u.s. doe scidac partnership predictive computing for ... · jcp (2013), under review by . jctc...
TRANSCRIPT
![Page 1: U.S. DOE SciDAC Partnership Predictive Computing for ... · JCP (2013), under review by . JCTC (2013) FCIQMC AFQMC . Monte Carlo MP2 . Willow, Kim and Hirata, JCP (2012) 17 . Single](https://reader033.vdocuments.site/reader033/viewer/2022051909/5ffd48f744049357d5716a8c/html5/thumbnails/1.jpg)
U.S. DOE SciDAC Partnership
Predictive Computing for Condensed Matter
So Hirata (Chemistry) Peter Abbamonte (Physics) David Ceperley (Physics)
Shinsei Ryu (Physics) Lucas Wagner (Physics)
July, 2013 SciDAC-3 meeting 1
Shiwei Zhang (Physics)
Garnet Chan (Chemistry) Bryan Clark (Physics)
![Page 2: U.S. DOE SciDAC Partnership Predictive Computing for ... · JCP (2013), under review by . JCTC (2013) FCIQMC AFQMC . Monte Carlo MP2 . Willow, Kim and Hirata, JCP (2012) 17 . Single](https://reader033.vdocuments.site/reader033/viewer/2022051909/5ffd48f744049357d5716a8c/html5/thumbnails/2.jpg)
Frontiers of predictive computing
2
Accuracy
App
licab
ility
DFT MP2 CCSD FCI
Mol
ecul
es
C
lust
ers
Pol
ymer
s
Sol
ids
New theories and algorithms
Condensed matter
Size consistency Finite temperature, finite pressure
Metals, superconductors Strong correlation
Computational costs
A DOE report on Computational Materials Science (2010): “We are at the threshold of a new era where the integrated synthesis, characterization, and modeling of complex materials and chemical processes will transform our ability to understand and design new materials and chemistries with predictive power” Materials Genome Initiative for Global Competitiveness (2011): “the development of advanced materials can be accelerated through advances in computational techniques”
![Page 3: U.S. DOE SciDAC Partnership Predictive Computing for ... · JCP (2013), under review by . JCTC (2013) FCIQMC AFQMC . Monte Carlo MP2 . Willow, Kim and Hirata, JCP (2012) 17 . Single](https://reader033.vdocuments.site/reader033/viewer/2022051909/5ffd48f744049357d5716a8c/html5/thumbnails/3.jpg)
The team
3
![Page 4: U.S. DOE SciDAC Partnership Predictive Computing for ... · JCP (2013), under review by . JCTC (2013) FCIQMC AFQMC . Monte Carlo MP2 . Willow, Kim and Hirata, JCP (2012) 17 . Single](https://reader033.vdocuments.site/reader033/viewer/2022051909/5ffd48f744049357d5716a8c/html5/thumbnails/4.jpg)
Project goals
4
Theories (Finite-T) MP, CC, RPA, etc.; QMC; DMRG, TN; FCIQMC,
AFQMC, MC-MP2; DMFT
Software Algorithm redesign; massive parallelism; fast integrals
Applications – experimental verification Metals; (copper oxide) superconductors;
graphene; molecular crystals; Peierls systems and Luttinger liquid; Mott-Hubbard systems;
spin liquid Abbamonte
![Page 5: U.S. DOE SciDAC Partnership Predictive Computing for ... · JCP (2013), under review by . JCTC (2013) FCIQMC AFQMC . Monte Carlo MP2 . Willow, Kim and Hirata, JCP (2012) 17 . Single](https://reader033.vdocuments.site/reader033/viewer/2022051909/5ffd48f744049357d5716a8c/html5/thumbnails/5.jpg)
5
Wave function theory
Quantum Monte Carlo
Strong correlation
Finite-temperature MP2, CCD, RPA
Monte Carlo MP2
DMC, VMC TN (MERA, PEPS)
FCIQMC
AFQMC DMRG DMFT
Theories Hirata
Zhang
Ceperley
Wagner
Chan Clark
Ryu
