation We gratefully acknowledge funding through grants DMS 0554377and DARPAAFOSR FA9550-05-1-0108 to Brown University and NSF grant PHY 0855678to the University of New Mexico

Appendix A Hyperbolicity of the first-order system

This appendix analyzes the matrix A(u) appearing in (10) in order to construct thecharacteristic fields (14) In matrix form the sector (8) of the principal part of (10) reads asfollows

partt

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

=

βr 0 minus 4λα3grr

0 0 λβr

6(grr)2λβr

3gθθgrr0 4λ

3grr

0 βr 0 23grrαχ

13α minus αχ

3grr

αχ3gθθ

minus23χ 0

0 0 βr 0 0 0 0 minus χgrr

0

0 0 minus 4α3grr

βr 0 βr

6(grr)2βr

3gθθgrr0 4

3grr

0 0 23αχ 0 βr minus βrχ

3grrminus2βrχ

3gθθ0 minus2

3χ

0 minus2α 0 0 0 23βr minus2grrβr

3gθθ0 4

3grr

0 gθθαgrr

0 0 0 minusgθθβr

3grr13βr 0 minus2

3gθθ

0 0 minus2α 0 0 0 0 βr 0

34

0 0 0 0 0 0 0 βr

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

prime

(A1)which defines the matrix A(u) appearing in (9) and so also the matrix A(u) in (10) Notethat in the last equation the matrix within the square brackets is minusA(u) For certainconfigurations of u and λ the system (10) is strongly hyperbolic [76] that is A(u) hasa complete set of eigenvectors and real eigenvalues Indeed five eigenpairs of A(u) aretrivially recovered upon inspection of A(u)rsquos leading 5times5 diagonal block These correspondto eigenvalue 0 and the left eigenspace ξj = eTj 1 le j le 5 where ej are the canonical

basis vectors Since each component of u arises as eTj W each is also a characteristic field

The remaining nine eigenpairs are determined by A(u) The eigenvalues of A(u) are

micro1 = 0 micro23 = minusβr microplusmn4 = minusβr plusmn

radic

2αχ

grr microplusmn

5 = minusβr plusmn α

radic

χ

grr microplusmn

6 = minusβr plusmnradic

λ

grr

(A2)

23

and the corresponding left eigenvectors are

x1 = (0 0 0 0 0 gθθ 2grr 0 0) (A3a)

x2 =

(

0 0 0 grr2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3b)

x3 =

(

grrλ 0 0 0

2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3c)

xplusmn4 =

(

0 0plusmnradic

2αgrrχ

0 0 0 0 1 0

)

(A3d)

xplusmn5 =

(

0∓ 3radicgrrχ

plusmn2

radic

grrχ 2grr

1

χminus 1

grr1

gθθ 0 0

)

(A3e)

xplusmn6 =

(

minus3

4

grrλ 0plusmn α

radicλgrr

(2αχminus λ) 0 0minus βr

8(βrgrr ∓radicλgrr)

minus βrgrr

4gθθ(βrgrr ∓radicλgrr)

αχ

(2αχminus λ)plusmnradic

grrλ

)

(A3f)

where for example xplusmn5 A(u) = microplusmn5 x

plusmn5 Assuming that grr gθθ χ and α are everywhere strictly

positive the eigenvalues are real and the eigenvectors are linearly independent providedthat (15) holds These eigenvectors are easily extended to eigenvectors of A(u) e g asxplusmn6 rarr (01times5 x

plusmn6 ) Then for example the characteristic field

Xplusmn6 equiv (01times5 x

plusmn6 )W = xplusmn6WvQ (A4)

and similarly Xplusmnj = xplusmnj WvQ for j = 4 5 and Xk = xkWvQ for k = 1 2 3 The characteristic

speeds for these fields are microk and microplusmnj With this convention the speeds listed in Table I

correspond to the Xk and Xplusmnj in (14)

Appendix B Schwarzschild solution in conformal Kerr-Schild coordinates

In Kerr-Schild coordinates here the system directly related to incoming Eddington-Finkelstein null coordinates the line element for the Schwarzschild solution reads

ds2 = minusα2dt2 + (1 + 2MR)(dR + βRdt)2 +R2dθ2 +R2 sin2 θdφ2 (B1)

where R is the area radius α = (1 + 2MR)minus12 is the lapse and βR = 2M(R + 2M) isthe shift vector The physical spatial metric gab is the spatial part of this line element

To define the corresponding solution to the BSSN system we use equation gab = χgab todefine the following relationship between line elements

dr2 + r2(dθ2 + sin2 θdφ2) = χ[(1 + 2MR)dR2 +R2dθ2 +R2 sin2 θdφ2] (B2)

so that

χ

(

1 +2M

R

)(

dR

dr

)2

= 1 χR2 = r2 (B3)

Then we have(

1 +2M

R

)12dR

R=dr

r (B4)

24

with integration yielding

r =R

4

(

1 +

radic

1 +2M

R

)2

e2minus2radic

1+2MR (B5)

where the constant of integration has been chosen so that the R r rarr infin limits are consistentThe second relation in (B3) shows that

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR =rArr χminus4 =2eradic

1+2MRminus1

1 +radic

1 + 2MR (B6)

The extrinsic curvature tensor is specified by the expression forK given in (B9h) the identityK = KR

R + 2Kθθ and

Kθθ =

(

1 +2M

R

)minus122M

R2 (B7)

Since KRR = Kr

r we compute that

Krr = K minus 2Kθ

θ = minus(

1 +2M

R

)minus12(R +M

R + 2M

)

2M

R2 (B8)

Next since Krr = grrKrr = χminus1Kr

r we have Krr = Arr+

13grrK This implies Arr = Kr

r minus 13K

from which we get (B9g) In all we have

α =

(

1 +2M

R

)minus12

(B9a)

βr = βR dr

dR= χ12

(

1 +2M

R

)minus122M

R(B9b)

grr = 1 (B9c)

gθθ = r2 = χR2 (B9d)

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR (B9e)

Br = 0 (B9f)

Arr = minus(

1 +2M

R

)minus124M

3R2

(

2R + 3M

R + 2M

)

(B9g)

K =

(

1 +2M

R

)minus32(

1 +3M

R

)

2M

R2(B9h)

Γr = minus2

r= minus 2

χ12R (B9i)

To differentiate these expressions with respect to r we use the identity

dR

dr= χminus12

(

1 +2M

R

)minus12

(B10)

along with the chain rule

25

Appendix C Stability of the model system

The following stability analysis for the model system (53) has been inspired by [78 79]After dropping all nonlinear source terms the system (53) becomes

parttu = uprime + av (C1a)

parttv = uprimeprime + vprime (C1b)

This section analyzes the stability of (C1) considering both the continuum system itself aswell as its semi-discrete dG approximation The latter analysis offers some insight into theempirically observed stability of our dG scheme for the spherically reduced BSSN equations

1 Analysis for a single interval

Throughout we work with the L2 inner product and norm

(f g)D =

int

D

fg fD =radic

(f f)D (C2)

where D is the spatial coordinate interval (here D may represent a subdomain Dk or the

whole domain Ω) and we have suppressed all integration measures For the continuummodel we will establish the following estimate

uprime(T middot)2D+ av(T middot)2

Dle C(T )

(

uprime(0 middot)2D+ av(0 middot)2

D

)

(C3)

where the time-dependent constant C(T ) is determined solely by the choice of boundaryconditions To show (C3) we first change variables with v =

radicav thereby rewriting (C1)

in the following symmetric form

parttu = uprime +radicav (C4a)

parttv =radicauprimeprime + vprime (C4b)

Equations (C4ab) then imply

1

2partt

int

D

(uprime)2 =

int

D

uprime(uprimeprime +radicavprime) =

int

D

radicauprimevprime +

1

2

int

partD

(uprime)2 (C5a)

1

2partt

int

D

(v)2 =

int

D

v(radicauprimeprime + vprime) = minus

int

D

radicauprimevprime +

1

2

int

partD

(v2 + 2radicauprimev) (C5b)

Here vvprime and uprimeuprimeprime have been expressed as exact derivatives and then integrated to boundaryterms the second equation employs an extra integration by parts and with only one spacedimension

int

partDdenotes a difference of endpoint evaluations Addition of Eqs (C5ab) gives

1

2partt

int

D

[

v2 + (uprime)2]

=1

2

int

partD

[

v2 + (uprime)2 + 2radicauprimev

]

(C6)

Substitutions with the identities

[

v2 + (uprime)2]

=1

2

[

(v + uprime)2 + (v minus uprime)2]

2uprimev =1

2

[

(v + uprime)2 minus (v minus uprime)2]

(C7)

26

and replacements to recover the original variable v = vradica yield

1

2partt

int

D

[

av2 + (uprime)2]

=1 +

radica

4

int

partD

(radicav + uprime)2 +

1minusradica

4

int

partD

(radicav minus uprime)2 (C8)

From (C8) we deduce that the time-dependent constant C(T ) in (C3) must satisfy

∣

∣

∣

∣

∣

∣

1 +

int T

0

[

1+radica

2

int

partD(radicav + uprime)2 + 1minus

radica

2

int

partD(radicav minus uprime)2

]

dt

uprime(0 middot)2 + av(0 middot)2

∣

∣

∣

∣

∣

∣

le C(T ) (C9)

For periodic boundary conditions we may choose C(T ) = 1 Moreover if a ge 1 anduprime = minusradic

av is specified at partD+ then uprime(t middot)2 + av(t middot)2 decaysStill working on a single interval (subdomain) we now consider the semi-discrete scheme

for (C4) i e (57) with all nonlinear source terms dropped and with v replaced by vradica

Derivation of a formula analogous to (C8) is our first step toward establishing L2 stabilityof the semi-discrete scheme While (57) features vectors for example u(t) taking values atthe Legendre-Gauss-Lobatto nodal points here we work with the numerical solution as apolynomial for example uh(t x) These two representations are related by the Lagrange in-terpolating polynomials for the nodal set here taken to span both the space of test functionsand the space of basis functions Our scheme is

int

Dk

ψparttuh =

int

Dk

ψ(Qh +radicavh) (C10a)

int

Dk

ξparttvh = minusint

Dk

ξprime(radicaQh + vh) +

int

partDk

ξ(radicaQlowast + vlowast) (C10b)

int

Dk

ϕQh =

int

Dk

ϕuprimeh +

int

partDk

ϕ (ulowast minus uh) (C10c)

where ψ ξ and ϕ are polynomial test functions These test functions are arbitrary exceptthat they must be degree-N polynomials In (C10) the variables uh vh and Qh should alsocarry a superscript k but we have suppressed this Derivation of a formula analogous to(C8) is complicated by the fact that Qh is not evolved Nevertheless at a given instant t wecan assemble Qh from (C10c)

Mimicking the calculation (C5b) from the continuum case we first use (C10b) with ξ = vhto write

1

2partt

int

Dk

v2h = minusint

Dk

(radicaQh + vh)v

primeh +

int

partDk

(radicaQlowast + vlowast)vh

= minusint

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2(radicaQlowast + vlowast)vh minus v2h

]

(C11)

The right-hand side of (C5a) is analogous to

1

2partt

int

Dk

Q2h =

int

Dk

QhparttQh (C12)

However since Qh is not evolved the term parttQh must be given a suitable interpretationOn the right hand side of (C10c) only uh u

primeh and u

lowast necessarily depend on time since the

27

test function ϕ need not be time-dependent Furthermore ulowast is explicitly given as a linearcombination of uh as seen in Eq (C21c) below Choosing ϕ = ℓj taking the time derivativeof (C10c) and appealing to the commutivity of mixed partial derivatives we therefore arriveat

int

Dk

ℓjparttQh =

int

Dk

ℓj(parttuh)prime +

int

partDk

ℓj(

(parttu)lowast minus parttuh

)

(C13)

where (parttu)lowast depends on parttuh in precisely the same way that ulowast depends on uh We have

written ℓj rather than ϕ in the last equation to emphasize that the result also holds for anylinear combination of ℓj (for example ϕ) and even for time-dependent combinations SinceQh is itself such a combination we obtain

1

2partt

int

Dk

Q2h =

int

Dk

Qh(parttuh)prime +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

Qh(Qprimeh +

radicavprimeh) +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2((parttu)lowast minus parttuh)Qh +Q2

h

]

(C14)

Addition of (C11) and (C14) gives

1

2partt

int

Dk

(Q2h + v2h) =

1

2

int

partDk

[

Q2h minus v2h + 2(

radicaQlowast + vlowast)vh + 2((parttu)

lowast minus parttuh)Qh

]

(C15)

the aforementioned analog of (C8) This formula holds on a single subdomain Dk and we

now combine multiple copies of it one for each value of k

2 Analysis for multiple intervals

To facilitate combination of (C15) over all k we change notation At every subdomaininterface Ik+12 equiv partDk cap partDk+1 let the superscripts L and R denote field values respectivelytaken from the left and right Then the fields evaluated at Ik+12 which belong to D

k will beuLk+12 v

Lk+12 and Q

Lk+12 while those belonging to D

k+1 will be uRk+12 vRk+12 and Q

Rk+12

However at Ikminus12 the values taken from D

k are uRkminus12 vRkminus12 and QR

kminus12 Note that we

have also replaced the subscript h denoting a numerical solution with k plusmn 12 denotingthe location of the endpoint value of the numerical solution With this notation we define

∆Lα =

1

2

[

(QLα)

2 minus (vLα)2]

+(radic

aQlowastα + vlowastα

)

vLα +[

(parttuα)lowast minus parttu

Lα

]

QLα (C16)

and similarly for ∆Rα The same numerical fluxes appear in both ∆L

α and ∆Rα (ie each

numerical flux takes the same value on either side of an interface) whence fluxes like Qlowastα do

not carry an L or R superscript In terms of these definitions (C15) becomes

1

2partt

int

Dk

(Q2h + v2h) = ∆L

k+12 minus∆Rkminus12 (C17)

28

Summation over all Dk yields

1

2partt

kmaxsum

k=1

int

Dk

(Q2h + v2h) =

kmaxminus1sum

k=1

(

∆Lk+12 minus∆R

k+12

)

+∆Lkmax+12 minus∆R

12

=

kmaxminus1sum

k=1

(

∆Lh minus∆R

h

)∣

∣

Ik+12 +∆Lkmax+12 minus∆R

12 (C18)

We have reverted to h-notation denoting the numerical solution since the LR superscriptsindicate unambiguously the relevant domain used for evaluation at Ik+12

We again seek an estimate of the form

kmaxsum

k=1

(

Qh(T middot)2Dk + avh(T middot)2Dk

)

le C(T )kmaxsum

k=1

(

Qh(0 middot)2Dk + avh(0 middot)2Dk

)

(C19)

that is essentially the same as the one (C3) considered in the continuum case Assume thatthe chosen boundary conditions ensure ∆L

kmax+12 minus ∆R12 is bounded by a time-dependent

constant which does not depend on the numerical parameters N and h (subdomain width)Establishment of stability then amounts to showing that the remaining sum over interfaceterms in (C18) is non-positive whence this remaining sum is consistent with C(T ) le 1although the boundary conditions may give rise to a different bound In fact we will choosethe numerical fluxes such that each individual interface term is non-positive At interfaceIk+12 and in LR notation the jump and average of vh for example are

1

2

(

v+ + vminus)

equiv vh =1

2

(

vLk+12 + vRk+12

)

(C20a)

nminusvminus + n+v+ equiv[[

vh]]

= vLk+12 minus vRk+12 (C20b)

Consider numerical fluxes of the form

Qlowast = Qh minusτQ2

[[

Qh

]]

(C21a)

vlowast = vh minusτv2

[[

vh]]

(C21b)

ulowast = uh minusτu2

[[

uh]]

(C21c)

(parttu)lowast = parttuh minus

τu2

[[

parttuh]]

(C21d)

where (C21c) induces (C21d) and where the penalty parameters τu τv and τQ are realnumbers The fluxes defined in (58) correspond to τu = 1 τv = 1 +

radica and τQ = 0 In

terms of these quantities the kth interface contribution in (C18) is

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

parttuh]]

minus τu2

[[

parttuh]][[

Qh

]]

(C22)

where we have suppressed the k dependence of the right-hand side Now consider the term[[

parttuh]]

Because parttuh and Qh+radicavh are both polynomials of degree N Eq (C10a) implies

29

τu

τ Q

minus1 0 1minus1

minus05

0

05

1

τu

minus1 0 1minus1

minus05

0

05

1

FIG 6 Stable evolutions for the model system For fixed τv = 10minus6 and τv = 1 +radic2

respectively the left and right plots depict stable choices (determined empirically) of τu and τQfor the linear model system (C1) The stable regions are colored black but the jagged edges result

from the discretization of the (τu τQ)-plane

the vector equation parttu = Q +radicav that is pointwise equivalence on the nodal points of

Dk which in turn implies

[[

parttuh]]

=[[

Qh+radicavh]]

Upon substituting this identity into thelast equation we arrive at an expression which features only vh and Qh

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

Qh +radicavh]]

minus τu2

[[

Qh +radicavh]][[

Qh

]]

(C23)

The identities vh[[

vh]]

= 12

[[

v2h]]

and[[

Qh +radicavh]]

=[[

Qh

]]

+radica[[

vh]]

then simplify(C23) to

(∆Lh minus∆R

h )∣

∣

Ik+12 = minusτv2

[[

vh]]2 minus

radica(τu + τQ)

2

[[

Qh

]][[

vh]]

minus τu2

[[

Qh

]]2 (C24)

The role of a penalty parameter is now clear Positive values of τv penalize jumps in vhthrough a negative contribution to the energy Likewise positive values of τu penalizejumps in Qh through a negative contribution to the energy However because the sign of[[

Qh

]][[

vh]]

can be positive or negative only the choice τQ = minusτu yields an expression for(∆L

h minus ∆Rh )|Ik+12 which is manifestly negative for τu ge 0 and τv ge 0 A simple estimate

based on Youngrsquos inequality with ε (that is 2αβ le εminus1α2 + εβ2 where α β ge 0 and ε gt 0)shows that for τQ = 0 the choice τv ge aτu4 also yields stability

Figure 6 depicts certain choices of stable penalty parameters for the linear model systemevolved to tfinal = 1000 (with a = 2 N = 10 and ∆t ≃ 00553) as determined empiricallywith simulations similar to those described in Sec IVA The left plot corresponds to asmall τv = 10minus6 for which the choice τu = 1 τQ = 0 is not stable as expected from thetheoretical analysis However the right plot corresponds to τv = 1 +

radica for which τu = 1

τQ = 0 is stable Motivated by the numerical flux choices (3842) used for the BSSN system

30

(4) we have (as mentioned above) set τu = 1 τv = 1 +radica and τQ = 0 in simulations of

the nonlinear model (53) For the nonlinear model system (53) the theoretically motivatedchoice τQ = minusτu also yields numerically stable evolutions when τu ge 0 and τv ge 0

For the nonlinear systems (4) and (53) we do not attempt a formal stability proofNevertheless the results of this appendix have served as a guide for our choices of penaltyparameters For the BSSN system (4) u v and Q are block indices [cf Eq (6)] Similarto the model problem we have penalized Q with τu = 1 with τv chosen large enough toheuristically overcome the cross-terms of indefinite size that arise from τQ = 0 (we interpretequations like τu = 1 componentwise) An analogous choice ldquoτQ = minusτurdquo for the BSSNsystem might be possible but would be considerably more complicated Indeed such achoice likely entails a matrix of penalty parameters but we do not give the details here

[1] F Pretorius Evolution of Binary Black-Hole Spacetimes Phys Rev Lett 95 121101 (2005)

(4 pages)

[2] F Pretorius Simulation of binary black hole spacetimes with a harmonic evolution scheme

Class Quantum Grav 23 S529-S552 (2006)

[3] M Campanelli C O Lousto P Marronetti and Y Zlochower Accurate Evolutions of Or-

biting Black-Hole Binaries without Excision Phys Rev Lett 96 111101 (2006) (4 pages)

[4] J G Baker J Centrella D-I Choi M Koppitz and J van Meter Gravitational-Wave

Extraction from an Inspiraling Configuration of Merging Black Holes Phys Rev Lett 96

111102 (2006) (4 pages)

[5] M Campanelli C O Lousto and Y Zlochower Last orbit of binary black holes Phys Rev D

73 061501(R) (2006) (5 pages)

[6] F Herrmann I Hinder D Shoemaker and P Laguna Unequal mass binary black hole plunges

and gravitational recoil Class Quantum Grav 24 S33-S42 (2007)

[7] P Diener F Herrmann D Pollney E Schnetter E Seidel R Takahashi J Thornburg and

J Ventrella Accurate Evolution of Orbiting Binary Black Holes Phys Rev Lett 96 121101

(2006) (4 pages)

[8] M A Scheel H P Pfeiffer L Lindblom L E Kidder O Rinne and S A Teukolsky Solving

Einsteinrsquos Equations With Dual Coordinate Frames Phys Rev D 74 104006 (2006)

[9] U Sperhake Binary black-hole evolutions of excision and puncture data Phys Rev D 76

104015 (2007) (20 pages)

[10] B Brugmann J A Gonzalez M Hannam S Husa U Sperhake and W Tichy Calibration

of moving puncture simulations Phys Rev D 77 024027 (2008) (25 pages)

[11] P Marronetti W Tichy B Brugmann J Gonzalez M Hannam S Husa and U Sperhake

Binary black holes on a budget simulations using workstations Class Quantum Grav 24

S43-S58 (2007)

[12] Z B Etienne J A Faber Y T Liu S L Shapiro and T W Baumgarte Filling the holes

Evolving excised binary black hole initial data with puncture techniques Phys Rev D 76

101503(R) (2007) (5 pages)

[13] B Szilagyi D Pollney L Rezzolla J Thornburg and J Winicour An explicit harmonic code

for black-hole evolution using excision Class Quantum Grav 24 S275-S293 (2007)

[14] L Boyle M Kesden and S Nissanke Binary-Black-Hole Merger Symmetry and the Spin

Expansion Phys Rev Lett 100 151101 (2008) (4 pages)

31

[15] M A Scheel M Boyle T Chu L E Kidder K D Matthews and H P Pfeiffer High-

accuracy waveforms for binary black hole inspiral merger and ringdown Phys Rev D 79

024003 (2009) (14 pages)

[16] M Hannam Status of black-hole-binary simulations for gravitational-wave detection

Class Quantum Grav 26 114001 (2009) (21 pages)

[17] I Hinder The current status of binary black hole simulations in numerical relativity

Class Quantum Grav 27 114004 (2010) (19 pages)

[18] B P Abbott et al LIGO The Laser Interferometer Gravitational-Wave Observatory

Rep Prog Phys 72 076901 (2009) (25 pages)

[19] F Acernese et al The status of VIRGO Class Quantum Grav 23 S63-S69 (2006)

[20] F Acernese et al Virgo status Class Quantum Grav 25 184001 (2008) (9 pages)

[21] R Arnowitt S Deser and C W Misner The Dynamics of General Relativity in Gravitation

an introduction to current research edited by L Witten (Wiley New York 1962) Chapter 7

227-265 Also available as gr-qc0405109

[22] J W York Jr Kinematics and Dynamics of General Relativity in Sources of Gravitational

Radiation edited by L L Smarr (Cambridge University Press Cambridge 1979)

[23] L E Kidder M A Scheel and S A Teukolsky Extending the lifetime of 3D black hole

computations with a new hyperbolic system of evolution equations Phys Rev D 64 064017

(2001) (13 pages)

[24] G Calabrese J Pullin O Sarbach and M Tiglio Convergence and stability in numerical

relativity Phys Rev D 66 041501(R) (2002) (4 pages)

[25] S Frittelli and O Reula On the Newtonian Limit of General Relativity Commun Math

Phys 166 221-235 (1994)

[26] M Shibata and T Nakamura Evolution of three-dimensional gravitational waves Harmonic

slicing case Phys Rev D 52 5428-5444 (1995)

[27] Y Choquet-Bruhat and J W York Geometrical well posed systems for the Einstein equations

C R Acad Sci Paris t 321 Serie I 1089-1095 (1995)

[28] A Abrahams A Anderson Y Choquet-Bruhat and J W York Jr Einstein and Yang-Mills

Theories in Hyperbolic Form without Gauge Fixing Phys Rev Lett 75 3377-3381 (1995)

[29] C Bona J Masso E Seidel and J Stela New Formalism for Numerical Relativity

Phys Rev Lett 75 600-603 (1995)

[30] M H P M van Putten and D M Eardley Nonlinear wave equations for relativity

Phys Rev D 53 3056-3063 (1996)

[31] S Frittelli and O A Reula First-Order Symmetric Hyperbolic Einstein Equations with Ar-

bitrary Fixed Gauge Phys Rev Lett 76 4667-4670 (1996)

[32] H Friedrich Hyperbolic reductions for Einsteinrsquos equations Class Quantum Grav 13 1451-

1469 (1996)

[33] F B Estabrook R S Robinson H D Wahlquist Hyperbolic equations for vacuum gravity

using special orthonormal frames Class Quantum Grav 14 1237-1247 (1997)

[34] M S Iriondo E O Leguizamon and O A Reula Einsteinrsquos Equations in Ashtekarrsquos Vari-

ables Constitute a Symmetric Hyperbolic System Phys Rev Lett 79 4732-4735 (1997)

[35] A Anderson Y Choquet-Bruhat and J W York Einstein-Bianchi hyperbolic system for

general relativity Topol Meth Nonlinear Anal 10 353-373 (1997)

[36] T W Baumgarte and S L Shapiro Numerical integration of Einsteinrsquos field equations

Phys Rev D 59 024007 (1998) (7 pages)

[37] M A G Bonilla Symmetric hyperbolic systems for Bianchi equations Class Quantum

9

We will refer to this inverse transformation when discussing outer boundary conditions forour numerical simulations in Sec IVB

III DISCONTINUOUS GALERKIN METHOD

This section describes the nodal discontinuous Galerkin method used to numerically solve(4) We adopt a method-of-lines strategy and here describe the relevant semi-discrete schemewhile leaving the temporal dimension continuous To approximate (4) we follow the generalprocedure first introduced in Ref [82] Our approach defines local auxiliary variables Q = uprimeand rewrites the spatially second-order system (4) as the first-order system (13a) Once weuse (13b) to eliminate Q from (13a) we recover the primal equations (4) The auxiliaryvariable approach was later generalized and coined the local discontinuous Galkerin (LDG)method in Ref [77] We may refer to our particular scheme as an LDG method but notethat many variations exist in the literature We stress that in LDG methods Q is not evolvedand is introduced primarily to assist in the construction of a stable scheme

Equations (12) and (13a) imply that the physical flux function is

F (W ) =

(

Fu(W )

Fv(W )

)

equiv A(u)WvQ =

(

05times1

f(W )

)

f =

fBr

fArr

fKfΓ

(18)

Only the evolution equations for Br Arr K and Γr give rise to non-zero components in F and we have collected these non-zero components into a smaller vector f = Fv Inspectionof (8) determines these components For example from (8c) we find

fK = minusβrK +χ

grrQα (19)

A Local approximation of the system (13)

Our treatment closely follows [83] but with the equations and notations relevant forthis paper Our computational domain Ω is the closed r-interval [a b] We cover Ω withkmax gt 1 non-overlapping intervals Dk = [ak bk] where a = a1 b = bkmax and bkminus1 = ak fork = 2 middot middot middot kmax

On each interval Dk we approximate each component of the system vector W by a localinterpolating polynomial of degree N For example

χkh(t r) =

Nsum

j=0

χ(t rkj )ℓkj (r) (20)

approximates χ(t r) Throughout this section approximations are denoted by a subscripth (see [66] for the notation) For example Wh and fh are approximations of W and f Although Q = uprime Qh and uprimeh are not necessarily the same In (20) ℓkj (r) is the jth Lagrange

polynomial belonging to Dk

ℓkj (r) =Nprod

i=0i 6=j

r minus rkirkj minus rki

(21)

10

Evidently the polynomial χkh interpolates χ at rkj To define the nodes rkj consider the

mapping from the unit interval [minus1 1] to Dk

rk(u) = ak + 12(1 + u)(bk minus ak) (22)

and the N+1 Legendre-Gauss-Lobatto (LGL) nodes uj The uj are the roots of the equation

(1minus u2)P primeN(u) = 0 (23)

where PN(u) is the Nth degree Legendre polynomial and the physical nodes are simplyrkj = rk(uj) In vector notation the approximation (20) takes the form

χkh(t r) = χk(t)Tℓk(r) (24)

in terms of the column vectors

χk(t) =[

χ(t rk0) middot middot middot χ(t rkN)]T ℓk(r) =

[

ℓk0(r) middot middot middot ℓkN(r)]T (25)

On each open interval (ak bk) sub Dk and for each component of the equations in (13) we

define local residuals measuring the extent to which our approximations satisfy the originalcontinuum system Dropping the subdomain label k on the polynomials and focusing onthe K equation as a representative example the local residual corresponding to (4h) is

minus(RK)kh equivminus parttKh + (βrK prime)h minus

(

χQprimeα

grr

)

h

+

(

χQgrrQα

2g2rr

)

h

minus(

χQgθθQα

grrgθθ

)

h

+

(

QαQχ

2grr

)

h

+

(

3αA2rr

2g2rr

)

h

+

(

1

3αK2

)

h

(26)

Here for example the expressions 2

(βrK prime)h = βrhK

primeh

(

QαQχ

2grr

)

h

=QαhQχh

2grrh (27)

We similarly construct the remaining eight residuals eg (Rgrr)h and (RΓr)h as well as fiveresiduals corresponding to (13b) For example one of these remaining five is

(RQα)kh equiv minusQαh + αprime

h (28)

Let the kth inner product be defined as

(

u v)

Dk equivint bk

akdru(r)v(r) (29)

and consider the expression (ℓkj (RK)kh)Dk We call the requirement that this inner product

vanish forallj the kth Galerkin condition For each component of the system and for each k

2 At this stage the first expression is generically a polynomial of degree 2N minus 1 and the latter is not a

polynomial The conventions adopted in Eq (27) prove useful while working with the residual However

later on in Sec III C to obtain the final form (47) of the numerical approximation corresponding to (26)

we will replace nonlinear terms with degree-N polynomials

11

there is a corresponding Galerkin condition in total 9kmax(N + 1) equations for (13a) and5kmax(N+1) for (13b) Enforcement of the Galerkin conditions on each D

k will not recover ameaningful global solution since they provide no mechanism for coupling the local solutionson the different intervals Borrowing from the finite volume toolbox we achieve couplingthrough integration by parts on r and introduction of the numerical flux f lowast at the interfacebetween subdomains

In (26) we only need to consider (βrK prime)h and (χQprimeαgrr)h as the other terms comprise a

component of the source vector Sh Using integration by parts we write

(

ℓkj (βrK prime)h

)

Dk =minusint bk

akdr[

(

ℓkjβrh

)primeKh

]

+ (βrhKh) ℓ

kj

∣

∣

∣

bk

ak (30a)

(

ℓkj (χQprimeαgrr)h

)

Dk =minusint bk

akdr

[(

ℓkjχh

grrh

)prime

Qαh

]

+

(

χhQαh

grrh

)

ℓkj

∣

∣

∣

bk

ak (30b)

In these formulas we have retained the domain index k on ℓkj while continuing to suppress iton Kh grrh etc Moreover we have suppressed the r-dependence in all terms on the right-hand side Addition of these formulas along with the definition fKh = minus(βrK)h+(χQαgrr)hgives

(

ℓkj (βrK prime)h minus (χQprime

αgrr)h)

Dk = minusint bk

akdr

[

(

ℓkjβrh

)primeKh minus

(

ℓkjχh

grrh

)primeQαh

]

minus fKhℓkj

∣

∣

∣

bk

ak (31)

In lieu of (31) we will instead work with the replacement

(

ℓkj (βrK prime)h minus (χQprime

αgrr)h)

Dk rarr minusint bk

akdr

[

(

ℓkjβrh

)primeKh minus

(

ℓkjχh

grrh

)primeQαh

]

minus f lowastKℓ

kj

∣

∣

∣

bk

ak (32)

This replacement features a component f lowastK of the numerical flux rather than a component

fKh of the boundary flux The numerical flux is determined by (as yet not chosen) functions3

f lowast = f lowast(W+Wminus) (33)

where for example Wminus is an interior boundary value [either W kh (t a

k) or W kh (t b

k)] of theapproximation defined on D

k andW+ is an exterior boundary value [either W kminus1h (t bkminus1) or

W k+1h (t ak+1)] of the approximation defined on either Dkminus1 or Dk+1 We discuss our choice

of numerical flux in the next subsection We now employ additional integration by parts towrite the above replacement as

(

ℓkj (βrK prime)h minus (χQprime

αgrr)h)

Dk rarrint bk

akdrℓkj

(

βrK prime minus χQprimeα

grr

)

h

+ (fKh minus f lowastK) ℓ

kj

∣

∣

∣

bk

ak (34)

Rather than the exact kth Galerkin condition(

ℓkj (RK)kh

)

Dk = 0 forallj for the K component

of (13) on Dk we will instead strive to enforce

(

ℓkj (RK)kh

)

Dk = (fKh minus f lowastK) ℓ

kj

∣

∣

∣

bk

ak forallj (35)

3 In the context of the dG method here + and minus denote ldquoexteriorrdquo and ldquointeriorrdquo and have no relation to

the plusmn using to denote the characteristic fields and speeds in Table I For characteristic fields and speeds

+ and minus mean ldquoright-movingrdquo and ldquoleft-movingrdquo

12

although our treatment of nonlinear terms will lead to a slight modification of these equations(we return to this issue shortly) The other components of (13a) are treated similarly asare the components of (13b) Recall that for example Qα = αprime Formally using the samedG method to solve for Qα we arrive at the replacement

(

ℓkj (RQα)kh

)

Dk rarrint bk

akdrℓkj (minusQαh + αprime

h)minus (αh minus αlowast) ℓkj

∣

∣

∣

bk

ak (36)

which again features a component αlowast of the numerical flux The auxiliary variables areconstructed and used at each stage of temporal integration We then have

(

ℓkj (RQα)kh

)

Dk = (αh minus αlowast) ℓkj

∣

∣

∣

bk

ak forallj (37)

as the corresponding enforced kth Galerkin condition

B Numerical Flux

To further complete our dG scheme we must specify functional forms for the compo-nents of the numerical flux introduced in the previous section We distinguish between thephysical fluxes (components of f) and the auxiliary fluxes (components of u) arising fromthe definition of the auxiliary variables These choices are not independent as the resultingscheme must be stable and consistent Our choice follows [84] which considered diffusionproblems Additional analysis of this flux choice appears in [66 85]

Let us first consider the numerical fluxes corresponding to the physical fluxes and of theform (33) The numerical flux vector is a function of the system and auxiliary variablesinterior and exterior to a subdomain A common choice for f lowast is

f lowast = fh+τ

2

[[

vh]]

K-component of f lowast f lowastK = fKh+

τ

2

[[

Kh

]]

(38)

where as an example we have also shown the component of f lowast corresponding to the analysisabove Respectively the average and jump across the interface are

fh =1

2

(

f+ + fminus) [[

vh]]

= nminusvminus + n+v+ (39)

Here τ is a position dependent penalty parameter (fixed below) and nminus(n+) is the localoutward pointing normal to the interior (exterior) subdomain The role of τ is to ldquopenalizerdquo(i e yield a negative contribution to the L2 energy norm) jumps across an interface Anappropriate choice of τ will ensure stability and we now provide some motivation for thechoice (41) of τ we make below

Were we treating the fully first-order system (10) the local Lax-Friedrichs flux wouldoften be a preferred choice due to its simplicity [66] In this case the constant ω in thenumerical flux formula Flowast = Fh + 1

2ω[[

Wh

]]

obeys ω ge max∣

∣micro(nablaWF(W ))∣

∣ HereF(W ) = A(u)W the notation micro(middot) indicates the spectral radius of the matrix within andthe max is taken over interior Wminus and exterior W+ states Motivated by (9) we adopt asimilar but simpler prescription substituting the field gradient

nablaWvQA(u)WvQ = A(u) (40)

13

for nablaWF(W ) Precisely we assume the scaling

τ(bk) = τ(ak+1) = τk+12 equiv C middotmax∣

∣micro(

A(u))∣

∣ (41)

where C = O(1) is a constant chosen for stability Larger values of C will result in schemeswith better stability properties whereas too large a value will impact the CFL conditionAt the interface point Ik+12 equiv D

k cap Dk+1 the vector uh has two representations uminus at bk

and u+ at ak+1 The max in (41) is taken over the corresponding two sets of field speedsMore precisely the speeds in Table I are computed for both uminus and u+ and the maximumtaken over all resulting speeds For the auxiliary variables a penalized central flux is usedThe definition with one representative component is

ulowast = uh minus1

2

[[

uh]]

α-component of ulowast αlowast = αh minus1

2

[[

αh

]]

(42)

with similar expressions for the remaining componentsWe stress the following point Since the interior coupling between subdomains is achieved

through the numerical flux forms (41) and (42) the inverse transformation (17) expressingthe fundamental fields in terms of the characteristic fields is not required to achieve thiscoupling On the other hand imposition of physical boundary conditions may still rely on(17) since this transformation allows one to fix only incoming characteristic modes

C Matrix form of the semi-discrete equations

Let us introduce the kth mass and stiffness matrices

Mkij =

int bk

akdrℓki (r)ℓ

kj (r) Sk

ij =

int bk

akdrℓki (r)ℓ

kjprime(r) (43)

These matrices belong to Dk and the corresponding matrices defined on the reference interval[minus1 1] are

Mij =

int 1

minus1

duℓi(u)ℓj(u) Sij =

int 1

minus1

duℓi(u)ℓprimej(u) (44)

where ℓj(u) is the jth Lagrange polynomial determined by the LGL nodes uj on [minus1 1]These matrices are related by Mk

ij =12(bk minus ak)Mij and S

kij = Sij whence only the reference

matrices require computation and storageWe will use the matrices Mk and Sk in obtaining an ODE system from (26) and (35)

Towards this end we first approximate the nonlinear terms (products and quotients) in(26) by degree-N interpolating polynomials Such approximations are achieved throughpointwise representations For example (QαQχgrr)h appears in (26) and is expressed inthe following way [cf footnote 2]

(

QαQχ

grr

)

h

(t r) =

(

QαhQχh

grrh

)

(t r) rarrNsum

j=0

Qαh(t rkj )Qχh(t r

kj )

grrh(t rkj )

ℓkj (r) (45)

Note that the expressions on the right and left are not equivalent due to aliasing error [65]Our vector notation for this replacement will be

(

QαQχ

grr

)

h

(t r) rarr(

QαQχ

grr

)

(t)Tℓk(r) (46)

14

Operations among bold variables are always performed pointwise Making similar replace-ments for all terms in (26) and then carrying out the integrations in (35) which bring inMk and Sk we arrive at

parttK = βrDK minus χDQα

grr

+1

2

χQgrrQα

g2rr

minusχQgθθ

Qα

grrgθθ

+1

2

QαQχ

grr

+3

2

αA2rr

g2rr

+1

3αK2 +Mminus1ℓk (fKh minus f lowast

K)∣

∣

∣

bk

ak (47)

where we have again suppressed the superscript k on all terms except ℓk(r) and the sub-script h is dropped on all boldfaced variables As described in [66] the spectral collocationderivative matrix

(Dk)ij =dℓkjdr

∣

∣

∣

∣

∣

r=rki

(48)

can also be expressed as Dk = (Mk)minus1Sk which appears above Eight other semi-discreteevolution equations are similarly obtained with nine in total (one for each component ofWuv) Additionally we have

Qα = Dα+Mminus1ℓk(αlowast minusαh)∣

∣

∣

bk

ak (49)

as one of the auxiliary equations with five in total (one for each component of Q = WQQ)

D Filtering

Like other nodal (pseudospectral) methods our scheme may suffer from instabilitiesdriven by aliasing error [65] Filtering is a simple yet robust remedy To filter a solu-tion component such as χ we use the modal (as opposed to nodal) representation of thesolution

χkh(t r) =

Nsum

j=0

χ(t rkj )ℓkj (r) =

Nsum

j=0

χkj (t)Pj(r) (50)

where Pj(r) is the jth Legendre polynomial Let ηj = jN and define the filter function

σ(ηj) =

1 for 0 le ηj le NcN

exp(

minus ǫ(

ηjminusNcN

1minusNcN

)2s)

for NcN le ηj le 1(51)

At each timestep we modify our solution component according to

χkh rarr

(

χkh

)filtered=

Nsum

j=0

σ(ηj)χkj (t)Pj(r) (52)

Evidently the modification only affects the top N minus Nc modes and is sufficient to controlthe type of weak instability driven by aliasing [66] The numerical parameters Nc and ǫ areproblem dependent For our simulations we have taken ǫ ≃ minuslog(εmach) = 36 where εmach

is machine accuracy in double precision

15

E Model system

To better illustrate the basic properties of our method we consider a toy model Namelythe following spatially second-order system

parttu = uprime + av minus u3 + g(t x) (53a)

parttv = uprimeprime + vprime minus (u+ v)(uprime)2 + v2u2 + h(t x) (53b)

where a ge 1 is constant and g and h are analytic source terms to be specified In contrastto (6) here u v and Q = uprime are scalars rather than vectors System (53) admits a first-order reduction in which uprime is defined as an extra variable Since this first-order reductionis strongly hyperbolic the spatially second-order system (53) is also strongly hyperbolic byone of the definitions considered in [75] The characteristic fields Xplusmn and speeds microplusmn are

X+ =radicav minus uprime micro+ =

radicaminus 1 Xminus =

radicav + uprime microminus = minus(

radica+ 1) (54)

To construct a local dG scheme for this system we first rewrite it as

parttu = Q+ av minus u3 + g(t x) (55a)

parttv = Qprime + vprime minus (u+ v)Q2 + v2u2 + h(t x) (55b)

Q = uprime (55c)

Evidently f = minus(Q + v) is the v-component of the physical flux vector

F (vQ) equiv(

Fu

Fv

)

=

(

0

f

)

(56)

Note that F has the same structure as (u v)T Borrowing from the presentation for theBSSN system we write the analogous semidiscrete scheme on each subdomain D

k for themodel system

parttu = Q+ av minus u3 + g(t) (57a)

parttv = DQ+Dv minus (u+ v)Q2 + v2u2 + h(t) +Mminus1ℓk(fh minus f lowast)∣

∣

∣

bk

ak (57b)

Q = Du+Mminus1ℓk(ulowast minus uh)∣

∣

∣

bk

ak (57c)

Here we have suppressed the subinterval label k from all variables except for the vectorℓk of Lagrange polynomial values Moreover following the guidelines discussed above thenumerical fluxes are given by

f lowast = fh+1 +

radica

2

[[

vh]]

ulowast = uh minus1

2

[[

uh]]

(58)

Appendix C analyzes the stability of our scheme for a more general numerical flux choiceas applied to (53) with the nonlinear and source terms dropped

IV RESULTS FROM NUMERICAL SIMULATIONS

This section presents results found by numerically solving both the model system (53)and BSSN system (4) with the dG scheme presented in Sec III

16

10minus10

10minus6

10minus2

∆uinfin

10minus10

10minus6

10minus2

∆vinfin

0 500 1000 1500 2000 2500 3000 3500 400010

minus10

10minus6

10minus2

∆(partxu)infin

t

N = 3N = 6N = 9N = 12

FIG 1 Spectral convergence of fields for model PDE Respectively for N = 3 6 9 12

a timestep of ∆t = 00578 00178 00084 00049 has been chosen for stability and accuracy In the

title headings for example ∆u equiv unumer minus uexact

A Simulations of the model system

The semi-discrete scheme (57) has been integrated with the classical fourth-order Runge-Kutta method When integrating this system we have first constructed Q at each Runge-Kutta stage and then substituted into the evolution equations (57ab) for u and v Theproblem has been solved on a computational domain [0 4π] comprised of two subdomainswith a timestep chosen small enough for stability The initial data has been taken from thefollowing exact solution to (53)

uprimeexact(t x) =1

2

[

sin(xminus microminust)minus sin(xminus micro+t)]

(59a)

vexact(t x) =1

2radica

[

sin(xminus microminust) + sin(xminus micro+t)]

(59b)

g(t x) = u3exact (59c)

h(t x) = (uexact + vexact)(uprimeexact)

2 minus v2exactu2exact (59d)

where the speeds microplusmn are found in (54) Specification of the boundary condition at a physicalendpoint amounts to choosing the external state for at the endpoint We have considered

17

8 16 32 64 128

10minus3

10minus2

10minus1

N

∆t

max stable ∆t8N2

FIG 2 Scaling of maximum stable ∆t with N for model PDE

two possibilities (i) the analytic state (Q+ v+) = (Qexact vexact) and (ii) an upwind stateFor example at x = 4π the upwind state is4

Q+ = Qupwind =1

2

[

(Xminus)exact minus (X+)numer

]

v+ = vupwind =1

2radica

[

(Xminus)exact + (X+)numer

]

(60)Either choice of (Q+ v+) leads to similar results and the plots here correspond to theanalytic state Figure 1 clearly shows spectral convergence with increasing polynomial orderN across all fields for the case a = 2 Other values of a including a = 1 for which X+

is a static characteristic field have also been considered with similar results Appendix Cdemonstrates that our proposed scheme for the system (57) with nonlinear and source termsdropped is stable in a semi-discrete sense Nevertheless the fully discrete scheme obtainedvia temporal discretization by the fourth-order Runge-Kutta method is still subject to thestandard absolute stability requirement Namely if microh is any eigenvalue corresponding tothe (linearized) discrete spatial operator then a necessary condition for stability is thatmicroh∆t lies in absolute stability region for Runge-Kutta 4 We here show empirically that theassociated timestep restriction scales like Nminus2 ie ∆t = O(Nminus2) for stability We note thatsuch scaling is welcome in light of the second-order spatial operators which appear in thesystem and suggest a possible worse scaling like Nminus4 Fig 2 plots the maximum stabletimestep for a range of N demonstrating the Nminus2 scaling in line with behavior known fromanalysis of first-order systems [66] This scaling also holds for the BSSN system

4 We remind the reader that unfortunately the plusmn on Xplusmn means something different than the plusmn indicating

exteriorinterior dG states [cf footnote 3]

18

B Simulations of the BSSN system

This subsection documents results for simulations of the unit-mass-parameter (M = 1)Schwarzschild solution (B9) expressed in terms of ingoing Kerr-Schild coordinates Sincethe solution is stationary temporal integration of the semi-discrete scheme has been carriedout with the forward Euler method The r-coordinate domain [04 34] has been split into3 equally spaced subdomains and we have set η = 10 λ = 01 and C = 2 [cf Eq (41)]For all simulations ∆t has been chosen for stability With the chosen λ the inner physicalboundary rmin = 04 is an excision surface At each timestep we have applied an (order2s = 20) exponential filter on the top two-thirds of the modal coefficient set for all fieldsexcept for grr and gθθ For stability we have empirically observed that grr and gθθ must not

be filtered A detailed understanding of this is still lackingIssues related to physical boundary conditions are similar to the one encountered in

Sec IVA for the model problem Similar to before we have retained Eqs (3842) as thechoice of numerical flux even at the endpoints Therefore at an endpoint the specificationof the boundary condition amounts to the choice W+ of external state We have typicallychosen the inner boundary of the radial domain as an excision boundary and in this caseW+ = Wminus is enforced at the inner physical boundary At the outer physical boundaryfor W+ we have again considered two choices (i) Wexact and (ii) Wupwind To enforcechoice (ii) the inverse transformation (17) must be used with incoming characteristic fieldsfixed to their exact values similar to (60) We have tried various versions of choice (ii)and in all cases the resulting simulations have been unstable We therefore present resultscorresponding to choice (i) Although the choice of an analytical external state Wexact

at the outer boundary is stable for our problem such a boundary condition is unlikelyto generalize to more complicated scenarios involving dynamical fields Indeed the issueof outer boundary conditions for the BSSN system is an active area of research with aproper treatment requiring fixation of incoming radiation control of the constraints andspecification of gauge (see Ref [86] for a recent analysis)

For BSSN simulations our main diagnostic is to monitor the Hamiltonian momentumand conformal connection constraints Figure 3 depicts long-time histories of constraintviolations whereas Figs 4 and 5 depict long-time error histories for the individual BSSNfield components From the middle plot in Fig 5 we infer that up to the indicated numericalerror the factor g sin4 θ = grr(gθθ)

2 remains at its initial fixed profile r4 throughout theevolution These figures indicate that the proposed scheme is stable for long times andexhibits spectral converge with increased polynomial order N Similar results are recoveredfromM = 0 Minkowski initial data The stability documented in these plots does not appearto rely on inordinate parameter tuning For example with the fixed parameters describedabove we obtain similar plots if we individually vary (i) rmin over 0325 035 04 0475(values still corresponding to an excision surface for the given choice of λ) (ii) η over1 3 7 10 (iii) s over 8 9 10 With the polynomial order N ranging over 23 26 29 31both stability and qualitatively similar exponential convergence is achieved with a singlesubdomain Likewise adoption of a larger coordinate domain with more subdomains doesnot impact our results Finally we have considered the addition of random noise to all fieldcomponents at the initial time Precisely with the system component χ as an example wehave set

χ(0) 7rarr χ(0) + δχ(0) (61)

where each component (nodal value) of δχ(0) is 10minus5 times a random variable drawn from a

19

10minus9

10minus7

10minus5

10minus3

Hinfin

10minus9

10minus7

10minus5

10minus3

Mrinfin

0 2000 4000 6000 8000 10000

10minus9

10minus7

10minus5

10minus3

Grinfin

t

N = 11N = 14N = 17N = 20

FIG 3 Spectral convergence of constraint violations for M = 1 Kerr-Schild ini-

tial data Respectively for N = 11 14 17 19 a timestep of ∆t ≃ 00041 00026 00018 00013

has been chosen for stability and accuracy

standard normal distribution Such perturbed initial data also gives rise to stable evolutions

V CONCLUSION

We have introduced a discontinuous Galerkin method for solving the spherically reducedBSSN system with second-order spatial operators Our scheme shares similarities withother discontinuous Galerkin methods that use local auxiliary variables to handle high-orderspatial derivatives [66 77ndash79 82 84 85] and which have typically been applied to eitherelliptic parabolic or mixed type problems The key ingredient of a stable dG scheme is anappropriate choice of numerical flux and our particular choice has been motivated by theanalysis presented in Appendix C When used to evolve the Schwarzschild solution in Kerr-Schild coordinates our numerical implementation of the BSSN system (4) is robustly stableand converges to the analytic solution exponentially with increased polynomial order Byapproximating the spatially second-order form of the BSSN system we have not introducedextra fields which are evolved Evolved auxiliary fields result in new constraints which mayspoil stability Our main goal has been stable evolution of the spherically reduced BSSNsystem as a first step towards understanding how a discontinuous Galerkin method might be

20

10minus9

10minus7

10minus5

∆Arrinfin

10minus9

10minus7

10minus5

∆Kinfin

0 2000 4000 6000 8000 10000

10minus9

10minus7

10minus5

max(∆Brinfin ∆Γrinfin)

t

N = 11N = 14N = 17N = 20

FIG 4 Spectral convergence of solution for M = 1 Kerr-Schild initial data

Timestep choices are described in the caption for Fig 3 In the title headings for example

∆Arr equiv (Arr)numer minus (Arr)exact

applied to the full BSSN system Towards that goal we now discuss treatment of singularitiesand generalization of the described dG method to higher space dimension

To deal with the fixed Schwarzschild singularity we have used excision which is easy inthe context of the spherically reduced BSSN system However excision for the binary blackhole problem in full general relativity requires attention to the technical challenge of hori-zon tracking State-of-the-art BSSN codes avoid such complication relying instead on themoving-puncture technique While the moving-puncture technique does involve mild centralsingularities it may still prove amenable to spectral methods Indeed spectral methods fornon-smooth problems is well-developed in both theory and for complex applications Sincethe moving-puncture technique can be performed in spherical symmetry [60] a first-steptoward a spectral moving-puncture code would be to implement a moving puncture withthe nodal dG method described here Such an implementation may adopt Legendre-Gauss-Radau nodes on the innermost subdomain thereby ensuring that the physical singularitydoes not lie on a nodal point (in much the same way finite difference codes use a staggeredgrid) Beyond traditional excision and moving punctures one might also retain a smoothproblem for which spectral methods are especially well-suited via the turducken approach[87 88] to singularities

21

10minus11

10minus9

10minus7

10minus5

∆χinfin

10minus11

10minus9

10minus7

10minus5

max(∆grrinfin ∆gθθinfin)

0 2000 4000 6000 8000 10000

10minus11

10minus9

10minus7

10minus5

max(∆αinfin ∆βrinfin)

t

N = 11N = 14N = 17N = 20

FIG 5 Spectral convergence of solution violations for M = 1 Kerr-Schild initial

data See the caption of Fig 4 for details

Discontinuous Galerkin methods for hyperbolic problems in two and three space dimen-sions are well-developed A generalization of the method described here to three-dimensionsand the full BSSN system would likely rely on an unstructured mesh Appropriate local poly-nomial expansions for the subelements are well-understood as are choices for the numericalfluxes which would now live on two-dimensional faces rather than single points Whether ornot it would ultimately prove successful generalization of our dG method to a higher dimen-sion would rely on an established conceptual framework Further computational advancesof relevance to a generalization of our dG method to the full BSSN system (possibly includ-ing matter) may include mesh hp-adaptivity local timestepping shock capturing and slopelimiting techniques [66] Moreover recent work [89] indicates that enhanced performancewould be expected were our scheme implemented on graphics processor units

VI ACKNOWLEDGMENTS

We thank Nick Taylor for discussions about treating second-order operators with spectralmethods Benjamin Stamm for helping polish up a few parts of Appendix C Khosro Shah-bazi for discussions on LDG and IP methods We also acknowledge helpful conversationswith David Brown and Manuel Tiglio about the BSSN system and previous work on its nu-

22

merical implementation We gratefully acknowledge funding through grants DMS 0554377and DARPAAFOSR FA9550-05-1-0108 to Brown University and NSF grant PHY 0855678to the University of New Mexico

Appendix A Hyperbolicity of the first-order system

This appendix analyzes the matrix A(u) appearing in (10) in order to construct thecharacteristic fields (14) In matrix form the sector (8) of the principal part of (10) reads asfollows

partt

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

=

βr 0 minus 4λα3grr

0 0 λβr

6(grr)2λβr

3gθθgrr0 4λ

3grr

0 βr 0 23grrαχ

13α minus αχ

3grr

αχ3gθθ

minus23χ 0

0 0 βr 0 0 0 0 minus χgrr

0

0 0 minus 4α3grr

βr 0 βr

6(grr)2βr

3gθθgrr0 4

3grr

0 0 23αχ 0 βr minus βrχ

3grrminus2βrχ

3gθθ0 minus2

3χ

0 minus2α 0 0 0 23βr minus2grrβr

3gθθ0 4

3grr

0 gθθαgrr

0 0 0 minusgθθβr

3grr13βr 0 minus2

3gθθ

0 0 minus2α 0 0 0 0 βr 0

34

0 0 0 0 0 0 0 βr

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

prime

(A1)which defines the matrix A(u) appearing in (9) and so also the matrix A(u) in (10) Notethat in the last equation the matrix within the square brackets is minusA(u) For certainconfigurations of u and λ the system (10) is strongly hyperbolic [76] that is A(u) hasa complete set of eigenvectors and real eigenvalues Indeed five eigenpairs of A(u) aretrivially recovered upon inspection of A(u)rsquos leading 5times5 diagonal block These correspondto eigenvalue 0 and the left eigenspace ξj = eTj 1 le j le 5 where ej are the canonical

basis vectors Since each component of u arises as eTj W each is also a characteristic field

The remaining nine eigenpairs are determined by A(u) The eigenvalues of A(u) are

micro1 = 0 micro23 = minusβr microplusmn4 = minusβr plusmn

radic

2αχ

grr microplusmn

5 = minusβr plusmn α

radic

χ

grr microplusmn

6 = minusβr plusmnradic

λ

grr

(A2)

23

and the corresponding left eigenvectors are

x1 = (0 0 0 0 0 gθθ 2grr 0 0) (A3a)

x2 =

(

0 0 0 grr2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3b)

x3 =

(

grrλ 0 0 0

2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3c)

xplusmn4 =

(

0 0plusmnradic

2αgrrχ

0 0 0 0 1 0

)

(A3d)

xplusmn5 =

(

0∓ 3radicgrrχ

plusmn2

radic

grrχ 2grr

1

χminus 1

grr1

gθθ 0 0

)

(A3e)

xplusmn6 =

(

minus3

4

grrλ 0plusmn α

radicλgrr

(2αχminus λ) 0 0minus βr

8(βrgrr ∓radicλgrr)

minus βrgrr

4gθθ(βrgrr ∓radicλgrr)

αχ

(2αχminus λ)plusmnradic

grrλ

)

(A3f)

where for example xplusmn5 A(u) = microplusmn5 x

plusmn5 Assuming that grr gθθ χ and α are everywhere strictly

positive the eigenvalues are real and the eigenvectors are linearly independent providedthat (15) holds These eigenvectors are easily extended to eigenvectors of A(u) e g asxplusmn6 rarr (01times5 x

plusmn6 ) Then for example the characteristic field

Xplusmn6 equiv (01times5 x

plusmn6 )W = xplusmn6WvQ (A4)

and similarly Xplusmnj = xplusmnj WvQ for j = 4 5 and Xk = xkWvQ for k = 1 2 3 The characteristic

speeds for these fields are microk and microplusmnj With this convention the speeds listed in Table I

correspond to the Xk and Xplusmnj in (14)

Appendix B Schwarzschild solution in conformal Kerr-Schild coordinates

In Kerr-Schild coordinates here the system directly related to incoming Eddington-Finkelstein null coordinates the line element for the Schwarzschild solution reads

ds2 = minusα2dt2 + (1 + 2MR)(dR + βRdt)2 +R2dθ2 +R2 sin2 θdφ2 (B1)

where R is the area radius α = (1 + 2MR)minus12 is the lapse and βR = 2M(R + 2M) isthe shift vector The physical spatial metric gab is the spatial part of this line element

To define the corresponding solution to the BSSN system we use equation gab = χgab todefine the following relationship between line elements

dr2 + r2(dθ2 + sin2 θdφ2) = χ[(1 + 2MR)dR2 +R2dθ2 +R2 sin2 θdφ2] (B2)

so that

χ

(

1 +2M

R

)(

dR

dr

)2

= 1 χR2 = r2 (B3)

Then we have(

1 +2M

R

)12dR

R=dr

r (B4)

24

with integration yielding

r =R

4

(

1 +

radic

1 +2M

R

)2

e2minus2radic

1+2MR (B5)

where the constant of integration has been chosen so that the R r rarr infin limits are consistentThe second relation in (B3) shows that

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR =rArr χminus4 =2eradic

1+2MRminus1

1 +radic

1 + 2MR (B6)

The extrinsic curvature tensor is specified by the expression forK given in (B9h) the identityK = KR

R + 2Kθθ and

Kθθ =

(

1 +2M

R

)minus122M

R2 (B7)

Since KRR = Kr

r we compute that

Krr = K minus 2Kθ

θ = minus(

1 +2M

R

)minus12(R +M

R + 2M

)

2M

R2 (B8)

Next since Krr = grrKrr = χminus1Kr

r we have Krr = Arr+

13grrK This implies Arr = Kr

r minus 13K

from which we get (B9g) In all we have

α =

(

1 +2M

R

)minus12

(B9a)

βr = βR dr

dR= χ12

(

1 +2M

R

)minus122M

R(B9b)

grr = 1 (B9c)

gθθ = r2 = χR2 (B9d)

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR (B9e)

Br = 0 (B9f)

Arr = minus(

1 +2M

R

)minus124M

3R2

(

2R + 3M

R + 2M

)

(B9g)

K =

(

1 +2M

R

)minus32(

1 +3M

R

)

2M

R2(B9h)

Γr = minus2

r= minus 2

χ12R (B9i)

To differentiate these expressions with respect to r we use the identity

dR

dr= χminus12

(

1 +2M

R

)minus12

(B10)

along with the chain rule

25

Appendix C Stability of the model system

The following stability analysis for the model system (53) has been inspired by [78 79]After dropping all nonlinear source terms the system (53) becomes

parttu = uprime + av (C1a)

parttv = uprimeprime + vprime (C1b)

This section analyzes the stability of (C1) considering both the continuum system itself aswell as its semi-discrete dG approximation The latter analysis offers some insight into theempirically observed stability of our dG scheme for the spherically reduced BSSN equations

1 Analysis for a single interval

Throughout we work with the L2 inner product and norm

(f g)D =

int

D

fg fD =radic

(f f)D (C2)

where D is the spatial coordinate interval (here D may represent a subdomain Dk or the

whole domain Ω) and we have suppressed all integration measures For the continuummodel we will establish the following estimate

uprime(T middot)2D+ av(T middot)2

Dle C(T )

(

uprime(0 middot)2D+ av(0 middot)2

D

)

(C3)

where the time-dependent constant C(T ) is determined solely by the choice of boundaryconditions To show (C3) we first change variables with v =

radicav thereby rewriting (C1)

in the following symmetric form

parttu = uprime +radicav (C4a)

parttv =radicauprimeprime + vprime (C4b)

Equations (C4ab) then imply

1

2partt

int

D

(uprime)2 =

int

D

uprime(uprimeprime +radicavprime) =

int

D

radicauprimevprime +

1

2

int

partD

(uprime)2 (C5a)

1

2partt

int

D

(v)2 =

int

D

v(radicauprimeprime + vprime) = minus

int

D

radicauprimevprime +

1

2

int

partD

(v2 + 2radicauprimev) (C5b)

Here vvprime and uprimeuprimeprime have been expressed as exact derivatives and then integrated to boundaryterms the second equation employs an extra integration by parts and with only one spacedimension

int

partDdenotes a difference of endpoint evaluations Addition of Eqs (C5ab) gives

1

2partt

int

D

[

v2 + (uprime)2]

=1

2

int

partD

[

v2 + (uprime)2 + 2radicauprimev

]

(C6)

Substitutions with the identities

[

v2 + (uprime)2]

=1

2

[

(v + uprime)2 + (v minus uprime)2]

2uprimev =1

2

[

(v + uprime)2 minus (v minus uprime)2]

(C7)

26

and replacements to recover the original variable v = vradica yield

1

2partt

int

D

[

av2 + (uprime)2]

=1 +

radica

4

int

partD

(radicav + uprime)2 +

1minusradica

4

int

partD

(radicav minus uprime)2 (C8)

From (C8) we deduce that the time-dependent constant C(T ) in (C3) must satisfy

∣

∣

∣

∣

∣

∣

1 +

int T

0

[

1+radica

2

int

partD(radicav + uprime)2 + 1minus

radica

2

int

partD(radicav minus uprime)2

]

dt

uprime(0 middot)2 + av(0 middot)2

∣

∣

∣

∣

∣

∣

le C(T ) (C9)

For periodic boundary conditions we may choose C(T ) = 1 Moreover if a ge 1 anduprime = minusradic

av is specified at partD+ then uprime(t middot)2 + av(t middot)2 decaysStill working on a single interval (subdomain) we now consider the semi-discrete scheme

for (C4) i e (57) with all nonlinear source terms dropped and with v replaced by vradica

Derivation of a formula analogous to (C8) is our first step toward establishing L2 stabilityof the semi-discrete scheme While (57) features vectors for example u(t) taking values atthe Legendre-Gauss-Lobatto nodal points here we work with the numerical solution as apolynomial for example uh(t x) These two representations are related by the Lagrange in-terpolating polynomials for the nodal set here taken to span both the space of test functionsand the space of basis functions Our scheme is

int

Dk

ψparttuh =

int

Dk

ψ(Qh +radicavh) (C10a)

int

Dk

ξparttvh = minusint

Dk

ξprime(radicaQh + vh) +

int

partDk

ξ(radicaQlowast + vlowast) (C10b)

int

Dk

ϕQh =

int

Dk

ϕuprimeh +

int

partDk

ϕ (ulowast minus uh) (C10c)

where ψ ξ and ϕ are polynomial test functions These test functions are arbitrary exceptthat they must be degree-N polynomials In (C10) the variables uh vh and Qh should alsocarry a superscript k but we have suppressed this Derivation of a formula analogous to(C8) is complicated by the fact that Qh is not evolved Nevertheless at a given instant t wecan assemble Qh from (C10c)

Mimicking the calculation (C5b) from the continuum case we first use (C10b) with ξ = vhto write

1

2partt

int

Dk

v2h = minusint

Dk

(radicaQh + vh)v

primeh +

int

partDk

(radicaQlowast + vlowast)vh

= minusint

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2(radicaQlowast + vlowast)vh minus v2h

]

(C11)

The right-hand side of (C5a) is analogous to

1

2partt

int

Dk

Q2h =

int

Dk

QhparttQh (C12)

However since Qh is not evolved the term parttQh must be given a suitable interpretationOn the right hand side of (C10c) only uh u

primeh and u

lowast necessarily depend on time since the

27

test function ϕ need not be time-dependent Furthermore ulowast is explicitly given as a linearcombination of uh as seen in Eq (C21c) below Choosing ϕ = ℓj taking the time derivativeof (C10c) and appealing to the commutivity of mixed partial derivatives we therefore arriveat

int

Dk

ℓjparttQh =

int

Dk

ℓj(parttuh)prime +

int

partDk

ℓj(

(parttu)lowast minus parttuh

)

(C13)

where (parttu)lowast depends on parttuh in precisely the same way that ulowast depends on uh We have

written ℓj rather than ϕ in the last equation to emphasize that the result also holds for anylinear combination of ℓj (for example ϕ) and even for time-dependent combinations SinceQh is itself such a combination we obtain

1

2partt

int

Dk

Q2h =

int

Dk

Qh(parttuh)prime +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

Qh(Qprimeh +

radicavprimeh) +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2((parttu)lowast minus parttuh)Qh +Q2

h

]

(C14)

Addition of (C11) and (C14) gives

1

2partt

int

Dk

(Q2h + v2h) =

1

2

int

partDk

[

Q2h minus v2h + 2(

radicaQlowast + vlowast)vh + 2((parttu)

lowast minus parttuh)Qh

]

(C15)

the aforementioned analog of (C8) This formula holds on a single subdomain Dk and we

now combine multiple copies of it one for each value of k

2 Analysis for multiple intervals

To facilitate combination of (C15) over all k we change notation At every subdomaininterface Ik+12 equiv partDk cap partDk+1 let the superscripts L and R denote field values respectivelytaken from the left and right Then the fields evaluated at Ik+12 which belong to D

k will beuLk+12 v

Lk+12 and Q

Lk+12 while those belonging to D

k+1 will be uRk+12 vRk+12 and Q

Rk+12

However at Ikminus12 the values taken from D

k are uRkminus12 vRkminus12 and QR

kminus12 Note that we

have also replaced the subscript h denoting a numerical solution with k plusmn 12 denotingthe location of the endpoint value of the numerical solution With this notation we define

∆Lα =

1

2

[

(QLα)

2 minus (vLα)2]

+(radic

aQlowastα + vlowastα

)

vLα +[

(parttuα)lowast minus parttu

Lα

]

QLα (C16)

and similarly for ∆Rα The same numerical fluxes appear in both ∆L

α and ∆Rα (ie each

numerical flux takes the same value on either side of an interface) whence fluxes like Qlowastα do

not carry an L or R superscript In terms of these definitions (C15) becomes

1

2partt

int

Dk

(Q2h + v2h) = ∆L

k+12 minus∆Rkminus12 (C17)

28

Summation over all Dk yields

1

2partt

kmaxsum

k=1

int

Dk

(Q2h + v2h) =

kmaxminus1sum

k=1

(

∆Lk+12 minus∆R

k+12

)

+∆Lkmax+12 minus∆R

12

=

kmaxminus1sum

k=1

(

∆Lh minus∆R

h

)∣

∣

Ik+12 +∆Lkmax+12 minus∆R

12 (C18)

We have reverted to h-notation denoting the numerical solution since the LR superscriptsindicate unambiguously the relevant domain used for evaluation at Ik+12

We again seek an estimate of the form

kmaxsum

k=1

(

Qh(T middot)2Dk + avh(T middot)2Dk

)

le C(T )kmaxsum

k=1

(

Qh(0 middot)2Dk + avh(0 middot)2Dk

)

(C19)

that is essentially the same as the one (C3) considered in the continuum case Assume thatthe chosen boundary conditions ensure ∆L

kmax+12 minus ∆R12 is bounded by a time-dependent

constant which does not depend on the numerical parameters N and h (subdomain width)Establishment of stability then amounts to showing that the remaining sum over interfaceterms in (C18) is non-positive whence this remaining sum is consistent with C(T ) le 1although the boundary conditions may give rise to a different bound In fact we will choosethe numerical fluxes such that each individual interface term is non-positive At interfaceIk+12 and in LR notation the jump and average of vh for example are

1

2

(

v+ + vminus)

equiv vh =1

2

(

vLk+12 + vRk+12

)

(C20a)

nminusvminus + n+v+ equiv[[

vh]]

= vLk+12 minus vRk+12 (C20b)

Consider numerical fluxes of the form

Qlowast = Qh minusτQ2

[[

Qh

]]

(C21a)

vlowast = vh minusτv2

[[

vh]]

(C21b)

ulowast = uh minusτu2

[[

uh]]

(C21c)

(parttu)lowast = parttuh minus

τu2

[[

parttuh]]

(C21d)

where (C21c) induces (C21d) and where the penalty parameters τu τv and τQ are realnumbers The fluxes defined in (58) correspond to τu = 1 τv = 1 +

radica and τQ = 0 In

terms of these quantities the kth interface contribution in (C18) is

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

parttuh]]

minus τu2

[[

parttuh]][[

Qh

]]

(C22)

where we have suppressed the k dependence of the right-hand side Now consider the term[[

parttuh]]

Because parttuh and Qh+radicavh are both polynomials of degree N Eq (C10a) implies

29

τu

τ Q

minus1 0 1minus1

minus05

0

05

1

τu

minus1 0 1minus1

minus05

0

05

1

FIG 6 Stable evolutions for the model system For fixed τv = 10minus6 and τv = 1 +radic2

respectively the left and right plots depict stable choices (determined empirically) of τu and τQfor the linear model system (C1) The stable regions are colored black but the jagged edges result

from the discretization of the (τu τQ)-plane

the vector equation parttu = Q +radicav that is pointwise equivalence on the nodal points of

Dk which in turn implies

[[

parttuh]]

=[[

Qh+radicavh]]

Upon substituting this identity into thelast equation we arrive at an expression which features only vh and Qh

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

Qh +radicavh]]

minus τu2

[[

Qh +radicavh]][[

Qh

]]

(C23)

The identities vh[[

vh]]

= 12

[[

v2h]]

and[[

Qh +radicavh]]

=[[

Qh

]]

+radica[[

vh]]

then simplify(C23) to

(∆Lh minus∆R

h )∣

∣

Ik+12 = minusτv2

[[

vh]]2 minus

radica(τu + τQ)

2

[[

Qh

]][[

vh]]

minus τu2

[[

Qh

]]2 (C24)

The role of a penalty parameter is now clear Positive values of τv penalize jumps in vhthrough a negative contribution to the energy Likewise positive values of τu penalizejumps in Qh through a negative contribution to the energy However because the sign of[[

Qh

]][[

vh]]

can be positive or negative only the choice τQ = minusτu yields an expression for(∆L

h minus ∆Rh )|Ik+12 which is manifestly negative for τu ge 0 and τv ge 0 A simple estimate

based on Youngrsquos inequality with ε (that is 2αβ le εminus1α2 + εβ2 where α β ge 0 and ε gt 0)shows that for τQ = 0 the choice τv ge aτu4 also yields stability

Figure 6 depicts certain choices of stable penalty parameters for the linear model systemevolved to tfinal = 1000 (with a = 2 N = 10 and ∆t ≃ 00553) as determined empiricallywith simulations similar to those described in Sec IVA The left plot corresponds to asmall τv = 10minus6 for which the choice τu = 1 τQ = 0 is not stable as expected from thetheoretical analysis However the right plot corresponds to τv = 1 +

radica for which τu = 1

τQ = 0 is stable Motivated by the numerical flux choices (3842) used for the BSSN system

30

(4) we have (as mentioned above) set τu = 1 τv = 1 +radica and τQ = 0 in simulations of

the nonlinear model (53) For the nonlinear model system (53) the theoretically motivatedchoice τQ = minusτu also yields numerically stable evolutions when τu ge 0 and τv ge 0

For the nonlinear systems (4) and (53) we do not attempt a formal stability proofNevertheless the results of this appendix have served as a guide for our choices of penaltyparameters For the BSSN system (4) u v and Q are block indices [cf Eq (6)] Similarto the model problem we have penalized Q with τu = 1 with τv chosen large enough toheuristically overcome the cross-terms of indefinite size that arise from τQ = 0 (we interpretequations like τu = 1 componentwise) An analogous choice ldquoτQ = minusτurdquo for the BSSNsystem might be possible but would be considerably more complicated Indeed such achoice likely entails a matrix of penalty parameters but we do not give the details here

[1] F Pretorius Evolution of Binary Black-Hole Spacetimes Phys Rev Lett 95 121101 (2005)

(4 pages)

[2] F Pretorius Simulation of binary black hole spacetimes with a harmonic evolution scheme

Class Quantum Grav 23 S529-S552 (2006)

[3] M Campanelli C O Lousto P Marronetti and Y Zlochower Accurate Evolutions of Or-

biting Black-Hole Binaries without Excision Phys Rev Lett 96 111101 (2006) (4 pages)

[4] J G Baker J Centrella D-I Choi M Koppitz and J van Meter Gravitational-Wave

Extraction from an Inspiraling Configuration of Merging Black Holes Phys Rev Lett 96

111102 (2006) (4 pages)

[5] M Campanelli C O Lousto and Y Zlochower Last orbit of binary black holes Phys Rev D

73 061501(R) (2006) (5 pages)

[6] F Herrmann I Hinder D Shoemaker and P Laguna Unequal mass binary black hole plunges

and gravitational recoil Class Quantum Grav 24 S33-S42 (2007)

[7] P Diener F Herrmann D Pollney E Schnetter E Seidel R Takahashi J Thornburg and

J Ventrella Accurate Evolution of Orbiting Binary Black Holes Phys Rev Lett 96 121101

(2006) (4 pages)

[8] M A Scheel H P Pfeiffer L Lindblom L E Kidder O Rinne and S A Teukolsky Solving

Einsteinrsquos Equations With Dual Coordinate Frames Phys Rev D 74 104006 (2006)

[9] U Sperhake Binary black-hole evolutions of excision and puncture data Phys Rev D 76

104015 (2007) (20 pages)

[10] B Brugmann J A Gonzalez M Hannam S Husa U Sperhake and W Tichy Calibration

of moving puncture simulations Phys Rev D 77 024027 (2008) (25 pages)

[11] P Marronetti W Tichy B Brugmann J Gonzalez M Hannam S Husa and U Sperhake

Binary black holes on a budget simulations using workstations Class Quantum Grav 24

S43-S58 (2007)

[12] Z B Etienne J A Faber Y T Liu S L Shapiro and T W Baumgarte Filling the holes

Evolving excised binary black hole initial data with puncture techniques Phys Rev D 76

101503(R) (2007) (5 pages)

[13] B Szilagyi D Pollney L Rezzolla J Thornburg and J Winicour An explicit harmonic code

for black-hole evolution using excision Class Quantum Grav 24 S275-S293 (2007)

[14] L Boyle M Kesden and S Nissanke Binary-Black-Hole Merger Symmetry and the Spin

Expansion Phys Rev Lett 100 151101 (2008) (4 pages)

31

[15] M A Scheel M Boyle T Chu L E Kidder K D Matthews and H P Pfeiffer High-

accuracy waveforms for binary black hole inspiral merger and ringdown Phys Rev D 79

024003 (2009) (14 pages)

[16] M Hannam Status of black-hole-binary simulations for gravitational-wave detection

Class Quantum Grav 26 114001 (2009) (21 pages)

[17] I Hinder The current status of binary black hole simulations in numerical relativity

Class Quantum Grav 27 114004 (2010) (19 pages)

[18] B P Abbott et al LIGO The Laser Interferometer Gravitational-Wave Observatory

Rep Prog Phys 72 076901 (2009) (25 pages)

[19] F Acernese et al The status of VIRGO Class Quantum Grav 23 S63-S69 (2006)

[20] F Acernese et al Virgo status Class Quantum Grav 25 184001 (2008) (9 pages)

[21] R Arnowitt S Deser and C W Misner The Dynamics of General Relativity in Gravitation

an introduction to current research edited by L Witten (Wiley New York 1962) Chapter 7

227-265 Also available as gr-qc0405109

[22] J W York Jr Kinematics and Dynamics of General Relativity in Sources of Gravitational

Radiation edited by L L Smarr (Cambridge University Press Cambridge 1979)

[23] L E Kidder M A Scheel and S A Teukolsky Extending the lifetime of 3D black hole

computations with a new hyperbolic system of evolution equations Phys Rev D 64 064017

(2001) (13 pages)

[24] G Calabrese J Pullin O Sarbach and M Tiglio Convergence and stability in numerical

relativity Phys Rev D 66 041501(R) (2002) (4 pages)

[25] S Frittelli and O Reula On the Newtonian Limit of General Relativity Commun Math

Phys 166 221-235 (1994)

[26] M Shibata and T Nakamura Evolution of three-dimensional gravitational waves Harmonic

slicing case Phys Rev D 52 5428-5444 (1995)

[27] Y Choquet-Bruhat and J W York Geometrical well posed systems for the Einstein equations

C R Acad Sci Paris t 321 Serie I 1089-1095 (1995)

[28] A Abrahams A Anderson Y Choquet-Bruhat and J W York Jr Einstein and Yang-Mills

Theories in Hyperbolic Form without Gauge Fixing Phys Rev Lett 75 3377-3381 (1995)

[29] C Bona J Masso E Seidel and J Stela New Formalism for Numerical Relativity

Phys Rev Lett 75 600-603 (1995)

[30] M H P M van Putten and D M Eardley Nonlinear wave equations for relativity

Phys Rev D 53 3056-3063 (1996)

[31] S Frittelli and O A Reula First-Order Symmetric Hyperbolic Einstein Equations with Ar-

bitrary Fixed Gauge Phys Rev Lett 76 4667-4670 (1996)

[32] H Friedrich Hyperbolic reductions for Einsteinrsquos equations Class Quantum Grav 13 1451-

1469 (1996)

[33] F B Estabrook R S Robinson H D Wahlquist Hyperbolic equations for vacuum gravity

using special orthonormal frames Class Quantum Grav 14 1237-1247 (1997)

[34] M S Iriondo E O Leguizamon and O A Reula Einsteinrsquos Equations in Ashtekarrsquos Vari-

ables Constitute a Symmetric Hyperbolic System Phys Rev Lett 79 4732-4735 (1997)

[35] A Anderson Y Choquet-Bruhat and J W York Einstein-Bianchi hyperbolic system for

general relativity Topol Meth Nonlinear Anal 10 353-373 (1997)

[36] T W Baumgarte and S L Shapiro Numerical integration of Einsteinrsquos field equations

Phys Rev D 59 024007 (1998) (7 pages)

[37] M A G Bonilla Symmetric hyperbolic systems for Bianchi equations Class Quantum

10

Evidently the polynomial χkh interpolates χ at rkj To define the nodes rkj consider the

mapping from the unit interval [minus1 1] to Dk

rk(u) = ak + 12(1 + u)(bk minus ak) (22)

and the N+1 Legendre-Gauss-Lobatto (LGL) nodes uj The uj are the roots of the equation

(1minus u2)P primeN(u) = 0 (23)

where PN(u) is the Nth degree Legendre polynomial and the physical nodes are simplyrkj = rk(uj) In vector notation the approximation (20) takes the form

χkh(t r) = χk(t)Tℓk(r) (24)

in terms of the column vectors

χk(t) =[

χ(t rk0) middot middot middot χ(t rkN)]T ℓk(r) =

[

ℓk0(r) middot middot middot ℓkN(r)]T (25)

On each open interval (ak bk) sub Dk and for each component of the equations in (13) we

define local residuals measuring the extent to which our approximations satisfy the originalcontinuum system Dropping the subdomain label k on the polynomials and focusing onthe K equation as a representative example the local residual corresponding to (4h) is

minus(RK)kh equivminus parttKh + (βrK prime)h minus

(

χQprimeα

grr

)

h

+

(

χQgrrQα

2g2rr

)

h

minus(

χQgθθQα

grrgθθ

)

h

+

(

QαQχ

2grr

)

h

+

(

3αA2rr

2g2rr

)

h

+

(

1

3αK2

)

h

(26)

Here for example the expressions 2

(βrK prime)h = βrhK

primeh

(

QαQχ

2grr

)

h

=QαhQχh

2grrh (27)

We similarly construct the remaining eight residuals eg (Rgrr)h and (RΓr)h as well as fiveresiduals corresponding to (13b) For example one of these remaining five is

(RQα)kh equiv minusQαh + αprime

h (28)

Let the kth inner product be defined as

(

u v)

Dk equivint bk

akdru(r)v(r) (29)

and consider the expression (ℓkj (RK)kh)Dk We call the requirement that this inner product

vanish forallj the kth Galerkin condition For each component of the system and for each k

2 At this stage the first expression is generically a polynomial of degree 2N minus 1 and the latter is not a

polynomial The conventions adopted in Eq (27) prove useful while working with the residual However

later on in Sec III C to obtain the final form (47) of the numerical approximation corresponding to (26)

we will replace nonlinear terms with degree-N polynomials

11

there is a corresponding Galerkin condition in total 9kmax(N + 1) equations for (13a) and5kmax(N+1) for (13b) Enforcement of the Galerkin conditions on each D

k will not recover ameaningful global solution since they provide no mechanism for coupling the local solutionson the different intervals Borrowing from the finite volume toolbox we achieve couplingthrough integration by parts on r and introduction of the numerical flux f lowast at the interfacebetween subdomains

In (26) we only need to consider (βrK prime)h and (χQprimeαgrr)h as the other terms comprise a

component of the source vector Sh Using integration by parts we write

(

ℓkj (βrK prime)h

)

Dk =minusint bk

akdr[

(

ℓkjβrh

)primeKh

]

+ (βrhKh) ℓ

kj

∣

∣

∣

bk

ak (30a)

(

ℓkj (χQprimeαgrr)h

)

Dk =minusint bk

akdr

[(

ℓkjχh

grrh

)prime

Qαh

]

+

(

χhQαh

grrh

)

ℓkj

∣

∣

∣

bk

ak (30b)

In these formulas we have retained the domain index k on ℓkj while continuing to suppress iton Kh grrh etc Moreover we have suppressed the r-dependence in all terms on the right-hand side Addition of these formulas along with the definition fKh = minus(βrK)h+(χQαgrr)hgives

(

ℓkj (βrK prime)h minus (χQprime

αgrr)h)

Dk = minusint bk

akdr

[

(

ℓkjβrh

)primeKh minus

(

ℓkjχh

grrh

)primeQαh

]

minus fKhℓkj

∣

∣

∣

bk

ak (31)

In lieu of (31) we will instead work with the replacement

(

ℓkj (βrK prime)h minus (χQprime

αgrr)h)

Dk rarr minusint bk

akdr

[

(

ℓkjβrh

)primeKh minus

(

ℓkjχh

grrh

)primeQαh

]

minus f lowastKℓ

kj

∣

∣

∣

bk

ak (32)

This replacement features a component f lowastK of the numerical flux rather than a component

fKh of the boundary flux The numerical flux is determined by (as yet not chosen) functions3

f lowast = f lowast(W+Wminus) (33)

where for example Wminus is an interior boundary value [either W kh (t a

k) or W kh (t b

k)] of theapproximation defined on D

k andW+ is an exterior boundary value [either W kminus1h (t bkminus1) or

W k+1h (t ak+1)] of the approximation defined on either Dkminus1 or Dk+1 We discuss our choice

of numerical flux in the next subsection We now employ additional integration by parts towrite the above replacement as

(

ℓkj (βrK prime)h minus (χQprime

αgrr)h)

Dk rarrint bk

akdrℓkj

(

βrK prime minus χQprimeα

grr

)

h

+ (fKh minus f lowastK) ℓ

kj

∣

∣

∣

bk

ak (34)

Rather than the exact kth Galerkin condition(

ℓkj (RK)kh

)

Dk = 0 forallj for the K component

of (13) on Dk we will instead strive to enforce

(

ℓkj (RK)kh

)

Dk = (fKh minus f lowastK) ℓ

kj

∣

∣

∣

bk

ak forallj (35)

3 In the context of the dG method here + and minus denote ldquoexteriorrdquo and ldquointeriorrdquo and have no relation to

the plusmn using to denote the characteristic fields and speeds in Table I For characteristic fields and speeds

+ and minus mean ldquoright-movingrdquo and ldquoleft-movingrdquo

12

although our treatment of nonlinear terms will lead to a slight modification of these equations(we return to this issue shortly) The other components of (13a) are treated similarly asare the components of (13b) Recall that for example Qα = αprime Formally using the samedG method to solve for Qα we arrive at the replacement

(

ℓkj (RQα)kh

)

Dk rarrint bk

akdrℓkj (minusQαh + αprime

h)minus (αh minus αlowast) ℓkj

∣

∣

∣

bk

ak (36)

which again features a component αlowast of the numerical flux The auxiliary variables areconstructed and used at each stage of temporal integration We then have

(

ℓkj (RQα)kh

)

Dk = (αh minus αlowast) ℓkj

∣

∣

∣

bk

ak forallj (37)

as the corresponding enforced kth Galerkin condition

B Numerical Flux

To further complete our dG scheme we must specify functional forms for the compo-nents of the numerical flux introduced in the previous section We distinguish between thephysical fluxes (components of f) and the auxiliary fluxes (components of u) arising fromthe definition of the auxiliary variables These choices are not independent as the resultingscheme must be stable and consistent Our choice follows [84] which considered diffusionproblems Additional analysis of this flux choice appears in [66 85]

Let us first consider the numerical fluxes corresponding to the physical fluxes and of theform (33) The numerical flux vector is a function of the system and auxiliary variablesinterior and exterior to a subdomain A common choice for f lowast is

f lowast = fh+τ

2

[[

vh]]

K-component of f lowast f lowastK = fKh+

τ

2

[[

Kh

]]

(38)

where as an example we have also shown the component of f lowast corresponding to the analysisabove Respectively the average and jump across the interface are

fh =1

2

(

f+ + fminus) [[

vh]]

= nminusvminus + n+v+ (39)

Here τ is a position dependent penalty parameter (fixed below) and nminus(n+) is the localoutward pointing normal to the interior (exterior) subdomain The role of τ is to ldquopenalizerdquo(i e yield a negative contribution to the L2 energy norm) jumps across an interface Anappropriate choice of τ will ensure stability and we now provide some motivation for thechoice (41) of τ we make below

Were we treating the fully first-order system (10) the local Lax-Friedrichs flux wouldoften be a preferred choice due to its simplicity [66] In this case the constant ω in thenumerical flux formula Flowast = Fh + 1

2ω[[

Wh

]]

obeys ω ge max∣

∣micro(nablaWF(W ))∣

∣ HereF(W ) = A(u)W the notation micro(middot) indicates the spectral radius of the matrix within andthe max is taken over interior Wminus and exterior W+ states Motivated by (9) we adopt asimilar but simpler prescription substituting the field gradient

nablaWvQA(u)WvQ = A(u) (40)

