the lattice gross-neven model and the langevin algorithm

5
Nuclear Physics B (Proc. Suppl.) 4 (1988) 595-599 595 North-Holland, Amsterdam THE LATTICE GROSS-NEVEU MODEL AND THE LANGEVIN ALGORITHM N. ATTIG*, R. LACAZE +, A. MOREL + , B. PETERSSON* and M. WOLFF* * Fakult~t fur Physik, Universit~t Bielefeld, D-4800 Bielefeld i, F.R. Germany + Service de Physique Th#orique, CEN-Saclay, F-91191 Gif-sur-Yvette Cedex, France The analytic solution of the Gross-Neveu model on the lattice is given including order 1/N. It is compared with a high statistics numerical simulation using the Langevin algorithm. 1. INTRODUCTION L = The motivations for studying the Gross-Neveu model on the lattice is both the interest in the model per se, and the use of it as a testing ground for numerical algorithms with fermions on the lattice. It is a non-trivial model with fer- mions, having a discrete chiral invariance spon- taneously broken for large enough number of fla- vours of fermi ons nf. The model is asymptotically free and analyti- cally solvable for large nf 1'2. For nf = ~ it 3-5 has a finite temperature phase transition Of course, the model also has particular fea- tures not directly connected to theories like QCD: it is constrained to two dimensions and contains no gauge fields. Earlier works on the lattice aspects of the model are listed in Refs. 6-9. The present in- 1- vestigation is the first to extend the analytic Tc =~ a exp(y E) , 1/nf expansion to the lattice regularization. Furthermore, we have performed a numerical simu, lation with much higher statistics than previous investigations, which makes it possible to esti- m~ = o , mate up to (i/nf) 2 effects. We have used the first order Langevin algorithm. In fact, one of the major goals of the investigation is to study the dependence of the results of this algorithm on the discrete time step. 2. THE ANALYTIC SOLUTION In the continuum version, the Lagrangian of the model is nf g2 nf (z=l c~=l (i) where the ~a are n, two-component spinors. A l . nff ,-rCt ~ scalar field o(x) conjugate to ~2-1= ~ ~ Is in- troduced I and after the formal integration over the fermion fields one obtains o(x) 2 Sef f = nf {S--~-- d2x + Tr In(3+o(x))] , (2) where ~ = g2 nf and the partition function is Z = f [d o(x)] exp [-Sef f] (3) If the theory is regularized e.g. by a spherical momentum cut-off A, one can make a saddle point expansion of Z, giving to leading order <o> ~ o = A e -T~/~ (4) and the critical temperature (5) where YE = .5772 is the Euler constant. A semi- 2 classical analysis gives the mass spectrum m n = n~ sin(n~/2nf) / (nl~J2nf) , m k = 2nf o/ T~ , (6) wehre m k is the kink-antikink mass. A lattice version of the model with Susskind fermions was first introduced in Ref. 6 and fur- ther studied in Ref. 7. The action is defined to be S=N{s~ -~ a + s Xx QxY x"}s ' (7) x x,y (x 0920-5632/88/$03.50 © Elsevier Science Publishers B.V. (North-Holland Physics Publishing Division)

Upload: atsushi-nakamura

Post on 02-Sep-2016

220 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: The lattice Gross-Neven model and the Langevin algorithm

Nuclear Physics B (Proc. Suppl.) 4 (1988) 595-599 595 North-Holland, Amsterdam

THE LATTICE GROSS-NEVEU MODEL AND THE LANGEVIN ALGORITHM

N. ATTIG*, R. LACAZE +, A. MOREL + , B. PETERSSON* and M. WOLFF*

* Fakul t~t fur Physik, Un ivers i t~ t B i e l e f e l d , D-4800 B i e l e f e l d i , F.R. Germany + Service de Physique Th#orique, CEN-Saclay, F-91191 Gi f -sur -Yvet te Cedex, France

The ana ly t i c so lu t ion o f the Gross-Neveu model on the l a t t i c e is given inc lud ing order 1/N. I t is compared wi th a high s t a t i s t i c s numerical s imulat ion using the Langevin a lgor i thm.