![Page 6: U.S. DOE SciDAC Partnership Predictive Computing for ... · JCP (2013), under review by . JCTC (2013) FCIQMC AFQMC . Monte Carlo MP2 . Willow, Kim and Hirata, JCP (2012) 17 . Single](https://reader033.vdocuments.site/reader033/viewer/2022051909/5ffd48f744049357d5716a8c/html5/thumbnails/6.jpg)
Schrödinger equation for water
10-particle, 30-dimensional partial differential equation
2 2
2 2 2 2 2 22 1 1 1 sin
sin sin ii i i i i i i i ir r r r
θθ φ θ θ θ
∂ ∂ ∂ ∂ ∂+ + + ∂ ∂ ∂ ∂ ∂
Conditions arising from the indistinguishability of electrons ( )( )( )
1 2 3 4 5 6 7 8 9 10 1 2
2 1 3 4 5 6 7 8 9 10 1 2
3 1 2 4 5 6 7 8 9 10 1 2
3 2 1 4 5 6 7 8 9
, , , , , , , , , , , ,
, , , , , , , , , , , ,
, , , , , , , , , , , ,
, , , , , , , , ,
e e e e e e e e e e H H O
e e e e e e e e e e H H O
e e e e e e e e e e H H O
e e e e e e e e e e
Ψ =
−Ψ =
Ψ =
−Ψ
r r r r r r r r r r r r r
r r r r r r r r r r r r r
r r r r r r r r r r r r r
r r r r r r r r r r( )10 1 2, , , ...H H O =r r r
3,628,800 terms!
Singularities
Many-body
![Page 7: U.S. DOE SciDAC Partnership Predictive Computing for ... · JCP (2013), under review by . JCTC (2013) FCIQMC AFQMC . Monte Carlo MP2 . Willow, Kim and Hirata, JCP (2012) 17 . Single](https://reader033.vdocuments.site/reader033/viewer/2022051909/5ffd48f744049357d5716a8c/html5/thumbnails/7.jpg)
Systematic many-body methods
![Page 8: U.S. DOE SciDAC Partnership Predictive Computing for ... · JCP (2013), under review by . JCTC (2013) FCIQMC AFQMC . Monte Carlo MP2 . Willow, Kim and Hirata, JCP (2012) 17 . Single](https://reader033.vdocuments.site/reader033/viewer/2022051909/5ffd48f744049357d5716a8c/html5/thumbnails/8.jpg)
Laplace-transform MP2
MP2
Canonical MP2
• A member of a systematic series of approximations converging toward exactness.
• Can describe covalent, ionic, hydrogen-bond, and dispersion interactions.
• Size consistent and thus applicable to solids. • Accurate energy bands and band gaps. • The operation cost grows as O(n5), where n is
the number of orbitals. • The memory cost grows as O(n2) to O(n4). • Parallelization is difficult.
![Page 9: U.S. DOE SciDAC Partnership Predictive Computing for ... · JCP (2013), under review by . JCTC (2013) FCIQMC AFQMC . Monte Carlo MP2 . Willow, Kim and Hirata, JCP (2012) 17 . Single](https://reader033.vdocuments.site/reader033/viewer/2022051909/5ffd48f744049357d5716a8c/html5/thumbnails/9.jpg)
Systematic many-body methods
QMC QMC
![Page 10: U.S. DOE SciDAC Partnership Predictive Computing for ... · JCP (2013), under review by . JCTC (2013) FCIQMC AFQMC . Monte Carlo MP2 . Willow, Kim and Hirata, JCP (2012) 17 . Single](https://reader033.vdocuments.site/reader033/viewer/2022051909/5ffd48f744049357d5716a8c/html5/thumbnails/10.jpg)
Monte Carlo
10
weight
function Metropolis
Requirement 1: the weight function g is analytically integrable
Requirement 2: the weight function g behaves like the integrand f
![Page 11: U.S. DOE SciDAC Partnership Predictive Computing for ... · JCP (2013), under review by . JCTC (2013) FCIQMC AFQMC . Monte Carlo MP2 . Willow, Kim and Hirata, JCP (2012) 17 . Single](https://reader033.vdocuments.site/reader033/viewer/2022051909/5ffd48f744049357d5716a8c/html5/thumbnails/11.jpg)
Quantum Monte Carlo
11
Requirement 1: the weight function g is analytically integrable
Requirement 2: the weight function g behaves like the integrand f
weight function
• Gives the best wave function and energy in the variational sense, which is usually nearly exact.