13

for nablaWF(W ) Precisely we assume the scaling

τ(bk) = τ(ak+1) = τk+12 equiv C middotmax∣

∣micro(

A(u))∣

∣ (41)

where C = O(1) is a constant chosen for stability Larger values of C will result in schemeswith better stability properties whereas too large a value will impact the CFL conditionAt the interface point Ik+12 equiv D

k cap Dk+1 the vector uh has two representations uminus at bk

and u+ at ak+1 The max in (41) is taken over the corresponding two sets of field speedsMore precisely the speeds in Table I are computed for both uminus and u+ and the maximumtaken over all resulting speeds For the auxiliary variables a penalized central flux is usedThe definition with one representative component is

ulowast = uh minus1

2

[[

uh]]

α-component of ulowast αlowast = αh minus1

2

[[

αh

]]

(42)

with similar expressions for the remaining componentsWe stress the following point Since the interior coupling between subdomains is achieved

through the numerical flux forms (41) and (42) the inverse transformation (17) expressingthe fundamental fields in terms of the characteristic fields is not required to achieve thiscoupling On the other hand imposition of physical boundary conditions may still rely on(17) since this transformation allows one to fix only incoming characteristic modes

C Matrix form of the semi-discrete equations

Let us introduce the kth mass and stiffness matrices

Mkij =

int bk

akdrℓki (r)ℓ

kj (r) Sk

ij =

int bk

akdrℓki (r)ℓ

kjprime(r) (43)

These matrices belong to Dk and the corresponding matrices defined on the reference interval[minus1 1] are

Mij =

int 1

minus1

duℓi(u)ℓj(u) Sij =

int 1

minus1

duℓi(u)ℓprimej(u) (44)

where ℓj(u) is the jth Lagrange polynomial determined by the LGL nodes uj on [minus1 1]These matrices are related by Mk

ij =12(bk minus ak)Mij and S

kij = Sij whence only the reference

matrices require computation and storageWe will use the matrices Mk and Sk in obtaining an ODE system from (26) and (35)

Towards this end we first approximate the nonlinear terms (products and quotients) in(26) by degree-N interpolating polynomials Such approximations are achieved throughpointwise representations For example (QαQχgrr)h appears in (26) and is expressed inthe following way [cf footnote 2]

(

QαQχ

grr

)

h

(t r) =

(

QαhQχh

grrh

)

(t r) rarrNsum

j=0

Qαh(t rkj )Qχh(t r

kj )

grrh(t rkj )

ℓkj (r) (45)

Note that the expressions on the right and left are not equivalent due to aliasing error [65]Our vector notation for this replacement will be

(

QαQχ

grr

)

h

(t r) rarr(

QαQχ

grr

)

(t)Tℓk(r) (46)

14

Operations among bold variables are always performed pointwise Making similar replace-ments for all terms in (26) and then carrying out the integrations in (35) which bring inMk and Sk we arrive at

parttK = βrDK minus χDQα

grr

+1

2

χQgrrQα

g2rr

minusχQgθθ

Qα

grrgθθ

+1

2

QαQχ

grr

+3

2

αA2rr

g2rr

+1

3αK2 +Mminus1ℓk (fKh minus f lowast

K)∣

∣

∣

bk

ak (47)

where we have again suppressed the superscript k on all terms except ℓk(r) and the sub-script h is dropped on all boldfaced variables As described in [66] the spectral collocationderivative matrix

(Dk)ij =dℓkjdr

∣

∣

∣

∣

∣

r=rki

(48)

can also be expressed as Dk = (Mk)minus1Sk which appears above Eight other semi-discreteevolution equations are similarly obtained with nine in total (one for each component ofWuv) Additionally we have

Qα = Dα+Mminus1ℓk(αlowast minusαh)∣

∣

∣

bk

ak (49)

as one of the auxiliary equations with five in total (one for each component of Q = WQQ)

D Filtering

Like other nodal (pseudospectral) methods our scheme may suffer from instabilitiesdriven by aliasing error [65] Filtering is a simple yet robust remedy To filter a solu-tion component such as χ we use the modal (as opposed to nodal) representation of thesolution

χkh(t r) =

Nsum

j=0

χ(t rkj )ℓkj (r) =

Nsum

j=0

χkj (t)Pj(r) (50)

where Pj(r) is the jth Legendre polynomial Let ηj = jN and define the filter function

σ(ηj) =

1 for 0 le ηj le NcN

exp(

minus ǫ(

ηjminusNcN

1minusNcN

)2s)

for NcN le ηj le 1(51)

At each timestep we modify our solution component according to

χkh rarr

(

χkh

)filtered=

Nsum

j=0

σ(ηj)χkj (t)Pj(r) (52)

Evidently the modification only affects the top N minus Nc modes and is sufficient to controlthe type of weak instability driven by aliasing [66] The numerical parameters Nc and ǫ areproblem dependent For our simulations we have taken ǫ ≃ minuslog(εmach) = 36 where εmach

is machine accuracy in double precision

15

E Model system

To better illustrate the basic properties of our method we consider a toy model Namelythe following spatially second-order system

parttu = uprime + av minus u3 + g(t x) (53a)

parttv = uprimeprime + vprime minus (u+ v)(uprime)2 + v2u2 + h(t x) (53b)

where a ge 1 is constant and g and h are analytic source terms to be specified In contrastto (6) here u v and Q = uprime are scalars rather than vectors System (53) admits a first-order reduction in which uprime is defined as an extra variable Since this first-order reductionis strongly hyperbolic the spatially second-order system (53) is also strongly hyperbolic byone of the definitions considered in [75] The characteristic fields Xplusmn and speeds microplusmn are

X+ =radicav minus uprime micro+ =

radicaminus 1 Xminus =

radicav + uprime microminus = minus(

radica+ 1) (54)

To construct a local dG scheme for this system we first rewrite it as

parttu = Q+ av minus u3 + g(t x) (55a)

parttv = Qprime + vprime minus (u+ v)Q2 + v2u2 + h(t x) (55b)

Q = uprime (55c)

Evidently f = minus(Q + v) is the v-component of the physical flux vector

F (vQ) equiv(

Fu

Fv

)

=

(

0

f

)

(56)

Note that F has the same structure as (u v)T Borrowing from the presentation for theBSSN system we write the analogous semidiscrete scheme on each subdomain D

k for themodel system

parttu = Q+ av minus u3 + g(t) (57a)

parttv = DQ+Dv minus (u+ v)Q2 + v2u2 + h(t) +Mminus1ℓk(fh minus f lowast)∣

∣

∣

bk

ak (57b)

Q = Du+Mminus1ℓk(ulowast minus uh)∣

∣

∣

bk

ak (57c)

Here we have suppressed the subinterval label k from all variables except for the vectorℓk of Lagrange polynomial values Moreover following the guidelines discussed above thenumerical fluxes are given by

f lowast = fh+1 +

radica

2

[[

vh]]

ulowast = uh minus1

2

[[

uh]]

(58)

Appendix C analyzes the stability of our scheme for a more general numerical flux choiceas applied to (53) with the nonlinear and source terms dropped

IV RESULTS FROM NUMERICAL SIMULATIONS

This section presents results found by numerically solving both the model system (53)and BSSN system (4) with the dG scheme presented in Sec III

16

10minus10

10minus6

10minus2

∆uinfin

10minus10

10minus6

10minus2

∆vinfin

0 500 1000 1500 2000 2500 3000 3500 400010

minus10

10minus6

10minus2

∆(partxu)infin

t

N = 3N = 6N = 9N = 12

FIG 1 Spectral convergence of fields for model PDE Respectively for N = 3 6 9 12

a timestep of ∆t = 00578 00178 00084 00049 has been chosen for stability and accuracy In the

title headings for example ∆u equiv unumer minus uexact

A Simulations of the model system

The semi-discrete scheme (57) has been integrated with the classical fourth-order Runge-Kutta method When integrating this system we have first constructed Q at each Runge-Kutta stage and then substituted into the evolution equations (57ab) for u and v Theproblem has been solved on a computational domain [0 4π] comprised of two subdomainswith a timestep chosen small enough for stability The initial data has been taken from thefollowing exact solution to (53)

uprimeexact(t x) =1

2

[

sin(xminus microminust)minus sin(xminus micro+t)]

(59a)

vexact(t x) =1

2radica

[

sin(xminus microminust) + sin(xminus micro+t)]

(59b)

g(t x) = u3exact (59c)

h(t x) = (uexact + vexact)(uprimeexact)

2 minus v2exactu2exact (59d)

where the speeds microplusmn are found in (54) Specification of the boundary condition at a physicalendpoint amounts to choosing the external state for at the endpoint We have considered

17

8 16 32 64 128

10minus3

10minus2

10minus1

N

∆t

max stable ∆t8N2

FIG 2 Scaling of maximum stable ∆t with N for model PDE

two possibilities (i) the analytic state (Q+ v+) = (Qexact vexact) and (ii) an upwind stateFor example at x = 4π the upwind state is4

Q+ = Qupwind =1

2

[

(Xminus)exact minus (X+)numer

]

v+ = vupwind =1

2radica

[

(Xminus)exact + (X+)numer

]

(60)Either choice of (Q+ v+) leads to similar results and the plots here correspond to theanalytic state Figure 1 clearly shows spectral convergence with increasing polynomial orderN across all fields for the case a = 2 Other values of a including a = 1 for which X+

is a static characteristic field have also been considered with similar results Appendix Cdemonstrates that our proposed scheme for the system (57) with nonlinear and source termsdropped is stable in a semi-discrete sense Nevertheless the fully discrete scheme obtainedvia temporal discretization by the fourth-order Runge-Kutta method is still subject to thestandard absolute stability requirement Namely if microh is any eigenvalue corresponding tothe (linearized) discrete spatial operator then a necessary condition for stability is thatmicroh∆t lies in absolute stability region for Runge-Kutta 4 We here show empirically that theassociated timestep restriction scales like Nminus2 ie ∆t = O(Nminus2) for stability We note thatsuch scaling is welcome in light of the second-order spatial operators which appear in thesystem and suggest a possible worse scaling like Nminus4 Fig 2 plots the maximum stabletimestep for a range of N demonstrating the Nminus2 scaling in line with behavior known fromanalysis of first-order systems [66] This scaling also holds for the BSSN system

4 We remind the reader that unfortunately the plusmn on Xplusmn means something different than the plusmn indicating

exteriorinterior dG states [cf footnote 3]

18

B Simulations of the BSSN system

This subsection documents results for simulations of the unit-mass-parameter (M = 1)Schwarzschild solution (B9) expressed in terms of ingoing Kerr-Schild coordinates Sincethe solution is stationary temporal integration of the semi-discrete scheme has been carriedout with the forward Euler method The r-coordinate domain [04 34] has been split into3 equally spaced subdomains and we have set η = 10 λ = 01 and C = 2 [cf Eq (41)]For all simulations ∆t has been chosen for stability With the chosen λ the inner physicalboundary rmin = 04 is an excision surface At each timestep we have applied an (order2s = 20) exponential filter on the top two-thirds of the modal coefficient set for all fieldsexcept for grr and gθθ For stability we have empirically observed that grr and gθθ must not

be filtered A detailed understanding of this is still lackingIssues related to physical boundary conditions are similar to the one encountered in

Sec IVA for the model problem Similar to before we have retained Eqs (3842) as thechoice of numerical flux even at the endpoints Therefore at an endpoint the specificationof the boundary condition amounts to the choice W+ of external state We have typicallychosen the inner boundary of the radial domain as an excision boundary and in this caseW+ = Wminus is enforced at the inner physical boundary At the outer physical boundaryfor W+ we have again considered two choices (i) Wexact and (ii) Wupwind To enforcechoice (ii) the inverse transformation (17) must be used with incoming characteristic fieldsfixed to their exact values similar to (60) We have tried various versions of choice (ii)and in all cases the resulting simulations have been unstable We therefore present resultscorresponding to choice (i) Although the choice of an analytical external state Wexact

at the outer boundary is stable for our problem such a boundary condition is unlikelyto generalize to more complicated scenarios involving dynamical fields Indeed the issueof outer boundary conditions for the BSSN system is an active area of research with aproper treatment requiring fixation of incoming radiation control of the constraints andspecification of gauge (see Ref [86] for a recent analysis)

For BSSN simulations our main diagnostic is to monitor the Hamiltonian momentumand conformal connection constraints Figure 3 depicts long-time histories of constraintviolations whereas Figs 4 and 5 depict long-time error histories for the individual BSSNfield components From the middle plot in Fig 5 we infer that up to the indicated numericalerror the factor g sin4 θ = grr(gθθ)

2 remains at its initial fixed profile r4 throughout theevolution These figures indicate that the proposed scheme is stable for long times andexhibits spectral converge with increased polynomial order N Similar results are recoveredfromM = 0 Minkowski initial data The stability documented in these plots does not appearto rely on inordinate parameter tuning For example with the fixed parameters describedabove we obtain similar plots if we individually vary (i) rmin over 0325 035 04 0475(values still corresponding to an excision surface for the given choice of λ) (ii) η over1 3 7 10 (iii) s over 8 9 10 With the polynomial order N ranging over 23 26 29 31both stability and qualitatively similar exponential convergence is achieved with a singlesubdomain Likewise adoption of a larger coordinate domain with more subdomains doesnot impact our results Finally we have considered the addition of random noise to all fieldcomponents at the initial time Precisely with the system component χ as an example wehave set

χ(0) 7rarr χ(0) + δχ(0) (61)

where each component (nodal value) of δχ(0) is 10minus5 times a random variable drawn from a

19

10minus9

10minus7

10minus5

10minus3

Hinfin

10minus9

10minus7

10minus5

10minus3

Mrinfin

0 2000 4000 6000 8000 10000

10minus9

10minus7

10minus5

10minus3

Grinfin

t

N = 11N = 14N = 17N = 20

FIG 3 Spectral convergence of constraint violations for M = 1 Kerr-Schild ini-

tial data Respectively for N = 11 14 17 19 a timestep of ∆t ≃ 00041 00026 00018 00013

has been chosen for stability and accuracy

standard normal distribution Such perturbed initial data also gives rise to stable evolutions

V CONCLUSION

We have introduced a discontinuous Galerkin method for solving the spherically reducedBSSN system with second-order spatial operators Our scheme shares similarities withother discontinuous Galerkin methods that use local auxiliary variables to handle high-orderspatial derivatives [66 77ndash79 82 84 85] and which have typically been applied to eitherelliptic parabolic or mixed type problems The key ingredient of a stable dG scheme is anappropriate choice of numerical flux and our particular choice has been motivated by theanalysis presented in Appendix C When used to evolve the Schwarzschild solution in Kerr-Schild coordinates our numerical implementation of the BSSN system (4) is robustly stableand converges to the analytic solution exponentially with increased polynomial order Byapproximating the spatially second-order form of the BSSN system we have not introducedextra fields which are evolved Evolved auxiliary fields result in new constraints which mayspoil stability Our main goal has been stable evolution of the spherically reduced BSSNsystem as a first step towards understanding how a discontinuous Galerkin method might be

20

10minus9

10minus7

10minus5

∆Arrinfin

10minus9

10minus7

10minus5

∆Kinfin

0 2000 4000 6000 8000 10000

10minus9

10minus7

10minus5

max(∆Brinfin ∆Γrinfin)

t

N = 11N = 14N = 17N = 20

FIG 4 Spectral convergence of solution for M = 1 Kerr-Schild initial data

Timestep choices are described in the caption for Fig 3 In the title headings for example

∆Arr equiv (Arr)numer minus (Arr)exact

applied to the full BSSN system Towards that goal we now discuss treatment of singularitiesand generalization of the described dG method to higher space dimension

To deal with the fixed Schwarzschild singularity we have used excision which is easy inthe context of the spherically reduced BSSN system However excision for the binary blackhole problem in full general relativity requires attention to the technical challenge of hori-zon tracking State-of-the-art BSSN codes avoid such complication relying instead on themoving-puncture technique While the moving-puncture technique does involve mild centralsingularities it may still prove amenable to spectral methods Indeed spectral methods fornon-smooth problems is well-developed in both theory and for complex applications Sincethe moving-puncture technique can be performed in spherical symmetry [60] a first-steptoward a spectral moving-puncture code would be to implement a moving puncture withthe nodal dG method described here Such an implementation may adopt Legendre-Gauss-Radau nodes on the innermost subdomain thereby ensuring that the physical singularitydoes not lie on a nodal point (in much the same way finite difference codes use a staggeredgrid) Beyond traditional excision and moving punctures one might also retain a smoothproblem for which spectral methods are especially well-suited via the turducken approach[87 88] to singularities

21

10minus11

10minus9

10minus7

10minus5

∆χinfin

10minus11

10minus9

10minus7

10minus5

max(∆grrinfin ∆gθθinfin)

0 2000 4000 6000 8000 10000

10minus11

10minus9

10minus7

10minus5

max(∆αinfin ∆βrinfin)

t

N = 11N = 14N = 17N = 20

FIG 5 Spectral convergence of solution violations for M = 1 Kerr-Schild initial

data See the caption of Fig 4 for details

Discontinuous Galerkin methods for hyperbolic problems in two and three space dimen-sions are well-developed A generalization of the method described here to three-dimensionsand the full BSSN system would likely rely on an unstructured mesh Appropriate local poly-nomial expansions for the subelements are well-understood as are choices for the numericalfluxes which would now live on two-dimensional faces rather than single points Whether ornot it would ultimately prove successful generalization of our dG method to a higher dimen-sion would rely on an established conceptual framework Further computational advancesof relevance to a generalization of our dG method to the full BSSN system (possibly includ-ing matter) may include mesh hp-adaptivity local timestepping shock capturing and slopelimiting techniques [66] Moreover recent work [89] indicates that enhanced performancewould be expected were our scheme implemented on graphics processor units

VI ACKNOWLEDGMENTS

We thank Nick Taylor for discussions about treating second-order operators with spectralmethods Benjamin Stamm for helping polish up a few parts of Appendix C Khosro Shah-bazi for discussions on LDG and IP methods We also acknowledge helpful conversationswith David Brown and Manuel Tiglio about the BSSN system and previous work on its nu-

22

merical implementation We gratefully acknowledge funding through grants DMS 0554377and DARPAAFOSR FA9550-05-1-0108 to Brown University and NSF grant PHY 0855678to the University of New Mexico

Appendix A Hyperbolicity of the first-order system

This appendix analyzes the matrix A(u) appearing in (10) in order to construct thecharacteristic fields (14) In matrix form the sector (8) of the principal part of (10) reads asfollows

partt

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

=

βr 0 minus 4λα3grr

0 0 λβr

6(grr)2λβr

3gθθgrr0 4λ

3grr

0 βr 0 23grrαχ

13α minus αχ

3grr

αχ3gθθ

minus23χ 0

0 0 βr 0 0 0 0 minus χgrr

0

0 0 minus 4α3grr

βr 0 βr

6(grr)2βr

3gθθgrr0 4

3grr

0 0 23αχ 0 βr minus βrχ

3grrminus2βrχ

3gθθ0 minus2

3χ

0 minus2α 0 0 0 23βr minus2grrβr

3gθθ0 4

3grr

0 gθθαgrr

0 0 0 minusgθθβr

3grr13βr 0 minus2

3gθθ

0 0 minus2α 0 0 0 0 βr 0

34

0 0 0 0 0 0 0 βr

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

prime

(A1)which defines the matrix A(u) appearing in (9) and so also the matrix A(u) in (10) Notethat in the last equation the matrix within the square brackets is minusA(u) For certainconfigurations of u and λ the system (10) is strongly hyperbolic [76] that is A(u) hasa complete set of eigenvectors and real eigenvalues Indeed five eigenpairs of A(u) aretrivially recovered upon inspection of A(u)rsquos leading 5times5 diagonal block These correspondto eigenvalue 0 and the left eigenspace ξj = eTj 1 le j le 5 where ej are the canonical

basis vectors Since each component of u arises as eTj W each is also a characteristic field

The remaining nine eigenpairs are determined by A(u) The eigenvalues of A(u) are

micro1 = 0 micro23 = minusβr microplusmn4 = minusβr plusmn

radic

2αχ

grr microplusmn

5 = minusβr plusmn α

radic

χ

grr microplusmn

6 = minusβr plusmnradic

λ

grr

(A2)

23

and the corresponding left eigenvectors are

x1 = (0 0 0 0 0 gθθ 2grr 0 0) (A3a)

x2 =

(

0 0 0 grr2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3b)

x3 =

(

grrλ 0 0 0

2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3c)

xplusmn4 =

(

0 0plusmnradic

2αgrrχ

0 0 0 0 1 0

)

(A3d)

xplusmn5 =

(

0∓ 3radicgrrχ

plusmn2

radic

grrχ 2grr

1

χminus 1

grr1

gθθ 0 0

)

(A3e)

xplusmn6 =

(

minus3

4

grrλ 0plusmn α

radicλgrr

(2αχminus λ) 0 0minus βr

8(βrgrr ∓radicλgrr)

minus βrgrr

4gθθ(βrgrr ∓radicλgrr)

αχ

(2αχminus λ)plusmnradic

grrλ

)

(A3f)

where for example xplusmn5 A(u) = microplusmn5 x

plusmn5 Assuming that grr gθθ χ and α are everywhere strictly

positive the eigenvalues are real and the eigenvectors are linearly independent providedthat (15) holds These eigenvectors are easily extended to eigenvectors of A(u) e g asxplusmn6 rarr (01times5 x

plusmn6 ) Then for example the characteristic field

Xplusmn6 equiv (01times5 x

plusmn6 )W = xplusmn6WvQ (A4)

and similarly Xplusmnj = xplusmnj WvQ for j = 4 5 and Xk = xkWvQ for k = 1 2 3 The characteristic

speeds for these fields are microk and microplusmnj With this convention the speeds listed in Table I

correspond to the Xk and Xplusmnj in (14)

Appendix B Schwarzschild solution in conformal Kerr-Schild coordinates

In Kerr-Schild coordinates here the system directly related to incoming Eddington-Finkelstein null coordinates the line element for the Schwarzschild solution reads

ds2 = minusα2dt2 + (1 + 2MR)(dR + βRdt)2 +R2dθ2 +R2 sin2 θdφ2 (B1)

where R is the area radius α = (1 + 2MR)minus12 is the lapse and βR = 2M(R + 2M) isthe shift vector The physical spatial metric gab is the spatial part of this line element

To define the corresponding solution to the BSSN system we use equation gab = χgab todefine the following relationship between line elements

dr2 + r2(dθ2 + sin2 θdφ2) = χ[(1 + 2MR)dR2 +R2dθ2 +R2 sin2 θdφ2] (B2)

so that

χ

(

1 +2M

R

)(

dR

dr

)2

= 1 χR2 = r2 (B3)

Then we have(

1 +2M

R

)12dR

R=dr

r (B4)

24

with integration yielding

r =R

4

(

1 +

radic

1 +2M

R

)2

e2minus2radic

1+2MR (B5)

where the constant of integration has been chosen so that the R r rarr infin limits are consistentThe second relation in (B3) shows that

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR =rArr χminus4 =2eradic

1+2MRminus1

1 +radic

1 + 2MR (B6)

The extrinsic curvature tensor is specified by the expression forK given in (B9h) the identityK = KR

R + 2Kθθ and

Kθθ =

(

1 +2M

R

)minus122M

R2 (B7)

Since KRR = Kr

r we compute that

Krr = K minus 2Kθ

θ = minus(

1 +2M

R

)minus12(R +M

R + 2M

)

2M

R2 (B8)

Next since Krr = grrKrr = χminus1Kr

r we have Krr = Arr+

13grrK This implies Arr = Kr

r minus 13K

from which we get (B9g) In all we have

α =

(

1 +2M

R

)minus12

(B9a)

βr = βR dr

dR= χ12

(

1 +2M

R

)minus122M

R(B9b)

grr = 1 (B9c)

gθθ = r2 = χR2 (B9d)

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR (B9e)

Br = 0 (B9f)

Arr = minus(

1 +2M

R

)minus124M

3R2

(

2R + 3M

R + 2M

)

(B9g)

K =

(

1 +2M

R

)minus32(

1 +3M

R

)

2M

R2(B9h)

Γr = minus2

r= minus 2

χ12R (B9i)

To differentiate these expressions with respect to r we use the identity

dR

dr= χminus12

(

1 +2M

R

)minus12

(B10)

along with the chain rule

25

Appendix C Stability of the model system

The following stability analysis for the model system (53) has been inspired by [78 79]After dropping all nonlinear source terms the system (53) becomes

parttu = uprime + av (C1a)

parttv = uprimeprime + vprime (C1b)

This section analyzes the stability of (C1) considering both the continuum system itself aswell as its semi-discrete dG approximation The latter analysis offers some insight into theempirically observed stability of our dG scheme for the spherically reduced BSSN equations

1 Analysis for a single interval

Throughout we work with the L2 inner product and norm

(f g)D =

int

D

fg fD =radic

(f f)D (C2)

where D is the spatial coordinate interval (here D may represent a subdomain Dk or the

whole domain Ω) and we have suppressed all integration measures For the continuummodel we will establish the following estimate

uprime(T middot)2D+ av(T middot)2

Dle C(T )

(

uprime(0 middot)2D+ av(0 middot)2

D

)

(C3)

where the time-dependent constant C(T ) is determined solely by the choice of boundaryconditions To show (C3) we first change variables with v =

radicav thereby rewriting (C1)

in the following symmetric form

parttu = uprime +radicav (C4a)

parttv =radicauprimeprime + vprime (C4b)

Equations (C4ab) then imply

1

2partt

int

D

(uprime)2 =

int

D

uprime(uprimeprime +radicavprime) =

int

D

radicauprimevprime +

1

2

int

partD

(uprime)2 (C5a)

1

2partt

int

D

(v)2 =

int

D

v(radicauprimeprime + vprime) = minus

int

D

radicauprimevprime +

1

2

int

partD

(v2 + 2radicauprimev) (C5b)

Here vvprime and uprimeuprimeprime have been expressed as exact derivatives and then integrated to boundaryterms the second equation employs an extra integration by parts and with only one spacedimension

int

partDdenotes a difference of endpoint evaluations Addition of Eqs (C5ab) gives

1

2partt

int

D

[

v2 + (uprime)2]

=1

2

int

partD

[

v2 + (uprime)2 + 2radicauprimev

]

(C6)

Substitutions with the identities

[

v2 + (uprime)2]

=1

2

[

(v + uprime)2 + (v minus uprime)2]

2uprimev =1

2

[

(v + uprime)2 minus (v minus uprime)2]

(C7)

26

and replacements to recover the original variable v = vradica yield

1

2partt

int

D

[

av2 + (uprime)2]

=1 +

radica

4

int

partD

(radicav + uprime)2 +

1minusradica

4

int

partD

(radicav minus uprime)2 (C8)

From (C8) we deduce that the time-dependent constant C(T ) in (C3) must satisfy

∣

∣

∣

∣

∣

∣

1 +

int T

0

[

1+radica

2

int

partD(radicav + uprime)2 + 1minus

radica

2

int

partD(radicav minus uprime)2

]

dt

uprime(0 middot)2 + av(0 middot)2

∣

∣

∣

∣

∣

∣

le C(T ) (C9)

For periodic boundary conditions we may choose C(T ) = 1 Moreover if a ge 1 anduprime = minusradic

av is specified at partD+ then uprime(t middot)2 + av(t middot)2 decaysStill working on a single interval (subdomain) we now consider the semi-discrete scheme

for (C4) i e (57) with all nonlinear source terms dropped and with v replaced by vradica

Derivation of a formula analogous to (C8) is our first step toward establishing L2 stabilityof the semi-discrete scheme While (57) features vectors for example u(t) taking values atthe Legendre-Gauss-Lobatto nodal points here we work with the numerical solution as apolynomial for example uh(t x) These two representations are related by the Lagrange in-terpolating polynomials for the nodal set here taken to span both the space of test functionsand the space of basis functions Our scheme is

int

Dk

ψparttuh =

int

Dk

ψ(Qh +radicavh) (C10a)

int

Dk

ξparttvh = minusint

Dk

ξprime(radicaQh + vh) +

int

partDk

ξ(radicaQlowast + vlowast) (C10b)

int

Dk

ϕQh =

int

Dk

ϕuprimeh +

int

partDk

ϕ (ulowast minus uh) (C10c)

where ψ ξ and ϕ are polynomial test functions These test functions are arbitrary exceptthat they must be degree-N polynomials In (C10) the variables uh vh and Qh should alsocarry a superscript k but we have suppressed this Derivation of a formula analogous to(C8) is complicated by the fact that Qh is not evolved Nevertheless at a given instant t wecan assemble Qh from (C10c)

Mimicking the calculation (C5b) from the continuum case we first use (C10b) with ξ = vhto write

1

2partt

int

Dk

v2h = minusint

Dk

(radicaQh + vh)v

primeh +

int

partDk

(radicaQlowast + vlowast)vh

= minusint

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2(radicaQlowast + vlowast)vh minus v2h

]

(C11)

The right-hand side of (C5a) is analogous to

1

2partt

int

Dk

Q2h =

int

Dk

QhparttQh (C12)

However since Qh is not evolved the term parttQh must be given a suitable interpretationOn the right hand side of (C10c) only uh u

primeh and u

lowast necessarily depend on time since the

27

test function ϕ need not be time-dependent Furthermore ulowast is explicitly given as a linearcombination of uh as seen in Eq (C21c) below Choosing ϕ = ℓj taking the time derivativeof (C10c) and appealing to the commutivity of mixed partial derivatives we therefore arriveat

int

Dk

ℓjparttQh =

int

Dk

ℓj(parttuh)prime +

int

partDk

ℓj(

(parttu)lowast minus parttuh

)

(C13)

where (parttu)lowast depends on parttuh in precisely the same way that ulowast depends on uh We have

written ℓj rather than ϕ in the last equation to emphasize that the result also holds for anylinear combination of ℓj (for example ϕ) and even for time-dependent combinations SinceQh is itself such a combination we obtain

1

2partt

int

Dk

Q2h =

int

Dk

Qh(parttuh)prime +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

Qh(Qprimeh +

radicavprimeh) +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2((parttu)lowast minus parttuh)Qh +Q2

h

]

(C14)

Addition of (C11) and (C14) gives

1

2partt

int

Dk

(Q2h + v2h) =

1

2

int

partDk

[

Q2h minus v2h + 2(

radicaQlowast + vlowast)vh + 2((parttu)

lowast minus parttuh)Qh

]

(C15)

the aforementioned analog of (C8) This formula holds on a single subdomain Dk and we

now combine multiple copies of it one for each value of k

2 Analysis for multiple intervals

To facilitate combination of (C15) over all k we change notation At every subdomaininterface Ik+12 equiv partDk cap partDk+1 let the superscripts L and R denote field values respectivelytaken from the left and right Then the fields evaluated at Ik+12 which belong to D

k will beuLk+12 v

Lk+12 and Q

Lk+12 while those belonging to D

k+1 will be uRk+12 vRk+12 and Q

Rk+12

However at Ikminus12 the values taken from D

k are uRkminus12 vRkminus12 and QR

kminus12 Note that we

have also replaced the subscript h denoting a numerical solution with k plusmn 12 denotingthe location of the endpoint value of the numerical solution With this notation we define

∆Lα =

1

2

[

(QLα)

2 minus (vLα)2]

+(radic

aQlowastα + vlowastα

)

vLα +[

(parttuα)lowast minus parttu

Lα

]

QLα (C16)

and similarly for ∆Rα The same numerical fluxes appear in both ∆L

α and ∆Rα (ie each

numerical flux takes the same value on either side of an interface) whence fluxes like Qlowastα do

not carry an L or R superscript In terms of these definitions (C15) becomes

1

2partt

int

Dk

(Q2h + v2h) = ∆L

k+12 minus∆Rkminus12 (C17)

28

Summation over all Dk yields

1

2partt

kmaxsum

k=1

int

Dk

(Q2h + v2h) =

kmaxminus1sum

k=1

(

∆Lk+12 minus∆R

k+12

)

+∆Lkmax+12 minus∆R

12

=

kmaxminus1sum

k=1

(

∆Lh minus∆R

h

)∣

∣

Ik+12 +∆Lkmax+12 minus∆R

12 (C18)

We have reverted to h-notation denoting the numerical solution since the LR superscriptsindicate unambiguously the relevant domain used for evaluation at Ik+12

We again seek an estimate of the form

kmaxsum

k=1

(

Qh(T middot)2Dk + avh(T middot)2Dk

)

le C(T )kmaxsum

k=1

(

Qh(0 middot)2Dk + avh(0 middot)2Dk

)

(C19)

that is essentially the same as the one (C3) considered in the continuum case Assume thatthe chosen boundary conditions ensure ∆L

kmax+12 minus ∆R12 is bounded by a time-dependent

constant which does not depend on the numerical parameters N and h (subdomain width)Establishment of stability then amounts to showing that the remaining sum over interfaceterms in (C18) is non-positive whence this remaining sum is consistent with C(T ) le 1although the boundary conditions may give rise to a different bound In fact we will choosethe numerical fluxes such that each individual interface term is non-positive At interfaceIk+12 and in LR notation the jump and average of vh for example are

1

2

(

v+ + vminus)

equiv vh =1

2

(

vLk+12 + vRk+12

)

(C20a)

nminusvminus + n+v+ equiv[[

vh]]

= vLk+12 minus vRk+12 (C20b)

Consider numerical fluxes of the form

Qlowast = Qh minusτQ2

[[

Qh

]]

(C21a)

vlowast = vh minusτv2

[[

vh]]

(C21b)

ulowast = uh minusτu2

[[

uh]]

(C21c)

(parttu)lowast = parttuh minus

τu2

[[

parttuh]]

(C21d)

where (C21c) induces (C21d) and where the penalty parameters τu τv and τQ are realnumbers The fluxes defined in (58) correspond to τu = 1 τv = 1 +

radica and τQ = 0 In

terms of these quantities the kth interface contribution in (C18) is

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

parttuh]]

minus τu2

[[

parttuh]][[

Qh

]]

(C22)

where we have suppressed the k dependence of the right-hand side Now consider the term[[

parttuh]]

Because parttuh and Qh+radicavh are both polynomials of degree N Eq (C10a) implies

29

τu

τ Q

minus1 0 1minus1

minus05

0

05

1

τu

minus1 0 1minus1

minus05

0

05

1

FIG 6 Stable evolutions for the model system For fixed τv = 10minus6 and τv = 1 +radic2

respectively the left and right plots depict stable choices (determined empirically) of τu and τQfor the linear model system (C1) The stable regions are colored black but the jagged edges result

from the discretization of the (τu τQ)-plane

the vector equation parttu = Q +radicav that is pointwise equivalence on the nodal points of

Dk which in turn implies

[[

parttuh]]

=[[

Qh+radicavh]]

Upon substituting this identity into thelast equation we arrive at an expression which features only vh and Qh

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

Qh +radicavh]]

minus τu2

[[

Qh +radicavh]][[

Qh

]]

(C23)

The identities vh[[

vh]]

= 12

[[

v2h]]

and[[

Qh +radicavh]]

=[[

Qh

]]

+radica[[

vh]]

then simplify(C23) to

(∆Lh minus∆R

h )∣

∣

Ik+12 = minusτv2

[[

vh]]2 minus

radica(τu + τQ)

2

[[

Qh

]][[

vh]]

minus τu2

[[

Qh

]]2 (C24)

The role of a penalty parameter is now clear Positive values of τv penalize jumps in vhthrough a negative contribution to the energy Likewise positive values of τu penalizejumps in Qh through a negative contribution to the energy However because the sign of[[

Qh

]][[

vh]]

can be positive or negative only the choice τQ = minusτu yields an expression for(∆L

h minus ∆Rh )|Ik+12 which is manifestly negative for τu ge 0 and τv ge 0 A simple estimate

based on Youngrsquos inequality with ε (that is 2αβ le εminus1α2 + εβ2 where α β ge 0 and ε gt 0)shows that for τQ = 0 the choice τv ge aτu4 also yields stability

Figure 6 depicts certain choices of stable penalty parameters for the linear model systemevolved to tfinal = 1000 (with a = 2 N = 10 and ∆t ≃ 00553) as determined empiricallywith simulations similar to those described in Sec IVA The left plot corresponds to asmall τv = 10minus6 for which the choice τu = 1 τQ = 0 is not stable as expected from thetheoretical analysis However the right plot corresponds to τv = 1 +

radica for which τu = 1

τQ = 0 is stable Motivated by the numerical flux choices (3842) used for the BSSN system

30

(4) we have (as mentioned above) set τu = 1 τv = 1 +radica and τQ = 0 in simulations of

the nonlinear model (53) For the nonlinear model system (53) the theoretically motivatedchoice τQ = minusτu also yields numerically stable evolutions when τu ge 0 and τv ge 0

For the nonlinear systems (4) and (53) we do not attempt a formal stability proofNevertheless the results of this appendix have served as a guide for our choices of penaltyparameters For the BSSN system (4) u v and Q are block indices [cf Eq (6)] Similarto the model problem we have penalized Q with τu = 1 with τv chosen large enough toheuristically overcome the cross-terms of indefinite size that arise from τQ = 0 (we interpretequations like τu = 1 componentwise) An analogous choice ldquoτQ = minusτurdquo for the BSSNsystem might be possible but would be considerably more complicated Indeed such achoice likely entails a matrix of penalty parameters but we do not give the details here

[1] F Pretorius Evolution of Binary Black-Hole Spacetimes Phys Rev Lett 95 121101 (2005)

(4 pages)

[2] F Pretorius Simulation of binary black hole spacetimes with a harmonic evolution scheme

Class Quantum Grav 23 S529-S552 (2006)

[3] M Campanelli C O Lousto P Marronetti and Y Zlochower Accurate Evolutions of Or-

biting Black-Hole Binaries without Excision Phys Rev Lett 96 111101 (2006) (4 pages)

[4] J G Baker J Centrella D-I Choi M Koppitz and J van Meter Gravitational-Wave

Extraction from an Inspiraling Configuration of Merging Black Holes Phys Rev Lett 96

111102 (2006) (4 pages)

[5] M Campanelli C O Lousto and Y Zlochower Last orbit of binary black holes Phys Rev D

73 061501(R) (2006) (5 pages)

[6] F Herrmann I Hinder D Shoemaker and P Laguna Unequal mass binary black hole plunges

and gravitational recoil Class Quantum Grav 24 S33-S42 (2007)

[7] P Diener F Herrmann D Pollney E Schnetter E Seidel R Takahashi J Thornburg and

J Ventrella Accurate Evolution of Orbiting Binary Black Holes Phys Rev Lett 96 121101

(2006) (4 pages)

[8] M A Scheel H P Pfeiffer L Lindblom L E Kidder O Rinne and S A Teukolsky Solving

Einsteinrsquos Equations With Dual Coordinate Frames Phys Rev D 74 104006 (2006)

[9] U Sperhake Binary black-hole evolutions of excision and puncture data Phys Rev D 76

104015 (2007) (20 pages)

[10] B Brugmann J A Gonzalez M Hannam S Husa U Sperhake and W Tichy Calibration

of moving puncture simulations Phys Rev D 77 024027 (2008) (25 pages)

[11] P Marronetti W Tichy B Brugmann J Gonzalez M Hannam S Husa and U Sperhake

Binary black holes on a budget simulations using workstations Class Quantum Grav 24

S43-S58 (2007)

[12] Z B Etienne J A Faber Y T Liu S L Shapiro and T W Baumgarte Filling the holes

Evolving excised binary black hole initial data with puncture techniques Phys Rev D 76

101503(R) (2007) (5 pages)

[13] B Szilagyi D Pollney L Rezzolla J Thornburg and J Winicour An explicit harmonic code

for black-hole evolution using excision Class Quantum Grav 24 S275-S293 (2007)

[14] L Boyle M Kesden and S Nissanke Binary-Black-Hole Merger Symmetry and the Spin

Expansion Phys Rev Lett 100 151101 (2008) (4 pages)

31

[15] M A Scheel M Boyle T Chu L E Kidder K D Matthews and H P Pfeiffer High-

accuracy waveforms for binary black hole inspiral merger and ringdown Phys Rev D 79

024003 (2009) (14 pages)

[16] M Hannam Status of black-hole-binary simulations for gravitational-wave detection

Class Quantum Grav 26 114001 (2009) (21 pages)

[17] I Hinder The current status of binary black hole simulations in numerical relativity

Class Quantum Grav 27 114004 (2010) (19 pages)

[18] B P Abbott et al LIGO The Laser Interferometer Gravitational-Wave Observatory

Rep Prog Phys 72 076901 (2009) (25 pages)

[19] F Acernese et al The status of VIRGO Class Quantum Grav 23 S63-S69 (2006)

[20] F Acernese et al Virgo status Class Quantum Grav 25 184001 (2008) (9 pages)

[21] R Arnowitt S Deser and C W Misner The Dynamics of General Relativity in Gravitation

an introduction to current research edited by L Witten (Wiley New York 1962) Chapter 7

227-265 Also available as gr-qc0405109

[22] J W York Jr Kinematics and Dynamics of General Relativity in Sources of Gravitational

Radiation edited by L L Smarr (Cambridge University Press Cambridge 1979)

[23] L E Kidder M A Scheel and S A Teukolsky Extending the lifetime of 3D black hole

computations with a new hyperbolic system of evolution equations Phys Rev D 64 064017

(2001) (13 pages)

[24] G Calabrese J Pullin O Sarbach and M Tiglio Convergence and stability in numerical

relativity Phys Rev D 66 041501(R) (2002) (4 pages)

[25] S Frittelli and O Reula On the Newtonian Limit of General Relativity Commun Math

Phys 166 221-235 (1994)

[26] M Shibata and T Nakamura Evolution of three-dimensional gravitational waves Harmonic

slicing case Phys Rev D 52 5428-5444 (1995)

[27] Y Choquet-Bruhat and J W York Geometrical well posed systems for the Einstein equations

C R Acad Sci Paris t 321 Serie I 1089-1095 (1995)

[28] A Abrahams A Anderson Y Choquet-Bruhat and J W York Jr Einstein and Yang-Mills

Theories in Hyperbolic Form without Gauge Fixing Phys Rev Lett 75 3377-3381 (1995)

[29] C Bona J Masso E Seidel and J Stela New Formalism for Numerical Relativity

Phys Rev Lett 75 600-603 (1995)

[30] M H P M van Putten and D M Eardley Nonlinear wave equations for relativity

Phys Rev D 53 3056-3063 (1996)

[31] S Frittelli and O A Reula First-Order Symmetric Hyperbolic Einstein Equations with Ar-

bitrary Fixed Gauge Phys Rev Lett 76 4667-4670 (1996)

[32] H Friedrich Hyperbolic reductions for Einsteinrsquos equations Class Quantum Grav 13 1451-

1469 (1996)

[33] F B Estabrook R S Robinson H D Wahlquist Hyperbolic equations for vacuum gravity

using special orthonormal frames Class Quantum Grav 14 1237-1247 (1997)

[34] M S Iriondo E O Leguizamon and O A Reula Einsteinrsquos Equations in Ashtekarrsquos Vari-

ables Constitute a Symmetric Hyperbolic System Phys Rev Lett 79 4732-4735 (1997)

[35] A Anderson Y Choquet-Bruhat and J W York Einstein-Bianchi hyperbolic system for

general relativity Topol Meth Nonlinear Anal 10 353-373 (1997)

[36] T W Baumgarte and S L Shapiro Numerical integration of Einsteinrsquos field equations

Phys Rev D 59 024007 (1998) (7 pages)

[37] M A G Bonilla Symmetric hyperbolic systems for Bianchi equations Class Quantum

11

there is a corresponding Galerkin condition in total 9kmax(N + 1) equations for (13a) and5kmax(N+1) for (13b) Enforcement of the Galerkin conditions on each D

k will not recover ameaningful global solution since they provide no mechanism for coupling the local solutionson the different intervals Borrowing from the finite volume toolbox we achieve couplingthrough integration by parts on r and introduction of the numerical flux f lowast at the interfacebetween subdomains

In (26) we only need to consider (βrK prime)h and (χQprimeαgrr)h as the other terms comprise a

component of the source vector Sh Using integration by parts we write

(

ℓkj (βrK prime)h

)

Dk =minusint bk

akdr[

(

ℓkjβrh

)primeKh

]

+ (βrhKh) ℓ

kj

∣

∣

∣

bk

ak (30a)

(

ℓkj (χQprimeαgrr)h

)

Dk =minusint bk

akdr

[(

ℓkjχh

grrh

)prime

Qαh

]

+

(

χhQαh

grrh

)

ℓkj

∣

∣

∣

bk

ak (30b)

In these formulas we have retained the domain index k on ℓkj while continuing to suppress iton Kh grrh etc Moreover we have suppressed the r-dependence in all terms on the right-hand side Addition of these formulas along with the definition fKh = minus(βrK)h+(χQαgrr)hgives

(

ℓkj (βrK prime)h minus (χQprime

αgrr)h)

Dk = minusint bk

akdr

[

(

ℓkjβrh

)primeKh minus

(

ℓkjχh

grrh

)primeQαh

]

minus fKhℓkj

∣

∣

∣

bk

ak (31)

In lieu of (31) we will instead work with the replacement

(

ℓkj (βrK prime)h minus (χQprime

αgrr)h)

Dk rarr minusint bk

akdr

[

(

ℓkjβrh

)primeKh minus

(

ℓkjχh

grrh

)primeQαh

]

minus f lowastKℓ

kj

∣

∣

∣

bk

ak (32)

This replacement features a component f lowastK of the numerical flux rather than a component

fKh of the boundary flux The numerical flux is determined by (as yet not chosen) functions3

f lowast = f lowast(W+Wminus) (33)

where for example Wminus is an interior boundary value [either W kh (t a

k) or W kh (t b

k)] of theapproximation defined on D

k andW+ is an exterior boundary value [either W kminus1h (t bkminus1) or

W k+1h (t ak+1)] of the approximation defined on either Dkminus1 or Dk+1 We discuss our choice

of numerical flux in the next subsection We now employ additional integration by parts towrite the above replacement as

(

ℓkj (βrK prime)h minus (χQprime

αgrr)h)

Dk rarrint bk

akdrℓkj

(

βrK prime minus χQprimeα

grr

)

h

+ (fKh minus f lowastK) ℓ

kj

∣

∣

∣

bk

ak (34)

Rather than the exact kth Galerkin condition(

ℓkj (RK)kh

)

Dk = 0 forallj for the K component

of (13) on Dk we will instead strive to enforce

(

ℓkj (RK)kh

)

Dk = (fKh minus f lowastK) ℓ

kj

∣

∣

∣

bk

ak forallj (35)

3 In the context of the dG method here + and minus denote ldquoexteriorrdquo and ldquointeriorrdquo and have no relation to

the plusmn using to denote the characteristic fields and speeds in Table I For characteristic fields and speeds

+ and minus mean ldquoright-movingrdquo and ldquoleft-movingrdquo

12

although our treatment of nonlinear terms will lead to a slight modification of these equations(we return to this issue shortly) The other components of (13a) are treated similarly asare the components of (13b) Recall that for example Qα = αprime Formally using the samedG method to solve for Qα we arrive at the replacement

(

ℓkj (RQα)kh

)

Dk rarrint bk

akdrℓkj (minusQαh + αprime

h)minus (αh minus αlowast) ℓkj

∣

∣

∣

bk

ak (36)

which again features a component αlowast of the numerical flux The auxiliary variables areconstructed and used at each stage of temporal integration We then have

(

ℓkj (RQα)kh

)

Dk = (αh minus αlowast) ℓkj

∣

∣

∣

bk

ak forallj (37)

as the corresponding enforced kth Galerkin condition

B Numerical Flux

To further complete our dG scheme we must specify functional forms for the compo-nents of the numerical flux introduced in the previous section We distinguish between thephysical fluxes (components of f) and the auxiliary fluxes (components of u) arising fromthe definition of the auxiliary variables These choices are not independent as the resultingscheme must be stable and consistent Our choice follows [84] which considered diffusionproblems Additional analysis of this flux choice appears in [66 85]

Let us first consider the numerical fluxes corresponding to the physical fluxes and of theform (33) The numerical flux vector is a function of the system and auxiliary variablesinterior and exterior to a subdomain A common choice for f lowast is

f lowast = fh+τ

2

[[

vh]]

K-component of f lowast f lowastK = fKh+

τ

2

[[

Kh

]]

(38)

where as an example we have also shown the component of f lowast corresponding to the analysisabove Respectively the average and jump across the interface are

fh =1

2

(

f+ + fminus) [[

vh]]

= nminusvminus + n+v+ (39)

Here τ is a position dependent penalty parameter (fixed below) and nminus(n+) is the localoutward pointing normal to the interior (exterior) subdomain The role of τ is to ldquopenalizerdquo(i e yield a negative contribution to the L2 energy norm) jumps across an interface Anappropriate choice of τ will ensure stability and we now provide some motivation for thechoice (41) of τ we make below

Were we treating the fully first-order system (10) the local Lax-Friedrichs flux wouldoften be a preferred choice due to its simplicity [66] In this case the constant ω in thenumerical flux formula Flowast = Fh + 1

2ω[[

Wh

]]

obeys ω ge max∣

∣micro(nablaWF(W ))∣

∣ HereF(W ) = A(u)W the notation micro(middot) indicates the spectral radius of the matrix within andthe max is taken over interior Wminus and exterior W+ states Motivated by (9) we adopt asimilar but simpler prescription substituting the field gradient

nablaWvQA(u)WvQ = A(u) (40)

13

for nablaWF(W ) Precisely we assume the scaling

τ(bk) = τ(ak+1) = τk+12 equiv C middotmax∣

∣micro(

A(u))∣

∣ (41)

where C = O(1) is a constant chosen for stability Larger values of C will result in schemeswith better stability properties whereas too large a value will impact the CFL conditionAt the interface point Ik+12 equiv D

k cap Dk+1 the vector uh has two representations uminus at bk

and u+ at ak+1 The max in (41) is taken over the corresponding two sets of field speedsMore precisely the speeds in Table I are computed for both uminus and u+ and the maximumtaken over all resulting speeds For the auxiliary variables a penalized central flux is usedThe definition with one representative component is

ulowast = uh minus1

2

[[

uh]]

α-component of ulowast αlowast = αh minus1

2

[[

αh

]]

(42)

with similar expressions for the remaining componentsWe stress the following point Since the interior coupling between subdomains is achieved

through the numerical flux forms (41) and (42) the inverse transformation (17) expressingthe fundamental fields in terms of the characteristic fields is not required to achieve thiscoupling On the other hand imposition of physical boundary conditions may still rely on(17) since this transformation allows one to fix only incoming characteristic modes

C Matrix form of the semi-discrete equations

Let us introduce the kth mass and stiffness matrices

Mkij =

int bk

akdrℓki (r)ℓ

kj (r) Sk

ij =

int bk

akdrℓki (r)ℓ

kjprime(r) (43)

These matrices belong to Dk and the corresponding matrices defined on the reference interval[minus1 1] are

Mij =

int 1

minus1

duℓi(u)ℓj(u) Sij =

int 1

minus1

duℓi(u)ℓprimej(u) (44)

where ℓj(u) is the jth Lagrange polynomial determined by the LGL nodes uj on [minus1 1]These matrices are related by Mk

ij =12(bk minus ak)Mij and S

kij = Sij whence only the reference

matrices require computation and storageWe will use the matrices Mk and Sk in obtaining an ODE system from (26) and (35)

Towards this end we first approximate the nonlinear terms (products and quotients) in(26) by degree-N interpolating polynomials Such approximations are achieved throughpointwise representations For example (QαQχgrr)h appears in (26) and is expressed inthe following way [cf footnote 2]

(

QαQχ

grr

)

h

(t r) =

(

QαhQχh

grrh

)

(t r) rarrNsum

j=0

Qαh(t rkj )Qχh(t r

kj )

grrh(t rkj )

ℓkj (r) (45)

Note that the expressions on the right and left are not equivalent due to aliasing error [65]Our vector notation for this replacement will be

(

QαQχ

grr

)

h

(t r) rarr(

QαQχ

grr

)

(t)Tℓk(r) (46)

14

Operations among bold variables are always performed pointwise Making similar replace-ments for all terms in (26) and then carrying out the integrations in (35) which bring inMk and Sk we arrive at

parttK = βrDK minus χDQα

grr

+1

2

χQgrrQα

g2rr

minusχQgθθ

Qα

grrgθθ

+1

2

QαQχ

grr

+3

2

αA2rr

g2rr

+1

3αK2 +Mminus1ℓk (fKh minus f lowast

K)∣

∣

∣

bk

ak (47)

where we have again suppressed the superscript k on all terms except ℓk(r) and the sub-script h is dropped on all boldfaced variables As described in [66] the spectral collocationderivative matrix

(Dk)ij =dℓkjdr

∣

∣

∣

∣

∣

r=rki

(48)

can also be expressed as Dk = (Mk)minus1Sk which appears above Eight other semi-discreteevolution equations are similarly obtained with nine in total (one for each component ofWuv) Additionally we have

Qα = Dα+Mminus1ℓk(αlowast minusαh)∣

∣

∣

bk

ak (49)

as one of the auxiliary equations with five in total (one for each component of Q = WQQ)

D Filtering

Like other nodal (pseudospectral) methods our scheme may suffer from instabilitiesdriven by aliasing error [65] Filtering is a simple yet robust remedy To filter a solu-tion component such as χ we use the modal (as opposed to nodal) representation of thesolution

χkh(t r) =

Nsum

j=0

χ(t rkj )ℓkj (r) =

Nsum

j=0

χkj (t)Pj(r) (50)

where Pj(r) is the jth Legendre polynomial Let ηj = jN and define the filter function

σ(ηj) =

1 for 0 le ηj le NcN

exp(

minus ǫ(

ηjminusNcN

1minusNcN

)2s)

for NcN le ηj le 1(51)

At each timestep we modify our solution component according to

χkh rarr

(

χkh

)filtered=

Nsum

j=0

σ(ηj)χkj (t)Pj(r) (52)

Evidently the modification only affects the top N minus Nc modes and is sufficient to controlthe type of weak instability driven by aliasing [66] The numerical parameters Nc and ǫ areproblem dependent For our simulations we have taken ǫ ≃ minuslog(εmach) = 36 where εmach

is machine accuracy in double precision

15

E Model system

To better illustrate the basic properties of our method we consider a toy model Namelythe following spatially second-order system

parttu = uprime + av minus u3 + g(t x) (53a)

parttv = uprimeprime + vprime minus (u+ v)(uprime)2 + v2u2 + h(t x) (53b)

where a ge 1 is constant and g and h are analytic source terms to be specified In contrastto (6) here u v and Q = uprime are scalars rather than vectors System (53) admits a first-order reduction in which uprime is defined as an extra variable Since this first-order reductionis strongly hyperbolic the spatially second-order system (53) is also strongly hyperbolic byone of the definitions considered in [75] The characteristic fields Xplusmn and speeds microplusmn are

X+ =radicav minus uprime micro+ =

radicaminus 1 Xminus =

radicav + uprime microminus = minus(

radica+ 1) (54)

To construct a local dG scheme for this system we first rewrite it as

parttu = Q+ av minus u3 + g(t x) (55a)

parttv = Qprime + vprime minus (u+ v)Q2 + v2u2 + h(t x) (55b)

Q = uprime (55c)

Evidently f = minus(Q + v) is the v-component of the physical flux vector

F (vQ) equiv(

Fu

Fv

)

=

(

0

f

)

(56)

Note that F has the same structure as (u v)T Borrowing from the presentation for theBSSN system we write the analogous semidiscrete scheme on each subdomain D

k for themodel system

parttu = Q+ av minus u3 + g(t) (57a)

parttv = DQ+Dv minus (u+ v)Q2 + v2u2 + h(t) +Mminus1ℓk(fh minus f lowast)∣

∣

∣

bk

ak (57b)

Q = Du+Mminus1ℓk(ulowast minus uh)∣

∣

∣

bk

ak (57c)

Here we have suppressed the subinterval label k from all variables except for the vectorℓk of Lagrange polynomial values Moreover following the guidelines discussed above thenumerical fluxes are given by

f lowast = fh+1 +

radica

2

[[

vh]]

ulowast = uh minus1

2

[[

uh]]

(58)

Appendix C analyzes the stability of our scheme for a more general numerical flux choiceas applied to (53) with the nonlinear and source terms dropped

IV RESULTS FROM NUMERICAL SIMULATIONS

This section presents results found by numerically solving both the model system (53)and BSSN system (4) with the dG scheme presented in Sec III

16

10minus10

10minus6

10minus2

∆uinfin

10minus10

10minus6

10minus2

∆vinfin

0 500 1000 1500 2000 2500 3000 3500 400010

minus10

10minus6

10minus2

∆(partxu)infin

t

N = 3N = 6N = 9N = 12

FIG 1 Spectral convergence of fields for model PDE Respectively for N = 3 6 9 12

a timestep of ∆t = 00578 00178 00084 00049 has been chosen for stability and accuracy In the

title headings for example ∆u equiv unumer minus uexact

A Simulations of the model system

The semi-discrete scheme (57) has been integrated with the classical fourth-order Runge-Kutta method When integrating this system we have first constructed Q at each Runge-Kutta stage and then substituted into the evolution equations (57ab) for u and v Theproblem has been solved on a computational domain [0 4π] comprised of two subdomainswith a timestep chosen small enough for stability The initial data has been taken from thefollowing exact solution to (53)

uprimeexact(t x) =1

2

[

sin(xminus microminust)minus sin(xminus micro+t)]

(59a)

vexact(t x) =1

2radica

[

sin(xminus microminust) + sin(xminus micro+t)]

(59b)

g(t x) = u3exact (59c)

h(t x) = (uexact + vexact)(uprimeexact)

2 minus v2exactu2exact (59d)

where the speeds microplusmn are found in (54) Specification of the boundary condition at a physicalendpoint amounts to choosing the external state for at the endpoint We have considered

17

8 16 32 64 128

10minus3

10minus2

10minus1

N

∆t

max stable ∆t8N2

FIG 2 Scaling of maximum stable ∆t with N for model PDE

two possibilities (i) the analytic state (Q+ v+) = (Qexact vexact) and (ii) an upwind stateFor example at x = 4π the upwind state is4

Q+ = Qupwind =1

2

[

(Xminus)exact minus (X+)numer

]

v+ = vupwind =1

2radica

[

(Xminus)exact + (X+)numer

]

(60)Either choice of (Q+ v+) leads to similar results and the plots here correspond to theanalytic state Figure 1 clearly shows spectral convergence with increasing polynomial orderN across all fields for the case a = 2 Other values of a including a = 1 for which X+

is a static characteristic field have also been considered with similar results Appendix Cdemonstrates that our proposed scheme for the system (57) with nonlinear and source termsdropped is stable in a semi-discrete sense Nevertheless the fully discrete scheme obtainedvia temporal discretization by the fourth-order Runge-Kutta method is still subject to thestandard absolute stability requirement Namely if microh is any eigenvalue corresponding tothe (linearized) discrete spatial operator then a necessary condition for stability is thatmicroh∆t lies in absolute stability region for Runge-Kutta 4 We here show empirically that theassociated timestep restriction scales like Nminus2 ie ∆t = O(Nminus2) for stability We note thatsuch scaling is welcome in light of the second-order spatial operators which appear in thesystem and suggest a possible worse scaling like Nminus4 Fig 2 plots the maximum stabletimestep for a range of N demonstrating the Nminus2 scaling in line with behavior known fromanalysis of first-order systems [66] This scaling also holds for the BSSN system

4 We remind the reader that unfortunately the plusmn on Xplusmn means something different than the plusmn indicating

exteriorinterior dG states [cf footnote 3]

18

B Simulations of the BSSN system

This subsection documents results for simulations of the unit-mass-parameter (M = 1)Schwarzschild solution (B9) expressed in terms of ingoing Kerr-Schild coordinates Sincethe solution is stationary temporal integration of the semi-discrete scheme has been carriedout with the forward Euler method The r-coordinate domain [04 34] has been split into3 equally spaced subdomains and we have set η = 10 λ = 01 and C = 2 [cf Eq (41)]For all simulations ∆t has been chosen for stability With the chosen λ the inner physicalboundary rmin = 04 is an excision surface At each timestep we have applied an (order2s = 20) exponential filter on the top two-thirds of the modal coefficient set for all fieldsexcept for grr and gθθ For stability we have empirically observed that grr and gθθ must not

be filtered A detailed understanding of this is still lackingIssues related to physical boundary conditions are similar to the one encountered in

Sec IVA for the model problem Similar to before we have retained Eqs (3842) as thechoice of numerical flux even at the endpoints Therefore at an endpoint the specificationof the boundary condition amounts to the choice W+ of external state We have typicallychosen the inner boundary of the radial domain as an excision boundary and in this caseW+ = Wminus is enforced at the inner physical boundary At the outer physical boundaryfor W+ we have again considered two choices (i) Wexact and (ii) Wupwind To enforcechoice (ii) the inverse transformation (17) must be used with incoming characteristic fieldsfixed to their exact values similar to (60) We have tried various versions of choice (ii)and in all cases the resulting simulations have been unstable We therefore present resultscorresponding to choice (i) Although the choice of an analytical external state Wexact

at the outer boundary is stable for our problem such a boundary condition is unlikelyto generalize to more complicated scenarios involving dynamical fields Indeed the issueof outer boundary conditions for the BSSN system is an active area of research with aproper treatment requiring fixation of incoming radiation control of the constraints andspecification of gauge (see Ref [86] for a recent analysis)

For BSSN simulations our main diagnostic is to monitor the Hamiltonian momentumand conformal connection constraints Figure 3 depicts long-time histories of constraintviolations whereas Figs 4 and 5 depict long-time error histories for the individual BSSNfield components From the middle plot in Fig 5 we infer that up to the indicated numericalerror the factor g sin4 θ = grr(gθθ)

2 remains at its initial fixed profile r4 throughout theevolution These figures indicate that the proposed scheme is stable for long times andexhibits spectral converge with increased polynomial order N Similar results are recoveredfromM = 0 Minkowski initial data The stability documented in these plots does not appearto rely on inordinate parameter tuning For example with the fixed parameters describedabove we obtain similar plots if we individually vary (i) rmin over 0325 035 04 0475(values still corresponding to an excision surface for the given choice of λ) (ii) η over1 3 7 10 (iii) s over 8 9 10 With the polynomial order N ranging over 23 26 29 31both stability and qualitatively similar exponential convergence is achieved with a singlesubdomain Likewise adoption of a larger coordinate domain with more subdomains doesnot impact our results Finally we have considered the addition of random noise to all fieldcomponents at the initial time Precisely with the system component χ as an example wehave set

χ(0) 7rarr χ(0) + δχ(0) (61)

where each component (nodal value) of δχ(0) is 10minus5 times a random variable drawn from a

19

10minus9

10minus7

10minus5

10minus3

Hinfin

10minus9

10minus7

10minus5

10minus3

Mrinfin

0 2000 4000 6000 8000 10000

10minus9

10minus7

10minus5

10minus3

Grinfin

t

N = 11N = 14N = 17N = 20

FIG 3 Spectral convergence of constraint violations for M = 1 Kerr-Schild ini-

tial data Respectively for N = 11 14 17 19 a timestep of ∆t ≃ 00041 00026 00018 00013

has been chosen for stability and accuracy

standard normal distribution Such perturbed initial data also gives rise to stable evolutions

V CONCLUSION

We have introduced a discontinuous Galerkin method for solving the spherically reducedBSSN system with second-order spatial operators Our scheme shares similarities withother discontinuous Galerkin methods that use local auxiliary variables to handle high-orderspatial derivatives [66 77ndash79 82 84 85] and which have typically been applied to eitherelliptic parabolic or mixed type problems The key ingredient of a stable dG scheme is anappropriate choice of numerical flux and our particular choice has been motivated by theanalysis presented in Appendix C When used to evolve the Schwarzschild solution in Kerr-Schild coordinates our numerical implementation of the BSSN system (4) is robustly stableand converges to the analytic solution exponentially with increased polynomial order Byapproximating the spatially second-order form of the BSSN system we have not introducedextra fields which are evolved Evolved auxiliary fields result in new constraints which mayspoil stability Our main goal has been stable evolution of the spherically reduced BSSNsystem as a first step towards understanding how a discontinuous Galerkin method might be

20

10minus9

10minus7

10minus5

∆Arrinfin

10minus9

10minus7

10minus5

∆Kinfin

0 2000 4000 6000 8000 10000

10minus9

10minus7

10minus5

max(∆Brinfin ∆Γrinfin)

t

N = 11N = 14N = 17N = 20

FIG 4 Spectral convergence of solution for M = 1 Kerr-Schild initial data

Timestep choices are described in the caption for Fig 3 In the title headings for example

∆Arr equiv (Arr)numer minus (Arr)exact

applied to the full BSSN system Towards that goal we now discuss treatment of singularitiesand generalization of the described dG method to higher space dimension

To deal with the fixed Schwarzschild singularity we have used excision which is easy inthe context of the spherically reduced BSSN system However excision for the binary blackhole problem in full general relativity requires attention to the technical challenge of hori-zon tracking State-of-the-art BSSN codes avoid such complication relying instead on themoving-puncture technique While the moving-puncture technique does involve mild centralsingularities it may still prove amenable to spectral methods Indeed spectral methods fornon-smooth problems is well-developed in both theory and for complex applications Sincethe moving-puncture technique can be performed in spherical symmetry [60] a first-steptoward a spectral moving-puncture code would be to implement a moving puncture withthe nodal dG method described here Such an implementation may adopt Legendre-Gauss-Radau nodes on the innermost subdomain thereby ensuring that the physical singularitydoes not lie on a nodal point (in much the same way finite difference codes use a staggeredgrid) Beyond traditional excision and moving punctures one might also retain a smoothproblem for which spectral methods are especially well-suited via the turducken approach[87 88] to singularities

21

10minus11

10minus9

10minus7

10minus5

∆χinfin

10minus11

10minus9

10minus7

10minus5

max(∆grrinfin ∆gθθinfin)

0 2000 4000 6000 8000 10000

10minus11

10minus9

10minus7

10minus5

max(∆αinfin ∆βrinfin)

t

N = 11N = 14N = 17N = 20

FIG 5 Spectral convergence of solution violations for M = 1 Kerr-Schild initial

data See the caption of Fig 4 for details

Discontinuous Galerkin methods for hyperbolic problems in two and three space dimen-sions are well-developed A generalization of the method described here to three-dimensionsand the full BSSN system would likely rely on an unstructured mesh Appropriate local poly-nomial expansions for the subelements are well-understood as are choices for the numericalfluxes which would now live on two-dimensional faces rather than single points Whether ornot it would ultimately prove successful generalization of our dG method to a higher dimen-sion would rely on an established conceptual framework Further computational advancesof relevance to a generalization of our dG method to the full BSSN system (possibly includ-ing matter) may include mesh hp-adaptivity local timestepping shock capturing and slopelimiting techniques [66] Moreover recent work [89] indicates that enhanced performancewould be expected were our scheme implemented on graphics processor units

VI ACKNOWLEDGMENTS

We thank Nick Taylor for discussions about treating second-order operators with spectralmethods Benjamin Stamm for helping polish up a few parts of Appendix C Khosro Shah-bazi for discussions on LDG and IP methods We also acknowledge helpful conversationswith David Brown and Manuel Tiglio about the BSSN system and previous work on its nu-

22

merical implementation We gratefully acknowledge funding through grants DMS 0554377and DARPAAFOSR FA9550-05-1-0108 to Brown University and NSF grant PHY 0855678to the University of New Mexico

Appendix A Hyperbolicity of the first-order system

This appendix analyzes the matrix A(u) appearing in (10) in order to construct thecharacteristic fields (14) In matrix form the sector (8) of the principal part of (10) reads asfollows

partt

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

=

βr 0 minus 4λα3grr

0 0 λβr

6(grr)2λβr

3gθθgrr0 4λ

3grr

0 βr 0 23grrαχ

13α minus αχ

3grr

αχ3gθθ

minus23χ 0

0 0 βr 0 0 0 0 minus χgrr

0

0 0 minus 4α3grr

βr 0 βr

6(grr)2βr

3gθθgrr0 4

3grr

0 0 23αχ 0 βr minus βrχ

3grrminus2βrχ

3gθθ0 minus2

3χ

0 minus2α 0 0 0 23βr minus2grrβr

3gθθ0 4

3grr

0 gθθαgrr

0 0 0 minusgθθβr

3grr13βr 0 minus2

3gθθ

0 0 minus2α 0 0 0 0 βr 0

34

0 0 0 0 0 0 0 βr

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

prime

(A1)which defines the matrix A(u) appearing in (9) and so also the matrix A(u) in (10) Notethat in the last equation the matrix within the square brackets is minusA(u) For certainconfigurations of u and λ the system (10) is strongly hyperbolic [76] that is A(u) hasa complete set of eigenvectors and real eigenvalues Indeed five eigenpairs of A(u) aretrivially recovered upon inspection of A(u)rsquos leading 5times5 diagonal block These correspondto eigenvalue 0 and the left eigenspace ξj = eTj 1 le j le 5 where ej are the canonical

basis vectors Since each component of u arises as eTj W each is also a characteristic field

The remaining nine eigenpairs are determined by A(u) The eigenvalues of A(u) are

micro1 = 0 micro23 = minusβr microplusmn4 = minusβr plusmn

radic

2αχ

grr microplusmn

5 = minusβr plusmn α

radic

χ

grr microplusmn

6 = minusβr plusmnradic

λ

grr

(A2)

23

and the corresponding left eigenvectors are

x1 = (0 0 0 0 0 gθθ 2grr 0 0) (A3a)

x2 =

(

0 0 0 grr2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3b)

x3 =

(

grrλ 0 0 0

2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3c)

xplusmn4 =

(

0 0plusmnradic

2αgrrχ

0 0 0 0 1 0

)

(A3d)

xplusmn5 =

(

0∓ 3radicgrrχ

plusmn2

radic

grrχ 2grr

1

χminus 1

grr1

gθθ 0 0

)

(A3e)

xplusmn6 =

(

minus3

4

grrλ 0plusmn α

radicλgrr

(2αχminus λ) 0 0minus βr

8(βrgrr ∓radicλgrr)

minus βrgrr

4gθθ(βrgrr ∓radicλgrr)

αχ

(2αχminus λ)plusmnradic

grrλ

)

(A3f)

where for example xplusmn5 A(u) = microplusmn5 x

plusmn5 Assuming that grr gθθ χ and α are everywhere strictly

positive the eigenvalues are real and the eigenvectors are linearly independent providedthat (15) holds These eigenvectors are easily extended to eigenvectors of A(u) e g asxplusmn6 rarr (01times5 x

plusmn6 ) Then for example the characteristic field

Xplusmn6 equiv (01times5 x

plusmn6 )W = xplusmn6WvQ (A4)

and similarly Xplusmnj = xplusmnj WvQ for j = 4 5 and Xk = xkWvQ for k = 1 2 3 The characteristic

speeds for these fields are microk and microplusmnj With this convention the speeds listed in Table I

correspond to the Xk and Xplusmnj in (14)

Appendix B Schwarzschild solution in conformal Kerr-Schild coordinates

In Kerr-Schild coordinates here the system directly related to incoming Eddington-Finkelstein null coordinates the line element for the Schwarzschild solution reads

ds2 = minusα2dt2 + (1 + 2MR)(dR + βRdt)2 +R2dθ2 +R2 sin2 θdφ2 (B1)

where R is the area radius α = (1 + 2MR)minus12 is the lapse and βR = 2M(R + 2M) isthe shift vector The physical spatial metric gab is the spatial part of this line element

To define the corresponding solution to the BSSN system we use equation gab = χgab todefine the following relationship between line elements

dr2 + r2(dθ2 + sin2 θdφ2) = χ[(1 + 2MR)dR2 +R2dθ2 +R2 sin2 θdφ2] (B2)

so that

χ

(

1 +2M

R

)(

dR

dr

)2

= 1 χR2 = r2 (B3)

Then we have(

1 +2M

R

)12dR

R=dr

r (B4)

24

with integration yielding

r =R

4

(

1 +

radic

1 +2M

R

)2

e2minus2radic

1+2MR (B5)

where the constant of integration has been chosen so that the R r rarr infin limits are consistentThe second relation in (B3) shows that

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR =rArr χminus4 =2eradic

1+2MRminus1

1 +radic

1 + 2MR (B6)

The extrinsic curvature tensor is specified by the expression forK given in (B9h) the identityK = KR

R + 2Kθθ and

Kθθ =

(

1 +2M

R

)minus122M

R2 (B7)

Since KRR = Kr

r we compute that

Krr = K minus 2Kθ

θ = minus(

1 +2M

R

)minus12(R +M

R + 2M

)

2M

R2 (B8)

Next since Krr = grrKrr = χminus1Kr

r we have Krr = Arr+

13grrK This implies Arr = Kr

r minus 13K

from which we get (B9g) In all we have

α =

(

1 +2M

R

)minus12

(B9a)

βr = βR dr

dR= χ12

(

1 +2M

R

)minus122M

R(B9b)

grr = 1 (B9c)

gθθ = r2 = χR2 (B9d)

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR (B9e)

Br = 0 (B9f)

Arr = minus(

1 +2M

R

)minus124M

3R2

(

2R + 3M

R + 2M

)

(B9g)

K =

(

1 +2M

R

)minus32(

1 +3M

R

)

2M

R2(B9h)

Γr = minus2

r= minus 2

χ12R (B9i)

To differentiate these expressions with respect to r we use the identity

dR

dr= χminus12

(

1 +2M

R

)minus12

(B10)

along with the chain rule

25

Appendix C Stability of the model system

The following stability analysis for the model system (53) has been inspired by [78 79]After dropping all nonlinear source terms the system (53) becomes

parttu = uprime + av (C1a)

parttv = uprimeprime + vprime (C1b)

This section analyzes the stability of (C1) considering both the continuum system itself aswell as its semi-discrete dG approximation The latter analysis offers some insight into theempirically observed stability of our dG scheme for the spherically reduced BSSN equations

1 Analysis for a single interval

Throughout we work with the L2 inner product and norm

(f g)D =

int

D

fg fD =radic

(f f)D (C2)

where D is the spatial coordinate interval (here D may represent a subdomain Dk or the

whole domain Ω) and we have suppressed all integration measures For the continuummodel we will establish the following estimate

uprime(T middot)2D+ av(T middot)2

Dle C(T )

(

uprime(0 middot)2D+ av(0 middot)2

D

)

(C3)

where the time-dependent constant C(T ) is determined solely by the choice of boundaryconditions To show (C3) we first change variables with v =

radicav thereby rewriting (C1)

in the following symmetric form

parttu = uprime +radicav (C4a)

parttv =radicauprimeprime + vprime (C4b)

Equations (C4ab) then imply

1

2partt

int

D

(uprime)2 =

int

D

uprime(uprimeprime +radicavprime) =

int

D

radicauprimevprime +

1

2

int

partD

(uprime)2 (C5a)

1

2partt

int

D

(v)2 =

int

D

v(radicauprimeprime + vprime) = minus

int

D

radicauprimevprime +

1

2

int

partD

(v2 + 2radicauprimev) (C5b)

Here vvprime and uprimeuprimeprime have been expressed as exact derivatives and then integrated to boundaryterms the second equation employs an extra integration by parts and with only one spacedimension

int

partDdenotes a difference of endpoint evaluations Addition of Eqs (C5ab) gives

1

2partt

int

D

[

v2 + (uprime)2]

=1

2

int

partD

[

v2 + (uprime)2 + 2radicauprimev

]

(C6)

Substitutions with the identities

[

v2 + (uprime)2]

=1

2

[

(v + uprime)2 + (v minus uprime)2]

2uprimev =1

2

[

(v + uprime)2 minus (v minus uprime)2]

(C7)

26

and replacements to recover the original variable v = vradica yield

1

2partt

int

D

[

av2 + (uprime)2]

=1 +

radica

4

int

partD

(radicav + uprime)2 +

1minusradica

4

int

partD

(radicav minus uprime)2 (C8)

From (C8) we deduce that the time-dependent constant C(T ) in (C3) must satisfy

∣

∣

∣

∣

∣

∣

1 +

int T

0

[

1+radica

2

int

partD(radicav + uprime)2 + 1minus

radica

2

int

partD(radicav minus uprime)2

]

dt

uprime(0 middot)2 + av(0 middot)2

∣

∣

∣

∣

∣

∣

le C(T ) (C9)

For periodic boundary conditions we may choose C(T ) = 1 Moreover if a ge 1 anduprime = minusradic

av is specified at partD+ then uprime(t middot)2 + av(t middot)2 decaysStill working on a single interval (subdomain) we now consider the semi-discrete scheme

for (C4) i e (57) with all nonlinear source terms dropped and with v replaced by vradica

Derivation of a formula analogous to (C8) is our first step toward establishing L2 stabilityof the semi-discrete scheme While (57) features vectors for example u(t) taking values atthe Legendre-Gauss-Lobatto nodal points here we work with the numerical solution as apolynomial for example uh(t x) These two representations are related by the Lagrange in-terpolating polynomials for the nodal set here taken to span both the space of test functionsand the space of basis functions Our scheme is

int

Dk

ψparttuh =

int

Dk

ψ(Qh +radicavh) (C10a)

int

Dk

ξparttvh = minusint

Dk

ξprime(radicaQh + vh) +

int

partDk

ξ(radicaQlowast + vlowast) (C10b)

int

Dk

ϕQh =

int

Dk

ϕuprimeh +

int

partDk

ϕ (ulowast minus uh) (C10c)

where ψ ξ and ϕ are polynomial test functions These test functions are arbitrary exceptthat they must be degree-N polynomials In (C10) the variables uh vh and Qh should alsocarry a superscript k but we have suppressed this Derivation of a formula analogous to(C8) is complicated by the fact that Qh is not evolved Nevertheless at a given instant t wecan assemble Qh from (C10c)

Mimicking the calculation (C5b) from the continuum case we first use (C10b) with ξ = vhto write

1

2partt

int

Dk

v2h = minusint

Dk

(radicaQh + vh)v

primeh +

int

partDk

(radicaQlowast + vlowast)vh

= minusint

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2(radicaQlowast + vlowast)vh minus v2h

]

(C11)

The right-hand side of (C5a) is analogous to

1

2partt

int

Dk

Q2h =

int

Dk

QhparttQh (C12)

However since Qh is not evolved the term parttQh must be given a suitable interpretationOn the right hand side of (C10c) only uh u

primeh and u

lowast necessarily depend on time since the

27

test function ϕ need not be time-dependent Furthermore ulowast is explicitly given as a linearcombination of uh as seen in Eq (C21c) below Choosing ϕ = ℓj taking the time derivativeof (C10c) and appealing to the commutivity of mixed partial derivatives we therefore arriveat

int

Dk

ℓjparttQh =

int

Dk

ℓj(parttuh)prime +

int

partDk

ℓj(

(parttu)lowast minus parttuh

)

(C13)

where (parttu)lowast depends on parttuh in precisely the same way that ulowast depends on uh We have

written ℓj rather than ϕ in the last equation to emphasize that the result also holds for anylinear combination of ℓj (for example ϕ) and even for time-dependent combinations SinceQh is itself such a combination we obtain

1

2partt

int

Dk

Q2h =

int

Dk

Qh(parttuh)prime +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

Qh(Qprimeh +

radicavprimeh) +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2((parttu)lowast minus parttuh)Qh +Q2

h

]

(C14)

Addition of (C11) and (C14) gives

1

2partt

int

Dk

(Q2h + v2h) =

1

2

int

partDk

[

Q2h minus v2h + 2(

radicaQlowast + vlowast)vh + 2((parttu)

lowast minus parttuh)Qh

]

(C15)

the aforementioned analog of (C8) This formula holds on a single subdomain Dk and we

now combine multiple copies of it one for each value of k

2 Analysis for multiple intervals

To facilitate combination of (C15) over all k we change notation At every subdomaininterface Ik+12 equiv partDk cap partDk+1 let the superscripts L and R denote field values respectivelytaken from the left and right Then the fields evaluated at Ik+12 which belong to D

k will beuLk+12 v

Lk+12 and Q

Lk+12 while those belonging to D

k+1 will be uRk+12 vRk+12 and Q

Rk+12

However at Ikminus12 the values taken from D

k are uRkminus12 vRkminus12 and QR

kminus12 Note that we

have also replaced the subscript h denoting a numerical solution with k plusmn 12 denotingthe location of the endpoint value of the numerical solution With this notation we define

∆Lα =

1

2

[

(QLα)

2 minus (vLα)2]

+(radic

aQlowastα + vlowastα

)

vLα +[

(parttuα)lowast minus parttu

Lα

]

QLα (C16)

and similarly for ∆Rα The same numerical fluxes appear in both ∆L

α and ∆Rα (ie each

numerical flux takes the same value on either side of an interface) whence fluxes like Qlowastα do

not carry an L or R superscript In terms of these definitions (C15) becomes

1

2partt

int

Dk

(Q2h + v2h) = ∆L

k+12 minus∆Rkminus12 (C17)

28

Summation over all Dk yields

1

2partt

kmaxsum

k=1

int

Dk

(Q2h + v2h) =

kmaxminus1sum

k=1

(

∆Lk+12 minus∆R

k+12

)

+∆Lkmax+12 minus∆R

12

=

kmaxminus1sum

k=1

(

∆Lh minus∆R

h

)∣

∣

Ik+12 +∆Lkmax+12 minus∆R

12 (C18)

We have reverted to h-notation denoting the numerical solution since the LR superscriptsindicate unambiguously the relevant domain used for evaluation at Ik+12

We again seek an estimate of the form

kmaxsum

k=1

(

Qh(T middot)2Dk + avh(T middot)2Dk

)

le C(T )kmaxsum

k=1

(

Qh(0 middot)2Dk + avh(0 middot)2Dk

)

(C19)

that is essentially the same as the one (C3) considered in the continuum case Assume thatthe chosen boundary conditions ensure ∆L

kmax+12 minus ∆R12 is bounded by a time-dependent

constant which does not depend on the numerical parameters N and h (subdomain width)Establishment of stability then amounts to showing that the remaining sum over interfaceterms in (C18) is non-positive whence this remaining sum is consistent with C(T ) le 1although the boundary conditions may give rise to a different bound In fact we will choosethe numerical fluxes such that each individual interface term is non-positive At interfaceIk+12 and in LR notation the jump and average of vh for example are

1

2

(

v+ + vminus)

equiv vh =1

2

(

vLk+12 + vRk+12

)

(C20a)

nminusvminus + n+v+ equiv[[

vh]]

= vLk+12 minus vRk+12 (C20b)

Consider numerical fluxes of the form

Qlowast = Qh minusτQ2

[[

Qh

]]

(C21a)

vlowast = vh minusτv2

[[

vh]]

(C21b)

ulowast = uh minusτu2

[[

uh]]

(C21c)

(parttu)lowast = parttuh minus

τu2

[[

parttuh]]

(C21d)

where (C21c) induces (C21d) and where the penalty parameters τu τv and τQ are realnumbers The fluxes defined in (58) correspond to τu = 1 τv = 1 +

radica and τQ = 0 In

terms of these quantities the kth interface contribution in (C18) is

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

parttuh]]

minus τu2

[[

parttuh]][[

Qh

]]

(C22)

where we have suppressed the k dependence of the right-hand side Now consider the term[[

parttuh]]

Because parttuh and Qh+radicavh are both polynomials of degree N Eq (C10a) implies

29

τu

τ Q

minus1 0 1minus1

minus05

0

05

1

τu

minus1 0 1minus1

minus05

0

05

1

FIG 6 Stable evolutions for the model system For fixed τv = 10minus6 and τv = 1 +radic2

respectively the left and right plots depict stable choices (determined empirically) of τu and τQfor the linear model system (C1) The stable regions are colored black but the jagged edges result

from the discretization of the (τu τQ)-plane

the vector equation parttu = Q +radicav that is pointwise equivalence on the nodal points of

Dk which in turn implies

[[

parttuh]]

=[[

Qh+radicavh]]

Upon substituting this identity into thelast equation we arrive at an expression which features only vh and Qh

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

Qh +radicavh]]

minus τu2

[[

Qh +radicavh]][[

Qh

]]

(C23)

The identities vh[[

vh]]

= 12

[[

v2h]]

and[[

Qh +radicavh]]

=[[

Qh

]]

+radica[[

vh]]

then simplify(C23) to

(∆Lh minus∆R

h )∣

∣

Ik+12 = minusτv2

[[

vh]]2 minus

radica(τu + τQ)

2

[[

Qh

]][[

vh]]

minus τu2

[[

Qh

]]2 (C24)

The role of a penalty parameter is now clear Positive values of τv penalize jumps in vhthrough a negative contribution to the energy Likewise positive values of τu penalizejumps in Qh through a negative contribution to the energy However because the sign of[[

Qh

]][[

vh]]

can be positive or negative only the choice τQ = minusτu yields an expression for(∆L

h minus ∆Rh )|Ik+12 which is manifestly negative for τu ge 0 and τv ge 0 A simple estimate

based on Youngrsquos inequality with ε (that is 2αβ le εminus1α2 + εβ2 where α β ge 0 and ε gt 0)shows that for τQ = 0 the choice τv ge aτu4 also yields stability

Figure 6 depicts certain choices of stable penalty parameters for the linear model systemevolved to tfinal = 1000 (with a = 2 N = 10 and ∆t ≃ 00553) as determined empiricallywith simulations similar to those described in Sec IVA The left plot corresponds to asmall τv = 10minus6 for which the choice τu = 1 τQ = 0 is not stable as expected from thetheoretical analysis However the right plot corresponds to τv = 1 +

radica for which τu = 1

τQ = 0 is stable Motivated by the numerical flux choices (3842) used for the BSSN system

30

(4) we have (as mentioned above) set τu = 1 τv = 1 +radica and τQ = 0 in simulations of

the nonlinear model (53) For the nonlinear model system (53) the theoretically motivatedchoice τQ = minusτu also yields numerically stable evolutions when τu ge 0 and τv ge 0

For the nonlinear systems (4) and (53) we do not attempt a formal stability proofNevertheless the results of this appendix have served as a guide for our choices of penaltyparameters For the BSSN system (4) u v and Q are block indices [cf Eq (6)] Similarto the model problem we have penalized Q with τu = 1 with τv chosen large enough toheuristically overcome the cross-terms of indefinite size that arise from τQ = 0 (we interpretequations like τu = 1 componentwise) An analogous choice ldquoτQ = minusτurdquo for the BSSNsystem might be possible but would be considerably more complicated Indeed such achoice likely entails a matrix of penalty parameters but we do not give the details here

[1] F Pretorius Evolution of Binary Black-Hole Spacetimes Phys Rev Lett 95 121101 (2005)

(4 pages)

[2] F Pretorius Simulation of binary black hole spacetimes with a harmonic evolution scheme

Class Quantum Grav 23 S529-S552 (2006)

[3] M Campanelli C O Lousto P Marronetti and Y Zlochower Accurate Evolutions of Or-

biting Black-Hole Binaries without Excision Phys Rev Lett 96 111101 (2006) (4 pages)

[4] J G Baker J Centrella D-I Choi M Koppitz and J van Meter Gravitational-Wave

Extraction from an Inspiraling Configuration of Merging Black Holes Phys Rev Lett 96

111102 (2006) (4 pages)

[5] M Campanelli C O Lousto and Y Zlochower Last orbit of binary black holes Phys Rev D

73 061501(R) (2006) (5 pages)

[6] F Herrmann I Hinder D Shoemaker and P Laguna Unequal mass binary black hole plunges

and gravitational recoil Class Quantum Grav 24 S33-S42 (2007)

[7] P Diener F Herrmann D Pollney E Schnetter E Seidel R Takahashi J Thornburg and

J Ventrella Accurate Evolution of Orbiting Binary Black Holes Phys Rev Lett 96 121101

(2006) (4 pages)

[8] M A Scheel H P Pfeiffer L Lindblom L E Kidder O Rinne and S A Teukolsky Solving

Einsteinrsquos Equations With Dual Coordinate Frames Phys Rev D 74 104006 (2006)

