[molpro-user] Both of RS2 and DF-RS2 are very slow
Hans-Joachim Werner
werner at theochem.uni-stuttgart.de
Fri Sep 6 07:53:05 BST 2013
The problem is that RS2 is based on our "old" mrci program, in which inactive (closed-shell) orbitals
are treated on equal footing with active orbitals, and singly external CSFs are uncontracted. Due to
the very large number of uncontracted CSFs this becomes very expensive for large active spaces combined
with many closed shells. The newer RS2C program [JCP 112, 5546 (2000)] is VERY much faster for such cases,
but unfortunately there are no gradients yet for this implemented. It is on our high priority list to
extend RS2C to do gradients and properties as soon as we can.
Joachim Werner
--
Prof. Dr. Hans-Joachim Werner
Institute for Theoretical Chemistry
University of Stuttgart
Pfaffenwaldring 55
D-70569 Stuttgart, Germany
Tel.: (0049) 711 / 685 64400
Fax.: (0049) 711 / 685 64442
e-mail: werner at theochem.uni-stuttgart.de
Am 05.09.2013 um 21:41 schrieb Kenny W <minjiwu at outlook.com>:
> Dear Molpro Experts,
>
> I tried to calculate RS2 (CASPT2) gradient using a converged CAS(10,10) wavefunction for a molecule in Ci symmetry. The basis sets used were def2-TZVP (and def2-TZVP/JKFIT for density fitting). The CASSCF part converged nicely, however, the calculation stuck at RS2 part for a long time without any further output. I issued 24GB of memory and 16 cpu OpenMP parallel on a dual E5-2680 computer so this should not be a resource problem. Also another NEVPT2 calculation based on the same CASSCF wavefunction without density fitting was able to finish normally without any error, while non-DF RS2 based on the same CASSCF calculation also stuck.
>
> The input file is:
> ***,MOLPRO CASPT2 Job
> memory,3000,m
> start;
> symmetry,auto
> bohr
> geomtyp=xyz
> geometry={
> 14
> MOLPRO Job DF-RS2
> .......
> ..... molecular specifications in XYZ
> .......
> }
>
> basis={default=def2-TZVP
> set,fit
> default,def2-TZVP/JKFIT}
> {cfit,basis=fit}
>
> {MATROP ! read orbitals from previous CASSCF
> read,matroporb,file=rs2.nat
> save,matroporb,2140.2,orbitals}
>
> !CAS(10,10)
> {df-multi,energy=1.0e-8,gradient=1.0e-5,step=1.0e-5
> maxit,40
> occ,27,25; ! RHF occ 24,23
> closed,21,21;
> wf,94,1,0;
> start,2140.2
> natorb,2141.2,print=20,ci,state=1.1
> }
>
> {df-RS2,thrden=1.0d-11,thrvar=1.0d-11,maxit=100,maxiti=100
> occ,27,25;
> closed,21,21;
> wf,94,1,0;
> orbit,2141.2;
> state,1,1;
> }
>
> forces;
> ---
>
> The CASSCF(10,10) part finished without any problems.