• Easily and efficiently parallelized. • Nearly zero memory cost. • Has statistical errors, which decay slowly as
O(N−1/2) with the number of MC steps (N). • Calculations of structures and properties
(including energy bands) and applications to excited states difficult.
![Page 12: U.S. DOE SciDAC Partnership Predictive Computing for ... · JCP (2013), under review by . JCTC (2013) FCIQMC AFQMC . Monte Carlo MP2 . Willow, Kim and Hirata, JCP (2012) 17 . Single](https://reader033.vdocuments.site/reader033/viewer/2022051909/5ffd48f744049357d5716a8c/html5/thumbnails/12.jpg)
Contents
• Research highlights of 2012-2013 – QMC for copper oxide superconductors – Auxiliary field QMC for band gaps – Full CI QMC for excited states
• Monte Carlo MP2 and MP3 • MP2 for molecular crystals: solid-solid phase
transition in CO2
12
![Page 13: U.S. DOE SciDAC Partnership Predictive Computing for ... · JCP (2013), under review by . JCTC (2013) FCIQMC AFQMC . Monte Carlo MP2 . Willow, Kim and Hirata, JCP (2012) 17 . Single](https://reader033.vdocuments.site/reader033/viewer/2022051909/5ffd48f744049357d5716a8c/html5/thumbnails/13.jpg)
QMC for copper oxide superconductors Wagner and Abbamonte, to be published (2013)
13
Research Highlights of 2012-2013
Wagner
La2CuO4
Method J / eV Magnetic moment
DFT (LDA) 0.82 0.3
DFT (GGA) 0.6 0.5
QMC 0.15 0.62
Expt 0.13 0.6
For more details, please visit the poster by Lucas Wagner
![Page 14: U.S. DOE SciDAC Partnership Predictive Computing for ... · JCP (2013), under review by . JCTC (2013) FCIQMC AFQMC . Monte Carlo MP2 . Willow, Kim and Hirata, JCP (2012) 17 . Single](https://reader033.vdocuments.site/reader033/viewer/2022051909/5ffd48f744049357d5716a8c/html5/thumbnails/14.jpg)
Auxiliary field QMC for band gaps Zhang et al., to be published (2013)
14
Research Highlights of 2012-2013
Method Band gap / eV
DFT (LDA) 0.67
GW 4.8
QMC 3.97
Expt 3.7, 3.87, 4.0, 4.3
Zhang
NiO ZnO
Method Band gap / eV
DFT (LDA) 1.0
GW 2.4, 2.6, 2.8
QMC 3.26
Expt 3.3 – 3.57
![Page 15: U.S. DOE SciDAC Partnership Predictive Computing for ... · JCP (2013), under review by . JCTC (2013) FCIQMC AFQMC . Monte Carlo MP2 . Willow, Kim and Hirata, JCP (2012) 17 . Single](https://reader033.vdocuments.site/reader033/viewer/2022051909/5ffd48f744049357d5716a8c/html5/thumbnails/15.jpg)
Full CI QMC for excited states Booth and Chan, JCP Comm (2012)
15
Research Highlights of 2012-2013
Chan
Standard propagator: max at ground-state
Gaussian propagator: max at specified omega
![Page 16: U.S. DOE SciDAC Partnership Predictive Computing for ... · JCP (2013), under review by . JCTC (2013) FCIQMC AFQMC . Monte Carlo MP2 . Willow, Kim and Hirata, JCP (2012) 17 . Single](https://reader033.vdocuments.site/reader033/viewer/2022051909/5ffd48f744049357d5716a8c/html5/thumbnails/16.jpg)
16
Wave function theory
Quantum Monte Carlo
Strong correlation
MP2, MP3, CCSD, CCSD(T)
Monte Carlo MP
DMC, VMC
