Towards PPCC So Hirata Quantum Theory Project University of Florida Sanibel Symposium 2008, Parallel...

13
Towards PPCC So Hirata Quantum Theory Project University of Florida Sanibel Symposium 2008, Parallel CC Workshop

Transcript of Towards PPCC So Hirata Quantum Theory Project University of Florida Sanibel Symposium 2008, Parallel...

Towards PPCC

So HirataQuantum Theory Project

University of Florida

Sanibel Symposium 2008, Parallel CC Workshop

Issues at petascale Increased complexity of algorithms and

implementations – computerized derivation and implementation / optimization, portability, extensibility, maintainability / parallel middleware

Multilevel hierarchical parallelism – hierarchical structures of chemical computing / linear scaling, PES scan, ab initio MD/MC

Fault tolerance, restart capability, calculation log, and verification – parallel middleware / compartmentation of calculations / redundant calculations / hierarchical methods

Automated symbolic algebraHirata, JPCA (2003); Hirata, TCA (2006); Hirata, JP Conf. Ser. (2006)

Definition of a many-electron theory 0 1 2 0 1 2 0 1 2 0exp ; 0 exp ; 0 expa ab

i ijC C CE H T T H T T H T T

Mathematical expressions

A parallel computer program

Automated symbolic algebraHirata, JPCA (2003); Hirata, TCA (2006); Hirata, JP Conf. Ser. (2006)

Agent Smith“The Matrix”

Correct scaling by factorized, reusable intermediates

Spin, point-group, and permutation symmetries

Dynamic load balancing parallelism; scalable

Runtime adjustment of memory usage

Never send a human to do a machine’s job

Implemented methodsHirata JPCA (2003); Hirata TCA (2006); Hirata JP Conf Ser (2006)

CI

PTCC Combined CC+PT

CCSD(T)CCSD(2)T, CCSD(3)T

CCSD(2)TQ, CCSD(3)TQ

CCSDT(2)Q, CR-CCSD(T)Hirata et al. JCP (2004)

Shiozaki, Hirao & Hirata JCP (2007)

Linear ExpansionCIS, CISD, CISDT, CISDTQ

Hirata JPCA (2003)

CIS+perturbationCIS(D), CIS(3), CIS(4)

Hirata JCP (2005)

PerturbationMP2, MP3, MP4

Hirata JPCA (2003)

Excited State TheoriesEOM-CCSD

EOM-CCSDTEOM-CCSDTQ

Hirata JCP (2004)

Ionization TheoryIP-EOM-CCSD

IP-EOM-CCSDTIP-EOM-CCSDTQKamiya & Hirata JCP (2006)

Cluster ExpansionCCD, CCSD, CCSDT,

CCSDTQ, LCCD,LCCSD, QCISD

Hirata JPCA (2003)

Electron Attachment TheoryEA-EOM-CCSD

EA-EOM-CCSDTEA-EOM-CCSDTQ

Kamiya & Hirata (2007)

EOM-CC+perturbationEOM-CCSD(2)T, EOM-CCSD(2)TQ

EOM-CCSD(3)TShiozaki, Hirao & Hirata JCP (2007)

Other CIS+2nd orderD-CIS(2), SCS-CIS(D)

SOS-CIS(D)Fan & Hirata (2007)

CC-R12, EOM-CC-R12, Λ-CC-R12Shiozaki, Kamiya, Hirata, & Valeev, in preparation (2008)

Toru ShiozakiUniversity of Florida

New types of ansatz – new symbolic algabra codeSignificantly more complex equationsLonger computational sequencesMultiple hotspotsPossibilities of various approximations

CC-R12

R12-CCSD on 100 processors ~ CCSD on 100,000 processors

Linear scaling CC

Linear scaling CCSD on 1 processor ~ CCSD on 100,000 processors

Fast methods for water clustersHirata et al. MP (2005)

n n

ij i j ii j i

E E E E E

Pair energy in the presence of dipole field

N-body (N > 2) Coulomb in dipole-dipole approximation

1 and 2-bodyCoulombExchangeCorrelation

Fast methods: excited statesHirata et al. MP (2005)

A record EOM-CCSD aug-cc-pVDZ calculation for a 247-atom system

0 5 10 15 20

Number of monomer units

CP

U t

ime

(arb

itrar

y un

its)

Conventional

Pair interaction (no cutoff)

Pair interaction (cutoff at 30 bohr)

Constant scaling!

Coupled-cluster for solids

Maddox (Nature, 1988): “One of the continuing scandals in the physical sciences is that it remains in general impossible to predict the structure of even the simplest crystalline solids from a knowledge of their chemical composition. … Solids such as crystalline water (ice) are still thought to lie beyond mortals’ ken.”

1 and 2 type solid formic acid

Molecular crystals

Ice XI

Formamide

Hydrogen fluoride