> NUCLEAR CHARGE: 94
> NUMBER OF PRIMITIVE AOS: 558
> NUMBER OF SYMMETRY AOS: 508
> NUMBER OF CONTRACTIONS: 312 ( 156Ag + 156Au )
> NUMBER OF CORE ORBITALS: 24 ( 12Ag + 12Au )
> NUMBER OF VALENCE ORBITALS: 42 ( 21Ag + 21Au )
>
> NUCLEAR REPULSION ENERGY 828.85467872
>
> Basis set: FIT
> Basis size: 1094
> Attributes:
> Sphericals: T
> 1PROGRAM * MULTI (Direct Multiconfiguration SCF) Authors: P.J. Knowles, H.-J. Werner (1984) S.T. Elbert (1988)
>
> (Density-fitting MCSCF) Authors: W. Gyorffy and T. Shiozaki (2012)
>
>
> Number of closed-shell orbitals: 42 ( 21 21 )
> Number of active orbitals: 10 ( 6 4 )
> Number of external orbitals: 260 ( 129 131 )
>
> State symmetry 1
>
> Number of electrons: 10 Spin symmetry=Singlet Space symmetry=1
> Number of states: 1
> Number of CSFs: 9772 (31824 determinants, 63504 intermediate states)
> <...skip...>
> Number of orbital rotations: 6968 ( 210 Core/Active 5460 Core/Virtual 0 Active/Active 1298 Active/Virtual)
> Total number of variables: 38792
>
>
> ITER. MIC NCI NEG ENERGY(VAR) ENERGY(PROJ) ENERGY CHANGE GRAD(0) GRAD(ORB) GRAD(CI) STEP TIME
>
> 1 54 31 0 -2383.67887780 -2383.67887791 -0.00000012 0.00089353 0.00000000 0.00000090 0.19D-03 1100.72
> 2 39 12 0 -2383.67887791 -2383.67887791 0.00000000 0.00000035 0.00000002 0.00000083 0.15D-05 1898.73
>
> ** WVFN **** CONVERGENCE REACHED, FINAL GRADIENT: 0.58D-07
> <...skip...>
> CI vector
> =========
>
> 220000 2220 0.8981314
> 222000 2200 -0.1354345
> 200000 2222 -0.1345665
> 220200 2020 -0.1324518
> 2b0b00 2a2a -0.0898835
> 2a0a00 2b2b -0.0898835
> 2aa000 22bb -0.0887362
> 2bb000 22aa -0.0887362
> 22ba00 2ba0 0.0783718
> 22ab00 2ab0 0.0783718
> 2b0a00 2b2a 0.0637264
> 2a0b00 2a2b 0.0637264
> 220020 0220 -0.0629049
> 2ba000 22ba 0.0608414
> 2ab000 22ab 0.0608414
> 22b0a0 b2a0 0.0556144
> 22a0b0 a2b0 0.0556144
> 220ab0 ab20 0.0555299
> 220ba0 ba20 0.0555299
> 2a00a0 b22b -0.0517008
> 2b00b0 a22a -0.0517008
>
> TOTAL ENERGIES -2383.67887791
> <...skip...>
>
> However, the DF-RS2 part that followed stuck at the status shown below without any further outputs for more than 20 hours, not to mention the next calculation of its analytic gradients.
>
> Basis set: FIT
> Basis size: 1094
> Attributes:
> Sphericals: T
> 1PROGRAM * RS2 (Multireference RS Perturbation Theory) Authors: H.-J. Werner (1993), P. Celani (1998)
>
> (Density-fitting CASPT2) Authors: W. Gyorffy and T. Shiozaki (2012)
>
> Convergence thresholds: THRVAR = 1.00D-11 THRDEN = 1.00D-11
>
> Number of optimized states: 1 Roots: 1
> Number of reference states: 1 Roots: 1
>
> Using ITF DF-MP2 gradient implementation
>
> Switching back to non-DF calculations did not solve the problem. The CASSCF and NEVPT2 single point calculations could finish normally and quickly, while the non-DF RS2 still stuck at the same place.
>
> PROGRAM * NEVPT (Multireference Perturbation Theory)
> Convergence thresholds: THRVAR = 1.00D-11 THRDEN = 1.00D-11
> Number of optimized states: 1 Roots: 1
> Number of reference states: 1 Roots: 1
> Reference symmetry: 1 Singlet
> Number of electrons: 94
> Maximum number of shells: 7
> Maximum number of spin couplings: 132
> Reference space: 4521 conf 9772 CSFs
> N elec internal: 4521 conf 9772 CSFs
> N-1 el internal: 8350 conf 27720 CSFs
> N-2 el internal: 6765 conf 34650 CSFs
> Number of electrons in valence space: 46
> Maximum number of open shell orbitals in reference space: 10
> Maximum number of open shell orbitals in internal spaces: 12
> Number of core orbitals: 24 ( 12 12 )
> Number of closed-shell orbitals: 18 ( 9 9 )
> Number of active orbitals: 10 ( 6 4 )
> Number of external orbitals: 260 ( 129 131 )
> Molecular orbitals read from record 2142.2 Type=MCSCF/NATURAL (state 1.1)
> Coulomb and exchange operators available. No transformation done.