Monte Carlo MP Willow, Kim and Hirata, JCP (2012), JCP (2013), under review by JCTC (2013)
FCIQMC
AFQMC
![Page 17: U.S. DOE SciDAC Partnership Predictive Computing for ... · JCP (2013), under review by . JCTC (2013) FCIQMC AFQMC . Monte Carlo MP2 . Willow, Kim and Hirata, JCP (2012) 17 . Single](https://reader033.vdocuments.site/reader033/viewer/2022051909/5ffd48f744049357d5716a8c/html5/thumbnails/17.jpg)
Monte Carlo MP2 Willow, Kim and Hirata, JCP (2012)
17
Single 13-dimensional integral evaluated by Monte Carlo
Very long O(n4) summation of products of 2 x 6-dimensional integrals
Explicit two-electron integrals
Laplace transformation of the denominator
Change of orders of summations and integrations
![Page 18: U.S. DOE SciDAC Partnership Predictive Computing for ... · JCP (2013), under review by . JCTC (2013) FCIQMC AFQMC . Monte Carlo MP2 . Willow, Kim and Hirata, JCP (2012) 17 . Single](https://reader033.vdocuments.site/reader033/viewer/2022051909/5ffd48f744049357d5716a8c/html5/thumbnails/18.jpg)
Monte Carlo MP2 Willow, Kim and Hirata, JCP (2012)
18
weight
function
Requirement 1: analytically integrable
Requirement 2: cancellation of singularities
Singularities
Metropolis
![Page 19: U.S. DOE SciDAC Partnership Predictive Computing for ... · JCP (2013), under review by . JCTC (2013) FCIQMC AFQMC . Monte Carlo MP2 . Willow, Kim and Hirata, JCP (2012) 17 . Single](https://reader033.vdocuments.site/reader033/viewer/2022051909/5ffd48f744049357d5716a8c/html5/thumbnails/19.jpg)
19
Monte Carlo MP2 Willow, Kim and Hirata, JCP (2012)
Nitrogen 6-31G** Water monomer, dimer, trimer 6-31G**
![Page 20: U.S. DOE SciDAC Partnership Predictive Computing for ... · JCP (2013), under review by . JCTC (2013) FCIQMC AFQMC . Monte Carlo MP2 . Willow, Kim and Hirata, JCP (2012) 17 . Single](https://reader033.vdocuments.site/reader033/viewer/2022051909/5ffd48f744049357d5716a8c/html5/thumbnails/20.jpg)
20
HOMO & HOMO−1 of H2O Efficiency
Dyson self-energies (quasiparticle energies) for multiple states obtainable from a single MC run
Size (number of orbitals)
Monte Carlo MP2 for self-energies Willow, Kim and Hirata, JCP (2013)
![Page 21: U.S. DOE SciDAC Partnership Predictive Computing for ... · JCP (2013), under review by . JCTC (2013) FCIQMC AFQMC . Monte Carlo MP2 . Willow, Kim and Hirata, JCP (2012) 17 . Single](https://reader033.vdocuments.site/reader033/viewer/2022051909/5ffd48f744049357d5716a8c/html5/thumbnails/21.jpg)
Real-space Green’s function Willow, Kim and Hirata, to be published (2013)
![Page 22: U.S. DOE SciDAC Partnership Predictive Computing for ... · JCP (2013), under review by . JCTC (2013) FCIQMC AFQMC . Monte Carlo MP2 . Willow, Kim and Hirata, JCP (2012) 17 . Single](https://reader033.vdocuments.site/reader033/viewer/2022051909/5ffd48f744049357d5716a8c/html5/thumbnails/22.jpg)
22
Monte Carlo MP3 Willow, Kim and Hirata, to be published (2013)
Water 6-31G**