1. INTRODUCTION L =

The mot ivat ions for studying the Gross-Neveu

model on the l a t t i c e is both the i n t e res t in the

model per se, and the use of i t as a tes t ing

ground fo r numerical algor i thms wi th fermions on

the l a t t i c e . I t is a n o n - t r i v i a l model wi th fe r -

mions, having a d iscrete ch i ra l invar iance spon-

taneously broken for large enough number of f l a -

vours of fermi ons nf.

The model is asympto t ica l ly f ree and a n a l y t i -

ca l l y so lvable fo r large n f 1'2. For nf = ~ i t 3-5

has a f i n i t e temperature phase t r a n s i t i o n

Of course, the model also has p a r t i c u l a r fea-

tures not d i r e c t l y connected to theor ies l i ke

QCD: i t is constrained to two dimensions and

contains no gauge f i e l d s .

E a r l i e r works on the l a t t i c e aspects of the

model are l i s t e d in Refs. 6-9. The present in - 1 -

ves t iga t ion is the f i r s t to extend the ana ly t i c T c = ~ a exp(y E) ,

1 /nf expansion to the l a t t i c e regu la r i za t i on .

Furthermore, we have performed a numerical simu,

l a t i on wi th much higher s t a t i s t i c s than previous

i nves t i ga t i ons , which makes i t possible to e s t i - m~ = o ,

mate up to ( i / n f ) 2 e f fec ts . We have used the

f i r s t order Langevin a lgor i thm. In fac t , one of

the major goals of the i nves t i ga t i on is to study

the dependence o f the resu l ts o f th is a lgor i thm

on the d iscrete time step.

2. THE ANALYTIC SOLUTION

In the continuum vers ion, the Lagrangian o f

the model is

nf g2 nf

(z=l c~=l ( i )

where the ~a are n , two-component spinors. A l . n f f , - rCt ~ • •

sca lar f i e l d o(x) conjugate to ~2-1= ~ ~ Is in -

troduced I and a f t e r the formal i n teg ra t i on over

the fermion f i e l ds one obtains

o(x) 2 Sef f = nf { S - - ~ - - d2x + Tr I n ( 3 + o ( x ) ) ] , (2)

where ~ = g2 nf and the p a r t i t i o n funct ion is

Z = f [d o (x ) ] exp [ -Sef f ] (3)

I f the theory is regu lar ized e.g. by a spher ical

momentum c u t - o f f A, one can make a saddle po in t

expansion of Z, g iv ing to leading order

< o > ~ o = A e -T~/~ ( 4 )

and the c r i t i c a l temperature

(5)

where YE = .5772 is the Euler constant. A semi- 2 c lass ica l analysis gives the mass spectrum

m n = n~ s in(n~/2nf ) / (nl~J2nf) ,

m k = 2nf o / T~ , (6)

wehre m k is the k ink -an t i k ink mass.

A l a t t i c e version of the model wi th Susskind

fermions was f i r s t introduced in Ref. 6 and fu r -

ther studied in Ref. 7. The act ion is def ined to be

S = N { s ~ - ~ a + s Xx QxY x"}s ' (7) x x,y (x

0920-5632/88/$03.50 © Elsevier Science Publishers B.V. (North-Holland Physics Publishing Division)

Page 2: The lattice Gross-Neven model and the Langevin algorithm

596 N. Attig et al. / The lattice Gross-Neveu model

- - 0 . where ×x and < are two independent N-component

Grassmann var iables per s i t e . The matr ix is

Qxy = ½ [6x,y+l + 6x,y-1] +

+ ½(- l)X1[6x,y+~ - 6x,y_~] +

+ ~ 6xy [Ox+Ox_~+Ox_~+Ox_~_~] , (8)

where x = ( x l , x2 ) and y are two points on the

two-d i~ns iona l l a t t i c e and I and 2 uni t vectors

in the resp. d i rec t ions .

As was discussed in Refs. 6, 7 in t h e l / n f e x -

pansion the continuum l i m i t of the model is the

Gross-Neveu model with

nf = 2N (9)

Note that our I = g2N = ~ / 2 . On an i n f i n i t e sym-

metric l a t t i c e , the saddle point equation is

I 1 d2p (10) l - ~ f sin2p 1+sin2p 2+~-2 ,

where ~ is a constant. The in tegra l can in fact

be performed exact ly to give

1 1 (½, ~; 1; 1 X = v~2(~2+2) F ~2(~2+2)) (11)

The strong coupling l i m i t (o large) gives

= v~ (12)

and for weak coupling

o ~ Oas = 2 ~ e -Tt/2/ (13)

containing the expected factor 2 in the exponent,

and giving also the lat t ice scale. In Figure 1

we show the exact solution (11) compared with

the results from strong coupling (12) and weak

coupling (13). In Figure 2 we show °/°as" One

sees a strong "crossover" effect followed by a

slow approach to asymptotic scaling.