> Number of p-space configurations: 56
> Reference wavefunction optimized for reference space (refopt=1)
> State Reference Energy
> 1 -2383.67915818
> Number of internal configurations: 9772
> Number of singly external configurations: 0
> Number of doubly external configurations: 0
> Total number of contracted configurations: 9772
> Total number of uncontracted configurations: 9772
> Weight factors for SA-density in H0: 1.000000
> FIMAX= 0.16D+00 FXMAX= 0.33D+02 DIAG= F F NOREF=1 NOINT=2 IHPPD=2
> Nuclear energy: 828.85467872
> Core energy: -2940.82861073
> Zeroth-order valence energy: -27.53507067
> Zeroth-order total energy: -2139.50900267
> First-order energy: -244.17015551
> Number of irreps: 2
> Number of frozen orbitals: 12 12
> Number of inactive orbitals: 21 21
> Number of occupied orbitals: 27 25
> Number of valence orbitals: 15 13
> Number of external orbitals: 129 131
> Number of alpha electrons: 5
> Number of beta electrons: 5
> Multiplicity: 1
> Number of states: 1
> Number of determinants: 31824
> Number of total orbitals: 288
> Number of core orbitals: 18
> Number of active orbitals: 10
> Number of virtual orbitals: 260
> **STARTING THE DENSITY MATRICES CALCULATION**
> CPU time for the density matrices calculation: 169.41 secs.
> Building up Koopmans matrices
> KoopE has taken: 0.00 secs.
> Koop2E has taken: 0.50 secs.
> Koop0pE has taken: 1.65 secs.
> Checking elements of F matrix
> Maximum value of the F matrix is: 0.8704349182D-08
> Building matrix AMAT
> BAMAT has taken: 102.62 secs.
> Building matrix BMAT
> BBMAT has taken: 0.46 secs.
> Building matrix CMAT
> BCMAT has taken: 1.58 secs.
> Largest difference between BMAT and CMAT : 0.5320559943D-07
> Largest difference between BTMAT and CTMAT: 0.5320559726D-07
> Total CPU time for Koopro4: 277.04 secs.
> **STARTING THE PERTURBATIVE CALCULATION**
> Maximum of off-diagonal core Fock matrix is: 0.4170875439D-09
> Maximum of off-diagonal virtual Fock matrix is: 0.1052481504D-08
> CPU time for V(0) 0.09 secs.
> CPU time for V(+1) 0.12 secs.
> CPU time for V(-1) 0.29 secs.
> CPU time for V(+2) 0.03 secs.
> CPU time for V(-2) 0.10 secs.
> CPU time for V(0)' 2.02 secs.
> CPU time for V(-1)' 16.72 secs.
> CPU time for V(+1)' 12.19 secs.
> SC and PC correlation energy for each class
> =================================================================
> Using H Dyall
> =================================================================
> Norm SC Energy SC Norm PC Energy PC
> (0) 0.2768641960 -0.9510166330 0.2768641960 -0.9510166330
> (+1) 0.0261897554 -0.0653943833 0.0263229796 -0.0654967075
> (-1) 0.0536273643 -0.1680576257 0.0538969350 -0.1682657199
> (+2) 0.0025936655 -0.0056460593 0.0026248742 -0.0056694796
> (-2) 0.0165171161 -0.0749480177 0.0167960046 -0.0752581146
> (+1)' 0.0015687187 -0.0031541628 0.0018816982 -0.0033382817
> (-1)' 0.0080081528 -0.0222789233 0.0094632751 -0.0231687787
> (0)' 0.0574059521 -0.1020098507 0.0612787166 -0.1036343115
> =================================================================
> Total 0.4427749210 -1.3925056558 0.4491286793 -1.3958480264
> Total CPU time for dypc: 31.58 secs.
> !NEVPT2 STATE 1.1 Energy -2385.075006209327
> Strongly contracted energy -2385.071663838717
>
> The non-DF RS2 part still stucks at the same place, without mentioning ITF DF-MP2:
>
> 1PROGRAM * RS2 (Multireference RS Perturbation Theory) Authors: H.-J. Werner (1993), P. Celani (1998)
>
> Convergence thresholds: THRVAR = 1.00D-11 THRDEN = 1.00D-11
>
> Number of optimized states: 1 Roots: 1
> Number of reference states: 1 Roots: 1
>
> May I request for some advise about what I did wrong with the RS2/DF-RS2 calculations?
> Thank you very much !
>
> ==========
> MJ Wu
>
> _______________________________________________
> Molpro-user mailing list
> Molpro-user at molpro.net
> http://www.molpro.net/mailman/listinfo/molpro-user
More information about the Molpro-user
mailing list