[9] U Sperhake Binary black-hole evolutions of excision and puncture data Phys Rev D 76

104015 (2007) (20 pages)

[10] B Brugmann J A Gonzalez M Hannam S Husa U Sperhake and W Tichy Calibration

of moving puncture simulations Phys Rev D 77 024027 (2008) (25 pages)

[11] P Marronetti W Tichy B Brugmann J Gonzalez M Hannam S Husa and U Sperhake

Binary black holes on a budget simulations using workstations Class Quantum Grav 24

S43-S58 (2007)

[12] Z B Etienne J A Faber Y T Liu S L Shapiro and T W Baumgarte Filling the holes

Evolving excised binary black hole initial data with puncture techniques Phys Rev D 76

101503(R) (2007) (5 pages)

[13] B Szilagyi D Pollney L Rezzolla J Thornburg and J Winicour An explicit harmonic code

for black-hole evolution using excision Class Quantum Grav 24 S275-S293 (2007)

[14] L Boyle M Kesden and S Nissanke Binary-Black-Hole Merger Symmetry and the Spin

Expansion Phys Rev Lett 100 151101 (2008) (4 pages)

31

[15] M A Scheel M Boyle T Chu L E Kidder K D Matthews and H P Pfeiffer High-

accuracy waveforms for binary black hole inspiral merger and ringdown Phys Rev D 79

024003 (2009) (14 pages)

[16] M Hannam Status of black-hole-binary simulations for gravitational-wave detection

Class Quantum Grav 26 114001 (2009) (21 pages)

[17] I Hinder The current status of binary black hole simulations in numerical relativity

Class Quantum Grav 27 114004 (2010) (19 pages)

[18] B P Abbott et al LIGO The Laser Interferometer Gravitational-Wave Observatory

Rep Prog Phys 72 076901 (2009) (25 pages)

[19] F Acernese et al The status of VIRGO Class Quantum Grav 23 S63-S69 (2006)

[20] F Acernese et al Virgo status Class Quantum Grav 25 184001 (2008) (9 pages)

[21] R Arnowitt S Deser and C W Misner The Dynamics of General Relativity in Gravitation

an introduction to current research edited by L Witten (Wiley New York 1962) Chapter 7

227-265 Also available as gr-qc0405109

[22] J W York Jr Kinematics and Dynamics of General Relativity in Sources of Gravitational

Radiation edited by L L Smarr (Cambridge University Press Cambridge 1979)

[23] L E Kidder M A Scheel and S A Teukolsky Extending the lifetime of 3D black hole

computations with a new hyperbolic system of evolution equations Phys Rev D 64 064017

(2001) (13 pages)

[24] G Calabrese J Pullin O Sarbach and M Tiglio Convergence and stability in numerical

relativity Phys Rev D 66 041501(R) (2002) (4 pages)

[25] S Frittelli and O Reula On the Newtonian Limit of General Relativity Commun Math

Phys 166 221-235 (1994)

[26] M Shibata and T Nakamura Evolution of three-dimensional gravitational waves Harmonic

slicing case Phys Rev D 52 5428-5444 (1995)

[27] Y Choquet-Bruhat and J W York Geometrical well posed systems for the Einstein equations

C R Acad Sci Paris t 321 Serie I 1089-1095 (1995)

[28] A Abrahams A Anderson Y Choquet-Bruhat and J W York Jr Einstein and Yang-Mills

Theories in Hyperbolic Form without Gauge Fixing Phys Rev Lett 75 3377-3381 (1995)

[29] C Bona J Masso E Seidel and J Stela New Formalism for Numerical Relativity

Phys Rev Lett 75 600-603 (1995)

[30] M H P M van Putten and D M Eardley Nonlinear wave equations for relativity

Phys Rev D 53 3056-3063 (1996)

[31] S Frittelli and O A Reula First-Order Symmetric Hyperbolic Einstein Equations with Ar-

bitrary Fixed Gauge Phys Rev Lett 76 4667-4670 (1996)

[32] H Friedrich Hyperbolic reductions for Einsteinrsquos equations Class Quantum Grav 13 1451-

1469 (1996)

[33] F B Estabrook R S Robinson H D Wahlquist Hyperbolic equations for vacuum gravity

using special orthonormal frames Class Quantum Grav 14 1237-1247 (1997)

[34] M S Iriondo E O Leguizamon and O A Reula Einsteinrsquos Equations in Ashtekarrsquos Vari-

ables Constitute a Symmetric Hyperbolic System Phys Rev Lett 79 4732-4735 (1997)

[35] A Anderson Y Choquet-Bruhat and J W York Einstein-Bianchi hyperbolic system for

general relativity Topol Meth Nonlinear Anal 10 353-373 (1997)

[36] T W Baumgarte and S L Shapiro Numerical integration of Einsteinrsquos field equations

Phys Rev D 59 024007 (1998) (7 pages)

[37] M A G Bonilla Symmetric hyperbolic systems for Bianchi equations Class Quantum

12

although our treatment of nonlinear terms will lead to a slight modification of these equations(we return to this issue shortly) The other components of (13a) are treated similarly asare the components of (13b) Recall that for example Qα = αprime Formally using the samedG method to solve for Qα we arrive at the replacement

(

ℓkj (RQα)kh

)

Dk rarrint bk

akdrℓkj (minusQαh + αprime

h)minus (αh minus αlowast) ℓkj

∣

∣

∣

bk

ak (36)

which again features a component αlowast of the numerical flux The auxiliary variables areconstructed and used at each stage of temporal integration We then have

(

ℓkj (RQα)kh

)

Dk = (αh minus αlowast) ℓkj

∣

∣

∣

bk

ak forallj (37)

as the corresponding enforced kth Galerkin condition

B Numerical Flux

To further complete our dG scheme we must specify functional forms for the compo-nents of the numerical flux introduced in the previous section We distinguish between thephysical fluxes (components of f) and the auxiliary fluxes (components of u) arising fromthe definition of the auxiliary variables These choices are not independent as the resultingscheme must be stable and consistent Our choice follows [84] which considered diffusionproblems Additional analysis of this flux choice appears in [66 85]

Let us first consider the numerical fluxes corresponding to the physical fluxes and of theform (33) The numerical flux vector is a function of the system and auxiliary variablesinterior and exterior to a subdomain A common choice for f lowast is

f lowast = fh+τ

2

[[

vh]]

K-component of f lowast f lowastK = fKh+

τ

2

[[

Kh

]]

(38)

where as an example we have also shown the component of f lowast corresponding to the analysisabove Respectively the average and jump across the interface are

fh =1

2

(

f+ + fminus) [[

vh]]

= nminusvminus + n+v+ (39)

Here τ is a position dependent penalty parameter (fixed below) and nminus(n+) is the localoutward pointing normal to the interior (exterior) subdomain The role of τ is to ldquopenalizerdquo(i e yield a negative contribution to the L2 energy norm) jumps across an interface Anappropriate choice of τ will ensure stability and we now provide some motivation for thechoice (41) of τ we make below

Were we treating the fully first-order system (10) the local Lax-Friedrichs flux wouldoften be a preferred choice due to its simplicity [66] In this case the constant ω in thenumerical flux formula Flowast = Fh + 1

2ω[[

Wh

]]

obeys ω ge max∣

∣micro(nablaWF(W ))∣

∣ HereF(W ) = A(u)W the notation micro(middot) indicates the spectral radius of the matrix within andthe max is taken over interior Wminus and exterior W+ states Motivated by (9) we adopt asimilar but simpler prescription substituting the field gradient

nablaWvQA(u)WvQ = A(u) (40)

13

for nablaWF(W ) Precisely we assume the scaling

τ(bk) = τ(ak+1) = τk+12 equiv C middotmax∣

∣micro(

A(u))∣

∣ (41)

where C = O(1) is a constant chosen for stability Larger values of C will result in schemeswith better stability properties whereas too large a value will impact the CFL conditionAt the interface point Ik+12 equiv D

k cap Dk+1 the vector uh has two representations uminus at bk

and u+ at ak+1 The max in (41) is taken over the corresponding two sets of field speedsMore precisely the speeds in Table I are computed for both uminus and u+ and the maximumtaken over all resulting speeds For the auxiliary variables a penalized central flux is usedThe definition with one representative component is

ulowast = uh minus1

2

[[

uh]]

α-component of ulowast αlowast = αh minus1

2

[[

αh

]]

(42)

with similar expressions for the remaining componentsWe stress the following point Since the interior coupling between subdomains is achieved

through the numerical flux forms (41) and (42) the inverse transformation (17) expressingthe fundamental fields in terms of the characteristic fields is not required to achieve thiscoupling On the other hand imposition of physical boundary conditions may still rely on(17) since this transformation allows one to fix only incoming characteristic modes

C Matrix form of the semi-discrete equations

Let us introduce the kth mass and stiffness matrices

Mkij =

int bk

akdrℓki (r)ℓ

kj (r) Sk

ij =

int bk

akdrℓki (r)ℓ

kjprime(r) (43)

These matrices belong to Dk and the corresponding matrices defined on the reference interval[minus1 1] are

Mij =

int 1

minus1

duℓi(u)ℓj(u) Sij =

int 1

minus1

duℓi(u)ℓprimej(u) (44)

where ℓj(u) is the jth Lagrange polynomial determined by the LGL nodes uj on [minus1 1]These matrices are related by Mk

ij =12(bk minus ak)Mij and S

kij = Sij whence only the reference

matrices require computation and storageWe will use the matrices Mk and Sk in obtaining an ODE system from (26) and (35)

Towards this end we first approximate the nonlinear terms (products and quotients) in(26) by degree-N interpolating polynomials Such approximations are achieved throughpointwise representations For example (QαQχgrr)h appears in (26) and is expressed inthe following way [cf footnote 2]

(

QαQχ

grr

)

h

(t r) =

(

QαhQχh

grrh

)

(t r) rarrNsum

j=0

Qαh(t rkj )Qχh(t r

kj )

grrh(t rkj )

ℓkj (r) (45)

Note that the expressions on the right and left are not equivalent due to aliasing error [65]Our vector notation for this replacement will be

(

QαQχ

grr

)

h

(t r) rarr(

QαQχ

grr

)

(t)Tℓk(r) (46)

14

Operations among bold variables are always performed pointwise Making similar replace-ments for all terms in (26) and then carrying out the integrations in (35) which bring inMk and Sk we arrive at

parttK = βrDK minus χDQα

grr

+1

2

χQgrrQα

g2rr

minusχQgθθ

Qα

grrgθθ

+1

2

QαQχ

grr

+3

2

αA2rr

g2rr

+1

3αK2 +Mminus1ℓk (fKh minus f lowast

K)∣

∣

∣

bk

ak (47)

where we have again suppressed the superscript k on all terms except ℓk(r) and the sub-script h is dropped on all boldfaced variables As described in [66] the spectral collocationderivative matrix

(Dk)ij =dℓkjdr

∣

∣

∣

∣

∣

r=rki

(48)

can also be expressed as Dk = (Mk)minus1Sk which appears above Eight other semi-discreteevolution equations are similarly obtained with nine in total (one for each component ofWuv) Additionally we have

Qα = Dα+Mminus1ℓk(αlowast minusαh)∣

∣

∣

bk

ak (49)

as one of the auxiliary equations with five in total (one for each component of Q = WQQ)

D Filtering

Like other nodal (pseudospectral) methods our scheme may suffer from instabilitiesdriven by aliasing error [65] Filtering is a simple yet robust remedy To filter a solu-tion component such as χ we use the modal (as opposed to nodal) representation of thesolution

χkh(t r) =

Nsum

j=0

χ(t rkj )ℓkj (r) =

Nsum

j=0

χkj (t)Pj(r) (50)

where Pj(r) is the jth Legendre polynomial Let ηj = jN and define the filter function

σ(ηj) =

1 for 0 le ηj le NcN

exp(

minus ǫ(

ηjminusNcN

1minusNcN

)2s)

for NcN le ηj le 1(51)

At each timestep we modify our solution component according to

χkh rarr

(

χkh

)filtered=

Nsum

j=0

σ(ηj)χkj (t)Pj(r) (52)

Evidently the modification only affects the top N minus Nc modes and is sufficient to controlthe type of weak instability driven by aliasing [66] The numerical parameters Nc and ǫ areproblem dependent For our simulations we have taken ǫ ≃ minuslog(εmach) = 36 where εmach

is machine accuracy in double precision

15

E Model system

To better illustrate the basic properties of our method we consider a toy model Namelythe following spatially second-order system

parttu = uprime + av minus u3 + g(t x) (53a)

parttv = uprimeprime + vprime minus (u+ v)(uprime)2 + v2u2 + h(t x) (53b)

where a ge 1 is constant and g and h are analytic source terms to be specified In contrastto (6) here u v and Q = uprime are scalars rather than vectors System (53) admits a first-order reduction in which uprime is defined as an extra variable Since this first-order reductionis strongly hyperbolic the spatially second-order system (53) is also strongly hyperbolic byone of the definitions considered in [75] The characteristic fields Xplusmn and speeds microplusmn are

X+ =radicav minus uprime micro+ =

radicaminus 1 Xminus =

radicav + uprime microminus = minus(

radica+ 1) (54)

To construct a local dG scheme for this system we first rewrite it as

parttu = Q+ av minus u3 + g(t x) (55a)

parttv = Qprime + vprime minus (u+ v)Q2 + v2u2 + h(t x) (55b)

Q = uprime (55c)

Evidently f = minus(Q + v) is the v-component of the physical flux vector

F (vQ) equiv(

Fu

Fv

)

=

(

0

f

)

(56)

Note that F has the same structure as (u v)T Borrowing from the presentation for theBSSN system we write the analogous semidiscrete scheme on each subdomain D

k for themodel system

parttu = Q+ av minus u3 + g(t) (57a)

parttv = DQ+Dv minus (u+ v)Q2 + v2u2 + h(t) +Mminus1ℓk(fh minus f lowast)∣

∣

∣

bk

ak (57b)

Q = Du+Mminus1ℓk(ulowast minus uh)∣

∣

∣

bk

ak (57c)

Here we have suppressed the subinterval label k from all variables except for the vectorℓk of Lagrange polynomial values Moreover following the guidelines discussed above thenumerical fluxes are given by

f lowast = fh+1 +

radica

2

[[

vh]]

ulowast = uh minus1

2

[[

uh]]

(58)

Appendix C analyzes the stability of our scheme for a more general numerical flux choiceas applied to (53) with the nonlinear and source terms dropped

IV RESULTS FROM NUMERICAL SIMULATIONS

This section presents results found by numerically solving both the model system (53)and BSSN system (4) with the dG scheme presented in Sec III

16

10minus10

10minus6

10minus2

∆uinfin

10minus10

10minus6

10minus2

∆vinfin

0 500 1000 1500 2000 2500 3000 3500 400010

minus10

10minus6

10minus2

∆(partxu)infin

t

N = 3N = 6N = 9N = 12

FIG 1 Spectral convergence of fields for model PDE Respectively for N = 3 6 9 12

a timestep of ∆t = 00578 00178 00084 00049 has been chosen for stability and accuracy In the

title headings for example ∆u equiv unumer minus uexact

A Simulations of the model system

The semi-discrete scheme (57) has been integrated with the classical fourth-order Runge-Kutta method When integrating this system we have first constructed Q at each Runge-Kutta stage and then substituted into the evolution equations (57ab) for u and v Theproblem has been solved on a computational domain [0 4π] comprised of two subdomainswith a timestep chosen small enough for stability The initial data has been taken from thefollowing exact solution to (53)

uprimeexact(t x) =1

2

[

sin(xminus microminust)minus sin(xminus micro+t)]

(59a)

vexact(t x) =1

2radica

[

sin(xminus microminust) + sin(xminus micro+t)]

(59b)

g(t x) = u3exact (59c)

h(t x) = (uexact + vexact)(uprimeexact)

2 minus v2exactu2exact (59d)

where the speeds microplusmn are found in (54) Specification of the boundary condition at a physicalendpoint amounts to choosing the external state for at the endpoint We have considered

17

8 16 32 64 128

10minus3

10minus2

10minus1

N

∆t

max stable ∆t8N2

FIG 2 Scaling of maximum stable ∆t with N for model PDE

two possibilities (i) the analytic state (Q+ v+) = (Qexact vexact) and (ii) an upwind stateFor example at x = 4π the upwind state is4

Q+ = Qupwind =1

2

[

(Xminus)exact minus (X+)numer

]

v+ = vupwind =1

2radica

[

(Xminus)exact + (X+)numer

]

(60)Either choice of (Q+ v+) leads to similar results and the plots here correspond to theanalytic state Figure 1 clearly shows spectral convergence with increasing polynomial orderN across all fields for the case a = 2 Other values of a including a = 1 for which X+

is a static characteristic field have also been considered with similar results Appendix Cdemonstrates that our proposed scheme for the system (57) with nonlinear and source termsdropped is stable in a semi-discrete sense Nevertheless the fully discrete scheme obtainedvia temporal discretization by the fourth-order Runge-Kutta method is still subject to thestandard absolute stability requirement Namely if microh is any eigenvalue corresponding tothe (linearized) discrete spatial operator then a necessary condition for stability is thatmicroh∆t lies in absolute stability region for Runge-Kutta 4 We here show empirically that theassociated timestep restriction scales like Nminus2 ie ∆t = O(Nminus2) for stability We note thatsuch scaling is welcome in light of the second-order spatial operators which appear in thesystem and suggest a possible worse scaling like Nminus4 Fig 2 plots the maximum stabletimestep for a range of N demonstrating the Nminus2 scaling in line with behavior known fromanalysis of first-order systems [66] This scaling also holds for the BSSN system

4 We remind the reader that unfortunately the plusmn on Xplusmn means something different than the plusmn indicating

exteriorinterior dG states [cf footnote 3]

18

B Simulations of the BSSN system

This subsection documents results for simulations of the unit-mass-parameter (M = 1)Schwarzschild solution (B9) expressed in terms of ingoing Kerr-Schild coordinates Sincethe solution is stationary temporal integration of the semi-discrete scheme has been carriedout with the forward Euler method The r-coordinate domain [04 34] has been split into3 equally spaced subdomains and we have set η = 10 λ = 01 and C = 2 [cf Eq (41)]For all simulations ∆t has been chosen for stability With the chosen λ the inner physicalboundary rmin = 04 is an excision surface At each timestep we have applied an (order2s = 20) exponential filter on the top two-thirds of the modal coefficient set for all fieldsexcept for grr and gθθ For stability we have empirically observed that grr and gθθ must not

be filtered A detailed understanding of this is still lackingIssues related to physical boundary conditions are similar to the one encountered in

Sec IVA for the model problem Similar to before we have retained Eqs (3842) as thechoice of numerical flux even at the endpoints Therefore at an endpoint the specificationof the boundary condition amounts to the choice W+ of external state We have typicallychosen the inner boundary of the radial domain as an excision boundary and in this caseW+ = Wminus is enforced at the inner physical boundary At the outer physical boundaryfor W+ we have again considered two choices (i) Wexact and (ii) Wupwind To enforcechoice (ii) the inverse transformation (17) must be used with incoming characteristic fieldsfixed to their exact values similar to (60) We have tried various versions of choice (ii)and in all cases the resulting simulations have been unstable We therefore present resultscorresponding to choice (i) Although the choice of an analytical external state Wexact

at the outer boundary is stable for our problem such a boundary condition is unlikelyto generalize to more complicated scenarios involving dynamical fields Indeed the issueof outer boundary conditions for the BSSN system is an active area of research with aproper treatment requiring fixation of incoming radiation control of the constraints andspecification of gauge (see Ref [86] for a recent analysis)

For BSSN simulations our main diagnostic is to monitor the Hamiltonian momentumand conformal connection constraints Figure 3 depicts long-time histories of constraintviolations whereas Figs 4 and 5 depict long-time error histories for the individual BSSNfield components From the middle plot in Fig 5 we infer that up to the indicated numericalerror the factor g sin4 θ = grr(gθθ)

2 remains at its initial fixed profile r4 throughout theevolution These figures indicate that the proposed scheme is stable for long times andexhibits spectral converge with increased polynomial order N Similar results are recoveredfromM = 0 Minkowski initial data The stability documented in these plots does not appearto rely on inordinate parameter tuning For example with the fixed parameters describedabove we obtain similar plots if we individually vary (i) rmin over 0325 035 04 0475(values still corresponding to an excision surface for the given choice of λ) (ii) η over1 3 7 10 (iii) s over 8 9 10 With the polynomial order N ranging over 23 26 29 31both stability and qualitatively similar exponential convergence is achieved with a singlesubdomain Likewise adoption of a larger coordinate domain with more subdomains doesnot impact our results Finally we have considered the addition of random noise to all fieldcomponents at the initial time Precisely with the system component χ as an example wehave set

χ(0) 7rarr χ(0) + δχ(0) (61)

where each component (nodal value) of δχ(0) is 10minus5 times a random variable drawn from a

19

10minus9

10minus7

10minus5

10minus3

Hinfin

10minus9

10minus7

10minus5

10minus3

Mrinfin

0 2000 4000 6000 8000 10000

10minus9

10minus7

10minus5

10minus3

Grinfin

t

N = 11N = 14N = 17N = 20

FIG 3 Spectral convergence of constraint violations for M = 1 Kerr-Schild ini-

tial data Respectively for N = 11 14 17 19 a timestep of ∆t ≃ 00041 00026 00018 00013

has been chosen for stability and accuracy

standard normal distribution Such perturbed initial data also gives rise to stable evolutions

V CONCLUSION

We have introduced a discontinuous Galerkin method for solving the spherically reducedBSSN system with second-order spatial operators Our scheme shares similarities withother discontinuous Galerkin methods that use local auxiliary variables to handle high-orderspatial derivatives [66 77ndash79 82 84 85] and which have typically been applied to eitherelliptic parabolic or mixed type problems The key ingredient of a stable dG scheme is anappropriate choice of numerical flux and our particular choice has been motivated by theanalysis presented in Appendix C When used to evolve the Schwarzschild solution in Kerr-Schild coordinates our numerical implementation of the BSSN system (4) is robustly stableand converges to the analytic solution exponentially with increased polynomial order Byapproximating the spatially second-order form of the BSSN system we have not introducedextra fields which are evolved Evolved auxiliary fields result in new constraints which mayspoil stability Our main goal has been stable evolution of the spherically reduced BSSNsystem as a first step towards understanding how a discontinuous Galerkin method might be

20

10minus9

10minus7

10minus5

∆Arrinfin

10minus9

10minus7

10minus5

∆Kinfin

0 2000 4000 6000 8000 10000

10minus9

10minus7

10minus5

max(∆Brinfin ∆Γrinfin)

t

N = 11N = 14N = 17N = 20

FIG 4 Spectral convergence of solution for M = 1 Kerr-Schild initial data

Timestep choices are described in the caption for Fig 3 In the title headings for example

∆Arr equiv (Arr)numer minus (Arr)exact

applied to the full BSSN system Towards that goal we now discuss treatment of singularitiesand generalization of the described dG method to higher space dimension

To deal with the fixed Schwarzschild singularity we have used excision which is easy inthe context of the spherically reduced BSSN system However excision for the binary blackhole problem in full general relativity requires attention to the technical challenge of hori-zon tracking State-of-the-art BSSN codes avoid such complication relying instead on themoving-puncture technique While the moving-puncture technique does involve mild centralsingularities it may still prove amenable to spectral methods Indeed spectral methods fornon-smooth problems is well-developed in both theory and for complex applications Sincethe moving-puncture technique can be performed in spherical symmetry [60] a first-steptoward a spectral moving-puncture code would be to implement a moving puncture withthe nodal dG method described here Such an implementation may adopt Legendre-Gauss-Radau nodes on the innermost subdomain thereby ensuring that the physical singularitydoes not lie on a nodal point (in much the same way finite difference codes use a staggeredgrid) Beyond traditional excision and moving punctures one might also retain a smoothproblem for which spectral methods are especially well-suited via the turducken approach[87 88] to singularities

21

10minus11

10minus9

10minus7

10minus5

∆χinfin

10minus11

10minus9

10minus7

10minus5

max(∆grrinfin ∆gθθinfin)

0 2000 4000 6000 8000 10000

10minus11

10minus9

10minus7

10minus5

max(∆αinfin ∆βrinfin)

t

N = 11N = 14N = 17N = 20

FIG 5 Spectral convergence of solution violations for M = 1 Kerr-Schild initial

data See the caption of Fig 4 for details

Discontinuous Galerkin methods for hyperbolic problems in two and three space dimen-sions are well-developed A generalization of the method described here to three-dimensionsand the full BSSN system would likely rely on an unstructured mesh Appropriate local poly-nomial expansions for the subelements are well-understood as are choices for the numericalfluxes which would now live on two-dimensional faces rather than single points Whether ornot it would ultimately prove successful generalization of our dG method to a higher dimen-sion would rely on an established conceptual framework Further computational advancesof relevance to a generalization of our dG method to the full BSSN system (possibly includ-ing matter) may include mesh hp-adaptivity local timestepping shock capturing and slopelimiting techniques [66] Moreover recent work [89] indicates that enhanced performancewould be expected were our scheme implemented on graphics processor units

VI ACKNOWLEDGMENTS

We thank Nick Taylor for discussions about treating second-order operators with spectralmethods Benjamin Stamm for helping polish up a few parts of Appendix C Khosro Shah-bazi for discussions on LDG and IP methods We also acknowledge helpful conversationswith David Brown and Manuel Tiglio about the BSSN system and previous work on its nu-

22

merical implementation We gratefully acknowledge funding through grants DMS 0554377and DARPAAFOSR FA9550-05-1-0108 to Brown University and NSF grant PHY 0855678to the University of New Mexico

Appendix A Hyperbolicity of the first-order system

This appendix analyzes the matrix A(u) appearing in (10) in order to construct thecharacteristic fields (14) In matrix form the sector (8) of the principal part of (10) reads asfollows

partt

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

=

βr 0 minus 4λα3grr

0 0 λβr

6(grr)2λβr

3gθθgrr0 4λ

3grr

0 βr 0 23grrαχ

13α minus αχ

3grr

αχ3gθθ

minus23χ 0

0 0 βr 0 0 0 0 minus χgrr

0

0 0 minus 4α3grr

βr 0 βr

6(grr)2βr

3gθθgrr0 4

3grr

0 0 23αχ 0 βr minus βrχ

3grrminus2βrχ

3gθθ0 minus2

3χ

0 minus2α 0 0 0 23βr minus2grrβr

3gθθ0 4

3grr

0 gθθαgrr

0 0 0 minusgθθβr

3grr13βr 0 minus2

3gθθ

0 0 minus2α 0 0 0 0 βr 0

34

0 0 0 0 0 0 0 βr

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

prime

(A1)which defines the matrix A(u) appearing in (9) and so also the matrix A(u) in (10) Notethat in the last equation the matrix within the square brackets is minusA(u) For certainconfigurations of u and λ the system (10) is strongly hyperbolic [76] that is A(u) hasa complete set of eigenvectors and real eigenvalues Indeed five eigenpairs of A(u) aretrivially recovered upon inspection of A(u)rsquos leading 5times5 diagonal block These correspondto eigenvalue 0 and the left eigenspace ξj = eTj 1 le j le 5 where ej are the canonical

basis vectors Since each component of u arises as eTj W each is also a characteristic field

The remaining nine eigenpairs are determined by A(u) The eigenvalues of A(u) are

micro1 = 0 micro23 = minusβr microplusmn4 = minusβr plusmn

radic

2αχ

grr microplusmn

5 = minusβr plusmn α

radic

χ

grr microplusmn

6 = minusβr plusmnradic

λ

grr

(A2)

23

and the corresponding left eigenvectors are

x1 = (0 0 0 0 0 gθθ 2grr 0 0) (A3a)

x2 =

(

0 0 0 grr2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3b)

x3 =

(

grrλ 0 0 0

2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3c)

xplusmn4 =

(

0 0plusmnradic

2αgrrχ

0 0 0 0 1 0

)

(A3d)

xplusmn5 =

(

0∓ 3radicgrrχ

plusmn2

radic

grrχ 2grr

1

χminus 1

grr1

gθθ 0 0

)

(A3e)

xplusmn6 =

(

minus3

4

grrλ 0plusmn α

radicλgrr

(2αχminus λ) 0 0minus βr

8(βrgrr ∓radicλgrr)

minus βrgrr

4gθθ(βrgrr ∓radicλgrr)

αχ

(2αχminus λ)plusmnradic

grrλ

)

(A3f)

where for example xplusmn5 A(u) = microplusmn5 x

plusmn5 Assuming that grr gθθ χ and α are everywhere strictly

positive the eigenvalues are real and the eigenvectors are linearly independent providedthat (15) holds These eigenvectors are easily extended to eigenvectors of A(u) e g asxplusmn6 rarr (01times5 x

plusmn6 ) Then for example the characteristic field

Xplusmn6 equiv (01times5 x

plusmn6 )W = xplusmn6WvQ (A4)

and similarly Xplusmnj = xplusmnj WvQ for j = 4 5 and Xk = xkWvQ for k = 1 2 3 The characteristic

speeds for these fields are microk and microplusmnj With this convention the speeds listed in Table I

correspond to the Xk and Xplusmnj in (14)

Appendix B Schwarzschild solution in conformal Kerr-Schild coordinates

In Kerr-Schild coordinates here the system directly related to incoming Eddington-Finkelstein null coordinates the line element for the Schwarzschild solution reads

ds2 = minusα2dt2 + (1 + 2MR)(dR + βRdt)2 +R2dθ2 +R2 sin2 θdφ2 (B1)

where R is the area radius α = (1 + 2MR)minus12 is the lapse and βR = 2M(R + 2M) isthe shift vector The physical spatial metric gab is the spatial part of this line element

To define the corresponding solution to the BSSN system we use equation gab = χgab todefine the following relationship between line elements

dr2 + r2(dθ2 + sin2 θdφ2) = χ[(1 + 2MR)dR2 +R2dθ2 +R2 sin2 θdφ2] (B2)

so that

χ

(

1 +2M

R

)(

dR

dr

)2

= 1 χR2 = r2 (B3)

Then we have(

1 +2M

R

)12dR

R=dr

r (B4)

24

with integration yielding

r =R

4

(

1 +

radic

1 +2M

R

)2

e2minus2radic

1+2MR (B5)

where the constant of integration has been chosen so that the R r rarr infin limits are consistentThe second relation in (B3) shows that

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR =rArr χminus4 =2eradic

1+2MRminus1

1 +radic

1 + 2MR (B6)

The extrinsic curvature tensor is specified by the expression forK given in (B9h) the identityK = KR

R + 2Kθθ and

Kθθ =

(

1 +2M

R

)minus122M

R2 (B7)

Since KRR = Kr

r we compute that

Krr = K minus 2Kθ

θ = minus(

1 +2M

R

)minus12(R +M

R + 2M

)

2M

R2 (B8)

Next since Krr = grrKrr = χminus1Kr

r we have Krr = Arr+

13grrK This implies Arr = Kr

r minus 13K

from which we get (B9g) In all we have

α =

(

1 +2M

R

)minus12

(B9a)

βr = βR dr

dR= χ12

(

1 +2M

R

)minus122M

R(B9b)

grr = 1 (B9c)

gθθ = r2 = χR2 (B9d)

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR (B9e)

Br = 0 (B9f)

Arr = minus(

1 +2M

R

)minus124M

3R2

(

2R + 3M

R + 2M

)

(B9g)

K =

(

1 +2M

R

)minus32(

1 +3M

R

)

2M

R2(B9h)

Γr = minus2

r= minus 2

χ12R (B9i)

To differentiate these expressions with respect to r we use the identity

dR

dr= χminus12

(

1 +2M

R

)minus12

(B10)

along with the chain rule

25

Appendix C Stability of the model system

The following stability analysis for the model system (53) has been inspired by [78 79]After dropping all nonlinear source terms the system (53) becomes

parttu = uprime + av (C1a)

parttv = uprimeprime + vprime (C1b)

This section analyzes the stability of (C1) considering both the continuum system itself aswell as its semi-discrete dG approximation The latter analysis offers some insight into theempirically observed stability of our dG scheme for the spherically reduced BSSN equations

1 Analysis for a single interval

Throughout we work with the L2 inner product and norm

(f g)D =

int

D

fg fD =radic

(f f)D (C2)

where D is the spatial coordinate interval (here D may represent a subdomain Dk or the

whole domain Ω) and we have suppressed all integration measures For the continuummodel we will establish the following estimate

uprime(T middot)2D+ av(T middot)2

Dle C(T )

(

uprime(0 middot)2D+ av(0 middot)2

D

)

(C3)

where the time-dependent constant C(T ) is determined solely by the choice of boundaryconditions To show (C3) we first change variables with v =

radicav thereby rewriting (C1)

in the following symmetric form

parttu = uprime +radicav (C4a)

parttv =radicauprimeprime + vprime (C4b)

Equations (C4ab) then imply

1

2partt

int

D

(uprime)2 =

int

D

uprime(uprimeprime +radicavprime) =

int

D

radicauprimevprime +

1

2

int

partD

(uprime)2 (C5a)

1

2partt

int

D

(v)2 =

int

D

v(radicauprimeprime + vprime) = minus

int

D

radicauprimevprime +

1

2

int

partD

(v2 + 2radicauprimev) (C5b)

Here vvprime and uprimeuprimeprime have been expressed as exact derivatives and then integrated to boundaryterms the second equation employs an extra integration by parts and with only one spacedimension

int

partDdenotes a difference of endpoint evaluations Addition of Eqs (C5ab) gives

1

2partt

int

D

[

v2 + (uprime)2]

=1

2

int

partD

[

v2 + (uprime)2 + 2radicauprimev

]

(C6)

Substitutions with the identities

[

v2 + (uprime)2]

=1

2

[

(v + uprime)2 + (v minus uprime)2]

2uprimev =1

2

[

(v + uprime)2 minus (v minus uprime)2]

(C7)

26

and replacements to recover the original variable v = vradica yield

1

2partt

int

D

[

av2 + (uprime)2]

=1 +

radica

4

int

partD

(radicav + uprime)2 +

1minusradica

4

int

partD

(radicav minus uprime)2 (C8)

From (C8) we deduce that the time-dependent constant C(T ) in (C3) must satisfy

∣

∣

∣

∣

∣

∣

1 +

int T

0

[

1+radica

2

int

partD(radicav + uprime)2 + 1minus

radica

2

int

partD(radicav minus uprime)2

]

dt

uprime(0 middot)2 + av(0 middot)2

∣

∣

∣

∣

∣

∣

le C(T ) (C9)

For periodic boundary conditions we may choose C(T ) = 1 Moreover if a ge 1 anduprime = minusradic

av is specified at partD+ then uprime(t middot)2 + av(t middot)2 decaysStill working on a single interval (subdomain) we now consider the semi-discrete scheme

for (C4) i e (57) with all nonlinear source terms dropped and with v replaced by vradica

Derivation of a formula analogous to (C8) is our first step toward establishing L2 stabilityof the semi-discrete scheme While (57) features vectors for example u(t) taking values atthe Legendre-Gauss-Lobatto nodal points here we work with the numerical solution as apolynomial for example uh(t x) These two representations are related by the Lagrange in-terpolating polynomials for the nodal set here taken to span both the space of test functionsand the space of basis functions Our scheme is

int

Dk

ψparttuh =

int

Dk

ψ(Qh +radicavh) (C10a)

int

Dk

ξparttvh = minusint

Dk

ξprime(radicaQh + vh) +

int

partDk

ξ(radicaQlowast + vlowast) (C10b)

int

Dk

ϕQh =

int

Dk

ϕuprimeh +

int

partDk

ϕ (ulowast minus uh) (C10c)

where ψ ξ and ϕ are polynomial test functions These test functions are arbitrary exceptthat they must be degree-N polynomials In (C10) the variables uh vh and Qh should alsocarry a superscript k but we have suppressed this Derivation of a formula analogous to(C8) is complicated by the fact that Qh is not evolved Nevertheless at a given instant t wecan assemble Qh from (C10c)

Mimicking the calculation (C5b) from the continuum case we first use (C10b) with ξ = vhto write

1

2partt

int

Dk

v2h = minusint

Dk

(radicaQh + vh)v

primeh +

int

partDk

(radicaQlowast + vlowast)vh

= minusint

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2(radicaQlowast + vlowast)vh minus v2h

]

(C11)

The right-hand side of (C5a) is analogous to

1

2partt

int

Dk

Q2h =

int

Dk

QhparttQh (C12)

However since Qh is not evolved the term parttQh must be given a suitable interpretationOn the right hand side of (C10c) only uh u

primeh and u

lowast necessarily depend on time since the

27

test function ϕ need not be time-dependent Furthermore ulowast is explicitly given as a linearcombination of uh as seen in Eq (C21c) below Choosing ϕ = ℓj taking the time derivativeof (C10c) and appealing to the commutivity of mixed partial derivatives we therefore arriveat

int

Dk

ℓjparttQh =

int

Dk

ℓj(parttuh)prime +

int

partDk

ℓj(

(parttu)lowast minus parttuh

)

(C13)

where (parttu)lowast depends on parttuh in precisely the same way that ulowast depends on uh We have

written ℓj rather than ϕ in the last equation to emphasize that the result also holds for anylinear combination of ℓj (for example ϕ) and even for time-dependent combinations SinceQh is itself such a combination we obtain

1

2partt

int

Dk

Q2h =

int

Dk

Qh(parttuh)prime +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

Qh(Qprimeh +

radicavprimeh) +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2((parttu)lowast minus parttuh)Qh +Q2

h

]

(C14)

Addition of (C11) and (C14) gives

1

2partt

int

Dk

(Q2h + v2h) =

1

2

int

partDk

[

Q2h minus v2h + 2(

radicaQlowast + vlowast)vh + 2((parttu)

lowast minus parttuh)Qh

]

(C15)

the aforementioned analog of (C8) This formula holds on a single subdomain Dk and we

now combine multiple copies of it one for each value of k

2 Analysis for multiple intervals

To facilitate combination of (C15) over all k we change notation At every subdomaininterface Ik+12 equiv partDk cap partDk+1 let the superscripts L and R denote field values respectivelytaken from the left and right Then the fields evaluated at Ik+12 which belong to D

k will beuLk+12 v

Lk+12 and Q

Lk+12 while those belonging to D

k+1 will be uRk+12 vRk+12 and Q

Rk+12

However at Ikminus12 the values taken from D

k are uRkminus12 vRkminus12 and QR

kminus12 Note that we

have also replaced the subscript h denoting a numerical solution with k plusmn 12 denotingthe location of the endpoint value of the numerical solution With this notation we define

∆Lα =

1

2

[

(QLα)

2 minus (vLα)2]

+(radic

aQlowastα + vlowastα

)

vLα +[

(parttuα)lowast minus parttu

Lα

]

QLα (C16)

and similarly for ∆Rα The same numerical fluxes appear in both ∆L

α and ∆Rα (ie each

numerical flux takes the same value on either side of an interface) whence fluxes like Qlowastα do

not carry an L or R superscript In terms of these definitions (C15) becomes

1

2partt

int

Dk

(Q2h + v2h) = ∆L

k+12 minus∆Rkminus12 (C17)

28

Summation over all Dk yields

1

2partt

kmaxsum

k=1

int

Dk

(Q2h + v2h) =

kmaxminus1sum

k=1

(

∆Lk+12 minus∆R

k+12

)

+∆Lkmax+12 minus∆R

12

=

kmaxminus1sum

k=1

(

∆Lh minus∆R

h

)∣

∣

Ik+12 +∆Lkmax+12 minus∆R

12 (C18)

We have reverted to h-notation denoting the numerical solution since the LR superscriptsindicate unambiguously the relevant domain used for evaluation at Ik+12

We again seek an estimate of the form

kmaxsum

k=1

(

Qh(T middot)2Dk + avh(T middot)2Dk

)

le C(T )kmaxsum

k=1

(

Qh(0 middot)2Dk + avh(0 middot)2Dk

)

(C19)

that is essentially the same as the one (C3) considered in the continuum case Assume thatthe chosen boundary conditions ensure ∆L

kmax+12 minus ∆R12 is bounded by a time-dependent

constant which does not depend on the numerical parameters N and h (subdomain width)Establishment of stability then amounts to showing that the remaining sum over interfaceterms in (C18) is non-positive whence this remaining sum is consistent with C(T ) le 1although the boundary conditions may give rise to a different bound In fact we will choosethe numerical fluxes such that each individual interface term is non-positive At interfaceIk+12 and in LR notation the jump and average of vh for example are

1

2

(

v+ + vminus)

equiv vh =1

2

(

vLk+12 + vRk+12

)

(C20a)

nminusvminus + n+v+ equiv[[

vh]]

= vLk+12 minus vRk+12 (C20b)

Consider numerical fluxes of the form

Qlowast = Qh minusτQ2

[[

Qh

]]

(C21a)

vlowast = vh minusτv2

[[

vh]]

(C21b)

ulowast = uh minusτu2

[[

uh]]

(C21c)

(parttu)lowast = parttuh minus

τu2

[[

parttuh]]

(C21d)

where (C21c) induces (C21d) and where the penalty parameters τu τv and τQ are realnumbers The fluxes defined in (58) correspond to τu = 1 τv = 1 +

radica and τQ = 0 In

terms of these quantities the kth interface contribution in (C18) is

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

parttuh]]

minus τu2

[[

parttuh]][[

Qh

]]

(C22)

where we have suppressed the k dependence of the right-hand side Now consider the term[[

parttuh]]

Because parttuh and Qh+radicavh are both polynomials of degree N Eq (C10a) implies

29

τu

τ Q

minus1 0 1minus1

minus05

0

05

1

τu

minus1 0 1minus1

minus05

0

05

1

FIG 6 Stable evolutions for the model system For fixed τv = 10minus6 and τv = 1 +radic2

respectively the left and right plots depict stable choices (determined empirically) of τu and τQfor the linear model system (C1) The stable regions are colored black but the jagged edges result

from the discretization of the (τu τQ)-plane

the vector equation parttu = Q +radicav that is pointwise equivalence on the nodal points of

Dk which in turn implies

[[

parttuh]]

=[[

Qh+radicavh]]

Upon substituting this identity into thelast equation we arrive at an expression which features only vh and Qh

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

Qh +radicavh]]

minus τu2

[[

Qh +radicavh]][[

Qh

]]

(C23)

The identities vh[[

vh]]

= 12

[[

v2h]]

and[[

Qh +radicavh]]

=[[

Qh

]]

+radica[[

vh]]

then simplify(C23) to

(∆Lh minus∆R

h )∣

∣

Ik+12 = minusτv2

[[

vh]]2 minus

radica(τu + τQ)

2

[[

Qh

]][[

vh]]

minus τu2

[[

Qh

]]2 (C24)

The role of a penalty parameter is now clear Positive values of τv penalize jumps in vhthrough a negative contribution to the energy Likewise positive values of τu penalizejumps in Qh through a negative contribution to the energy However because the sign of[[

Qh

]][[

vh]]

can be positive or negative only the choice τQ = minusτu yields an expression for(∆L

h minus ∆Rh )|Ik+12 which is manifestly negative for τu ge 0 and τv ge 0 A simple estimate

based on Youngrsquos inequality with ε (that is 2αβ le εminus1α2 + εβ2 where α β ge 0 and ε gt 0)shows that for τQ = 0 the choice τv ge aτu4 also yields stability

Figure 6 depicts certain choices of stable penalty parameters for the linear model systemevolved to tfinal = 1000 (with a = 2 N = 10 and ∆t ≃ 00553) as determined empiricallywith simulations similar to those described in Sec IVA The left plot corresponds to asmall τv = 10minus6 for which the choice τu = 1 τQ = 0 is not stable as expected from thetheoretical analysis However the right plot corresponds to τv = 1 +

radica for which τu = 1

τQ = 0 is stable Motivated by the numerical flux choices (3842) used for the BSSN system

30

(4) we have (as mentioned above) set τu = 1 τv = 1 +radica and τQ = 0 in simulations of

the nonlinear model (53) For the nonlinear model system (53) the theoretically motivatedchoice τQ = minusτu also yields numerically stable evolutions when τu ge 0 and τv ge 0

For the nonlinear systems (4) and (53) we do not attempt a formal stability proofNevertheless the results of this appendix have served as a guide for our choices of penaltyparameters For the BSSN system (4) u v and Q are block indices [cf Eq (6)] Similarto the model problem we have penalized Q with τu = 1 with τv chosen large enough toheuristically overcome the cross-terms of indefinite size that arise from τQ = 0 (we interpretequations like τu = 1 componentwise) An analogous choice ldquoτQ = minusτurdquo for the BSSNsystem might be possible but would be considerably more complicated Indeed such achoice likely entails a matrix of penalty parameters but we do not give the details here

[1] F Pretorius Evolution of Binary Black-Hole Spacetimes Phys Rev Lett 95 121101 (2005)

(4 pages)

[2] F Pretorius Simulation of binary black hole spacetimes with a harmonic evolution scheme

Class Quantum Grav 23 S529-S552 (2006)

[3] M Campanelli C O Lousto P Marronetti and Y Zlochower Accurate Evolutions of Or-

biting Black-Hole Binaries without Excision Phys Rev Lett 96 111101 (2006) (4 pages)

[4] J G Baker J Centrella D-I Choi M Koppitz and J van Meter Gravitational-Wave

Extraction from an Inspiraling Configuration of Merging Black Holes Phys Rev Lett 96

111102 (2006) (4 pages)

[5] M Campanelli C O Lousto and Y Zlochower Last orbit of binary black holes Phys Rev D

73 061501(R) (2006) (5 pages)

[6] F Herrmann I Hinder D Shoemaker and P Laguna Unequal mass binary black hole plunges

and gravitational recoil Class Quantum Grav 24 S33-S42 (2007)

[7] P Diener F Herrmann D Pollney E Schnetter E Seidel R Takahashi J Thornburg and

J Ventrella Accurate Evolution of Orbiting Binary Black Holes Phys Rev Lett 96 121101

(2006) (4 pages)

[8] M A Scheel H P Pfeiffer L Lindblom L E Kidder O Rinne and S A Teukolsky Solving

Einsteinrsquos Equations With Dual Coordinate Frames Phys Rev D 74 104006 (2006)

[9] U Sperhake Binary black-hole evolutions of excision and puncture data Phys Rev D 76

104015 (2007) (20 pages)

[10] B Brugmann J A Gonzalez M Hannam S Husa U Sperhake and W Tichy Calibration

of moving puncture simulations Phys Rev D 77 024027 (2008) (25 pages)

[11] P Marronetti W Tichy B Brugmann J Gonzalez M Hannam S Husa and U Sperhake

Binary black holes on a budget simulations using workstations Class Quantum Grav 24

S43-S58 (2007)

[12] Z B Etienne J A Faber Y T Liu S L Shapiro and T W Baumgarte Filling the holes

Evolving excised binary black hole initial data with puncture techniques Phys Rev D 76

101503(R) (2007) (5 pages)

[13] B Szilagyi D Pollney L Rezzolla J Thornburg and J Winicour An explicit harmonic code

for black-hole evolution using excision Class Quantum Grav 24 S275-S293 (2007)

[14] L Boyle M Kesden and S Nissanke Binary-Black-Hole Merger Symmetry and the Spin

Expansion Phys Rev Lett 100 151101 (2008) (4 pages)

31

[15] M A Scheel M Boyle T Chu L E Kidder K D Matthews and H P Pfeiffer High-

accuracy waveforms for binary black hole inspiral merger and ringdown Phys Rev D 79

024003 (2009) (14 pages)

[16] M Hannam Status of black-hole-binary simulations for gravitational-wave detection

Class Quantum Grav 26 114001 (2009) (21 pages)

[17] I Hinder The current status of binary black hole simulations in numerical relativity

Class Quantum Grav 27 114004 (2010) (19 pages)

[18] B P Abbott et al LIGO The Laser Interferometer Gravitational-Wave Observatory

Rep Prog Phys 72 076901 (2009) (25 pages)

[19] F Acernese et al The status of VIRGO Class Quantum Grav 23 S63-S69 (2006)

[20] F Acernese et al Virgo status Class Quantum Grav 25 184001 (2008) (9 pages)

[21] R Arnowitt S Deser and C W Misner The Dynamics of General Relativity in Gravitation

an introduction to current research edited by L Witten (Wiley New York 1962) Chapter 7

227-265 Also available as gr-qc0405109

[22] J W York Jr Kinematics and Dynamics of General Relativity in Sources of Gravitational

Radiation edited by L L Smarr (Cambridge University Press Cambridge 1979)

[23] L E Kidder M A Scheel and S A Teukolsky Extending the lifetime of 3D black hole

computations with a new hyperbolic system of evolution equations Phys Rev D 64 064017

(2001) (13 pages)

[24] G Calabrese J Pullin O Sarbach and M Tiglio Convergence and stability in numerical

relativity Phys Rev D 66 041501(R) (2002) (4 pages)

[25] S Frittelli and O Reula On the Newtonian Limit of General Relativity Commun Math

Phys 166 221-235 (1994)

[26] M Shibata and T Nakamura Evolution of three-dimensional gravitational waves Harmonic

slicing case Phys Rev D 52 5428-5444 (1995)

[27] Y Choquet-Bruhat and J W York Geometrical well posed systems for the Einstein equations

C R Acad Sci Paris t 321 Serie I 1089-1095 (1995)

[28] A Abrahams A Anderson Y Choquet-Bruhat and J W York Jr Einstein and Yang-Mills

Theories in Hyperbolic Form without Gauge Fixing Phys Rev Lett 75 3377-3381 (1995)

[29] C Bona J Masso E Seidel and J Stela New Formalism for Numerical Relativity

Phys Rev Lett 75 600-603 (1995)

[30] M H P M van Putten and D M Eardley Nonlinear wave equations for relativity

Phys Rev D 53 3056-3063 (1996)

[31] S Frittelli and O A Reula First-Order Symmetric Hyperbolic Einstein Equations with Ar-

bitrary Fixed Gauge Phys Rev Lett 76 4667-4670 (1996)

[32] H Friedrich Hyperbolic reductions for Einsteinrsquos equations Class Quantum Grav 13 1451-

1469 (1996)

[33] F B Estabrook R S Robinson H D Wahlquist Hyperbolic equations for vacuum gravity

using special orthonormal frames Class Quantum Grav 14 1237-1247 (1997)

[34] M S Iriondo E O Leguizamon and O A Reula Einsteinrsquos Equations in Ashtekarrsquos Vari-

ables Constitute a Symmetric Hyperbolic System Phys Rev Lett 79 4732-4735 (1997)

[35] A Anderson Y Choquet-Bruhat and J W York Einstein-Bianchi hyperbolic system for

general relativity Topol Meth Nonlinear Anal 10 353-373 (1997)

[36] T W Baumgarte and S L Shapiro Numerical integration of Einsteinrsquos field equations

Phys Rev D 59 024007 (1998) (7 pages)

[37] M A G Bonilla Symmetric hyperbolic systems for Bianchi equations Class Quantum

13

for nablaWF(W ) Precisely we assume the scaling

τ(bk) = τ(ak+1) = τk+12 equiv C middotmax∣

∣micro(

A(u))∣

∣ (41)

where C = O(1) is a constant chosen for stability Larger values of C will result in schemeswith better stability properties whereas too large a value will impact the CFL conditionAt the interface point Ik+12 equiv D

k cap Dk+1 the vector uh has two representations uminus at bk

and u+ at ak+1 The max in (41) is taken over the corresponding two sets of field speedsMore precisely the speeds in Table I are computed for both uminus and u+ and the maximumtaken over all resulting speeds For the auxiliary variables a penalized central flux is usedThe definition with one representative component is

ulowast = uh minus1

2

[[

uh]]

α-component of ulowast αlowast = αh minus1

2

[[

αh

]]

(42)

with similar expressions for the remaining componentsWe stress the following point Since the interior coupling between subdomains is achieved

through the numerical flux forms (41) and (42) the inverse transformation (17) expressingthe fundamental fields in terms of the characteristic fields is not required to achieve thiscoupling On the other hand imposition of physical boundary conditions may still rely on(17) since this transformation allows one to fix only incoming characteristic modes

C Matrix form of the semi-discrete equations

Let us introduce the kth mass and stiffness matrices

Mkij =

int bk

akdrℓki (r)ℓ

kj (r) Sk

ij =

int bk

akdrℓki (r)ℓ

kjprime(r) (43)

These matrices belong to Dk and the corresponding matrices defined on the reference interval[minus1 1] are

Mij =

int 1

minus1

duℓi(u)ℓj(u) Sij =

int 1

minus1

duℓi(u)ℓprimej(u) (44)

where ℓj(u) is the jth Lagrange polynomial determined by the LGL nodes uj on [minus1 1]These matrices are related by Mk

ij =12(bk minus ak)Mij and S

kij = Sij whence only the reference

matrices require computation and storageWe will use the matrices Mk and Sk in obtaining an ODE system from (26) and (35)

Towards this end we first approximate the nonlinear terms (products and quotients) in(26) by degree-N interpolating polynomials Such approximations are achieved throughpointwise representations For example (QαQχgrr)h appears in (26) and is expressed inthe following way [cf footnote 2]

(

QαQχ

grr

)

h

(t r) =

(

QαhQχh

grrh

)

(t r) rarrNsum

j=0

Qαh(t rkj )Qχh(t r

kj )

grrh(t rkj )

ℓkj (r) (45)

Note that the expressions on the right and left are not equivalent due to aliasing error [65]Our vector notation for this replacement will be

(

QαQχ

grr

)

h

(t r) rarr(

QαQχ

grr

)

(t)Tℓk(r) (46)

14

Operations among bold variables are always performed pointwise Making similar replace-ments for all terms in (26) and then carrying out the integrations in (35) which bring inMk and Sk we arrive at

parttK = βrDK minus χDQα

grr

+1

2

χQgrrQα

g2rr

minusχQgθθ

Qα

grrgθθ

+1

2

QαQχ

grr

+3

2

αA2rr

g2rr

+1

3αK2 +Mminus1ℓk (fKh minus f lowast

K)∣

∣

∣

bk

ak (47)

where we have again suppressed the superscript k on all terms except ℓk(r) and the sub-script h is dropped on all boldfaced variables As described in [66] the spectral collocationderivative matrix

(Dk)ij =dℓkjdr

∣

∣

∣

∣

∣

r=rki

(48)

can also be expressed as Dk = (Mk)minus1Sk which appears above Eight other semi-discreteevolution equations are similarly obtained with nine in total (one for each component ofWuv) Additionally we have

Qα = Dα+Mminus1ℓk(αlowast minusαh)∣

∣

∣

bk

ak (49)

as one of the auxiliary equations with five in total (one for each component of Q = WQQ)

D Filtering

Like other nodal (pseudospectral) methods our scheme may suffer from instabilitiesdriven by aliasing error [65] Filtering is a simple yet robust remedy To filter a solu-tion component such as χ we use the modal (as opposed to nodal) representation of thesolution

χkh(t r) =

Nsum

j=0

χ(t rkj )ℓkj (r) =

Nsum

j=0

χkj (t)Pj(r) (50)

where Pj(r) is the jth Legendre polynomial Let ηj = jN and define the filter function

σ(ηj) =

1 for 0 le ηj le NcN

exp(

minus ǫ(

ηjminusNcN

1minusNcN

)2s)

for NcN le ηj le 1(51)

At each timestep we modify our solution component according to

χkh rarr

(

χkh

)filtered=

Nsum

j=0

σ(ηj)χkj (t)Pj(r) (52)

Evidently the modification only affects the top N minus Nc modes and is sufficient to controlthe type of weak instability driven by aliasing [66] The numerical parameters Nc and ǫ areproblem dependent For our simulations we have taken ǫ ≃ minuslog(εmach) = 36 where εmach

is machine accuracy in double precision

15

E Model system

To better illustrate the basic properties of our method we consider a toy model Namelythe following spatially second-order system

parttu = uprime + av minus u3 + g(t x) (53a)

parttv = uprimeprime + vprime minus (u+ v)(uprime)2 + v2u2 + h(t x) (53b)

where a ge 1 is constant and g and h are analytic source terms to be specified In contrastto (6) here u v and Q = uprime are scalars rather than vectors System (53) admits a first-order reduction in which uprime is defined as an extra variable Since this first-order reductionis strongly hyperbolic the spatially second-order system (53) is also strongly hyperbolic byone of the definitions considered in [75] The characteristic fields Xplusmn and speeds microplusmn are

X+ =radicav minus uprime micro+ =

radicaminus 1 Xminus =

radicav + uprime microminus = minus(

radica+ 1) (54)

To construct a local dG scheme for this system we first rewrite it as

parttu = Q+ av minus u3 + g(t x) (55a)

parttv = Qprime + vprime minus (u+ v)Q2 + v2u2 + h(t x) (55b)

Q = uprime (55c)

Evidently f = minus(Q + v) is the v-component of the physical flux vector

F (vQ) equiv(

Fu

Fv

)

=

(

0

f

)

(56)

Note that F has the same structure as (u v)T Borrowing from the presentation for theBSSN system we write the analogous semidiscrete scheme on each subdomain D

k for themodel system

parttu = Q+ av minus u3 + g(t) (57a)

parttv = DQ+Dv minus (u+ v)Q2 + v2u2 + h(t) +Mminus1ℓk(fh minus f lowast)∣

∣

∣

bk

ak (57b)

Q = Du+Mminus1ℓk(ulowast minus uh)∣

∣

∣

bk

ak (57c)

Here we have suppressed the subinterval label k from all variables except for the vectorℓk of Lagrange polynomial values Moreover following the guidelines discussed above thenumerical fluxes are given by

f lowast = fh+1 +

radica

2

[[

vh]]

ulowast = uh minus1

2

[[

uh]]

(58)

Appendix C analyzes the stability of our scheme for a more general numerical flux choiceas applied to (53) with the nonlinear and source terms dropped

IV RESULTS FROM NUMERICAL SIMULATIONS

This section presents results found by numerically solving both the model system (53)and BSSN system (4) with the dG scheme presented in Sec III

16

10minus10

10minus6

10minus2

∆uinfin

10minus10

10minus6

10minus2

∆vinfin

0 500 1000 1500 2000 2500 3000 3500 400010

minus10

10minus6

10minus2

∆(partxu)infin

t

N = 3N = 6N = 9N = 12

FIG 1 Spectral convergence of fields for model PDE Respectively for N = 3 6 9 12

a timestep of ∆t = 00578 00178 00084 00049 has been chosen for stability and accuracy In the

title headings for example ∆u equiv unumer minus uexact

A Simulations of the model system

The semi-discrete scheme (57) has been integrated with the classical fourth-order Runge-Kutta method When integrating this system we have first constructed Q at each Runge-Kutta stage and then substituted into the evolution equations (57ab) for u and v Theproblem has been solved on a computational domain [0 4π] comprised of two subdomainswith a timestep chosen small enough for stability The initial data has been taken from thefollowing exact solution to (53)

uprimeexact(t x) =1

2

[

sin(xminus microminust)minus sin(xminus micro+t)]

(59a)

vexact(t x) =1

2radica

[

sin(xminus microminust) + sin(xminus micro+t)]

(59b)

g(t x) = u3exact (59c)

h(t x) = (uexact + vexact)(uprimeexact)

2 minus v2exactu2exact (59d)

where the speeds microplusmn are found in (54) Specification of the boundary condition at a physicalendpoint amounts to choosing the external state for at the endpoint We have considered

17

8 16 32 64 128

10minus3

10minus2

10minus1

N

∆t

max stable ∆t8N2

FIG 2 Scaling of maximum stable ∆t with N for model PDE

two possibilities (i) the analytic state (Q+ v+) = (Qexact vexact) and (ii) an upwind stateFor example at x = 4π the upwind state is4

Q+ = Qupwind =1

2

[

(Xminus)exact minus (X+)numer

]

v+ = vupwind =1

2radica

[

(Xminus)exact + (X+)numer

]

(60)Either choice of (Q+ v+) leads to similar results and the plots here correspond to theanalytic state Figure 1 clearly shows spectral convergence with increasing polynomial orderN across all fields for the case a = 2 Other values of a including a = 1 for which X+

is a static characteristic field have also been considered with similar results Appendix Cdemonstrates that our proposed scheme for the system (57) with nonlinear and source termsdropped is stable in a semi-discrete sense Nevertheless the fully discrete scheme obtainedvia temporal discretization by the fourth-order Runge-Kutta method is still subject to thestandard absolute stability requirement Namely if microh is any eigenvalue corresponding tothe (linearized) discrete spatial operator then a necessary condition for stability is thatmicroh∆t lies in absolute stability region for Runge-Kutta 4 We here show empirically that theassociated timestep restriction scales like Nminus2 ie ∆t = O(Nminus2) for stability We note thatsuch scaling is welcome in light of the second-order spatial operators which appear in thesystem and suggest a possible worse scaling like Nminus4 Fig 2 plots the maximum stabletimestep for a range of N demonstrating the Nminus2 scaling in line with behavior known fromanalysis of first-order systems [66] This scaling also holds for the BSSN system

4 We remind the reader that unfortunately the plusmn on Xplusmn means something different than the plusmn indicating

exteriorinterior dG states [cf footnote 3]

18

B Simulations of the BSSN system

This subsection documents results for simulations of the unit-mass-parameter (M = 1)Schwarzschild solution (B9) expressed in terms of ingoing Kerr-Schild coordinates Sincethe solution is stationary temporal integration of the semi-discrete scheme has been carriedout with the forward Euler method The r-coordinate domain [04 34] has been split into3 equally spaced subdomains and we have set η = 10 λ = 01 and C = 2 [cf Eq (41)]For all simulations ∆t has been chosen for stability With the chosen λ the inner physicalboundary rmin = 04 is an excision surface At each timestep we have applied an (order2s = 20) exponential filter on the top two-thirds of the modal coefficient set for all fieldsexcept for grr and gθθ For stability we have empirically observed that grr and gθθ must not

be filtered A detailed understanding of this is still lackingIssues related to physical boundary conditions are similar to the one encountered in

Sec IVA for the model problem Similar to before we have retained Eqs (3842) as thechoice of numerical flux even at the endpoints Therefore at an endpoint the specificationof the boundary condition amounts to the choice W+ of external state We have typicallychosen the inner boundary of the radial domain as an excision boundary and in this caseW+ = Wminus is enforced at the inner physical boundary At the outer physical boundaryfor W+ we have again considered two choices (i) Wexact and (ii) Wupwind To enforcechoice (ii) the inverse transformation (17) must be used with incoming characteristic fieldsfixed to their exact values similar to (60) We have tried various versions of choice (ii)and in all cases the resulting simulations have been unstable We therefore present resultscorresponding to choice (i) Although the choice of an analytical external state Wexact

at the outer boundary is stable for our problem such a boundary condition is unlikelyto generalize to more complicated scenarios involving dynamical fields Indeed the issueof outer boundary conditions for the BSSN system is an active area of research with aproper treatment requiring fixation of incoming radiation control of the constraints andspecification of gauge (see Ref [86] for a recent analysis)

For BSSN simulations our main diagnostic is to monitor the Hamiltonian momentumand conformal connection constraints Figure 3 depicts long-time histories of constraintviolations whereas Figs 4 and 5 depict long-time error histories for the individual BSSNfield components From the middle plot in Fig 5 we infer that up to the indicated numericalerror the factor g sin4 θ = grr(gθθ)

2 remains at its initial fixed profile r4 throughout theevolution These figures indicate that the proposed scheme is stable for long times andexhibits spectral converge with increased polynomial order N Similar results are recoveredfromM = 0 Minkowski initial data The stability documented in these plots does not appearto rely on inordinate parameter tuning For example with the fixed parameters describedabove we obtain similar plots if we individually vary (i) rmin over 0325 035 04 0475(values still corresponding to an excision surface for the given choice of λ) (ii) η over1 3 7 10 (iii) s over 8 9 10 With the polynomial order N ranging over 23 26 29 31both stability and qualitatively similar exponential convergence is achieved with a singlesubdomain Likewise adoption of a larger coordinate domain with more subdomains doesnot impact our results Finally we have considered the addition of random noise to all fieldcomponents at the initial time Precisely with the system component χ as an example wehave set

χ(0) 7rarr χ(0) + δχ(0) (61)

where each component (nodal value) of δχ(0) is 10minus5 times a random variable drawn from a

19

10minus9

10minus7

10minus5

10minus3

Hinfin

10minus9

10minus7

10minus5

10minus3

Mrinfin

0 2000 4000 6000 8000 10000

10minus9

10minus7

10minus5

10minus3

Grinfin

t

N = 11N = 14N = 17N = 20

FIG 3 Spectral convergence of constraint violations for M = 1 Kerr-Schild ini-

tial data Respectively for N = 11 14 17 19 a timestep of ∆t ≃ 00041 00026 00018 00013

has been chosen for stability and accuracy

standard normal distribution Such perturbed initial data also gives rise to stable evolutions

V CONCLUSION

We have introduced a discontinuous Galerkin method for solving the spherically reducedBSSN system with second-order spatial operators Our scheme shares similarities withother discontinuous Galerkin methods that use local auxiliary variables to handle high-orderspatial derivatives [66 77ndash79 82 84 85] and which have typically been applied to eitherelliptic parabolic or mixed type problems The key ingredient of a stable dG scheme is anappropriate choice of numerical flux and our particular choice has been motivated by theanalysis presented in Appendix C When used to evolve the Schwarzschild solution in Kerr-Schild coordinates our numerical implementation of the BSSN system (4) is robustly stableand converges to the analytic solution exponentially with increased polynomial order Byapproximating the spatially second-order form of the BSSN system we have not introducedextra fields which are evolved Evolved auxiliary fields result in new constraints which mayspoil stability Our main goal has been stable evolution of the spherically reduced BSSNsystem as a first step towards understanding how a discontinuous Galerkin method might be

20

10minus9

10minus7

10minus5

∆Arrinfin

10minus9

10minus7

10minus5

∆Kinfin

0 2000 4000 6000 8000 10000

10minus9

10minus7

10minus5

max(∆Brinfin ∆Γrinfin)

t

N = 11N = 14N = 17N = 20

FIG 4 Spectral convergence of solution for M = 1 Kerr-Schild initial data

Timestep choices are described in the caption for Fig 3 In the title headings for example

∆Arr equiv (Arr)numer minus (Arr)exact

applied to the full BSSN system Towards that goal we now discuss treatment of singularitiesand generalization of the described dG method to higher space dimension

To deal with the fixed Schwarzschild singularity we have used excision which is easy inthe context of the spherically reduced BSSN system However excision for the binary blackhole problem in full general relativity requires attention to the technical challenge of hori-zon tracking State-of-the-art BSSN codes avoid such complication relying instead on themoving-puncture technique While the moving-puncture technique does involve mild centralsingularities it may still prove amenable to spectral methods Indeed spectral methods fornon-smooth problems is well-developed in both theory and for complex applications Sincethe moving-puncture technique can be performed in spherical symmetry [60] a first-steptoward a spectral moving-puncture code would be to implement a moving puncture withthe nodal dG method described here Such an implementation may adopt Legendre-Gauss-Radau nodes on the innermost subdomain thereby ensuring that the physical singularitydoes not lie on a nodal point (in much the same way finite difference codes use a staggeredgrid) Beyond traditional excision and moving punctures one might also retain a smoothproblem for which spectral methods are especially well-suited via the turducken approach[87 88] to singularities

21

10minus11

10minus9

10minus7

10minus5

∆χinfin

10minus11

10minus9

10minus7

10minus5

max(∆grrinfin ∆gθθinfin)

0 2000 4000 6000 8000 10000

10minus11

10minus9

10minus7

10minus5

max(∆αinfin ∆βrinfin)

t

N = 11N = 14N = 17N = 20

FIG 5 Spectral convergence of solution violations for M = 1 Kerr-Schild initial

data See the caption of Fig 4 for details

Discontinuous Galerkin methods for hyperbolic problems in two and three space dimen-sions are well-developed A generalization of the method described here to three-dimensionsand the full BSSN system would likely rely on an unstructured mesh Appropriate local poly-nomial expansions for the subelements are well-understood as are choices for the numericalfluxes which would now live on two-dimensional faces rather than single points Whether ornot it would ultimately prove successful generalization of our dG method to a higher dimen-sion would rely on an established conceptual framework Further computational advancesof relevance to a generalization of our dG method to the full BSSN system (possibly includ-ing matter) may include mesh hp-adaptivity local timestepping shock capturing and slopelimiting techniques [66] Moreover recent work [89] indicates that enhanced performancewould be expected were our scheme implemented on graphics processor units

VI ACKNOWLEDGMENTS

We thank Nick Taylor for discussions about treating second-order operators with spectralmethods Benjamin Stamm for helping polish up a few parts of Appendix C Khosro Shah-bazi for discussions on LDG and IP methods We also acknowledge helpful conversationswith David Brown and Manuel Tiglio about the BSSN system and previous work on its nu-

22

merical implementation We gratefully acknowledge funding through grants DMS 0554377and DARPAAFOSR FA9550-05-1-0108 to Brown University and NSF grant PHY 0855678to the University of New Mexico

Appendix A Hyperbolicity of the first-order system

This appendix analyzes the matrix A(u) appearing in (10) in order to construct thecharacteristic fields (14) In matrix form the sector (8) of the principal part of (10) reads asfollows

partt

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

=

βr 0 minus 4λα3grr

0 0 λβr

6(grr)2λβr

3gθθgrr0 4λ

3grr

0 βr 0 23grrαχ

13α minus αχ

3grr

αχ3gθθ

minus23χ 0

0 0 βr 0 0 0 0 minus χgrr

0

0 0 minus 4α3grr

βr 0 βr

6(grr)2βr

3gθθgrr0 4

3grr

0 0 23αχ 0 βr minus βrχ

3grrminus2βrχ

3gθθ0 minus2

3χ

0 minus2α 0 0 0 23βr minus2grrβr

3gθθ0 4

3grr

0 gθθαgrr

0 0 0 minusgθθβr

3grr13βr 0 minus2

3gθθ

0 0 minus2α 0 0 0 0 βr 0

34

0 0 0 0 0 0 0 βr

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

prime

(A1)which defines the matrix A(u) appearing in (9) and so also the matrix A(u) in (10) Notethat in the last equation the matrix within the square brackets is minusA(u) For certainconfigurations of u and λ the system (10) is strongly hyperbolic [76] that is A(u) hasa complete set of eigenvectors and real eigenvalues Indeed five eigenpairs of A(u) aretrivially recovered upon inspection of A(u)rsquos leading 5times5 diagonal block These correspondto eigenvalue 0 and the left eigenspace ξj = eTj 1 le j le 5 where ej are the canonical

basis vectors Since each component of u arises as eTj W each is also a characteristic field

The remaining nine eigenpairs are determined by A(u) The eigenvalues of A(u) are

micro1 = 0 micro23 = minusβr microplusmn4 = minusβr plusmn

radic

2αχ

grr microplusmn

5 = minusβr plusmn α

radic

χ

grr microplusmn

6 = minusβr plusmnradic

λ

grr

(A2)

23

and the corresponding left eigenvectors are

x1 = (0 0 0 0 0 gθθ 2grr 0 0) (A3a)

x2 =

(

0 0 0 grr2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3b)

x3 =

(

grrλ 0 0 0

2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3c)

xplusmn4 =

(

0 0plusmnradic

2αgrrχ

0 0 0 0 1 0

)

(A3d)

xplusmn5 =

(

0∓ 3radicgrrχ

plusmn2

radic

grrχ 2grr

1

χminus 1

grr1

gθθ 0 0

)

(A3e)

xplusmn6 =

(

minus3

4

grrλ 0plusmn α

radicλgrr

(2αχminus λ) 0 0minus βr

8(βrgrr ∓radicλgrr)

minus βrgrr

4gθθ(βrgrr ∓radicλgrr)

αχ

(2αχminus λ)plusmnradic

grrλ

)

(A3f)

where for example xplusmn5 A(u) = microplusmn5 x

plusmn5 Assuming that grr gθθ χ and α are everywhere strictly

positive the eigenvalues are real and the eigenvectors are linearly independent providedthat (15) holds These eigenvectors are easily extended to eigenvectors of A(u) e g asxplusmn6 rarr (01times5 x

plusmn6 ) Then for example the characteristic field

Xplusmn6 equiv (01times5 x

plusmn6 )W = xplusmn6WvQ (A4)

and similarly Xplusmnj = xplusmnj WvQ for j = 4 5 and Xk = xkWvQ for k = 1 2 3 The characteristic

speeds for these fields are microk and microplusmnj With this convention the speeds listed in Table I

correspond to the Xk and Xplusmnj in (14)

Appendix B Schwarzschild solution in conformal Kerr-Schild coordinates

In Kerr-Schild coordinates here the system directly related to incoming Eddington-Finkelstein null coordinates the line element for the Schwarzschild solution reads

ds2 = minusα2dt2 + (1 + 2MR)(dR + βRdt)2 +R2dθ2 +R2 sin2 θdφ2 (B1)

where R is the area radius α = (1 + 2MR)minus12 is the lapse and βR = 2M(R + 2M) isthe shift vector The physical spatial metric gab is the spatial part of this line element

To define the corresponding solution to the BSSN system we use equation gab = χgab todefine the following relationship between line elements

dr2 + r2(dθ2 + sin2 θdφ2) = χ[(1 + 2MR)dR2 +R2dθ2 +R2 sin2 θdφ2] (B2)

so that

χ

(

1 +2M

R

)(

dR

dr

)2

= 1 χR2 = r2 (B3)

Then we have(

1 +2M

R

)12dR

R=dr

r (B4)

24

with integration yielding

r =R

4

(

1 +

radic

1 +2M

R

)2

e2minus2radic

1+2MR (B5)

where the constant of integration has been chosen so that the R r rarr infin limits are consistentThe second relation in (B3) shows that

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR =rArr χminus4 =2eradic

1+2MRminus1

1 +radic

1 + 2MR (B6)

The extrinsic curvature tensor is specified by the expression forK given in (B9h) the identityK = KR

R + 2Kθθ and

Kθθ =

(

1 +2M

R

)minus122M

R2 (B7)

Since KRR = Kr

r we compute that

Krr = K minus 2Kθ

θ = minus(

1 +2M

R

)minus12(R +M

R + 2M

)

2M

R2 (B8)

Next since Krr = grrKrr = χminus1Kr

r we have Krr = Arr+

13grrK This implies Arr = Kr

r minus 13K

from which we get (B9g) In all we have

α =

(

1 +2M

R

)minus12

(B9a)

βr = βR dr

dR= χ12

(

1 +2M

R

)minus122M

R(B9b)

grr = 1 (B9c)

gθθ = r2 = χR2 (B9d)

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR (B9e)

Br = 0 (B9f)

Arr = minus(

1 +2M

R

)minus124M

3R2

(

2R + 3M

R + 2M

)

(B9g)

K =

(

1 +2M

R

)minus32(

1 +3M

R

)

2M

R2(B9h)

Γr = minus2

r= minus 2

χ12R (B9i)

To differentiate these expressions with respect to r we use the identity

dR

dr= χminus12

(

1 +2M

R

)minus12

(B10)

along with the chain rule

25

Appendix C Stability of the model system

The following stability analysis for the model system (53) has been inspired by [78 79]After dropping all nonlinear source terms the system (53) becomes

parttu = uprime + av (C1a)

parttv = uprimeprime + vprime (C1b)

This section analyzes the stability of (C1) considering both the continuum system itself aswell as its semi-discrete dG approximation The latter analysis offers some insight into theempirically observed stability of our dG scheme for the spherically reduced BSSN equations

1 Analysis for a single interval

Throughout we work with the L2 inner product and norm

(f g)D =

int

D

fg fD =radic

(f f)D (C2)

where D is the spatial coordinate interval (here D may represent a subdomain Dk or the

whole domain Ω) and we have suppressed all integration measures For the continuummodel we will establish the following estimate

uprime(T middot)2D+ av(T middot)2

Dle C(T )

(

uprime(0 middot)2D+ av(0 middot)2

D

)

(C3)

where the time-dependent constant C(T ) is determined solely by the choice of boundaryconditions To show (C3) we first change variables with v =

radicav thereby rewriting (C1)

in the following symmetric form

parttu = uprime +radicav (C4a)

parttv =radicauprimeprime + vprime (C4b)

Equations (C4ab) then imply

1

2partt

int

D

(uprime)2 =

int

D

uprime(uprimeprime +radicavprime) =

int

D

radicauprimevprime +

1

2

int

partD

(uprime)2 (C5a)

1

2partt

int

D

(v)2 =

int

D

v(radicauprimeprime + vprime) = minus

int

D

radicauprimevprime +

1

2

int

partD

(v2 + 2radicauprimev) (C5b)

Here vvprime and uprimeuprimeprime have been expressed as exact derivatives and then integrated to boundaryterms the second equation employs an extra integration by parts and with only one spacedimension

int

partDdenotes a difference of endpoint evaluations Addition of Eqs (C5ab) gives

1

2partt

int

D

[

v2 + (uprime)2]

=1

2

int

partD

[

v2 + (uprime)2 + 2radicauprimev

]

(C6)

Substitutions with the identities

[

v2 + (uprime)2]

=1

2

[

(v + uprime)2 + (v minus uprime)2]

2uprimev =1

2

[

(v + uprime)2 minus (v minus uprime)2]

(C7)

26

and replacements to recover the original variable v = vradica yield

1

2partt

int

D

[

av2 + (uprime)2]

=1 +

radica

4

int

partD

(radicav + uprime)2 +

1minusradica

4

int

partD

(radicav minus uprime)2 (C8)

From (C8) we deduce that the time-dependent constant C(T ) in (C3) must satisfy

∣

∣

∣

∣

∣

∣

1 +

int T

0

[

1+radica

2

int

partD(radicav + uprime)2 + 1minus

radica

2

int

partD(radicav minus uprime)2

]

dt

uprime(0 middot)2 + av(0 middot)2

∣

∣

∣

∣

∣

∣

le C(T ) (C9)

For periodic boundary conditions we may choose C(T ) = 1 Moreover if a ge 1 anduprime = minusradic

av is specified at partD+ then uprime(t middot)2 + av(t middot)2 decaysStill working on a single interval (subdomain) we now consider the semi-discrete scheme

for (C4) i e (57) with all nonlinear source terms dropped and with v replaced by vradica

Derivation of a formula analogous to (C8) is our first step toward establishing L2 stabilityof the semi-discrete scheme While (57) features vectors for example u(t) taking values atthe Legendre-Gauss-Lobatto nodal points here we work with the numerical solution as apolynomial for example uh(t x) These two representations are related by the Lagrange in-terpolating polynomials for the nodal set here taken to span both the space of test functionsand the space of basis functions Our scheme is

int

Dk

ψparttuh =

int

Dk

ψ(Qh +radicavh) (C10a)

int

Dk

ξparttvh = minusint

Dk

ξprime(radicaQh + vh) +

int

partDk

ξ(radicaQlowast + vlowast) (C10b)

int

Dk

ϕQh =

int

Dk

ϕuprimeh +

int

partDk

ϕ (ulowast minus uh) (C10c)

where ψ ξ and ϕ are polynomial test functions These test functions are arbitrary exceptthat they must be degree-N polynomials In (C10) the variables uh vh and Qh should alsocarry a superscript k but we have suppressed this Derivation of a formula analogous to(C8) is complicated by the fact that Qh is not evolved Nevertheless at a given instant t wecan assemble Qh from (C10c)

Mimicking the calculation (C5b) from the continuum case we first use (C10b) with ξ = vhto write

1

2partt

int

Dk

v2h = minusint

Dk

(radicaQh + vh)v

primeh +

int

partDk

(radicaQlowast + vlowast)vh

= minusint

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2(radicaQlowast + vlowast)vh minus v2h

]

(C11)

The right-hand side of (C5a) is analogous to

1

2partt

int

Dk

Q2h =

int

Dk

QhparttQh (C12)

However since Qh is not evolved the term parttQh must be given a suitable interpretationOn the right hand side of (C10c) only uh u

primeh and u

lowast necessarily depend on time since the

27

test function ϕ need not be time-dependent Furthermore ulowast is explicitly given as a linearcombination of uh as seen in Eq (C21c) below Choosing ϕ = ℓj taking the time derivativeof (C10c) and appealing to the commutivity of mixed partial derivatives we therefore arriveat

int

Dk

ℓjparttQh =

int

Dk

ℓj(parttuh)prime +

int

partDk

ℓj(

(parttu)lowast minus parttuh

)

(C13)

where (parttu)lowast depends on parttuh in precisely the same way that ulowast depends on uh We have

written ℓj rather than ϕ in the last equation to emphasize that the result also holds for anylinear combination of ℓj (for example ϕ) and even for time-dependent combinations SinceQh is itself such a combination we obtain

1

2partt

int

Dk

Q2h =

int

Dk

Qh(parttuh)prime +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

Qh(Qprimeh +

radicavprimeh) +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2((parttu)lowast minus parttuh)Qh +Q2

h

]

(C14)

Addition of (C11) and (C14) gives

1

2partt

int

Dk

(Q2h + v2h) =

1

2

int

partDk

[

Q2h minus v2h + 2(

radicaQlowast + vlowast)vh + 2((parttu)

lowast minus parttuh)Qh

]

(C15)

the aforementioned analog of (C8) This formula holds on a single subdomain Dk and we

now combine multiple copies of it one for each value of k

2 Analysis for multiple intervals

To facilitate combination of (C15) over all k we change notation At every subdomaininterface Ik+12 equiv partDk cap partDk+1 let the superscripts L and R denote field values respectivelytaken from the left and right Then the fields evaluated at Ik+12 which belong to D

k will beuLk+12 v

Lk+12 and Q

Lk+12 while those belonging to D

k+1 will be uRk+12 vRk+12 and Q

Rk+12

However at Ikminus12 the values taken from D

k are uRkminus12 vRkminus12 and QR

kminus12 Note that we

have also replaced the subscript h denoting a numerical solution with k plusmn 12 denotingthe location of the endpoint value of the numerical solution With this notation we define

∆Lα =

1

2

[

(QLα)

2 minus (vLα)2]

+(radic

aQlowastα + vlowastα

)

vLα +[

(parttuα)lowast minus parttu

Lα

]

QLα (C16)

and similarly for ∆Rα The same numerical fluxes appear in both ∆L

α and ∆Rα (ie each

numerical flux takes the same value on either side of an interface) whence fluxes like Qlowastα do

not carry an L or R superscript In terms of these definitions (C15) becomes

1

2partt

int

Dk

(Q2h + v2h) = ∆L

k+12 minus∆Rkminus12 (C17)

28

Summation over all Dk yields

1

2partt

kmaxsum

k=1

int

Dk

(Q2h + v2h) =

kmaxminus1sum

k=1

(

∆Lk+12 minus∆R

k+12

)

+∆Lkmax+12 minus∆R

12

=

kmaxminus1sum

k=1

(

∆Lh minus∆R

h

)∣

∣

Ik+12 +∆Lkmax+12 minus∆R

12 (C18)

We have reverted to h-notation denoting the numerical solution since the LR superscriptsindicate unambiguously the relevant domain used for evaluation at Ik+12

We again seek an estimate of the form

kmaxsum

k=1

(

Qh(T middot)2Dk + avh(T middot)2Dk

)

le C(T )kmaxsum

k=1

(

Qh(0 middot)2Dk + avh(0 middot)2Dk

)

(C19)

that is essentially the same as the one (C3) considered in the continuum case Assume thatthe chosen boundary conditions ensure ∆L

kmax+12 minus ∆R12 is bounded by a time-dependent

constant which does not depend on the numerical parameters N and h (subdomain width)Establishment of stability then amounts to showing that the remaining sum over interfaceterms in (C18) is non-positive whence this remaining sum is consistent with C(T ) le 1although the boundary conditions may give rise to a different bound In fact we will choosethe numerical fluxes such that each individual interface term is non-positive At interfaceIk+12 and in LR notation the jump and average of vh for example are

1

2

(

v+ + vminus)

equiv vh =1

2

(

vLk+12 + vRk+12

)

(C20a)

nminusvminus + n+v+ equiv[[

vh]]

= vLk+12 minus vRk+12 (C20b)

Consider numerical fluxes of the form

Qlowast = Qh minusτQ2

[[

Qh

]]

(C21a)

vlowast = vh minusτv2

[[

vh]]

(C21b)

ulowast = uh minusτu2

[[

uh]]

(C21c)

(parttu)lowast = parttuh minus

τu2

[[

parttuh]]

(C21d)

where (C21c) induces (C21d) and where the penalty parameters τu τv and τQ are realnumbers The fluxes defined in (58) correspond to τu = 1 τv = 1 +

radica and τQ = 0 In

terms of these quantities the kth interface contribution in (C18) is

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

parttuh]]

minus τu2

[[

parttuh]][[

Qh

]]

(C22)

where we have suppressed the k dependence of the right-hand side Now consider the term[[

parttuh]]

Because parttuh and Qh+radicavh are both polynomials of degree N Eq (C10a) implies

29

τu

τ Q

minus1 0 1minus1

minus05

0

05

1

τu

minus1 0 1minus1

minus05

0

05

1

FIG 6 Stable evolutions for the model system For fixed τv = 10minus6 and τv = 1 +radic2

respectively the left and right plots depict stable choices (determined empirically) of τu and τQfor the linear model system (C1) The stable regions are colored black but the jagged edges result

from the discretization of the (τu τQ)-plane

the vector equation parttu = Q +radicav that is pointwise equivalence on the nodal points of

Dk which in turn implies

[[

parttuh]]

=[[

Qh+radicavh]]

Upon substituting this identity into thelast equation we arrive at an expression which features only vh and Qh

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

Qh +radicavh]]

minus τu2

[[

Qh +radicavh]][[

Qh

]]

(C23)

The identities vh[[

vh]]

= 12

[[

v2h]]

and[[

Qh +radicavh]]

=[[

Qh

]]

+radica[[

vh]]

then simplify(C23) to

(∆Lh minus∆R

h )∣

∣

Ik+12 = minusτv2

[[

vh]]2 minus

radica(τu + τQ)

2

[[

Qh

]][[

vh]]

minus τu2

[[

Qh

]]2 (C24)

The role of a penalty parameter is now clear Positive values of τv penalize jumps in vhthrough a negative contribution to the energy Likewise positive values of τu penalizejumps in Qh through a negative contribution to the energy However because the sign of[[

Qh

]][[

vh]]

can be positive or negative only the choice τQ = minusτu yields an expression for(∆L

h minus ∆Rh )|Ik+12 which is manifestly negative for τu ge 0 and τv ge 0 A simple estimate

based on Youngrsquos inequality with ε (that is 2αβ le εminus1α2 + εβ2 where α β ge 0 and ε gt 0)shows that for τQ = 0 the choice τv ge aτu4 also yields stability

Figure 6 depicts certain choices of stable penalty parameters for the linear model systemevolved to tfinal = 1000 (with a = 2 N = 10 and ∆t ≃ 00553) as determined empiricallywith simulations similar to those described in Sec IVA The left plot corresponds to asmall τv = 10minus6 for which the choice τu = 1 τQ = 0 is not stable as expected from thetheoretical analysis However the right plot corresponds to τv = 1 +

radica for which τu = 1

τQ = 0 is stable Motivated by the numerical flux choices (3842) used for the BSSN system

30

(4) we have (as mentioned above) set τu = 1 τv = 1 +radica and τQ = 0 in simulations of

the nonlinear model (53) For the nonlinear model system (53) the theoretically motivatedchoice τQ = minusτu also yields numerically stable evolutions when τu ge 0 and τv ge 0

For the nonlinear systems (4) and (53) we do not attempt a formal stability proofNevertheless the results of this appendix have served as a guide for our choices of penaltyparameters For the BSSN system (4) u v and Q are block indices [cf Eq (6)] Similarto the model problem we have penalized Q with τu = 1 with τv chosen large enough toheuristically overcome the cross-terms of indefinite size that arise from τQ = 0 (we interpretequations like τu = 1 componentwise) An analogous choice ldquoτQ = minusτurdquo for the BSSNsystem might be possible but would be considerably more complicated Indeed such achoice likely entails a matrix of penalty parameters but we do not give the details here