We have also calculated the I/N corrections

to 5, by developing in the fluctuations around

the saddle point. The calculation is rather

lengthy and w i l l be reported in Ref. 10. We get

on an in f in i te lat t ice

G

N = ~

5 0 ~

20 ng coupling 1.0 "%_,. I° * o ~

° °l°%m e • O5

0 2

I I I 0 0 5 1 0 1S

1/A

FI GURE 1

a/Gas

1.(

0 c. I I 0 5 1

I 0 5

N = ~

I 2 I 10

I/G i 15

1/.l

FIGURE 2

I d2P 9~'(P) [ ~ + ~.(p) ] - ' (14) < o > = ~ - ~T S W E F --Ca--

where

: ( l + c o s p , ) (1+cosp2) _~ T2-~-~ - x

2 k k ~2 _ )~ sin -~ sin (-2 ~--pot )

(~= 1 x 2 k 2 k (15)

( S sin2~2~+~2 ) ( 7, sin 2 ~=1 ~=1 ( T - Pa) +°2)

For comparison with the Lanqevin data we have

Page 3: The lattice Gross-Neven model and the Langevin algorithm

N. Attig et al. / The lattice Gross-Neoeu model 597

performed the in teg ra l s (or sums on a f i n i t e

l a t t i c e ) numer ica l ly .

3. THE NUMERICAL SIMULATION

To simulate the e f f e c t i v e act ion

Sef f = N ( ~ g ~ - Tr logQ) (16) X

we used the f i r s t order Langevin a lgor i thm with

an ext ra noise for the fermion par t

o T Q-l 3Q ~n ) Ox(n+ 1) = Ox(n ) - e N(-~-- ½ 5n ' ~

+ ~ q n ' (17)

where

<~n ~n' > = <qn qn' > = 26nn' (18)

The a lgor i thm is chosen to be as s im i l a r as pos-

s ib le to the Langevin algori thms fo r QCD, be-

cause one of the main object ives was to tes t

th is a lgor i thm and the E-dependence of physical

quan t i t i es . We have therefore not used features

special to two dimensions.

The numerical s imula t ion consists of runs at

N = 6, 12, 24 and 60 wi th .08 < 1/X < 2, with

l a t t i c e s of s ize 20x20 and 40x40 , and fo r f i -

n i te temperatures also 6x60 . The data were ta-

ken fo r various values of the time step ~, with

150,000- 300,000 sweeps fo r each po in t . The ma-

t r i x invers ion ~ = Q-I~ was done by the conju-

gate gradient a lgor i thm, We demanded that the

to ta l absolute value o f the rest vector

IIQ ~ - El i < .01 (19)

Errors were c a r e f u l l y monitored by the binning

method, so that co r re la t i ons in Langevin time

were taken in to account.

In Figure 3 we show < o > for N = 60 and for

two values of X as a funct ion of the time step

e. At ~ = 0 we have also p lo t ted the values com-

ing from the ana ly t i c expression on the l a t t i c e

to order I /N. The agreement wi th the ex t rapo la -

t ion of the numerical data is qu i te good. In the

strong coupl ing region the approach to c = 0 is

G

1 761 N =60

A=4 O

• e

02'

0 26

A=07 #

I I 1 0.001 0 003

FIGURE 3

from above, fo r weak coupl ing from below. This

can be understood from a per tu rba t ion study in E

of the Langevin a lgor i thm along the l ines of

Ref. I i . Deta i ls of th is study w i l l be given in

Ref. 10.

<lal>

0 5 1 - u

05(

OZ~!