![Page 23: U.S. DOE SciDAC Partnership Predictive Computing for ... · JCP (2013), under review by . JCTC (2013) FCIQMC AFQMC . Monte Carlo MP2 . Willow, Kim and Hirata, JCP (2012) 17 . Single](https://reader033.vdocuments.site/reader033/viewer/2022051909/5ffd48f744049357d5716a8c/html5/thumbnails/23.jpg)
23
Redundant-walker MC-MP2 Willow, Hermes, Kim and Hirata, under review by JCTC (2013)
Benzene 6-31G**
7 hours
23 hours 59 hours
m = 2: 2 electron walker pairs
m = 10: 10 electron walker pairs Cost: 5 times
Benefit: 10 x 9 / 2 = 45 times MC steps
![Page 24: U.S. DOE SciDAC Partnership Predictive Computing for ... · JCP (2013), under review by . JCTC (2013) FCIQMC AFQMC . Monte Carlo MP2 . Willow, Kim and Hirata, JCP (2012) 17 . Single](https://reader033.vdocuments.site/reader033/viewer/2022051909/5ffd48f744049357d5716a8c/html5/thumbnails/24.jpg)
24
Massively parallel MC-MP2 Willow, Hermes, Kim and Hirata, under review by JCTC (2013)
C60 cc-pVDZ on 320 processors of Blue Waters
m = 120: 120 electron walker pairs Cost: 60 times
Benefit: 120 x 119 / 2 = 7140 times MC steps
![Page 25: U.S. DOE SciDAC Partnership Predictive Computing for ... · JCP (2013), under review by . JCTC (2013) FCIQMC AFQMC . Monte Carlo MP2 . Willow, Kim and Hirata, JCP (2012) 17 . Single](https://reader033.vdocuments.site/reader033/viewer/2022051909/5ffd48f744049357d5716a8c/html5/thumbnails/25.jpg)
MP2 for molecular crystals
25
![Page 26: U.S. DOE SciDAC Partnership Predictive Computing for ... · JCP (2013), under review by . JCTC (2013) FCIQMC AFQMC . Monte Carlo MP2 . Willow, Kim and Hirata, JCP (2012) 17 . Single](https://reader033.vdocuments.site/reader033/viewer/2022051909/5ffd48f744049357d5716a8c/html5/thumbnails/26.jpg)
Embedded-fragment approach Hirata et al., MP (2005); Kamiya, Hirata, and Valiev, JCP (2008)
Pair energy in the presence of self-consistent atomic charges or dipoles
N-body (N > 2) Coulomb in point-charge or dipole approximation
1 and 2-body Coulomb Exchange Correlation
![Page 27: U.S. DOE SciDAC Partnership Predictive Computing for ... · JCP (2013), under review by . JCTC (2013) FCIQMC AFQMC . Monte Carlo MP2 . Willow, Kim and Hirata, JCP (2012) 17 . Single](https://reader033.vdocuments.site/reader033/viewer/2022051909/5ffd48f744049357d5716a8c/html5/thumbnails/27.jpg)
Molecular crystals Hirata, JCP (2008)
{ } ( ) { }(0) (0) (0) (0) (0) 0 (0) ( ) (0) ( ),
1 12
L
i i j i j m i j m i j mi i j m L i j
E E E E E E E Eδ+
< =−
= + − − + − − −∑ ∑ ∑ ∑
( )(0) (0) (0) (0) (0) (0) ( ) (0) ( )0
,
1 12
Li i j i j i j m i j m
mi i j m L i j
E E E E E E EEx x x x x x x x
δ+
< =−
∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂≈ + − − + − − − ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∑ ∑ ∑ ∑
(0) ( ) ( ) LR( ) ( )
,
12
Li j m j m
j m j mm L i j
E EE Ema z z aγ γ γ
+
=−
∂ ∂∂ ∂ ≈ − + ∂ ∂ ∂ ∂ ∑ ∑∑
Energy per unit cell
Energy gradients
Energy Hessian
( )2 2 2 2 2 2 22
(0) (0) (0) (0) (0) (0) ( ) (0) ( )0
,
1 12
Li i j i j i j m i j m
mi i j m L i j
E E E E E E EEx y x y x y x y x y x y x y x y