[1] F Pretorius Evolution of Binary Black-Hole Spacetimes Phys Rev Lett 95 121101 (2005)

(4 pages)

[2] F Pretorius Simulation of binary black hole spacetimes with a harmonic evolution scheme

Class Quantum Grav 23 S529-S552 (2006)

[3] M Campanelli C O Lousto P Marronetti and Y Zlochower Accurate Evolutions of Or-

biting Black-Hole Binaries without Excision Phys Rev Lett 96 111101 (2006) (4 pages)

[4] J G Baker J Centrella D-I Choi M Koppitz and J van Meter Gravitational-Wave

Extraction from an Inspiraling Configuration of Merging Black Holes Phys Rev Lett 96

111102 (2006) (4 pages)

[5] M Campanelli C O Lousto and Y Zlochower Last orbit of binary black holes Phys Rev D

73 061501(R) (2006) (5 pages)

[6] F Herrmann I Hinder D Shoemaker and P Laguna Unequal mass binary black hole plunges

and gravitational recoil Class Quantum Grav 24 S33-S42 (2007)

[7] P Diener F Herrmann D Pollney E Schnetter E Seidel R Takahashi J Thornburg and

J Ventrella Accurate Evolution of Orbiting Binary Black Holes Phys Rev Lett 96 121101

(2006) (4 pages)

[8] M A Scheel H P Pfeiffer L Lindblom L E Kidder O Rinne and S A Teukolsky Solving

Einsteinrsquos Equations With Dual Coordinate Frames Phys Rev D 74 104006 (2006)

[9] U Sperhake Binary black-hole evolutions of excision and puncture data Phys Rev D 76

104015 (2007) (20 pages)

[10] B Brugmann J A Gonzalez M Hannam S Husa U Sperhake and W Tichy Calibration

of moving puncture simulations Phys Rev D 77 024027 (2008) (25 pages)

[11] P Marronetti W Tichy B Brugmann J Gonzalez M Hannam S Husa and U Sperhake

Binary black holes on a budget simulations using workstations Class Quantum Grav 24

S43-S58 (2007)

[12] Z B Etienne J A Faber Y T Liu S L Shapiro and T W Baumgarte Filling the holes

Evolving excised binary black hole initial data with puncture techniques Phys Rev D 76

101503(R) (2007) (5 pages)

[13] B Szilagyi D Pollney L Rezzolla J Thornburg and J Winicour An explicit harmonic code

for black-hole evolution using excision Class Quantum Grav 24 S275-S293 (2007)

[14] L Boyle M Kesden and S Nissanke Binary-Black-Hole Merger Symmetry and the Spin

Expansion Phys Rev Lett 100 151101 (2008) (4 pages)

31

[15] M A Scheel M Boyle T Chu L E Kidder K D Matthews and H P Pfeiffer High-

accuracy waveforms for binary black hole inspiral merger and ringdown Phys Rev D 79

024003 (2009) (14 pages)

[16] M Hannam Status of black-hole-binary simulations for gravitational-wave detection

Class Quantum Grav 26 114001 (2009) (21 pages)

[17] I Hinder The current status of binary black hole simulations in numerical relativity

Class Quantum Grav 27 114004 (2010) (19 pages)

[18] B P Abbott et al LIGO The Laser Interferometer Gravitational-Wave Observatory

Rep Prog Phys 72 076901 (2009) (25 pages)

[19] F Acernese et al The status of VIRGO Class Quantum Grav 23 S63-S69 (2006)

[20] F Acernese et al Virgo status Class Quantum Grav 25 184001 (2008) (9 pages)

[21] R Arnowitt S Deser and C W Misner The Dynamics of General Relativity in Gravitation

an introduction to current research edited by L Witten (Wiley New York 1962) Chapter 7

227-265 Also available as gr-qc0405109

[22] J W York Jr Kinematics and Dynamics of General Relativity in Sources of Gravitational

Radiation edited by L L Smarr (Cambridge University Press Cambridge 1979)

[23] L E Kidder M A Scheel and S A Teukolsky Extending the lifetime of 3D black hole

computations with a new hyperbolic system of evolution equations Phys Rev D 64 064017

(2001) (13 pages)

[24] G Calabrese J Pullin O Sarbach and M Tiglio Convergence and stability in numerical

relativity Phys Rev D 66 041501(R) (2002) (4 pages)

[25] S Frittelli and O Reula On the Newtonian Limit of General Relativity Commun Math

Phys 166 221-235 (1994)

[26] M Shibata and T Nakamura Evolution of three-dimensional gravitational waves Harmonic

slicing case Phys Rev D 52 5428-5444 (1995)

[27] Y Choquet-Bruhat and J W York Geometrical well posed systems for the Einstein equations

C R Acad Sci Paris t 321 Serie I 1089-1095 (1995)

[28] A Abrahams A Anderson Y Choquet-Bruhat and J W York Jr Einstein and Yang-Mills

Theories in Hyperbolic Form without Gauge Fixing Phys Rev Lett 75 3377-3381 (1995)

[29] C Bona J Masso E Seidel and J Stela New Formalism for Numerical Relativity

Phys Rev Lett 75 600-603 (1995)

[30] M H P M van Putten and D M Eardley Nonlinear wave equations for relativity

Phys Rev D 53 3056-3063 (1996)

[31] S Frittelli and O A Reula First-Order Symmetric Hyperbolic Einstein Equations with Ar-

bitrary Fixed Gauge Phys Rev Lett 76 4667-4670 (1996)

[32] H Friedrich Hyperbolic reductions for Einsteinrsquos equations Class Quantum Grav 13 1451-

1469 (1996)

[33] F B Estabrook R S Robinson H D Wahlquist Hyperbolic equations for vacuum gravity

using special orthonormal frames Class Quantum Grav 14 1237-1247 (1997)

[34] M S Iriondo E O Leguizamon and O A Reula Einsteinrsquos Equations in Ashtekarrsquos Vari-

ables Constitute a Symmetric Hyperbolic System Phys Rev Lett 79 4732-4735 (1997)

[35] A Anderson Y Choquet-Bruhat and J W York Einstein-Bianchi hyperbolic system for

general relativity Topol Meth Nonlinear Anal 10 353-373 (1997)

[36] T W Baumgarte and S L Shapiro Numerical integration of Einsteinrsquos field equations

Phys Rev D 59 024007 (1998) (7 pages)

[37] M A G Bonilla Symmetric hyperbolic systems for Bianchi equations Class Quantum

14

Operations among bold variables are always performed pointwise Making similar replace-ments for all terms in (26) and then carrying out the integrations in (35) which bring inMk and Sk we arrive at

parttK = βrDK minus χDQα

grr

+1

2

χQgrrQα

g2rr

minusχQgθθ

Qα

grrgθθ

+1

2

QαQχ

grr

+3

2

αA2rr

g2rr

+1

3αK2 +Mminus1ℓk (fKh minus f lowast

K)∣

∣

∣

bk

ak (47)

where we have again suppressed the superscript k on all terms except ℓk(r) and the sub-script h is dropped on all boldfaced variables As described in [66] the spectral collocationderivative matrix

(Dk)ij =dℓkjdr

∣

∣

∣

∣

∣

r=rki

(48)

can also be expressed as Dk = (Mk)minus1Sk which appears above Eight other semi-discreteevolution equations are similarly obtained with nine in total (one for each component ofWuv) Additionally we have

Qα = Dα+Mminus1ℓk(αlowast minusαh)∣

∣

∣

bk

ak (49)

as one of the auxiliary equations with five in total (one for each component of Q = WQQ)

D Filtering

Like other nodal (pseudospectral) methods our scheme may suffer from instabilitiesdriven by aliasing error [65] Filtering is a simple yet robust remedy To filter a solu-tion component such as χ we use the modal (as opposed to nodal) representation of thesolution

χkh(t r) =

Nsum

j=0

χ(t rkj )ℓkj (r) =

Nsum

j=0

χkj (t)Pj(r) (50)

where Pj(r) is the jth Legendre polynomial Let ηj = jN and define the filter function

σ(ηj) =

1 for 0 le ηj le NcN

exp(

minus ǫ(

ηjminusNcN

1minusNcN

)2s)

for NcN le ηj le 1(51)

At each timestep we modify our solution component according to

χkh rarr

(

χkh

)filtered=

Nsum

j=0

σ(ηj)χkj (t)Pj(r) (52)

Evidently the modification only affects the top N minus Nc modes and is sufficient to controlthe type of weak instability driven by aliasing [66] The numerical parameters Nc and ǫ areproblem dependent For our simulations we have taken ǫ ≃ minuslog(εmach) = 36 where εmach

is machine accuracy in double precision

15

E Model system

To better illustrate the basic properties of our method we consider a toy model Namelythe following spatially second-order system

parttu = uprime + av minus u3 + g(t x) (53a)

parttv = uprimeprime + vprime minus (u+ v)(uprime)2 + v2u2 + h(t x) (53b)

where a ge 1 is constant and g and h are analytic source terms to be specified In contrastto (6) here u v and Q = uprime are scalars rather than vectors System (53) admits a first-order reduction in which uprime is defined as an extra variable Since this first-order reductionis strongly hyperbolic the spatially second-order system (53) is also strongly hyperbolic byone of the definitions considered in [75] The characteristic fields Xplusmn and speeds microplusmn are

X+ =radicav minus uprime micro+ =

radicaminus 1 Xminus =

radicav + uprime microminus = minus(

radica+ 1) (54)

To construct a local dG scheme for this system we first rewrite it as

parttu = Q+ av minus u3 + g(t x) (55a)

parttv = Qprime + vprime minus (u+ v)Q2 + v2u2 + h(t x) (55b)

Q = uprime (55c)

Evidently f = minus(Q + v) is the v-component of the physical flux vector

F (vQ) equiv(

Fu

Fv

)

=

(

0

f

)

(56)

Note that F has the same structure as (u v)T Borrowing from the presentation for theBSSN system we write the analogous semidiscrete scheme on each subdomain D

k for themodel system

parttu = Q+ av minus u3 + g(t) (57a)

parttv = DQ+Dv minus (u+ v)Q2 + v2u2 + h(t) +Mminus1ℓk(fh minus f lowast)∣

∣

∣

bk

ak (57b)

Q = Du+Mminus1ℓk(ulowast minus uh)∣

∣

∣

bk

ak (57c)

Here we have suppressed the subinterval label k from all variables except for the vectorℓk of Lagrange polynomial values Moreover following the guidelines discussed above thenumerical fluxes are given by

f lowast = fh+1 +

radica

2

[[

vh]]

ulowast = uh minus1

2

[[

uh]]

(58)

Appendix C analyzes the stability of our scheme for a more general numerical flux choiceas applied to (53) with the nonlinear and source terms dropped

IV RESULTS FROM NUMERICAL SIMULATIONS

This section presents results found by numerically solving both the model system (53)and BSSN system (4) with the dG scheme presented in Sec III

16

10minus10

10minus6

10minus2

∆uinfin

10minus10

10minus6

10minus2

∆vinfin

0 500 1000 1500 2000 2500 3000 3500 400010

minus10

10minus6

10minus2

∆(partxu)infin

t

N = 3N = 6N = 9N = 12

FIG 1 Spectral convergence of fields for model PDE Respectively for N = 3 6 9 12

a timestep of ∆t = 00578 00178 00084 00049 has been chosen for stability and accuracy In the

title headings for example ∆u equiv unumer minus uexact

A Simulations of the model system

The semi-discrete scheme (57) has been integrated with the classical fourth-order Runge-Kutta method When integrating this system we have first constructed Q at each Runge-Kutta stage and then substituted into the evolution equations (57ab) for u and v Theproblem has been solved on a computational domain [0 4π] comprised of two subdomainswith a timestep chosen small enough for stability The initial data has been taken from thefollowing exact solution to (53)

uprimeexact(t x) =1

2

[

sin(xminus microminust)minus sin(xminus micro+t)]

(59a)

vexact(t x) =1

2radica

[

sin(xminus microminust) + sin(xminus micro+t)]

(59b)

g(t x) = u3exact (59c)

h(t x) = (uexact + vexact)(uprimeexact)

2 minus v2exactu2exact (59d)

where the speeds microplusmn are found in (54) Specification of the boundary condition at a physicalendpoint amounts to choosing the external state for at the endpoint We have considered

17

8 16 32 64 128

10minus3

10minus2

10minus1

N

∆t

max stable ∆t8N2

FIG 2 Scaling of maximum stable ∆t with N for model PDE

two possibilities (i) the analytic state (Q+ v+) = (Qexact vexact) and (ii) an upwind stateFor example at x = 4π the upwind state is4

Q+ = Qupwind =1

2

[

(Xminus)exact minus (X+)numer

]

v+ = vupwind =1

2radica

[

(Xminus)exact + (X+)numer

]

(60)Either choice of (Q+ v+) leads to similar results and the plots here correspond to theanalytic state Figure 1 clearly shows spectral convergence with increasing polynomial orderN across all fields for the case a = 2 Other values of a including a = 1 for which X+

is a static characteristic field have also been considered with similar results Appendix Cdemonstrates that our proposed scheme for the system (57) with nonlinear and source termsdropped is stable in a semi-discrete sense Nevertheless the fully discrete scheme obtainedvia temporal discretization by the fourth-order Runge-Kutta method is still subject to thestandard absolute stability requirement Namely if microh is any eigenvalue corresponding tothe (linearized) discrete spatial operator then a necessary condition for stability is thatmicroh∆t lies in absolute stability region for Runge-Kutta 4 We here show empirically that theassociated timestep restriction scales like Nminus2 ie ∆t = O(Nminus2) for stability We note thatsuch scaling is welcome in light of the second-order spatial operators which appear in thesystem and suggest a possible worse scaling like Nminus4 Fig 2 plots the maximum stabletimestep for a range of N demonstrating the Nminus2 scaling in line with behavior known fromanalysis of first-order systems [66] This scaling also holds for the BSSN system

4 We remind the reader that unfortunately the plusmn on Xplusmn means something different than the plusmn indicating

exteriorinterior dG states [cf footnote 3]

18

B Simulations of the BSSN system

This subsection documents results for simulations of the unit-mass-parameter (M = 1)Schwarzschild solution (B9) expressed in terms of ingoing Kerr-Schild coordinates Sincethe solution is stationary temporal integration of the semi-discrete scheme has been carriedout with the forward Euler method The r-coordinate domain [04 34] has been split into3 equally spaced subdomains and we have set η = 10 λ = 01 and C = 2 [cf Eq (41)]For all simulations ∆t has been chosen for stability With the chosen λ the inner physicalboundary rmin = 04 is an excision surface At each timestep we have applied an (order2s = 20) exponential filter on the top two-thirds of the modal coefficient set for all fieldsexcept for grr and gθθ For stability we have empirically observed that grr and gθθ must not

be filtered A detailed understanding of this is still lackingIssues related to physical boundary conditions are similar to the one encountered in

Sec IVA for the model problem Similar to before we have retained Eqs (3842) as thechoice of numerical flux even at the endpoints Therefore at an endpoint the specificationof the boundary condition amounts to the choice W+ of external state We have typicallychosen the inner boundary of the radial domain as an excision boundary and in this caseW+ = Wminus is enforced at the inner physical boundary At the outer physical boundaryfor W+ we have again considered two choices (i) Wexact and (ii) Wupwind To enforcechoice (ii) the inverse transformation (17) must be used with incoming characteristic fieldsfixed to their exact values similar to (60) We have tried various versions of choice (ii)and in all cases the resulting simulations have been unstable We therefore present resultscorresponding to choice (i) Although the choice of an analytical external state Wexact

at the outer boundary is stable for our problem such a boundary condition is unlikelyto generalize to more complicated scenarios involving dynamical fields Indeed the issueof outer boundary conditions for the BSSN system is an active area of research with aproper treatment requiring fixation of incoming radiation control of the constraints andspecification of gauge (see Ref [86] for a recent analysis)

For BSSN simulations our main diagnostic is to monitor the Hamiltonian momentumand conformal connection constraints Figure 3 depicts long-time histories of constraintviolations whereas Figs 4 and 5 depict long-time error histories for the individual BSSNfield components From the middle plot in Fig 5 we infer that up to the indicated numericalerror the factor g sin4 θ = grr(gθθ)

2 remains at its initial fixed profile r4 throughout theevolution These figures indicate that the proposed scheme is stable for long times andexhibits spectral converge with increased polynomial order N Similar results are recoveredfromM = 0 Minkowski initial data The stability documented in these plots does not appearto rely on inordinate parameter tuning For example with the fixed parameters describedabove we obtain similar plots if we individually vary (i) rmin over 0325 035 04 0475(values still corresponding to an excision surface for the given choice of λ) (ii) η over1 3 7 10 (iii) s over 8 9 10 With the polynomial order N ranging over 23 26 29 31both stability and qualitatively similar exponential convergence is achieved with a singlesubdomain Likewise adoption of a larger coordinate domain with more subdomains doesnot impact our results Finally we have considered the addition of random noise to all fieldcomponents at the initial time Precisely with the system component χ as an example wehave set

χ(0) 7rarr χ(0) + δχ(0) (61)

where each component (nodal value) of δχ(0) is 10minus5 times a random variable drawn from a

19

10minus9

10minus7

10minus5

10minus3

Hinfin

10minus9

10minus7

10minus5

10minus3

Mrinfin

0 2000 4000 6000 8000 10000

10minus9

10minus7

10minus5

10minus3

Grinfin

t

N = 11N = 14N = 17N = 20

FIG 3 Spectral convergence of constraint violations for M = 1 Kerr-Schild ini-

tial data Respectively for N = 11 14 17 19 a timestep of ∆t ≃ 00041 00026 00018 00013

has been chosen for stability and accuracy

standard normal distribution Such perturbed initial data also gives rise to stable evolutions

V CONCLUSION

We have introduced a discontinuous Galerkin method for solving the spherically reducedBSSN system with second-order spatial operators Our scheme shares similarities withother discontinuous Galerkin methods that use local auxiliary variables to handle high-orderspatial derivatives [66 77ndash79 82 84 85] and which have typically been applied to eitherelliptic parabolic or mixed type problems The key ingredient of a stable dG scheme is anappropriate choice of numerical flux and our particular choice has been motivated by theanalysis presented in Appendix C When used to evolve the Schwarzschild solution in Kerr-Schild coordinates our numerical implementation of the BSSN system (4) is robustly stableand converges to the analytic solution exponentially with increased polynomial order Byapproximating the spatially second-order form of the BSSN system we have not introducedextra fields which are evolved Evolved auxiliary fields result in new constraints which mayspoil stability Our main goal has been stable evolution of the spherically reduced BSSNsystem as a first step towards understanding how a discontinuous Galerkin method might be

20

10minus9

10minus7

10minus5

∆Arrinfin

10minus9

10minus7

10minus5

∆Kinfin

0 2000 4000 6000 8000 10000

10minus9

10minus7

10minus5

max(∆Brinfin ∆Γrinfin)

t

N = 11N = 14N = 17N = 20

FIG 4 Spectral convergence of solution for M = 1 Kerr-Schild initial data

Timestep choices are described in the caption for Fig 3 In the title headings for example

∆Arr equiv (Arr)numer minus (Arr)exact

applied to the full BSSN system Towards that goal we now discuss treatment of singularitiesand generalization of the described dG method to higher space dimension

To deal with the fixed Schwarzschild singularity we have used excision which is easy inthe context of the spherically reduced BSSN system However excision for the binary blackhole problem in full general relativity requires attention to the technical challenge of hori-zon tracking State-of-the-art BSSN codes avoid such complication relying instead on themoving-puncture technique While the moving-puncture technique does involve mild centralsingularities it may still prove amenable to spectral methods Indeed spectral methods fornon-smooth problems is well-developed in both theory and for complex applications Sincethe moving-puncture technique can be performed in spherical symmetry [60] a first-steptoward a spectral moving-puncture code would be to implement a moving puncture withthe nodal dG method described here Such an implementation may adopt Legendre-Gauss-Radau nodes on the innermost subdomain thereby ensuring that the physical singularitydoes not lie on a nodal point (in much the same way finite difference codes use a staggeredgrid) Beyond traditional excision and moving punctures one might also retain a smoothproblem for which spectral methods are especially well-suited via the turducken approach[87 88] to singularities

21

10minus11

10minus9

10minus7

10minus5

∆χinfin

10minus11

10minus9

10minus7

10minus5

max(∆grrinfin ∆gθθinfin)

0 2000 4000 6000 8000 10000

10minus11

10minus9

10minus7

10minus5

max(∆αinfin ∆βrinfin)

t

N = 11N = 14N = 17N = 20

FIG 5 Spectral convergence of solution violations for M = 1 Kerr-Schild initial

data See the caption of Fig 4 for details

Discontinuous Galerkin methods for hyperbolic problems in two and three space dimen-sions are well-developed A generalization of the method described here to three-dimensionsand the full BSSN system would likely rely on an unstructured mesh Appropriate local poly-nomial expansions for the subelements are well-understood as are choices for the numericalfluxes which would now live on two-dimensional faces rather than single points Whether ornot it would ultimately prove successful generalization of our dG method to a higher dimen-sion would rely on an established conceptual framework Further computational advancesof relevance to a generalization of our dG method to the full BSSN system (possibly includ-ing matter) may include mesh hp-adaptivity local timestepping shock capturing and slopelimiting techniques [66] Moreover recent work [89] indicates that enhanced performancewould be expected were our scheme implemented on graphics processor units

VI ACKNOWLEDGMENTS

We thank Nick Taylor for discussions about treating second-order operators with spectralmethods Benjamin Stamm for helping polish up a few parts of Appendix C Khosro Shah-bazi for discussions on LDG and IP methods We also acknowledge helpful conversationswith David Brown and Manuel Tiglio about the BSSN system and previous work on its nu-

22

merical implementation We gratefully acknowledge funding through grants DMS 0554377and DARPAAFOSR FA9550-05-1-0108 to Brown University and NSF grant PHY 0855678to the University of New Mexico

Appendix A Hyperbolicity of the first-order system

This appendix analyzes the matrix A(u) appearing in (10) in order to construct thecharacteristic fields (14) In matrix form the sector (8) of the principal part of (10) reads asfollows

partt

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

=

βr 0 minus 4λα3grr

0 0 λβr

6(grr)2λβr

3gθθgrr0 4λ

3grr

0 βr 0 23grrαχ

13α minus αχ

3grr

αχ3gθθ

minus23χ 0

0 0 βr 0 0 0 0 minus χgrr

0

0 0 minus 4α3grr

βr 0 βr

6(grr)2βr

3gθθgrr0 4

3grr

0 0 23αχ 0 βr minus βrχ

3grrminus2βrχ

3gθθ0 minus2

3χ

0 minus2α 0 0 0 23βr minus2grrβr

3gθθ0 4

3grr

0 gθθαgrr

0 0 0 minusgθθβr

3grr13βr 0 minus2

3gθθ

0 0 minus2α 0 0 0 0 βr 0

34

0 0 0 0 0 0 0 βr

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

prime

(A1)which defines the matrix A(u) appearing in (9) and so also the matrix A(u) in (10) Notethat in the last equation the matrix within the square brackets is minusA(u) For certainconfigurations of u and λ the system (10) is strongly hyperbolic [76] that is A(u) hasa complete set of eigenvectors and real eigenvalues Indeed five eigenpairs of A(u) aretrivially recovered upon inspection of A(u)rsquos leading 5times5 diagonal block These correspondto eigenvalue 0 and the left eigenspace ξj = eTj 1 le j le 5 where ej are the canonical

basis vectors Since each component of u arises as eTj W each is also a characteristic field

The remaining nine eigenpairs are determined by A(u) The eigenvalues of A(u) are

micro1 = 0 micro23 = minusβr microplusmn4 = minusβr plusmn

radic

2αχ

grr microplusmn

5 = minusβr plusmn α

radic

χ

grr microplusmn

6 = minusβr plusmnradic

λ

grr

(A2)

23

and the corresponding left eigenvectors are

x1 = (0 0 0 0 0 gθθ 2grr 0 0) (A3a)

x2 =

(

0 0 0 grr2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3b)

x3 =

(

grrλ 0 0 0

2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3c)

xplusmn4 =

(

0 0plusmnradic

2αgrrχ

0 0 0 0 1 0

)

(A3d)

xplusmn5 =

(

0∓ 3radicgrrχ

plusmn2

radic

grrχ 2grr

1

χminus 1

grr1

gθθ 0 0

)

(A3e)

xplusmn6 =

(

minus3

4

grrλ 0plusmn α

radicλgrr

(2αχminus λ) 0 0minus βr

8(βrgrr ∓radicλgrr)

minus βrgrr

4gθθ(βrgrr ∓radicλgrr)

αχ

(2αχminus λ)plusmnradic

grrλ

)

(A3f)

where for example xplusmn5 A(u) = microplusmn5 x

plusmn5 Assuming that grr gθθ χ and α are everywhere strictly

positive the eigenvalues are real and the eigenvectors are linearly independent providedthat (15) holds These eigenvectors are easily extended to eigenvectors of A(u) e g asxplusmn6 rarr (01times5 x

plusmn6 ) Then for example the characteristic field

Xplusmn6 equiv (01times5 x

plusmn6 )W = xplusmn6WvQ (A4)

and similarly Xplusmnj = xplusmnj WvQ for j = 4 5 and Xk = xkWvQ for k = 1 2 3 The characteristic

speeds for these fields are microk and microplusmnj With this convention the speeds listed in Table I

correspond to the Xk and Xplusmnj in (14)

Appendix B Schwarzschild solution in conformal Kerr-Schild coordinates

In Kerr-Schild coordinates here the system directly related to incoming Eddington-Finkelstein null coordinates the line element for the Schwarzschild solution reads

ds2 = minusα2dt2 + (1 + 2MR)(dR + βRdt)2 +R2dθ2 +R2 sin2 θdφ2 (B1)

where R is the area radius α = (1 + 2MR)minus12 is the lapse and βR = 2M(R + 2M) isthe shift vector The physical spatial metric gab is the spatial part of this line element

To define the corresponding solution to the BSSN system we use equation gab = χgab todefine the following relationship between line elements

dr2 + r2(dθ2 + sin2 θdφ2) = χ[(1 + 2MR)dR2 +R2dθ2 +R2 sin2 θdφ2] (B2)

so that

χ

(

1 +2M

R

)(

dR

dr

)2

= 1 χR2 = r2 (B3)

Then we have(

1 +2M

R

)12dR

R=dr

r (B4)

24

with integration yielding

r =R

4

(

1 +

radic

1 +2M

R

)2

e2minus2radic

1+2MR (B5)

where the constant of integration has been chosen so that the R r rarr infin limits are consistentThe second relation in (B3) shows that

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR =rArr χminus4 =2eradic

1+2MRminus1

1 +radic

1 + 2MR (B6)

The extrinsic curvature tensor is specified by the expression forK given in (B9h) the identityK = KR

R + 2Kθθ and

Kθθ =

(

1 +2M

R

)minus122M

R2 (B7)

Since KRR = Kr

r we compute that

Krr = K minus 2Kθ

θ = minus(

1 +2M

R

)minus12(R +M

R + 2M

)

2M

R2 (B8)

Next since Krr = grrKrr = χminus1Kr

r we have Krr = Arr+

13grrK This implies Arr = Kr

r minus 13K

from which we get (B9g) In all we have

α =

(

1 +2M

R

)minus12

(B9a)

βr = βR dr

dR= χ12

(

1 +2M

R

)minus122M

R(B9b)

grr = 1 (B9c)

gθθ = r2 = χR2 (B9d)

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR (B9e)

Br = 0 (B9f)

Arr = minus(

1 +2M

R

)minus124M

3R2

(

2R + 3M

R + 2M

)

(B9g)

K =

(

1 +2M

R

)minus32(

1 +3M

R

)

2M

R2(B9h)

Γr = minus2

r= minus 2

χ12R (B9i)

To differentiate these expressions with respect to r we use the identity

dR

dr= χminus12

(

1 +2M

R

)minus12

(B10)

along with the chain rule

25

Appendix C Stability of the model system

The following stability analysis for the model system (53) has been inspired by [78 79]After dropping all nonlinear source terms the system (53) becomes

parttu = uprime + av (C1a)

parttv = uprimeprime + vprime (C1b)

This section analyzes the stability of (C1) considering both the continuum system itself aswell as its semi-discrete dG approximation The latter analysis offers some insight into theempirically observed stability of our dG scheme for the spherically reduced BSSN equations

1 Analysis for a single interval

Throughout we work with the L2 inner product and norm

(f g)D =

int

D

fg fD =radic

(f f)D (C2)

where D is the spatial coordinate interval (here D may represent a subdomain Dk or the

whole domain Ω) and we have suppressed all integration measures For the continuummodel we will establish the following estimate

uprime(T middot)2D+ av(T middot)2

Dle C(T )

(

uprime(0 middot)2D+ av(0 middot)2

D

)

(C3)

where the time-dependent constant C(T ) is determined solely by the choice of boundaryconditions To show (C3) we first change variables with v =

radicav thereby rewriting (C1)

in the following symmetric form

parttu = uprime +radicav (C4a)

parttv =radicauprimeprime + vprime (C4b)

Equations (C4ab) then imply

1

2partt

int

D

(uprime)2 =

int

D

uprime(uprimeprime +radicavprime) =

int

D

radicauprimevprime +

1

2

int

partD

(uprime)2 (C5a)

1

2partt

int

D

(v)2 =

int

D

v(radicauprimeprime + vprime) = minus

int

D

radicauprimevprime +

1

2

int

partD

(v2 + 2radicauprimev) (C5b)

Here vvprime and uprimeuprimeprime have been expressed as exact derivatives and then integrated to boundaryterms the second equation employs an extra integration by parts and with only one spacedimension

int

partDdenotes a difference of endpoint evaluations Addition of Eqs (C5ab) gives

1

2partt

int

D

[

v2 + (uprime)2]

=1

2

int

partD

[

v2 + (uprime)2 + 2radicauprimev

]

(C6)

Substitutions with the identities

[

v2 + (uprime)2]

=1

2

[

(v + uprime)2 + (v minus uprime)2]

2uprimev =1

2

[

(v + uprime)2 minus (v minus uprime)2]

(C7)

26

and replacements to recover the original variable v = vradica yield

1

2partt

int

D

[

av2 + (uprime)2]

=1 +

radica

4

int

partD

(radicav + uprime)2 +

1minusradica

4

int

partD

(radicav minus uprime)2 (C8)

From (C8) we deduce that the time-dependent constant C(T ) in (C3) must satisfy

∣

∣

∣

∣

∣

∣

1 +

int T

0

[

1+radica

2

int

partD(radicav + uprime)2 + 1minus

radica

2

int

partD(radicav minus uprime)2

]

dt

uprime(0 middot)2 + av(0 middot)2

∣

∣

∣

∣

∣

∣

le C(T ) (C9)

For periodic boundary conditions we may choose C(T ) = 1 Moreover if a ge 1 anduprime = minusradic

av is specified at partD+ then uprime(t middot)2 + av(t middot)2 decaysStill working on a single interval (subdomain) we now consider the semi-discrete scheme

for (C4) i e (57) with all nonlinear source terms dropped and with v replaced by vradica

Derivation of a formula analogous to (C8) is our first step toward establishing L2 stabilityof the semi-discrete scheme While (57) features vectors for example u(t) taking values atthe Legendre-Gauss-Lobatto nodal points here we work with the numerical solution as apolynomial for example uh(t x) These two representations are related by the Lagrange in-terpolating polynomials for the nodal set here taken to span both the space of test functionsand the space of basis functions Our scheme is

int

Dk

ψparttuh =

int

Dk

ψ(Qh +radicavh) (C10a)

int

Dk

ξparttvh = minusint

Dk

ξprime(radicaQh + vh) +

int

partDk

ξ(radicaQlowast + vlowast) (C10b)

int

Dk

ϕQh =

int

Dk

ϕuprimeh +

int

partDk

ϕ (ulowast minus uh) (C10c)

where ψ ξ and ϕ are polynomial test functions These test functions are arbitrary exceptthat they must be degree-N polynomials In (C10) the variables uh vh and Qh should alsocarry a superscript k but we have suppressed this Derivation of a formula analogous to(C8) is complicated by the fact that Qh is not evolved Nevertheless at a given instant t wecan assemble Qh from (C10c)

Mimicking the calculation (C5b) from the continuum case we first use (C10b) with ξ = vhto write

1

2partt

int

Dk

v2h = minusint

Dk

(radicaQh + vh)v

primeh +

int

partDk

(radicaQlowast + vlowast)vh

= minusint

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2(radicaQlowast + vlowast)vh minus v2h

]

(C11)

The right-hand side of (C5a) is analogous to

1

2partt

int

Dk

Q2h =

int

Dk

QhparttQh (C12)

However since Qh is not evolved the term parttQh must be given a suitable interpretationOn the right hand side of (C10c) only uh u

primeh and u

lowast necessarily depend on time since the

27

test function ϕ need not be time-dependent Furthermore ulowast is explicitly given as a linearcombination of uh as seen in Eq (C21c) below Choosing ϕ = ℓj taking the time derivativeof (C10c) and appealing to the commutivity of mixed partial derivatives we therefore arriveat

int

Dk

ℓjparttQh =

int

Dk

ℓj(parttuh)prime +

int

partDk

ℓj(

(parttu)lowast minus parttuh

)

(C13)

where (parttu)lowast depends on parttuh in precisely the same way that ulowast depends on uh We have

written ℓj rather than ϕ in the last equation to emphasize that the result also holds for anylinear combination of ℓj (for example ϕ) and even for time-dependent combinations SinceQh is itself such a combination we obtain

1

2partt

int

Dk

Q2h =

int

Dk

Qh(parttuh)prime +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

Qh(Qprimeh +

radicavprimeh) +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2((parttu)lowast minus parttuh)Qh +Q2

h

]

(C14)

Addition of (C11) and (C14) gives

1

2partt

int

Dk

(Q2h + v2h) =

1

2

int

partDk

[

Q2h minus v2h + 2(

radicaQlowast + vlowast)vh + 2((parttu)

lowast minus parttuh)Qh

]

(C15)

the aforementioned analog of (C8) This formula holds on a single subdomain Dk and we

now combine multiple copies of it one for each value of k

2 Analysis for multiple intervals

To facilitate combination of (C15) over all k we change notation At every subdomaininterface Ik+12 equiv partDk cap partDk+1 let the superscripts L and R denote field values respectivelytaken from the left and right Then the fields evaluated at Ik+12 which belong to D

k will beuLk+12 v

Lk+12 and Q

Lk+12 while those belonging to D

k+1 will be uRk+12 vRk+12 and Q

Rk+12

However at Ikminus12 the values taken from D

k are uRkminus12 vRkminus12 and QR

kminus12 Note that we

have also replaced the subscript h denoting a numerical solution with k plusmn 12 denotingthe location of the endpoint value of the numerical solution With this notation we define

∆Lα =

1

2

[

(QLα)

2 minus (vLα)2]

+(radic

aQlowastα + vlowastα

)

vLα +[

(parttuα)lowast minus parttu

Lα

]

QLα (C16)

and similarly for ∆Rα The same numerical fluxes appear in both ∆L

α and ∆Rα (ie each

numerical flux takes the same value on either side of an interface) whence fluxes like Qlowastα do

not carry an L or R superscript In terms of these definitions (C15) becomes

1

2partt

int

Dk

(Q2h + v2h) = ∆L

k+12 minus∆Rkminus12 (C17)

28

Summation over all Dk yields

1

2partt

kmaxsum

k=1

int

Dk

(Q2h + v2h) =

kmaxminus1sum

k=1

(

∆Lk+12 minus∆R

k+12

)

+∆Lkmax+12 minus∆R

12

=

kmaxminus1sum

k=1

(

∆Lh minus∆R

h

)∣

∣

Ik+12 +∆Lkmax+12 minus∆R

12 (C18)

We have reverted to h-notation denoting the numerical solution since the LR superscriptsindicate unambiguously the relevant domain used for evaluation at Ik+12

We again seek an estimate of the form

kmaxsum

k=1

(

Qh(T middot)2Dk + avh(T middot)2Dk

)

le C(T )kmaxsum

k=1

(

Qh(0 middot)2Dk + avh(0 middot)2Dk

)

(C19)

that is essentially the same as the one (C3) considered in the continuum case Assume thatthe chosen boundary conditions ensure ∆L

kmax+12 minus ∆R12 is bounded by a time-dependent

constant which does not depend on the numerical parameters N and h (subdomain width)Establishment of stability then amounts to showing that the remaining sum over interfaceterms in (C18) is non-positive whence this remaining sum is consistent with C(T ) le 1although the boundary conditions may give rise to a different bound In fact we will choosethe numerical fluxes such that each individual interface term is non-positive At interfaceIk+12 and in LR notation the jump and average of vh for example are

1

2

(

v+ + vminus)

equiv vh =1

2

(

vLk+12 + vRk+12

)

(C20a)

nminusvminus + n+v+ equiv[[

vh]]

= vLk+12 minus vRk+12 (C20b)

Consider numerical fluxes of the form

Qlowast = Qh minusτQ2

[[

Qh

]]

(C21a)

vlowast = vh minusτv2

[[

vh]]

(C21b)

ulowast = uh minusτu2

[[

uh]]

(C21c)

(parttu)lowast = parttuh minus

τu2

[[

parttuh]]

(C21d)

where (C21c) induces (C21d) and where the penalty parameters τu τv and τQ are realnumbers The fluxes defined in (58) correspond to τu = 1 τv = 1 +

radica and τQ = 0 In

terms of these quantities the kth interface contribution in (C18) is

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

parttuh]]

minus τu2

[[

parttuh]][[

Qh

]]

(C22)

where we have suppressed the k dependence of the right-hand side Now consider the term[[

parttuh]]

Because parttuh and Qh+radicavh are both polynomials of degree N Eq (C10a) implies

29

τu

τ Q

minus1 0 1minus1

minus05

0

05

1

τu

minus1 0 1minus1

minus05

0

05

1

FIG 6 Stable evolutions for the model system For fixed τv = 10minus6 and τv = 1 +radic2

respectively the left and right plots depict stable choices (determined empirically) of τu and τQfor the linear model system (C1) The stable regions are colored black but the jagged edges result

from the discretization of the (τu τQ)-plane

the vector equation parttu = Q +radicav that is pointwise equivalence on the nodal points of

Dk which in turn implies

[[

parttuh]]

=[[

Qh+radicavh]]

Upon substituting this identity into thelast equation we arrive at an expression which features only vh and Qh

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

Qh +radicavh]]

minus τu2

[[

Qh +radicavh]][[

Qh

]]

(C23)

The identities vh[[

vh]]

= 12

[[

v2h]]

and[[

Qh +radicavh]]

=[[

Qh

]]

+radica[[

vh]]

then simplify(C23) to

(∆Lh minus∆R

h )∣

∣

Ik+12 = minusτv2

[[

vh]]2 minus

radica(τu + τQ)

2

[[

Qh

]][[

vh]]

minus τu2

[[

Qh

]]2 (C24)

The role of a penalty parameter is now clear Positive values of τv penalize jumps in vhthrough a negative contribution to the energy Likewise positive values of τu penalizejumps in Qh through a negative contribution to the energy However because the sign of[[

Qh

]][[

vh]]

can be positive or negative only the choice τQ = minusτu yields an expression for(∆L

h minus ∆Rh )|Ik+12 which is manifestly negative for τu ge 0 and τv ge 0 A simple estimate

based on Youngrsquos inequality with ε (that is 2αβ le εminus1α2 + εβ2 where α β ge 0 and ε gt 0)shows that for τQ = 0 the choice τv ge aτu4 also yields stability

Figure 6 depicts certain choices of stable penalty parameters for the linear model systemevolved to tfinal = 1000 (with a = 2 N = 10 and ∆t ≃ 00553) as determined empiricallywith simulations similar to those described in Sec IVA The left plot corresponds to asmall τv = 10minus6 for which the choice τu = 1 τQ = 0 is not stable as expected from thetheoretical analysis However the right plot corresponds to τv = 1 +

radica for which τu = 1

τQ = 0 is stable Motivated by the numerical flux choices (3842) used for the BSSN system

30

(4) we have (as mentioned above) set τu = 1 τv = 1 +radica and τQ = 0 in simulations of

the nonlinear model (53) For the nonlinear model system (53) the theoretically motivatedchoice τQ = minusτu also yields numerically stable evolutions when τu ge 0 and τv ge 0

For the nonlinear systems (4) and (53) we do not attempt a formal stability proofNevertheless the results of this appendix have served as a guide for our choices of penaltyparameters For the BSSN system (4) u v and Q are block indices [cf Eq (6)] Similarto the model problem we have penalized Q with τu = 1 with τv chosen large enough toheuristically overcome the cross-terms of indefinite size that arise from τQ = 0 (we interpretequations like τu = 1 componentwise) An analogous choice ldquoτQ = minusτurdquo for the BSSNsystem might be possible but would be considerably more complicated Indeed such achoice likely entails a matrix of penalty parameters but we do not give the details here

[1] F Pretorius Evolution of Binary Black-Hole Spacetimes Phys Rev Lett 95 121101 (2005)

(4 pages)

[2] F Pretorius Simulation of binary black hole spacetimes with a harmonic evolution scheme

Class Quantum Grav 23 S529-S552 (2006)

[3] M Campanelli C O Lousto P Marronetti and Y Zlochower Accurate Evolutions of Or-

biting Black-Hole Binaries without Excision Phys Rev Lett 96 111101 (2006) (4 pages)

[4] J G Baker J Centrella D-I Choi M Koppitz and J van Meter Gravitational-Wave

Extraction from an Inspiraling Configuration of Merging Black Holes Phys Rev Lett 96

111102 (2006) (4 pages)

[5] M Campanelli C O Lousto and Y Zlochower Last orbit of binary black holes Phys Rev D

73 061501(R) (2006) (5 pages)

[6] F Herrmann I Hinder D Shoemaker and P Laguna Unequal mass binary black hole plunges

and gravitational recoil Class Quantum Grav 24 S33-S42 (2007)

[7] P Diener F Herrmann D Pollney E Schnetter E Seidel R Takahashi J Thornburg and

J Ventrella Accurate Evolution of Orbiting Binary Black Holes Phys Rev Lett 96 121101

(2006) (4 pages)

[8] M A Scheel H P Pfeiffer L Lindblom L E Kidder O Rinne and S A Teukolsky Solving

Einsteinrsquos Equations With Dual Coordinate Frames Phys Rev D 74 104006 (2006)

[9] U Sperhake Binary black-hole evolutions of excision and puncture data Phys Rev D 76

104015 (2007) (20 pages)

[10] B Brugmann J A Gonzalez M Hannam S Husa U Sperhake and W Tichy Calibration

of moving puncture simulations Phys Rev D 77 024027 (2008) (25 pages)

[11] P Marronetti W Tichy B Brugmann J Gonzalez M Hannam S Husa and U Sperhake

Binary black holes on a budget simulations using workstations Class Quantum Grav 24

S43-S58 (2007)

[12] Z B Etienne J A Faber Y T Liu S L Shapiro and T W Baumgarte Filling the holes

Evolving excised binary black hole initial data with puncture techniques Phys Rev D 76

101503(R) (2007) (5 pages)

[13] B Szilagyi D Pollney L Rezzolla J Thornburg and J Winicour An explicit harmonic code

for black-hole evolution using excision Class Quantum Grav 24 S275-S293 (2007)

[14] L Boyle M Kesden and S Nissanke Binary-Black-Hole Merger Symmetry and the Spin

Expansion Phys Rev Lett 100 151101 (2008) (4 pages)

31

[15] M A Scheel M Boyle T Chu L E Kidder K D Matthews and H P Pfeiffer High-

accuracy waveforms for binary black hole inspiral merger and ringdown Phys Rev D 79

024003 (2009) (14 pages)

[16] M Hannam Status of black-hole-binary simulations for gravitational-wave detection

Class Quantum Grav 26 114001 (2009) (21 pages)

[17] I Hinder The current status of binary black hole simulations in numerical relativity

Class Quantum Grav 27 114004 (2010) (19 pages)

[18] B P Abbott et al LIGO The Laser Interferometer Gravitational-Wave Observatory

Rep Prog Phys 72 076901 (2009) (25 pages)

[19] F Acernese et al The status of VIRGO Class Quantum Grav 23 S63-S69 (2006)

[20] F Acernese et al Virgo status Class Quantum Grav 25 184001 (2008) (9 pages)

[21] R Arnowitt S Deser and C W Misner The Dynamics of General Relativity in Gravitation

an introduction to current research edited by L Witten (Wiley New York 1962) Chapter 7

227-265 Also available as gr-qc0405109

[22] J W York Jr Kinematics and Dynamics of General Relativity in Sources of Gravitational

Radiation edited by L L Smarr (Cambridge University Press Cambridge 1979)

[23] L E Kidder M A Scheel and S A Teukolsky Extending the lifetime of 3D black hole

computations with a new hyperbolic system of evolution equations Phys Rev D 64 064017

(2001) (13 pages)

[24] G Calabrese J Pullin O Sarbach and M Tiglio Convergence and stability in numerical

relativity Phys Rev D 66 041501(R) (2002) (4 pages)

[25] S Frittelli and O Reula On the Newtonian Limit of General Relativity Commun Math

Phys 166 221-235 (1994)

[26] M Shibata and T Nakamura Evolution of three-dimensional gravitational waves Harmonic

slicing case Phys Rev D 52 5428-5444 (1995)

[27] Y Choquet-Bruhat and J W York Geometrical well posed systems for the Einstein equations

C R Acad Sci Paris t 321 Serie I 1089-1095 (1995)

[28] A Abrahams A Anderson Y Choquet-Bruhat and J W York Jr Einstein and Yang-Mills

Theories in Hyperbolic Form without Gauge Fixing Phys Rev Lett 75 3377-3381 (1995)

[29] C Bona J Masso E Seidel and J Stela New Formalism for Numerical Relativity

Phys Rev Lett 75 600-603 (1995)

[30] M H P M van Putten and D M Eardley Nonlinear wave equations for relativity

Phys Rev D 53 3056-3063 (1996)

[31] S Frittelli and O A Reula First-Order Symmetric Hyperbolic Einstein Equations with Ar-

bitrary Fixed Gauge Phys Rev Lett 76 4667-4670 (1996)

[32] H Friedrich Hyperbolic reductions for Einsteinrsquos equations Class Quantum Grav 13 1451-

1469 (1996)

[33] F B Estabrook R S Robinson H D Wahlquist Hyperbolic equations for vacuum gravity

using special orthonormal frames Class Quantum Grav 14 1237-1247 (1997)

[34] M S Iriondo E O Leguizamon and O A Reula Einsteinrsquos Equations in Ashtekarrsquos Vari-

ables Constitute a Symmetric Hyperbolic System Phys Rev Lett 79 4732-4735 (1997)

[35] A Anderson Y Choquet-Bruhat and J W York Einstein-Bianchi hyperbolic system for

general relativity Topol Meth Nonlinear Anal 10 353-373 (1997)

[36] T W Baumgarte and S L Shapiro Numerical integration of Einsteinrsquos field equations

Phys Rev D 59 024007 (1998) (7 pages)

[37] M A G Bonilla Symmetric hyperbolic systems for Bianchi equations Class Quantum

15

E Model system

To better illustrate the basic properties of our method we consider a toy model Namelythe following spatially second-order system

parttu = uprime + av minus u3 + g(t x) (53a)

parttv = uprimeprime + vprime minus (u+ v)(uprime)2 + v2u2 + h(t x) (53b)

where a ge 1 is constant and g and h are analytic source terms to be specified In contrastto (6) here u v and Q = uprime are scalars rather than vectors System (53) admits a first-order reduction in which uprime is defined as an extra variable Since this first-order reductionis strongly hyperbolic the spatially second-order system (53) is also strongly hyperbolic byone of the definitions considered in [75] The characteristic fields Xplusmn and speeds microplusmn are

X+ =radicav minus uprime micro+ =

radicaminus 1 Xminus =

radicav + uprime microminus = minus(

radica+ 1) (54)

To construct a local dG scheme for this system we first rewrite it as

parttu = Q+ av minus u3 + g(t x) (55a)

parttv = Qprime + vprime minus (u+ v)Q2 + v2u2 + h(t x) (55b)

Q = uprime (55c)

Evidently f = minus(Q + v) is the v-component of the physical flux vector

F (vQ) equiv(

Fu

Fv

)

=

(

0

f

)

(56)

Note that F has the same structure as (u v)T Borrowing from the presentation for theBSSN system we write the analogous semidiscrete scheme on each subdomain D

k for themodel system

parttu = Q+ av minus u3 + g(t) (57a)

parttv = DQ+Dv minus (u+ v)Q2 + v2u2 + h(t) +Mminus1ℓk(fh minus f lowast)∣

∣

∣

bk

ak (57b)

Q = Du+Mminus1ℓk(ulowast minus uh)∣

∣

∣

bk

ak (57c)

Here we have suppressed the subinterval label k from all variables except for the vectorℓk of Lagrange polynomial values Moreover following the guidelines discussed above thenumerical fluxes are given by

f lowast = fh+1 +

radica

2

[[

vh]]

ulowast = uh minus1

2

[[

uh]]

(58)

Appendix C analyzes the stability of our scheme for a more general numerical flux choiceas applied to (53) with the nonlinear and source terms dropped

IV RESULTS FROM NUMERICAL SIMULATIONS

This section presents results found by numerically solving both the model system (53)and BSSN system (4) with the dG scheme presented in Sec III

16

10minus10

10minus6

10minus2

∆uinfin

10minus10

10minus6

10minus2

∆vinfin

0 500 1000 1500 2000 2500 3000 3500 400010

minus10

10minus6

10minus2

∆(partxu)infin

t

N = 3N = 6N = 9N = 12

FIG 1 Spectral convergence of fields for model PDE Respectively for N = 3 6 9 12

a timestep of ∆t = 00578 00178 00084 00049 has been chosen for stability and accuracy In the

title headings for example ∆u equiv unumer minus uexact

A Simulations of the model system

The semi-discrete scheme (57) has been integrated with the classical fourth-order Runge-Kutta method When integrating this system we have first constructed Q at each Runge-Kutta stage and then substituted into the evolution equations (57ab) for u and v Theproblem has been solved on a computational domain [0 4π] comprised of two subdomainswith a timestep chosen small enough for stability The initial data has been taken from thefollowing exact solution to (53)

uprimeexact(t x) =1

2

[

sin(xminus microminust)minus sin(xminus micro+t)]

(59a)

vexact(t x) =1

2radica

[

sin(xminus microminust) + sin(xminus micro+t)]

(59b)

g(t x) = u3exact (59c)

h(t x) = (uexact + vexact)(uprimeexact)

2 minus v2exactu2exact (59d)

where the speeds microplusmn are found in (54) Specification of the boundary condition at a physicalendpoint amounts to choosing the external state for at the endpoint We have considered

17

8 16 32 64 128

10minus3

10minus2

10minus1

N

∆t

max stable ∆t8N2

FIG 2 Scaling of maximum stable ∆t with N for model PDE

two possibilities (i) the analytic state (Q+ v+) = (Qexact vexact) and (ii) an upwind stateFor example at x = 4π the upwind state is4

Q+ = Qupwind =1

2

[

(Xminus)exact minus (X+)numer

]

v+ = vupwind =1

2radica

[

(Xminus)exact + (X+)numer

]

(60)Either choice of (Q+ v+) leads to similar results and the plots here correspond to theanalytic state Figure 1 clearly shows spectral convergence with increasing polynomial orderN across all fields for the case a = 2 Other values of a including a = 1 for which X+

is a static characteristic field have also been considered with similar results Appendix Cdemonstrates that our proposed scheme for the system (57) with nonlinear and source termsdropped is stable in a semi-discrete sense Nevertheless the fully discrete scheme obtainedvia temporal discretization by the fourth-order Runge-Kutta method is still subject to thestandard absolute stability requirement Namely if microh is any eigenvalue corresponding tothe (linearized) discrete spatial operator then a necessary condition for stability is thatmicroh∆t lies in absolute stability region for Runge-Kutta 4 We here show empirically that theassociated timestep restriction scales like Nminus2 ie ∆t = O(Nminus2) for stability We note thatsuch scaling is welcome in light of the second-order spatial operators which appear in thesystem and suggest a possible worse scaling like Nminus4 Fig 2 plots the maximum stabletimestep for a range of N demonstrating the Nminus2 scaling in line with behavior known fromanalysis of first-order systems [66] This scaling also holds for the BSSN system

4 We remind the reader that unfortunately the plusmn on Xplusmn means something different than the plusmn indicating

exteriorinterior dG states [cf footnote 3]

18

B Simulations of the BSSN system

This subsection documents results for simulations of the unit-mass-parameter (M = 1)Schwarzschild solution (B9) expressed in terms of ingoing Kerr-Schild coordinates Sincethe solution is stationary temporal integration of the semi-discrete scheme has been carriedout with the forward Euler method The r-coordinate domain [04 34] has been split into3 equally spaced subdomains and we have set η = 10 λ = 01 and C = 2 [cf Eq (41)]For all simulations ∆t has been chosen for stability With the chosen λ the inner physicalboundary rmin = 04 is an excision surface At each timestep we have applied an (order2s = 20) exponential filter on the top two-thirds of the modal coefficient set for all fieldsexcept for grr and gθθ For stability we have empirically observed that grr and gθθ must not

be filtered A detailed understanding of this is still lackingIssues related to physical boundary conditions are similar to the one encountered in

Sec IVA for the model problem Similar to before we have retained Eqs (3842) as thechoice of numerical flux even at the endpoints Therefore at an endpoint the specificationof the boundary condition amounts to the choice W+ of external state We have typicallychosen the inner boundary of the radial domain as an excision boundary and in this caseW+ = Wminus is enforced at the inner physical boundary At the outer physical boundaryfor W+ we have again considered two choices (i) Wexact and (ii) Wupwind To enforcechoice (ii) the inverse transformation (17) must be used with incoming characteristic fieldsfixed to their exact values similar to (60) We have tried various versions of choice (ii)and in all cases the resulting simulations have been unstable We therefore present resultscorresponding to choice (i) Although the choice of an analytical external state Wexact

at the outer boundary is stable for our problem such a boundary condition is unlikelyto generalize to more complicated scenarios involving dynamical fields Indeed the issueof outer boundary conditions for the BSSN system is an active area of research with aproper treatment requiring fixation of incoming radiation control of the constraints andspecification of gauge (see Ref [86] for a recent analysis)

For BSSN simulations our main diagnostic is to monitor the Hamiltonian momentumand conformal connection constraints Figure 3 depicts long-time histories of constraintviolations whereas Figs 4 and 5 depict long-time error histories for the individual BSSNfield components From the middle plot in Fig 5 we infer that up to the indicated numericalerror the factor g sin4 θ = grr(gθθ)

2 remains at its initial fixed profile r4 throughout theevolution These figures indicate that the proposed scheme is stable for long times andexhibits spectral converge with increased polynomial order N Similar results are recoveredfromM = 0 Minkowski initial data The stability documented in these plots does not appearto rely on inordinate parameter tuning For example with the fixed parameters describedabove we obtain similar plots if we individually vary (i) rmin over 0325 035 04 0475(values still corresponding to an excision surface for the given choice of λ) (ii) η over1 3 7 10 (iii) s over 8 9 10 With the polynomial order N ranging over 23 26 29 31both stability and qualitatively similar exponential convergence is achieved with a singlesubdomain Likewise adoption of a larger coordinate domain with more subdomains doesnot impact our results Finally we have considered the addition of random noise to all fieldcomponents at the initial time Precisely with the system component χ as an example wehave set

χ(0) 7rarr χ(0) + δχ(0) (61)

where each component (nodal value) of δχ(0) is 10minus5 times a random variable drawn from a

19

10minus9

10minus7

10minus5

10minus3

Hinfin

10minus9

10minus7

10minus5

10minus3

Mrinfin

0 2000 4000 6000 8000 10000

10minus9

10minus7

10minus5

10minus3

Grinfin

t

N = 11N = 14N = 17N = 20

FIG 3 Spectral convergence of constraint violations for M = 1 Kerr-Schild ini-

tial data Respectively for N = 11 14 17 19 a timestep of ∆t ≃ 00041 00026 00018 00013

has been chosen for stability and accuracy

standard normal distribution Such perturbed initial data also gives rise to stable evolutions

V CONCLUSION

We have introduced a discontinuous Galerkin method for solving the spherically reducedBSSN system with second-order spatial operators Our scheme shares similarities withother discontinuous Galerkin methods that use local auxiliary variables to handle high-orderspatial derivatives [66 77ndash79 82 84 85] and which have typically been applied to eitherelliptic parabolic or mixed type problems The key ingredient of a stable dG scheme is anappropriate choice of numerical flux and our particular choice has been motivated by theanalysis presented in Appendix C When used to evolve the Schwarzschild solution in Kerr-Schild coordinates our numerical implementation of the BSSN system (4) is robustly stableand converges to the analytic solution exponentially with increased polynomial order Byapproximating the spatially second-order form of the BSSN system we have not introducedextra fields which are evolved Evolved auxiliary fields result in new constraints which mayspoil stability Our main goal has been stable evolution of the spherically reduced BSSNsystem as a first step towards understanding how a discontinuous Galerkin method might be

20

10minus9

10minus7

10minus5

∆Arrinfin

10minus9

10minus7

10minus5

∆Kinfin

0 2000 4000 6000 8000 10000

10minus9

10minus7

10minus5

max(∆Brinfin ∆Γrinfin)

t

N = 11N = 14N = 17N = 20

FIG 4 Spectral convergence of solution for M = 1 Kerr-Schild initial data

Timestep choices are described in the caption for Fig 3 In the title headings for example

∆Arr equiv (Arr)numer minus (Arr)exact

applied to the full BSSN system Towards that goal we now discuss treatment of singularitiesand generalization of the described dG method to higher space dimension

To deal with the fixed Schwarzschild singularity we have used excision which is easy inthe context of the spherically reduced BSSN system However excision for the binary blackhole problem in full general relativity requires attention to the technical challenge of hori-zon tracking State-of-the-art BSSN codes avoid such complication relying instead on themoving-puncture technique While the moving-puncture technique does involve mild centralsingularities it may still prove amenable to spectral methods Indeed spectral methods fornon-smooth problems is well-developed in both theory and for complex applications Sincethe moving-puncture technique can be performed in spherical symmetry [60] a first-steptoward a spectral moving-puncture code would be to implement a moving puncture withthe nodal dG method described here Such an implementation may adopt Legendre-Gauss-Radau nodes on the innermost subdomain thereby ensuring that the physical singularitydoes not lie on a nodal point (in much the same way finite difference codes use a staggeredgrid) Beyond traditional excision and moving punctures one might also retain a smoothproblem for which spectral methods are especially well-suited via the turducken approach[87 88] to singularities

21

10minus11

10minus9

10minus7

10minus5

∆χinfin

10minus11

10minus9

10minus7

10minus5

max(∆grrinfin ∆gθθinfin)

0 2000 4000 6000 8000 10000

10minus11

10minus9

10minus7

10minus5

max(∆αinfin ∆βrinfin)

t

N = 11N = 14N = 17N = 20

FIG 5 Spectral convergence of solution violations for M = 1 Kerr-Schild initial

data See the caption of Fig 4 for details

Discontinuous Galerkin methods for hyperbolic problems in two and three space dimen-sions are well-developed A generalization of the method described here to three-dimensionsand the full BSSN system would likely rely on an unstructured mesh Appropriate local poly-nomial expansions for the subelements are well-understood as are choices for the numericalfluxes which would now live on two-dimensional faces rather than single points Whether ornot it would ultimately prove successful generalization of our dG method to a higher dimen-sion would rely on an established conceptual framework Further computational advancesof relevance to a generalization of our dG method to the full BSSN system (possibly includ-ing matter) may include mesh hp-adaptivity local timestepping shock capturing and slopelimiting techniques [66] Moreover recent work [89] indicates that enhanced performancewould be expected were our scheme implemented on graphics processor units

VI ACKNOWLEDGMENTS

We thank Nick Taylor for discussions about treating second-order operators with spectralmethods Benjamin Stamm for helping polish up a few parts of Appendix C Khosro Shah-bazi for discussions on LDG and IP methods We also acknowledge helpful conversationswith David Brown and Manuel Tiglio about the BSSN system and previous work on its nu-

22

merical implementation We gratefully acknowledge funding through grants DMS 0554377and DARPAAFOSR FA9550-05-1-0108 to Brown University and NSF grant PHY 0855678to the University of New Mexico

Appendix A Hyperbolicity of the first-order system

This appendix analyzes the matrix A(u) appearing in (10) in order to construct thecharacteristic fields (14) In matrix form the sector (8) of the principal part of (10) reads asfollows

partt

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

=

βr 0 minus 4λα3grr

0 0 λβr

6(grr)2λβr

3gθθgrr0 4λ

3grr

0 βr 0 23grrαχ

13α minus αχ

3grr

αχ3gθθ

minus23χ 0

0 0 βr 0 0 0 0 minus χgrr

0

0 0 minus 4α3grr

βr 0 βr

6(grr)2βr

3gθθgrr0 4

3grr

0 0 23αχ 0 βr minus βrχ

3grrminus2βrχ

3gθθ0 minus2

3χ

0 minus2α 0 0 0 23βr minus2grrβr

3gθθ0 4

3grr

0 gθθαgrr

0 0 0 minusgθθβr

3grr13βr 0 minus2

3gθθ

0 0 minus2α 0 0 0 0 βr 0

34

0 0 0 0 0 0 0 βr

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

prime

(A1)which defines the matrix A(u) appearing in (9) and so also the matrix A(u) in (10) Notethat in the last equation the matrix within the square brackets is minusA(u) For certainconfigurations of u and λ the system (10) is strongly hyperbolic [76] that is A(u) hasa complete set of eigenvectors and real eigenvalues Indeed five eigenpairs of A(u) aretrivially recovered upon inspection of A(u)rsquos leading 5times5 diagonal block These correspondto eigenvalue 0 and the left eigenspace ξj = eTj 1 le j le 5 where ej are the canonical

basis vectors Since each component of u arises as eTj W each is also a characteristic field

The remaining nine eigenpairs are determined by A(u) The eigenvalues of A(u) are

micro1 = 0 micro23 = minusβr microplusmn4 = minusβr plusmn

radic

2αχ

grr microplusmn

5 = minusβr plusmn α

radic

χ

grr microplusmn

6 = minusβr plusmnradic

λ

grr

(A2)

23

and the corresponding left eigenvectors are

x1 = (0 0 0 0 0 gθθ 2grr 0 0) (A3a)

x2 =

(

0 0 0 grr2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3b)

x3 =

(

grrλ 0 0 0

2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3c)

xplusmn4 =

(

0 0plusmnradic

2αgrrχ

0 0 0 0 1 0

)

(A3d)

xplusmn5 =

(

0∓ 3radicgrrχ

plusmn2

radic

grrχ 2grr

1

χminus 1

grr1

gθθ 0 0

)

(A3e)

xplusmn6 =

(

minus3

4

grrλ 0plusmn α

radicλgrr

(2αχminus λ) 0 0minus βr

8(βrgrr ∓radicλgrr)

minus βrgrr

4gθθ(βrgrr ∓radicλgrr)

αχ

(2αχminus λ)plusmnradic

grrλ

)

(A3f)

where for example xplusmn5 A(u) = microplusmn5 x

plusmn5 Assuming that grr gθθ χ and α are everywhere strictly

positive the eigenvalues are real and the eigenvectors are linearly independent providedthat (15) holds These eigenvectors are easily extended to eigenvectors of A(u) e g asxplusmn6 rarr (01times5 x

plusmn6 ) Then for example the characteristic field

Xplusmn6 equiv (01times5 x

plusmn6 )W = xplusmn6WvQ (A4)

and similarly Xplusmnj = xplusmnj WvQ for j = 4 5 and Xk = xkWvQ for k = 1 2 3 The characteristic

speeds for these fields are microk and microplusmnj With this convention the speeds listed in Table I

correspond to the Xk and Xplusmnj in (14)

Appendix B Schwarzschild solution in conformal Kerr-Schild coordinates

In Kerr-Schild coordinates here the system directly related to incoming Eddington-Finkelstein null coordinates the line element for the Schwarzschild solution reads

ds2 = minusα2dt2 + (1 + 2MR)(dR + βRdt)2 +R2dθ2 +R2 sin2 θdφ2 (B1)

where R is the area radius α = (1 + 2MR)minus12 is the lapse and βR = 2M(R + 2M) isthe shift vector The physical spatial metric gab is the spatial part of this line element

To define the corresponding solution to the BSSN system we use equation gab = χgab todefine the following relationship between line elements

dr2 + r2(dθ2 + sin2 θdφ2) = χ[(1 + 2MR)dR2 +R2dθ2 +R2 sin2 θdφ2] (B2)

so that

χ

(

1 +2M

R

)(

dR

dr

)2

= 1 χR2 = r2 (B3)

Then we have(

1 +2M

R

)12dR

R=dr

r (B4)

24

with integration yielding

r =R

4

(

1 +

radic

1 +2M

R

)2

e2minus2radic

1+2MR (B5)

where the constant of integration has been chosen so that the R r rarr infin limits are consistentThe second relation in (B3) shows that

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR =rArr χminus4 =2eradic

1+2MRminus1

1 +radic

1 + 2MR (B6)

The extrinsic curvature tensor is specified by the expression forK given in (B9h) the identityK = KR

R + 2Kθθ and

Kθθ =

(

1 +2M

R

)minus122M

R2 (B7)

Since KRR = Kr

r we compute that

Krr = K minus 2Kθ

θ = minus(

1 +2M

R

)minus12(R +M

R + 2M

)

2M

R2 (B8)

Next since Krr = grrKrr = χminus1Kr

r we have Krr = Arr+

13grrK This implies Arr = Kr

r minus 13K

from which we get (B9g) In all we have

α =

(

1 +2M

R

)minus12

(B9a)

βr = βR dr

dR= χ12

(

1 +2M

R

)minus122M

R(B9b)

grr = 1 (B9c)

gθθ = r2 = χR2 (B9d)

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR (B9e)

Br = 0 (B9f)

Arr = minus(

1 +2M

R

)minus124M

3R2

(

2R + 3M

R + 2M

)

(B9g)

K =

(

1 +2M

R

)minus32(

1 +3M

R

)

2M

R2(B9h)

Γr = minus2

r= minus 2

χ12R (B9i)

To differentiate these expressions with respect to r we use the identity

dR

dr= χminus12

(

1 +2M

R

)minus12

(B10)

along with the chain rule

25

Appendix C Stability of the model system

The following stability analysis for the model system (53) has been inspired by [78 79]After dropping all nonlinear source terms the system (53) becomes

parttu = uprime + av (C1a)

parttv = uprimeprime + vprime (C1b)

This section analyzes the stability of (C1) considering both the continuum system itself aswell as its semi-discrete dG approximation The latter analysis offers some insight into theempirically observed stability of our dG scheme for the spherically reduced BSSN equations

1 Analysis for a single interval

Throughout we work with the L2 inner product and norm

(f g)D =

int

D

fg fD =radic

(f f)D (C2)

where D is the spatial coordinate interval (here D may represent a subdomain Dk or the

whole domain Ω) and we have suppressed all integration measures For the continuummodel we will establish the following estimate

uprime(T middot)2D+ av(T middot)2

Dle C(T )

(

uprime(0 middot)2D+ av(0 middot)2

D

)

(C3)

where the time-dependent constant C(T ) is determined solely by the choice of boundaryconditions To show (C3) we first change variables with v =

radicav thereby rewriting (C1)

in the following symmetric form

parttu = uprime +radicav (C4a)

parttv =radicauprimeprime + vprime (C4b)

Equations (C4ab) then imply

1

2partt

int

D

(uprime)2 =

int

D

uprime(uprimeprime +radicavprime) =

int

D

radicauprimevprime +

1

2

int

partD

(uprime)2 (C5a)

1

2partt

int

D

(v)2 =

int

D

v(radicauprimeprime + vprime) = minus

int

D

radicauprimevprime +

1

2

int

partD

(v2 + 2radicauprimev) (C5b)

Here vvprime and uprimeuprimeprime have been expressed as exact derivatives and then integrated to boundaryterms the second equation employs an extra integration by parts and with only one spacedimension

int

partDdenotes a difference of endpoint evaluations Addition of Eqs (C5ab) gives

1

2partt

int

D

[

v2 + (uprime)2]

=1

2

int

partD

[

v2 + (uprime)2 + 2radicauprimev

]

(C6)

Substitutions with the identities

[

v2 + (uprime)2]

=1

2

[

(v + uprime)2 + (v minus uprime)2]

2uprimev =1

2

[

(v + uprime)2 minus (v minus uprime)2]

(C7)

26

and replacements to recover the original variable v = vradica yield

1

2partt

int

D

[

av2 + (uprime)2]

=1 +

radica

4

int

partD

(radicav + uprime)2 +

1minusradica

4

int

partD

(radicav minus uprime)2 (C8)

From (C8) we deduce that the time-dependent constant C(T ) in (C3) must satisfy

∣

∣

∣

∣

∣

∣

1 +

int T

0

[

1+radica

2

int

partD(radicav + uprime)2 + 1minus

radica

2

int

partD(radicav minus uprime)2

]

dt

uprime(0 middot)2 + av(0 middot)2

∣

∣

∣

∣

∣

∣

le C(T ) (C9)

For periodic boundary conditions we may choose C(T ) = 1 Moreover if a ge 1 anduprime = minusradic

av is specified at partD+ then uprime(t middot)2 + av(t middot)2 decaysStill working on a single interval (subdomain) we now consider the semi-discrete scheme

for (C4) i e (57) with all nonlinear source terms dropped and with v replaced by vradica

Derivation of a formula analogous to (C8) is our first step toward establishing L2 stabilityof the semi-discrete scheme While (57) features vectors for example u(t) taking values atthe Legendre-Gauss-Lobatto nodal points here we work with the numerical solution as apolynomial for example uh(t x) These two representations are related by the Lagrange in-terpolating polynomials for the nodal set here taken to span both the space of test functionsand the space of basis functions Our scheme is

int

Dk

ψparttuh =

int

Dk

ψ(Qh +radicavh) (C10a)

int

Dk

ξparttvh = minusint

Dk

ξprime(radicaQh + vh) +

int

partDk

ξ(radicaQlowast + vlowast) (C10b)

int

Dk

ϕQh =

int

Dk

ϕuprimeh +

int

partDk

ϕ (ulowast minus uh) (C10c)

where ψ ξ and ϕ are polynomial test functions These test functions are arbitrary exceptthat they must be degree-N polynomials In (C10) the variables uh vh and Qh should alsocarry a superscript k but we have suppressed this Derivation of a formula analogous to(C8) is complicated by the fact that Qh is not evolved Nevertheless at a given instant t wecan assemble Qh from (C10c)

Mimicking the calculation (C5b) from the continuum case we first use (C10b) with ξ = vhto write

1

2partt

int

Dk

v2h = minusint

Dk

(radicaQh + vh)v

primeh +

int

partDk

(radicaQlowast + vlowast)vh

= minusint

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2(radicaQlowast + vlowast)vh minus v2h

]

(C11)

The right-hand side of (C5a) is analogous to

1

2partt

int

Dk

Q2h =

int

Dk

QhparttQh (C12)

However since Qh is not evolved the term parttQh must be given a suitable interpretationOn the right hand side of (C10c) only uh u

primeh and u

lowast necessarily depend on time since the

27

test function ϕ need not be time-dependent Furthermore ulowast is explicitly given as a linearcombination of uh as seen in Eq (C21c) below Choosing ϕ = ℓj taking the time derivativeof (C10c) and appealing to the commutivity of mixed partial derivatives we therefore arriveat

int

Dk

ℓjparttQh =

int

Dk

ℓj(parttuh)prime +

int

partDk

ℓj(

(parttu)lowast minus parttuh

)

(C13)

where (parttu)lowast depends on parttuh in precisely the same way that ulowast depends on uh We have

written ℓj rather than ϕ in the last equation to emphasize that the result also holds for anylinear combination of ℓj (for example ϕ) and even for time-dependent combinations SinceQh is itself such a combination we obtain

1

2partt

int

Dk

Q2h =

int

Dk

Qh(parttuh)prime +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

Qh(Qprimeh +

radicavprimeh) +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2((parttu)lowast minus parttuh)Qh +Q2

h

]

(C14)

Addition of (C11) and (C14) gives

1

2partt

int

Dk

(Q2h + v2h) =

1

2

int

partDk

[

Q2h minus v2h + 2(

radicaQlowast + vlowast)vh + 2((parttu)

lowast minus parttuh)Qh

]

(C15)

the aforementioned analog of (C8) This formula holds on a single subdomain Dk and we

now combine multiple copies of it one for each value of k

2 Analysis for multiple intervals

To facilitate combination of (C15) over all k we change notation At every subdomaininterface Ik+12 equiv partDk cap partDk+1 let the superscripts L and R denote field values respectivelytaken from the left and right Then the fields evaluated at Ik+12 which belong to D

k will beuLk+12 v

Lk+12 and Q

Lk+12 while those belonging to D

k+1 will be uRk+12 vRk+12 and Q

Rk+12

However at Ikminus12 the values taken from D

k are uRkminus12 vRkminus12 and QR

kminus12 Note that we

have also replaced the subscript h denoting a numerical solution with k plusmn 12 denotingthe location of the endpoint value of the numerical solution With this notation we define

∆Lα =

1

2

[

(QLα)

2 minus (vLα)2]

+(radic

aQlowastα + vlowastα

)

vLα +[

(parttuα)lowast minus parttu

Lα

]

QLα (C16)

and similarly for ∆Rα The same numerical fluxes appear in both ∆L

α and ∆Rα (ie each

numerical flux takes the same value on either side of an interface) whence fluxes like Qlowastα do

not carry an L or R superscript In terms of these definitions (C15) becomes

1

2partt

int

Dk

(Q2h + v2h) = ∆L

k+12 minus∆Rkminus12 (C17)

28

Summation over all Dk yields

1

2partt

kmaxsum

k=1

int

Dk

(Q2h + v2h) =

kmaxminus1sum

k=1

(

∆Lk+12 minus∆R

k+12

)

+∆Lkmax+12 minus∆R

12

=

kmaxminus1sum

k=1

(

∆Lh minus∆R

h

)∣

∣

Ik+12 +∆Lkmax+12 minus∆R

12 (C18)

We have reverted to h-notation denoting the numerical solution since the LR superscriptsindicate unambiguously the relevant domain used for evaluation at Ik+12

We again seek an estimate of the form

kmaxsum

k=1

(

Qh(T middot)2Dk + avh(T middot)2Dk

)

le C(T )kmaxsum

k=1

(

Qh(0 middot)2Dk + avh(0 middot)2Dk

)

(C19)

that is essentially the same as the one (C3) considered in the continuum case Assume thatthe chosen boundary conditions ensure ∆L

kmax+12 minus ∆R12 is bounded by a time-dependent

constant which does not depend on the numerical parameters N and h (subdomain width)Establishment of stability then amounts to showing that the remaining sum over interfaceterms in (C18) is non-positive whence this remaining sum is consistent with C(T ) le 1although the boundary conditions may give rise to a different bound In fact we will choosethe numerical fluxes such that each individual interface term is non-positive At interfaceIk+12 and in LR notation the jump and average of vh for example are

1

2

(

v+ + vminus)

equiv vh =1

2

(

vLk+12 + vRk+12

)

(C20a)

nminusvminus + n+v+ equiv[[

vh]]

= vLk+12 minus vRk+12 (C20b)

Consider numerical fluxes of the form

Qlowast = Qh minusτQ2

[[

Qh

]]

(C21a)

vlowast = vh minusτv2

[[

vh]]

(C21b)

ulowast = uh minusτu2

[[

uh]]

(C21c)

(parttu)lowast = parttuh minus

τu2

[[

parttuh]]

(C21d)

where (C21c) induces (C21d) and where the penalty parameters τu τv and τQ are realnumbers The fluxes defined in (58) correspond to τu = 1 τv = 1 +

radica and τQ = 0 In

terms of these quantities the kth interface contribution in (C18) is

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

parttuh]]

minus τu2

[[

parttuh]][[

Qh

]]

(C22)

where we have suppressed the k dependence of the right-hand side Now consider the term[[

parttuh]]

Because parttuh and Qh+radicavh are both polynomials of degree N Eq (C10a) implies

29

τu

τ Q

minus1 0 1minus1

minus05

0

05

1

τu

minus1 0 1minus1

minus05

0

05

1

FIG 6 Stable evolutions for the model system For fixed τv = 10minus6 and τv = 1 +radic2

respectively the left and right plots depict stable choices (determined empirically) of τu and τQfor the linear model system (C1) The stable regions are colored black but the jagged edges result

from the discretization of the (τu τQ)-plane

the vector equation parttu = Q +radicav that is pointwise equivalence on the nodal points of

Dk which in turn implies

[[

parttuh]]

=[[

Qh+radicavh]]

Upon substituting this identity into thelast equation we arrive at an expression which features only vh and Qh

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

Qh +radicavh]]

minus τu2

[[

Qh +radicavh]][[

Qh

]]

(C23)

The identities vh[[

vh]]

= 12

[[

v2h]]

and[[

Qh +radicavh]]

=[[

Qh

]]

+radica[[

vh]]

then simplify(C23) to

(∆Lh minus∆R

h )∣

∣

Ik+12 = minusτv2

[[

vh]]2 minus

radica(τu + τQ)

2

[[

Qh

]][[

vh]]

minus τu2

[[

Qh

]]2 (C24)

The role of a penalty parameter is now clear Positive values of τv penalize jumps in vhthrough a negative contribution to the energy Likewise positive values of τu penalizejumps in Qh through a negative contribution to the energy However because the sign of[[

Qh

]][[

vh]]

can be positive or negative only the choice τQ = minusτu yields an expression for(∆L

h minus ∆Rh )|Ik+12 which is manifestly negative for τu ge 0 and τv ge 0 A simple estimate

based on Youngrsquos inequality with ε (that is 2αβ le εminus1α2 + εβ2 where α β ge 0 and ε gt 0)shows that for τQ = 0 the choice τv ge aτu4 also yields stability

Figure 6 depicts certain choices of stable penalty parameters for the linear model systemevolved to tfinal = 1000 (with a = 2 N = 10 and ∆t ≃ 00553) as determined empiricallywith simulations similar to those described in Sec IVA The left plot corresponds to asmall τv = 10minus6 for which the choice τu = 1 τQ = 0 is not stable as expected from thetheoretical analysis However the right plot corresponds to τv = 1 +

radica for which τu = 1

τQ = 0 is stable Motivated by the numerical flux choices (3842) used for the BSSN system

30

(4) we have (as mentioned above) set τu = 1 τv = 1 +radica and τQ = 0 in simulations of

the nonlinear model (53) For the nonlinear model system (53) the theoretically motivatedchoice τQ = minusτu also yields numerically stable evolutions when τu ge 0 and τv ge 0

For the nonlinear systems (4) and (53) we do not attempt a formal stability proofNevertheless the results of this appendix have served as a guide for our choices of penaltyparameters For the BSSN system (4) u v and Q are block indices [cf Eq (6)] Similarto the model problem we have penalized Q with τu = 1 with τv chosen large enough toheuristically overcome the cross-terms of indefinite size that arise from τQ = 0 (we interpretequations like τu = 1 componentwise) An analogous choice ldquoτQ = minusτurdquo for the BSSNsystem might be possible but would be considerably more complicated Indeed such achoice likely entails a matrix of penalty parameters but we do not give the details here

[1] F Pretorius Evolution of Binary Black-Hole Spacetimes Phys Rev Lett 95 121101 (2005)

(4 pages)

[2] F Pretorius Simulation of binary black hole spacetimes with a harmonic evolution scheme

Class Quantum Grav 23 S529-S552 (2006)

[3] M Campanelli C O Lousto P Marronetti and Y Zlochower Accurate Evolutions of Or-

biting Black-Hole Binaries without Excision Phys Rev Lett 96 111101 (2006) (4 pages)

[4] J G Baker J Centrella D-I Choi M Koppitz and J van Meter Gravitational-Wave

Extraction from an Inspiraling Configuration of Merging Black Holes Phys Rev Lett 96

111102 (2006) (4 pages)

[5] M Campanelli C O Lousto and Y Zlochower Last orbit of binary black holes Phys Rev D

73 061501(R) (2006) (5 pages)

[6] F Herrmann I Hinder D Shoemaker and P Laguna Unequal mass binary black hole plunges

and gravitational recoil Class Quantum Grav 24 S33-S42 (2007)

[7] P Diener F Herrmann D Pollney E Schnetter E Seidel R Takahashi J Thornburg and

J Ventrella Accurate Evolution of Orbiting Binary Black Holes Phys Rev Lett 96 121101

(2006) (4 pages)

[8] M A Scheel H P Pfeiffer L Lindblom L E Kidder O Rinne and S A Teukolsky Solving

Einsteinrsquos Equations With Dual Coordinate Frames Phys Rev D 74 104006 (2006)

[9] U Sperhake Binary black-hole evolutions of excision and puncture data Phys Rev D 76

104015 (2007) (20 pages)

[10] B Brugmann J A Gonzalez M Hannam S Husa U Sperhake and W Tichy Calibration

of moving puncture simulations Phys Rev D 77 024027 (2008) (25 pages)

[11] P Marronetti W Tichy B Brugmann J Gonzalez M Hannam S Husa and U Sperhake

Binary black holes on a budget simulations using workstations Class Quantum Grav 24

S43-S58 (2007)

[12] Z B Etienne J A Faber Y T Liu S L Shapiro and T W Baumgarte Filling the holes

Evolving excised binary black hole initial data with puncture techniques Phys Rev D 76

101503(R) (2007) (5 pages)

[13] B Szilagyi D Pollney L Rezzolla J Thornburg and J Winicour An explicit harmonic code

for black-hole evolution using excision Class Quantum Grav 24 S275-S293 (2007)

[14] L Boyle M Kesden and S Nissanke Binary-Black-Hole Merger Symmetry and the Spin

Expansion Phys Rev Lett 100 151101 (2008) (4 pages)

31

[15] M A Scheel M Boyle T Chu L E Kidder K D Matthews and H P Pfeiffer High-

accuracy waveforms for binary black hole inspiral merger and ringdown Phys Rev D 79

024003 (2009) (14 pages)

[16] M Hannam Status of black-hole-binary simulations for gravitational-wave detection

Class Quantum Grav 26 114001 (2009) (21 pages)

[17] I Hinder The current status of binary black hole simulations in numerical relativity

Class Quantum Grav 27 114004 (2010) (19 pages)

[18] B P Abbott et al LIGO The Laser Interferometer Gravitational-Wave Observatory

Rep Prog Phys 72 076901 (2009) (25 pages)

[19] F Acernese et al The status of VIRGO Class Quantum Grav 23 S63-S69 (2006)

[20] F Acernese et al Virgo status Class Quantum Grav 25 184001 (2008) (9 pages)

[21] R Arnowitt S Deser and C W Misner The Dynamics of General Relativity in Gravitation

an introduction to current research edited by L Witten (Wiley New York 1962) Chapter 7

227-265 Also available as gr-qc0405109

[22] J W York Jr Kinematics and Dynamics of General Relativity in Sources of Gravitational

Radiation edited by L L Smarr (Cambridge University Press Cambridge 1979)

[23] L E Kidder M A Scheel and S A Teukolsky Extending the lifetime of 3D black hole

computations with a new hyperbolic system of evolution equations Phys Rev D 64 064017

(2001) (13 pages)

[24] G Calabrese J Pullin O Sarbach and M Tiglio Convergence and stability in numerical

relativity Phys Rev D 66 041501(R) (2002) (4 pages)

[25] S Frittelli and O Reula On the Newtonian Limit of General Relativity Commun Math

Phys 166 221-235 (1994)

[26] M Shibata and T Nakamura Evolution of three-dimensional gravitational waves Harmonic

slicing case Phys Rev D 52 5428-5444 (1995)

[27] Y Choquet-Bruhat and J W York Geometrical well posed systems for the Einstein equations

C R Acad Sci Paris t 321 Serie I 1089-1095 (1995)

[28] A Abrahams A Anderson Y Choquet-Bruhat and J W York Jr Einstein and Yang-Mills

Theories in Hyperbolic Form without Gauge Fixing Phys Rev Lett 75 3377-3381 (1995)

[29] C Bona J Masso E Seidel and J Stela New Formalism for Numerical Relativity

Phys Rev Lett 75 600-603 (1995)

[30] M H P M van Putten and D M Eardley Nonlinear wave equations for relativity

Phys Rev D 53 3056-3063 (1996)

[31] S Frittelli and O A Reula First-Order Symmetric Hyperbolic Einstein Equations with Ar-

bitrary Fixed Gauge Phys Rev Lett 76 4667-4670 (1996)

[32] H Friedrich Hyperbolic reductions for Einsteinrsquos equations Class Quantum Grav 13 1451-

1469 (1996)

[33] F B Estabrook R S Robinson H D Wahlquist Hyperbolic equations for vacuum gravity

using special orthonormal frames Class Quantum Grav 14 1237-1247 (1997)

[34] M S Iriondo E O Leguizamon and O A Reula Einsteinrsquos Equations in Ashtekarrsquos Vari-

ables Constitute a Symmetric Hyperbolic System Phys Rev Lett 79 4732-4735 (1997)

[35] A Anderson Y Choquet-Bruhat and J W York Einstein-Bianchi hyperbolic system for

general relativity Topol Meth Nonlinear Anal 10 353-373 (1997)

[36] T W Baumgarte and S L Shapiro Numerical integration of Einsteinrsquos field equations

Phys Rev D 59 024007 (1998) (7 pages)

[37] M A G Bonilla Symmetric hyperbolic systems for Bianchi equations Class Quantum

16

10minus10

10minus6

10minus2

∆uinfin

10minus10

10minus6

10minus2

∆vinfin

0 500 1000 1500 2000 2500 3000 3500 400010

minus10

10minus6

10minus2

∆(partxu)infin

t

N = 3N = 6N = 9N = 12

FIG 1 Spectral convergence of fields for model PDE Respectively for N = 3 6 9 12

a timestep of ∆t = 00578 00178 00084 00049 has been chosen for stability and accuracy In the

title headings for example ∆u equiv unumer minus uexact

A Simulations of the model system

The semi-discrete scheme (57) has been integrated with the classical fourth-order Runge-Kutta method When integrating this system we have first constructed Q at each Runge-Kutta stage and then substituted into the evolution equations (57ab) for u and v Theproblem has been solved on a computational domain [0 4π] comprised of two subdomainswith a timestep chosen small enough for stability The initial data has been taken from thefollowing exact solution to (53)

uprimeexact(t x) =1

2

[

sin(xminus microminust)minus sin(xminus micro+t)]

(59a)

vexact(t x) =1

2radica

[

sin(xminus microminust) + sin(xminus micro+t)]

(59b)

g(t x) = u3exact (59c)

h(t x) = (uexact + vexact)(uprimeexact)

2 minus v2exactu2exact (59d)

where the speeds microplusmn are found in (54) Specification of the boundary condition at a physicalendpoint amounts to choosing the external state for at the endpoint We have considered

17

8 16 32 64 128

10minus3

10minus2

10minus1

N

∆t

max stable ∆t8N2

FIG 2 Scaling of maximum stable ∆t with N for model PDE

two possibilities (i) the analytic state (Q+ v+) = (Qexact vexact) and (ii) an upwind stateFor example at x = 4π the upwind state is4

Q+ = Qupwind =1

2

[

(Xminus)exact minus (X+)numer

]

v+ = vupwind =1

2radica

[

(Xminus)exact + (X+)numer

]