O~E O,~

O,Z~7

20x20 A=I.0

N =60 D

N =2#

N=12 x

L 0.06

I 0.12

N=6 o I

018 CN

FIGURE 4

Page 4: The lattice Gross-Neven model and the Langevin algorithm

598 N. Attig et aL / The lattice Gross-Neveu model

In Figure 4 we show < o > fo r ~ = 1 and fo r

several values of N as a func t ion o f oN. Again

the open po in t a t c = 0 is taken from an analy-

t i c c a l c u l a t i o n . Ex t rapo la t i ng our numerical da-

ta l i n e a r l y in ~ and then l i n e a r l y in I/N fo r

N = 12, 24 and 60 we get fo r N== < o > = . 5 1 1 ,

in e x c e l l e n t agreement w i th the ana l y t i c r e s u l t

fo r N = =, which is < o > = .5114. For N = 6

(n f = 12) the I/N 2 co r rec t i on which we have not

evaluated a n a l y t i c a l l y is as expected s i g n i f i -

can t ly outs ide our numerical e r ro r s .

<1~1>

06 o

OL.

02

O0 08

6 , 6 0

N : 6 0 e : 0 0001

0

0

0

0

o l I I I I 1 I

10 1.2 1/* l / A

FIGURE 5a

<1~1)

05

03

6x60

o N : 6

o ~ : 0 0005 o

+

#~o o 01 o

I I I I I I I 08 10 1.2 1~

FIGURE 5b

In Figures 5a and b is shown < Io i> on a 6 x 6 0

l a t t i c e f o r N = 60 and 6 r espec t i ve l y . Each

po in t corresponds to 300,000 sweeps. For N = 60

we see the expected behaviour corresponding to

a second order phase t r a n s i t i o n . One can e s t i -

mate 1/~ c = 1.39. For N = =, the a n a l y t i c ca l -

cu la t i on gives I / ~ c = 1.43. The d i f f e rence is

compat ib le w i th a 1 /N-cor rec t ion .

For N = 6 we get instead s t rong ly f l u c t u a t -

ing values f o r o, in disagreement w i th Ref. 9.

We are present ly i n v e s t i g a t i n g i f the e f f e c t we

see is due to k i n k - a n t i k i n k con f i gu ra t i ons , pre-

sent at f i n i t e N.

4. CONCLUSIONS

From our h i g h - s t a t i s t i c s s imu la t ion we see

tha t the behaviour o f the order parameter < o >

is in agreement w i th the a n a l y t i c 1/N expansion

on the l a t t i c e down to N = 6 (n f = 12). For the

c r i t i c a l temperature the s i t u a t i o n is more com-

p l i c a t e d . The data at N = 6 look l i k e those

expected fo r a f i r s t order t r a n s i t i o n or might

be due to K ink -an t i k i nk s ta tes .

This quest ion and the ~-dependence of the Lange-

v in a lgo r i t hm w i l l be f u r t h e r discussed in Ref.

i0 .

REFERENCES

1 .

2.

3 .

4.

5.

6 .

7.

8 .

9.

D.J. Gross and A. Neveu, Phys. Rev. DIO (1974) 3235.

R. Dashen, B. Hasslacher and A. Neveu, Phys. Rev. D12 (1975) 2443.

L. Jacobs, Phys. Rev. DIO (1974) 3956.

B.J. Horr ington and A. Y i l d i z , Phys. Rev. D l l (1975) 779.

R.F. Dashen, S. Ma and R. Rajaraman, Phys. Rev. D l l (1975) 1499.

Y. Cohen, S. E l i t z u r and E. Rab inov ic i , Nucl. Phys. B220 [FS8] (1983) 102.

T. J o l i c o e u r , A. Morel and B. Petersson, Nucl. Phys. B274 (1986) 225; A. Morel, i n : La t t i ce Gauge Theory '86, eds. H. Satz e t a l . (Plenum Press 1987) p. 245.

T. Jo l i coeu r , Phys. Le t t . B171 (1986) 431.

F. Karsch, J. Kogut and H.W. Wyld, l l l i n o i s p r e p r i n t ILL-(TH)-86-36 (1986)

Page 5: The lattice Gross-Neven model and the Langevin algorithm

N. Attig et al. / The lattice Gross-Neveu model 599

I0. N. A t t i g , R. Lacaze, A. Morel, B. Petersson and M. Wolff , to appear.

11. C.G. Batrauni, G.R. Katz, A.S. Kronfeld, G.P. Lepage, B. Svet i tsky and K. Wilson, Phys. Rev. D32 (1985) 2736,

Part of th is work was supported by the

Commission of the European Communities

(Twining contract n ° 86300272 FROI*PU*JUI).