δ+
< =−
∂ ∂ ∂ ∂ ∂ ∂ ∂∂≈ + − − + − − − ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∑ ∑ ∑ ∑
Linear scaling regardless of gradients algorithms (analytical vs. numerical) Long-range electrostatic correction essential for cell gradients
x and y need not be in-phase coordinates: phonon dispersions can be obtained
![Page 28: U.S. DOE SciDAC Partnership Predictive Computing for ... · JCP (2013), under review by . JCTC (2013) FCIQMC AFQMC . Monte Carlo MP2 . Willow, Kim and Hirata, JCP (2012) 17 . Single](https://reader033.vdocuments.site/reader033/viewer/2022051909/5ffd48f744049357d5716a8c/html5/thumbnails/28.jpg)
Finite pressures and temperatures Li, Sode, Voth and Hirata, under review by Nature Comm (2013)
Gibbs free energy per unit cell
Partition function
Zero-point energy, entropy, and heat capacity
For an orthorhombic unit cell
![Page 29: U.S. DOE SciDAC Partnership Predictive Computing for ... · JCP (2013), under review by . JCTC (2013) FCIQMC AFQMC . Monte Carlo MP2 . Willow, Kim and Hirata, JCP (2012) 17 . Single](https://reader033.vdocuments.site/reader033/viewer/2022051909/5ffd48f744049357d5716a8c/html5/thumbnails/29.jpg)
Phase transition in solid CO2 Li, Sode, Voth and Hirata, under review by Nature Comm (2013)
29
296 K
![Page 30: U.S. DOE SciDAC Partnership Predictive Computing for ... · JCP (2013), under review by . JCTC (2013) FCIQMC AFQMC . Monte Carlo MP2 . Willow, Kim and Hirata, JCP (2012) 17 . Single](https://reader033.vdocuments.site/reader033/viewer/2022051909/5ffd48f744049357d5716a8c/html5/thumbnails/30.jpg)
Phase transition in solid CO2 Li, Sode, Voth and Hirata, under review by Nature Comm (2013)
30
![Page 31: U.S. DOE SciDAC Partnership Predictive Computing for ... · JCP (2013), under review by . JCTC (2013) FCIQMC AFQMC . Monte Carlo MP2 . Willow, Kim and Hirata, JCP (2012) 17 . Single](https://reader033.vdocuments.site/reader033/viewer/2022051909/5ffd48f744049357d5716a8c/html5/thumbnails/31.jpg)
Phase transition in solid CO2 Li, Sode, Voth and Hirata, under review by Nature Comm (2013)
31
14.5 GPa
18.0 GPa
![Page 32: U.S. DOE SciDAC Partnership Predictive Computing for ... · JCP (2013), under review by . JCTC (2013) FCIQMC AFQMC . Monte Carlo MP2 . Willow, Kim and Hirata, JCP (2012) 17 . Single](https://reader033.vdocuments.site/reader033/viewer/2022051909/5ffd48f744049357d5716a8c/html5/thumbnails/32.jpg)
Phase transition in solid CO2 Li, Sode, Voth and Hirata, under review by Nature Comm (2013)
32
![Page 33: U.S. DOE SciDAC Partnership Predictive Computing for ... · JCP (2013), under review by . JCTC (2013) FCIQMC AFQMC . Monte Carlo MP2 . Willow, Kim and Hirata, JCP (2012) 17 . Single](https://reader033.vdocuments.site/reader033/viewer/2022051909/5ffd48f744049357d5716a8c/html5/thumbnails/33.jpg)
Acknowledgements
33 For more information, please visit http://www.predictive-scidac.org
Dr. James Davenport, Dr. Mark Pederson, and Dr. Ceren Susut