(60)Either choice of (Q+ v+) leads to similar results and the plots here correspond to theanalytic state Figure 1 clearly shows spectral convergence with increasing polynomial orderN across all fields for the case a = 2 Other values of a including a = 1 for which X+

is a static characteristic field have also been considered with similar results Appendix Cdemonstrates that our proposed scheme for the system (57) with nonlinear and source termsdropped is stable in a semi-discrete sense Nevertheless the fully discrete scheme obtainedvia temporal discretization by the fourth-order Runge-Kutta method is still subject to thestandard absolute stability requirement Namely if microh is any eigenvalue corresponding tothe (linearized) discrete spatial operator then a necessary condition for stability is thatmicroh∆t lies in absolute stability region for Runge-Kutta 4 We here show empirically that theassociated timestep restriction scales like Nminus2 ie ∆t = O(Nminus2) for stability We note thatsuch scaling is welcome in light of the second-order spatial operators which appear in thesystem and suggest a possible worse scaling like Nminus4 Fig 2 plots the maximum stabletimestep for a range of N demonstrating the Nminus2 scaling in line with behavior known fromanalysis of first-order systems [66] This scaling also holds for the BSSN system

4 We remind the reader that unfortunately the plusmn on Xplusmn means something different than the plusmn indicating

exteriorinterior dG states [cf footnote 3]

18

B Simulations of the BSSN system

This subsection documents results for simulations of the unit-mass-parameter (M = 1)Schwarzschild solution (B9) expressed in terms of ingoing Kerr-Schild coordinates Sincethe solution is stationary temporal integration of the semi-discrete scheme has been carriedout with the forward Euler method The r-coordinate domain [04 34] has been split into3 equally spaced subdomains and we have set η = 10 λ = 01 and C = 2 [cf Eq (41)]For all simulations ∆t has been chosen for stability With the chosen λ the inner physicalboundary rmin = 04 is an excision surface At each timestep we have applied an (order2s = 20) exponential filter on the top two-thirds of the modal coefficient set for all fieldsexcept for grr and gθθ For stability we have empirically observed that grr and gθθ must not

be filtered A detailed understanding of this is still lackingIssues related to physical boundary conditions are similar to the one encountered in

Sec IVA for the model problem Similar to before we have retained Eqs (3842) as thechoice of numerical flux even at the endpoints Therefore at an endpoint the specificationof the boundary condition amounts to the choice W+ of external state We have typicallychosen the inner boundary of the radial domain as an excision boundary and in this caseW+ = Wminus is enforced at the inner physical boundary At the outer physical boundaryfor W+ we have again considered two choices (i) Wexact and (ii) Wupwind To enforcechoice (ii) the inverse transformation (17) must be used with incoming characteristic fieldsfixed to their exact values similar to (60) We have tried various versions of choice (ii)and in all cases the resulting simulations have been unstable We therefore present resultscorresponding to choice (i) Although the choice of an analytical external state Wexact

at the outer boundary is stable for our problem such a boundary condition is unlikelyto generalize to more complicated scenarios involving dynamical fields Indeed the issueof outer boundary conditions for the BSSN system is an active area of research with aproper treatment requiring fixation of incoming radiation control of the constraints andspecification of gauge (see Ref [86] for a recent analysis)

For BSSN simulations our main diagnostic is to monitor the Hamiltonian momentumand conformal connection constraints Figure 3 depicts long-time histories of constraintviolations whereas Figs 4 and 5 depict long-time error histories for the individual BSSNfield components From the middle plot in Fig 5 we infer that up to the indicated numericalerror the factor g sin4 θ = grr(gθθ)

2 remains at its initial fixed profile r4 throughout theevolution These figures indicate that the proposed scheme is stable for long times andexhibits spectral converge with increased polynomial order N Similar results are recoveredfromM = 0 Minkowski initial data The stability documented in these plots does not appearto rely on inordinate parameter tuning For example with the fixed parameters describedabove we obtain similar plots if we individually vary (i) rmin over 0325 035 04 0475(values still corresponding to an excision surface for the given choice of λ) (ii) η over1 3 7 10 (iii) s over 8 9 10 With the polynomial order N ranging over 23 26 29 31both stability and qualitatively similar exponential convergence is achieved with a singlesubdomain Likewise adoption of a larger coordinate domain with more subdomains doesnot impact our results Finally we have considered the addition of random noise to all fieldcomponents at the initial time Precisely with the system component χ as an example wehave set

χ(0) 7rarr χ(0) + δχ(0) (61)

where each component (nodal value) of δχ(0) is 10minus5 times a random variable drawn from a

19

10minus9

10minus7

10minus5

10minus3

Hinfin

10minus9

10minus7

10minus5

10minus3

Mrinfin

0 2000 4000 6000 8000 10000

10minus9

10minus7

10minus5

10minus3

Grinfin

t

N = 11N = 14N = 17N = 20

FIG 3 Spectral convergence of constraint violations for M = 1 Kerr-Schild ini-

tial data Respectively for N = 11 14 17 19 a timestep of ∆t ≃ 00041 00026 00018 00013

has been chosen for stability and accuracy

standard normal distribution Such perturbed initial data also gives rise to stable evolutions

V CONCLUSION

We have introduced a discontinuous Galerkin method for solving the spherically reducedBSSN system with second-order spatial operators Our scheme shares similarities withother discontinuous Galerkin methods that use local auxiliary variables to handle high-orderspatial derivatives [66 77ndash79 82 84 85] and which have typically been applied to eitherelliptic parabolic or mixed type problems The key ingredient of a stable dG scheme is anappropriate choice of numerical flux and our particular choice has been motivated by theanalysis presented in Appendix C When used to evolve the Schwarzschild solution in Kerr-Schild coordinates our numerical implementation of the BSSN system (4) is robustly stableand converges to the analytic solution exponentially with increased polynomial order Byapproximating the spatially second-order form of the BSSN system we have not introducedextra fields which are evolved Evolved auxiliary fields result in new constraints which mayspoil stability Our main goal has been stable evolution of the spherically reduced BSSNsystem as a first step towards understanding how a discontinuous Galerkin method might be

20

10minus9

10minus7

10minus5

∆Arrinfin

10minus9

10minus7

10minus5

∆Kinfin

0 2000 4000 6000 8000 10000

10minus9

10minus7

10minus5

max(∆Brinfin ∆Γrinfin)

t

N = 11N = 14N = 17N = 20

FIG 4 Spectral convergence of solution for M = 1 Kerr-Schild initial data

Timestep choices are described in the caption for Fig 3 In the title headings for example

∆Arr equiv (Arr)numer minus (Arr)exact

applied to the full BSSN system Towards that goal we now discuss treatment of singularitiesand generalization of the described dG method to higher space dimension

To deal with the fixed Schwarzschild singularity we have used excision which is easy inthe context of the spherically reduced BSSN system However excision for the binary blackhole problem in full general relativity requires attention to the technical challenge of hori-zon tracking State-of-the-art BSSN codes avoid such complication relying instead on themoving-puncture technique While the moving-puncture technique does involve mild centralsingularities it may still prove amenable to spectral methods Indeed spectral methods fornon-smooth problems is well-developed in both theory and for complex applications Sincethe moving-puncture technique can be performed in spherical symmetry [60] a first-steptoward a spectral moving-puncture code would be to implement a moving puncture withthe nodal dG method described here Such an implementation may adopt Legendre-Gauss-Radau nodes on the innermost subdomain thereby ensuring that the physical singularitydoes not lie on a nodal point (in much the same way finite difference codes use a staggeredgrid) Beyond traditional excision and moving punctures one might also retain a smoothproblem for which spectral methods are especially well-suited via the turducken approach[87 88] to singularities

21

10minus11

10minus9

10minus7

10minus5

∆χinfin

10minus11

10minus9

10minus7

10minus5

max(∆grrinfin ∆gθθinfin)

0 2000 4000 6000 8000 10000

10minus11

10minus9

10minus7

10minus5

max(∆αinfin ∆βrinfin)

t

N = 11N = 14N = 17N = 20

FIG 5 Spectral convergence of solution violations for M = 1 Kerr-Schild initial

data See the caption of Fig 4 for details

Discontinuous Galerkin methods for hyperbolic problems in two and three space dimen-sions are well-developed A generalization of the method described here to three-dimensionsand the full BSSN system would likely rely on an unstructured mesh Appropriate local poly-nomial expansions for the subelements are well-understood as are choices for the numericalfluxes which would now live on two-dimensional faces rather than single points Whether ornot it would ultimately prove successful generalization of our dG method to a higher dimen-sion would rely on an established conceptual framework Further computational advancesof relevance to a generalization of our dG method to the full BSSN system (possibly includ-ing matter) may include mesh hp-adaptivity local timestepping shock capturing and slopelimiting techniques [66] Moreover recent work [89] indicates that enhanced performancewould be expected were our scheme implemented on graphics processor units

VI ACKNOWLEDGMENTS

We thank Nick Taylor for discussions about treating second-order operators with spectralmethods Benjamin Stamm for helping polish up a few parts of Appendix C Khosro Shah-bazi for discussions on LDG and IP methods We also acknowledge helpful conversationswith David Brown and Manuel Tiglio about the BSSN system and previous work on its nu-

22

merical implementation We gratefully acknowledge funding through grants DMS 0554377and DARPAAFOSR FA9550-05-1-0108 to Brown University and NSF grant PHY 0855678to the University of New Mexico

Appendix A Hyperbolicity of the first-order system

This appendix analyzes the matrix A(u) appearing in (10) in order to construct thecharacteristic fields (14) In matrix form the sector (8) of the principal part of (10) reads asfollows

partt

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

=

βr 0 minus 4λα3grr

0 0 λβr

6(grr)2λβr

3gθθgrr0 4λ

3grr

0 βr 0 23grrαχ

13α minus αχ

3grr

αχ3gθθ

minus23χ 0

0 0 βr 0 0 0 0 minus χgrr

0

0 0 minus 4α3grr

βr 0 βr

6(grr)2βr

3gθθgrr0 4

3grr

0 0 23αχ 0 βr minus βrχ

3grrminus2βrχ

3gθθ0 minus2

3χ

0 minus2α 0 0 0 23βr minus2grrβr

3gθθ0 4

3grr

0 gθθαgrr

0 0 0 minusgθθβr

3grr13βr 0 minus2

3gθθ

0 0 minus2α 0 0 0 0 βr 0

34

0 0 0 0 0 0 0 βr

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

prime

(A1)which defines the matrix A(u) appearing in (9) and so also the matrix A(u) in (10) Notethat in the last equation the matrix within the square brackets is minusA(u) For certainconfigurations of u and λ the system (10) is strongly hyperbolic [76] that is A(u) hasa complete set of eigenvectors and real eigenvalues Indeed five eigenpairs of A(u) aretrivially recovered upon inspection of A(u)rsquos leading 5times5 diagonal block These correspondto eigenvalue 0 and the left eigenspace ξj = eTj 1 le j le 5 where ej are the canonical

basis vectors Since each component of u arises as eTj W each is also a characteristic field

The remaining nine eigenpairs are determined by A(u) The eigenvalues of A(u) are

micro1 = 0 micro23 = minusβr microplusmn4 = minusβr plusmn

radic

2αχ

grr microplusmn

5 = minusβr plusmn α

radic

χ

grr microplusmn

6 = minusβr plusmnradic

λ

grr

(A2)

23

and the corresponding left eigenvectors are

x1 = (0 0 0 0 0 gθθ 2grr 0 0) (A3a)

x2 =

(

0 0 0 grr2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3b)

x3 =

(

grrλ 0 0 0

2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3c)

xplusmn4 =

(

0 0plusmnradic

2αgrrχ

0 0 0 0 1 0

)

(A3d)

xplusmn5 =

(

0∓ 3radicgrrχ

plusmn2

radic

grrχ 2grr

1

χminus 1

grr1

gθθ 0 0

)

(A3e)

xplusmn6 =

(

minus3

4

grrλ 0plusmn α

radicλgrr

(2αχminus λ) 0 0minus βr

8(βrgrr ∓radicλgrr)

minus βrgrr

4gθθ(βrgrr ∓radicλgrr)

αχ

(2αχminus λ)plusmnradic

grrλ

)

(A3f)

where for example xplusmn5 A(u) = microplusmn5 x

plusmn5 Assuming that grr gθθ χ and α are everywhere strictly

positive the eigenvalues are real and the eigenvectors are linearly independent providedthat (15) holds These eigenvectors are easily extended to eigenvectors of A(u) e g asxplusmn6 rarr (01times5 x

plusmn6 ) Then for example the characteristic field

Xplusmn6 equiv (01times5 x

plusmn6 )W = xplusmn6WvQ (A4)

and similarly Xplusmnj = xplusmnj WvQ for j = 4 5 and Xk = xkWvQ for k = 1 2 3 The characteristic

speeds for these fields are microk and microplusmnj With this convention the speeds listed in Table I

correspond to the Xk and Xplusmnj in (14)

Appendix B Schwarzschild solution in conformal Kerr-Schild coordinates

In Kerr-Schild coordinates here the system directly related to incoming Eddington-Finkelstein null coordinates the line element for the Schwarzschild solution reads

ds2 = minusα2dt2 + (1 + 2MR)(dR + βRdt)2 +R2dθ2 +R2 sin2 θdφ2 (B1)

where R is the area radius α = (1 + 2MR)minus12 is the lapse and βR = 2M(R + 2M) isthe shift vector The physical spatial metric gab is the spatial part of this line element

To define the corresponding solution to the BSSN system we use equation gab = χgab todefine the following relationship between line elements

dr2 + r2(dθ2 + sin2 θdφ2) = χ[(1 + 2MR)dR2 +R2dθ2 +R2 sin2 θdφ2] (B2)

so that

χ

(

1 +2M

R

)(

dR

dr

)2

= 1 χR2 = r2 (B3)

Then we have(

1 +2M

R

)12dR

R=dr

r (B4)

24

with integration yielding

r =R

4

(

1 +

radic

1 +2M

R

)2

e2minus2radic

1+2MR (B5)

where the constant of integration has been chosen so that the R r rarr infin limits are consistentThe second relation in (B3) shows that

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR =rArr χminus4 =2eradic

1+2MRminus1

1 +radic

1 + 2MR (B6)

The extrinsic curvature tensor is specified by the expression forK given in (B9h) the identityK = KR

R + 2Kθθ and

Kθθ =

(

1 +2M

R

)minus122M

R2 (B7)

Since KRR = Kr

r we compute that

Krr = K minus 2Kθ

θ = minus(

1 +2M

R

)minus12(R +M

R + 2M

)

2M

R2 (B8)

Next since Krr = grrKrr = χminus1Kr

r we have Krr = Arr+

13grrK This implies Arr = Kr

r minus 13K

from which we get (B9g) In all we have

α =

(

1 +2M

R

)minus12

(B9a)

βr = βR dr

dR= χ12

(

1 +2M

R

)minus122M

R(B9b)

grr = 1 (B9c)

gθθ = r2 = χR2 (B9d)

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR (B9e)

Br = 0 (B9f)

Arr = minus(

1 +2M

R

)minus124M

3R2

(

2R + 3M

R + 2M

)

(B9g)

K =

(

1 +2M

R

)minus32(

1 +3M

R

)

2M

R2(B9h)

Γr = minus2

r= minus 2

χ12R (B9i)

To differentiate these expressions with respect to r we use the identity

dR

dr= χminus12

(

1 +2M

R

)minus12

(B10)

along with the chain rule

25

Appendix C Stability of the model system

The following stability analysis for the model system (53) has been inspired by [78 79]After dropping all nonlinear source terms the system (53) becomes

parttu = uprime + av (C1a)

parttv = uprimeprime + vprime (C1b)

This section analyzes the stability of (C1) considering both the continuum system itself aswell as its semi-discrete dG approximation The latter analysis offers some insight into theempirically observed stability of our dG scheme for the spherically reduced BSSN equations

1 Analysis for a single interval

Throughout we work with the L2 inner product and norm

(f g)D =

int

D

fg fD =radic

(f f)D (C2)

where D is the spatial coordinate interval (here D may represent a subdomain Dk or the

whole domain Ω) and we have suppressed all integration measures For the continuummodel we will establish the following estimate

uprime(T middot)2D+ av(T middot)2

Dle C(T )

(

uprime(0 middot)2D+ av(0 middot)2

D

)

(C3)

where the time-dependent constant C(T ) is determined solely by the choice of boundaryconditions To show (C3) we first change variables with v =

radicav thereby rewriting (C1)

in the following symmetric form

parttu = uprime +radicav (C4a)

parttv =radicauprimeprime + vprime (C4b)

Equations (C4ab) then imply

1

2partt

int

D

(uprime)2 =

int

D

uprime(uprimeprime +radicavprime) =

int

D

radicauprimevprime +

1

2

int

partD

(uprime)2 (C5a)

1

2partt

int

D

(v)2 =

int

D

v(radicauprimeprime + vprime) = minus

int

D

radicauprimevprime +

1

2

int

partD

(v2 + 2radicauprimev) (C5b)

Here vvprime and uprimeuprimeprime have been expressed as exact derivatives and then integrated to boundaryterms the second equation employs an extra integration by parts and with only one spacedimension

int

partDdenotes a difference of endpoint evaluations Addition of Eqs (C5ab) gives

1

2partt

int

D

[

v2 + (uprime)2]

=1

2

int

partD

[

v2 + (uprime)2 + 2radicauprimev

]

(C6)

Substitutions with the identities

[

v2 + (uprime)2]

=1

2

[

(v + uprime)2 + (v minus uprime)2]

2uprimev =1

2

[

(v + uprime)2 minus (v minus uprime)2]

(C7)

26

and replacements to recover the original variable v = vradica yield

1

2partt

int

D

[

av2 + (uprime)2]

=1 +

radica

4

int

partD

(radicav + uprime)2 +

1minusradica

4

int

partD

(radicav minus uprime)2 (C8)

From (C8) we deduce that the time-dependent constant C(T ) in (C3) must satisfy

∣

∣

∣

∣

∣

∣

1 +

int T

0

[

1+radica

2

int

partD(radicav + uprime)2 + 1minus

radica

2

int

partD(radicav minus uprime)2

]

dt

uprime(0 middot)2 + av(0 middot)2

∣

∣

∣

∣

∣

∣

le C(T ) (C9)

For periodic boundary conditions we may choose C(T ) = 1 Moreover if a ge 1 anduprime = minusradic

av is specified at partD+ then uprime(t middot)2 + av(t middot)2 decaysStill working on a single interval (subdomain) we now consider the semi-discrete scheme

for (C4) i e (57) with all nonlinear source terms dropped and with v replaced by vradica

Derivation of a formula analogous to (C8) is our first step toward establishing L2 stabilityof the semi-discrete scheme While (57) features vectors for example u(t) taking values atthe Legendre-Gauss-Lobatto nodal points here we work with the numerical solution as apolynomial for example uh(t x) These two representations are related by the Lagrange in-terpolating polynomials for the nodal set here taken to span both the space of test functionsand the space of basis functions Our scheme is

int

Dk

ψparttuh =

int

Dk

ψ(Qh +radicavh) (C10a)

int

Dk

ξparttvh = minusint

Dk

ξprime(radicaQh + vh) +

int

partDk

ξ(radicaQlowast + vlowast) (C10b)

int

Dk

ϕQh =

int

Dk

ϕuprimeh +

int

partDk

ϕ (ulowast minus uh) (C10c)

where ψ ξ and ϕ are polynomial test functions These test functions are arbitrary exceptthat they must be degree-N polynomials In (C10) the variables uh vh and Qh should alsocarry a superscript k but we have suppressed this Derivation of a formula analogous to(C8) is complicated by the fact that Qh is not evolved Nevertheless at a given instant t wecan assemble Qh from (C10c)

Mimicking the calculation (C5b) from the continuum case we first use (C10b) with ξ = vhto write

1

2partt

int

Dk

v2h = minusint

Dk

(radicaQh + vh)v

primeh +

int

partDk

(radicaQlowast + vlowast)vh

= minusint

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2(radicaQlowast + vlowast)vh minus v2h

]

(C11)

The right-hand side of (C5a) is analogous to

1

2partt

int

Dk

Q2h =

int

Dk

QhparttQh (C12)

However since Qh is not evolved the term parttQh must be given a suitable interpretationOn the right hand side of (C10c) only uh u

primeh and u

lowast necessarily depend on time since the

27

test function ϕ need not be time-dependent Furthermore ulowast is explicitly given as a linearcombination of uh as seen in Eq (C21c) below Choosing ϕ = ℓj taking the time derivativeof (C10c) and appealing to the commutivity of mixed partial derivatives we therefore arriveat

int

Dk

ℓjparttQh =

int

Dk

ℓj(parttuh)prime +

int

partDk

ℓj(

(parttu)lowast minus parttuh

)

(C13)

where (parttu)lowast depends on parttuh in precisely the same way that ulowast depends on uh We have

written ℓj rather than ϕ in the last equation to emphasize that the result also holds for anylinear combination of ℓj (for example ϕ) and even for time-dependent combinations SinceQh is itself such a combination we obtain

1

2partt

int

Dk

Q2h =

int

Dk

Qh(parttuh)prime +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

Qh(Qprimeh +

radicavprimeh) +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2((parttu)lowast minus parttuh)Qh +Q2

h

]

(C14)

Addition of (C11) and (C14) gives

1

2partt

int

Dk

(Q2h + v2h) =

1

2

int

partDk

[

Q2h minus v2h + 2(

radicaQlowast + vlowast)vh + 2((parttu)

lowast minus parttuh)Qh

]

(C15)

the aforementioned analog of (C8) This formula holds on a single subdomain Dk and we

now combine multiple copies of it one for each value of k

2 Analysis for multiple intervals

To facilitate combination of (C15) over all k we change notation At every subdomaininterface Ik+12 equiv partDk cap partDk+1 let the superscripts L and R denote field values respectivelytaken from the left and right Then the fields evaluated at Ik+12 which belong to D

k will beuLk+12 v

Lk+12 and Q

Lk+12 while those belonging to D

k+1 will be uRk+12 vRk+12 and Q

Rk+12

However at Ikminus12 the values taken from D

k are uRkminus12 vRkminus12 and QR

kminus12 Note that we

have also replaced the subscript h denoting a numerical solution with k plusmn 12 denotingthe location of the endpoint value of the numerical solution With this notation we define

∆Lα =

1

2

[

(QLα)

2 minus (vLα)2]

+(radic

aQlowastα + vlowastα

)

vLα +[

(parttuα)lowast minus parttu

Lα

]

QLα (C16)

and similarly for ∆Rα The same numerical fluxes appear in both ∆L

α and ∆Rα (ie each

numerical flux takes the same value on either side of an interface) whence fluxes like Qlowastα do

not carry an L or R superscript In terms of these definitions (C15) becomes

1

2partt

int

Dk

(Q2h + v2h) = ∆L

k+12 minus∆Rkminus12 (C17)

28

Summation over all Dk yields

1

2partt

kmaxsum

k=1

int

Dk

(Q2h + v2h) =

kmaxminus1sum

k=1

(

∆Lk+12 minus∆R

k+12

)

+∆Lkmax+12 minus∆R

12

=

kmaxminus1sum

k=1

(

∆Lh minus∆R

h

)∣

∣

Ik+12 +∆Lkmax+12 minus∆R

12 (C18)

We have reverted to h-notation denoting the numerical solution since the LR superscriptsindicate unambiguously the relevant domain used for evaluation at Ik+12

We again seek an estimate of the form

kmaxsum

k=1

(

Qh(T middot)2Dk + avh(T middot)2Dk

)

le C(T )kmaxsum

k=1

(

Qh(0 middot)2Dk + avh(0 middot)2Dk

)

(C19)

that is essentially the same as the one (C3) considered in the continuum case Assume thatthe chosen boundary conditions ensure ∆L

kmax+12 minus ∆R12 is bounded by a time-dependent

constant which does not depend on the numerical parameters N and h (subdomain width)Establishment of stability then amounts to showing that the remaining sum over interfaceterms in (C18) is non-positive whence this remaining sum is consistent with C(T ) le 1although the boundary conditions may give rise to a different bound In fact we will choosethe numerical fluxes such that each individual interface term is non-positive At interfaceIk+12 and in LR notation the jump and average of vh for example are

1

2

(

v+ + vminus)

equiv vh =1

2

(

vLk+12 + vRk+12

)

(C20a)

nminusvminus + n+v+ equiv[[

vh]]

= vLk+12 minus vRk+12 (C20b)

Consider numerical fluxes of the form

Qlowast = Qh minusτQ2

[[

Qh

]]

(C21a)

vlowast = vh minusτv2

[[

vh]]

(C21b)

ulowast = uh minusτu2

[[

uh]]

(C21c)

(parttu)lowast = parttuh minus

τu2

[[

parttuh]]

(C21d)

where (C21c) induces (C21d) and where the penalty parameters τu τv and τQ are realnumbers The fluxes defined in (58) correspond to τu = 1 τv = 1 +

radica and τQ = 0 In

terms of these quantities the kth interface contribution in (C18) is

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

parttuh]]

minus τu2

[[

parttuh]][[

Qh

]]

(C22)

where we have suppressed the k dependence of the right-hand side Now consider the term[[

parttuh]]

Because parttuh and Qh+radicavh are both polynomials of degree N Eq (C10a) implies

29

τu

τ Q

minus1 0 1minus1

minus05

0

05

1

τu

minus1 0 1minus1

minus05

0

05

1

FIG 6 Stable evolutions for the model system For fixed τv = 10minus6 and τv = 1 +radic2

respectively the left and right plots depict stable choices (determined empirically) of τu and τQfor the linear model system (C1) The stable regions are colored black but the jagged edges result

from the discretization of the (τu τQ)-plane

the vector equation parttu = Q +radicav that is pointwise equivalence on the nodal points of

Dk which in turn implies

[[

parttuh]]

=[[

Qh+radicavh]]

Upon substituting this identity into thelast equation we arrive at an expression which features only vh and Qh

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

Qh +radicavh]]

minus τu2

[[

Qh +radicavh]][[

Qh

]]

(C23)

The identities vh[[

vh]]

= 12

[[

v2h]]

and[[

Qh +radicavh]]

=[[

Qh

]]

+radica[[

vh]]

then simplify(C23) to

(∆Lh minus∆R

h )∣

∣

Ik+12 = minusτv2

[[

vh]]2 minus

radica(τu + τQ)

2

[[

Qh

]][[

vh]]

minus τu2

[[

Qh

]]2 (C24)

The role of a penalty parameter is now clear Positive values of τv penalize jumps in vhthrough a negative contribution to the energy Likewise positive values of τu penalizejumps in Qh through a negative contribution to the energy However because the sign of[[

Qh

]][[

vh]]

can be positive or negative only the choice τQ = minusτu yields an expression for(∆L

h minus ∆Rh )|Ik+12 which is manifestly negative for τu ge 0 and τv ge 0 A simple estimate

based on Youngrsquos inequality with ε (that is 2αβ le εminus1α2 + εβ2 where α β ge 0 and ε gt 0)shows that for τQ = 0 the choice τv ge aτu4 also yields stability

Figure 6 depicts certain choices of stable penalty parameters for the linear model systemevolved to tfinal = 1000 (with a = 2 N = 10 and ∆t ≃ 00553) as determined empiricallywith simulations similar to those described in Sec IVA The left plot corresponds to asmall τv = 10minus6 for which the choice τu = 1 τQ = 0 is not stable as expected from thetheoretical analysis However the right plot corresponds to τv = 1 +

radica for which τu = 1

τQ = 0 is stable Motivated by the numerical flux choices (3842) used for the BSSN system

30

(4) we have (as mentioned above) set τu = 1 τv = 1 +radica and τQ = 0 in simulations of

the nonlinear model (53) For the nonlinear model system (53) the theoretically motivatedchoice τQ = minusτu also yields numerically stable evolutions when τu ge 0 and τv ge 0

For the nonlinear systems (4) and (53) we do not attempt a formal stability proofNevertheless the results of this appendix have served as a guide for our choices of penaltyparameters For the BSSN system (4) u v and Q are block indices [cf Eq (6)] Similarto the model problem we have penalized Q with τu = 1 with τv chosen large enough toheuristically overcome the cross-terms of indefinite size that arise from τQ = 0 (we interpretequations like τu = 1 componentwise) An analogous choice ldquoτQ = minusτurdquo for the BSSNsystem might be possible but would be considerably more complicated Indeed such achoice likely entails a matrix of penalty parameters but we do not give the details here

[1] F Pretorius Evolution of Binary Black-Hole Spacetimes Phys Rev Lett 95 121101 (2005)

(4 pages)

[2] F Pretorius Simulation of binary black hole spacetimes with a harmonic evolution scheme

Class Quantum Grav 23 S529-S552 (2006)

[3] M Campanelli C O Lousto P Marronetti and Y Zlochower Accurate Evolutions of Or-

biting Black-Hole Binaries without Excision Phys Rev Lett 96 111101 (2006) (4 pages)

[4] J G Baker J Centrella D-I Choi M Koppitz and J van Meter Gravitational-Wave

Extraction from an Inspiraling Configuration of Merging Black Holes Phys Rev Lett 96

111102 (2006) (4 pages)

[5] M Campanelli C O Lousto and Y Zlochower Last orbit of binary black holes Phys Rev D

73 061501(R) (2006) (5 pages)

[6] F Herrmann I Hinder D Shoemaker and P Laguna Unequal mass binary black hole plunges

and gravitational recoil Class Quantum Grav 24 S33-S42 (2007)

[7] P Diener F Herrmann D Pollney E Schnetter E Seidel R Takahashi J Thornburg and

J Ventrella Accurate Evolution of Orbiting Binary Black Holes Phys Rev Lett 96 121101

(2006) (4 pages)

[8] M A Scheel H P Pfeiffer L Lindblom L E Kidder O Rinne and S A Teukolsky Solving

Einsteinrsquos Equations With Dual Coordinate Frames Phys Rev D 74 104006 (2006)

[9] U Sperhake Binary black-hole evolutions of excision and puncture data Phys Rev D 76

104015 (2007) (20 pages)

[10] B Brugmann J A Gonzalez M Hannam S Husa U Sperhake and W Tichy Calibration

of moving puncture simulations Phys Rev D 77 024027 (2008) (25 pages)

[11] P Marronetti W Tichy B Brugmann J Gonzalez M Hannam S Husa and U Sperhake

Binary black holes on a budget simulations using workstations Class Quantum Grav 24

S43-S58 (2007)

[12] Z B Etienne J A Faber Y T Liu S L Shapiro and T W Baumgarte Filling the holes

Evolving excised binary black hole initial data with puncture techniques Phys Rev D 76

101503(R) (2007) (5 pages)

[13] B Szilagyi D Pollney L Rezzolla J Thornburg and J Winicour An explicit harmonic code

for black-hole evolution using excision Class Quantum Grav 24 S275-S293 (2007)

[14] L Boyle M Kesden and S Nissanke Binary-Black-Hole Merger Symmetry and the Spin

Expansion Phys Rev Lett 100 151101 (2008) (4 pages)

31

[15] M A Scheel M Boyle T Chu L E Kidder K D Matthews and H P Pfeiffer High-

accuracy waveforms for binary black hole inspiral merger and ringdown Phys Rev D 79

024003 (2009) (14 pages)

[16] M Hannam Status of black-hole-binary simulations for gravitational-wave detection

Class Quantum Grav 26 114001 (2009) (21 pages)

[17] I Hinder The current status of binary black hole simulations in numerical relativity

Class Quantum Grav 27 114004 (2010) (19 pages)

[18] B P Abbott et al LIGO The Laser Interferometer Gravitational-Wave Observatory

Rep Prog Phys 72 076901 (2009) (25 pages)

[19] F Acernese et al The status of VIRGO Class Quantum Grav 23 S63-S69 (2006)

[20] F Acernese et al Virgo status Class Quantum Grav 25 184001 (2008) (9 pages)

[21] R Arnowitt S Deser and C W Misner The Dynamics of General Relativity in Gravitation

an introduction to current research edited by L Witten (Wiley New York 1962) Chapter 7

227-265 Also available as gr-qc0405109

[22] J W York Jr Kinematics and Dynamics of General Relativity in Sources of Gravitational

Radiation edited by L L Smarr (Cambridge University Press Cambridge 1979)

[23] L E Kidder M A Scheel and S A Teukolsky Extending the lifetime of 3D black hole

computations with a new hyperbolic system of evolution equations Phys Rev D 64 064017

(2001) (13 pages)

[24] G Calabrese J Pullin O Sarbach and M Tiglio Convergence and stability in numerical

relativity Phys Rev D 66 041501(R) (2002) (4 pages)

[25] S Frittelli and O Reula On the Newtonian Limit of General Relativity Commun Math

Phys 166 221-235 (1994)

[26] M Shibata and T Nakamura Evolution of three-dimensional gravitational waves Harmonic

slicing case Phys Rev D 52 5428-5444 (1995)

[27] Y Choquet-Bruhat and J W York Geometrical well posed systems for the Einstein equations

C R Acad Sci Paris t 321 Serie I 1089-1095 (1995)

[28] A Abrahams A Anderson Y Choquet-Bruhat and J W York Jr Einstein and Yang-Mills

Theories in Hyperbolic Form without Gauge Fixing Phys Rev Lett 75 3377-3381 (1995)

[29] C Bona J Masso E Seidel and J Stela New Formalism for Numerical Relativity

Phys Rev Lett 75 600-603 (1995)

[30] M H P M van Putten and D M Eardley Nonlinear wave equations for relativity

Phys Rev D 53 3056-3063 (1996)

[31] S Frittelli and O A Reula First-Order Symmetric Hyperbolic Einstein Equations with Ar-

bitrary Fixed Gauge Phys Rev Lett 76 4667-4670 (1996)

[32] H Friedrich Hyperbolic reductions for Einsteinrsquos equations Class Quantum Grav 13 1451-

1469 (1996)

[33] F B Estabrook R S Robinson H D Wahlquist Hyperbolic equations for vacuum gravity

using special orthonormal frames Class Quantum Grav 14 1237-1247 (1997)

[34] M S Iriondo E O Leguizamon and O A Reula Einsteinrsquos Equations in Ashtekarrsquos Vari-

ables Constitute a Symmetric Hyperbolic System Phys Rev Lett 79 4732-4735 (1997)

[35] A Anderson Y Choquet-Bruhat and J W York Einstein-Bianchi hyperbolic system for

general relativity Topol Meth Nonlinear Anal 10 353-373 (1997)

[36] T W Baumgarte and S L Shapiro Numerical integration of Einsteinrsquos field equations

Phys Rev D 59 024007 (1998) (7 pages)

[37] M A G Bonilla Symmetric hyperbolic systems for Bianchi equations Class Quantum

17

8 16 32 64 128

10minus3

10minus2

10minus1

N

∆t

max stable ∆t8N2

FIG 2 Scaling of maximum stable ∆t with N for model PDE

two possibilities (i) the analytic state (Q+ v+) = (Qexact vexact) and (ii) an upwind stateFor example at x = 4π the upwind state is4

Q+ = Qupwind =1

2

[

(Xminus)exact minus (X+)numer

]

v+ = vupwind =1

2radica

[

(Xminus)exact + (X+)numer

]

(60)Either choice of (Q+ v+) leads to similar results and the plots here correspond to theanalytic state Figure 1 clearly shows spectral convergence with increasing polynomial orderN across all fields for the case a = 2 Other values of a including a = 1 for which X+

is a static characteristic field have also been considered with similar results Appendix Cdemonstrates that our proposed scheme for the system (57) with nonlinear and source termsdropped is stable in a semi-discrete sense Nevertheless the fully discrete scheme obtainedvia temporal discretization by the fourth-order Runge-Kutta method is still subject to thestandard absolute stability requirement Namely if microh is any eigenvalue corresponding tothe (linearized) discrete spatial operator then a necessary condition for stability is thatmicroh∆t lies in absolute stability region for Runge-Kutta 4 We here show empirically that theassociated timestep restriction scales like Nminus2 ie ∆t = O(Nminus2) for stability We note thatsuch scaling is welcome in light of the second-order spatial operators which appear in thesystem and suggest a possible worse scaling like Nminus4 Fig 2 plots the maximum stabletimestep for a range of N demonstrating the Nminus2 scaling in line with behavior known fromanalysis of first-order systems [66] This scaling also holds for the BSSN system

4 We remind the reader that unfortunately the plusmn on Xplusmn means something different than the plusmn indicating

exteriorinterior dG states [cf footnote 3]

18

B Simulations of the BSSN system

This subsection documents results for simulations of the unit-mass-parameter (M = 1)Schwarzschild solution (B9) expressed in terms of ingoing Kerr-Schild coordinates Sincethe solution is stationary temporal integration of the semi-discrete scheme has been carriedout with the forward Euler method The r-coordinate domain [04 34] has been split into3 equally spaced subdomains and we have set η = 10 λ = 01 and C = 2 [cf Eq (41)]For all simulations ∆t has been chosen for stability With the chosen λ the inner physicalboundary rmin = 04 is an excision surface At each timestep we have applied an (order2s = 20) exponential filter on the top two-thirds of the modal coefficient set for all fieldsexcept for grr and gθθ For stability we have empirically observed that grr and gθθ must not

be filtered A detailed understanding of this is still lackingIssues related to physical boundary conditions are similar to the one encountered in

Sec IVA for the model problem Similar to before we have retained Eqs (3842) as thechoice of numerical flux even at the endpoints Therefore at an endpoint the specificationof the boundary condition amounts to the choice W+ of external state We have typicallychosen the inner boundary of the radial domain as an excision boundary and in this caseW+ = Wminus is enforced at the inner physical boundary At the outer physical boundaryfor W+ we have again considered two choices (i) Wexact and (ii) Wupwind To enforcechoice (ii) the inverse transformation (17) must be used with incoming characteristic fieldsfixed to their exact values similar to (60) We have tried various versions of choice (ii)and in all cases the resulting simulations have been unstable We therefore present resultscorresponding to choice (i) Although the choice of an analytical external state Wexact

at the outer boundary is stable for our problem such a boundary condition is unlikelyto generalize to more complicated scenarios involving dynamical fields Indeed the issueof outer boundary conditions for the BSSN system is an active area of research with aproper treatment requiring fixation of incoming radiation control of the constraints andspecification of gauge (see Ref [86] for a recent analysis)

For BSSN simulations our main diagnostic is to monitor the Hamiltonian momentumand conformal connection constraints Figure 3 depicts long-time histories of constraintviolations whereas Figs 4 and 5 depict long-time error histories for the individual BSSNfield components From the middle plot in Fig 5 we infer that up to the indicated numericalerror the factor g sin4 θ = grr(gθθ)

2 remains at its initial fixed profile r4 throughout theevolution These figures indicate that the proposed scheme is stable for long times andexhibits spectral converge with increased polynomial order N Similar results are recoveredfromM = 0 Minkowski initial data The stability documented in these plots does not appearto rely on inordinate parameter tuning For example with the fixed parameters describedabove we obtain similar plots if we individually vary (i) rmin over 0325 035 04 0475(values still corresponding to an excision surface for the given choice of λ) (ii) η over1 3 7 10 (iii) s over 8 9 10 With the polynomial order N ranging over 23 26 29 31both stability and qualitatively similar exponential convergence is achieved with a singlesubdomain Likewise adoption of a larger coordinate domain with more subdomains doesnot impact our results Finally we have considered the addition of random noise to all fieldcomponents at the initial time Precisely with the system component χ as an example wehave set

χ(0) 7rarr χ(0) + δχ(0) (61)

where each component (nodal value) of δχ(0) is 10minus5 times a random variable drawn from a

19

10minus9

10minus7

10minus5

10minus3

Hinfin

10minus9

10minus7

10minus5

10minus3

Mrinfin

0 2000 4000 6000 8000 10000

10minus9

10minus7

10minus5

10minus3

Grinfin

t

N = 11N = 14N = 17N = 20

FIG 3 Spectral convergence of constraint violations for M = 1 Kerr-Schild ini-

tial data Respectively for N = 11 14 17 19 a timestep of ∆t ≃ 00041 00026 00018 00013

has been chosen for stability and accuracy

standard normal distribution Such perturbed initial data also gives rise to stable evolutions

V CONCLUSION

We have introduced a discontinuous Galerkin method for solving the spherically reducedBSSN system with second-order spatial operators Our scheme shares similarities withother discontinuous Galerkin methods that use local auxiliary variables to handle high-orderspatial derivatives [66 77ndash79 82 84 85] and which have typically been applied to eitherelliptic parabolic or mixed type problems The key ingredient of a stable dG scheme is anappropriate choice of numerical flux and our particular choice has been motivated by theanalysis presented in Appendix C When used to evolve the Schwarzschild solution in Kerr-Schild coordinates our numerical implementation of the BSSN system (4) is robustly stableand converges to the analytic solution exponentially with increased polynomial order Byapproximating the spatially second-order form of the BSSN system we have not introducedextra fields which are evolved Evolved auxiliary fields result in new constraints which mayspoil stability Our main goal has been stable evolution of the spherically reduced BSSNsystem as a first step towards understanding how a discontinuous Galerkin method might be

20

10minus9

10minus7

10minus5

∆Arrinfin

10minus9

10minus7

10minus5

∆Kinfin

0 2000 4000 6000 8000 10000

10minus9

10minus7

10minus5

max(∆Brinfin ∆Γrinfin)

t

N = 11N = 14N = 17N = 20

FIG 4 Spectral convergence of solution for M = 1 Kerr-Schild initial data

Timestep choices are described in the caption for Fig 3 In the title headings for example

∆Arr equiv (Arr)numer minus (Arr)exact

applied to the full BSSN system Towards that goal we now discuss treatment of singularitiesand generalization of the described dG method to higher space dimension

To deal with the fixed Schwarzschild singularity we have used excision which is easy inthe context of the spherically reduced BSSN system However excision for the binary blackhole problem in full general relativity requires attention to the technical challenge of hori-zon tracking State-of-the-art BSSN codes avoid such complication relying instead on themoving-puncture technique While the moving-puncture technique does involve mild centralsingularities it may still prove amenable to spectral methods Indeed spectral methods fornon-smooth problems is well-developed in both theory and for complex applications Sincethe moving-puncture technique can be performed in spherical symmetry [60] a first-steptoward a spectral moving-puncture code would be to implement a moving puncture withthe nodal dG method described here Such an implementation may adopt Legendre-Gauss-Radau nodes on the innermost subdomain thereby ensuring that the physical singularitydoes not lie on a nodal point (in much the same way finite difference codes use a staggeredgrid) Beyond traditional excision and moving punctures one might also retain a smoothproblem for which spectral methods are especially well-suited via the turducken approach[87 88] to singularities

21

10minus11

10minus9

10minus7

10minus5

∆χinfin

10minus11

10minus9

10minus7

10minus5

max(∆grrinfin ∆gθθinfin)

0 2000 4000 6000 8000 10000

10minus11

10minus9

10minus7

10minus5

max(∆αinfin ∆βrinfin)

t

N = 11N = 14N = 17N = 20

FIG 5 Spectral convergence of solution violations for M = 1 Kerr-Schild initial

data See the caption of Fig 4 for details

Discontinuous Galerkin methods for hyperbolic problems in two and three space dimen-sions are well-developed A generalization of the method described here to three-dimensionsand the full BSSN system would likely rely on an unstructured mesh Appropriate local poly-nomial expansions for the subelements are well-understood as are choices for the numericalfluxes which would now live on two-dimensional faces rather than single points Whether ornot it would ultimately prove successful generalization of our dG method to a higher dimen-sion would rely on an established conceptual framework Further computational advancesof relevance to a generalization of our dG method to the full BSSN system (possibly includ-ing matter) may include mesh hp-adaptivity local timestepping shock capturing and slopelimiting techniques [66] Moreover recent work [89] indicates that enhanced performancewould be expected were our scheme implemented on graphics processor units

VI ACKNOWLEDGMENTS

We thank Nick Taylor for discussions about treating second-order operators with spectralmethods Benjamin Stamm for helping polish up a few parts of Appendix C Khosro Shah-bazi for discussions on LDG and IP methods We also acknowledge helpful conversationswith David Brown and Manuel Tiglio about the BSSN system and previous work on its nu-

22

merical implementation We gratefully acknowledge funding through grants DMS 0554377and DARPAAFOSR FA9550-05-1-0108 to Brown University and NSF grant PHY 0855678to the University of New Mexico

Appendix A Hyperbolicity of the first-order system

This appendix analyzes the matrix A(u) appearing in (10) in order to construct thecharacteristic fields (14) In matrix form the sector (8) of the principal part of (10) reads asfollows

partt

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

=

βr 0 minus 4λα3grr

0 0 λβr

6(grr)2λβr

3gθθgrr0 4λ

3grr

0 βr 0 23grrαχ

13α minus αχ

3grr

αχ3gθθ

minus23χ 0

0 0 βr 0 0 0 0 minus χgrr

0

0 0 minus 4α3grr

βr 0 βr

6(grr)2βr

3gθθgrr0 4

3grr

0 0 23αχ 0 βr minus βrχ

3grrminus2βrχ

3gθθ0 minus2

3χ

0 minus2α 0 0 0 23βr minus2grrβr

3gθθ0 4

3grr

0 gθθαgrr

0 0 0 minusgθθβr

3grr13βr 0 minus2

3gθθ

0 0 minus2α 0 0 0 0 βr 0

34

0 0 0 0 0 0 0 βr

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

prime

(A1)which defines the matrix A(u) appearing in (9) and so also the matrix A(u) in (10) Notethat in the last equation the matrix within the square brackets is minusA(u) For certainconfigurations of u and λ the system (10) is strongly hyperbolic [76] that is A(u) hasa complete set of eigenvectors and real eigenvalues Indeed five eigenpairs of A(u) aretrivially recovered upon inspection of A(u)rsquos leading 5times5 diagonal block These correspondto eigenvalue 0 and the left eigenspace ξj = eTj 1 le j le 5 where ej are the canonical

basis vectors Since each component of u arises as eTj W each is also a characteristic field

The remaining nine eigenpairs are determined by A(u) The eigenvalues of A(u) are

micro1 = 0 micro23 = minusβr microplusmn4 = minusβr plusmn

radic

2αχ

grr microplusmn

5 = minusβr plusmn α

radic

χ

grr microplusmn

6 = minusβr plusmnradic

λ

grr

(A2)

23

and the corresponding left eigenvectors are

x1 = (0 0 0 0 0 gθθ 2grr 0 0) (A3a)

x2 =

(

0 0 0 grr2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3b)

x3 =

(

grrλ 0 0 0

2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3c)

xplusmn4 =

(

0 0plusmnradic

2αgrrχ

0 0 0 0 1 0

)

(A3d)

xplusmn5 =

(

0∓ 3radicgrrχ

plusmn2

radic

grrχ 2grr

1

χminus 1

grr1

gθθ 0 0

)

(A3e)

xplusmn6 =

(

minus3

4

grrλ 0plusmn α

radicλgrr

(2αχminus λ) 0 0minus βr

8(βrgrr ∓radicλgrr)

minus βrgrr

4gθθ(βrgrr ∓radicλgrr)

αχ

(2αχminus λ)plusmnradic

grrλ

)

(A3f)

where for example xplusmn5 A(u) = microplusmn5 x

plusmn5 Assuming that grr gθθ χ and α are everywhere strictly

positive the eigenvalues are real and the eigenvectors are linearly independent providedthat (15) holds These eigenvectors are easily extended to eigenvectors of A(u) e g asxplusmn6 rarr (01times5 x

plusmn6 ) Then for example the characteristic field

Xplusmn6 equiv (01times5 x

plusmn6 )W = xplusmn6WvQ (A4)

and similarly Xplusmnj = xplusmnj WvQ for j = 4 5 and Xk = xkWvQ for k = 1 2 3 The characteristic

speeds for these fields are microk and microplusmnj With this convention the speeds listed in Table I

correspond to the Xk and Xplusmnj in (14)

Appendix B Schwarzschild solution in conformal Kerr-Schild coordinates

In Kerr-Schild coordinates here the system directly related to incoming Eddington-Finkelstein null coordinates the line element for the Schwarzschild solution reads

ds2 = minusα2dt2 + (1 + 2MR)(dR + βRdt)2 +R2dθ2 +R2 sin2 θdφ2 (B1)

where R is the area radius α = (1 + 2MR)minus12 is the lapse and βR = 2M(R + 2M) isthe shift vector The physical spatial metric gab is the spatial part of this line element

To define the corresponding solution to the BSSN system we use equation gab = χgab todefine the following relationship between line elements

dr2 + r2(dθ2 + sin2 θdφ2) = χ[(1 + 2MR)dR2 +R2dθ2 +R2 sin2 θdφ2] (B2)

so that

χ

(

1 +2M

R

)(

dR

dr

)2

= 1 χR2 = r2 (B3)

Then we have(

1 +2M

R

)12dR

R=dr

r (B4)

24

with integration yielding

r =R

4

(

1 +

radic

1 +2M

R

)2

e2minus2radic

1+2MR (B5)

where the constant of integration has been chosen so that the R r rarr infin limits are consistentThe second relation in (B3) shows that

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR =rArr χminus4 =2eradic

1+2MRminus1

1 +radic

1 + 2MR (B6)

The extrinsic curvature tensor is specified by the expression forK given in (B9h) the identityK = KR

R + 2Kθθ and

Kθθ =

(

1 +2M

R

)minus122M

R2 (B7)

Since KRR = Kr

r we compute that

Krr = K minus 2Kθ

θ = minus(

1 +2M

R

)minus12(R +M

R + 2M

)

2M

R2 (B8)

Next since Krr = grrKrr = χminus1Kr

r we have Krr = Arr+

13grrK This implies Arr = Kr

r minus 13K

from which we get (B9g) In all we have

α =

(

1 +2M

R

)minus12

(B9a)

βr = βR dr

dR= χ12

(

1 +2M

R

)minus122M

R(B9b)

grr = 1 (B9c)

gθθ = r2 = χR2 (B9d)

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR (B9e)

Br = 0 (B9f)

Arr = minus(

1 +2M

R

)minus124M

3R2

(

2R + 3M

R + 2M

)

(B9g)

K =

(

1 +2M

R

)minus32(

1 +3M

R

)

2M

R2(B9h)

Γr = minus2

r= minus 2

χ12R (B9i)

To differentiate these expressions with respect to r we use the identity

dR

dr= χminus12

(

1 +2M

R

)minus12

(B10)

along with the chain rule

25

Appendix C Stability of the model system

The following stability analysis for the model system (53) has been inspired by [78 79]After dropping all nonlinear source terms the system (53) becomes

parttu = uprime + av (C1a)

parttv = uprimeprime + vprime (C1b)

This section analyzes the stability of (C1) considering both the continuum system itself aswell as its semi-discrete dG approximation The latter analysis offers some insight into theempirically observed stability of our dG scheme for the spherically reduced BSSN equations

1 Analysis for a single interval

Throughout we work with the L2 inner product and norm

(f g)D =

int

D

fg fD =radic

(f f)D (C2)

where D is the spatial coordinate interval (here D may represent a subdomain Dk or the

whole domain Ω) and we have suppressed all integration measures For the continuummodel we will establish the following estimate

uprime(T middot)2D+ av(T middot)2

Dle C(T )

(

uprime(0 middot)2D+ av(0 middot)2

D

)

(C3)

where the time-dependent constant C(T ) is determined solely by the choice of boundaryconditions To show (C3) we first change variables with v =

radicav thereby rewriting (C1)

in the following symmetric form

parttu = uprime +radicav (C4a)

parttv =radicauprimeprime + vprime (C4b)

Equations (C4ab) then imply

1

2partt

int

D

(uprime)2 =

int

D

uprime(uprimeprime +radicavprime) =

int

D

radicauprimevprime +

1

2

int

partD

(uprime)2 (C5a)

1

2partt

int

D

(v)2 =

int

D

v(radicauprimeprime + vprime) = minus

int

D

radicauprimevprime +

1

2

int

partD

(v2 + 2radicauprimev) (C5b)

Here vvprime and uprimeuprimeprime have been expressed as exact derivatives and then integrated to boundaryterms the second equation employs an extra integration by parts and with only one spacedimension

int

partDdenotes a difference of endpoint evaluations Addition of Eqs (C5ab) gives

1

2partt

int

D

[

v2 + (uprime)2]

=1

2

int

partD

[

v2 + (uprime)2 + 2radicauprimev

]

(C6)

Substitutions with the identities

[

v2 + (uprime)2]

=1

2

[

(v + uprime)2 + (v minus uprime)2]

2uprimev =1

2

[

(v + uprime)2 minus (v minus uprime)2]

(C7)

26

and replacements to recover the original variable v = vradica yield

1

2partt

int

D

[

av2 + (uprime)2]

=1 +

radica

4

int

partD

(radicav + uprime)2 +

1minusradica

4

int

partD

(radicav minus uprime)2 (C8)

From (C8) we deduce that the time-dependent constant C(T ) in (C3) must satisfy

∣

∣

∣

∣

∣

∣

1 +

int T

0

[

1+radica

2

int

partD(radicav + uprime)2 + 1minus

radica

2

int

partD(radicav minus uprime)2

]

dt

uprime(0 middot)2 + av(0 middot)2

∣

∣

∣

∣

∣

∣

le C(T ) (C9)

For periodic boundary conditions we may choose C(T ) = 1 Moreover if a ge 1 anduprime = minusradic

av is specified at partD+ then uprime(t middot)2 + av(t middot)2 decaysStill working on a single interval (subdomain) we now consider the semi-discrete scheme

for (C4) i e (57) with all nonlinear source terms dropped and with v replaced by vradica

Derivation of a formula analogous to (C8) is our first step toward establishing L2 stabilityof the semi-discrete scheme While (57) features vectors for example u(t) taking values atthe Legendre-Gauss-Lobatto nodal points here we work with the numerical solution as apolynomial for example uh(t x) These two representations are related by the Lagrange in-terpolating polynomials for the nodal set here taken to span both the space of test functionsand the space of basis functions Our scheme is

int

Dk

ψparttuh =

int

Dk

ψ(Qh +radicavh) (C10a)

int

Dk

ξparttvh = minusint

Dk

ξprime(radicaQh + vh) +

int

partDk

ξ(radicaQlowast + vlowast) (C10b)

int

Dk

ϕQh =

int

Dk

ϕuprimeh +

int

partDk

ϕ (ulowast minus uh) (C10c)

where ψ ξ and ϕ are polynomial test functions These test functions are arbitrary exceptthat they must be degree-N polynomials In (C10) the variables uh vh and Qh should alsocarry a superscript k but we have suppressed this Derivation of a formula analogous to(C8) is complicated by the fact that Qh is not evolved Nevertheless at a given instant t wecan assemble Qh from (C10c)

Mimicking the calculation (C5b) from the continuum case we first use (C10b) with ξ = vhto write

1

2partt

int

Dk

v2h = minusint

Dk

(radicaQh + vh)v

primeh +

int

partDk

(radicaQlowast + vlowast)vh

= minusint

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2(radicaQlowast + vlowast)vh minus v2h

]

(C11)

The right-hand side of (C5a) is analogous to

1

2partt

int

Dk

Q2h =

int

Dk

QhparttQh (C12)

However since Qh is not evolved the term parttQh must be given a suitable interpretationOn the right hand side of (C10c) only uh u

primeh and u

lowast necessarily depend on time since the

27

test function ϕ need not be time-dependent Furthermore ulowast is explicitly given as a linearcombination of uh as seen in Eq (C21c) below Choosing ϕ = ℓj taking the time derivativeof (C10c) and appealing to the commutivity of mixed partial derivatives we therefore arriveat

int

Dk

ℓjparttQh =

int

Dk

ℓj(parttuh)prime +

int

partDk

ℓj(

(parttu)lowast minus parttuh

)

(C13)

where (parttu)lowast depends on parttuh in precisely the same way that ulowast depends on uh We have

written ℓj rather than ϕ in the last equation to emphasize that the result also holds for anylinear combination of ℓj (for example ϕ) and even for time-dependent combinations SinceQh is itself such a combination we obtain

1

2partt

int

Dk

Q2h =

int

Dk

Qh(parttuh)prime +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

Qh(Qprimeh +

radicavprimeh) +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2((parttu)lowast minus parttuh)Qh +Q2

h

]

(C14)

Addition of (C11) and (C14) gives

1

2partt

int

Dk

(Q2h + v2h) =

1

2

int

partDk

[

Q2h minus v2h + 2(

radicaQlowast + vlowast)vh + 2((parttu)

lowast minus parttuh)Qh

]

(C15)

the aforementioned analog of (C8) This formula holds on a single subdomain Dk and we

now combine multiple copies of it one for each value of k

2 Analysis for multiple intervals

To facilitate combination of (C15) over all k we change notation At every subdomaininterface Ik+12 equiv partDk cap partDk+1 let the superscripts L and R denote field values respectivelytaken from the left and right Then the fields evaluated at Ik+12 which belong to D

k will beuLk+12 v

Lk+12 and Q

Lk+12 while those belonging to D

k+1 will be uRk+12 vRk+12 and Q

Rk+12

However at Ikminus12 the values taken from D

k are uRkminus12 vRkminus12 and QR

kminus12 Note that we

have also replaced the subscript h denoting a numerical solution with k plusmn 12 denotingthe location of the endpoint value of the numerical solution With this notation we define

∆Lα =

1

2

[

(QLα)

2 minus (vLα)2]

+(radic

aQlowastα + vlowastα

)

vLα +[

(parttuα)lowast minus parttu

Lα

]

QLα (C16)

and similarly for ∆Rα The same numerical fluxes appear in both ∆L

α and ∆Rα (ie each

numerical flux takes the same value on either side of an interface) whence fluxes like Qlowastα do

not carry an L or R superscript In terms of these definitions (C15) becomes

1

2partt

int

Dk

(Q2h + v2h) = ∆L

k+12 minus∆Rkminus12 (C17)

28

Summation over all Dk yields

1

2partt

kmaxsum

k=1

int

Dk

(Q2h + v2h) =

kmaxminus1sum

k=1

(

∆Lk+12 minus∆R

k+12

)

+∆Lkmax+12 minus∆R

12

=

kmaxminus1sum

k=1

(

∆Lh minus∆R

h

)∣

∣

Ik+12 +∆Lkmax+12 minus∆R

12 (C18)

We have reverted to h-notation denoting the numerical solution since the LR superscriptsindicate unambiguously the relevant domain used for evaluation at Ik+12

We again seek an estimate of the form

kmaxsum

k=1

(

Qh(T middot)2Dk + avh(T middot)2Dk

)

le C(T )kmaxsum

k=1

(

Qh(0 middot)2Dk + avh(0 middot)2Dk

)

(C19)

that is essentially the same as the one (C3) considered in the continuum case Assume thatthe chosen boundary conditions ensure ∆L

kmax+12 minus ∆R12 is bounded by a time-dependent

constant which does not depend on the numerical parameters N and h (subdomain width)Establishment of stability then amounts to showing that the remaining sum over interfaceterms in (C18) is non-positive whence this remaining sum is consistent with C(T ) le 1although the boundary conditions may give rise to a different bound In fact we will choosethe numerical fluxes such that each individual interface term is non-positive At interfaceIk+12 and in LR notation the jump and average of vh for example are

1

2

(

v+ + vminus)

equiv vh =1

2

(

vLk+12 + vRk+12

)

(C20a)

nminusvminus + n+v+ equiv[[

vh]]

= vLk+12 minus vRk+12 (C20b)

Consider numerical fluxes of the form

Qlowast = Qh minusτQ2

[[

Qh

]]

(C21a)

vlowast = vh minusτv2

[[

vh]]

(C21b)

ulowast = uh minusτu2

[[

uh]]

(C21c)

(parttu)lowast = parttuh minus

τu2

[[

parttuh]]

(C21d)

where (C21c) induces (C21d) and where the penalty parameters τu τv and τQ are realnumbers The fluxes defined in (58) correspond to τu = 1 τv = 1 +

radica and τQ = 0 In

terms of these quantities the kth interface contribution in (C18) is

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

parttuh]]

minus τu2

[[

parttuh]][[

Qh

]]

(C22)

where we have suppressed the k dependence of the right-hand side Now consider the term[[

parttuh]]

Because parttuh and Qh+radicavh are both polynomials of degree N Eq (C10a) implies

29

τu

τ Q

minus1 0 1minus1

minus05

0

05

1

τu

minus1 0 1minus1

minus05

0

05

1

FIG 6 Stable evolutions for the model system For fixed τv = 10minus6 and τv = 1 +radic2

respectively the left and right plots depict stable choices (determined empirically) of τu and τQfor the linear model system (C1) The stable regions are colored black but the jagged edges result

from the discretization of the (τu τQ)-plane

the vector equation parttu = Q +radicav that is pointwise equivalence on the nodal points of

Dk which in turn implies

[[

parttuh]]

=[[

Qh+radicavh]]

Upon substituting this identity into thelast equation we arrive at an expression which features only vh and Qh

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

Qh +radicavh]]

minus τu2

[[

Qh +radicavh]][[

Qh

]]

(C23)

The identities vh[[

vh]]

= 12

[[

v2h]]

and[[

Qh +radicavh]]

=[[

Qh

]]

+radica[[

vh]]

then simplify(C23) to

(∆Lh minus∆R

h )∣

∣

Ik+12 = minusτv2

[[

vh]]2 minus

radica(τu + τQ)

2

[[

Qh

]][[

vh]]

minus τu2

[[

Qh

]]2 (C24)

The role of a penalty parameter is now clear Positive values of τv penalize jumps in vhthrough a negative contribution to the energy Likewise positive values of τu penalizejumps in Qh through a negative contribution to the energy However because the sign of[[

Qh

]][[

vh]]

can be positive or negative only the choice τQ = minusτu yields an expression for(∆L

h minus ∆Rh )|Ik+12 which is manifestly negative for τu ge 0 and τv ge 0 A simple estimate

based on Youngrsquos inequality with ε (that is 2αβ le εminus1α2 + εβ2 where α β ge 0 and ε gt 0)shows that for τQ = 0 the choice τv ge aτu4 also yields stability

Figure 6 depicts certain choices of stable penalty parameters for the linear model systemevolved to tfinal = 1000 (with a = 2 N = 10 and ∆t ≃ 00553) as determined empiricallywith simulations similar to those described in Sec IVA The left plot corresponds to asmall τv = 10minus6 for which the choice τu = 1 τQ = 0 is not stable as expected from thetheoretical analysis However the right plot corresponds to τv = 1 +

radica for which τu = 1

τQ = 0 is stable Motivated by the numerical flux choices (3842) used for the BSSN system

30

(4) we have (as mentioned above) set τu = 1 τv = 1 +radica and τQ = 0 in simulations of

the nonlinear model (53) For the nonlinear model system (53) the theoretically motivatedchoice τQ = minusτu also yields numerically stable evolutions when τu ge 0 and τv ge 0

For the nonlinear systems (4) and (53) we do not attempt a formal stability proofNevertheless the results of this appendix have served as a guide for our choices of penaltyparameters For the BSSN system (4) u v and Q are block indices [cf Eq (6)] Similarto the model problem we have penalized Q with τu = 1 with τv chosen large enough toheuristically overcome the cross-terms of indefinite size that arise from τQ = 0 (we interpretequations like τu = 1 componentwise) An analogous choice ldquoτQ = minusτurdquo for the BSSNsystem might be possible but would be considerably more complicated Indeed such achoice likely entails a matrix of penalty parameters but we do not give the details here

[1] F Pretorius Evolution of Binary Black-Hole Spacetimes Phys Rev Lett 95 121101 (2005)

(4 pages)

[2] F Pretorius Simulation of binary black hole spacetimes with a harmonic evolution scheme

Class Quantum Grav 23 S529-S552 (2006)

[3] M Campanelli C O Lousto P Marronetti and Y Zlochower Accurate Evolutions of Or-

biting Black-Hole Binaries without Excision Phys Rev Lett 96 111101 (2006) (4 pages)

[4] J G Baker J Centrella D-I Choi M Koppitz and J van Meter Gravitational-Wave

Extraction from an Inspiraling Configuration of Merging Black Holes Phys Rev Lett 96

111102 (2006) (4 pages)

[5] M Campanelli C O Lousto and Y Zlochower Last orbit of binary black holes Phys Rev D

73 061501(R) (2006) (5 pages)

[6] F Herrmann I Hinder D Shoemaker and P Laguna Unequal mass binary black hole plunges

and gravitational recoil Class Quantum Grav 24 S33-S42 (2007)

[7] P Diener F Herrmann D Pollney E Schnetter E Seidel R Takahashi J Thornburg and

J Ventrella Accurate Evolution of Orbiting Binary Black Holes Phys Rev Lett 96 121101

(2006) (4 pages)

[8] M A Scheel H P Pfeiffer L Lindblom L E Kidder O Rinne and S A Teukolsky Solving

Einsteinrsquos Equations With Dual Coordinate Frames Phys Rev D 74 104006 (2006)

[9] U Sperhake Binary black-hole evolutions of excision and puncture data Phys Rev D 76

104015 (2007) (20 pages)

[10] B Brugmann J A Gonzalez M Hannam S Husa U Sperhake and W Tichy Calibration

of moving puncture simulations Phys Rev D 77 024027 (2008) (25 pages)

[11] P Marronetti W Tichy B Brugmann J Gonzalez M Hannam S Husa and U Sperhake

Binary black holes on a budget simulations using workstations Class Quantum Grav 24

S43-S58 (2007)

[12] Z B Etienne J A Faber Y T Liu S L Shapiro and T W Baumgarte Filling the holes

Evolving excised binary black hole initial data with puncture techniques Phys Rev D 76

101503(R) (2007) (5 pages)

[13] B Szilagyi D Pollney L Rezzolla J Thornburg and J Winicour An explicit harmonic code

for black-hole evolution using excision Class Quantum Grav 24 S275-S293 (2007)

[14] L Boyle M Kesden and S Nissanke Binary-Black-Hole Merger Symmetry and the Spin

Expansion Phys Rev Lett 100 151101 (2008) (4 pages)

31

[15] M A Scheel M Boyle T Chu L E Kidder K D Matthews and H P Pfeiffer High-

accuracy waveforms for binary black hole inspiral merger and ringdown Phys Rev D 79

024003 (2009) (14 pages)

[16] M Hannam Status of black-hole-binary simulations for gravitational-wave detection

Class Quantum Grav 26 114001 (2009) (21 pages)

[17] I Hinder The current status of binary black hole simulations in numerical relativity

Class Quantum Grav 27 114004 (2010) (19 pages)

[18] B P Abbott et al LIGO The Laser Interferometer Gravitational-Wave Observatory

Rep Prog Phys 72 076901 (2009) (25 pages)

[19] F Acernese et al The status of VIRGO Class Quantum Grav 23 S63-S69 (2006)

[20] F Acernese et al Virgo status Class Quantum Grav 25 184001 (2008) (9 pages)

[21] R Arnowitt S Deser and C W Misner The Dynamics of General Relativity in Gravitation

an introduction to current research edited by L Witten (Wiley New York 1962) Chapter 7

227-265 Also available as gr-qc0405109

[22] J W York Jr Kinematics and Dynamics of General Relativity in Sources of Gravitational

Radiation edited by L L Smarr (Cambridge University Press Cambridge 1979)

[23] L E Kidder M A Scheel and S A Teukolsky Extending the lifetime of 3D black hole

computations with a new hyperbolic system of evolution equations Phys Rev D 64 064017

(2001) (13 pages)

[24] G Calabrese J Pullin O Sarbach and M Tiglio Convergence and stability in numerical

relativity Phys Rev D 66 041501(R) (2002) (4 pages)

[25] S Frittelli and O Reula On the Newtonian Limit of General Relativity Commun Math

Phys 166 221-235 (1994)

[26] M Shibata and T Nakamura Evolution of three-dimensional gravitational waves Harmonic

slicing case Phys Rev D 52 5428-5444 (1995)

[27] Y Choquet-Bruhat and J W York Geometrical well posed systems for the Einstein equations

C R Acad Sci Paris t 321 Serie I 1089-1095 (1995)

[28] A Abrahams A Anderson Y Choquet-Bruhat and J W York Jr Einstein and Yang-Mills

Theories in Hyperbolic Form without Gauge Fixing Phys Rev Lett 75 3377-3381 (1995)

[29] C Bona J Masso E Seidel and J Stela New Formalism for Numerical Relativity

Phys Rev Lett 75 600-603 (1995)

[30] M H P M van Putten and D M Eardley Nonlinear wave equations for relativity

Phys Rev D 53 3056-3063 (1996)

[31] S Frittelli and O A Reula First-Order Symmetric Hyperbolic Einstein Equations with Ar-

bitrary Fixed Gauge Phys Rev Lett 76 4667-4670 (1996)

[32] H Friedrich Hyperbolic reductions for Einsteinrsquos equations Class Quantum Grav 13 1451-

1469 (1996)

[33] F B Estabrook R S Robinson H D Wahlquist Hyperbolic equations for vacuum gravity

using special orthonormal frames Class Quantum Grav 14 1237-1247 (1997)

[34] M S Iriondo E O Leguizamon and O A Reula Einsteinrsquos Equations in Ashtekarrsquos Vari-

ables Constitute a Symmetric Hyperbolic System Phys Rev Lett 79 4732-4735 (1997)

[35] A Anderson Y Choquet-Bruhat and J W York Einstein-Bianchi hyperbolic system for

general relativity Topol Meth Nonlinear Anal 10 353-373 (1997)

[36] T W Baumgarte and S L Shapiro Numerical integration of Einsteinrsquos field equations

Phys Rev D 59 024007 (1998) (7 pages)

[37] M A G Bonilla Symmetric hyperbolic systems for Bianchi equations Class Quantum

18

B Simulations of the BSSN system

This subsection documents results for simulations of the unit-mass-parameter (M = 1)Schwarzschild solution (B9) expressed in terms of ingoing Kerr-Schild coordinates Sincethe solution is stationary temporal integration of the semi-discrete scheme has been carriedout with the forward Euler method The r-coordinate domain [04 34] has been split into3 equally spaced subdomains and we have set η = 10 λ = 01 and C = 2 [cf Eq (41)]For all simulations ∆t has been chosen for stability With the chosen λ the inner physicalboundary rmin = 04 is an excision surface At each timestep we have applied an (order2s = 20) exponential filter on the top two-thirds of the modal coefficient set for all fieldsexcept for grr and gθθ For stability we have empirically observed that grr and gθθ must not

be filtered A detailed understanding of this is still lackingIssues related to physical boundary conditions are similar to the one encountered in

Sec IVA for the model problem Similar to before we have retained Eqs (3842) as thechoice of numerical flux even at the endpoints Therefore at an endpoint the specificationof the boundary condition amounts to the choice W+ of external state We have typicallychosen the inner boundary of the radial domain as an excision boundary and in this caseW+ = Wminus is enforced at the inner physical boundary At the outer physical boundaryfor W+ we have again considered two choices (i) Wexact and (ii) Wupwind To enforcechoice (ii) the inverse transformation (17) must be used with incoming characteristic fieldsfixed to their exact values similar to (60) We have tried various versions of choice (ii)and in all cases the resulting simulations have been unstable We therefore present resultscorresponding to choice (i) Although the choice of an analytical external state Wexact

at the outer boundary is stable for our problem such a boundary condition is unlikelyto generalize to more complicated scenarios involving dynamical fields Indeed the issueof outer boundary conditions for the BSSN system is an active area of research with aproper treatment requiring fixation of incoming radiation control of the constraints andspecification of gauge (see Ref [86] for a recent analysis)

For BSSN simulations our main diagnostic is to monitor the Hamiltonian momentumand conformal connection constraints Figure 3 depicts long-time histories of constraintviolations whereas Figs 4 and 5 depict long-time error histories for the individual BSSNfield components From the middle plot in Fig 5 we infer that up to the indicated numericalerror the factor g sin4 θ = grr(gθθ)

2 remains at its initial fixed profile r4 throughout theevolution These figures indicate that the proposed scheme is stable for long times andexhibits spectral converge with increased polynomial order N Similar results are recoveredfromM = 0 Minkowski initial data The stability documented in these plots does not appearto rely on inordinate parameter tuning For example with the fixed parameters describedabove we obtain similar plots if we individually vary (i) rmin over 0325 035 04 0475(values still corresponding to an excision surface for the given choice of λ) (ii) η over1 3 7 10 (iii) s over 8 9 10 With the polynomial order N ranging over 23 26 29 31both stability and qualitatively similar exponential convergence is achieved with a singlesubdomain Likewise adoption of a larger coordinate domain with more subdomains doesnot impact our results Finally we have considered the addition of random noise to all fieldcomponents at the initial time Precisely with the system component χ as an example wehave set

χ(0) 7rarr χ(0) + δχ(0) (61)

where each component (nodal value) of δχ(0) is 10minus5 times a random variable drawn from a

19

10minus9

10minus7

10minus5

10minus3

Hinfin

10minus9

10minus7

10minus5

10minus3

Mrinfin

0 2000 4000 6000 8000 10000

10minus9

10minus7

10minus5

10minus3

Grinfin

t

N = 11N = 14N = 17N = 20

FIG 3 Spectral convergence of constraint violations for M = 1 Kerr-Schild ini-

tial data Respectively for N = 11 14 17 19 a timestep of ∆t ≃ 00041 00026 00018 00013

has been chosen for stability and accuracy

standard normal distribution Such perturbed initial data also gives rise to stable evolutions

V CONCLUSION

We have introduced a discontinuous Galerkin method for solving the spherically reducedBSSN system with second-order spatial operators Our scheme shares similarities withother discontinuous Galerkin methods that use local auxiliary variables to handle high-orderspatial derivatives [66 77ndash79 82 84 85] and which have typically been applied to eitherelliptic parabolic or mixed type problems The key ingredient of a stable dG scheme is anappropriate choice of numerical flux and our particular choice has been motivated by theanalysis presented in Appendix C When used to evolve the Schwarzschild solution in Kerr-Schild coordinates our numerical implementation of the BSSN system (4) is robustly stableand converges to the analytic solution exponentially with increased polynomial order Byapproximating the spatially second-order form of the BSSN system we have not introducedextra fields which are evolved Evolved auxiliary fields result in new constraints which mayspoil stability Our main goal has been stable evolution of the spherically reduced BSSNsystem as a first step towards understanding how a discontinuous Galerkin method might be

20

10minus9

10minus7

10minus5

∆Arrinfin

10minus9

10minus7

10minus5

∆Kinfin

0 2000 4000 6000 8000 10000

10minus9

10minus7

10minus5

max(∆Brinfin ∆Γrinfin)

t

N = 11N = 14N = 17N = 20

FIG 4 Spectral convergence of solution for M = 1 Kerr-Schild initial data

Timestep choices are described in the caption for Fig 3 In the title headings for example

∆Arr equiv (Arr)numer minus (Arr)exact

applied to the full BSSN system Towards that goal we now discuss treatment of singularitiesand generalization of the described dG method to higher space dimension

To deal with the fixed Schwarzschild singularity we have used excision which is easy inthe context of the spherically reduced BSSN system However excision for the binary blackhole problem in full general relativity requires attention to the technical challenge of hori-zon tracking State-of-the-art BSSN codes avoid such complication relying instead on themoving-puncture technique While the moving-puncture technique does involve mild centralsingularities it may still prove amenable to spectral methods Indeed spectral methods fornon-smooth problems is well-developed in both theory and for complex applications Sincethe moving-puncture technique can be performed in spherical symmetry [60] a first-steptoward a spectral moving-puncture code would be to implement a moving puncture withthe nodal dG method described here Such an implementation may adopt Legendre-Gauss-Radau nodes on the innermost subdomain thereby ensuring that the physical singularitydoes not lie on a nodal point (in much the same way finite difference codes use a staggeredgrid) Beyond traditional excision and moving punctures one might also retain a smoothproblem for which spectral methods are especially well-suited via the turducken approach[87 88] to singularities

21

10minus11

10minus9

10minus7

10minus5

∆χinfin

10minus11

10minus9

10minus7

10minus5

max(∆grrinfin ∆gθθinfin)

0 2000 4000 6000 8000 10000

10minus11

10minus9

10minus7

10minus5

max(∆αinfin ∆βrinfin)

t

N = 11N = 14N = 17N = 20

FIG 5 Spectral convergence of solution violations for M = 1 Kerr-Schild initial

data See the caption of Fig 4 for details

Discontinuous Galerkin methods for hyperbolic problems in two and three space dimen-sions are well-developed A generalization of the method described here to three-dimensionsand the full BSSN system would likely rely on an unstructured mesh Appropriate local poly-nomial expansions for the subelements are well-understood as are choices for the numericalfluxes which would now live on two-dimensional faces rather than single points Whether ornot it would ultimately prove successful generalization of our dG method to a higher dimen-sion would rely on an established conceptual framework Further computational advancesof relevance to a generalization of our dG method to the full BSSN system (possibly includ-ing matter) may include mesh hp-adaptivity local timestepping shock capturing and slopelimiting techniques [66] Moreover recent work [89] indicates that enhanced performancewould be expected were our scheme implemented on graphics processor units

VI ACKNOWLEDGMENTS

We thank Nick Taylor for discussions about treating second-order operators with spectralmethods Benjamin Stamm for helping polish up a few parts of Appendix C Khosro Shah-bazi for discussions on LDG and IP methods We also acknowledge helpful conversationswith David Brown and Manuel Tiglio about the BSSN system and previous work on its nu-

22

merical implementation We gratefully acknowledge funding through grants DMS 0554377and DARPAAFOSR FA9550-05-1-0108 to Brown University and NSF grant PHY 0855678to the University of New Mexico

Appendix A Hyperbolicity of the first-order system

This appendix analyzes the matrix A(u) appearing in (10) in order to construct thecharacteristic fields (14) In matrix form the sector (8) of the principal part of (10) reads asfollows

partt

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

=

βr 0 minus 4λα3grr

0 0 λβr

6(grr)2λβr

3gθθgrr0 4λ

3grr

0 βr 0 23grrαχ

13α minus αχ

3grr

αχ3gθθ

minus23χ 0

0 0 βr 0 0 0 0 minus χgrr

0

0 0 minus 4α3grr

βr 0 βr

6(grr)2βr

3gθθgrr0 4

3grr

0 0 23αχ 0 βr minus βrχ

3grrminus2βrχ

3gθθ0 minus2

3χ

0 minus2α 0 0 0 23βr minus2grrβr

3gθθ0 4

3grr

0 gθθαgrr

0 0 0 minusgθθβr

3grr13βr 0 minus2

3gθθ

0 0 minus2α 0 0 0 0 βr 0

34

0 0 0 0 0 0 0 βr

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

prime

(A1)which defines the matrix A(u) appearing in (9) and so also the matrix A(u) in (10) Notethat in the last equation the matrix within the square brackets is minusA(u) For certainconfigurations of u and λ the system (10) is strongly hyperbolic [76] that is A(u) hasa complete set of eigenvectors and real eigenvalues Indeed five eigenpairs of A(u) aretrivially recovered upon inspection of A(u)rsquos leading 5times5 diagonal block These correspondto eigenvalue 0 and the left eigenspace ξj = eTj 1 le j le 5 where ej are the canonical

basis vectors Since each component of u arises as eTj W each is also a characteristic field

The remaining nine eigenpairs are determined by A(u) The eigenvalues of A(u) are

micro1 = 0 micro23 = minusβr microplusmn4 = minusβr plusmn

radic

2αχ

grr microplusmn

5 = minusβr plusmn α

radic

χ

grr microplusmn

6 = minusβr plusmnradic

λ

grr

(A2)

23

and the corresponding left eigenvectors are

x1 = (0 0 0 0 0 gθθ 2grr 0 0) (A3a)

x2 =

(

0 0 0 grr2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3b)

x3 =

(

grrλ 0 0 0

2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3c)

xplusmn4 =

(

0 0plusmnradic

2αgrrχ

0 0 0 0 1 0

)

(A3d)

xplusmn5 =

(

0∓ 3radicgrrχ

plusmn2

radic

grrχ 2grr

1

χminus 1

grr1

gθθ 0 0

)

(A3e)

xplusmn6 =

(

minus3

4

grrλ 0plusmn α

radicλgrr

(2αχminus λ) 0 0minus βr

8(βrgrr ∓radicλgrr)

minus βrgrr

4gθθ(βrgrr ∓radicλgrr)

αχ

(2αχminus λ)plusmnradic

grrλ

)

(A3f)

where for example xplusmn5 A(u) = microplusmn5 x

plusmn5 Assuming that grr gθθ χ and α are everywhere strictly

positive the eigenvalues are real and the eigenvectors are linearly independent providedthat (15) holds These eigenvectors are easily extended to eigenvectors of A(u) e g asxplusmn6 rarr (01times5 x

plusmn6 ) Then for example the characteristic field

Xplusmn6 equiv (01times5 x

plusmn6 )W = xplusmn6WvQ (A4)

and similarly Xplusmnj = xplusmnj WvQ for j = 4 5 and Xk = xkWvQ for k = 1 2 3 The characteristic

speeds for these fields are microk and microplusmnj With this convention the speeds listed in Table I

correspond to the Xk and Xplusmnj in (14)

Appendix B Schwarzschild solution in conformal Kerr-Schild coordinates

In Kerr-Schild coordinates here the system directly related to incoming Eddington-Finkelstein null coordinates the line element for the Schwarzschild solution reads

ds2 = minusα2dt2 + (1 + 2MR)(dR + βRdt)2 +R2dθ2 +R2 sin2 θdφ2 (B1)

where R is the area radius α = (1 + 2MR)minus12 is the lapse and βR = 2M(R + 2M) isthe shift vector The physical spatial metric gab is the spatial part of this line element

To define the corresponding solution to the BSSN system we use equation gab = χgab todefine the following relationship between line elements

dr2 + r2(dθ2 + sin2 θdφ2) = χ[(1 + 2MR)dR2 +R2dθ2 +R2 sin2 θdφ2] (B2)

so that

χ

(

1 +2M

R

)(

dR

dr

)2

= 1 χR2 = r2 (B3)

Then we have(

1 +2M

R

)12dR

R=dr

r (B4)

24

with integration yielding

r =R

4

(

1 +

radic

1 +2M

R

)2

e2minus2radic

1+2MR (B5)

where the constant of integration has been chosen so that the R r rarr infin limits are consistentThe second relation in (B3) shows that

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR =rArr χminus4 =2eradic

1+2MRminus1

1 +radic

1 + 2MR (B6)

The extrinsic curvature tensor is specified by the expression forK given in (B9h) the identityK = KR

R + 2Kθθ and

Kθθ =

(

1 +2M

R

)minus122M

R2 (B7)

Since KRR = Kr

r we compute that

Krr = K minus 2Kθ

θ = minus(

1 +2M

R

)minus12(R +M

R + 2M

)

2M

R2 (B8)

Next since Krr = grrKrr = χminus1Kr

r we have Krr = Arr+

13grrK This implies Arr = Kr

r minus 13K

from which we get (B9g) In all we have

α =

(

1 +2M

R

)minus12

(B9a)

βr = βR dr

dR= χ12

(

1 +2M

R

)minus122M

R(B9b)

grr = 1 (B9c)

gθθ = r2 = χR2 (B9d)

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR (B9e)

Br = 0 (B9f)

Arr = minus(

1 +2M

R

)minus124M

3R2

(

2R + 3M

R + 2M

)

(B9g)

K =

(

1 +2M

R

)minus32(

1 +3M

R

)

2M

R2(B9h)

Γr = minus2

r= minus 2

χ12R (B9i)

To differentiate these expressions with respect to r we use the identity

dR

dr= χminus12

(

1 +2M

R

)minus12

(B10)

along with the chain rule

25

Appendix C Stability of the model system

The following stability analysis for the model system (53) has been inspired by [78 79]After dropping all nonlinear source terms the system (53) becomes

parttu = uprime + av (C1a)

parttv = uprimeprime + vprime (C1b)

This section analyzes the stability of (C1) considering both the continuum system itself aswell as its semi-discrete dG approximation The latter analysis offers some insight into theempirically observed stability of our dG scheme for the spherically reduced BSSN equations

1 Analysis for a single interval

Throughout we work with the L2 inner product and norm

(f g)D =

int

D

fg fD =radic

(f f)D (C2)

where D is the spatial coordinate interval (here D may represent a subdomain Dk or the

whole domain Ω) and we have suppressed all integration measures For the continuummodel we will establish the following estimate

uprime(T middot)2D+ av(T middot)2

Dle C(T )

(

uprime(0 middot)2D+ av(0 middot)2

D

)

(C3)

where the time-dependent constant C(T ) is determined solely by the choice of boundaryconditions To show (C3) we first change variables with v =

radicav thereby rewriting (C1)

in the following symmetric form

parttu = uprime +radicav (C4a)

parttv =radicauprimeprime + vprime (C4b)

Equations (C4ab) then imply

1

2partt

int

D

(uprime)2 =

int

D

uprime(uprimeprime +radicavprime) =

int

D

radicauprimevprime +

1

2

int

partD

(uprime)2 (C5a)

1

2partt

int

D

(v)2 =

int

D

v(radicauprimeprime + vprime) = minus

int

D

radicauprimevprime +

1

2

int

partD

(v2 + 2radicauprimev) (C5b)

Here vvprime and uprimeuprimeprime have been expressed as exact derivatives and then integrated to boundaryterms the second equation employs an extra integration by parts and with only one spacedimension

int

partDdenotes a difference of endpoint evaluations Addition of Eqs (C5ab) gives

1

2partt

int

D

[

v2 + (uprime)2]

=1

2

int

partD

[

v2 + (uprime)2 + 2radicauprimev

]

(C6)

Substitutions with the identities

[

v2 + (uprime)2]

=1

2

[

(v + uprime)2 + (v minus uprime)2]

2uprimev =1

2

[

(v + uprime)2 minus (v minus uprime)2]

(C7)

26

and replacements to recover the original variable v = vradica yield

1

2partt

int

D

[

av2 + (uprime)2]

=1 +

radica

4

int

partD

(radicav + uprime)2 +

1minusradica

4

int

partD

(radicav minus uprime)2 (C8)

From (C8) we deduce that the time-dependent constant C(T ) in (C3) must satisfy

∣

∣

∣

∣

∣

∣

1 +

int T

0

[

1+radica

2

int

partD(radicav + uprime)2 + 1minus

radica

2

int

partD(radicav minus uprime)2

]

dt

uprime(0 middot)2 + av(0 middot)2

∣

∣

∣

∣

∣

∣

le C(T ) (C9)

For periodic boundary conditions we may choose C(T ) = 1 Moreover if a ge 1 anduprime = minusradic

av is specified at partD+ then uprime(t middot)2 + av(t middot)2 decaysStill working on a single interval (subdomain) we now consider the semi-discrete scheme

for (C4) i e (57) with all nonlinear source terms dropped and with v replaced by vradica

Derivation of a formula analogous to (C8) is our first step toward establishing L2 stabilityof the semi-discrete scheme While (57) features vectors for example u(t) taking values atthe Legendre-Gauss-Lobatto nodal points here we work with the numerical solution as apolynomial for example uh(t x) These two representations are related by the Lagrange in-terpolating polynomials for the nodal set here taken to span both the space of test functionsand the space of basis functions Our scheme is

int

Dk

ψparttuh =

int

Dk

ψ(Qh +radicavh) (C10a)

int

Dk

ξparttvh = minusint

Dk

ξprime(radicaQh + vh) +

int

partDk

ξ(radicaQlowast + vlowast) (C10b)

int

Dk

ϕQh =

int

Dk

ϕuprimeh +

int

partDk

ϕ (ulowast minus uh) (C10c)

where ψ ξ and ϕ are polynomial test functions These test functions are arbitrary exceptthat they must be degree-N polynomials In (C10) the variables uh vh and Qh should alsocarry a superscript k but we have suppressed this Derivation of a formula analogous to(C8) is complicated by the fact that Qh is not evolved Nevertheless at a given instant t wecan assemble Qh from (C10c)

Mimicking the calculation (C5b) from the continuum case we first use (C10b) with ξ = vhto write

1

2partt

int

Dk

v2h = minusint

Dk

(radicaQh + vh)v

primeh +

int

partDk

(radicaQlowast + vlowast)vh

= minusint

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2(radicaQlowast + vlowast)vh minus v2h

]

(C11)

The right-hand side of (C5a) is analogous to

1

2partt

int

Dk

Q2h =

int

Dk

QhparttQh (C12)

However since Qh is not evolved the term parttQh must be given a suitable interpretationOn the right hand side of (C10c) only uh u

primeh and u

lowast necessarily depend on time since the

27

test function ϕ need not be time-dependent Furthermore ulowast is explicitly given as a linearcombination of uh as seen in Eq (C21c) below Choosing ϕ = ℓj taking the time derivativeof (C10c) and appealing to the commutivity of mixed partial derivatives we therefore arriveat

int

Dk

ℓjparttQh =

int

Dk

ℓj(parttuh)prime +

int

partDk

ℓj(

(parttu)lowast minus parttuh

)

(C13)

where (parttu)lowast depends on parttuh in precisely the same way that ulowast depends on uh We have

written ℓj rather than ϕ in the last equation to emphasize that the result also holds for anylinear combination of ℓj (for example ϕ) and even for time-dependent combinations SinceQh is itself such a combination we obtain

1

2partt

int

Dk

Q2h =

int

Dk

Qh(parttuh)prime +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

Qh(Qprimeh +

radicavprimeh) +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2((parttu)lowast minus parttuh)Qh +Q2

h

]

(C14)

Addition of (C11) and (C14) gives

1

2partt

int

Dk

(Q2h + v2h) =

1

2

int

partDk

[

Q2h minus v2h + 2(

radicaQlowast + vlowast)vh + 2((parttu)

lowast minus parttuh)Qh

]

(C15)

the aforementioned analog of (C8) This formula holds on a single subdomain Dk and we

now combine multiple copies of it one for each value of k

2 Analysis for multiple intervals

To facilitate combination of (C15) over all k we change notation At every subdomaininterface Ik+12 equiv partDk cap partDk+1 let the superscripts L and R denote field values respectivelytaken from the left and right Then the fields evaluated at Ik+12 which belong to D

k will beuLk+12 v

Lk+12 and Q

Lk+12 while those belonging to D

k+1 will be uRk+12 vRk+12 and Q

Rk+12

However at Ikminus12 the values taken from D

k are uRkminus12 vRkminus12 and QR

kminus12 Note that we

have also replaced the subscript h denoting a numerical solution with k plusmn 12 denotingthe location of the endpoint value of the numerical solution With this notation we define

∆Lα =

1

2

[

(QLα)

2 minus (vLα)2]

+(radic

aQlowastα + vlowastα

)

vLα +[

(parttuα)lowast minus parttu

Lα

]

QLα (C16)

and similarly for ∆Rα The same numerical fluxes appear in both ∆L

α and ∆Rα (ie each

numerical flux takes the same value on either side of an interface) whence fluxes like Qlowastα do

not carry an L or R superscript In terms of these definitions (C15) becomes

1

2partt

int

Dk

(Q2h + v2h) = ∆L

k+12 minus∆Rkminus12 (C17)

28

Summation over all Dk yields

1

2partt

kmaxsum

k=1

int

Dk

(Q2h + v2h) =

kmaxminus1sum

k=1

(

∆Lk+12 minus∆R

k+12

)

+∆Lkmax+12 minus∆R

12

=

kmaxminus1sum

k=1

(

∆Lh minus∆R

h

)∣

∣

Ik+12 +∆Lkmax+12 minus∆R

12 (C18)

We have reverted to h-notation denoting the numerical solution since the LR superscriptsindicate unambiguously the relevant domain used for evaluation at Ik+12

We again seek an estimate of the form

kmaxsum

k=1

(

Qh(T middot)2Dk + avh(T middot)2Dk

)

le C(T )kmaxsum

k=1

(

Qh(0 middot)2Dk + avh(0 middot)2Dk

)

(C19)

that is essentially the same as the one (C3) considered in the continuum case Assume thatthe chosen boundary conditions ensure ∆L

kmax+12 minus ∆R12 is bounded by a time-dependent

constant which does not depend on the numerical parameters N and h (subdomain width)Establishment of stability then amounts to showing that the remaining sum over interfaceterms in (C18) is non-positive whence this remaining sum is consistent with C(T ) le 1although the boundary conditions may give rise to a different bound In fact we will choosethe numerical fluxes such that each individual interface term is non-positive At interfaceIk+12 and in LR notation the jump and average of vh for example are

1

2

(

v+ + vminus)

equiv vh =1

2

(

vLk+12 + vRk+12

)

(C20a)

nminusvminus + n+v+ equiv[[

vh]]

= vLk+12 minus vRk+12 (C20b)

Consider numerical fluxes of the form

Qlowast = Qh minusτQ2

[[

Qh

]]

(C21a)

vlowast = vh minusτv2

[[

vh]]

(C21b)

ulowast = uh minusτu2

[[

uh]]

(C21c)

(parttu)lowast = parttuh minus

τu2

[[

parttuh]]

(C21d)

where (C21c) induces (C21d) and where the penalty parameters τu τv and τQ are realnumbers The fluxes defined in (58) correspond to τu = 1 τv = 1 +

radica and τQ = 0 In

terms of these quantities the kth interface contribution in (C18) is

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

parttuh]]

minus τu2

[[

parttuh]][[

Qh

]]

(C22)

where we have suppressed the k dependence of the right-hand side Now consider the term[[

parttuh]]

Because parttuh and Qh+radicavh are both polynomials of degree N Eq (C10a) implies

29

τu

τ Q

minus1 0 1minus1

minus05

0

05

1

τu

minus1 0 1minus1

minus05

0

05

1

FIG 6 Stable evolutions for the model system For fixed τv = 10minus6 and τv = 1 +radic2

respectively the left and right plots depict stable choices (determined empirically) of τu and τQfor the linear model system (C1) The stable regions are colored black but the jagged edges result

from the discretization of the (τu τQ)-plane

the vector equation parttu = Q +radicav that is pointwise equivalence on the nodal points of

Dk which in turn implies

[[

parttuh]]

=[[

Qh+radicavh]]

Upon substituting this identity into thelast equation we arrive at an expression which features only vh and Qh

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

Qh +radicavh]]

minus τu2

[[

Qh +radicavh]][[

Qh

]]

(C23)

The identities vh[[

vh]]

= 12

[[

v2h]]

and[[

Qh +radicavh]]

=[[

Qh

]]

+radica[[

vh]]

then simplify(C23) to

(∆Lh minus∆R

h )∣

∣

Ik+12 = minusτv2

[[

vh]]2 minus

radica(τu + τQ)

2

[[

Qh

]][[

vh]]

minus τu2

[[

Qh

]]2 (C24)

The role of a penalty parameter is now clear Positive values of τv penalize jumps in vhthrough a negative contribution to the energy Likewise positive values of τu penalizejumps in Qh through a negative contribution to the energy However because the sign of[[

Qh

]][[

vh]]

can be positive or negative only the choice τQ = minusτu yields an expression for(∆L

h minus ∆Rh )|Ik+12 which is manifestly negative for τu ge 0 and τv ge 0 A simple estimate

based on Youngrsquos inequality with ε (that is 2αβ le εminus1α2 + εβ2 where α β ge 0 and ε gt 0)shows that for τQ = 0 the choice τv ge aτu4 also yields stability

Figure 6 depicts certain choices of stable penalty parameters for the linear model systemevolved to tfinal = 1000 (with a = 2 N = 10 and ∆t ≃ 00553) as determined empiricallywith simulations similar to those described in Sec IVA The left plot corresponds to asmall τv = 10minus6 for which the choice τu = 1 τQ = 0 is not stable as expected from thetheoretical analysis However the right plot corresponds to τv = 1 +

radica for which τu = 1

τQ = 0 is stable Motivated by the numerical flux choices (3842) used for the BSSN system

30

(4) we have (as mentioned above) set τu = 1 τv = 1 +radica and τQ = 0 in simulations of

the nonlinear model (53) For the nonlinear model system (53) the theoretically motivatedchoice τQ = minusτu also yields numerically stable evolutions when τu ge 0 and τv ge 0

For the nonlinear systems (4) and (53) we do not attempt a formal stability proofNevertheless the results of this appendix have served as a guide for our choices of penaltyparameters For the BSSN system (4) u v and Q are block indices [cf Eq (6)] Similarto the model problem we have penalized Q with τu = 1 with τv chosen large enough toheuristically overcome the cross-terms of indefinite size that arise from τQ = 0 (we interpretequations like τu = 1 componentwise) An analogous choice ldquoτQ = minusτurdquo for the BSSNsystem might be possible but would be considerably more complicated Indeed such achoice likely entails a matrix of penalty parameters but we do not give the details here

[1] F Pretorius Evolution of Binary Black-Hole Spacetimes Phys Rev Lett 95 121101 (2005)

(4 pages)

[2] F Pretorius Simulation of binary black hole spacetimes with a harmonic evolution scheme

Class Quantum Grav 23 S529-S552 (2006)

[3] M Campanelli C O Lousto P Marronetti and Y Zlochower Accurate Evolutions of Or-

biting Black-Hole Binaries without Excision Phys Rev Lett 96 111101 (2006) (4 pages)

[4] J G Baker J Centrella D-I Choi M Koppitz and J van Meter Gravitational-Wave

Extraction from an Inspiraling Configuration of Merging Black Holes Phys Rev Lett 96

111102 (2006) (4 pages)

[5] M Campanelli C O Lousto and Y Zlochower Last orbit of binary black holes Phys Rev D

73 061501(R) (2006) (5 pages)

[6] F Herrmann I Hinder D Shoemaker and P Laguna Unequal mass binary black hole plunges

and gravitational recoil Class Quantum Grav 24 S33-S42 (2007)

[7] P Diener F Herrmann D Pollney E Schnetter E Seidel R Takahashi J Thornburg and

J Ventrella Accurate Evolution of Orbiting Binary Black Holes Phys Rev Lett 96 121101

(2006) (4 pages)

[8] M A Scheel H P Pfeiffer L Lindblom L E Kidder O Rinne and S A Teukolsky Solving

Einsteinrsquos Equations With Dual Coordinate Frames Phys Rev D 74 104006 (2006)

[9] U Sperhake Binary black-hole evolutions of excision and puncture data Phys Rev D 76

104015 (2007) (20 pages)

[10] B Brugmann J A Gonzalez M Hannam S Husa U Sperhake and W Tichy Calibration

of moving puncture simulations Phys Rev D 77 024027 (2008) (25 pages)

[11] P Marronetti W Tichy B Brugmann J Gonzalez M Hannam S Husa and U Sperhake

Binary black holes on a budget simulations using workstations Class Quantum Grav 24

S43-S58 (2007)

[12] Z B Etienne J A Faber Y T Liu S L Shapiro and T W Baumgarte Filling the holes

Evolving excised binary black hole initial data with puncture techniques Phys Rev D 76

101503(R) (2007) (5 pages)

[13] B Szilagyi D Pollney L Rezzolla J Thornburg and J Winicour An explicit harmonic code

for black-hole evolution using excision Class Quantum Grav 24 S275-S293 (2007)

[14] L Boyle M Kesden and S Nissanke Binary-Black-Hole Merger Symmetry and the Spin

Expansion Phys Rev Lett 100 151101 (2008) (4 pages)

31

[15] M A Scheel M Boyle T Chu L E Kidder K D Matthews and H P Pfeiffer High-

accuracy waveforms for binary black hole inspiral merger and ringdown Phys Rev D 79

024003 (2009) (14 pages)

[16] M Hannam Status of black-hole-binary simulations for gravitational-wave detection

Class Quantum Grav 26 114001 (2009) (21 pages)

[17] I Hinder The current status of binary black hole simulations in numerical relativity

Class Quantum Grav 27 114004 (2010) (19 pages)

[18] B P Abbott et al LIGO The Laser Interferometer Gravitational-Wave Observatory

Rep Prog Phys 72 076901 (2009) (25 pages)

[19] F Acernese et al The status of VIRGO Class Quantum Grav 23 S63-S69 (2006)

[20] F Acernese et al Virgo status Class Quantum Grav 25 184001 (2008) (9 pages)

[21] R Arnowitt S Deser and C W Misner The Dynamics of General Relativity in Gravitation

an introduction to current research edited by L Witten (Wiley New York 1962) Chapter 7

227-265 Also available as gr-qc0405109

[22] J W York Jr Kinematics and Dynamics of General Relativity in Sources of Gravitational

Radiation edited by L L Smarr (Cambridge University Press Cambridge 1979)

[23] L E Kidder M A Scheel and S A Teukolsky Extending the lifetime of 3D black hole

computations with a new hyperbolic system of evolution equations Phys Rev D 64 064017

(2001) (13 pages)

[24] G Calabrese J Pullin O Sarbach and M Tiglio Convergence and stability in numerical

relativity Phys Rev D 66 041501(R) (2002) (4 pages)

[25] S Frittelli and O Reula On the Newtonian Limit of General Relativity Commun Math

Phys 166 221-235 (1994)

[26] M Shibata and T Nakamura Evolution of three-dimensional gravitational waves Harmonic

slicing case Phys Rev D 52 5428-5444 (1995)

[27] Y Choquet-Bruhat and J W York Geometrical well posed systems for the Einstein equations

C R Acad Sci Paris t 321 Serie I 1089-1095 (1995)

[28] A Abrahams A Anderson Y Choquet-Bruhat and J W York Jr Einstein and Yang-Mills

Theories in Hyperbolic Form without Gauge Fixing Phys Rev Lett 75 3377-3381 (1995)

[29] C Bona J Masso E Seidel and J Stela New Formalism for Numerical Relativity

Phys Rev Lett 75 600-603 (1995)

[30] M H P M van Putten and D M Eardley Nonlinear wave equations for relativity

Phys Rev D 53 3056-3063 (1996)

[31] S Frittelli and O A Reula First-Order Symmetric Hyperbolic Einstein Equations with Ar-

bitrary Fixed Gauge Phys Rev Lett 76 4667-4670 (1996)

[32] H Friedrich Hyperbolic reductions for Einsteinrsquos equations Class Quantum Grav 13 1451-

1469 (1996)

[33] F B Estabrook R S Robinson H D Wahlquist Hyperbolic equations for vacuum gravity

using special orthonormal frames Class Quantum Grav 14 1237-1247 (1997)

[34] M S Iriondo E O Leguizamon and O A Reula Einsteinrsquos Equations in Ashtekarrsquos Vari-

ables Constitute a Symmetric Hyperbolic System Phys Rev Lett 79 4732-4735 (1997)

[35] A Anderson Y Choquet-Bruhat and J W York Einstein-Bianchi hyperbolic system for

general relativity Topol Meth Nonlinear Anal 10 353-373 (1997)

[36] T W Baumgarte and S L Shapiro Numerical integration of Einsteinrsquos field equations

Phys Rev D 59 024007 (1998) (7 pages)

[37] M A G Bonilla Symmetric hyperbolic systems for Bianchi equations Class Quantum

19

10minus9

10minus7

10minus5

10minus3

Hinfin

10minus9

10minus7

10minus5

10minus3

Mrinfin

0 2000 4000 6000 8000 10000

10minus9

10minus7

10minus5

10minus3

Grinfin

t

N = 11N = 14N = 17N = 20

FIG 3 Spectral convergence of constraint violations for M = 1 Kerr-Schild ini-

tial data Respectively for N = 11 14 17 19 a timestep of ∆t ≃ 00041 00026 00018 00013

has been chosen for stability and accuracy

standard normal distribution Such perturbed initial data also gives rise to stable evolutions

V CONCLUSION

We have introduced a discontinuous Galerkin method for solving the spherically reducedBSSN system with second-order spatial operators Our scheme shares similarities withother discontinuous Galerkin methods that use local auxiliary variables to handle high-orderspatial derivatives [66 77ndash79 82 84 85] and which have typically been applied to eitherelliptic parabolic or mixed type problems The key ingredient of a stable dG scheme is anappropriate choice of numerical flux and our particular choice has been motivated by theanalysis presented in Appendix C When used to evolve the Schwarzschild solution in Kerr-Schild coordinates our numerical implementation of the BSSN system (4) is robustly stableand converges to the analytic solution exponentially with increased polynomial order Byapproximating the spatially second-order form of the BSSN system we have not introducedextra fields which are evolved Evolved auxiliary fields result in new constraints which mayspoil stability Our main goal has been stable evolution of the spherically reduced BSSNsystem as a first step towards understanding how a discontinuous Galerkin method might be

20

10minus9

10minus7

10minus5

∆Arrinfin

10minus9

10minus7

10minus5

∆Kinfin

0 2000 4000 6000 8000 10000

10minus9

10minus7

10minus5

max(∆Brinfin ∆Γrinfin)

t

N = 11N = 14N = 17N = 20

FIG 4 Spectral convergence of solution for M = 1 Kerr-Schild initial data

Timestep choices are described in the caption for Fig 3 In the title headings for example

∆Arr equiv (Arr)numer minus (Arr)exact

applied to the full BSSN system Towards that goal we now discuss treatment of singularitiesand generalization of the described dG method to higher space dimension

To deal with the fixed Schwarzschild singularity we have used excision which is easy inthe context of the spherically reduced BSSN system However excision for the binary blackhole problem in full general relativity requires attention to the technical challenge of hori-zon tracking State-of-the-art BSSN codes avoid such complication relying instead on themoving-puncture technique While the moving-puncture technique does involve mild centralsingularities it may still prove amenable to spectral methods Indeed spectral methods fornon-smooth problems is well-developed in both theory and for complex applications Sincethe moving-puncture technique can be performed in spherical symmetry [60] a first-steptoward a spectral moving-puncture code would be to implement a moving puncture withthe nodal dG method described here Such an implementation may adopt Legendre-Gauss-Radau nodes on the innermost subdomain thereby ensuring that the physical singularitydoes not lie on a nodal point (in much the same way finite difference codes use a staggeredgrid) Beyond traditional excision and moving punctures one might also retain a smoothproblem for which spectral methods are especially well-suited via the turducken approach[87 88] to singularities

21

10minus11

10minus9

10minus7

10minus5

∆χinfin

10minus11

10minus9

10minus7

10minus5

max(∆grrinfin ∆gθθinfin)

0 2000 4000 6000 8000 10000

10minus11

10minus9

10minus7

10minus5

max(∆αinfin ∆βrinfin)

t

N = 11N = 14N = 17N = 20

FIG 5 Spectral convergence of solution violations for M = 1 Kerr-Schild initial

data See the caption of Fig 4 for details

Discontinuous Galerkin methods for hyperbolic problems in two and three space dimen-sions are well-developed A generalization of the method described here to three-dimensionsand the full BSSN system would likely rely on an unstructured mesh Appropriate local poly-nomial expansions for the subelements are well-understood as are choices for the numericalfluxes which would now live on two-dimensional faces rather than single points Whether ornot it would ultimately prove successful generalization of our dG method to a higher dimen-sion would rely on an established conceptual framework Further computational advancesof relevance to a generalization of our dG method to the full BSSN system (possibly includ-ing matter) may include mesh hp-adaptivity local timestepping shock capturing and slopelimiting techniques [66] Moreover recent work [89] indicates that enhanced performancewould be expected were our scheme implemented on graphics processor units

VI ACKNOWLEDGMENTS

We thank Nick Taylor for discussions about treating second-order operators with spectralmethods Benjamin Stamm for helping polish up a few parts of Appendix C Khosro Shah-bazi for discussions on LDG and IP methods We also acknowledge helpful conversationswith David Brown and Manuel Tiglio about the BSSN system and previous work on its nu-

22

merical implementation We gratefully acknowledge funding through grants DMS 0554377and DARPAAFOSR FA9550-05-1-0108 to Brown University and NSF grant PHY 0855678to the University of New Mexico

Appendix A Hyperbolicity of the first-order system

This appendix analyzes the matrix A(u) appearing in (10) in order to construct thecharacteristic fields (14) In matrix form the sector (8) of the principal part of (10) reads asfollows

partt

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

=

βr 0 minus 4λα3grr

0 0 λβr

6(grr)2λβr

3gθθgrr0 4λ

3grr

0 βr 0 23grrαχ

13α minus αχ

3grr

αχ3gθθ

minus23χ 0

0 0 βr 0 0 0 0 minus χgrr

0

0 0 minus 4α3grr

βr 0 βr

6(grr)2βr

3gθθgrr0 4

3grr

0 0 23αχ 0 βr minus βrχ

3grrminus2βrχ

3gθθ0 minus2

3χ

0 minus2α 0 0 0 23βr minus2grrβr

3gθθ0 4

3grr

0 gθθαgrr

0 0 0 minusgθθβr

3grr13βr 0 minus2

3gθθ

0 0 minus2α 0 0 0 0 βr 0

34

0 0 0 0 0 0 0 βr

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

prime

(A1)which defines the matrix A(u) appearing in (9) and so also the matrix A(u) in (10) Notethat in the last equation the matrix within the square brackets is minusA(u) For certainconfigurations of u and λ the system (10) is strongly hyperbolic [76] that is A(u) hasa complete set of eigenvectors and real eigenvalues Indeed five eigenpairs of A(u) aretrivially recovered upon inspection of A(u)rsquos leading 5times5 diagonal block These correspondto eigenvalue 0 and the left eigenspace ξj = eTj 1 le j le 5 where ej are the canonical

basis vectors Since each component of u arises as eTj W each is also a characteristic field

The remaining nine eigenpairs are determined by A(u) The eigenvalues of A(u) are

micro1 = 0 micro23 = minusβr microplusmn4 = minusβr plusmn

radic

2αχ

grr microplusmn

5 = minusβr plusmn α

radic

χ

grr microplusmn

6 = minusβr plusmnradic

λ

grr

(A2)

23

and the corresponding left eigenvectors are

x1 = (0 0 0 0 0 gθθ 2grr 0 0) (A3a)

x2 =

(

0 0 0 grr2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3b)

x3 =

(

grrλ 0 0 0

2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3c)

xplusmn4 =

(

0 0plusmnradic

2αgrrχ

0 0 0 0 1 0

)

(A3d)

xplusmn5 =

(

0∓ 3radicgrrχ

plusmn2

radic

grrχ 2grr

1

χminus 1

grr1

gθθ 0 0

)

(A3e)

xplusmn6 =

(

minus3

4

grrλ 0plusmn α

radicλgrr

(2αχminus λ) 0 0minus βr

8(βrgrr ∓radicλgrr)

minus βrgrr

4gθθ(βrgrr ∓radicλgrr)

αχ

(2αχminus λ)plusmnradic

grrλ

)

(A3f)

where for example xplusmn5 A(u) = microplusmn5 x

plusmn5 Assuming that grr gθθ χ and α are everywhere strictly

positive the eigenvalues are real and the eigenvectors are linearly independent providedthat (15) holds These eigenvectors are easily extended to eigenvectors of A(u) e g asxplusmn6 rarr (01times5 x

plusmn6 ) Then for example the characteristic field

Xplusmn6 equiv (01times5 x

plusmn6 )W = xplusmn6WvQ (A4)

and similarly Xplusmnj = xplusmnj WvQ for j = 4 5 and Xk = xkWvQ for k = 1 2 3 The characteristic

speeds for these fields are microk and microplusmnj With this convention the speeds listed in Table I

correspond to the Xk and Xplusmnj in (14)

Appendix B Schwarzschild solution in conformal Kerr-Schild coordinates

In Kerr-Schild coordinates here the system directly related to incoming Eddington-Finkelstein null coordinates the line element for the Schwarzschild solution reads

ds2 = minusα2dt2 + (1 + 2MR)(dR + βRdt)2 +R2dθ2 +R2 sin2 θdφ2 (B1)

where R is the area radius α = (1 + 2MR)minus12 is the lapse and βR = 2M(R + 2M) isthe shift vector The physical spatial metric gab is the spatial part of this line element

To define the corresponding solution to the BSSN system we use equation gab = χgab todefine the following relationship between line elements

dr2 + r2(dθ2 + sin2 θdφ2) = χ[(1 + 2MR)dR2 +R2dθ2 +R2 sin2 θdφ2] (B2)

so that

χ

(

1 +2M

R

)(

dR

dr

)2

= 1 χR2 = r2 (B3)

Then we have(

1 +2M

R

)12dR

R=dr

r (B4)

24

with integration yielding

r =R

4

(

1 +

radic

1 +2M

R

)2

e2minus2radic

1+2MR (B5)

where the constant of integration has been chosen so that the R r rarr infin limits are consistentThe second relation in (B3) shows that

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR =rArr χminus4 =2eradic

1+2MRminus1

1 +radic

1 + 2MR (B6)

The extrinsic curvature tensor is specified by the expression forK given in (B9h) the identityK = KR

R + 2Kθθ and

Kθθ =

(

1 +2M

R

)minus122M

R2 (B7)

Since KRR = Kr

r we compute that

Krr = K minus 2Kθ

θ = minus(

1 +2M

R

)minus12(R +M

R + 2M

)

2M

R2 (B8)

Next since Krr = grrKrr = χminus1Kr

r we have Krr = Arr+

13grrK This implies Arr = Kr

r minus 13K

from which we get (B9g) In all we have

α =

(

1 +2M

R

)minus12

(B9a)

βr = βR dr

dR= χ12

(

1 +2M

R

)minus122M

R(B9b)

grr = 1 (B9c)

gθθ = r2 = χR2 (B9d)

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR (B9e)

Br = 0 (B9f)

Arr = minus(

1 +2M

R

)minus124M

3R2

(

2R + 3M

R + 2M

)

(B9g)

K =

(

1 +2M

R

)minus32(

1 +3M

R

)

2M

R2(B9h)

Γr = minus2

r= minus 2

χ12R (B9i)

To differentiate these expressions with respect to r we use the identity

dR

dr= χminus12

(

1 +2M

R

)minus12

(B10)

along with the chain rule

25

Appendix C Stability of the model system

The following stability analysis for the model system (53) has been inspired by [78 79]After dropping all nonlinear source terms the system (53) becomes

parttu = uprime + av (C1a)

parttv = uprimeprime + vprime (C1b)

This section analyzes the stability of (C1) considering both the continuum system itself aswell as its semi-discrete dG approximation The latter analysis offers some insight into theempirically observed stability of our dG scheme for the spherically reduced BSSN equations

1 Analysis for a single interval

Throughout we work with the L2 inner product and norm

(f g)D =

int

D

fg fD =radic

(f f)D (C2)

where D is the spatial coordinate interval (here D may represent a subdomain Dk or the

whole domain Ω) and we have suppressed all integration measures For the continuummodel we will establish the following estimate

uprime(T middot)2D+ av(T middot)2

Dle C(T )

(

uprime(0 middot)2D+ av(0 middot)2

D

)

(C3)

where the time-dependent constant C(T ) is determined solely by the choice of boundaryconditions To show (C3) we first change variables with v =

radicav thereby rewriting (C1)

in the following symmetric form

parttu = uprime +radicav (C4a)

parttv =radicauprimeprime + vprime (C4b)

Equations (C4ab) then imply

1

2partt

int

D

(uprime)2 =

int

D

uprime(uprimeprime +radicavprime) =

int

D

radicauprimevprime +

1

2

int

partD

(uprime)2 (C5a)

1

2partt

int

D

(v)2 =

int

D

v(radicauprimeprime + vprime) = minus

int

D

radicauprimevprime +

1

2

int

partD

(v2 + 2radicauprimev) (C5b)

Here vvprime and uprimeuprimeprime have been expressed as exact derivatives and then integrated to boundaryterms the second equation employs an extra integration by parts and with only one spacedimension

int

partDdenotes a difference of endpoint evaluations Addition of Eqs (C5ab) gives

1

2partt

int

D

[

v2 + (uprime)2]

=1

2

int

partD

[

v2 + (uprime)2 + 2radicauprimev

]

(C6)

Substitutions with the identities

[

v2 + (uprime)2]

=1

2

[

(v + uprime)2 + (v minus uprime)2]

2uprimev =1

2

[

(v + uprime)2 minus (v minus uprime)2]

(C7)

26

and replacements to recover the original variable v = vradica yield

1

2partt

int

D

[

av2 + (uprime)2]

=1 +

radica

4

int

partD

(radicav + uprime)2 +

1minusradica

4

int

partD

(radicav minus uprime)2 (C8)

From (C8) we deduce that the time-dependent constant C(T ) in (C3) must satisfy

∣

∣

∣

∣

∣

∣

1 +

int T

0

[

1+radica

2

int

partD(radicav + uprime)2 + 1minus

radica

2

int

partD(radicav minus uprime)2

]

dt

uprime(0 middot)2 + av(0 middot)2

∣

∣

∣

∣

∣

∣

le C(T ) (C9)

For periodic boundary conditions we may choose C(T ) = 1 Moreover if a ge 1 anduprime = minusradic

av is specified at partD+ then uprime(t middot)2 + av(t middot)2 decaysStill working on a single interval (subdomain) we now consider the semi-discrete scheme

for (C4) i e (57) with all nonlinear source terms dropped and with v replaced by vradica

Derivation of a formula analogous to (C8) is our first step toward establishing L2 stabilityof the semi-discrete scheme While (57) features vectors for example u(t) taking values atthe Legendre-Gauss-Lobatto nodal points here we work with the numerical solution as apolynomial for example uh(t x) These two representations are related by the Lagrange in-terpolating polynomials for the nodal set here taken to span both the space of test functionsand the space of basis functions Our scheme is

int

Dk

ψparttuh =

int

Dk

ψ(Qh +radicavh) (C10a)

int

Dk

ξparttvh = minusint

Dk

ξprime(radicaQh + vh) +

int

partDk

ξ(radicaQlowast + vlowast) (C10b)

int

Dk

ϕQh =

int

Dk

ϕuprimeh +

int

partDk

ϕ (ulowast minus uh) (C10c)

where ψ ξ and ϕ are polynomial test functions These test functions are arbitrary exceptthat they must be degree-N polynomials In (C10) the variables uh vh and Qh should alsocarry a superscript k but we have suppressed this Derivation of a formula analogous to(C8) is complicated by the fact that Qh is not evolved Nevertheless at a given instant t wecan assemble Qh from (C10c)

Mimicking the calculation (C5b) from the continuum case we first use (C10b) with ξ = vhto write

1

2partt

int

Dk

v2h = minusint

Dk

(radicaQh + vh)v

primeh +

int

partDk

(radicaQlowast + vlowast)vh

= minusint

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2(radicaQlowast + vlowast)vh minus v2h

]

(C11)

The right-hand side of (C5a) is analogous to

1

2partt

int

Dk

Q2h =

int

Dk

QhparttQh (C12)

However since Qh is not evolved the term parttQh must be given a suitable interpretationOn the right hand side of (C10c) only uh u

primeh and u

lowast necessarily depend on time since the

27

test function ϕ need not be time-dependent Furthermore ulowast is explicitly given as a linearcombination of uh as seen in Eq (C21c) below Choosing ϕ = ℓj taking the time derivativeof (C10c) and appealing to the commutivity of mixed partial derivatives we therefore arriveat

int

Dk

ℓjparttQh =

int

Dk

ℓj(parttuh)prime +

int

partDk

ℓj(

(parttu)lowast minus parttuh

)

(C13)

where (parttu)lowast depends on parttuh in precisely the same way that ulowast depends on uh We have

written ℓj rather than ϕ in the last equation to emphasize that the result also holds for anylinear combination of ℓj (for example ϕ) and even for time-dependent combinations SinceQh is itself such a combination we obtain

1

2partt

int

Dk

Q2h =

int

Dk

Qh(parttuh)prime +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

Qh(Qprimeh +

radicavprimeh) +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2((parttu)lowast minus parttuh)Qh +Q2

h

]

(C14)

Addition of (C11) and (C14) gives

1

2partt

int

Dk

(Q2h + v2h) =

1

2

int

partDk

[

Q2h minus v2h + 2(

radicaQlowast + vlowast)vh + 2((parttu)

lowast minus parttuh)Qh

]

(C15)

the aforementioned analog of (C8) This formula holds on a single subdomain Dk and we

now combine multiple copies of it one for each value of k

2 Analysis for multiple intervals

To facilitate combination of (C15) over all k we change notation At every subdomaininterface Ik+12 equiv partDk cap partDk+1 let the superscripts L and R denote field values respectivelytaken from the left and right Then the fields evaluated at Ik+12 which belong to D

k will beuLk+12 v

Lk+12 and Q

Lk+12 while those belonging to D

k+1 will be uRk+12 vRk+12 and Q

Rk+12

However at Ikminus12 the values taken from D

k are uRkminus12 vRkminus12 and QR

kminus12 Note that we

have also replaced the subscript h denoting a numerical solution with k plusmn 12 denotingthe location of the endpoint value of the numerical solution With this notation we define

∆Lα =

1

2

[

(QLα)

2 minus (vLα)2]

+(radic

aQlowastα + vlowastα

)

vLα +[

(parttuα)lowast minus parttu

Lα

]

QLα (C16)

and similarly for ∆Rα The same numerical fluxes appear in both ∆L

α and ∆Rα (ie each

numerical flux takes the same value on either side of an interface) whence fluxes like Qlowastα do

not carry an L or R superscript In terms of these definitions (C15) becomes

1

2partt

int

Dk

(Q2h + v2h) = ∆L

k+12 minus∆Rkminus12 (C17)

28

Summation over all Dk yields

1

2partt

kmaxsum

k=1

int

Dk

(Q2h + v2h) =

kmaxminus1sum

k=1

(

∆Lk+12 minus∆R

k+12

)

+∆Lkmax+12 minus∆R

12

=

kmaxminus1sum

k=1

(

∆Lh minus∆R

h

)∣

∣

Ik+12 +∆Lkmax+12 minus∆R

12 (C18)

We have reverted to h-notation denoting the numerical solution since the LR superscriptsindicate unambiguously the relevant domain used for evaluation at Ik+12

We again seek an estimate of the form

kmaxsum

k=1

(

Qh(T middot)2Dk + avh(T middot)2Dk

)

le C(T )kmaxsum

k=1

(

Qh(0 middot)2Dk + avh(0 middot)2Dk

)

(C19)

that is essentially the same as the one (C3) considered in the continuum case Assume thatthe chosen boundary conditions ensure ∆L

kmax+12 minus ∆R12 is bounded by a time-dependent

constant which does not depend on the numerical parameters N and h (subdomain width)Establishment of stability then amounts to showing that the remaining sum over interfaceterms in (C18) is non-positive whence this remaining sum is consistent with C(T ) le 1although the boundary conditions may give rise to a different bound In fact we will choosethe numerical fluxes such that each individual interface term is non-positive At interfaceIk+12 and in LR notation the jump and average of vh for example are

1

2

(

v+ + vminus)

equiv vh =1

2

(

vLk+12 + vRk+12

)

(C20a)

nminusvminus + n+v+ equiv[[

vh]]

= vLk+12 minus vRk+12 (C20b)

Consider numerical fluxes of the form

Qlowast = Qh minusτQ2

[[

Qh

]]

(C21a)

vlowast = vh minusτv2

[[

vh]]

(C21b)

ulowast = uh minusτu2

[[

uh]]

(C21c)

(parttu)lowast = parttuh minus

τu2

[[

parttuh]]

(C21d)

where (C21c) induces (C21d) and where the penalty parameters τu τv and τQ are realnumbers The fluxes defined in (58) correspond to τu = 1 τv = 1 +

radica and τQ = 0 In

terms of these quantities the kth interface contribution in (C18) is

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

parttuh]]

minus τu2

[[

parttuh]][[

Qh

]]

(C22)

where we have suppressed the k dependence of the right-hand side Now consider the term[[

parttuh]]

Because parttuh and Qh+radicavh are both polynomials of degree N Eq (C10a) implies

29

τu

τ Q

minus1 0 1minus1

minus05

0

05

1

τu

minus1 0 1minus1

minus05

0

05

1

FIG 6 Stable evolutions for the model system For fixed τv = 10minus6 and τv = 1 +radic2

respectively the left and right plots depict stable choices (determined empirically) of τu and τQfor the linear model system (C1) The stable regions are colored black but the jagged edges result

from the discretization of the (τu τQ)-plane

the vector equation parttu = Q +radicav that is pointwise equivalence on the nodal points of

Dk which in turn implies

[[

parttuh]]

=[[

Qh+radicavh]]

Upon substituting this identity into thelast equation we arrive at an expression which features only vh and Qh

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

Qh +radicavh]]

minus τu2

[[

Qh +radicavh]][[

Qh

]]

(C23)

The identities vh[[

vh]]

= 12

[[

v2h]]

and[[

Qh +radicavh]]

=[[

Qh

]]

+radica[[

vh]]

then simplify(C23) to

(∆Lh minus∆R

h )∣

∣

Ik+12 = minusτv2

[[

vh]]2 minus

radica(τu + τQ)

2

[[

Qh

]][[

vh]]

minus τu2

[[

Qh

]]2 (C24)

The role of a penalty parameter is now clear Positive values of τv penalize jumps in vhthrough a negative contribution to the energy Likewise positive values of τu penalizejumps in Qh through a negative contribution to the energy However because the sign of[[

Qh

]][[

vh]]

can be positive or negative only the choice τQ = minusτu yields an expression for(∆L

h minus ∆Rh )|Ik+12 which is manifestly negative for τu ge 0 and τv ge 0 A simple estimate

based on Youngrsquos inequality with ε (that is 2αβ le εminus1α2 + εβ2 where α β ge 0 and ε gt 0)shows that for τQ = 0 the choice τv ge aτu4 also yields stability

Figure 6 depicts certain choices of stable penalty parameters for the linear model systemevolved to tfinal = 1000 (with a = 2 N = 10 and ∆t ≃ 00553) as determined empiricallywith simulations similar to those described in Sec IVA The left plot corresponds to asmall τv = 10minus6 for which the choice τu = 1 τQ = 0 is not stable as expected from thetheoretical analysis However the right plot corresponds to τv = 1 +

radica for which τu = 1

τQ = 0 is stable Motivated by the numerical flux choices (3842) used for the BSSN system

30

(4) we have (as mentioned above) set τu = 1 τv = 1 +radica and τQ = 0 in simulations of

the nonlinear model (53) For the nonlinear model system (53) the theoretically motivatedchoice τQ = minusτu also yields numerically stable evolutions when τu ge 0 and τv ge 0

For the nonlinear systems (4) and (53) we do not attempt a formal stability proofNevertheless the results of this appendix have served as a guide for our choices of penaltyparameters For the BSSN system (4) u v and Q are block indices [cf Eq (6)] Similarto the model problem we have penalized Q with τu = 1 with τv chosen large enough toheuristically overcome the cross-terms of indefinite size that arise from τQ = 0 (we interpretequations like τu = 1 componentwise) An analogous choice ldquoτQ = minusτurdquo for the BSSNsystem might be possible but would be considerably more complicated Indeed such achoice likely entails a matrix of penalty parameters but we do not give the details here

[1] F Pretorius Evolution of Binary Black-Hole Spacetimes Phys Rev Lett 95 121101 (2005)

(4 pages)

[2] F Pretorius Simulation of binary black hole spacetimes with a harmonic evolution scheme

Class Quantum Grav 23 S529-S552 (2006)

[3] M Campanelli C O Lousto P Marronetti and Y Zlochower Accurate Evolutions of Or-

biting Black-Hole Binaries without Excision Phys Rev Lett 96 111101 (2006) (4 pages)

[4] J G Baker J Centrella D-I Choi M Koppitz and J van Meter Gravitational-Wave

Extraction from an Inspiraling Configuration of Merging Black Holes Phys Rev Lett 96

111102 (2006) (4 pages)

[5] M Campanelli C O Lousto and Y Zlochower Last orbit of binary black holes Phys Rev D

73 061501(R) (2006) (5 pages)

[6] F Herrmann I Hinder D Shoemaker and P Laguna Unequal mass binary black hole plunges

and gravitational recoil Class Quantum Grav 24 S33-S42 (2007)

[7] P Diener F Herrmann D Pollney E Schnetter E Seidel R Takahashi J Thornburg and

J Ventrella Accurate Evolution of Orbiting Binary Black Holes Phys Rev Lett 96 121101

(2006) (4 pages)

[8] M A Scheel H P Pfeiffer L Lindblom L E Kidder O Rinne and S A Teukolsky Solving

Einsteinrsquos Equations With Dual Coordinate Frames Phys Rev D 74 104006 (2006)

[9] U Sperhake Binary black-hole evolutions of excision and puncture data Phys Rev D 76

104015 (2007) (20 pages)

[10] B Brugmann J A Gonzalez M Hannam S Husa U Sperhake and W Tichy Calibration

of moving puncture simulations Phys Rev D 77 024027 (2008) (25 pages)

[11] P Marronetti W Tichy B Brugmann J Gonzalez M Hannam S Husa and U Sperhake

Binary black holes on a budget simulations using workstations Class Quantum Grav 24

S43-S58 (2007)

[12] Z B Etienne J A Faber Y T Liu S L Shapiro and T W Baumgarte Filling the holes

Evolving excised binary black hole initial data with puncture techniques Phys Rev D 76

101503(R) (2007) (5 pages)

[13] B Szilagyi D Pollney L Rezzolla J Thornburg and J Winicour An explicit harmonic code

for black-hole evolution using excision Class Quantum Grav 24 S275-S293 (2007)

[14] L Boyle M Kesden and S Nissanke Binary-Black-Hole Merger Symmetry and the Spin

Expansion Phys Rev Lett 100 151101 (2008) (4 pages)

31

[15] M A Scheel M Boyle T Chu L E Kidder K D Matthews and H P Pfeiffer High-

accuracy waveforms for binary black hole inspiral merger and ringdown Phys Rev D 79

024003 (2009) (14 pages)

[16] M Hannam Status of black-hole-binary simulations for gravitational-wave detection

Class Quantum Grav 26 114001 (2009) (21 pages)

[17] I Hinder The current status of binary black hole simulations in numerical relativity

Class Quantum Grav 27 114004 (2010) (19 pages)

[18] B P Abbott et al LIGO The Laser Interferometer Gravitational-Wave Observatory

Rep Prog Phys 72 076901 (2009) (25 pages)

[19] F Acernese et al The status of VIRGO Class Quantum Grav 23 S63-S69 (2006)

[20] F Acernese et al Virgo status Class Quantum Grav 25 184001 (2008) (9 pages)

[21] R Arnowitt S Deser and C W Misner The Dynamics of General Relativity in Gravitation

an introduction to current research edited by L Witten (Wiley New York 1962) Chapter 7

227-265 Also available as gr-qc0405109

[22] J W York Jr Kinematics and Dynamics of General Relativity in Sources of Gravitational

Radiation edited by L L Smarr (Cambridge University Press Cambridge 1979)

[23] L E Kidder M A Scheel and S A Teukolsky Extending the lifetime of 3D black hole

computations with a new hyperbolic system of evolution equations Phys Rev D 64 064017

(2001) (13 pages)

[24] G Calabrese J Pullin O Sarbach and M Tiglio Convergence and stability in numerical

relativity Phys Rev D 66 041501(R) (2002) (4 pages)

[25] S Frittelli and O Reula On the Newtonian Limit of General Relativity Commun Math

Phys 166 221-235 (1994)

[26] M Shibata and T Nakamura Evolution of three-dimensional gravitational waves Harmonic

slicing case Phys Rev D 52 5428-5444 (1995)

[27] Y Choquet-Bruhat and J W York Geometrical well posed systems for the Einstein equations

C R Acad Sci Paris t 321 Serie I 1089-1095 (1995)

[28] A Abrahams A Anderson Y Choquet-Bruhat and J W York Jr Einstein and Yang-Mills

Theories in Hyperbolic Form without Gauge Fixing Phys Rev Lett 75 3377-3381 (1995)

[29] C Bona J Masso E Seidel and J Stela New Formalism for Numerical Relativity

Phys Rev Lett 75 600-603 (1995)

[30] M H P M van Putten and D M Eardley Nonlinear wave equations for relativity

Phys Rev D 53 3056-3063 (1996)

[31] S Frittelli and O A Reula First-Order Symmetric Hyperbolic Einstein Equations with Ar-

bitrary Fixed Gauge Phys Rev Lett 76 4667-4670 (1996)

[32] H Friedrich Hyperbolic reductions for Einsteinrsquos equations Class Quantum Grav 13 1451-

1469 (1996)

[33] F B Estabrook R S Robinson H D Wahlquist Hyperbolic equations for vacuum gravity

using special orthonormal frames Class Quantum Grav 14 1237-1247 (1997)

[34] M S Iriondo E O Leguizamon and O A Reula Einsteinrsquos Equations in Ashtekarrsquos Vari-

ables Constitute a Symmetric Hyperbolic System Phys Rev Lett 79 4732-4735 (1997)

[35] A Anderson Y Choquet-Bruhat and J W York Einstein-Bianchi hyperbolic system for

general relativity Topol Meth Nonlinear Anal 10 353-373 (1997)

[36] T W Baumgarte and S L Shapiro Numerical integration of Einsteinrsquos field equations

Phys Rev D 59 024007 (1998) (7 pages)

[37] M A G Bonilla Symmetric hyperbolic systems for Bianchi equations Class Quantum

20

10minus9

10minus7

10minus5

∆Arrinfin

10minus9

10minus7

10minus5

∆Kinfin

0 2000 4000 6000 8000 10000

10minus9

10minus7

10minus5

max(∆Brinfin ∆Γrinfin)

t

N = 11N = 14N = 17N = 20

FIG 4 Spectral convergence of solution for M = 1 Kerr-Schild initial data

Timestep choices are described in the caption for Fig 3 In the title headings for example

∆Arr equiv (Arr)numer minus (Arr)exact

applied to the full BSSN system Towards that goal we now discuss treatment of singularitiesand generalization of the described dG method to higher space dimension

To deal with the fixed Schwarzschild singularity we have used excision which is easy inthe context of the spherically reduced BSSN system However excision for the binary blackhole problem in full general relativity requires attention to the technical challenge of hori-zon tracking State-of-the-art BSSN codes avoid such complication relying instead on themoving-puncture technique While the moving-puncture technique does involve mild centralsingularities it may still prove amenable to spectral methods Indeed spectral methods fornon-smooth problems is well-developed in both theory and for complex applications Sincethe moving-puncture technique can be performed in spherical symmetry [60] a first-steptoward a spectral moving-puncture code would be to implement a moving puncture withthe nodal dG method described here Such an implementation may adopt Legendre-Gauss-Radau nodes on the innermost subdomain thereby ensuring that the physical singularitydoes not lie on a nodal point (in much the same way finite difference codes use a staggeredgrid) Beyond traditional excision and moving punctures one might also retain a smoothproblem for which spectral methods are especially well-suited via the turducken approach[87 88] to singularities

21

10minus11

10minus9

10minus7

10minus5

∆χinfin

10minus11

10minus9

10minus7

10minus5

max(∆grrinfin ∆gθθinfin)

0 2000 4000 6000 8000 10000

10minus11

10minus9

10minus7

10minus5

max(∆αinfin ∆βrinfin)

t

N = 11N = 14N = 17N = 20

FIG 5 Spectral convergence of solution violations for M = 1 Kerr-Schild initial

data See the caption of Fig 4 for details

Discontinuous Galerkin methods for hyperbolic problems in two and three space dimen-sions are well-developed A generalization of the method described here to three-dimensionsand the full BSSN system would likely rely on an unstructured mesh Appropriate local poly-nomial expansions for the subelements are well-understood as are choices for the numericalfluxes which would now live on two-dimensional faces rather than single points Whether ornot it would ultimately prove successful generalization of our dG method to a higher dimen-sion would rely on an established conceptual framework Further computational advancesof relevance to a generalization of our dG method to the full BSSN system (possibly includ-ing matter) may include mesh hp-adaptivity local timestepping shock capturing and slopelimiting techniques [66] Moreover recent work [89] indicates that enhanced performancewould be expected were our scheme implemented on graphics processor units

VI ACKNOWLEDGMENTS

We thank Nick Taylor for discussions about treating second-order operators with spectralmethods Benjamin Stamm for helping polish up a few parts of Appendix C Khosro Shah-bazi for discussions on LDG and IP methods We also acknowledge helpful conversationswith David Brown and Manuel Tiglio about the BSSN system and previous work on its nu-

22

merical implementation We gratefully acknowledge funding through grants DMS 0554377and DARPAAFOSR FA9550-05-1-0108 to Brown University and NSF grant PHY 0855678to the University of New Mexico

Appendix A Hyperbolicity of the first-order system

This appendix analyzes the matrix A(u) appearing in (10) in order to construct thecharacteristic fields (14) In matrix form the sector (8) of the principal part of (10) reads asfollows

partt

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

=

βr 0 minus 4λα3grr

0 0 λβr

6(grr)2λβr

3gθθgrr0 4λ

3grr

0 βr 0 23grrαχ

13α minus αχ

3grr

αχ3gθθ

minus23χ 0

0 0 βr 0 0 0 0 minus χgrr

0

0 0 minus 4α3grr

βr 0 βr

6(grr)2βr

3gθθgrr0 4

3grr

0 0 23αχ 0 βr minus βrχ

3grrminus2βrχ

3gθθ0 minus2

3χ

0 minus2α 0 0 0 23βr minus2grrβr

3gθθ0 4

3grr

0 gθθαgrr

0 0 0 minusgθθβr

3grr13βr 0 minus2

3gθθ

0 0 minus2α 0 0 0 0 βr 0

34

0 0 0 0 0 0 0 βr

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

prime

(A1)which defines the matrix A(u) appearing in (9) and so also the matrix A(u) in (10) Notethat in the last equation the matrix within the square brackets is minusA(u) For certainconfigurations of u and λ the system (10) is strongly hyperbolic [76] that is A(u) hasa complete set of eigenvectors and real eigenvalues Indeed five eigenpairs of A(u) aretrivially recovered upon inspection of A(u)rsquos leading 5times5 diagonal block These correspondto eigenvalue 0 and the left eigenspace ξj = eTj 1 le j le 5 where ej are the canonical

basis vectors Since each component of u arises as eTj W each is also a characteristic field

The remaining nine eigenpairs are determined by A(u) The eigenvalues of A(u) are

micro1 = 0 micro23 = minusβr microplusmn4 = minusβr plusmn

radic

2αχ

grr microplusmn

5 = minusβr plusmn α

radic

χ

grr microplusmn

6 = minusβr plusmnradic

λ

grr

(A2)

23

and the corresponding left eigenvectors are

x1 = (0 0 0 0 0 gθθ 2grr 0 0) (A3a)

x2 =

(

0 0 0 grr2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3b)

x3 =

(

grrλ 0 0 0

2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3c)

xplusmn4 =

(

0 0plusmnradic

2αgrrχ

0 0 0 0 1 0

)

(A3d)

xplusmn5 =

(

0∓ 3radicgrrχ

plusmn2

radic

grrχ 2grr

1

χminus 1

grr1

gθθ 0 0

)

(A3e)

xplusmn6 =

(

minus3

4

grrλ 0plusmn α

radicλgrr

(2αχminus λ) 0 0minus βr

8(βrgrr ∓radicλgrr)

minus βrgrr

4gθθ(βrgrr ∓radicλgrr)

αχ

(2αχminus λ)plusmnradic

grrλ

)

(A3f)

where for example xplusmn5 A(u) = microplusmn5 x

plusmn5 Assuming that grr gθθ χ and α are everywhere strictly

positive the eigenvalues are real and the eigenvectors are linearly independent providedthat (15) holds These eigenvectors are easily extended to eigenvectors of A(u) e g asxplusmn6 rarr (01times5 x

plusmn6 ) Then for example the characteristic field

Xplusmn6 equiv (01times5 x

plusmn6 )W = xplusmn6WvQ (A4)

and similarly Xplusmnj = xplusmnj WvQ for j = 4 5 and Xk = xkWvQ for k = 1 2 3 The characteristic

speeds for these fields are microk and microplusmnj With this convention the speeds listed in Table I

correspond to the Xk and Xplusmnj in (14)

Appendix B Schwarzschild solution in conformal Kerr-Schild coordinates

In Kerr-Schild coordinates here the system directly related to incoming Eddington-Finkelstein null coordinates the line element for the Schwarzschild solution reads

ds2 = minusα2dt2 + (1 + 2MR)(dR + βRdt)2 +R2dθ2 +R2 sin2 θdφ2 (B1)

where R is the area radius α = (1 + 2MR)minus12 is the lapse and βR = 2M(R + 2M) isthe shift vector The physical spatial metric gab is the spatial part of this line element

To define the corresponding solution to the BSSN system we use equation gab = χgab todefine the following relationship between line elements

dr2 + r2(dθ2 + sin2 θdφ2) = χ[(1 + 2MR)dR2 +R2dθ2 +R2 sin2 θdφ2] (B2)

so that

χ

(

1 +2M

R

)(

dR

dr

)2

= 1 χR2 = r2 (B3)

Then we have(

1 +2M

R

)12dR

R=dr

r (B4)

24

with integration yielding

r =R

4

(

1 +

radic

1 +2M

R

)2

e2minus2radic

1+2MR (B5)

where the constant of integration has been chosen so that the R r rarr infin limits are consistentThe second relation in (B3) shows that

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR =rArr χminus4 =2eradic

1+2MRminus1

1 +radic

1 + 2MR (B6)

The extrinsic curvature tensor is specified by the expression forK given in (B9h) the identityK = KR

R + 2Kθθ and

Kθθ =

(

1 +2M

R

)minus122M

R2 (B7)

Since KRR = Kr

r we compute that

Krr = K minus 2Kθ

θ = minus(

1 +2M

R

)minus12(R +M

R + 2M

)

2M

R2 (B8)

Next since Krr = grrKrr = χminus1Kr

r we have Krr = Arr+

13grrK This implies Arr = Kr

r minus 13K

from which we get (B9g) In all we have

α =

(

1 +2M

R

)minus12

(B9a)

βr = βR dr

dR= χ12

(

1 +2M

R

)minus122M

R(B9b)

grr = 1 (B9c)

gθθ = r2 = χR2 (B9d)

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR (B9e)

Br = 0 (B9f)

Arr = minus(

1 +2M

R

)minus124M

3R2

(

2R + 3M

R + 2M

)

(B9g)

K =

(

1 +2M

R

)minus32(

1 +3M

R

)

2M

R2(B9h)

Γr = minus2

r= minus 2

χ12R (B9i)

To differentiate these expressions with respect to r we use the identity

dR

dr= χminus12

(

1 +2M

R

)minus12

(B10)

along with the chain rule

25

Appendix C Stability of the model system

The following stability analysis for the model system (53) has been inspired by [78 79]After dropping all nonlinear source terms the system (53) becomes

parttu = uprime + av (C1a)

parttv = uprimeprime + vprime (C1b)

This section analyzes the stability of (C1) considering both the continuum system itself aswell as its semi-discrete dG approximation The latter analysis offers some insight into theempirically observed stability of our dG scheme for the spherically reduced BSSN equations

1 Analysis for a single interval

Throughout we work with the L2 inner product and norm

(f g)D =

int

D

fg fD =radic

(f f)D (C2)

where D is the spatial coordinate interval (here D may represent a subdomain Dk or the

whole domain Ω) and we have suppressed all integration measures For the continuummodel we will establish the following estimate

uprime(T middot)2D+ av(T middot)2

Dle C(T )

(

uprime(0 middot)2D+ av(0 middot)2

D

)

(C3)

where the time-dependent constant C(T ) is determined solely by the choice of boundaryconditions To show (C3) we first change variables with v =

radicav thereby rewriting (C1)

in the following symmetric form

parttu = uprime +radicav (C4a)

parttv =radicauprimeprime + vprime (C4b)

Equations (C4ab) then imply

1

2partt

int

D

(uprime)2 =

int

D

uprime(uprimeprime +radicavprime) =

int

D

radicauprimevprime +

1

2

int

partD

(uprime)2 (C5a)

1

2partt

int

D

(v)2 =

int

D

v(radicauprimeprime + vprime) = minus

int

D

radicauprimevprime +

1

2

int

partD

(v2 + 2radicauprimev) (C5b)

Here vvprime and uprimeuprimeprime have been expressed as exact derivatives and then integrated to boundaryterms the second equation employs an extra integration by parts and with only one spacedimension

int

partDdenotes a difference of endpoint evaluations Addition of Eqs (C5ab) gives

1

2partt

int

D

[

v2 + (uprime)2]

=1

2

int

partD

[

v2 + (uprime)2 + 2radicauprimev

]

(C6)

Substitutions with the identities

[

v2 + (uprime)2]

=1

2

[

(v + uprime)2 + (v minus uprime)2]

2uprimev =1

2

[

(v + uprime)2 minus (v minus uprime)2]

(C7)

26

and replacements to recover the original variable v = vradica yield

1

2partt

int

D

[

av2 + (uprime)2]

=1 +

radica

4

int

partD

(radicav + uprime)2 +

1minusradica

4

int

partD

(radicav minus uprime)2 (C8)

From (C8) we deduce that the time-dependent constant C(T ) in (C3) must satisfy

∣

∣

∣

∣

∣

∣

1 +

int T

0

[

1+radica

2

int

partD(radicav + uprime)2 + 1minus

radica

2

int

partD(radicav minus uprime)2

]

dt

uprime(0 middot)2 + av(0 middot)2

∣

∣

∣

∣

∣

∣

le C(T ) (C9)

For periodic boundary conditions we may choose C(T ) = 1 Moreover if a ge 1 anduprime = minusradic

av is specified at partD+ then uprime(t middot)2 + av(t middot)2 decaysStill working on a single interval (subdomain) we now consider the semi-discrete scheme

for (C4) i e (57) with all nonlinear source terms dropped and with v replaced by vradica

Derivation of a formula analogous to (C8) is our first step toward establishing L2 stabilityof the semi-discrete scheme While (57) features vectors for example u(t) taking values atthe Legendre-Gauss-Lobatto nodal points here we work with the numerical solution as apolynomial for example uh(t x) These two representations are related by the Lagrange in-terpolating polynomials for the nodal set here taken to span both the space of test functionsand the space of basis functions Our scheme is

int

Dk

ψparttuh =

int

Dk

ψ(Qh +radicavh) (C10a)

int

Dk

ξparttvh = minusint

Dk

ξprime(radicaQh + vh) +

int

partDk

ξ(radicaQlowast + vlowast) (C10b)

int

Dk

ϕQh =

int

Dk

ϕuprimeh +

int

partDk

ϕ (ulowast minus uh) (C10c)

where ψ ξ and ϕ are polynomial test functions These test functions are arbitrary exceptthat they must be degree-N polynomials In (C10) the variables uh vh and Qh should alsocarry a superscript k but we have suppressed this Derivation of a formula analogous to(C8) is complicated by the fact that Qh is not evolved Nevertheless at a given instant t wecan assemble Qh from (C10c)

Mimicking the calculation (C5b) from the continuum case we first use (C10b) with ξ = vhto write

1

2partt

int

Dk

v2h = minusint

Dk

(radicaQh + vh)v

primeh +

int

partDk

(radicaQlowast + vlowast)vh

= minusint

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2(radicaQlowast + vlowast)vh minus v2h

]

(C11)

The right-hand side of (C5a) is analogous to

1

2partt

int

Dk

Q2h =

int

Dk

QhparttQh (C12)

However since Qh is not evolved the term parttQh must be given a suitable interpretationOn the right hand side of (C10c) only uh u

primeh and u

lowast necessarily depend on time since the

27

test function ϕ need not be time-dependent Furthermore ulowast is explicitly given as a linearcombination of uh as seen in Eq (C21c) below Choosing ϕ = ℓj taking the time derivativeof (C10c) and appealing to the commutivity of mixed partial derivatives we therefore arriveat

int

Dk

ℓjparttQh =

int

Dk

ℓj(parttuh)prime +

int

partDk

ℓj(

(parttu)lowast minus parttuh

)

(C13)

where (parttu)lowast depends on parttuh in precisely the same way that ulowast depends on uh We have

written ℓj rather than ϕ in the last equation to emphasize that the result also holds for anylinear combination of ℓj (for example ϕ) and even for time-dependent combinations SinceQh is itself such a combination we obtain

1

2partt

int

Dk

Q2h =

int

Dk

Qh(parttuh)prime +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

Qh(Qprimeh +

radicavprimeh) +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2((parttu)lowast minus parttuh)Qh +Q2

h

]

(C14)

Addition of (C11) and (C14) gives

1

2partt

int

Dk

(Q2h + v2h) =

1

2

int

partDk

[

Q2h minus v2h + 2(

radicaQlowast + vlowast)vh + 2((parttu)

lowast minus parttuh)Qh

]

(C15)

the aforementioned analog of (C8) This formula holds on a single subdomain Dk and we

now combine multiple copies of it one for each value of k

2 Analysis for multiple intervals

To facilitate combination of (C15) over all k we change notation At every subdomaininterface Ik+12 equiv partDk cap partDk+1 let the superscripts L and R denote field values respectivelytaken from the left and right Then the fields evaluated at Ik+12 which belong to D

k will beuLk+12 v

Lk+12 and Q

Lk+12 while those belonging to D

k+1 will be uRk+12 vRk+12 and Q

Rk+12

However at Ikminus12 the values taken from D

k are uRkminus12 vRkminus12 and QR

kminus12 Note that we

have also replaced the subscript h denoting a numerical solution with k plusmn 12 denotingthe location of the endpoint value of the numerical solution With this notation we define

∆Lα =

1

2

[

(QLα)

2 minus (vLα)2]

+(radic

aQlowastα + vlowastα

)

vLα +[

(parttuα)lowast minus parttu

Lα

]

QLα (C16)

and similarly for ∆Rα The same numerical fluxes appear in both ∆L

α and ∆Rα (ie each

numerical flux takes the same value on either side of an interface) whence fluxes like Qlowastα do

not carry an L or R superscript In terms of these definitions (C15) becomes

1

2partt

int

Dk

(Q2h + v2h) = ∆L

k+12 minus∆Rkminus12 (C17)

28

Summation over all Dk yields

1

2partt

kmaxsum

k=1

int

Dk

(Q2h + v2h) =

kmaxminus1sum

k=1

(

∆Lk+12 minus∆R

k+12

)

+∆Lkmax+12 minus∆R

12

=

kmaxminus1sum

k=1

(

∆Lh minus∆R

h

)∣

∣

Ik+12 +∆Lkmax+12 minus∆R

12 (C18)

We have reverted to h-notation denoting the numerical solution since the LR superscriptsindicate unambiguously the relevant domain used for evaluation at Ik+12

We again seek an estimate of the form

kmaxsum

k=1

(

Qh(T middot)2Dk + avh(T middot)2Dk

)

le C(T )kmaxsum

k=1

(

Qh(0 middot)2Dk + avh(0 middot)2Dk

)

(C19)

that is essentially the same as the one (C3) considered in the continuum case Assume thatthe chosen boundary conditions ensure ∆L

kmax+12 minus ∆R12 is bounded by a time-dependent

constant which does not depend on the numerical parameters N and h (subdomain width)Establishment of stability then amounts to showing that the remaining sum over interfaceterms in (C18) is non-positive whence this remaining sum is consistent with C(T ) le 1although the boundary conditions may give rise to a different bound In fact we will choosethe numerical fluxes such that each individual interface term is non-positive At interfaceIk+12 and in LR notation the jump and average of vh for example are

1

2

(

v+ + vminus)

equiv vh =1

2

(

vLk+12 + vRk+12

)

(C20a)

nminusvminus + n+v+ equiv[[

vh]]

= vLk+12 minus vRk+12 (C20b)

Consider numerical fluxes of the form

Qlowast = Qh minusτQ2

[[

Qh

]]

(C21a)

vlowast = vh minusτv2

[[

vh]]

(C21b)

ulowast = uh minusτu2

[[

uh]]

(C21c)

(parttu)lowast = parttuh minus

τu2

[[

parttuh]]

(C21d)

where (C21c) induces (C21d) and where the penalty parameters τu τv and τQ are realnumbers The fluxes defined in (58) correspond to τu = 1 τv = 1 +

radica and τQ = 0 In

terms of these quantities the kth interface contribution in (C18) is

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

parttuh]]

minus τu2

[[

parttuh]][[

Qh

]]

(C22)

where we have suppressed the k dependence of the right-hand side Now consider the term[[

parttuh]]

Because parttuh and Qh+radicavh are both polynomials of degree N Eq (C10a) implies

29

τu

τ Q

minus1 0 1minus1

minus05

0

05

1

τu

minus1 0 1minus1

minus05

0

05

1

FIG 6 Stable evolutions for the model system For fixed τv = 10minus6 and τv = 1 +radic2

respectively the left and right plots depict stable choices (determined empirically) of τu and τQfor the linear model system (C1) The stable regions are colored black but the jagged edges result

from the discretization of the (τu τQ)-plane

the vector equation parttu = Q +radicav that is pointwise equivalence on the nodal points of

Dk which in turn implies

[[

parttuh]]

=[[

Qh+radicavh]]

Upon substituting this identity into thelast equation we arrive at an expression which features only vh and Qh

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

Qh +radicavh]]

minus τu2

[[

Qh +radicavh]][[

Qh

]]

(C23)

The identities vh[[

vh]]

= 12

[[

v2h]]

and[[

Qh +radicavh]]

=[[

Qh

]]

+radica[[

vh]]

then simplify(C23) to

(∆Lh minus∆R

h )∣

∣

Ik+12 = minusτv2

[[

vh]]2 minus

radica(τu + τQ)

2

[[

Qh

]][[

vh]]

minus τu2

[[

Qh

]]2 (C24)

The role of a penalty parameter is now clear Positive values of τv penalize jumps in vhthrough a negative contribution to the energy Likewise positive values of τu penalizejumps in Qh through a negative contribution to the energy However because the sign of[[

Qh

]][[

vh]]

can be positive or negative only the choice τQ = minusτu yields an expression for(∆L

h minus ∆Rh )|Ik+12 which is manifestly negative for τu ge 0 and τv ge 0 A simple estimate

based on Youngrsquos inequality with ε (that is 2αβ le εminus1α2 + εβ2 where α β ge 0 and ε gt 0)shows that for τQ = 0 the choice τv ge aτu4 also yields stability

Figure 6 depicts certain choices of stable penalty parameters for the linear model systemevolved to tfinal = 1000 (with a = 2 N = 10 and ∆t ≃ 00553) as determined empiricallywith simulations similar to those described in Sec IVA The left plot corresponds to asmall τv = 10minus6 for which the choice τu = 1 τQ = 0 is not stable as expected from thetheoretical analysis However the right plot corresponds to τv = 1 +

radica for which τu = 1

τQ = 0 is stable Motivated by the numerical flux choices (3842) used for the BSSN system

30

(4) we have (as mentioned above) set τu = 1 τv = 1 +radica and τQ = 0 in simulations of

the nonlinear model (53) For the nonlinear model system (53) the theoretically motivatedchoice τQ = minusτu also yields numerically stable evolutions when τu ge 0 and τv ge 0

For the nonlinear systems (4) and (53) we do not attempt a formal stability proofNevertheless the results of this appendix have served as a guide for our choices of penaltyparameters For the BSSN system (4) u v and Q are block indices [cf Eq (6)] Similarto the model problem we have penalized Q with τu = 1 with τv chosen large enough toheuristically overcome the cross-terms of indefinite size that arise from τQ = 0 (we interpretequations like τu = 1 componentwise) An analogous choice ldquoτQ = minusτurdquo for the BSSNsystem might be possible but would be considerably more complicated Indeed such achoice likely entails a matrix of penalty parameters but we do not give the details here

[1] F Pretorius Evolution of Binary Black-Hole Spacetimes Phys Rev Lett 95 121101 (2005)

(4 pages)

[2] F Pretorius Simulation of binary black hole spacetimes with a harmonic evolution scheme

Class Quantum Grav 23 S529-S552 (2006)

[3] M Campanelli C O Lousto P Marronetti and Y Zlochower Accurate Evolutions of Or-

biting Black-Hole Binaries without Excision Phys Rev Lett 96 111101 (2006) (4 pages)

[4] J G Baker J Centrella D-I Choi M Koppitz and J van Meter Gravitational-Wave

Extraction from an Inspiraling Configuration of Merging Black Holes Phys Rev Lett 96

111102 (2006) (4 pages)

[5] M Campanelli C O Lousto and Y Zlochower Last orbit of binary black holes Phys Rev D

73 061501(R) (2006) (5 pages)

[6] F Herrmann I Hinder D Shoemaker and P Laguna Unequal mass binary black hole plunges

and gravitational recoil Class Quantum Grav 24 S33-S42 (2007)

[7] P Diener F Herrmann D Pollney E Schnetter E Seidel R Takahashi J Thornburg and

J Ventrella Accurate Evolution of Orbiting Binary Black Holes Phys Rev Lett 96 121101

(2006) (4 pages)

[8] M A Scheel H P Pfeiffer L Lindblom L E Kidder O Rinne and S A Teukolsky Solving

Einsteinrsquos Equations With Dual Coordinate Frames Phys Rev D 74 104006 (2006)

[9] U Sperhake Binary black-hole evolutions of excision and puncture data Phys Rev D 76

104015 (2007) (20 pages)

[10] B Brugmann J A Gonzalez M Hannam S Husa U Sperhake and W Tichy Calibration

of moving puncture simulations Phys Rev D 77 024027 (2008) (25 pages)

[11] P Marronetti W Tichy B Brugmann J Gonzalez M Hannam S Husa and U Sperhake

Binary black holes on a budget simulations using workstations Class Quantum Grav 24

S43-S58 (2007)

[12] Z B Etienne J A Faber Y T Liu S L Shapiro and T W Baumgarte Filling the holes

Evolving excised binary black hole initial data with puncture techniques Phys Rev D 76

101503(R) (2007) (5 pages)

[13] B Szilagyi D Pollney L Rezzolla J Thornburg and J Winicour An explicit harmonic code

for black-hole evolution using excision Class Quantum Grav 24 S275-S293 (2007)

[14] L Boyle M Kesden and S Nissanke Binary-Black-Hole Merger Symmetry and the Spin

Expansion Phys Rev Lett 100 151101 (2008) (4 pages)

31

[15] M A Scheel M Boyle T Chu L E Kidder K D Matthews and H P Pfeiffer High-

accuracy waveforms for binary black hole inspiral merger and ringdown Phys Rev D 79

024003 (2009) (14 pages)

[16] M Hannam Status of black-hole-binary simulations for gravitational-wave detection

Class Quantum Grav 26 114001 (2009) (21 pages)

[17] I Hinder The current status of binary black hole simulations in numerical relativity

Class Quantum Grav 27 114004 (2010) (19 pages)

[18] B P Abbott et al LIGO The Laser Interferometer Gravitational-Wave Observatory

Rep Prog Phys 72 076901 (2009) (25 pages)

[19] F Acernese et al The status of VIRGO Class Quantum Grav 23 S63-S69 (2006)

[20] F Acernese et al Virgo status Class Quantum Grav 25 184001 (2008) (9 pages)

[21] R Arnowitt S Deser and C W Misner The Dynamics of General Relativity in Gravitation

an introduction to current research edited by L Witten (Wiley New York 1962) Chapter 7

227-265 Also available as gr-qc0405109

[22] J W York Jr Kinematics and Dynamics of General Relativity in Sources of Gravitational

Radiation edited by L L Smarr (Cambridge University Press Cambridge 1979)

[23] L E Kidder M A Scheel and S A Teukolsky Extending the lifetime of 3D black hole

computations with a new hyperbolic system of evolution equations Phys Rev D 64 064017

(2001) (13 pages)

[24] G Calabrese J Pullin O Sarbach and M Tiglio Convergence and stability in numerical

relativity Phys Rev D 66 041501(R) (2002) (4 pages)

[25] S Frittelli and O Reula On the Newtonian Limit of General Relativity Commun Math

Phys 166 221-235 (1994)

[26] M Shibata and T Nakamura Evolution of three-dimensional gravitational waves Harmonic

slicing case Phys Rev D 52 5428-5444 (1995)

[27] Y Choquet-Bruhat and J W York Geometrical well posed systems for the Einstein equations

C R Acad Sci Paris t 321 Serie I 1089-1095 (1995)

[28] A Abrahams A Anderson Y Choquet-Bruhat and J W York Jr Einstein and Yang-Mills

Theories in Hyperbolic Form without Gauge Fixing Phys Rev Lett 75 3377-3381 (1995)

[29] C Bona J Masso E Seidel and J Stela New Formalism for Numerical Relativity

Phys Rev Lett 75 600-603 (1995)

[30] M H P M van Putten and D M Eardley Nonlinear wave equations for relativity

Phys Rev D 53 3056-3063 (1996)

[31] S Frittelli and O A Reula First-Order Symmetric Hyperbolic Einstein Equations with Ar-

bitrary Fixed Gauge Phys Rev Lett 76 4667-4670 (1996)

[32] H Friedrich Hyperbolic reductions for Einsteinrsquos equations Class Quantum Grav 13 1451-

1469 (1996)

[33] F B Estabrook R S Robinson H D Wahlquist Hyperbolic equations for vacuum gravity

using special orthonormal frames Class Quantum Grav 14 1237-1247 (1997)

[34] M S Iriondo E O Leguizamon and O A Reula Einsteinrsquos Equations in Ashtekarrsquos Vari-

ables Constitute a Symmetric Hyperbolic System Phys Rev Lett 79 4732-4735 (1997)

[35] A Anderson Y Choquet-Bruhat and J W York Einstein-Bianchi hyperbolic system for

general relativity Topol Meth Nonlinear Anal 10 353-373 (1997)

[36] T W Baumgarte and S L Shapiro Numerical integration of Einsteinrsquos field equations

Phys Rev D 59 024007 (1998) (7 pages)

[37] M A G Bonilla Symmetric hyperbolic systems for Bianchi equations Class Quantum

21

10minus11

10minus9

10minus7

10minus5

∆χinfin

10minus11

10minus9

10minus7

10minus5

max(∆grrinfin ∆gθθinfin)

0 2000 4000 6000 8000 10000

10minus11

10minus9

10minus7

10minus5

max(∆αinfin ∆βrinfin)

t

N = 11N = 14N = 17N = 20

FIG 5 Spectral convergence of solution violations for M = 1 Kerr-Schild initial

data See the caption of Fig 4 for details

Discontinuous Galerkin methods for hyperbolic problems in two and three space dimen-sions are well-developed A generalization of the method described here to three-dimensionsand the full BSSN system would likely rely on an unstructured mesh Appropriate local poly-nomial expansions for the subelements are well-understood as are choices for the numericalfluxes which would now live on two-dimensional faces rather than single points Whether ornot it would ultimately prove successful generalization of our dG method to a higher dimen-sion would rely on an established conceptual framework Further computational advancesof relevance to a generalization of our dG method to the full BSSN system (possibly includ-ing matter) may include mesh hp-adaptivity local timestepping shock capturing and slopelimiting techniques [66] Moreover recent work [89] indicates that enhanced performancewould be expected were our scheme implemented on graphics processor units

VI ACKNOWLEDGMENTS

We thank Nick Taylor for discussions about treating second-order operators with spectralmethods Benjamin Stamm for helping polish up a few parts of Appendix C Khosro Shah-bazi for discussions on LDG and IP methods We also acknowledge helpful conversationswith David Brown and Manuel Tiglio about the BSSN system and previous work on its nu-

22

merical implementation We gratefully acknowledge funding through grants DMS 0554377and DARPAAFOSR FA9550-05-1-0108 to Brown University and NSF grant PHY 0855678to the University of New Mexico

Appendix A Hyperbolicity of the first-order system

This appendix analyzes the matrix A(u) appearing in (10) in order to construct thecharacteristic fields (14) In matrix form the sector (8) of the principal part of (10) reads asfollows

partt

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

=

βr 0 minus 4λα3grr

0 0 λβr

6(grr)2λβr

3gθθgrr0 4λ

3grr

0 βr 0 23grrαχ

13α minus αχ

3grr

αχ3gθθ

minus23χ 0

0 0 βr 0 0 0 0 minus χgrr

0

0 0 minus 4α3grr

βr 0 βr

6(grr)2βr

3gθθgrr0 4

3grr

0 0 23αχ 0 βr minus βrχ

3grrminus2βrχ

3gθθ0 minus2

3χ

0 minus2α 0 0 0 23βr minus2grrβr

3gθθ0 4

3grr

0 gθθαgrr

0 0 0 minusgθθβr

3grr13βr 0 minus2

3gθθ

0 0 minus2α 0 0 0 0 βr 0

34

0 0 0 0 0 0 0 βr

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

prime

(A1)which defines the matrix A(u) appearing in (9) and so also the matrix A(u) in (10) Notethat in the last equation the matrix within the square brackets is minusA(u) For certainconfigurations of u and λ the system (10) is strongly hyperbolic [76] that is A(u) hasa complete set of eigenvectors and real eigenvalues Indeed five eigenpairs of A(u) aretrivially recovered upon inspection of A(u)rsquos leading 5times5 diagonal block These correspondto eigenvalue 0 and the left eigenspace ξj = eTj 1 le j le 5 where ej are the canonical

basis vectors Since each component of u arises as eTj W each is also a characteristic field

The remaining nine eigenpairs are determined by A(u) The eigenvalues of A(u) are

micro1 = 0 micro23 = minusβr microplusmn4 = minusβr plusmn

radic

2αχ

grr microplusmn

5 = minusβr plusmn α

radic

χ

grr microplusmn

6 = minusβr plusmnradic

λ

grr

(A2)

23

and the corresponding left eigenvectors are

x1 = (0 0 0 0 0 gθθ 2grr 0 0) (A3a)

x2 =

(

0 0 0 grr2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3b)

x3 =

(

grrλ 0 0 0

2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3c)

xplusmn4 =

(

0 0plusmnradic

2αgrrχ

0 0 0 0 1 0

)

(A3d)

xplusmn5 =

(

0∓ 3radicgrrχ

plusmn2

radic

grrχ 2grr

1

χminus 1

grr1

gθθ 0 0

)

(A3e)

xplusmn6 =

(

minus3

4

grrλ 0plusmn α

radicλgrr

(2αχminus λ) 0 0minus βr

8(βrgrr ∓radicλgrr)

minus βrgrr

4gθθ(βrgrr ∓radicλgrr)

αχ

(2αχminus λ)plusmnradic

grrλ

)

(A3f)

where for example xplusmn5 A(u) = microplusmn5 x

plusmn5 Assuming that grr gθθ χ and α are everywhere strictly

positive the eigenvalues are real and the eigenvectors are linearly independent providedthat (15) holds These eigenvectors are easily extended to eigenvectors of A(u) e g asxplusmn6 rarr (01times5 x

plusmn6 ) Then for example the characteristic field

Xplusmn6 equiv (01times5 x

plusmn6 )W = xplusmn6WvQ (A4)

and similarly Xplusmnj = xplusmnj WvQ for j = 4 5 and Xk = xkWvQ for k = 1 2 3 The characteristic

speeds for these fields are microk and microplusmnj With this convention the speeds listed in Table I

correspond to the Xk and Xplusmnj in (14)

Appendix B Schwarzschild solution in conformal Kerr-Schild coordinates

In Kerr-Schild coordinates here the system directly related to incoming Eddington-Finkelstein null coordinates the line element for the Schwarzschild solution reads

ds2 = minusα2dt2 + (1 + 2MR)(dR + βRdt)2 +R2dθ2 +R2 sin2 θdφ2 (B1)

where R is the area radius α = (1 + 2MR)minus12 is the lapse and βR = 2M(R + 2M) isthe shift vector The physical spatial metric gab is the spatial part of this line element

To define the corresponding solution to the BSSN system we use equation gab = χgab todefine the following relationship between line elements

dr2 + r2(dθ2 + sin2 θdφ2) = χ[(1 + 2MR)dR2 +R2dθ2 +R2 sin2 θdφ2] (B2)

so that

χ

(

1 +2M

R

)(

dR

dr

)2

= 1 χR2 = r2 (B3)

Then we have(

1 +2M

R

)12dR

R=dr

r (B4)

24

with integration yielding

r =R

4

(

1 +

radic

1 +2M

R

)2

e2minus2radic

1+2MR (B5)

where the constant of integration has been chosen so that the R r rarr infin limits are consistentThe second relation in (B3) shows that

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR =rArr χminus4 =2eradic

1+2MRminus1

1 +radic

1 + 2MR (B6)

The extrinsic curvature tensor is specified by the expression forK given in (B9h) the identityK = KR

R + 2Kθθ and

Kθθ =

(

1 +2M

R

)minus122M

R2 (B7)

Since KRR = Kr

r we compute that

Krr = K minus 2Kθ

θ = minus(

1 +2M

R

)minus12(R +M

R + 2M

)

2M

R2 (B8)

Next since Krr = grrKrr = χminus1Kr

r we have Krr = Arr+

13grrK This implies Arr = Kr

r minus 13K

from which we get (B9g) In all we have

α =

(

1 +2M

R

)minus12

(B9a)

βr = βR dr

dR= χ12

(

1 +2M

R

)minus122M

R(B9b)

grr = 1 (B9c)

gθθ = r2 = χR2 (B9d)

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR (B9e)

Br = 0 (B9f)

Arr = minus(

1 +2M

R

)minus124M

3R2

(

2R + 3M

R + 2M

)

(B9g)

K =

(

1 +2M

R

)minus32(

1 +3M

R

)

2M

R2(B9h)

Γr = minus2

r= minus 2

χ12R (B9i)

To differentiate these expressions with respect to r we use the identity

dR

dr= χminus12

(

1 +2M

R

)minus12

(B10)

along with the chain rule

25

Appendix C Stability of the model system

The following stability analysis for the model system (53) has been inspired by [78 79]After dropping all nonlinear source terms the system (53) becomes

parttu = uprime + av (C1a)

parttv = uprimeprime + vprime (C1b)

This section analyzes the stability of (C1) considering both the continuum system itself aswell as its semi-discrete dG approximation The latter analysis offers some insight into theempirically observed stability of our dG scheme for the spherically reduced BSSN equations

1 Analysis for a single interval

Throughout we work with the L2 inner product and norm

(f g)D =

int

D

fg fD =radic

(f f)D (C2)

where D is the spatial coordinate interval (here D may represent a subdomain Dk or the

whole domain Ω) and we have suppressed all integration measures For the continuummodel we will establish the following estimate

uprime(T middot)2D+ av(T middot)2

Dle C(T )

(

uprime(0 middot)2D+ av(0 middot)2

D

)

(C3)

where the time-dependent constant C(T ) is determined solely by the choice of boundaryconditions To show (C3) we first change variables with v =

radicav thereby rewriting (C1)

in the following symmetric form

parttu = uprime +radicav (C4a)

parttv =radicauprimeprime + vprime (C4b)

Equations (C4ab) then imply

1

2partt

int

D

(uprime)2 =

int

D

uprime(uprimeprime +radicavprime) =

int

D

radicauprimevprime +

1

2

int

partD

(uprime)2 (C5a)

1

2partt

int

D

(v)2 =

int

D

v(radicauprimeprime + vprime) = minus

int

D

radicauprimevprime +

1

2

int

partD

(v2 + 2radicauprimev) (C5b)

Here vvprime and uprimeuprimeprime have been expressed as exact derivatives and then integrated to boundaryterms the second equation employs an extra integration by parts and with only one spacedimension

int

partDdenotes a difference of endpoint evaluations Addition of Eqs (C5ab) gives

1

2partt

int

D

[

v2 + (uprime)2]

=1

2

int

partD

[

v2 + (uprime)2 + 2radicauprimev

]

(C6)

Substitutions with the identities

[

v2 + (uprime)2]

=1

2

[

(v + uprime)2 + (v minus uprime)2]

2uprimev =1

2

[

(v + uprime)2 minus (v minus uprime)2]

(C7)

26

and replacements to recover the original variable v = vradica yield

1

2partt

int

D

[

av2 + (uprime)2]

=1 +

radica

4

int

partD

(radicav + uprime)2 +

1minusradica

4

int

partD

(radicav minus uprime)2 (C8)

From (C8) we deduce that the time-dependent constant C(T ) in (C3) must satisfy

∣

∣

∣

∣

∣

∣

1 +

int T

0

[

1+radica

2

int

partD(radicav + uprime)2 + 1minus

radica

2

int

partD(radicav minus uprime)2

]

dt

uprime(0 middot)2 + av(0 middot)2

∣

∣

∣

∣

∣

∣

le C(T ) (C9)

For periodic boundary conditions we may choose C(T ) = 1 Moreover if a ge 1 anduprime = minusradic

av is specified at partD+ then uprime(t middot)2 + av(t middot)2 decaysStill working on a single interval (subdomain) we now consider the semi-discrete scheme

for (C4) i e (57) with all nonlinear source terms dropped and with v replaced by vradica

Derivation of a formula analogous to (C8) is our first step toward establishing L2 stabilityof the semi-discrete scheme While (57) features vectors for example u(t) taking values atthe Legendre-Gauss-Lobatto nodal points here we work with the numerical solution as apolynomial for example uh(t x) These two representations are related by the Lagrange in-terpolating polynomials for the nodal set here taken to span both the space of test functionsand the space of basis functions Our scheme is

int

Dk

ψparttuh =

int

Dk

ψ(Qh +radicavh) (C10a)

int

Dk

ξparttvh = minusint

Dk

ξprime(radicaQh + vh) +

int

partDk

ξ(radicaQlowast + vlowast) (C10b)

int

Dk

ϕQh =

int

Dk

ϕuprimeh +

int

partDk

ϕ (ulowast minus uh) (C10c)

where ψ ξ and ϕ are polynomial test functions These test functions are arbitrary exceptthat they must be degree-N polynomials In (C10) the variables uh vh and Qh should alsocarry a superscript k but we have suppressed this Derivation of a formula analogous to(C8) is complicated by the fact that Qh is not evolved Nevertheless at a given instant t wecan assemble Qh from (C10c)

Mimicking the calculation (C5b) from the continuum case we first use (C10b) with ξ = vhto write

1

2partt

int

Dk

v2h = minusint

Dk

(radicaQh + vh)v

primeh +

int

partDk

(radicaQlowast + vlowast)vh

= minusint

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2(radicaQlowast + vlowast)vh minus v2h

]

(C11)

The right-hand side of (C5a) is analogous to

1

2partt

int

Dk

Q2h =

int

Dk

QhparttQh (C12)

However since Qh is not evolved the term parttQh must be given a suitable interpretationOn the right hand side of (C10c) only uh u

primeh and u

lowast necessarily depend on time since the

27

test function ϕ need not be time-dependent Furthermore ulowast is explicitly given as a linearcombination of uh as seen in Eq (C21c) below Choosing ϕ = ℓj taking the time derivativeof (C10c) and appealing to the commutivity of mixed partial derivatives we therefore arriveat

int

Dk

ℓjparttQh =

int

Dk

ℓj(parttuh)prime +

int

partDk

ℓj(

(parttu)lowast minus parttuh

)

(C13)

where (parttu)lowast depends on parttuh in precisely the same way that ulowast depends on uh We have

written ℓj rather than ϕ in the last equation to emphasize that the result also holds for anylinear combination of ℓj (for example ϕ) and even for time-dependent combinations SinceQh is itself such a combination we obtain

1

2partt

int

Dk

Q2h =

int

Dk

Qh(parttuh)prime +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

Qh(Qprimeh +

radicavprimeh) +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2((parttu)lowast minus parttuh)Qh +Q2

h

]

(C14)

Addition of (C11) and (C14) gives

1

2partt

int

Dk

(Q2h + v2h) =

1

2

int

partDk

[

Q2h minus v2h + 2(

radicaQlowast + vlowast)vh + 2((parttu)

lowast minus parttuh)Qh

]

(C15)

the aforementioned analog of (C8) This formula holds on a single subdomain Dk and we

now combine multiple copies of it one for each value of k

2 Analysis for multiple intervals

To facilitate combination of (C15) over all k we change notation At every subdomaininterface Ik+12 equiv partDk cap partDk+1 let the superscripts L and R denote field values respectivelytaken from the left and right Then the fields evaluated at Ik+12 which belong to D

k will beuLk+12 v

Lk+12 and Q

Lk+12 while those belonging to D

k+1 will be uRk+12 vRk+12 and Q

Rk+12

However at Ikminus12 the values taken from D

k are uRkminus12 vRkminus12 and QR

kminus12 Note that we

have also replaced the subscript h denoting a numerical solution with k plusmn 12 denotingthe location of the endpoint value of the numerical solution With this notation we define

∆Lα =

1

2

[

(QLα)

2 minus (vLα)2]

+(radic

aQlowastα + vlowastα

)

vLα +[

(parttuα)lowast minus parttu

Lα

]

QLα (C16)

and similarly for ∆Rα The same numerical fluxes appear in both ∆L

α and ∆Rα (ie each

numerical flux takes the same value on either side of an interface) whence fluxes like Qlowastα do

not carry an L or R superscript In terms of these definitions (C15) becomes

1

2partt

int

Dk

(Q2h + v2h) = ∆L

k+12 minus∆Rkminus12 (C17)

28

Summation over all Dk yields

1

2partt

kmaxsum

k=1

int

Dk

(Q2h + v2h) =

kmaxminus1sum

k=1

(

∆Lk+12 minus∆R

k+12

)

+∆Lkmax+12 minus∆R

12

=

kmaxminus1sum

k=1

(

∆Lh minus∆R

h

)∣

∣

Ik+12 +∆Lkmax+12 minus∆R

12 (C18)

We have reverted to h-notation denoting the numerical solution since the LR superscriptsindicate unambiguously the relevant domain used for evaluation at Ik+12

We again seek an estimate of the form

kmaxsum

k=1

(

Qh(T middot)2Dk + avh(T middot)2Dk

)

le C(T )kmaxsum

k=1

(

Qh(0 middot)2Dk + avh(0 middot)2Dk

)

(C19)

that is essentially the same as the one (C3) considered in the continuum case Assume thatthe chosen boundary conditions ensure ∆L

kmax+12 minus ∆R12 is bounded by a time-dependent

constant which does not depend on the numerical parameters N and h (subdomain width)Establishment of stability then amounts to showing that the remaining sum over interfaceterms in (C18) is non-positive whence this remaining sum is consistent with C(T ) le 1although the boundary conditions may give rise to a different bound In fact we will choosethe numerical fluxes such that each individual interface term is non-positive At interfaceIk+12 and in LR notation the jump and average of vh for example are

1

2

(

v+ + vminus)

equiv vh =1

2

(

vLk+12 + vRk+12

)

(C20a)

nminusvminus + n+v+ equiv[[

vh]]

= vLk+12 minus vRk+12 (C20b)

Consider numerical fluxes of the form

Qlowast = Qh minusτQ2

[[

Qh

]]

(C21a)

vlowast = vh minusτv2

[[

vh]]

(C21b)

ulowast = uh minusτu2

[[

uh]]

(C21c)

(parttu)lowast = parttuh minus

τu2

[[

parttuh]]

(C21d)

where (C21c) induces (C21d) and where the penalty parameters τu τv and τQ are realnumbers The fluxes defined in (58) correspond to τu = 1 τv = 1 +

radica and τQ = 0 In

terms of these quantities the kth interface contribution in (C18) is

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

parttuh]]

minus τu2

[[

parttuh]][[

Qh

]]

(C22)

where we have suppressed the k dependence of the right-hand side Now consider the term[[

parttuh]]

Because parttuh and Qh+radicavh are both polynomials of degree N Eq (C10a) implies

29

τu

τ Q

minus1 0 1minus1

minus05

0

05

1

τu

minus1 0 1minus1

minus05

0

05

1

FIG 6 Stable evolutions for the model system For fixed τv = 10minus6 and τv = 1 +radic2

respectively the left and right plots depict stable choices (determined empirically) of τu and τQfor the linear model system (C1) The stable regions are colored black but the jagged edges result

from the discretization of the (τu τQ)-plane

the vector equation parttu = Q +radicav that is pointwise equivalence on the nodal points of

Dk which in turn implies

[[

parttuh]]

=[[

Qh+radicavh]]

Upon substituting this identity into thelast equation we arrive at an expression which features only vh and Qh

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

Qh +radicavh]]

minus τu2

[[

Qh +radicavh]][[

Qh

]]

(C23)

The identities vh[[

vh]]

= 12

[[

v2h]]

and[[

Qh +radicavh]]

=[[

Qh

]]

+radica[[

vh]]

then simplify(C23) to

(∆Lh minus∆R

h )∣

∣

Ik+12 = minusτv2

[[

vh]]2 minus

radica(τu + τQ)

2

[[

Qh

]][[

vh]]

minus τu2

[[

Qh

]]2 (C24)

The role of a penalty parameter is now clear Positive values of τv penalize jumps in vhthrough a negative contribution to the energy Likewise positive values of τu penalizejumps in Qh through a negative contribution to the energy However because the sign of[[

Qh

]][[

vh]]

can be positive or negative only the choice τQ = minusτu yields an expression for(∆L

h minus ∆Rh )|Ik+12 which is manifestly negative for τu ge 0 and τv ge 0 A simple estimate

based on Youngrsquos inequality with ε (that is 2αβ le εminus1α2 + εβ2 where α β ge 0 and ε gt 0)shows that for τQ = 0 the choice τv ge aτu4 also yields stability

Figure 6 depicts certain choices of stable penalty parameters for the linear model systemevolved to tfinal = 1000 (with a = 2 N = 10 and ∆t ≃ 00553) as determined empiricallywith simulations similar to those described in Sec IVA The left plot corresponds to asmall τv = 10minus6 for which the choice τu = 1 τQ = 0 is not stable as expected from thetheoretical analysis However the right plot corresponds to τv = 1 +

radica for which τu = 1

τQ = 0 is stable Motivated by the numerical flux choices (3842) used for the BSSN system

30

(4) we have (as mentioned above) set τu = 1 τv = 1 +radica and τQ = 0 in simulations of

the nonlinear model (53) For the nonlinear model system (53) the theoretically motivatedchoice τQ = minusτu also yields numerically stable evolutions when τu ge 0 and τv ge 0

For the nonlinear systems (4) and (53) we do not attempt a formal stability proofNevertheless the results of this appendix have served as a guide for our choices of penaltyparameters For the BSSN system (4) u v and Q are block indices [cf Eq (6)] Similarto the model problem we have penalized Q with τu = 1 with τv chosen large enough toheuristically overcome the cross-terms of indefinite size that arise from τQ = 0 (we interpretequations like τu = 1 componentwise) An analogous choice ldquoτQ = minusτurdquo for the BSSNsystem might be possible but would be considerably more complicated Indeed such achoice likely entails a matrix of penalty parameters but we do not give the details here

[1] F Pretorius Evolution of Binary Black-Hole Spacetimes Phys Rev Lett 95 121101 (2005)

(4 pages)

[2] F Pretorius Simulation of binary black hole spacetimes with a harmonic evolution scheme

Class Quantum Grav 23 S529-S552 (2006)

[3] M Campanelli C O Lousto P Marronetti and Y Zlochower Accurate Evolutions of Or-

biting Black-Hole Binaries without Excision Phys Rev Lett 96 111101 (2006) (4 pages)

[4] J G Baker J Centrella D-I Choi M Koppitz and J van Meter Gravitational-Wave

Extraction from an Inspiraling Configuration of Merging Black Holes Phys Rev Lett 96

111102 (2006) (4 pages)

[5] M Campanelli C O Lousto and Y Zlochower Last orbit of binary black holes Phys Rev D

73 061501(R) (2006) (5 pages)

[6] F Herrmann I Hinder D Shoemaker and P Laguna Unequal mass binary black hole plunges

and gravitational recoil Class Quantum Grav 24 S33-S42 (2007)

[7] P Diener F Herrmann D Pollney E Schnetter E Seidel R Takahashi J Thornburg and

J Ventrella Accurate Evolution of Orbiting Binary Black Holes Phys Rev Lett 96 121101

(2006) (4 pages)

[8] M A Scheel H P Pfeiffer L Lindblom L E Kidder O Rinne and S A Teukolsky Solving

Einsteinrsquos Equations With Dual Coordinate Frames Phys Rev D 74 104006 (2006)

[9] U Sperhake Binary black-hole evolutions of excision and puncture data Phys Rev D 76

104015 (2007) (20 pages)

[10] B Brugmann J A Gonzalez M Hannam S Husa U Sperhake and W Tichy Calibration

of moving puncture simulations Phys Rev D 77 024027 (2008) (25 pages)

[11] P Marronetti W Tichy B Brugmann J Gonzalez M Hannam S Husa and U Sperhake

Binary black holes on a budget simulations using workstations Class Quantum Grav 24

S43-S58 (2007)

[12] Z B Etienne J A Faber Y T Liu S L Shapiro and T W Baumgarte Filling the holes

Evolving excised binary black hole initial data with puncture techniques Phys Rev D 76

101503(R) (2007) (5 pages)

[13] B Szilagyi D Pollney L Rezzolla J Thornburg and J Winicour An explicit harmonic code

for black-hole evolution using excision Class Quantum Grav 24 S275-S293 (2007)

[14] L Boyle M Kesden and S Nissanke Binary-Black-Hole Merger Symmetry and the Spin

Expansion Phys Rev Lett 100 151101 (2008) (4 pages)

31

[15] M A Scheel M Boyle T Chu L E Kidder K D Matthews and H P Pfeiffer High-

accuracy waveforms for binary black hole inspiral merger and ringdown Phys Rev D 79

024003 (2009) (14 pages)

[16] M Hannam Status of black-hole-binary simulations for gravitational-wave detection

Class Quantum Grav 26 114001 (2009) (21 pages)

[17] I Hinder The current status of binary black hole simulations in numerical relativity

Class Quantum Grav 27 114004 (2010) (19 pages)

[18] B P Abbott et al LIGO The Laser Interferometer Gravitational-Wave Observatory

Rep Prog Phys 72 076901 (2009) (25 pages)

[19] F Acernese et al The status of VIRGO Class Quantum Grav 23 S63-S69 (2006)

[20] F Acernese et al Virgo status Class Quantum Grav 25 184001 (2008) (9 pages)

[21] R Arnowitt S Deser and C W Misner The Dynamics of General Relativity in Gravitation

an introduction to current research edited by L Witten (Wiley New York 1962) Chapter 7

227-265 Also available as gr-qc0405109

[22] J W York Jr Kinematics and Dynamics of General Relativity in Sources of Gravitational

Radiation edited by L L Smarr (Cambridge University Press Cambridge 1979)

[23] L E Kidder M A Scheel and S A Teukolsky Extending the lifetime of 3D black hole

computations with a new hyperbolic system of evolution equations Phys Rev D 64 064017

(2001) (13 pages)

[24] G Calabrese J Pullin O Sarbach and M Tiglio Convergence and stability in numerical

relativity Phys Rev D 66 041501(R) (2002) (4 pages)

[25] S Frittelli and O Reula On the Newtonian Limit of General Relativity Commun Math

Phys 166 221-235 (1994)

[26] M Shibata and T Nakamura Evolution of three-dimensional gravitational waves Harmonic

slicing case Phys Rev D 52 5428-5444 (1995)

[27] Y Choquet-Bruhat and J W York Geometrical well posed systems for the Einstein equations

C R Acad Sci Paris t 321 Serie I 1089-1095 (1995)

[28] A Abrahams A Anderson Y Choquet-Bruhat and J W York Jr Einstein and Yang-Mills

Theories in Hyperbolic Form without Gauge Fixing Phys Rev Lett 75 3377-3381 (1995)

[29] C Bona J Masso E Seidel and J Stela New Formalism for Numerical Relativity

Phys Rev Lett 75 600-603 (1995)

[30] M H P M van Putten and D M Eardley Nonlinear wave equations for relativity

Phys Rev D 53 3056-3063 (1996)

[31] S Frittelli and O A Reula First-Order Symmetric Hyperbolic Einstein Equations with Ar-

bitrary Fixed Gauge Phys Rev Lett 76 4667-4670 (1996)

[32] H Friedrich Hyperbolic reductions for Einsteinrsquos equations Class Quantum Grav 13 1451-

1469 (1996)

[33] F B Estabrook R S Robinson H D Wahlquist Hyperbolic equations for vacuum gravity

using special orthonormal frames Class Quantum Grav 14 1237-1247 (1997)

[34] M S Iriondo E O Leguizamon and O A Reula Einsteinrsquos Equations in Ashtekarrsquos Vari-

ables Constitute a Symmetric Hyperbolic System Phys Rev Lett 79 4732-4735 (1997)

[35] A Anderson Y Choquet-Bruhat and J W York Einstein-Bianchi hyperbolic system for

general relativity Topol Meth Nonlinear Anal 10 353-373 (1997)

[36] T W Baumgarte and S L Shapiro Numerical integration of Einsteinrsquos field equations

Phys Rev D 59 024007 (1998) (7 pages)

[37] M A G Bonilla Symmetric hyperbolic systems for Bianchi equations Class Quantum

22

merical implementation We gratefully acknowledge funding through grants DMS 0554377and DARPAAFOSR FA9550-05-1-0108 to Brown University and NSF grant PHY 0855678to the University of New Mexico

Appendix A Hyperbolicity of the first-order system

This appendix analyzes the matrix A(u) appearing in (10) in order to construct thecharacteristic fields (14) In matrix form the sector (8) of the principal part of (10) reads asfollows

partt

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

=

βr 0 minus 4λα3grr

0 0 λβr

6(grr)2λβr

3gθθgrr0 4λ

3grr

0 βr 0 23grrαχ

13α minus αχ

3grr

αχ3gθθ

minus23χ 0

0 0 βr 0 0 0 0 minus χgrr

0

0 0 minus 4α3grr

βr 0 βr

6(grr)2βr

3gθθgrr0 4

3grr

0 0 23αχ 0 βr minus βrχ

3grrminus2βrχ

3gθθ0 minus2

3χ

0 minus2α 0 0 0 23βr minus2grrβr

3gθθ0 4

3grr

0 gθθαgrr

0 0 0 minusgθθβr

3grr13βr 0 minus2

3gθθ

0 0 minus2α 0 0 0 0 βr 0

34

0 0 0 0 0 0 0 βr

Br

Arr

K

Γr

Qχ

Qgrr

Qgθθ

Qα

Qβr

prime

(A1)which defines the matrix A(u) appearing in (9) and so also the matrix A(u) in (10) Notethat in the last equation the matrix within the square brackets is minusA(u) For certainconfigurations of u and λ the system (10) is strongly hyperbolic [76] that is A(u) hasa complete set of eigenvectors and real eigenvalues Indeed five eigenpairs of A(u) aretrivially recovered upon inspection of A(u)rsquos leading 5times5 diagonal block These correspondto eigenvalue 0 and the left eigenspace ξj = eTj 1 le j le 5 where ej are the canonical

basis vectors Since each component of u arises as eTj W each is also a characteristic field

The remaining nine eigenpairs are determined by A(u) The eigenvalues of A(u) are

micro1 = 0 micro23 = minusβr microplusmn4 = minusβr plusmn

radic

2αχ

grr microplusmn

5 = minusβr plusmn α

radic

χ

grr microplusmn

6 = minusβr plusmnradic

λ

grr

(A2)

23

and the corresponding left eigenvectors are

x1 = (0 0 0 0 0 gθθ 2grr 0 0) (A3a)

x2 =

(

0 0 0 grr2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3b)

x3 =

(

grrλ 0 0 0

2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3c)

xplusmn4 =

(

0 0plusmnradic

2αgrrχ

0 0 0 0 1 0

)

(A3d)

xplusmn5 =

(

0∓ 3radicgrrχ

plusmn2

radic

grrχ 2grr

1

χminus 1

grr1

gθθ 0 0

)

(A3e)

xplusmn6 =

(

minus3

4

grrλ 0plusmn α

radicλgrr

(2αχminus λ) 0 0minus βr

8(βrgrr ∓radicλgrr)

minus βrgrr

4gθθ(βrgrr ∓radicλgrr)

αχ

(2αχminus λ)plusmnradic

grrλ

)

(A3f)

where for example xplusmn5 A(u) = microplusmn5 x

plusmn5 Assuming that grr gθθ χ and α are everywhere strictly

positive the eigenvalues are real and the eigenvectors are linearly independent providedthat (15) holds These eigenvectors are easily extended to eigenvectors of A(u) e g asxplusmn6 rarr (01times5 x

plusmn6 ) Then for example the characteristic field

Xplusmn6 equiv (01times5 x

plusmn6 )W = xplusmn6WvQ (A4)

and similarly Xplusmnj = xplusmnj WvQ for j = 4 5 and Xk = xkWvQ for k = 1 2 3 The characteristic

speeds for these fields are microk and microplusmnj With this convention the speeds listed in Table I

correspond to the Xk and Xplusmnj in (14)

Appendix B Schwarzschild solution in conformal Kerr-Schild coordinates

In Kerr-Schild coordinates here the system directly related to incoming Eddington-Finkelstein null coordinates the line element for the Schwarzschild solution reads

ds2 = minusα2dt2 + (1 + 2MR)(dR + βRdt)2 +R2dθ2 +R2 sin2 θdφ2 (B1)

where R is the area radius α = (1 + 2MR)minus12 is the lapse and βR = 2M(R + 2M) isthe shift vector The physical spatial metric gab is the spatial part of this line element

To define the corresponding solution to the BSSN system we use equation gab = χgab todefine the following relationship between line elements

dr2 + r2(dθ2 + sin2 θdφ2) = χ[(1 + 2MR)dR2 +R2dθ2 +R2 sin2 θdφ2] (B2)

so that

χ

(

1 +2M

R

)(

dR

dr

)2

= 1 χR2 = r2 (B3)

Then we have(

1 +2M

R

)12dR

R=dr

r (B4)

24

with integration yielding

r =R

4

(

1 +

radic

1 +2M

R

)2

e2minus2radic

1+2MR (B5)

where the constant of integration has been chosen so that the R r rarr infin limits are consistentThe second relation in (B3) shows that

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR =rArr χminus4 =2eradic

1+2MRminus1

1 +radic

1 + 2MR (B6)

The extrinsic curvature tensor is specified by the expression forK given in (B9h) the identityK = KR

R + 2Kθθ and

Kθθ =

(

1 +2M

R

)minus122M

R2 (B7)

Since KRR = Kr

r we compute that

Krr = K minus 2Kθ

θ = minus(

1 +2M

R

)minus12(R +M

R + 2M

)

2M

R2 (B8)

Next since Krr = grrKrr = χminus1Kr

r we have Krr = Arr+

13grrK This implies Arr = Kr

r minus 13K

from which we get (B9g) In all we have

α =

(

1 +2M

R

)minus12

(B9a)

βr = βR dr

dR= χ12

(

1 +2M

R

)minus122M

R(B9b)

grr = 1 (B9c)

gθθ = r2 = χR2 (B9d)

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR (B9e)

Br = 0 (B9f)

Arr = minus(

1 +2M

R

)minus124M

3R2

(

2R + 3M

R + 2M

)

(B9g)

K =

(

1 +2M

R

)minus32(

1 +3M

R

)

2M

R2(B9h)

Γr = minus2

r= minus 2

χ12R (B9i)

To differentiate these expressions with respect to r we use the identity

dR

dr= χminus12

(

1 +2M

R

)minus12

(B10)

along with the chain rule

25

Appendix C Stability of the model system

The following stability analysis for the model system (53) has been inspired by [78 79]After dropping all nonlinear source terms the system (53) becomes

parttu = uprime + av (C1a)

parttv = uprimeprime + vprime (C1b)

This section analyzes the stability of (C1) considering both the continuum system itself aswell as its semi-discrete dG approximation The latter analysis offers some insight into theempirically observed stability of our dG scheme for the spherically reduced BSSN equations

1 Analysis for a single interval

Throughout we work with the L2 inner product and norm

(f g)D =

int

D

fg fD =radic

(f f)D (C2)

where D is the spatial coordinate interval (here D may represent a subdomain Dk or the

whole domain Ω) and we have suppressed all integration measures For the continuummodel we will establish the following estimate

uprime(T middot)2D+ av(T middot)2

Dle C(T )

(

uprime(0 middot)2D+ av(0 middot)2

D

)

(C3)

where the time-dependent constant C(T ) is determined solely by the choice of boundaryconditions To show (C3) we first change variables with v =

radicav thereby rewriting (C1)

in the following symmetric form

parttu = uprime +radicav (C4a)

parttv =radicauprimeprime + vprime (C4b)

Equations (C4ab) then imply

1

2partt

int

D

(uprime)2 =

int

D

uprime(uprimeprime +radicavprime) =

int

D

radicauprimevprime +

1

2

int

partD

(uprime)2 (C5a)

1

2partt

int

D

(v)2 =

int

D

v(radicauprimeprime + vprime) = minus

int

D

radicauprimevprime +

1

2

int

partD

(v2 + 2radicauprimev) (C5b)

Here vvprime and uprimeuprimeprime have been expressed as exact derivatives and then integrated to boundaryterms the second equation employs an extra integration by parts and with only one spacedimension

int

partDdenotes a difference of endpoint evaluations Addition of Eqs (C5ab) gives

1

2partt

int

D

[

v2 + (uprime)2]

=1

2

int

partD

[

v2 + (uprime)2 + 2radicauprimev

]

(C6)

Substitutions with the identities

[

v2 + (uprime)2]

=1

2

[

(v + uprime)2 + (v minus uprime)2]

2uprimev =1

2

[

(v + uprime)2 minus (v minus uprime)2]

(C7)

26

and replacements to recover the original variable v = vradica yield

1

2partt

int

D

[

av2 + (uprime)2]

=1 +

radica

4

int

partD

(radicav + uprime)2 +

1minusradica

4

int

partD

(radicav minus uprime)2 (C8)

From (C8) we deduce that the time-dependent constant C(T ) in (C3) must satisfy

∣

∣

∣

∣

∣

∣

1 +

int T

0

[

1+radica

2

int

partD(radicav + uprime)2 + 1minus

radica

2

int

partD(radicav minus uprime)2

]

dt

uprime(0 middot)2 + av(0 middot)2

∣

∣

∣

∣

∣

∣

le C(T ) (C9)

For periodic boundary conditions we may choose C(T ) = 1 Moreover if a ge 1 anduprime = minusradic

av is specified at partD+ then uprime(t middot)2 + av(t middot)2 decaysStill working on a single interval (subdomain) we now consider the semi-discrete scheme

for (C4) i e (57) with all nonlinear source terms dropped and with v replaced by vradica

Derivation of a formula analogous to (C8) is our first step toward establishing L2 stabilityof the semi-discrete scheme While (57) features vectors for example u(t) taking values atthe Legendre-Gauss-Lobatto nodal points here we work with the numerical solution as apolynomial for example uh(t x) These two representations are related by the Lagrange in-terpolating polynomials for the nodal set here taken to span both the space of test functionsand the space of basis functions Our scheme is

int

Dk

ψparttuh =

int

Dk

ψ(Qh +radicavh) (C10a)

int

Dk

ξparttvh = minusint

Dk

ξprime(radicaQh + vh) +

int

partDk

ξ(radicaQlowast + vlowast) (C10b)

int

Dk

ϕQh =

int

Dk

ϕuprimeh +

int

partDk

ϕ (ulowast minus uh) (C10c)

where ψ ξ and ϕ are polynomial test functions These test functions are arbitrary exceptthat they must be degree-N polynomials In (C10) the variables uh vh and Qh should alsocarry a superscript k but we have suppressed this Derivation of a formula analogous to(C8) is complicated by the fact that Qh is not evolved Nevertheless at a given instant t wecan assemble Qh from (C10c)

Mimicking the calculation (C5b) from the continuum case we first use (C10b) with ξ = vhto write

1

2partt

int

Dk

v2h = minusint

Dk

(radicaQh + vh)v

primeh +

int

partDk

(radicaQlowast + vlowast)vh

= minusint

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2(radicaQlowast + vlowast)vh minus v2h

]

(C11)

The right-hand side of (C5a) is analogous to

1

2partt

int

Dk

Q2h =

int

Dk

QhparttQh (C12)

However since Qh is not evolved the term parttQh must be given a suitable interpretationOn the right hand side of (C10c) only uh u

primeh and u

lowast necessarily depend on time since the

27

test function ϕ need not be time-dependent Furthermore ulowast is explicitly given as a linearcombination of uh as seen in Eq (C21c) below Choosing ϕ = ℓj taking the time derivativeof (C10c) and appealing to the commutivity of mixed partial derivatives we therefore arriveat

int

Dk

ℓjparttQh =

int

Dk

ℓj(parttuh)prime +

int

partDk

ℓj(

(parttu)lowast minus parttuh

)

(C13)

where (parttu)lowast depends on parttuh in precisely the same way that ulowast depends on uh We have

written ℓj rather than ϕ in the last equation to emphasize that the result also holds for anylinear combination of ℓj (for example ϕ) and even for time-dependent combinations SinceQh is itself such a combination we obtain

1

2partt

int

Dk

Q2h =

int

Dk

Qh(parttuh)prime +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

Qh(Qprimeh +

radicavprimeh) +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2((parttu)lowast minus parttuh)Qh +Q2

h

]

(C14)

Addition of (C11) and (C14) gives

1

2partt

int

Dk

(Q2h + v2h) =

1

2

int

partDk

[

Q2h minus v2h + 2(

radicaQlowast + vlowast)vh + 2((parttu)

lowast minus parttuh)Qh

]

(C15)

the aforementioned analog of (C8) This formula holds on a single subdomain Dk and we

now combine multiple copies of it one for each value of k

2 Analysis for multiple intervals

To facilitate combination of (C15) over all k we change notation At every subdomaininterface Ik+12 equiv partDk cap partDk+1 let the superscripts L and R denote field values respectivelytaken from the left and right Then the fields evaluated at Ik+12 which belong to D

k will beuLk+12 v

Lk+12 and Q

Lk+12 while those belonging to D

k+1 will be uRk+12 vRk+12 and Q

Rk+12

However at Ikminus12 the values taken from D

k are uRkminus12 vRkminus12 and QR

kminus12 Note that we

have also replaced the subscript h denoting a numerical solution with k plusmn 12 denotingthe location of the endpoint value of the numerical solution With this notation we define

∆Lα =

1

2

[

(QLα)

2 minus (vLα)2]

+(radic

aQlowastα + vlowastα

)

vLα +[

(parttuα)lowast minus parttu

Lα

]

QLα (C16)

and similarly for ∆Rα The same numerical fluxes appear in both ∆L

α and ∆Rα (ie each

numerical flux takes the same value on either side of an interface) whence fluxes like Qlowastα do

not carry an L or R superscript In terms of these definitions (C15) becomes

1

2partt

int

Dk

(Q2h + v2h) = ∆L

k+12 minus∆Rkminus12 (C17)

28

Summation over all Dk yields

1

2partt

kmaxsum

k=1

int

Dk

(Q2h + v2h) =

kmaxminus1sum

k=1

(

∆Lk+12 minus∆R

k+12

)

+∆Lkmax+12 minus∆R

12

=

kmaxminus1sum

k=1

(

∆Lh minus∆R

h

)∣

∣

Ik+12 +∆Lkmax+12 minus∆R

12 (C18)

We have reverted to h-notation denoting the numerical solution since the LR superscriptsindicate unambiguously the relevant domain used for evaluation at Ik+12

We again seek an estimate of the form

kmaxsum

k=1

(

Qh(T middot)2Dk + avh(T middot)2Dk

)

le C(T )kmaxsum

k=1

(

Qh(0 middot)2Dk + avh(0 middot)2Dk

)

(C19)

that is essentially the same as the one (C3) considered in the continuum case Assume thatthe chosen boundary conditions ensure ∆L

kmax+12 minus ∆R12 is bounded by a time-dependent

constant which does not depend on the numerical parameters N and h (subdomain width)Establishment of stability then amounts to showing that the remaining sum over interfaceterms in (C18) is non-positive whence this remaining sum is consistent with C(T ) le 1although the boundary conditions may give rise to a different bound In fact we will choosethe numerical fluxes such that each individual interface term is non-positive At interfaceIk+12 and in LR notation the jump and average of vh for example are

1

2

(

v+ + vminus)

equiv vh =1

2

(

vLk+12 + vRk+12

)

(C20a)

nminusvminus + n+v+ equiv[[

vh]]

= vLk+12 minus vRk+12 (C20b)

Consider numerical fluxes of the form

Qlowast = Qh minusτQ2

[[

Qh

]]

(C21a)

vlowast = vh minusτv2

[[

vh]]

(C21b)

ulowast = uh minusτu2

[[

uh]]

(C21c)

(parttu)lowast = parttuh minus

τu2

[[

parttuh]]

(C21d)

where (C21c) induces (C21d) and where the penalty parameters τu τv and τQ are realnumbers The fluxes defined in (58) correspond to τu = 1 τv = 1 +

radica and τQ = 0 In

terms of these quantities the kth interface contribution in (C18) is

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

parttuh]]

minus τu2

[[

parttuh]][[

Qh

]]

(C22)

where we have suppressed the k dependence of the right-hand side Now consider the term[[

parttuh]]

Because parttuh and Qh+radicavh are both polynomials of degree N Eq (C10a) implies

29

τu

τ Q

minus1 0 1minus1

minus05

0

05

1

τu

minus1 0 1minus1

minus05

0

05

1

FIG 6 Stable evolutions for the model system For fixed τv = 10minus6 and τv = 1 +radic2

respectively the left and right plots depict stable choices (determined empirically) of τu and τQfor the linear model system (C1) The stable regions are colored black but the jagged edges result

from the discretization of the (τu τQ)-plane

the vector equation parttu = Q +radicav that is pointwise equivalence on the nodal points of

Dk which in turn implies

[[

parttuh]]

=[[

Qh+radicavh]]

Upon substituting this identity into thelast equation we arrive at an expression which features only vh and Qh

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

Qh +radicavh]]

minus τu2

[[

Qh +radicavh]][[

Qh

]]

(C23)

The identities vh[[

vh]]

= 12

[[

v2h]]

and[[

Qh +radicavh]]

=[[

Qh

]]

+radica[[

vh]]

then simplify(C23) to

(∆Lh minus∆R

h )∣

∣

Ik+12 = minusτv2

[[

vh]]2 minus

radica(τu + τQ)

2

[[

Qh

]][[

vh]]

minus τu2

[[

Qh

]]2 (C24)

The role of a penalty parameter is now clear Positive values of τv penalize jumps in vhthrough a negative contribution to the energy Likewise positive values of τu penalizejumps in Qh through a negative contribution to the energy However because the sign of[[

Qh

]][[

vh]]

can be positive or negative only the choice τQ = minusτu yields an expression for(∆L

h minus ∆Rh )|Ik+12 which is manifestly negative for τu ge 0 and τv ge 0 A simple estimate

based on Youngrsquos inequality with ε (that is 2αβ le εminus1α2 + εβ2 where α β ge 0 and ε gt 0)shows that for τQ = 0 the choice τv ge aτu4 also yields stability

Figure 6 depicts certain choices of stable penalty parameters for the linear model systemevolved to tfinal = 1000 (with a = 2 N = 10 and ∆t ≃ 00553) as determined empiricallywith simulations similar to those described in Sec IVA The left plot corresponds to asmall τv = 10minus6 for which the choice τu = 1 τQ = 0 is not stable as expected from thetheoretical analysis However the right plot corresponds to τv = 1 +

radica for which τu = 1

τQ = 0 is stable Motivated by the numerical flux choices (3842) used for the BSSN system

30

(4) we have (as mentioned above) set τu = 1 τv = 1 +radica and τQ = 0 in simulations of

the nonlinear model (53) For the nonlinear model system (53) the theoretically motivatedchoice τQ = minusτu also yields numerically stable evolutions when τu ge 0 and τv ge 0

For the nonlinear systems (4) and (53) we do not attempt a formal stability proofNevertheless the results of this appendix have served as a guide for our choices of penaltyparameters For the BSSN system (4) u v and Q are block indices [cf Eq (6)] Similarto the model problem we have penalized Q with τu = 1 with τv chosen large enough toheuristically overcome the cross-terms of indefinite size that arise from τQ = 0 (we interpretequations like τu = 1 componentwise) An analogous choice ldquoτQ = minusτurdquo for the BSSNsystem might be possible but would be considerably more complicated Indeed such achoice likely entails a matrix of penalty parameters but we do not give the details here

[1] F Pretorius Evolution of Binary Black-Hole Spacetimes Phys Rev Lett 95 121101 (2005)

(4 pages)

[2] F Pretorius Simulation of binary black hole spacetimes with a harmonic evolution scheme

Class Quantum Grav 23 S529-S552 (2006)

[3] M Campanelli C O Lousto P Marronetti and Y Zlochower Accurate Evolutions of Or-

biting Black-Hole Binaries without Excision Phys Rev Lett 96 111101 (2006) (4 pages)

[4] J G Baker J Centrella D-I Choi M Koppitz and J van Meter Gravitational-Wave

Extraction from an Inspiraling Configuration of Merging Black Holes Phys Rev Lett 96

111102 (2006) (4 pages)

[5] M Campanelli C O Lousto and Y Zlochower Last orbit of binary black holes Phys Rev D

73 061501(R) (2006) (5 pages)

[6] F Herrmann I Hinder D Shoemaker and P Laguna Unequal mass binary black hole plunges

and gravitational recoil Class Quantum Grav 24 S33-S42 (2007)

[7] P Diener F Herrmann D Pollney E Schnetter E Seidel R Takahashi J Thornburg and

J Ventrella Accurate Evolution of Orbiting Binary Black Holes Phys Rev Lett 96 121101

(2006) (4 pages)

[8] M A Scheel H P Pfeiffer L Lindblom L E Kidder O Rinne and S A Teukolsky Solving

Einsteinrsquos Equations With Dual Coordinate Frames Phys Rev D 74 104006 (2006)

[9] U Sperhake Binary black-hole evolutions of excision and puncture data Phys Rev D 76

104015 (2007) (20 pages)

[10] B Brugmann J A Gonzalez M Hannam S Husa U Sperhake and W Tichy Calibration

of moving puncture simulations Phys Rev D 77 024027 (2008) (25 pages)

[11] P Marronetti W Tichy B Brugmann J Gonzalez M Hannam S Husa and U Sperhake

Binary black holes on a budget simulations using workstations Class Quantum Grav 24

S43-S58 (2007)

[12] Z B Etienne J A Faber Y T Liu S L Shapiro and T W Baumgarte Filling the holes

Evolving excised binary black hole initial data with puncture techniques Phys Rev D 76

101503(R) (2007) (5 pages)

[13] B Szilagyi D Pollney L Rezzolla J Thornburg and J Winicour An explicit harmonic code

for black-hole evolution using excision Class Quantum Grav 24 S275-S293 (2007)

[14] L Boyle M Kesden and S Nissanke Binary-Black-Hole Merger Symmetry and the Spin

Expansion Phys Rev Lett 100 151101 (2008) (4 pages)

31

[15] M A Scheel M Boyle T Chu L E Kidder K D Matthews and H P Pfeiffer High-

accuracy waveforms for binary black hole inspiral merger and ringdown Phys Rev D 79

024003 (2009) (14 pages)

[16] M Hannam Status of black-hole-binary simulations for gravitational-wave detection

Class Quantum Grav 26 114001 (2009) (21 pages)

[17] I Hinder The current status of binary black hole simulations in numerical relativity

Class Quantum Grav 27 114004 (2010) (19 pages)

[18] B P Abbott et al LIGO The Laser Interferometer Gravitational-Wave Observatory

Rep Prog Phys 72 076901 (2009) (25 pages)

[19] F Acernese et al The status of VIRGO Class Quantum Grav 23 S63-S69 (2006)

[20] F Acernese et al Virgo status Class Quantum Grav 25 184001 (2008) (9 pages)

[21] R Arnowitt S Deser and C W Misner The Dynamics of General Relativity in Gravitation

an introduction to current research edited by L Witten (Wiley New York 1962) Chapter 7

227-265 Also available as gr-qc0405109

[22] J W York Jr Kinematics and Dynamics of General Relativity in Sources of Gravitational

Radiation edited by L L Smarr (Cambridge University Press Cambridge 1979)

[23] L E Kidder M A Scheel and S A Teukolsky Extending the lifetime of 3D black hole

computations with a new hyperbolic system of evolution equations Phys Rev D 64 064017

(2001) (13 pages)

[24] G Calabrese J Pullin O Sarbach and M Tiglio Convergence and stability in numerical

relativity Phys Rev D 66 041501(R) (2002) (4 pages)

[25] S Frittelli and O Reula On the Newtonian Limit of General Relativity Commun Math

Phys 166 221-235 (1994)

[26] M Shibata and T Nakamura Evolution of three-dimensional gravitational waves Harmonic

slicing case Phys Rev D 52 5428-5444 (1995)

[27] Y Choquet-Bruhat and J W York Geometrical well posed systems for the Einstein equations

C R Acad Sci Paris t 321 Serie I 1089-1095 (1995)

[28] A Abrahams A Anderson Y Choquet-Bruhat and J W York Jr Einstein and Yang-Mills

Theories in Hyperbolic Form without Gauge Fixing Phys Rev Lett 75 3377-3381 (1995)

[29] C Bona J Masso E Seidel and J Stela New Formalism for Numerical Relativity

Phys Rev Lett 75 600-603 (1995)

[30] M H P M van Putten and D M Eardley Nonlinear wave equations for relativity

Phys Rev D 53 3056-3063 (1996)

[31] S Frittelli and O A Reula First-Order Symmetric Hyperbolic Einstein Equations with Ar-

bitrary Fixed Gauge Phys Rev Lett 76 4667-4670 (1996)

[32] H Friedrich Hyperbolic reductions for Einsteinrsquos equations Class Quantum Grav 13 1451-

1469 (1996)

[33] F B Estabrook R S Robinson H D Wahlquist Hyperbolic equations for vacuum gravity

using special orthonormal frames Class Quantum Grav 14 1237-1247 (1997)

[34] M S Iriondo E O Leguizamon and O A Reula Einsteinrsquos Equations in Ashtekarrsquos Vari-

ables Constitute a Symmetric Hyperbolic System Phys Rev Lett 79 4732-4735 (1997)

[35] A Anderson Y Choquet-Bruhat and J W York Einstein-Bianchi hyperbolic system for

general relativity Topol Meth Nonlinear Anal 10 353-373 (1997)

[36] T W Baumgarte and S L Shapiro Numerical integration of Einsteinrsquos field equations

Phys Rev D 59 024007 (1998) (7 pages)

[37] M A G Bonilla Symmetric hyperbolic systems for Bianchi equations Class Quantum

23

and the corresponding left eigenvectors are

x1 = (0 0 0 0 0 gθθ 2grr 0 0) (A3a)

x2 =

(

0 0 0 grr2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3b)

x3 =

(

grrλ 0 0 0

2

χminus 1

2grrminus 1

gθθ 0 0

)

(A3c)

xplusmn4 =

(

0 0plusmnradic

2αgrrχ

0 0 0 0 1 0

)

(A3d)

xplusmn5 =

(

0∓ 3radicgrrχ

plusmn2

radic

grrχ 2grr

1

χminus 1

grr1

gθθ 0 0

)

(A3e)

xplusmn6 =

(

minus3

4

grrλ 0plusmn α

radicλgrr

(2αχminus λ) 0 0minus βr

8(βrgrr ∓radicλgrr)

minus βrgrr

4gθθ(βrgrr ∓radicλgrr)

αχ

(2αχminus λ)plusmnradic

grrλ

)

(A3f)

where for example xplusmn5 A(u) = microplusmn5 x

plusmn5 Assuming that grr gθθ χ and α are everywhere strictly

positive the eigenvalues are real and the eigenvectors are linearly independent providedthat (15) holds These eigenvectors are easily extended to eigenvectors of A(u) e g asxplusmn6 rarr (01times5 x

plusmn6 ) Then for example the characteristic field

Xplusmn6 equiv (01times5 x

plusmn6 )W = xplusmn6WvQ (A4)

and similarly Xplusmnj = xplusmnj WvQ for j = 4 5 and Xk = xkWvQ for k = 1 2 3 The characteristic

speeds for these fields are microk and microplusmnj With this convention the speeds listed in Table I

correspond to the Xk and Xplusmnj in (14)

Appendix B Schwarzschild solution in conformal Kerr-Schild coordinates

In Kerr-Schild coordinates here the system directly related to incoming Eddington-Finkelstein null coordinates the line element for the Schwarzschild solution reads

ds2 = minusα2dt2 + (1 + 2MR)(dR + βRdt)2 +R2dθ2 +R2 sin2 θdφ2 (B1)

where R is the area radius α = (1 + 2MR)minus12 is the lapse and βR = 2M(R + 2M) isthe shift vector The physical spatial metric gab is the spatial part of this line element

To define the corresponding solution to the BSSN system we use equation gab = χgab todefine the following relationship between line elements

dr2 + r2(dθ2 + sin2 θdφ2) = χ[(1 + 2MR)dR2 +R2dθ2 +R2 sin2 θdφ2] (B2)

so that

χ

(

1 +2M

R

)(

dR

dr

)2

= 1 χR2 = r2 (B3)

Then we have(

1 +2M

R

)12dR

R=dr

r (B4)

24

with integration yielding

r =R

4

(

1 +

radic

1 +2M

R

)2

e2minus2radic

1+2MR (B5)

where the constant of integration has been chosen so that the R r rarr infin limits are consistentThe second relation in (B3) shows that

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR =rArr χminus4 =2eradic

1+2MRminus1

1 +radic

1 + 2MR (B6)

The extrinsic curvature tensor is specified by the expression forK given in (B9h) the identityK = KR

R + 2Kθθ and

Kθθ =

(

1 +2M

R

)minus122M

R2 (B7)

Since KRR = Kr

r we compute that

Krr = K minus 2Kθ

θ = minus(

1 +2M

R

)minus12(R +M

R + 2M

)

2M

R2 (B8)

Next since Krr = grrKrr = χminus1Kr

r we have Krr = Arr+

13grrK This implies Arr = Kr

r minus 13K

from which we get (B9g) In all we have

α =

(

1 +2M

R

)minus12

(B9a)

βr = βR dr

dR= χ12

(

1 +2M

R

)minus122M

R(B9b)

grr = 1 (B9c)

gθθ = r2 = χR2 (B9d)

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR (B9e)

Br = 0 (B9f)

Arr = minus(

1 +2M

R

)minus124M

3R2

(

2R + 3M

R + 2M

)

(B9g)

K =

(

1 +2M

R

)minus32(

1 +3M

R

)

2M

R2(B9h)

Γr = minus2

r= minus 2

χ12R (B9i)

To differentiate these expressions with respect to r we use the identity

dR

dr= χminus12

(

1 +2M

R

)minus12

(B10)

along with the chain rule

25

Appendix C Stability of the model system

The following stability analysis for the model system (53) has been inspired by [78 79]After dropping all nonlinear source terms the system (53) becomes

parttu = uprime + av (C1a)

parttv = uprimeprime + vprime (C1b)

This section analyzes the stability of (C1) considering both the continuum system itself aswell as its semi-discrete dG approximation The latter analysis offers some insight into theempirically observed stability of our dG scheme for the spherically reduced BSSN equations

1 Analysis for a single interval

Throughout we work with the L2 inner product and norm

(f g)D =

int

D

fg fD =radic

(f f)D (C2)

where D is the spatial coordinate interval (here D may represent a subdomain Dk or the

whole domain Ω) and we have suppressed all integration measures For the continuummodel we will establish the following estimate

uprime(T middot)2D+ av(T middot)2

Dle C(T )

(

uprime(0 middot)2D+ av(0 middot)2

D

)

(C3)

where the time-dependent constant C(T ) is determined solely by the choice of boundaryconditions To show (C3) we first change variables with v =

radicav thereby rewriting (C1)

in the following symmetric form

parttu = uprime +radicav (C4a)

parttv =radicauprimeprime + vprime (C4b)

Equations (C4ab) then imply

1

2partt

int

D

(uprime)2 =

int

D

uprime(uprimeprime +radicavprime) =

int

D

radicauprimevprime +

1

2

int

partD

(uprime)2 (C5a)

1

2partt

int

D

(v)2 =

int

D

v(radicauprimeprime + vprime) = minus

int

D

radicauprimevprime +

1

2

int

partD

(v2 + 2radicauprimev) (C5b)

Here vvprime and uprimeuprimeprime have been expressed as exact derivatives and then integrated to boundaryterms the second equation employs an extra integration by parts and with only one spacedimension

int

partDdenotes a difference of endpoint evaluations Addition of Eqs (C5ab) gives

1

2partt

int

D

[

v2 + (uprime)2]

=1

2

int

partD

[

v2 + (uprime)2 + 2radicauprimev

]

(C6)

Substitutions with the identities

[

v2 + (uprime)2]

=1

2

[

(v + uprime)2 + (v minus uprime)2]

2uprimev =1

2

[

(v + uprime)2 minus (v minus uprime)2]

(C7)

26

and replacements to recover the original variable v = vradica yield

1

2partt

int

D

[

av2 + (uprime)2]

=1 +

radica

4

int

partD

(radicav + uprime)2 +

1minusradica

4

int

partD

(radicav minus uprime)2 (C8)

From (C8) we deduce that the time-dependent constant C(T ) in (C3) must satisfy

∣

∣

∣

∣

∣

∣

1 +

int T

0

[

1+radica

2

int

partD(radicav + uprime)2 + 1minus

radica

2

int

partD(radicav minus uprime)2

]

dt

uprime(0 middot)2 + av(0 middot)2

∣

∣

∣

∣

∣

∣

le C(T ) (C9)

For periodic boundary conditions we may choose C(T ) = 1 Moreover if a ge 1 anduprime = minusradic

av is specified at partD+ then uprime(t middot)2 + av(t middot)2 decaysStill working on a single interval (subdomain) we now consider the semi-discrete scheme

for (C4) i e (57) with all nonlinear source terms dropped and with v replaced by vradica

Derivation of a formula analogous to (C8) is our first step toward establishing L2 stabilityof the semi-discrete scheme While (57) features vectors for example u(t) taking values atthe Legendre-Gauss-Lobatto nodal points here we work with the numerical solution as apolynomial for example uh(t x) These two representations are related by the Lagrange in-terpolating polynomials for the nodal set here taken to span both the space of test functionsand the space of basis functions Our scheme is

int

Dk

ψparttuh =

int

Dk

ψ(Qh +radicavh) (C10a)

int

Dk

ξparttvh = minusint

Dk

ξprime(radicaQh + vh) +

int

partDk

ξ(radicaQlowast + vlowast) (C10b)

int

Dk

ϕQh =

int

Dk

ϕuprimeh +

int

partDk

ϕ (ulowast minus uh) (C10c)

where ψ ξ and ϕ are polynomial test functions These test functions are arbitrary exceptthat they must be degree-N polynomials In (C10) the variables uh vh and Qh should alsocarry a superscript k but we have suppressed this Derivation of a formula analogous to(C8) is complicated by the fact that Qh is not evolved Nevertheless at a given instant t wecan assemble Qh from (C10c)

Mimicking the calculation (C5b) from the continuum case we first use (C10b) with ξ = vhto write

1

2partt

int

Dk

v2h = minusint

Dk

(radicaQh + vh)v

primeh +

int

partDk

(radicaQlowast + vlowast)vh

= minusint

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2(radicaQlowast + vlowast)vh minus v2h

]

(C11)

The right-hand side of (C5a) is analogous to

1

2partt

int

Dk

Q2h =

int

Dk

QhparttQh (C12)

However since Qh is not evolved the term parttQh must be given a suitable interpretationOn the right hand side of (C10c) only uh u

primeh and u

lowast necessarily depend on time since the

27

test function ϕ need not be time-dependent Furthermore ulowast is explicitly given as a linearcombination of uh as seen in Eq (C21c) below Choosing ϕ = ℓj taking the time derivativeof (C10c) and appealing to the commutivity of mixed partial derivatives we therefore arriveat

int

Dk

ℓjparttQh =

int

Dk

ℓj(parttuh)prime +

int

partDk

ℓj(

(parttu)lowast minus parttuh

)

(C13)

where (parttu)lowast depends on parttuh in precisely the same way that ulowast depends on uh We have

written ℓj rather than ϕ in the last equation to emphasize that the result also holds for anylinear combination of ℓj (for example ϕ) and even for time-dependent combinations SinceQh is itself such a combination we obtain

1

2partt

int

Dk

Q2h =

int

Dk

Qh(parttuh)prime +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

Qh(Qprimeh +

radicavprimeh) +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2((parttu)lowast minus parttuh)Qh +Q2

h

]

(C14)

Addition of (C11) and (C14) gives

1

2partt

int

Dk

(Q2h + v2h) =

1

2

int

partDk

[

Q2h minus v2h + 2(

radicaQlowast + vlowast)vh + 2((parttu)

lowast minus parttuh)Qh

]

(C15)

the aforementioned analog of (C8) This formula holds on a single subdomain Dk and we

now combine multiple copies of it one for each value of k

2 Analysis for multiple intervals

To facilitate combination of (C15) over all k we change notation At every subdomaininterface Ik+12 equiv partDk cap partDk+1 let the superscripts L and R denote field values respectivelytaken from the left and right Then the fields evaluated at Ik+12 which belong to D

k will beuLk+12 v

Lk+12 and Q

Lk+12 while those belonging to D

k+1 will be uRk+12 vRk+12 and Q

Rk+12

However at Ikminus12 the values taken from D

k are uRkminus12 vRkminus12 and QR

kminus12 Note that we

have also replaced the subscript h denoting a numerical solution with k plusmn 12 denotingthe location of the endpoint value of the numerical solution With this notation we define

∆Lα =

1

2

[

(QLα)

2 minus (vLα)2]

+(radic

aQlowastα + vlowastα

)

vLα +[

(parttuα)lowast minus parttu

Lα

]

QLα (C16)

and similarly for ∆Rα The same numerical fluxes appear in both ∆L

α and ∆Rα (ie each

numerical flux takes the same value on either side of an interface) whence fluxes like Qlowastα do

not carry an L or R superscript In terms of these definitions (C15) becomes

1

2partt

int

Dk

(Q2h + v2h) = ∆L

k+12 minus∆Rkminus12 (C17)

28

Summation over all Dk yields

1

2partt

kmaxsum

k=1

int

Dk

(Q2h + v2h) =

kmaxminus1sum

k=1

(

∆Lk+12 minus∆R

k+12

)

+∆Lkmax+12 minus∆R

12

=

kmaxminus1sum

k=1

(

∆Lh minus∆R

h

)∣

∣

Ik+12 +∆Lkmax+12 minus∆R

12 (C18)

We have reverted to h-notation denoting the numerical solution since the LR superscriptsindicate unambiguously the relevant domain used for evaluation at Ik+12

We again seek an estimate of the form

kmaxsum

k=1

(

Qh(T middot)2Dk + avh(T middot)2Dk

)

le C(T )kmaxsum

k=1

(

Qh(0 middot)2Dk + avh(0 middot)2Dk

)

(C19)

that is essentially the same as the one (C3) considered in the continuum case Assume thatthe chosen boundary conditions ensure ∆L

kmax+12 minus ∆R12 is bounded by a time-dependent

constant which does not depend on the numerical parameters N and h (subdomain width)Establishment of stability then amounts to showing that the remaining sum over interfaceterms in (C18) is non-positive whence this remaining sum is consistent with C(T ) le 1although the boundary conditions may give rise to a different bound In fact we will choosethe numerical fluxes such that each individual interface term is non-positive At interfaceIk+12 and in LR notation the jump and average of vh for example are

1

2

(

v+ + vminus)

equiv vh =1

2

(

vLk+12 + vRk+12

)

(C20a)

nminusvminus + n+v+ equiv[[

vh]]

= vLk+12 minus vRk+12 (C20b)

Consider numerical fluxes of the form

Qlowast = Qh minusτQ2

[[

Qh

]]

(C21a)

vlowast = vh minusτv2

[[

vh]]

(C21b)

ulowast = uh minusτu2

[[

uh]]

(C21c)

(parttu)lowast = parttuh minus

τu2

[[

parttuh]]

(C21d)

where (C21c) induces (C21d) and where the penalty parameters τu τv and τQ are realnumbers The fluxes defined in (58) correspond to τu = 1 τv = 1 +

radica and τQ = 0 In

terms of these quantities the kth interface contribution in (C18) is

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

parttuh]]

minus τu2

[[

parttuh]][[

Qh

]]

(C22)

where we have suppressed the k dependence of the right-hand side Now consider the term[[

parttuh]]

Because parttuh and Qh+radicavh are both polynomials of degree N Eq (C10a) implies

29

τu

τ Q

minus1 0 1minus1

minus05

0

05

1

τu

minus1 0 1minus1

minus05

0

05

1

FIG 6 Stable evolutions for the model system For fixed τv = 10minus6 and τv = 1 +radic2

respectively the left and right plots depict stable choices (determined empirically) of τu and τQfor the linear model system (C1) The stable regions are colored black but the jagged edges result

from the discretization of the (τu τQ)-plane

the vector equation parttu = Q +radicav that is pointwise equivalence on the nodal points of

Dk which in turn implies

[[

parttuh]]

=[[

Qh+radicavh]]

Upon substituting this identity into thelast equation we arrive at an expression which features only vh and Qh

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

Qh +radicavh]]

minus τu2

[[

Qh +radicavh]][[

Qh

]]

(C23)

The identities vh[[

vh]]

= 12

[[

v2h]]

and[[

Qh +radicavh]]

=[[

Qh

]]

+radica[[

vh]]

then simplify(C23) to

(∆Lh minus∆R

h )∣

∣

Ik+12 = minusτv2

[[

vh]]2 minus

radica(τu + τQ)

2

[[

Qh

]][[

vh]]

minus τu2

[[

Qh

]]2 (C24)

The role of a penalty parameter is now clear Positive values of τv penalize jumps in vhthrough a negative contribution to the energy Likewise positive values of τu penalizejumps in Qh through a negative contribution to the energy However because the sign of[[

Qh

]][[

vh]]

can be positive or negative only the choice τQ = minusτu yields an expression for(∆L

h minus ∆Rh )|Ik+12 which is manifestly negative for τu ge 0 and τv ge 0 A simple estimate

based on Youngrsquos inequality with ε (that is 2αβ le εminus1α2 + εβ2 where α β ge 0 and ε gt 0)shows that for τQ = 0 the choice τv ge aτu4 also yields stability

Figure 6 depicts certain choices of stable penalty parameters for the linear model systemevolved to tfinal = 1000 (with a = 2 N = 10 and ∆t ≃ 00553) as determined empiricallywith simulations similar to those described in Sec IVA The left plot corresponds to asmall τv = 10minus6 for which the choice τu = 1 τQ = 0 is not stable as expected from thetheoretical analysis However the right plot corresponds to τv = 1 +

radica for which τu = 1

τQ = 0 is stable Motivated by the numerical flux choices (3842) used for the BSSN system

30

(4) we have (as mentioned above) set τu = 1 τv = 1 +radica and τQ = 0 in simulations of

the nonlinear model (53) For the nonlinear model system (53) the theoretically motivatedchoice τQ = minusτu also yields numerically stable evolutions when τu ge 0 and τv ge 0

For the nonlinear systems (4) and (53) we do not attempt a formal stability proofNevertheless the results of this appendix have served as a guide for our choices of penaltyparameters For the BSSN system (4) u v and Q are block indices [cf Eq (6)] Similarto the model problem we have penalized Q with τu = 1 with τv chosen large enough toheuristically overcome the cross-terms of indefinite size that arise from τQ = 0 (we interpretequations like τu = 1 componentwise) An analogous choice ldquoτQ = minusτurdquo for the BSSNsystem might be possible but would be considerably more complicated Indeed such achoice likely entails a matrix of penalty parameters but we do not give the details here

[1] F Pretorius Evolution of Binary Black-Hole Spacetimes Phys Rev Lett 95 121101 (2005)

(4 pages)

[2] F Pretorius Simulation of binary black hole spacetimes with a harmonic evolution scheme

Class Quantum Grav 23 S529-S552 (2006)

[3] M Campanelli C O Lousto P Marronetti and Y Zlochower Accurate Evolutions of Or-

biting Black-Hole Binaries without Excision Phys Rev Lett 96 111101 (2006) (4 pages)

[4] J G Baker J Centrella D-I Choi M Koppitz and J van Meter Gravitational-Wave

Extraction from an Inspiraling Configuration of Merging Black Holes Phys Rev Lett 96

111102 (2006) (4 pages)

[5] M Campanelli C O Lousto and Y Zlochower Last orbit of binary black holes Phys Rev D

73 061501(R) (2006) (5 pages)

[6] F Herrmann I Hinder D Shoemaker and P Laguna Unequal mass binary black hole plunges

and gravitational recoil Class Quantum Grav 24 S33-S42 (2007)

[7] P Diener F Herrmann D Pollney E Schnetter E Seidel R Takahashi J Thornburg and

J Ventrella Accurate Evolution of Orbiting Binary Black Holes Phys Rev Lett 96 121101

(2006) (4 pages)

[8] M A Scheel H P Pfeiffer L Lindblom L E Kidder O Rinne and S A Teukolsky Solving

Einsteinrsquos Equations With Dual Coordinate Frames Phys Rev D 74 104006 (2006)

[9] U Sperhake Binary black-hole evolutions of excision and puncture data Phys Rev D 76

104015 (2007) (20 pages)

[10] B Brugmann J A Gonzalez M Hannam S Husa U Sperhake and W Tichy Calibration

of moving puncture simulations Phys Rev D 77 024027 (2008) (25 pages)

[11] P Marronetti W Tichy B Brugmann J Gonzalez M Hannam S Husa and U Sperhake

Binary black holes on a budget simulations using workstations Class Quantum Grav 24

S43-S58 (2007)

[12] Z B Etienne J A Faber Y T Liu S L Shapiro and T W Baumgarte Filling the holes

Evolving excised binary black hole initial data with puncture techniques Phys Rev D 76

101503(R) (2007) (5 pages)

[13] B Szilagyi D Pollney L Rezzolla J Thornburg and J Winicour An explicit harmonic code

for black-hole evolution using excision Class Quantum Grav 24 S275-S293 (2007)

[14] L Boyle M Kesden and S Nissanke Binary-Black-Hole Merger Symmetry and the Spin

Expansion Phys Rev Lett 100 151101 (2008) (4 pages)

31

[15] M A Scheel M Boyle T Chu L E Kidder K D Matthews and H P Pfeiffer High-

accuracy waveforms for binary black hole inspiral merger and ringdown Phys Rev D 79

024003 (2009) (14 pages)

[16] M Hannam Status of black-hole-binary simulations for gravitational-wave detection

Class Quantum Grav 26 114001 (2009) (21 pages)

[17] I Hinder The current status of binary black hole simulations in numerical relativity

Class Quantum Grav 27 114004 (2010) (19 pages)

[18] B P Abbott et al LIGO The Laser Interferometer Gravitational-Wave Observatory

Rep Prog Phys 72 076901 (2009) (25 pages)

[19] F Acernese et al The status of VIRGO Class Quantum Grav 23 S63-S69 (2006)

[20] F Acernese et al Virgo status Class Quantum Grav 25 184001 (2008) (9 pages)

[21] R Arnowitt S Deser and C W Misner The Dynamics of General Relativity in Gravitation

an introduction to current research edited by L Witten (Wiley New York 1962) Chapter 7

227-265 Also available as gr-qc0405109

[22] J W York Jr Kinematics and Dynamics of General Relativity in Sources of Gravitational

Radiation edited by L L Smarr (Cambridge University Press Cambridge 1979)

[23] L E Kidder M A Scheel and S A Teukolsky Extending the lifetime of 3D black hole

computations with a new hyperbolic system of evolution equations Phys Rev D 64 064017

(2001) (13 pages)

[24] G Calabrese J Pullin O Sarbach and M Tiglio Convergence and stability in numerical

relativity Phys Rev D 66 041501(R) (2002) (4 pages)

[25] S Frittelli and O Reula On the Newtonian Limit of General Relativity Commun Math

Phys 166 221-235 (1994)

[26] M Shibata and T Nakamura Evolution of three-dimensional gravitational waves Harmonic

slicing case Phys Rev D 52 5428-5444 (1995)

[27] Y Choquet-Bruhat and J W York Geometrical well posed systems for the Einstein equations

C R Acad Sci Paris t 321 Serie I 1089-1095 (1995)

[28] A Abrahams A Anderson Y Choquet-Bruhat and J W York Jr Einstein and Yang-Mills

Theories in Hyperbolic Form without Gauge Fixing Phys Rev Lett 75 3377-3381 (1995)

[29] C Bona J Masso E Seidel and J Stela New Formalism for Numerical Relativity

Phys Rev Lett 75 600-603 (1995)

[30] M H P M van Putten and D M Eardley Nonlinear wave equations for relativity

Phys Rev D 53 3056-3063 (1996)

[31] S Frittelli and O A Reula First-Order Symmetric Hyperbolic Einstein Equations with Ar-

bitrary Fixed Gauge Phys Rev Lett 76 4667-4670 (1996)

[32] H Friedrich Hyperbolic reductions for Einsteinrsquos equations Class Quantum Grav 13 1451-

1469 (1996)

[33] F B Estabrook R S Robinson H D Wahlquist Hyperbolic equations for vacuum gravity

using special orthonormal frames Class Quantum Grav 14 1237-1247 (1997)

[34] M S Iriondo E O Leguizamon and O A Reula Einsteinrsquos Equations in Ashtekarrsquos Vari-

ables Constitute a Symmetric Hyperbolic System Phys Rev Lett 79 4732-4735 (1997)

[35] A Anderson Y Choquet-Bruhat and J W York Einstein-Bianchi hyperbolic system for

general relativity Topol Meth Nonlinear Anal 10 353-373 (1997)

[36] T W Baumgarte and S L Shapiro Numerical integration of Einsteinrsquos field equations

Phys Rev D 59 024007 (1998) (7 pages)

[37] M A G Bonilla Symmetric hyperbolic systems for Bianchi equations Class Quantum

24

with integration yielding

r =R

4

(

1 +

radic

1 +2M

R

)2

e2minus2radic

1+2MR (B5)

where the constant of integration has been chosen so that the R r rarr infin limits are consistentThe second relation in (B3) shows that

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR =rArr χminus4 =2eradic

1+2MRminus1

1 +radic

1 + 2MR (B6)

The extrinsic curvature tensor is specified by the expression forK given in (B9h) the identityK = KR

R + 2Kθθ and

Kθθ =

(

1 +2M

R

)minus122M

R2 (B7)

Since KRR = Kr

r we compute that

Krr = K minus 2Kθ

θ = minus(

1 +2M

R

)minus12(R +M

R + 2M

)

2M

R2 (B8)

Next since Krr = grrKrr = χminus1Kr

r we have Krr = Arr+

13grrK This implies Arr = Kr

r minus 13K

from which we get (B9g) In all we have

α =

(

1 +2M

R

)minus12

(B9a)

βr = βR dr

dR= χ12

(

1 +2M

R

)minus122M

R(B9b)

grr = 1 (B9c)

gθθ = r2 = χR2 (B9d)

χ =1

16

(

1 +

radic

1 +2M

R

)4

e4minus4radic

1+2MR (B9e)

Br = 0 (B9f)

Arr = minus(

1 +2M

R

)minus124M

3R2

(

2R + 3M

R + 2M

)

(B9g)

K =

(

1 +2M

R

)minus32(

1 +3M

R

)

2M

R2(B9h)

Γr = minus2

r= minus 2

χ12R (B9i)

To differentiate these expressions with respect to r we use the identity

dR

dr= χminus12

(

1 +2M

R

)minus12

(B10)

along with the chain rule

25

Appendix C Stability of the model system

The following stability analysis for the model system (53) has been inspired by [78 79]After dropping all nonlinear source terms the system (53) becomes

parttu = uprime + av (C1a)

parttv = uprimeprime + vprime (C1b)

This section analyzes the stability of (C1) considering both the continuum system itself aswell as its semi-discrete dG approximation The latter analysis offers some insight into theempirically observed stability of our dG scheme for the spherically reduced BSSN equations

1 Analysis for a single interval

Throughout we work with the L2 inner product and norm

(f g)D =

int

D

fg fD =radic

(f f)D (C2)

where D is the spatial coordinate interval (here D may represent a subdomain Dk or the

whole domain Ω) and we have suppressed all integration measures For the continuummodel we will establish the following estimate

uprime(T middot)2D+ av(T middot)2

Dle C(T )

(

uprime(0 middot)2D+ av(0 middot)2

D

)

(C3)

where the time-dependent constant C(T ) is determined solely by the choice of boundaryconditions To show (C3) we first change variables with v =

radicav thereby rewriting (C1)

in the following symmetric form

parttu = uprime +radicav (C4a)

parttv =radicauprimeprime + vprime (C4b)

Equations (C4ab) then imply

1

2partt

int

D

(uprime)2 =

int

D

uprime(uprimeprime +radicavprime) =

int

D

radicauprimevprime +

1

2

int

partD

(uprime)2 (C5a)

1

2partt

int

D

(v)2 =

int

D

v(radicauprimeprime + vprime) = minus

int

D

radicauprimevprime +

1

2

int

partD

(v2 + 2radicauprimev) (C5b)

Here vvprime and uprimeuprimeprime have been expressed as exact derivatives and then integrated to boundaryterms the second equation employs an extra integration by parts and with only one spacedimension

int

partDdenotes a difference of endpoint evaluations Addition of Eqs (C5ab) gives

1

2partt

int

D

[

v2 + (uprime)2]

=1

2

int

partD

[

v2 + (uprime)2 + 2radicauprimev

]

(C6)

Substitutions with the identities

[

v2 + (uprime)2]

=1

2

[

(v + uprime)2 + (v minus uprime)2]

2uprimev =1

2

[

(v + uprime)2 minus (v minus uprime)2]

(C7)

26

and replacements to recover the original variable v = vradica yield

1

2partt

int

D

[

av2 + (uprime)2]

=1 +

radica

4

int

partD

(radicav + uprime)2 +

1minusradica

4

int

partD

(radicav minus uprime)2 (C8)

From (C8) we deduce that the time-dependent constant C(T ) in (C3) must satisfy

∣

∣

∣

∣

∣

∣

1 +

int T

0

[

1+radica

2

int

partD(radicav + uprime)2 + 1minus

radica

2

int

partD(radicav minus uprime)2

]

dt

uprime(0 middot)2 + av(0 middot)2

∣

∣

∣

∣

∣

∣

le C(T ) (C9)

For periodic boundary conditions we may choose C(T ) = 1 Moreover if a ge 1 anduprime = minusradic

av is specified at partD+ then uprime(t middot)2 + av(t middot)2 decaysStill working on a single interval (subdomain) we now consider the semi-discrete scheme

for (C4) i e (57) with all nonlinear source terms dropped and with v replaced by vradica

Derivation of a formula analogous to (C8) is our first step toward establishing L2 stabilityof the semi-discrete scheme While (57) features vectors for example u(t) taking values atthe Legendre-Gauss-Lobatto nodal points here we work with the numerical solution as apolynomial for example uh(t x) These two representations are related by the Lagrange in-terpolating polynomials for the nodal set here taken to span both the space of test functionsand the space of basis functions Our scheme is

int

Dk

ψparttuh =

int

Dk

ψ(Qh +radicavh) (C10a)

int

Dk

ξparttvh = minusint

Dk

ξprime(radicaQh + vh) +

int

partDk

ξ(radicaQlowast + vlowast) (C10b)

int

Dk

ϕQh =

int

Dk

ϕuprimeh +

int

partDk

ϕ (ulowast minus uh) (C10c)

where ψ ξ and ϕ are polynomial test functions These test functions are arbitrary exceptthat they must be degree-N polynomials In (C10) the variables uh vh and Qh should alsocarry a superscript k but we have suppressed this Derivation of a formula analogous to(C8) is complicated by the fact that Qh is not evolved Nevertheless at a given instant t wecan assemble Qh from (C10c)

Mimicking the calculation (C5b) from the continuum case we first use (C10b) with ξ = vhto write

1

2partt

int

Dk

v2h = minusint

Dk

(radicaQh + vh)v

primeh +

int

partDk

(radicaQlowast + vlowast)vh

= minusint

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2(radicaQlowast + vlowast)vh minus v2h

]

(C11)

The right-hand side of (C5a) is analogous to

1

2partt

int

Dk

Q2h =

int

Dk

QhparttQh (C12)

However since Qh is not evolved the term parttQh must be given a suitable interpretationOn the right hand side of (C10c) only uh u

primeh and u

lowast necessarily depend on time since the

27

test function ϕ need not be time-dependent Furthermore ulowast is explicitly given as a linearcombination of uh as seen in Eq (C21c) below Choosing ϕ = ℓj taking the time derivativeof (C10c) and appealing to the commutivity of mixed partial derivatives we therefore arriveat

int

Dk

ℓjparttQh =

int

Dk

ℓj(parttuh)prime +

int

partDk

ℓj(

(parttu)lowast minus parttuh

)

(C13)

where (parttu)lowast depends on parttuh in precisely the same way that ulowast depends on uh We have

written ℓj rather than ϕ in the last equation to emphasize that the result also holds for anylinear combination of ℓj (for example ϕ) and even for time-dependent combinations SinceQh is itself such a combination we obtain

1

2partt

int

Dk

Q2h =

int

Dk

Qh(parttuh)prime +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

Qh(Qprimeh +

radicavprimeh) +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2((parttu)lowast minus parttuh)Qh +Q2

h

]

(C14)

Addition of (C11) and (C14) gives

1

2partt

int

Dk

(Q2h + v2h) =

1

2

int

partDk

[

Q2h minus v2h + 2(

radicaQlowast + vlowast)vh + 2((parttu)

lowast minus parttuh)Qh

]

(C15)

the aforementioned analog of (C8) This formula holds on a single subdomain Dk and we

now combine multiple copies of it one for each value of k

2 Analysis for multiple intervals

To facilitate combination of (C15) over all k we change notation At every subdomaininterface Ik+12 equiv partDk cap partDk+1 let the superscripts L and R denote field values respectivelytaken from the left and right Then the fields evaluated at Ik+12 which belong to D

k will beuLk+12 v

Lk+12 and Q

Lk+12 while those belonging to D

k+1 will be uRk+12 vRk+12 and Q

Rk+12

However at Ikminus12 the values taken from D

k are uRkminus12 vRkminus12 and QR

kminus12 Note that we

have also replaced the subscript h denoting a numerical solution with k plusmn 12 denotingthe location of the endpoint value of the numerical solution With this notation we define

∆Lα =

1

2

[

(QLα)

2 minus (vLα)2]

+(radic

aQlowastα + vlowastα

)

vLα +[

(parttuα)lowast minus parttu

Lα

]

QLα (C16)

and similarly for ∆Rα The same numerical fluxes appear in both ∆L

α and ∆Rα (ie each

numerical flux takes the same value on either side of an interface) whence fluxes like Qlowastα do

not carry an L or R superscript In terms of these definitions (C15) becomes

1

2partt

int

Dk

(Q2h + v2h) = ∆L

k+12 minus∆Rkminus12 (C17)

28

Summation over all Dk yields

1

2partt

kmaxsum

k=1

int

Dk

(Q2h + v2h) =

kmaxminus1sum

k=1

(

∆Lk+12 minus∆R

k+12

)

+∆Lkmax+12 minus∆R

12

=

kmaxminus1sum

k=1

(

∆Lh minus∆R

h

)∣

∣

Ik+12 +∆Lkmax+12 minus∆R

12 (C18)

We have reverted to h-notation denoting the numerical solution since the LR superscriptsindicate unambiguously the relevant domain used for evaluation at Ik+12

We again seek an estimate of the form

kmaxsum

k=1

(

Qh(T middot)2Dk + avh(T middot)2Dk

)

le C(T )kmaxsum

k=1

(

Qh(0 middot)2Dk + avh(0 middot)2Dk

)

(C19)

that is essentially the same as the one (C3) considered in the continuum case Assume thatthe chosen boundary conditions ensure ∆L

kmax+12 minus ∆R12 is bounded by a time-dependent

constant which does not depend on the numerical parameters N and h (subdomain width)Establishment of stability then amounts to showing that the remaining sum over interfaceterms in (C18) is non-positive whence this remaining sum is consistent with C(T ) le 1although the boundary conditions may give rise to a different bound In fact we will choosethe numerical fluxes such that each individual interface term is non-positive At interfaceIk+12 and in LR notation the jump and average of vh for example are

1

2

(

v+ + vminus)

equiv vh =1

2

(

vLk+12 + vRk+12

)

(C20a)

nminusvminus + n+v+ equiv[[

vh]]

= vLk+12 minus vRk+12 (C20b)

Consider numerical fluxes of the form

Qlowast = Qh minusτQ2

[[

Qh

]]

(C21a)

vlowast = vh minusτv2

[[

vh]]

(C21b)

ulowast = uh minusτu2

[[

uh]]

(C21c)

(parttu)lowast = parttuh minus

τu2

[[

parttuh]]

(C21d)

where (C21c) induces (C21d) and where the penalty parameters τu τv and τQ are realnumbers The fluxes defined in (58) correspond to τu = 1 τv = 1 +

radica and τQ = 0 In

terms of these quantities the kth interface contribution in (C18) is

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

parttuh]]

minus τu2

[[

parttuh]][[

Qh

]]

(C22)

where we have suppressed the k dependence of the right-hand side Now consider the term[[

parttuh]]

Because parttuh and Qh+radicavh are both polynomials of degree N Eq (C10a) implies

29

τu

τ Q

minus1 0 1minus1

minus05

0

05

1

τu

minus1 0 1minus1

minus05

0

05

1

FIG 6 Stable evolutions for the model system For fixed τv = 10minus6 and τv = 1 +radic2

respectively the left and right plots depict stable choices (determined empirically) of τu and τQfor the linear model system (C1) The stable regions are colored black but the jagged edges result

from the discretization of the (τu τQ)-plane

the vector equation parttu = Q +radicav that is pointwise equivalence on the nodal points of

Dk which in turn implies

[[

parttuh]]

=[[

Qh+radicavh]]

Upon substituting this identity into thelast equation we arrive at an expression which features only vh and Qh

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

Qh +radicavh]]

minus τu2

[[

Qh +radicavh]][[

Qh

]]

(C23)

The identities vh[[

vh]]

= 12

[[

v2h]]

and[[

Qh +radicavh]]

=[[

Qh

]]

+radica[[

vh]]

then simplify(C23) to

(∆Lh minus∆R

h )∣

∣

Ik+12 = minusτv2

[[

vh]]2 minus

radica(τu + τQ)

2

[[

Qh

]][[

vh]]

minus τu2

[[

Qh

]]2 (C24)

The role of a penalty parameter is now clear Positive values of τv penalize jumps in vhthrough a negative contribution to the energy Likewise positive values of τu penalizejumps in Qh through a negative contribution to the energy However because the sign of[[

Qh

]][[

vh]]

can be positive or negative only the choice τQ = minusτu yields an expression for(∆L

h minus ∆Rh )|Ik+12 which is manifestly negative for τu ge 0 and τv ge 0 A simple estimate

based on Youngrsquos inequality with ε (that is 2αβ le εminus1α2 + εβ2 where α β ge 0 and ε gt 0)shows that for τQ = 0 the choice τv ge aτu4 also yields stability

Figure 6 depicts certain choices of stable penalty parameters for the linear model systemevolved to tfinal = 1000 (with a = 2 N = 10 and ∆t ≃ 00553) as determined empiricallywith simulations similar to those described in Sec IVA The left plot corresponds to asmall τv = 10minus6 for which the choice τu = 1 τQ = 0 is not stable as expected from thetheoretical analysis However the right plot corresponds to τv = 1 +

radica for which τu = 1

τQ = 0 is stable Motivated by the numerical flux choices (3842) used for the BSSN system

30

(4) we have (as mentioned above) set τu = 1 τv = 1 +radica and τQ = 0 in simulations of

the nonlinear model (53) For the nonlinear model system (53) the theoretically motivatedchoice τQ = minusτu also yields numerically stable evolutions when τu ge 0 and τv ge 0

For the nonlinear systems (4) and (53) we do not attempt a formal stability proofNevertheless the results of this appendix have served as a guide for our choices of penaltyparameters For the BSSN system (4) u v and Q are block indices [cf Eq (6)] Similarto the model problem we have penalized Q with τu = 1 with τv chosen large enough toheuristically overcome the cross-terms of indefinite size that arise from τQ = 0 (we interpretequations like τu = 1 componentwise) An analogous choice ldquoτQ = minusτurdquo for the BSSNsystem might be possible but would be considerably more complicated Indeed such achoice likely entails a matrix of penalty parameters but we do not give the details here

[1] F Pretorius Evolution of Binary Black-Hole Spacetimes Phys Rev Lett 95 121101 (2005)

(4 pages)

[2] F Pretorius Simulation of binary black hole spacetimes with a harmonic evolution scheme

Class Quantum Grav 23 S529-S552 (2006)

[3] M Campanelli C O Lousto P Marronetti and Y Zlochower Accurate Evolutions of Or-

biting Black-Hole Binaries without Excision Phys Rev Lett 96 111101 (2006) (4 pages)

[4] J G Baker J Centrella D-I Choi M Koppitz and J van Meter Gravitational-Wave

Extraction from an Inspiraling Configuration of Merging Black Holes Phys Rev Lett 96

111102 (2006) (4 pages)

[5] M Campanelli C O Lousto and Y Zlochower Last orbit of binary black holes Phys Rev D

73 061501(R) (2006) (5 pages)

[6] F Herrmann I Hinder D Shoemaker and P Laguna Unequal mass binary black hole plunges

and gravitational recoil Class Quantum Grav 24 S33-S42 (2007)

[7] P Diener F Herrmann D Pollney E Schnetter E Seidel R Takahashi J Thornburg and

J Ventrella Accurate Evolution of Orbiting Binary Black Holes Phys Rev Lett 96 121101

(2006) (4 pages)

[8] M A Scheel H P Pfeiffer L Lindblom L E Kidder O Rinne and S A Teukolsky Solving

Einsteinrsquos Equations With Dual Coordinate Frames Phys Rev D 74 104006 (2006)

[9] U Sperhake Binary black-hole evolutions of excision and puncture data Phys Rev D 76

104015 (2007) (20 pages)

[10] B Brugmann J A Gonzalez M Hannam S Husa U Sperhake and W Tichy Calibration

of moving puncture simulations Phys Rev D 77 024027 (2008) (25 pages)

[11] P Marronetti W Tichy B Brugmann J Gonzalez M Hannam S Husa and U Sperhake

Binary black holes on a budget simulations using workstations Class Quantum Grav 24

S43-S58 (2007)

[12] Z B Etienne J A Faber Y T Liu S L Shapiro and T W Baumgarte Filling the holes

Evolving excised binary black hole initial data with puncture techniques Phys Rev D 76

101503(R) (2007) (5 pages)

[13] B Szilagyi D Pollney L Rezzolla J Thornburg and J Winicour An explicit harmonic code

for black-hole evolution using excision Class Quantum Grav 24 S275-S293 (2007)

[14] L Boyle M Kesden and S Nissanke Binary-Black-Hole Merger Symmetry and the Spin

Expansion Phys Rev Lett 100 151101 (2008) (4 pages)

31

[15] M A Scheel M Boyle T Chu L E Kidder K D Matthews and H P Pfeiffer High-

accuracy waveforms for binary black hole inspiral merger and ringdown Phys Rev D 79

024003 (2009) (14 pages)

[16] M Hannam Status of black-hole-binary simulations for gravitational-wave detection

Class Quantum Grav 26 114001 (2009) (21 pages)

[17] I Hinder The current status of binary black hole simulations in numerical relativity

Class Quantum Grav 27 114004 (2010) (19 pages)

[18] B P Abbott et al LIGO The Laser Interferometer Gravitational-Wave Observatory

Rep Prog Phys 72 076901 (2009) (25 pages)

[19] F Acernese et al The status of VIRGO Class Quantum Grav 23 S63-S69 (2006)

[20] F Acernese et al Virgo status Class Quantum Grav 25 184001 (2008) (9 pages)

[21] R Arnowitt S Deser and C W Misner The Dynamics of General Relativity in Gravitation

an introduction to current research edited by L Witten (Wiley New York 1962) Chapter 7

227-265 Also available as gr-qc0405109

[22] J W York Jr Kinematics and Dynamics of General Relativity in Sources of Gravitational

Radiation edited by L L Smarr (Cambridge University Press Cambridge 1979)

[23] L E Kidder M A Scheel and S A Teukolsky Extending the lifetime of 3D black hole

computations with a new hyperbolic system of evolution equations Phys Rev D 64 064017

(2001) (13 pages)

[24] G Calabrese J Pullin O Sarbach and M Tiglio Convergence and stability in numerical

relativity Phys Rev D 66 041501(R) (2002) (4 pages)

[25] S Frittelli and O Reula On the Newtonian Limit of General Relativity Commun Math

Phys 166 221-235 (1994)

[26] M Shibata and T Nakamura Evolution of three-dimensional gravitational waves Harmonic

slicing case Phys Rev D 52 5428-5444 (1995)

[27] Y Choquet-Bruhat and J W York Geometrical well posed systems for the Einstein equations

C R Acad Sci Paris t 321 Serie I 1089-1095 (1995)

[28] A Abrahams A Anderson Y Choquet-Bruhat and J W York Jr Einstein and Yang-Mills

Theories in Hyperbolic Form without Gauge Fixing Phys Rev Lett 75 3377-3381 (1995)

[29] C Bona J Masso E Seidel and J Stela New Formalism for Numerical Relativity

Phys Rev Lett 75 600-603 (1995)

[30] M H P M van Putten and D M Eardley Nonlinear wave equations for relativity

Phys Rev D 53 3056-3063 (1996)

[31] S Frittelli and O A Reula First-Order Symmetric Hyperbolic Einstein Equations with Ar-

bitrary Fixed Gauge Phys Rev Lett 76 4667-4670 (1996)

[32] H Friedrich Hyperbolic reductions for Einsteinrsquos equations Class Quantum Grav 13 1451-

1469 (1996)

[33] F B Estabrook R S Robinson H D Wahlquist Hyperbolic equations for vacuum gravity

using special orthonormal frames Class Quantum Grav 14 1237-1247 (1997)

[34] M S Iriondo E O Leguizamon and O A Reula Einsteinrsquos Equations in Ashtekarrsquos Vari-

ables Constitute a Symmetric Hyperbolic System Phys Rev Lett 79 4732-4735 (1997)

[35] A Anderson Y Choquet-Bruhat and J W York Einstein-Bianchi hyperbolic system for

general relativity Topol Meth Nonlinear Anal 10 353-373 (1997)

[36] T W Baumgarte and S L Shapiro Numerical integration of Einsteinrsquos field equations

Phys Rev D 59 024007 (1998) (7 pages)

[37] M A G Bonilla Symmetric hyperbolic systems for Bianchi equations Class Quantum

25

Appendix C Stability of the model system

The following stability analysis for the model system (53) has been inspired by [78 79]After dropping all nonlinear source terms the system (53) becomes

parttu = uprime + av (C1a)

parttv = uprimeprime + vprime (C1b)

This section analyzes the stability of (C1) considering both the continuum system itself aswell as its semi-discrete dG approximation The latter analysis offers some insight into theempirically observed stability of our dG scheme for the spherically reduced BSSN equations

1 Analysis for a single interval

Throughout we work with the L2 inner product and norm

(f g)D =

int

D

fg fD =radic

(f f)D (C2)

where D is the spatial coordinate interval (here D may represent a subdomain Dk or the

whole domain Ω) and we have suppressed all integration measures For the continuummodel we will establish the following estimate

uprime(T middot)2D+ av(T middot)2

Dle C(T )

(

uprime(0 middot)2D+ av(0 middot)2

D

)

(C3)

where the time-dependent constant C(T ) is determined solely by the choice of boundaryconditions To show (C3) we first change variables with v =

radicav thereby rewriting (C1)

in the following symmetric form

parttu = uprime +radicav (C4a)

parttv =radicauprimeprime + vprime (C4b)

Equations (C4ab) then imply

1

2partt

int

D

(uprime)2 =

int

D

uprime(uprimeprime +radicavprime) =

int

D

radicauprimevprime +

1

2

int

partD

(uprime)2 (C5a)

1

2partt

int

D

(v)2 =

int

D

v(radicauprimeprime + vprime) = minus

int

D

radicauprimevprime +

1

2

int

partD

(v2 + 2radicauprimev) (C5b)

Here vvprime and uprimeuprimeprime have been expressed as exact derivatives and then integrated to boundaryterms the second equation employs an extra integration by parts and with only one spacedimension

int

partDdenotes a difference of endpoint evaluations Addition of Eqs (C5ab) gives

1

2partt

int

D

[

v2 + (uprime)2]

=1

2

int

partD

[

v2 + (uprime)2 + 2radicauprimev

]

(C6)

Substitutions with the identities

[

v2 + (uprime)2]

=1

2

[

(v + uprime)2 + (v minus uprime)2]

2uprimev =1

2

[

(v + uprime)2 minus (v minus uprime)2]

(C7)

26

and replacements to recover the original variable v = vradica yield

1

2partt

int

D

[

av2 + (uprime)2]

=1 +

radica

4

int

partD

(radicav + uprime)2 +

1minusradica

4

int

partD

(radicav minus uprime)2 (C8)

From (C8) we deduce that the time-dependent constant C(T ) in (C3) must satisfy

∣

∣

∣

∣

∣

∣

1 +

int T

0

[

1+radica

2

int

partD(radicav + uprime)2 + 1minus

radica

2

int

partD(radicav minus uprime)2

]

dt

uprime(0 middot)2 + av(0 middot)2

∣

∣

∣

∣

∣

∣

le C(T ) (C9)

For periodic boundary conditions we may choose C(T ) = 1 Moreover if a ge 1 anduprime = minusradic

av is specified at partD+ then uprime(t middot)2 + av(t middot)2 decaysStill working on a single interval (subdomain) we now consider the semi-discrete scheme

for (C4) i e (57) with all nonlinear source terms dropped and with v replaced by vradica

Derivation of a formula analogous to (C8) is our first step toward establishing L2 stabilityof the semi-discrete scheme While (57) features vectors for example u(t) taking values atthe Legendre-Gauss-Lobatto nodal points here we work with the numerical solution as apolynomial for example uh(t x) These two representations are related by the Lagrange in-terpolating polynomials for the nodal set here taken to span both the space of test functionsand the space of basis functions Our scheme is

int

Dk

ψparttuh =

int

Dk

ψ(Qh +radicavh) (C10a)

int

Dk

ξparttvh = minusint

Dk

ξprime(radicaQh + vh) +

int

partDk

ξ(radicaQlowast + vlowast) (C10b)

int

Dk

ϕQh =

int

Dk

ϕuprimeh +

int

partDk

ϕ (ulowast minus uh) (C10c)

where ψ ξ and ϕ are polynomial test functions These test functions are arbitrary exceptthat they must be degree-N polynomials In (C10) the variables uh vh and Qh should alsocarry a superscript k but we have suppressed this Derivation of a formula analogous to(C8) is complicated by the fact that Qh is not evolved Nevertheless at a given instant t wecan assemble Qh from (C10c)

Mimicking the calculation (C5b) from the continuum case we first use (C10b) with ξ = vhto write

1

2partt

int

Dk

v2h = minusint

Dk

(radicaQh + vh)v

primeh +

int

partDk

(radicaQlowast + vlowast)vh

= minusint

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2(radicaQlowast + vlowast)vh minus v2h

]

(C11)

The right-hand side of (C5a) is analogous to

1

2partt

int

Dk

Q2h =

int

Dk

QhparttQh (C12)

However since Qh is not evolved the term parttQh must be given a suitable interpretationOn the right hand side of (C10c) only uh u

primeh and u

lowast necessarily depend on time since the

27

test function ϕ need not be time-dependent Furthermore ulowast is explicitly given as a linearcombination of uh as seen in Eq (C21c) below Choosing ϕ = ℓj taking the time derivativeof (C10c) and appealing to the commutivity of mixed partial derivatives we therefore arriveat

int

Dk

ℓjparttQh =

int

Dk

ℓj(parttuh)prime +

int

partDk

ℓj(

(parttu)lowast minus parttuh

)

(C13)

where (parttu)lowast depends on parttuh in precisely the same way that ulowast depends on uh We have

written ℓj rather than ϕ in the last equation to emphasize that the result also holds for anylinear combination of ℓj (for example ϕ) and even for time-dependent combinations SinceQh is itself such a combination we obtain

1

2partt

int

Dk

Q2h =

int

Dk

Qh(parttuh)prime +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

Qh(Qprimeh +

radicavprimeh) +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2((parttu)lowast minus parttuh)Qh +Q2

h

]

(C14)

Addition of (C11) and (C14) gives

1

2partt

int

Dk

(Q2h + v2h) =

1

2

int

partDk

[

Q2h minus v2h + 2(

radicaQlowast + vlowast)vh + 2((parttu)

lowast minus parttuh)Qh

]

(C15)

the aforementioned analog of (C8) This formula holds on a single subdomain Dk and we

now combine multiple copies of it one for each value of k

2 Analysis for multiple intervals

To facilitate combination of (C15) over all k we change notation At every subdomaininterface Ik+12 equiv partDk cap partDk+1 let the superscripts L and R denote field values respectivelytaken from the left and right Then the fields evaluated at Ik+12 which belong to D

k will beuLk+12 v

Lk+12 and Q

Lk+12 while those belonging to D

k+1 will be uRk+12 vRk+12 and Q

Rk+12

However at Ikminus12 the values taken from D

k are uRkminus12 vRkminus12 and QR

kminus12 Note that we

have also replaced the subscript h denoting a numerical solution with k plusmn 12 denotingthe location of the endpoint value of the numerical solution With this notation we define

∆Lα =

1

2

[

(QLα)

2 minus (vLα)2]

+(radic

aQlowastα + vlowastα

)

vLα +[

(parttuα)lowast minus parttu

Lα

]

QLα (C16)

and similarly for ∆Rα The same numerical fluxes appear in both ∆L

α and ∆Rα (ie each

numerical flux takes the same value on either side of an interface) whence fluxes like Qlowastα do

not carry an L or R superscript In terms of these definitions (C15) becomes

1

2partt

int

Dk

(Q2h + v2h) = ∆L

k+12 minus∆Rkminus12 (C17)

28

Summation over all Dk yields

1

2partt

kmaxsum

k=1

int

Dk

(Q2h + v2h) =

kmaxminus1sum

k=1

(

∆Lk+12 minus∆R

k+12

)

+∆Lkmax+12 minus∆R

12

=

kmaxminus1sum

k=1

(

∆Lh minus∆R

h

)∣

∣

Ik+12 +∆Lkmax+12 minus∆R

12 (C18)

We have reverted to h-notation denoting the numerical solution since the LR superscriptsindicate unambiguously the relevant domain used for evaluation at Ik+12

We again seek an estimate of the form

kmaxsum

k=1

(

Qh(T middot)2Dk + avh(T middot)2Dk

)

le C(T )kmaxsum

k=1

(

Qh(0 middot)2Dk + avh(0 middot)2Dk

)

(C19)

that is essentially the same as the one (C3) considered in the continuum case Assume thatthe chosen boundary conditions ensure ∆L

kmax+12 minus ∆R12 is bounded by a time-dependent

constant which does not depend on the numerical parameters N and h (subdomain width)Establishment of stability then amounts to showing that the remaining sum over interfaceterms in (C18) is non-positive whence this remaining sum is consistent with C(T ) le 1although the boundary conditions may give rise to a different bound In fact we will choosethe numerical fluxes such that each individual interface term is non-positive At interfaceIk+12 and in LR notation the jump and average of vh for example are

1

2

(

v+ + vminus)

equiv vh =1

2

(

vLk+12 + vRk+12

)

(C20a)

nminusvminus + n+v+ equiv[[

vh]]

= vLk+12 minus vRk+12 (C20b)

Consider numerical fluxes of the form

Qlowast = Qh minusτQ2

[[

Qh

]]

(C21a)

vlowast = vh minusτv2

[[

vh]]

(C21b)

ulowast = uh minusτu2

[[

uh]]

(C21c)

(parttu)lowast = parttuh minus

τu2

[[

parttuh]]

(C21d)

where (C21c) induces (C21d) and where the penalty parameters τu τv and τQ are realnumbers The fluxes defined in (58) correspond to τu = 1 τv = 1 +

radica and τQ = 0 In

terms of these quantities the kth interface contribution in (C18) is

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

parttuh]]

minus τu2

[[

parttuh]][[

Qh

]]

(C22)

where we have suppressed the k dependence of the right-hand side Now consider the term[[

parttuh]]

Because parttuh and Qh+radicavh are both polynomials of degree N Eq (C10a) implies

29

τu

τ Q

minus1 0 1minus1

minus05

0

05

1

τu

minus1 0 1minus1

minus05

0

05

1

FIG 6 Stable evolutions for the model system For fixed τv = 10minus6 and τv = 1 +radic2

respectively the left and right plots depict stable choices (determined empirically) of τu and τQfor the linear model system (C1) The stable regions are colored black but the jagged edges result

from the discretization of the (τu τQ)-plane

the vector equation parttu = Q +radicav that is pointwise equivalence on the nodal points of

Dk which in turn implies

[[

parttuh]]

=[[

Qh+radicavh]]

Upon substituting this identity into thelast equation we arrive at an expression which features only vh and Qh

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

Qh +radicavh]]

minus τu2

[[

Qh +radicavh]][[

Qh

]]

(C23)

The identities vh[[

vh]]

= 12

[[

v2h]]

and[[

Qh +radicavh]]

=[[

Qh

]]

+radica[[

vh]]

then simplify(C23) to

(∆Lh minus∆R

h )∣

∣

Ik+12 = minusτv2

[[

vh]]2 minus

radica(τu + τQ)

2

[[

Qh

]][[

vh]]

minus τu2

[[

Qh

]]2 (C24)

The role of a penalty parameter is now clear Positive values of τv penalize jumps in vhthrough a negative contribution to the energy Likewise positive values of τu penalizejumps in Qh through a negative contribution to the energy However because the sign of[[

Qh

]][[

vh]]

can be positive or negative only the choice τQ = minusτu yields an expression for(∆L

h minus ∆Rh )|Ik+12 which is manifestly negative for τu ge 0 and τv ge 0 A simple estimate

based on Youngrsquos inequality with ε (that is 2αβ le εminus1α2 + εβ2 where α β ge 0 and ε gt 0)shows that for τQ = 0 the choice τv ge aτu4 also yields stability

Figure 6 depicts certain choices of stable penalty parameters for the linear model systemevolved to tfinal = 1000 (with a = 2 N = 10 and ∆t ≃ 00553) as determined empiricallywith simulations similar to those described in Sec IVA The left plot corresponds to asmall τv = 10minus6 for which the choice τu = 1 τQ = 0 is not stable as expected from thetheoretical analysis However the right plot corresponds to τv = 1 +

radica for which τu = 1

τQ = 0 is stable Motivated by the numerical flux choices (3842) used for the BSSN system

30

(4) we have (as mentioned above) set τu = 1 τv = 1 +radica and τQ = 0 in simulations of

the nonlinear model (53) For the nonlinear model system (53) the theoretically motivatedchoice τQ = minusτu also yields numerically stable evolutions when τu ge 0 and τv ge 0

For the nonlinear systems (4) and (53) we do not attempt a formal stability proofNevertheless the results of this appendix have served as a guide for our choices of penaltyparameters For the BSSN system (4) u v and Q are block indices [cf Eq (6)] Similarto the model problem we have penalized Q with τu = 1 with τv chosen large enough toheuristically overcome the cross-terms of indefinite size that arise from τQ = 0 (we interpretequations like τu = 1 componentwise) An analogous choice ldquoτQ = minusτurdquo for the BSSNsystem might be possible but would be considerably more complicated Indeed such achoice likely entails a matrix of penalty parameters but we do not give the details here

[1] F Pretorius Evolution of Binary Black-Hole Spacetimes Phys Rev Lett 95 121101 (2005)

(4 pages)

[2] F Pretorius Simulation of binary black hole spacetimes with a harmonic evolution scheme

Class Quantum Grav 23 S529-S552 (2006)

[3] M Campanelli C O Lousto P Marronetti and Y Zlochower Accurate Evolutions of Or-

biting Black-Hole Binaries without Excision Phys Rev Lett 96 111101 (2006) (4 pages)

[4] J G Baker J Centrella D-I Choi M Koppitz and J van Meter Gravitational-Wave

Extraction from an Inspiraling Configuration of Merging Black Holes Phys Rev Lett 96

111102 (2006) (4 pages)

[5] M Campanelli C O Lousto and Y Zlochower Last orbit of binary black holes Phys Rev D

73 061501(R) (2006) (5 pages)

[6] F Herrmann I Hinder D Shoemaker and P Laguna Unequal mass binary black hole plunges

and gravitational recoil Class Quantum Grav 24 S33-S42 (2007)

[7] P Diener F Herrmann D Pollney E Schnetter E Seidel R Takahashi J Thornburg and

J Ventrella Accurate Evolution of Orbiting Binary Black Holes Phys Rev Lett 96 121101

(2006) (4 pages)

[8] M A Scheel H P Pfeiffer L Lindblom L E Kidder O Rinne and S A Teukolsky Solving

Einsteinrsquos Equations With Dual Coordinate Frames Phys Rev D 74 104006 (2006)

[9] U Sperhake Binary black-hole evolutions of excision and puncture data Phys Rev D 76

104015 (2007) (20 pages)

[10] B Brugmann J A Gonzalez M Hannam S Husa U Sperhake and W Tichy Calibration

of moving puncture simulations Phys Rev D 77 024027 (2008) (25 pages)

[11] P Marronetti W Tichy B Brugmann J Gonzalez M Hannam S Husa and U Sperhake

Binary black holes on a budget simulations using workstations Class Quantum Grav 24

S43-S58 (2007)

[12] Z B Etienne J A Faber Y T Liu S L Shapiro and T W Baumgarte Filling the holes

Evolving excised binary black hole initial data with puncture techniques Phys Rev D 76

101503(R) (2007) (5 pages)

[13] B Szilagyi D Pollney L Rezzolla J Thornburg and J Winicour An explicit harmonic code

for black-hole evolution using excision Class Quantum Grav 24 S275-S293 (2007)

[14] L Boyle M Kesden and S Nissanke Binary-Black-Hole Merger Symmetry and the Spin

Expansion Phys Rev Lett 100 151101 (2008) (4 pages)

31

[15] M A Scheel M Boyle T Chu L E Kidder K D Matthews and H P Pfeiffer High-

accuracy waveforms for binary black hole inspiral merger and ringdown Phys Rev D 79

024003 (2009) (14 pages)

[16] M Hannam Status of black-hole-binary simulations for gravitational-wave detection

Class Quantum Grav 26 114001 (2009) (21 pages)

[17] I Hinder The current status of binary black hole simulations in numerical relativity

Class Quantum Grav 27 114004 (2010) (19 pages)

[18] B P Abbott et al LIGO The Laser Interferometer Gravitational-Wave Observatory

Rep Prog Phys 72 076901 (2009) (25 pages)

[19] F Acernese et al The status of VIRGO Class Quantum Grav 23 S63-S69 (2006)

[20] F Acernese et al Virgo status Class Quantum Grav 25 184001 (2008) (9 pages)

[21] R Arnowitt S Deser and C W Misner The Dynamics of General Relativity in Gravitation

an introduction to current research edited by L Witten (Wiley New York 1962) Chapter 7

227-265 Also available as gr-qc0405109

[22] J W York Jr Kinematics and Dynamics of General Relativity in Sources of Gravitational

Radiation edited by L L Smarr (Cambridge University Press Cambridge 1979)

[23] L E Kidder M A Scheel and S A Teukolsky Extending the lifetime of 3D black hole

computations with a new hyperbolic system of evolution equations Phys Rev D 64 064017

(2001) (13 pages)

[24] G Calabrese J Pullin O Sarbach and M Tiglio Convergence and stability in numerical

relativity Phys Rev D 66 041501(R) (2002) (4 pages)

[25] S Frittelli and O Reula On the Newtonian Limit of General Relativity Commun Math

Phys 166 221-235 (1994)

[26] M Shibata and T Nakamura Evolution of three-dimensional gravitational waves Harmonic

slicing case Phys Rev D 52 5428-5444 (1995)

[27] Y Choquet-Bruhat and J W York Geometrical well posed systems for the Einstein equations

C R Acad Sci Paris t 321 Serie I 1089-1095 (1995)

[28] A Abrahams A Anderson Y Choquet-Bruhat and J W York Jr Einstein and Yang-Mills

Theories in Hyperbolic Form without Gauge Fixing Phys Rev Lett 75 3377-3381 (1995)

[29] C Bona J Masso E Seidel and J Stela New Formalism for Numerical Relativity

Phys Rev Lett 75 600-603 (1995)

[30] M H P M van Putten and D M Eardley Nonlinear wave equations for relativity

Phys Rev D 53 3056-3063 (1996)

[31] S Frittelli and O A Reula First-Order Symmetric Hyperbolic Einstein Equations with Ar-

bitrary Fixed Gauge Phys Rev Lett 76 4667-4670 (1996)

[32] H Friedrich Hyperbolic reductions for Einsteinrsquos equations Class Quantum Grav 13 1451-

1469 (1996)

[33] F B Estabrook R S Robinson H D Wahlquist Hyperbolic equations for vacuum gravity

using special orthonormal frames Class Quantum Grav 14 1237-1247 (1997)

[34] M S Iriondo E O Leguizamon and O A Reula Einsteinrsquos Equations in Ashtekarrsquos Vari-

ables Constitute a Symmetric Hyperbolic System Phys Rev Lett 79 4732-4735 (1997)

[35] A Anderson Y Choquet-Bruhat and J W York Einstein-Bianchi hyperbolic system for

general relativity Topol Meth Nonlinear Anal 10 353-373 (1997)

[36] T W Baumgarte and S L Shapiro Numerical integration of Einsteinrsquos field equations

Phys Rev D 59 024007 (1998) (7 pages)

[37] M A G Bonilla Symmetric hyperbolic systems for Bianchi equations Class Quantum

26

and replacements to recover the original variable v = vradica yield

1

2partt

int

D

[

av2 + (uprime)2]

=1 +

radica

4

int

partD

(radicav + uprime)2 +

1minusradica

4

int

partD

(radicav minus uprime)2 (C8)

From (C8) we deduce that the time-dependent constant C(T ) in (C3) must satisfy

∣

∣

∣

∣

∣

∣

1 +

int T

0

[

1+radica

2

int

partD(radicav + uprime)2 + 1minus

radica

2

int

partD(radicav minus uprime)2

]

dt

uprime(0 middot)2 + av(0 middot)2

∣

∣

∣

∣

∣

∣

le C(T ) (C9)

For periodic boundary conditions we may choose C(T ) = 1 Moreover if a ge 1 anduprime = minusradic

av is specified at partD+ then uprime(t middot)2 + av(t middot)2 decaysStill working on a single interval (subdomain) we now consider the semi-discrete scheme

for (C4) i e (57) with all nonlinear source terms dropped and with v replaced by vradica

Derivation of a formula analogous to (C8) is our first step toward establishing L2 stabilityof the semi-discrete scheme While (57) features vectors for example u(t) taking values atthe Legendre-Gauss-Lobatto nodal points here we work with the numerical solution as apolynomial for example uh(t x) These two representations are related by the Lagrange in-terpolating polynomials for the nodal set here taken to span both the space of test functionsand the space of basis functions Our scheme is

int

Dk

ψparttuh =

int

Dk

ψ(Qh +radicavh) (C10a)

int

Dk

ξparttvh = minusint

Dk

ξprime(radicaQh + vh) +

int

partDk

ξ(radicaQlowast + vlowast) (C10b)

int

Dk

ϕQh =

int

Dk

ϕuprimeh +

int

partDk

ϕ (ulowast minus uh) (C10c)

where ψ ξ and ϕ are polynomial test functions These test functions are arbitrary exceptthat they must be degree-N polynomials In (C10) the variables uh vh and Qh should alsocarry a superscript k but we have suppressed this Derivation of a formula analogous to(C8) is complicated by the fact that Qh is not evolved Nevertheless at a given instant t wecan assemble Qh from (C10c)

Mimicking the calculation (C5b) from the continuum case we first use (C10b) with ξ = vhto write

1

2partt

int

Dk

v2h = minusint

Dk

(radicaQh + vh)v

primeh +

int

partDk

(radicaQlowast + vlowast)vh

= minusint

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2(radicaQlowast + vlowast)vh minus v2h

]

(C11)

The right-hand side of (C5a) is analogous to

1

2partt

int

Dk

Q2h =

int

Dk

QhparttQh (C12)

However since Qh is not evolved the term parttQh must be given a suitable interpretationOn the right hand side of (C10c) only uh u

primeh and u

lowast necessarily depend on time since the

27

test function ϕ need not be time-dependent Furthermore ulowast is explicitly given as a linearcombination of uh as seen in Eq (C21c) below Choosing ϕ = ℓj taking the time derivativeof (C10c) and appealing to the commutivity of mixed partial derivatives we therefore arriveat

int

Dk

ℓjparttQh =

int

Dk

ℓj(parttuh)prime +

int

partDk

ℓj(

(parttu)lowast minus parttuh

)

(C13)

where (parttu)lowast depends on parttuh in precisely the same way that ulowast depends on uh We have

written ℓj rather than ϕ in the last equation to emphasize that the result also holds for anylinear combination of ℓj (for example ϕ) and even for time-dependent combinations SinceQh is itself such a combination we obtain

1

2partt

int

Dk

Q2h =

int

Dk

Qh(parttuh)prime +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

Qh(Qprimeh +

radicavprimeh) +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2((parttu)lowast minus parttuh)Qh +Q2

h

]

(C14)

Addition of (C11) and (C14) gives

1

2partt

int

Dk

(Q2h + v2h) =

1

2

int

partDk

[

Q2h minus v2h + 2(

radicaQlowast + vlowast)vh + 2((parttu)

lowast minus parttuh)Qh

]

(C15)

the aforementioned analog of (C8) This formula holds on a single subdomain Dk and we

now combine multiple copies of it one for each value of k

2 Analysis for multiple intervals

To facilitate combination of (C15) over all k we change notation At every subdomaininterface Ik+12 equiv partDk cap partDk+1 let the superscripts L and R denote field values respectivelytaken from the left and right Then the fields evaluated at Ik+12 which belong to D

k will beuLk+12 v

Lk+12 and Q

Lk+12 while those belonging to D

k+1 will be uRk+12 vRk+12 and Q

Rk+12

However at Ikminus12 the values taken from D

k are uRkminus12 vRkminus12 and QR

kminus12 Note that we

have also replaced the subscript h denoting a numerical solution with k plusmn 12 denotingthe location of the endpoint value of the numerical solution With this notation we define

∆Lα =

1

2

[

(QLα)

2 minus (vLα)2]

+(radic

aQlowastα + vlowastα

)

vLα +[

(parttuα)lowast minus parttu

Lα

]

QLα (C16)

and similarly for ∆Rα The same numerical fluxes appear in both ∆L

α and ∆Rα (ie each

numerical flux takes the same value on either side of an interface) whence fluxes like Qlowastα do

not carry an L or R superscript In terms of these definitions (C15) becomes

1

2partt

int

Dk

(Q2h + v2h) = ∆L

k+12 minus∆Rkminus12 (C17)

28

Summation over all Dk yields

1

2partt

kmaxsum

k=1

int

Dk

(Q2h + v2h) =

kmaxminus1sum

k=1

(

∆Lk+12 minus∆R

k+12

)

+∆Lkmax+12 minus∆R

12

=

kmaxminus1sum

k=1

(

∆Lh minus∆R

h

)∣

∣

Ik+12 +∆Lkmax+12 minus∆R

12 (C18)

We have reverted to h-notation denoting the numerical solution since the LR superscriptsindicate unambiguously the relevant domain used for evaluation at Ik+12

We again seek an estimate of the form

kmaxsum

k=1

(

Qh(T middot)2Dk + avh(T middot)2Dk

)

le C(T )kmaxsum

k=1

(

Qh(0 middot)2Dk + avh(0 middot)2Dk

)

(C19)

that is essentially the same as the one (C3) considered in the continuum case Assume thatthe chosen boundary conditions ensure ∆L

kmax+12 minus ∆R12 is bounded by a time-dependent

constant which does not depend on the numerical parameters N and h (subdomain width)Establishment of stability then amounts to showing that the remaining sum over interfaceterms in (C18) is non-positive whence this remaining sum is consistent with C(T ) le 1although the boundary conditions may give rise to a different bound In fact we will choosethe numerical fluxes such that each individual interface term is non-positive At interfaceIk+12 and in LR notation the jump and average of vh for example are

1

2

(

v+ + vminus)

equiv vh =1

2

(

vLk+12 + vRk+12

)

(C20a)

nminusvminus + n+v+ equiv[[

vh]]

= vLk+12 minus vRk+12 (C20b)

Consider numerical fluxes of the form

Qlowast = Qh minusτQ2

[[

Qh

]]

(C21a)

vlowast = vh minusτv2

[[

vh]]

(C21b)

ulowast = uh minusτu2

[[

uh]]

(C21c)

(parttu)lowast = parttuh minus

τu2

[[

parttuh]]

(C21d)

where (C21c) induces (C21d) and where the penalty parameters τu τv and τQ are realnumbers The fluxes defined in (58) correspond to τu = 1 τv = 1 +

radica and τQ = 0 In

terms of these quantities the kth interface contribution in (C18) is

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

parttuh]]

minus τu2

[[

parttuh]][[

Qh

]]

(C22)

where we have suppressed the k dependence of the right-hand side Now consider the term[[

parttuh]]

Because parttuh and Qh+radicavh are both polynomials of degree N Eq (C10a) implies

29

τu

τ Q

minus1 0 1minus1

minus05

0

05

1

τu

minus1 0 1minus1

minus05

0

05

1

FIG 6 Stable evolutions for the model system For fixed τv = 10minus6 and τv = 1 +radic2

respectively the left and right plots depict stable choices (determined empirically) of τu and τQfor the linear model system (C1) The stable regions are colored black but the jagged edges result

from the discretization of the (τu τQ)-plane

the vector equation parttu = Q +radicav that is pointwise equivalence on the nodal points of

Dk which in turn implies

[[

parttuh]]

=[[

Qh+radicavh]]

Upon substituting this identity into thelast equation we arrive at an expression which features only vh and Qh

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

Qh +radicavh]]

minus τu2

[[

Qh +radicavh]][[

Qh

]]

(C23)

The identities vh[[

vh]]

= 12

[[

v2h]]

and[[

Qh +radicavh]]

=[[

Qh

]]

+radica[[

vh]]

then simplify(C23) to

(∆Lh minus∆R

h )∣

∣

Ik+12 = minusτv2

[[

vh]]2 minus

radica(τu + τQ)

2

[[

Qh

]][[

vh]]

minus τu2

[[

Qh

]]2 (C24)

The role of a penalty parameter is now clear Positive values of τv penalize jumps in vhthrough a negative contribution to the energy Likewise positive values of τu penalizejumps in Qh through a negative contribution to the energy However because the sign of[[

Qh

]][[

vh]]

can be positive or negative only the choice τQ = minusτu yields an expression for(∆L

h minus ∆Rh )|Ik+12 which is manifestly negative for τu ge 0 and τv ge 0 A simple estimate

based on Youngrsquos inequality with ε (that is 2αβ le εminus1α2 + εβ2 where α β ge 0 and ε gt 0)shows that for τQ = 0 the choice τv ge aτu4 also yields stability

Figure 6 depicts certain choices of stable penalty parameters for the linear model systemevolved to tfinal = 1000 (with a = 2 N = 10 and ∆t ≃ 00553) as determined empiricallywith simulations similar to those described in Sec IVA The left plot corresponds to asmall τv = 10minus6 for which the choice τu = 1 τQ = 0 is not stable as expected from thetheoretical analysis However the right plot corresponds to τv = 1 +

radica for which τu = 1

τQ = 0 is stable Motivated by the numerical flux choices (3842) used for the BSSN system

30

(4) we have (as mentioned above) set τu = 1 τv = 1 +radica and τQ = 0 in simulations of

the nonlinear model (53) For the nonlinear model system (53) the theoretically motivatedchoice τQ = minusτu also yields numerically stable evolutions when τu ge 0 and τv ge 0

For the nonlinear systems (4) and (53) we do not attempt a formal stability proofNevertheless the results of this appendix have served as a guide for our choices of penaltyparameters For the BSSN system (4) u v and Q are block indices [cf Eq (6)] Similarto the model problem we have penalized Q with τu = 1 with τv chosen large enough toheuristically overcome the cross-terms of indefinite size that arise from τQ = 0 (we interpretequations like τu = 1 componentwise) An analogous choice ldquoτQ = minusτurdquo for the BSSNsystem might be possible but would be considerably more complicated Indeed such achoice likely entails a matrix of penalty parameters but we do not give the details here

[1] F Pretorius Evolution of Binary Black-Hole Spacetimes Phys Rev Lett 95 121101 (2005)

(4 pages)

[2] F Pretorius Simulation of binary black hole spacetimes with a harmonic evolution scheme

Class Quantum Grav 23 S529-S552 (2006)

[3] M Campanelli C O Lousto P Marronetti and Y Zlochower Accurate Evolutions of Or-

biting Black-Hole Binaries without Excision Phys Rev Lett 96 111101 (2006) (4 pages)

[4] J G Baker J Centrella D-I Choi M Koppitz and J van Meter Gravitational-Wave

Extraction from an Inspiraling Configuration of Merging Black Holes Phys Rev Lett 96

111102 (2006) (4 pages)

[5] M Campanelli C O Lousto and Y Zlochower Last orbit of binary black holes Phys Rev D

73 061501(R) (2006) (5 pages)

[6] F Herrmann I Hinder D Shoemaker and P Laguna Unequal mass binary black hole plunges

and gravitational recoil Class Quantum Grav 24 S33-S42 (2007)

[7] P Diener F Herrmann D Pollney E Schnetter E Seidel R Takahashi J Thornburg and

J Ventrella Accurate Evolution of Orbiting Binary Black Holes Phys Rev Lett 96 121101

(2006) (4 pages)

[8] M A Scheel H P Pfeiffer L Lindblom L E Kidder O Rinne and S A Teukolsky Solving

Einsteinrsquos Equations With Dual Coordinate Frames Phys Rev D 74 104006 (2006)

[9] U Sperhake Binary black-hole evolutions of excision and puncture data Phys Rev D 76

104015 (2007) (20 pages)

[10] B Brugmann J A Gonzalez M Hannam S Husa U Sperhake and W Tichy Calibration

of moving puncture simulations Phys Rev D 77 024027 (2008) (25 pages)

[11] P Marronetti W Tichy B Brugmann J Gonzalez M Hannam S Husa and U Sperhake

Binary black holes on a budget simulations using workstations Class Quantum Grav 24

S43-S58 (2007)

[12] Z B Etienne J A Faber Y T Liu S L Shapiro and T W Baumgarte Filling the holes

Evolving excised binary black hole initial data with puncture techniques Phys Rev D 76

101503(R) (2007) (5 pages)

[13] B Szilagyi D Pollney L Rezzolla J Thornburg and J Winicour An explicit harmonic code

for black-hole evolution using excision Class Quantum Grav 24 S275-S293 (2007)

[14] L Boyle M Kesden and S Nissanke Binary-Black-Hole Merger Symmetry and the Spin

Expansion Phys Rev Lett 100 151101 (2008) (4 pages)

31

[15] M A Scheel M Boyle T Chu L E Kidder K D Matthews and H P Pfeiffer High-

accuracy waveforms for binary black hole inspiral merger and ringdown Phys Rev D 79

024003 (2009) (14 pages)

[16] M Hannam Status of black-hole-binary simulations for gravitational-wave detection

Class Quantum Grav 26 114001 (2009) (21 pages)

[17] I Hinder The current status of binary black hole simulations in numerical relativity

Class Quantum Grav 27 114004 (2010) (19 pages)

[18] B P Abbott et al LIGO The Laser Interferometer Gravitational-Wave Observatory

Rep Prog Phys 72 076901 (2009) (25 pages)

[19] F Acernese et al The status of VIRGO Class Quantum Grav 23 S63-S69 (2006)

[20] F Acernese et al Virgo status Class Quantum Grav 25 184001 (2008) (9 pages)

[21] R Arnowitt S Deser and C W Misner The Dynamics of General Relativity in Gravitation

an introduction to current research edited by L Witten (Wiley New York 1962) Chapter 7

227-265 Also available as gr-qc0405109

[22] J W York Jr Kinematics and Dynamics of General Relativity in Sources of Gravitational

Radiation edited by L L Smarr (Cambridge University Press Cambridge 1979)

[23] L E Kidder M A Scheel and S A Teukolsky Extending the lifetime of 3D black hole

computations with a new hyperbolic system of evolution equations Phys Rev D 64 064017

(2001) (13 pages)

[24] G Calabrese J Pullin O Sarbach and M Tiglio Convergence and stability in numerical

relativity Phys Rev D 66 041501(R) (2002) (4 pages)

[25] S Frittelli and O Reula On the Newtonian Limit of General Relativity Commun Math

Phys 166 221-235 (1994)

[26] M Shibata and T Nakamura Evolution of three-dimensional gravitational waves Harmonic

slicing case Phys Rev D 52 5428-5444 (1995)

[27] Y Choquet-Bruhat and J W York Geometrical well posed systems for the Einstein equations

C R Acad Sci Paris t 321 Serie I 1089-1095 (1995)

[28] A Abrahams A Anderson Y Choquet-Bruhat and J W York Jr Einstein and Yang-Mills

Theories in Hyperbolic Form without Gauge Fixing Phys Rev Lett 75 3377-3381 (1995)

[29] C Bona J Masso E Seidel and J Stela New Formalism for Numerical Relativity

Phys Rev Lett 75 600-603 (1995)

[30] M H P M van Putten and D M Eardley Nonlinear wave equations for relativity

Phys Rev D 53 3056-3063 (1996)

[31] S Frittelli and O A Reula First-Order Symmetric Hyperbolic Einstein Equations with Ar-

bitrary Fixed Gauge Phys Rev Lett 76 4667-4670 (1996)

[32] H Friedrich Hyperbolic reductions for Einsteinrsquos equations Class Quantum Grav 13 1451-

1469 (1996)

[33] F B Estabrook R S Robinson H D Wahlquist Hyperbolic equations for vacuum gravity

using special orthonormal frames Class Quantum Grav 14 1237-1247 (1997)

[34] M S Iriondo E O Leguizamon and O A Reula Einsteinrsquos Equations in Ashtekarrsquos Vari-

ables Constitute a Symmetric Hyperbolic System Phys Rev Lett 79 4732-4735 (1997)

[35] A Anderson Y Choquet-Bruhat and J W York Einstein-Bianchi hyperbolic system for

general relativity Topol Meth Nonlinear Anal 10 353-373 (1997)

[36] T W Baumgarte and S L Shapiro Numerical integration of Einsteinrsquos field equations

Phys Rev D 59 024007 (1998) (7 pages)

[37] M A G Bonilla Symmetric hyperbolic systems for Bianchi equations Class Quantum

27

test function ϕ need not be time-dependent Furthermore ulowast is explicitly given as a linearcombination of uh as seen in Eq (C21c) below Choosing ϕ = ℓj taking the time derivativeof (C10c) and appealing to the commutivity of mixed partial derivatives we therefore arriveat

int

Dk

ℓjparttQh =

int

Dk

ℓj(parttuh)prime +

int

partDk

ℓj(

(parttu)lowast minus parttuh

)

(C13)

where (parttu)lowast depends on parttuh in precisely the same way that ulowast depends on uh We have

written ℓj rather than ϕ in the last equation to emphasize that the result also holds for anylinear combination of ℓj (for example ϕ) and even for time-dependent combinations SinceQh is itself such a combination we obtain

1

2partt

int

Dk

Q2h =

int

Dk

Qh(parttuh)prime +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

Qh(Qprimeh +

radicavprimeh) +

int

partDk

(

(parttu)lowast minus parttuh

)

Qh

=

int

Dk

radicaQhv

primeh +

1

2

int

partDk

[

2((parttu)lowast minus parttuh)Qh +Q2

h

]

(C14)

Addition of (C11) and (C14) gives

1

2partt

int

Dk

(Q2h + v2h) =

1

2

int

partDk

[

Q2h minus v2h + 2(

radicaQlowast + vlowast)vh + 2((parttu)

lowast minus parttuh)Qh

]

(C15)

the aforementioned analog of (C8) This formula holds on a single subdomain Dk and we

now combine multiple copies of it one for each value of k

2 Analysis for multiple intervals

To facilitate combination of (C15) over all k we change notation At every subdomaininterface Ik+12 equiv partDk cap partDk+1 let the superscripts L and R denote field values respectivelytaken from the left and right Then the fields evaluated at Ik+12 which belong to D

k will beuLk+12 v

Lk+12 and Q

Lk+12 while those belonging to D

k+1 will be uRk+12 vRk+12 and Q

Rk+12

However at Ikminus12 the values taken from D

k are uRkminus12 vRkminus12 and QR

kminus12 Note that we

have also replaced the subscript h denoting a numerical solution with k plusmn 12 denotingthe location of the endpoint value of the numerical solution With this notation we define

∆Lα =

1

2

[

(QLα)

2 minus (vLα)2]

+(radic

aQlowastα + vlowastα

)

vLα +[

(parttuα)lowast minus parttu

Lα

]

QLα (C16)

and similarly for ∆Rα The same numerical fluxes appear in both ∆L

α and ∆Rα (ie each

numerical flux takes the same value on either side of an interface) whence fluxes like Qlowastα do

not carry an L or R superscript In terms of these definitions (C15) becomes

1

2partt

int

Dk

(Q2h + v2h) = ∆L

k+12 minus∆Rkminus12 (C17)

28

Summation over all Dk yields

1

2partt

kmaxsum

k=1

int

Dk

(Q2h + v2h) =

kmaxminus1sum

k=1

(

∆Lk+12 minus∆R

k+12

)

+∆Lkmax+12 minus∆R

12

=

kmaxminus1sum

k=1

(

∆Lh minus∆R

h

)∣

∣

Ik+12 +∆Lkmax+12 minus∆R

12 (C18)

We have reverted to h-notation denoting the numerical solution since the LR superscriptsindicate unambiguously the relevant domain used for evaluation at Ik+12

We again seek an estimate of the form

kmaxsum

k=1

(

Qh(T middot)2Dk + avh(T middot)2Dk

)

le C(T )kmaxsum

k=1

(

Qh(0 middot)2Dk + avh(0 middot)2Dk

)

(C19)

that is essentially the same as the one (C3) considered in the continuum case Assume thatthe chosen boundary conditions ensure ∆L

kmax+12 minus ∆R12 is bounded by a time-dependent

constant which does not depend on the numerical parameters N and h (subdomain width)Establishment of stability then amounts to showing that the remaining sum over interfaceterms in (C18) is non-positive whence this remaining sum is consistent with C(T ) le 1although the boundary conditions may give rise to a different bound In fact we will choosethe numerical fluxes such that each individual interface term is non-positive At interfaceIk+12 and in LR notation the jump and average of vh for example are

1

2

(

v+ + vminus)

equiv vh =1

2

(

vLk+12 + vRk+12

)

(C20a)

nminusvminus + n+v+ equiv[[

vh]]

= vLk+12 minus vRk+12 (C20b)

Consider numerical fluxes of the form

Qlowast = Qh minusτQ2

[[

Qh

]]

(C21a)

vlowast = vh minusτv2

[[

vh]]

(C21b)

ulowast = uh minusτu2

[[

uh]]

(C21c)

(parttu)lowast = parttuh minus

τu2

[[

parttuh]]

(C21d)

where (C21c) induces (C21d) and where the penalty parameters τu τv and τQ are realnumbers The fluxes defined in (58) correspond to τu = 1 τv = 1 +

radica and τQ = 0 In

terms of these quantities the kth interface contribution in (C18) is

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

parttuh]]

minus τu2

[[

parttuh]][[

Qh

]]

(C22)

where we have suppressed the k dependence of the right-hand side Now consider the term[[

parttuh]]

Because parttuh and Qh+radicavh are both polynomials of degree N Eq (C10a) implies

29

τu

τ Q

minus1 0 1minus1

minus05

0

05

1

τu

minus1 0 1minus1

minus05

0

05

1

FIG 6 Stable evolutions for the model system For fixed τv = 10minus6 and τv = 1 +radic2

respectively the left and right plots depict stable choices (determined empirically) of τu and τQfor the linear model system (C1) The stable regions are colored black but the jagged edges result

from the discretization of the (τu τQ)-plane

the vector equation parttu = Q +radicav that is pointwise equivalence on the nodal points of

Dk which in turn implies

[[

parttuh]]

=[[

Qh+radicavh]]

Upon substituting this identity into thelast equation we arrive at an expression which features only vh and Qh

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

Qh +radicavh]]

minus τu2

[[

Qh +radicavh]][[

Qh

]]

(C23)

The identities vh[[

vh]]

= 12

[[

v2h]]

and[[

Qh +radicavh]]

=[[

Qh

]]

+radica[[

vh]]

then simplify(C23) to

(∆Lh minus∆R

h )∣

∣

Ik+12 = minusτv2

[[

vh]]2 minus

radica(τu + τQ)

2

[[

Qh

]][[

vh]]

minus τu2

[[

Qh

]]2 (C24)

The role of a penalty parameter is now clear Positive values of τv penalize jumps in vhthrough a negative contribution to the energy Likewise positive values of τu penalizejumps in Qh through a negative contribution to the energy However because the sign of[[

Qh

]][[

vh]]

can be positive or negative only the choice τQ = minusτu yields an expression for(∆L

h minus ∆Rh )|Ik+12 which is manifestly negative for τu ge 0 and τv ge 0 A simple estimate

based on Youngrsquos inequality with ε (that is 2αβ le εminus1α2 + εβ2 where α β ge 0 and ε gt 0)shows that for τQ = 0 the choice τv ge aτu4 also yields stability

Figure 6 depicts certain choices of stable penalty parameters for the linear model systemevolved to tfinal = 1000 (with a = 2 N = 10 and ∆t ≃ 00553) as determined empiricallywith simulations similar to those described in Sec IVA The left plot corresponds to asmall τv = 10minus6 for which the choice τu = 1 τQ = 0 is not stable as expected from thetheoretical analysis However the right plot corresponds to τv = 1 +

radica for which τu = 1

τQ = 0 is stable Motivated by the numerical flux choices (3842) used for the BSSN system

30

(4) we have (as mentioned above) set τu = 1 τv = 1 +radica and τQ = 0 in simulations of

the nonlinear model (53) For the nonlinear model system (53) the theoretically motivatedchoice τQ = minusτu also yields numerically stable evolutions when τu ge 0 and τv ge 0

For the nonlinear systems (4) and (53) we do not attempt a formal stability proofNevertheless the results of this appendix have served as a guide for our choices of penaltyparameters For the BSSN system (4) u v and Q are block indices [cf Eq (6)] Similarto the model problem we have penalized Q with τu = 1 with τv chosen large enough toheuristically overcome the cross-terms of indefinite size that arise from τQ = 0 (we interpretequations like τu = 1 componentwise) An analogous choice ldquoτQ = minusτurdquo for the BSSNsystem might be possible but would be considerably more complicated Indeed such achoice likely entails a matrix of penalty parameters but we do not give the details here

[1] F Pretorius Evolution of Binary Black-Hole Spacetimes Phys Rev Lett 95 121101 (2005)

(4 pages)

[2] F Pretorius Simulation of binary black hole spacetimes with a harmonic evolution scheme

Class Quantum Grav 23 S529-S552 (2006)

[3] M Campanelli C O Lousto P Marronetti and Y Zlochower Accurate Evolutions of Or-

biting Black-Hole Binaries without Excision Phys Rev Lett 96 111101 (2006) (4 pages)

[4] J G Baker J Centrella D-I Choi M Koppitz and J van Meter Gravitational-Wave

Extraction from an Inspiraling Configuration of Merging Black Holes Phys Rev Lett 96

111102 (2006) (4 pages)

[5] M Campanelli C O Lousto and Y Zlochower Last orbit of binary black holes Phys Rev D

73 061501(R) (2006) (5 pages)

[6] F Herrmann I Hinder D Shoemaker and P Laguna Unequal mass binary black hole plunges

and gravitational recoil Class Quantum Grav 24 S33-S42 (2007)

[7] P Diener F Herrmann D Pollney E Schnetter E Seidel R Takahashi J Thornburg and

J Ventrella Accurate Evolution of Orbiting Binary Black Holes Phys Rev Lett 96 121101

(2006) (4 pages)

[8] M A Scheel H P Pfeiffer L Lindblom L E Kidder O Rinne and S A Teukolsky Solving

Einsteinrsquos Equations With Dual Coordinate Frames Phys Rev D 74 104006 (2006)

[9] U Sperhake Binary black-hole evolutions of excision and puncture data Phys Rev D 76

104015 (2007) (20 pages)

[10] B Brugmann J A Gonzalez M Hannam S Husa U Sperhake and W Tichy Calibration

of moving puncture simulations Phys Rev D 77 024027 (2008) (25 pages)

[11] P Marronetti W Tichy B Brugmann J Gonzalez M Hannam S Husa and U Sperhake

Binary black holes on a budget simulations using workstations Class Quantum Grav 24

S43-S58 (2007)

[12] Z B Etienne J A Faber Y T Liu S L Shapiro and T W Baumgarte Filling the holes

Evolving excised binary black hole initial data with puncture techniques Phys Rev D 76

101503(R) (2007) (5 pages)

[13] B Szilagyi D Pollney L Rezzolla J Thornburg and J Winicour An explicit harmonic code

for black-hole evolution using excision Class Quantum Grav 24 S275-S293 (2007)

[14] L Boyle M Kesden and S Nissanke Binary-Black-Hole Merger Symmetry and the Spin

Expansion Phys Rev Lett 100 151101 (2008) (4 pages)

31

[15] M A Scheel M Boyle T Chu L E Kidder K D Matthews and H P Pfeiffer High-

accuracy waveforms for binary black hole inspiral merger and ringdown Phys Rev D 79

024003 (2009) (14 pages)

[16] M Hannam Status of black-hole-binary simulations for gravitational-wave detection

Class Quantum Grav 26 114001 (2009) (21 pages)

[17] I Hinder The current status of binary black hole simulations in numerical relativity

Class Quantum Grav 27 114004 (2010) (19 pages)

[18] B P Abbott et al LIGO The Laser Interferometer Gravitational-Wave Observatory

Rep Prog Phys 72 076901 (2009) (25 pages)

[19] F Acernese et al The status of VIRGO Class Quantum Grav 23 S63-S69 (2006)

[20] F Acernese et al Virgo status Class Quantum Grav 25 184001 (2008) (9 pages)

[21] R Arnowitt S Deser and C W Misner The Dynamics of General Relativity in Gravitation

an introduction to current research edited by L Witten (Wiley New York 1962) Chapter 7

227-265 Also available as gr-qc0405109

[22] J W York Jr Kinematics and Dynamics of General Relativity in Sources of Gravitational

Radiation edited by L L Smarr (Cambridge University Press Cambridge 1979)

[23] L E Kidder M A Scheel and S A Teukolsky Extending the lifetime of 3D black hole

computations with a new hyperbolic system of evolution equations Phys Rev D 64 064017

(2001) (13 pages)

[24] G Calabrese J Pullin O Sarbach and M Tiglio Convergence and stability in numerical

relativity Phys Rev D 66 041501(R) (2002) (4 pages)

[25] S Frittelli and O Reula On the Newtonian Limit of General Relativity Commun Math

Phys 166 221-235 (1994)

[26] M Shibata and T Nakamura Evolution of three-dimensional gravitational waves Harmonic

slicing case Phys Rev D 52 5428-5444 (1995)

[27] Y Choquet-Bruhat and J W York Geometrical well posed systems for the Einstein equations

C R Acad Sci Paris t 321 Serie I 1089-1095 (1995)

[28] A Abrahams A Anderson Y Choquet-Bruhat and J W York Jr Einstein and Yang-Mills

Theories in Hyperbolic Form without Gauge Fixing Phys Rev Lett 75 3377-3381 (1995)

[29] C Bona J Masso E Seidel and J Stela New Formalism for Numerical Relativity

Phys Rev Lett 75 600-603 (1995)

[30] M H P M van Putten and D M Eardley Nonlinear wave equations for relativity

Phys Rev D 53 3056-3063 (1996)

[31] S Frittelli and O A Reula First-Order Symmetric Hyperbolic Einstein Equations with Ar-

bitrary Fixed Gauge Phys Rev Lett 76 4667-4670 (1996)

[32] H Friedrich Hyperbolic reductions for Einsteinrsquos equations Class Quantum Grav 13 1451-

1469 (1996)

[33] F B Estabrook R S Robinson H D Wahlquist Hyperbolic equations for vacuum gravity

using special orthonormal frames Class Quantum Grav 14 1237-1247 (1997)

[34] M S Iriondo E O Leguizamon and O A Reula Einsteinrsquos Equations in Ashtekarrsquos Vari-

ables Constitute a Symmetric Hyperbolic System Phys Rev Lett 79 4732-4735 (1997)

[35] A Anderson Y Choquet-Bruhat and J W York Einstein-Bianchi hyperbolic system for

general relativity Topol Meth Nonlinear Anal 10 353-373 (1997)

[36] T W Baumgarte and S L Shapiro Numerical integration of Einsteinrsquos field equations

Phys Rev D 59 024007 (1998) (7 pages)

[37] M A G Bonilla Symmetric hyperbolic systems for Bianchi equations Class Quantum

28

Summation over all Dk yields

1

2partt

kmaxsum

k=1

int

Dk

(Q2h + v2h) =

kmaxminus1sum

k=1

(

∆Lk+12 minus∆R

k+12

)

+∆Lkmax+12 minus∆R

12

=

kmaxminus1sum

k=1

(

∆Lh minus∆R

h

)∣

∣

Ik+12 +∆Lkmax+12 minus∆R

12 (C18)

We have reverted to h-notation denoting the numerical solution since the LR superscriptsindicate unambiguously the relevant domain used for evaluation at Ik+12

We again seek an estimate of the form

kmaxsum

k=1

(

Qh(T middot)2Dk + avh(T middot)2Dk

)

le C(T )kmaxsum

k=1

(

Qh(0 middot)2Dk + avh(0 middot)2Dk

)

(C19)

that is essentially the same as the one (C3) considered in the continuum case Assume thatthe chosen boundary conditions ensure ∆L

kmax+12 minus ∆R12 is bounded by a time-dependent

constant which does not depend on the numerical parameters N and h (subdomain width)Establishment of stability then amounts to showing that the remaining sum over interfaceterms in (C18) is non-positive whence this remaining sum is consistent with C(T ) le 1although the boundary conditions may give rise to a different bound In fact we will choosethe numerical fluxes such that each individual interface term is non-positive At interfaceIk+12 and in LR notation the jump and average of vh for example are

1

2

(

v+ + vminus)

equiv vh =1

2

(

vLk+12 + vRk+12

)

(C20a)

nminusvminus + n+v+ equiv[[

vh]]

= vLk+12 minus vRk+12 (C20b)

Consider numerical fluxes of the form

Qlowast = Qh minusτQ2

[[

Qh

]]

(C21a)

vlowast = vh minusτv2

[[

vh]]

(C21b)

ulowast = uh minusτu2

[[

uh]]

(C21c)

(parttu)lowast = parttuh minus

τu2

[[

parttuh]]

(C21d)

where (C21c) induces (C21d) and where the penalty parameters τu τv and τQ are realnumbers The fluxes defined in (58) correspond to τu = 1 τv = 1 +

radica and τQ = 0 In

terms of these quantities the kth interface contribution in (C18) is

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

parttuh]]

minus τu2

[[

parttuh]][[

Qh

]]

(C22)

where we have suppressed the k dependence of the right-hand side Now consider the term[[

parttuh]]

Because parttuh and Qh+radicavh are both polynomials of degree N Eq (C10a) implies

29

τu

τ Q

minus1 0 1minus1

minus05

0

05

1

τu

minus1 0 1minus1

minus05

0

05

1

FIG 6 Stable evolutions for the model system For fixed τv = 10minus6 and τv = 1 +radic2

respectively the left and right plots depict stable choices (determined empirically) of τu and τQfor the linear model system (C1) The stable regions are colored black but the jagged edges result

from the discretization of the (τu τQ)-plane

the vector equation parttu = Q +radicav that is pointwise equivalence on the nodal points of

Dk which in turn implies

[[

parttuh]]

=[[

Qh+radicavh]]

Upon substituting this identity into thelast equation we arrive at an expression which features only vh and Qh

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

Qh +radicavh]]

minus τu2

[[

Qh +radicavh]][[

Qh

]]

(C23)

The identities vh[[

vh]]

= 12

[[

v2h]]

and[[

Qh +radicavh]]

=[[

Qh

]]

+radica[[

vh]]

then simplify(C23) to

(∆Lh minus∆R

h )∣

∣

Ik+12 = minusτv2

[[

vh]]2 minus

radica(τu + τQ)

2

[[

Qh

]][[

vh]]

minus τu2

[[

Qh

]]2 (C24)

The role of a penalty parameter is now clear Positive values of τv penalize jumps in vhthrough a negative contribution to the energy Likewise positive values of τu penalizejumps in Qh through a negative contribution to the energy However because the sign of[[

Qh

]][[

vh]]

can be positive or negative only the choice τQ = minusτu yields an expression for(∆L

h minus ∆Rh )|Ik+12 which is manifestly negative for τu ge 0 and τv ge 0 A simple estimate

based on Youngrsquos inequality with ε (that is 2αβ le εminus1α2 + εβ2 where α β ge 0 and ε gt 0)shows that for τQ = 0 the choice τv ge aτu4 also yields stability

Figure 6 depicts certain choices of stable penalty parameters for the linear model systemevolved to tfinal = 1000 (with a = 2 N = 10 and ∆t ≃ 00553) as determined empiricallywith simulations similar to those described in Sec IVA The left plot corresponds to asmall τv = 10minus6 for which the choice τu = 1 τQ = 0 is not stable as expected from thetheoretical analysis However the right plot corresponds to τv = 1 +

radica for which τu = 1

τQ = 0 is stable Motivated by the numerical flux choices (3842) used for the BSSN system

30

(4) we have (as mentioned above) set τu = 1 τv = 1 +radica and τQ = 0 in simulations of

the nonlinear model (53) For the nonlinear model system (53) the theoretically motivatedchoice τQ = minusτu also yields numerically stable evolutions when τu ge 0 and τv ge 0

For the nonlinear systems (4) and (53) we do not attempt a formal stability proofNevertheless the results of this appendix have served as a guide for our choices of penaltyparameters For the BSSN system (4) u v and Q are block indices [cf Eq (6)] Similarto the model problem we have penalized Q with τu = 1 with τv chosen large enough toheuristically overcome the cross-terms of indefinite size that arise from τQ = 0 (we interpretequations like τu = 1 componentwise) An analogous choice ldquoτQ = minusτurdquo for the BSSNsystem might be possible but would be considerably more complicated Indeed such achoice likely entails a matrix of penalty parameters but we do not give the details here

[1] F Pretorius Evolution of Binary Black-Hole Spacetimes Phys Rev Lett 95 121101 (2005)

(4 pages)

[2] F Pretorius Simulation of binary black hole spacetimes with a harmonic evolution scheme

Class Quantum Grav 23 S529-S552 (2006)

[3] M Campanelli C O Lousto P Marronetti and Y Zlochower Accurate Evolutions of Or-

biting Black-Hole Binaries without Excision Phys Rev Lett 96 111101 (2006) (4 pages)

[4] J G Baker J Centrella D-I Choi M Koppitz and J van Meter Gravitational-Wave

Extraction from an Inspiraling Configuration of Merging Black Holes Phys Rev Lett 96

111102 (2006) (4 pages)

[5] M Campanelli C O Lousto and Y Zlochower Last orbit of binary black holes Phys Rev D

73 061501(R) (2006) (5 pages)

[6] F Herrmann I Hinder D Shoemaker and P Laguna Unequal mass binary black hole plunges

and gravitational recoil Class Quantum Grav 24 S33-S42 (2007)

[7] P Diener F Herrmann D Pollney E Schnetter E Seidel R Takahashi J Thornburg and

J Ventrella Accurate Evolution of Orbiting Binary Black Holes Phys Rev Lett 96 121101

(2006) (4 pages)

[8] M A Scheel H P Pfeiffer L Lindblom L E Kidder O Rinne and S A Teukolsky Solving

Einsteinrsquos Equations With Dual Coordinate Frames Phys Rev D 74 104006 (2006)

[9] U Sperhake Binary black-hole evolutions of excision and puncture data Phys Rev D 76

104015 (2007) (20 pages)

[10] B Brugmann J A Gonzalez M Hannam S Husa U Sperhake and W Tichy Calibration

of moving puncture simulations Phys Rev D 77 024027 (2008) (25 pages)

[11] P Marronetti W Tichy B Brugmann J Gonzalez M Hannam S Husa and U Sperhake

Binary black holes on a budget simulations using workstations Class Quantum Grav 24

S43-S58 (2007)

[12] Z B Etienne J A Faber Y T Liu S L Shapiro and T W Baumgarte Filling the holes

Evolving excised binary black hole initial data with puncture techniques Phys Rev D 76

101503(R) (2007) (5 pages)

[13] B Szilagyi D Pollney L Rezzolla J Thornburg and J Winicour An explicit harmonic code

for black-hole evolution using excision Class Quantum Grav 24 S275-S293 (2007)

[14] L Boyle M Kesden and S Nissanke Binary-Black-Hole Merger Symmetry and the Spin

Expansion Phys Rev Lett 100 151101 (2008) (4 pages)

31

[15] M A Scheel M Boyle T Chu L E Kidder K D Matthews and H P Pfeiffer High-

accuracy waveforms for binary black hole inspiral merger and ringdown Phys Rev D 79

024003 (2009) (14 pages)

[16] M Hannam Status of black-hole-binary simulations for gravitational-wave detection

Class Quantum Grav 26 114001 (2009) (21 pages)

[17] I Hinder The current status of binary black hole simulations in numerical relativity

Class Quantum Grav 27 114004 (2010) (19 pages)

[18] B P Abbott et al LIGO The Laser Interferometer Gravitational-Wave Observatory

Rep Prog Phys 72 076901 (2009) (25 pages)

[19] F Acernese et al The status of VIRGO Class Quantum Grav 23 S63-S69 (2006)

[20] F Acernese et al Virgo status Class Quantum Grav 25 184001 (2008) (9 pages)

[21] R Arnowitt S Deser and C W Misner The Dynamics of General Relativity in Gravitation

an introduction to current research edited by L Witten (Wiley New York 1962) Chapter 7

227-265 Also available as gr-qc0405109

[22] J W York Jr Kinematics and Dynamics of General Relativity in Sources of Gravitational

Radiation edited by L L Smarr (Cambridge University Press Cambridge 1979)

[23] L E Kidder M A Scheel and S A Teukolsky Extending the lifetime of 3D black hole

computations with a new hyperbolic system of evolution equations Phys Rev D 64 064017

(2001) (13 pages)

[24] G Calabrese J Pullin O Sarbach and M Tiglio Convergence and stability in numerical

relativity Phys Rev D 66 041501(R) (2002) (4 pages)

[25] S Frittelli and O Reula On the Newtonian Limit of General Relativity Commun Math

Phys 166 221-235 (1994)

[26] M Shibata and T Nakamura Evolution of three-dimensional gravitational waves Harmonic

slicing case Phys Rev D 52 5428-5444 (1995)

[27] Y Choquet-Bruhat and J W York Geometrical well posed systems for the Einstein equations

C R Acad Sci Paris t 321 Serie I 1089-1095 (1995)

[28] A Abrahams A Anderson Y Choquet-Bruhat and J W York Jr Einstein and Yang-Mills

Theories in Hyperbolic Form without Gauge Fixing Phys Rev Lett 75 3377-3381 (1995)

[29] C Bona J Masso E Seidel and J Stela New Formalism for Numerical Relativity

Phys Rev Lett 75 600-603 (1995)

[30] M H P M van Putten and D M Eardley Nonlinear wave equations for relativity

Phys Rev D 53 3056-3063 (1996)

[31] S Frittelli and O A Reula First-Order Symmetric Hyperbolic Einstein Equations with Ar-

bitrary Fixed Gauge Phys Rev Lett 76 4667-4670 (1996)

[32] H Friedrich Hyperbolic reductions for Einsteinrsquos equations Class Quantum Grav 13 1451-

1469 (1996)

[33] F B Estabrook R S Robinson H D Wahlquist Hyperbolic equations for vacuum gravity

using special orthonormal frames Class Quantum Grav 14 1237-1247 (1997)

[34] M S Iriondo E O Leguizamon and O A Reula Einsteinrsquos Equations in Ashtekarrsquos Vari-

ables Constitute a Symmetric Hyperbolic System Phys Rev Lett 79 4732-4735 (1997)

[35] A Anderson Y Choquet-Bruhat and J W York Einstein-Bianchi hyperbolic system for

general relativity Topol Meth Nonlinear Anal 10 353-373 (1997)

[36] T W Baumgarte and S L Shapiro Numerical integration of Einsteinrsquos field equations

Phys Rev D 59 024007 (1998) (7 pages)

[37] M A G Bonilla Symmetric hyperbolic systems for Bianchi equations Class Quantum

29

τu

τ Q

minus1 0 1minus1

minus05

0

05

1

τu

minus1 0 1minus1

minus05

0

05

1

FIG 6 Stable evolutions for the model system For fixed τv = 10minus6 and τv = 1 +radic2

respectively the left and right plots depict stable choices (determined empirically) of τu and τQfor the linear model system (C1) The stable regions are colored black but the jagged edges result

from the discretization of the (τu τQ)-plane

the vector equation parttu = Q +radicav that is pointwise equivalence on the nodal points of

Dk which in turn implies

[[

parttuh]]

=[[

Qh+radicavh]]

Upon substituting this identity into thelast equation we arrive at an expression which features only vh and Qh

(∆Lh minus∆R

h )∣

∣

Ik+12 =1

2

([[

Q2h

]]

minus[[

v2h]])

+ vh[[

vh]]

minus τv2

[[

vh]]2

+radicaQh

[[

vh]]

minusradicaτQ2

[[

Qh

]][[

vh]]

minus Qh[[

Qh +radicavh]]

minus τu2

[[

Qh +radicavh]][[

Qh

]]

(C23)

The identities vh[[

vh]]

= 12

[[

v2h]]

and[[

Qh +radicavh]]

=[[

Qh

]]

+radica[[

vh]]

then simplify(C23) to

(∆Lh minus∆R

h )∣

∣

Ik+12 = minusτv2

[[

vh]]2 minus

radica(τu + τQ)

2

[[

Qh

]][[

vh]]

minus τu2

[[

Qh

]]2 (C24)

The role of a penalty parameter is now clear Positive values of τv penalize jumps in vhthrough a negative contribution to the energy Likewise positive values of τu penalizejumps in Qh through a negative contribution to the energy However because the sign of[[

Qh

]][[

vh]]

can be positive or negative only the choice τQ = minusτu yields an expression for(∆L

h minus ∆Rh )|Ik+12 which is manifestly negative for τu ge 0 and τv ge 0 A simple estimate

based on Youngrsquos inequality with ε (that is 2αβ le εminus1α2 + εβ2 where α β ge 0 and ε gt 0)shows that for τQ = 0 the choice τv ge aτu4 also yields stability

Figure 6 depicts certain choices of stable penalty parameters for the linear model systemevolved to tfinal = 1000 (with a = 2 N = 10 and ∆t ≃ 00553) as determined empiricallywith simulations similar to those described in Sec IVA The left plot corresponds to asmall τv = 10minus6 for which the choice τu = 1 τQ = 0 is not stable as expected from thetheoretical analysis However the right plot corresponds to τv = 1 +

radica for which τu = 1

τQ = 0 is stable Motivated by the numerical flux choices (3842) used for the BSSN system

30

(4) we have (as mentioned above) set τu = 1 τv = 1 +radica and τQ = 0 in simulations of

the nonlinear model (53) For the nonlinear model system (53) the theoretically motivatedchoice τQ = minusτu also yields numerically stable evolutions when τu ge 0 and τv ge 0

For the nonlinear systems (4) and (53) we do not attempt a formal stability proofNevertheless the results of this appendix have served as a guide for our choices of penaltyparameters For the BSSN system (4) u v and Q are block indices [cf Eq (6)] Similarto the model problem we have penalized Q with τu = 1 with τv chosen large enough toheuristically overcome the cross-terms of indefinite size that arise from τQ = 0 (we interpretequations like τu = 1 componentwise) An analogous choice ldquoτQ = minusτurdquo for the BSSNsystem might be possible but would be considerably more complicated Indeed such achoice likely entails a matrix of penalty parameters but we do not give the details here

[1] F Pretorius Evolution of Binary Black-Hole Spacetimes Phys Rev Lett 95 121101 (2005)

(4 pages)

[2] F Pretorius Simulation of binary black hole spacetimes with a harmonic evolution scheme

Class Quantum Grav 23 S529-S552 (2006)

[3] M Campanelli C O Lousto P Marronetti and Y Zlochower Accurate Evolutions of Or-

biting Black-Hole Binaries without Excision Phys Rev Lett 96 111101 (2006) (4 pages)

[4] J G Baker J Centrella D-I Choi M Koppitz and J van Meter Gravitational-Wave

Extraction from an Inspiraling Configuration of Merging Black Holes Phys Rev Lett 96

111102 (2006) (4 pages)

[5] M Campanelli C O Lousto and Y Zlochower Last orbit of binary black holes Phys Rev D

73 061501(R) (2006) (5 pages)

[6] F Herrmann I Hinder D Shoemaker and P Laguna Unequal mass binary black hole plunges

and gravitational recoil Class Quantum Grav 24 S33-S42 (2007)

[7] P Diener F Herrmann D Pollney E Schnetter E Seidel R Takahashi J Thornburg and

J Ventrella Accurate Evolution of Orbiting Binary Black Holes Phys Rev Lett 96 121101

(2006) (4 pages)

[8] M A Scheel H P Pfeiffer L Lindblom L E Kidder O Rinne and S A Teukolsky Solving

Einsteinrsquos Equations With Dual Coordinate Frames Phys Rev D 74 104006 (2006)

[9] U Sperhake Binary black-hole evolutions of excision and puncture data Phys Rev D 76

104015 (2007) (20 pages)

[10] B Brugmann J A Gonzalez M Hannam S Husa U Sperhake and W Tichy Calibration

of moving puncture simulations Phys Rev D 77 024027 (2008) (25 pages)

[11] P Marronetti W Tichy B Brugmann J Gonzalez M Hannam S Husa and U Sperhake

Binary black holes on a budget simulations using workstations Class Quantum Grav 24

S43-S58 (2007)

[12] Z B Etienne J A Faber Y T Liu S L Shapiro and T W Baumgarte Filling the holes

Evolving excised binary black hole initial data with puncture techniques Phys Rev D 76

101503(R) (2007) (5 pages)

[13] B Szilagyi D Pollney L Rezzolla J Thornburg and J Winicour An explicit harmonic code

for black-hole evolution using excision Class Quantum Grav 24 S275-S293 (2007)

[14] L Boyle M Kesden and S Nissanke Binary-Black-Hole Merger Symmetry and the Spin

Expansion Phys Rev Lett 100 151101 (2008) (4 pages)

31

[15] M A Scheel M Boyle T Chu L E Kidder K D Matthews and H P Pfeiffer High-

accuracy waveforms for binary black hole inspiral merger and ringdown Phys Rev D 79

024003 (2009) (14 pages)

[16] M Hannam Status of black-hole-binary simulations for gravitational-wave detection

Class Quantum Grav 26 114001 (2009) (21 pages)

[17] I Hinder The current status of binary black hole simulations in numerical relativity

Class Quantum Grav 27 114004 (2010) (19 pages)

[18] B P Abbott et al LIGO The Laser Interferometer Gravitational-Wave Observatory

Rep Prog Phys 72 076901 (2009) (25 pages)

[19] F Acernese et al The status of VIRGO Class Quantum Grav 23 S63-S69 (2006)

[20] F Acernese et al Virgo status Class Quantum Grav 25 184001 (2008) (9 pages)

[21] R Arnowitt S Deser and C W Misner The Dynamics of General Relativity in Gravitation

an introduction to current research edited by L Witten (Wiley New York 1962) Chapter 7

227-265 Also available as gr-qc0405109

[22] J W York Jr Kinematics and Dynamics of General Relativity in Sources of Gravitational

Radiation edited by L L Smarr (Cambridge University Press Cambridge 1979)

[23] L E Kidder M A Scheel and S A Teukolsky Extending the lifetime of 3D black hole

computations with a new hyperbolic system of evolution equations Phys Rev D 64 064017

(2001) (13 pages)

[24] G Calabrese J Pullin O Sarbach and M Tiglio Convergence and stability in numerical

relativity Phys Rev D 66 041501(R) (2002) (4 pages)

[25] S Frittelli and O Reula On the Newtonian Limit of General Relativity Commun Math

Phys 166 221-235 (1994)

[26] M Shibata and T Nakamura Evolution of three-dimensional gravitational waves Harmonic

slicing case Phys Rev D 52 5428-5444 (1995)

[27] Y Choquet-Bruhat and J W York Geometrical well posed systems for the Einstein equations

C R Acad Sci Paris t 321 Serie I 1089-1095 (1995)

[28] A Abrahams A Anderson Y Choquet-Bruhat and J W York Jr Einstein and Yang-Mills

Theories in Hyperbolic Form without Gauge Fixing Phys Rev Lett 75 3377-3381 (1995)

[29] C Bona J Masso E Seidel and J Stela New Formalism for Numerical Relativity

Phys Rev Lett 75 600-603 (1995)

[30] M H P M van Putten and D M Eardley Nonlinear wave equations for relativity

Phys Rev D 53 3056-3063 (1996)

[31] S Frittelli and O A Reula First-Order Symmetric Hyperbolic Einstein Equations with Ar-

bitrary Fixed Gauge Phys Rev Lett 76 4667-4670 (1996)

[32] H Friedrich Hyperbolic reductions for Einsteinrsquos equations Class Quantum Grav 13 1451-

1469 (1996)

[33] F B Estabrook R S Robinson H D Wahlquist Hyperbolic equations for vacuum gravity

using special orthonormal frames Class Quantum Grav 14 1237-1247 (1997)

[34] M S Iriondo E O Leguizamon and O A Reula Einsteinrsquos Equations in Ashtekarrsquos Vari-

ables Constitute a Symmetric Hyperbolic System Phys Rev Lett 79 4732-4735 (1997)

[35] A Anderson Y Choquet-Bruhat and J W York Einstein-Bianchi hyperbolic system for

general relativity Topol Meth Nonlinear Anal 10 353-373 (1997)

[36] T W Baumgarte and S L Shapiro Numerical integration of Einsteinrsquos field equations

Phys Rev D 59 024007 (1998) (7 pages)

[37] M A G Bonilla Symmetric hyperbolic systems for Bianchi equations Class Quantum

30

(4) we have (as mentioned above) set τu = 1 τv = 1 +radica and τQ = 0 in simulations of

the nonlinear model (53) For the nonlinear model system (53) the theoretically motivatedchoice τQ = minusτu also yields numerically stable evolutions when τu ge 0 and τv ge 0

For the nonlinear systems (4) and (53) we do not attempt a formal stability proofNevertheless the results of this appendix have served as a guide for our choices of penaltyparameters For the BSSN system (4) u v and Q are block indices [cf Eq (6)] Similarto the model problem we have penalized Q with τu = 1 with τv chosen large enough toheuristically overcome the cross-terms of indefinite size that arise from τQ = 0 (we interpretequations like τu = 1 componentwise) An analogous choice ldquoτQ = minusτurdquo for the BSSNsystem might be possible but would be considerably more complicated Indeed such achoice likely entails a matrix of penalty parameters but we do not give the details here

[1] F Pretorius Evolution of Binary Black-Hole Spacetimes Phys Rev Lett 95 121101 (2005)

(4 pages)

[2] F Pretorius Simulation of binary black hole spacetimes with a harmonic evolution scheme

Class Quantum Grav 23 S529-S552 (2006)

[3] M Campanelli C O Lousto P Marronetti and Y Zlochower Accurate Evolutions of Or-

biting Black-Hole Binaries without Excision Phys Rev Lett 96 111101 (2006) (4 pages)

[4] J G Baker J Centrella D-I Choi M Koppitz and J van Meter Gravitational-Wave

Extraction from an Inspiraling Configuration of Merging Black Holes Phys Rev Lett 96

111102 (2006) (4 pages)

[5] M Campanelli C O Lousto and Y Zlochower Last orbit of binary black holes Phys Rev D

73 061501(R) (2006) (5 pages)

[6] F Herrmann I Hinder D Shoemaker and P Laguna Unequal mass binary black hole plunges

and gravitational recoil Class Quantum Grav 24 S33-S42 (2007)

[7] P Diener F Herrmann D Pollney E Schnetter E Seidel R Takahashi J Thornburg and

J Ventrella Accurate Evolution of Orbiting Binary Black Holes Phys Rev Lett 96 121101

(2006) (4 pages)

[8] M A Scheel H P Pfeiffer L Lindblom L E Kidder O Rinne and S A Teukolsky Solving

Einsteinrsquos Equations With Dual Coordinate Frames Phys Rev D 74 104006 (2006)

[9] U Sperhake Binary black-hole evolutions of excision and puncture data Phys Rev D 76

104015 (2007) (20 pages)

[10] B Brugmann J A Gonzalez M Hannam S Husa U Sperhake and W Tichy Calibration

of moving puncture simulations Phys Rev D 77 024027 (2008) (25 pages)

[11] P Marronetti W Tichy B Brugmann J Gonzalez M Hannam S Husa and U Sperhake

Binary black holes on a budget simulations using workstations Class Quantum Grav 24

S43-S58 (2007)

[12] Z B Etienne J A Faber Y T Liu S L Shapiro and T W Baumgarte Filling the holes

Evolving excised binary black hole initial data with puncture techniques Phys Rev D 76

101503(R) (2007) (5 pages)

[13] B Szilagyi D Pollney L Rezzolla J Thornburg and J Winicour An explicit harmonic code

for black-hole evolution using excision Class Quantum Grav 24 S275-S293 (2007)

[14] L Boyle M Kesden and S Nissanke Binary-Black-Hole Merger Symmetry and the Spin

Expansion Phys Rev Lett 100 151101 (2008) (4 pages)

31

[15] M A Scheel M Boyle T Chu L E Kidder K D Matthews and H P Pfeiffer High-

accuracy waveforms for binary black hole inspiral merger and ringdown Phys Rev D 79

024003 (2009) (14 pages)

[16] M Hannam Status of black-hole-binary simulations for gravitational-wave detection

Class Quantum Grav 26 114001 (2009) (21 pages)

[17] I Hinder The current status of binary black hole simulations in numerical relativity

Class Quantum Grav 27 114004 (2010) (19 pages)

[18] B P Abbott et al LIGO The Laser Interferometer Gravitational-Wave Observatory

Rep Prog Phys 72 076901 (2009) (25 pages)

[19] F Acernese et al The status of VIRGO Class Quantum Grav 23 S63-S69 (2006)

[20] F Acernese et al Virgo status Class Quantum Grav 25 184001 (2008) (9 pages)

[21] R Arnowitt S Deser and C W Misner The Dynamics of General Relativity in Gravitation

an introduction to current research edited by L Witten (Wiley New York 1962) Chapter 7

227-265 Also available as gr-qc0405109

[22] J W York Jr Kinematics and Dynamics of General Relativity in Sources of Gravitational

Radiation edited by L L Smarr (Cambridge University Press Cambridge 1979)

[23] L E Kidder M A Scheel and S A Teukolsky Extending the lifetime of 3D black hole

computations with a new hyperbolic system of evolution equations Phys Rev D 64 064017

(2001) (13 pages)

[24] G Calabrese J Pullin O Sarbach and M Tiglio Convergence and stability in numerical

relativity Phys Rev D 66 041501(R) (2002) (4 pages)

[25] S Frittelli and O Reula On the Newtonian Limit of General Relativity Commun Math

Phys 166 221-235 (1994)

[26] M Shibata and T Nakamura Evolution of three-dimensional gravitational waves Harmonic

slicing case Phys Rev D 52 5428-5444 (1995)

[27] Y Choquet-Bruhat and J W York Geometrical well posed systems for the Einstein equations

C R Acad Sci Paris t 321 Serie I 1089-1095 (1995)

[28] A Abrahams A Anderson Y Choquet-Bruhat and J W York Jr Einstein and Yang-Mills

Theories in Hyperbolic Form without Gauge Fixing Phys Rev Lett 75 3377-3381 (1995)

[29] C Bona J Masso E Seidel and J Stela New Formalism for Numerical Relativity

Phys Rev Lett 75 600-603 (1995)

[30] M H P M van Putten and D M Eardley Nonlinear wave equations for relativity

Phys Rev D 53 3056-3063 (1996)

[31] S Frittelli and O A Reula First-Order Symmetric Hyperbolic Einstein Equations with Ar-

bitrary Fixed Gauge Phys Rev Lett 76 4667-4670 (1996)

[32] H Friedrich Hyperbolic reductions for Einsteinrsquos equations Class Quantum Grav 13 1451-

1469 (1996)

[33] F B Estabrook R S Robinson H D Wahlquist Hyperbolic equations for vacuum gravity

using special orthonormal frames Class Quantum Grav 14 1237-1247 (1997)

[34] M S Iriondo E O Leguizamon and O A Reula Einsteinrsquos Equations in Ashtekarrsquos Vari-

ables Constitute a Symmetric Hyperbolic System Phys Rev Lett 79 4732-4735 (1997)

[35] A Anderson Y Choquet-Bruhat and J W York Einstein-Bianchi hyperbolic system for

general relativity Topol Meth Nonlinear Anal 10 353-373 (1997)

[36] T W Baumgarte and S L Shapiro Numerical integration of Einsteinrsquos field equations

Phys Rev D 59 024007 (1998) (7 pages)

[37] M A G Bonilla Symmetric hyperbolic systems for Bianchi equations Class Quantum

31

[15] M A Scheel M Boyle T Chu L E Kidder K D Matthews and H P Pfeiffer High-

accuracy waveforms for binary black hole inspiral merger and ringdown Phys Rev D 79

024003 (2009) (14 pages)

[16] M Hannam Status of black-hole-binary simulations for gravitational-wave detection

Class Quantum Grav 26 114001 (2009) (21 pages)

[17] I Hinder The current status of binary black hole simulations in numerical relativity

Class Quantum Grav 27 114004 (2010) (19 pages)

[18] B P Abbott et al LIGO The Laser Interferometer Gravitational-Wave Observatory

Rep Prog Phys 72 076901 (2009) (25 pages)

[19] F Acernese et al The status of VIRGO Class Quantum Grav 23 S63-S69 (2006)

[20] F Acernese et al Virgo status Class Quantum Grav 25 184001 (2008) (9 pages)

[21] R Arnowitt S Deser and C W Misner The Dynamics of General Relativity in Gravitation

an introduction to current research edited by L Witten (Wiley New York 1962) Chapter 7

227-265 Also available as gr-qc0405109

[22] J W York Jr Kinematics and Dynamics of General Relativity in Sources of Gravitational

Radiation edited by L L Smarr (Cambridge University Press Cambridge 1979)

[23] L E Kidder M A Scheel and S A Teukolsky Extending the lifetime of 3D black hole

computations with a new hyperbolic system of evolution equations Phys Rev D 64 064017

(2001) (13 pages)

[24] G Calabrese J Pullin O Sarbach and M Tiglio Convergence and stability in numerical

relativity Phys Rev D 66 041501(R) (2002) (4 pages)

[25] S Frittelli and O Reula On the Newtonian Limit of General Relativity Commun Math

Phys 166 221-235 (1994)

[26] M Shibata and T Nakamura Evolution of three-dimensional gravitational waves Harmonic

slicing case Phys Rev D 52 5428-5444 (1995)

[27] Y Choquet-Bruhat and J W York Geometrical well posed systems for the Einstein equations

C R Acad Sci Paris t 321 Serie I 1089-1095 (1995)

[28] A Abrahams A Anderson Y Choquet-Bruhat and J W York Jr Einstein and Yang-Mills

Theories in Hyperbolic Form without Gauge Fixing Phys Rev Lett 75 3377-3381 (1995)

[29] C Bona J Masso E Seidel and J Stela New Formalism for Numerical Relativity

Phys Rev Lett 75 600-603 (1995)

[30] M H P M van Putten and D M Eardley Nonlinear wave equations for relativity

Phys Rev D 53 3056-3063 (1996)

[31] S Frittelli and O A Reula First-Order Symmetric Hyperbolic Einstein Equations with Ar-

bitrary Fixed Gauge Phys Rev Lett 76 4667-4670 (1996)

[32] H Friedrich Hyperbolic reductions for Einsteinrsquos equations Class Quantum Grav 13 1451-

1469 (1996)

[33] F B Estabrook R S Robinson H D Wahlquist Hyperbolic equations for vacuum gravity

using special orthonormal frames Class Quantum Grav 14 1237-1247 (1997)

[34] M S Iriondo E O Leguizamon and O A Reula Einsteinrsquos Equations in Ashtekarrsquos Vari-

ables Constitute a Symmetric Hyperbolic System Phys Rev Lett 79 4732-4735 (1997)

[35] A Anderson Y Choquet-Bruhat and J W York Einstein-Bianchi hyperbolic system for

general relativity Topol Meth Nonlinear Anal 10 353-373 (1997)

[36] T W Baumgarte and S L Shapiro Numerical integration of Einsteinrsquos field equations

Phys Rev D 59 024007 (1998) (7 pages)

[37] M A G Bonilla Symmetric hyperbolic systems for Bianchi equations Class Quantum

32

Grav 15 2001-2005 (1998)

[38] G Yoneda and H Shinkai Symmetric Hyperbolic System in the Ashtekar Formulation

Phys Rev Lett 82 263-266 (1999)

[39] M Alcubierre B Brugmann M Miller and W-M Suen Conformal hyperbolic formulation

of the Einstein equations Phys Rev D 60 064017 (1999) (4 pages)

[40] S Frittelli and O A Reula Well-posed forms of the 3+1 conformally-decomposed Einstein

equations J Math Phys 40 5143-5156 (1999)

[41] A Anderson and J W York Jr Fixing Einsteinrsquos Equations Phys Rev Lett 82 4384-4387

(1999)

[42] S D Hern Numerical Relativity and Inhomogeneous Cosmologies PhD dissertation Uni-

versity of Cambridge 1999 Also available as gr-qc0004036

[43] H Friedrich and A Rendall The Cauchy problem for the Einstein equations in Einsteinrsquos

Field Equations and their Physical Implications edited by B G Schmidt Springer Lecture

Notes in Physics vol 540 pp 127-223 (Springer-Verlag Berlin 2000)

[44] G Yoneda and H Shinkai Constructing Hyperbolic Systems in the Ashtekar Formulation of

General Relativity Int J Mod Phys D 9 13-34 (2000)

[45] H Shinkai and G Yoneda Hyperbolic formulations and numerical relativity experiments

using Ashtekarrsquos connection variables Class Quantum Grav 17 4799-4822 (2000)

[46] G Yoneda and H Shinkai Hyperbolic formulations and numerical relativity II asymptotically

constrained systems of Einstein equations Class Quantum Grav 18 441-462 (2001)

[47] L E Kidder M A Scheel S A Teukolsky Extending the lifetime of 3D black hole compu-

tations with a new hyperbolic system of evolution equations Phys Rev D 64 064017 (2001)

(13 pages)

[48] C Gundlach G Calabrese I Hinder and J M Martın-Garcıa Constraint damping in the Z4

formulation and harmonic gauge Class Quantum Grav 22 3767-3773 (2005)

[49] F Pretorius Numerical relativity using a generalized harmonic decomposition Class Quantum

Grav 22 425-451 (2005)

[50] J D Brown Conformal invariance and the conformal-traceless decomposition of the gravita-

tional field Phys Rev D 71 104011 (2005) (12 pages)

[51] L Lindblom M A Scheel L E Kidder R Owen and O Rinne A new generalized harmonic

evolution system Class Quantum Grav 23 S447-S462 (2006)

[52] R Owen Constraint damping in first-order evolution systems for numerical relativity

Phys Rev D 76 044019 (2007) (11 pages)

[53] V Fock The Theory of Space Time and Gravitation (Pergamon Press New York 1959)

[54] httpwwwblack-holesorgSpEChtml

[55] N W Taylor L E Kidder and S A Teukolsky Spectral methods for the wave equation in

second-order form arXiv10052922v1[gr-qc]

[56] N W Taylor personal communication

[57] W Tichy Black hole evolution with the BSSN system by pseudospectral methods Phys Rev D

74 084005 (2006) (10 pages)

[58] W Tichy Long term black hole evolution with the BSSN system by pseudospectral methods

Phys Rev D 80 104034 (2009) (10 pages)

[59] A H Mroue personal correspondence

[60] J D Brown BSSN in spherical symmetry Class Quantum Grav 25 205004 (2008) (23

pages)

[61] T W Baumgarte and S G Naculich Analytical representation of a black hole puncture

33

solution Phys Rev D 75 067502 (2007) (4 pages)

[62] M Hannam S Husa F Ohme B Brugmann and N O Murchadha Wormholes and trum-

pets the Schwarzschild spacetime for the moving-puncture generation Phys Rev D 78 064020

(2008) (19 pages)

[63] J D Brown Probing the puncture for black hole simulations Phys Rev D 80 084042 (2009)

(24 pages)

[64] J D Brown Puncture evolution of Schwarzschild black holes Phys Rev D 77 044018 (2008)

(5 pages)

[65] J S Hesthaven S Gottlieb and D Gottlieb Spectral Methods for Time-Dependent Problems

(Cambridge University Press Cambridge UK 2007)

[66] J S Hesthaven and T Warburton Nodal Discontinuous Galerkin Methods Algorithms Anal-

ysis and Applications (Springer New York 2008)

[67] K Shahbazi An explicit expression for the penalty parameter of the interior penalty method

J Comp Phys 205 issue 2 401-407 (2005)

[68] Y Cheng and C-W Shu A discontinuous Galerkin finite element method for time dependent

partial differential equations with higher order derivatives Math Comp 77 no 262 699-730

(2007)

[69] M J Grote A Schneebeli and D Schotzau Discontinuous Galerkin finite element method

for the wave equation SIAM J Numer Anal 44 no 6 2408-2431 (2006)

[70] M J Grote A Schneebeli and D Schotzau Interior penalty discontinuous Galerkin method

for Maxwellrsquos equations Energy norm error estimates J Comp Appl Math 204 issue 2

375-386 (2007)

[71] J S Hesthaven and T Warburton Nodal High-Order Methods on Unstructured

Grids I Time-Domain Solution of Maxwellrsquos Equations J Comp Phys 181 issue 1 186-221

(2002)

[72] A Schneebeli Interior Penalty Discontinuous Galerkin Methods for Electromagnetic and

Acoustic Wave Equations PhD dissertation University of Basel 2006 Also available as

httpedocunibaschdissDissB_7760

[73] H Friedrich and A D Rendall The Cauchy Problem for the Einstein Equations Lec Notes

Phys 540 127-224 (2000) Also available as arXivgr-qc0002074v1

[74] C Gundlach and J M Martin-Garcia Symmetric hyperbolic form of systems of second-order

evolution equations subject to constraints Phys Rev D 70 044031 (2004) (14 pages)

[75] C Gundlach J M Martin-Garcia Hyperbolicity of second-order in space systems of evolution

equations Class Quantum Grav 23 S387-S404 (2006)

[76] H-O Kreiss and J Lorenz Initial-Boundary Value Problems and the Navier-Stokes Equations

SIAM Classics in Applied Mathematics (SIAM Philadelphia 2004)

[77] B Cockburn and C-W Shu The local discontinuous Galerkin method for time-dependent

convection-diffusion systems SIAM J Numer Anal 35 no 6 2440-2463 (1998)

[78] Y Xu C-W Shu Local discontinuous Galerkin methods for nonlinear Schrodinger equations

J Comp Phys 205 issue 1 72-97 (2005)

[79] Y Xu C-W Shu Local discontinuous Galerkin methods for the Kuramoto-Sivashinsky equa-

tions and the Ito-type coupled KdV equations Comp Methods Appl Mech Engin 195 3430-

3447 (2006)

[80] M Alcubierre B Brugmann P Diener M Koppitz D Pollney E Seidel and R Takahashi

Gauge conditions for long-term numerical black hole evolutions without excision Phys Rev D

67 084023 (2003) (18 pages)

34

[81] G H Golub and C F Van Loan Matrix Computations third edition (John Hopkins Univer-

sity Press Baltimore and London 1996)

[82] F Bassi and S Rebay A High-Order Accurate Discontinuous Finite Element Method for the

Numerical Solution of the Compressible Navier-Stokes Equations J Comp Phys 131 issue

2 267-279 (1997)

[83] S E Field J S Hesthaven and S R Lau Discontinuous Galerkin method for computing

gravitational waveforms from extreme mass ratio binaries Class Quantum Grav 26 165010

(2009) (28 pages)

[84] F Brezzi G Manzini D Marini P Pietra and A Russo Discontinuous finite elements for

diffusion problems in Atti Convegno in onore di F Brioschi (Milan 1997) 197-217 Istituto

Lombardo Accademia di Scienze e Lettere Milan Italy 1999

[85] D N Arnold F Brezzi B Cockburn and L D Marini Unified analysis of discontinuous

Galerkin methods for elliptic problems SIAM J Numer Anal 39 no 5 17491779 (2002)

[86] D Nunez and O Sarbach Boundary conditions for the Baumgarte-Shapiro-Shibata-Nakamura

formulation of Einsteinrsquos field equations Phys Rev D 81 044011 (2010) (16 pages)

[87] D Brown P Diener O Sarbach E Schnetter and M Tiglio Turduckening black holes An

analytical and computational study Phys Rev D 79 044023 (2009) (19 pages)

[88] D Brown O Sarbach E Schnetter M Tiglio P Diener I Hawke and D Pollney Excision

without excision Phys Rev D 76 081503(R) (2007) (5 pages)

[89] A Klockner T Warburton J Bridge and J S Hesthaven Nodal discontinuous Galerkin

methods on graphics processors J Comp Phys 228 issue 21 7863-7882 (2009)

• I Introduction
• II Spherically symmetric (generalized) BSSN equations
• • A Basic variables and spherically reduced system
  • B Abstract expressions of the system
  • C Hyperbolicity and characteristic fields
  • • III Discontinuous Galerkin Method
    • • A Local approximation of the system (13)
      • B Numerical Flux
      • C Matrix form of the semi-discrete equations
      • D Filtering
      • E Model system
      • • IV Results from numerical simulations
        • • A Simulations of the model system
          • B Simulations of the BSSN system
          • • V Conclusion
            • VI Acknowledgments
            • A Hyperbolicity of the first-order system
            • B Schwarzschild solution in conformal Kerr-Schild coordinates
            • C Stability of the model system
            • • 1 Analysis for a single interval
              • 2 Analysis for multiple intervals
              • • References