CCSD correlation energy infinite (Molpro works).


  • Guest -
When running NWChem6.0 (compiled from source) with the following input file I get infinite CCSD correlation energies (NaNs) from the very first iteration. The RHF energy is -78.064796148475 hartree.

When I run the same job using Molpro2010.2 I get
RHF energy = -78.064796230852 hartree, and
CCSD total energy = -78.394519990527 hartree


CCSD iterations
-----------------------------------------------------------------
Iter Residuum Correlation Cpu Wall V2*C2
-----------------------------------------------------------------
1************************************ 0.9 1.0 0.5
2************************************ 0.9 1.0 0.5
3 Infinity****************** 0.9 1.0 0.5
4 NaN NaN 0.9 1.0 0.5
5 NaN NaN 0.9 1.0 0.5
MICROCYCLE DIIS UPDATE: 5 5

The only unusual thing about this test as far as I can tell is that it has a user-defined basis set (taken from a paper by B. Roos) with some rather diffuse functions.

memory stack 4000 mb heap 100 mb global 4000 mb noverify
SCRATCH_DIR /scratch
title "Expt geometry"
geometry units angs
SYMMETRY c2h
C -0.669500 0.000000 0.000000
C 0.669500 0.000000 0.000000
H -1.232077 0.928926 0.000000
H 1.232077 0.928926 0.000000
H -1.232077 -0.928926 0.000000
H 1.232077 -0.928926 0.000000
end

basis spherical nosegment
hydrogen s
188.61445 0.0009639 -0.0013119 0.0024224
28.276596 0.0074920 -0.0103451 0.0203382
6.4248300 0.0375954 -0.0504953 0.0896394
1.8150410 0.1433950 -0.2073855 0.4422907
0.5910630 0.3486363 -0.4350885 0.5757144
0.2121490 0.4382974 -0.0247297 -0.9802890
0.0798910 0.1651066 0.3225260 -0.6721538
0.0279620 0.0210229 0.7072754 1.1417685
hydrogen p
2.3050000 0.1127902 -0.2108688
0.8067500 0.4185075 -0.5943796
0.2823620 0.4700077 0.0896889
0.0988270 0.1826260 0.8611634
carbon s
50557.501 0.0000553 -0.0000120 0.0000119 -0.0000156
7524.7856 0.0004343 -0.0000940 0.0000927 -0.0001140
1694.3276 0.0023159 -0.0005028 0.0004989 -0.0006731
472.82279 0.0098729 -0.0021476 0.0021180 -0.0025323
151.71075 0.0352195 -0.0077942 0.0077784 -0.0109003
53.918746 0.1041938 -0.0237634 0.0236328 -0.0277807
20.659311 0.2412741 -0.0600235 0.0616306 -0.0958713
8.3839760 0.3840174 -0.1153985 0.1189680 -0.1247806
3.5770150 0.3082371 -0.1539009 0.1880621 -0.3932460
1.5471180 0.0683055 -0.0145946 -0.0540304 0.6796004
0.6130130 0.0007782 0.3895849 -0.9814137 1.1978690
0.2460680 0.0009905 0.5397291 -0.1096758 -1.8979520
0.0990870 -0.0000893 0.1884060 0.8847356 -0.0044614
0.0346800 0.0000471 0.0258575 0.2964983 1.0142148
carbon s
0.012138 1.0
carbon s
0.0042482 1.0
carbon p
83.333155 0.0012241 -0.0011444 0.0014669
19.557611 0.0094389 -0.0089796 0.0144553
6.0803650 0.0417744 -0.0378456 0.0466373
2.1793170 0.1318330 -0.1292708 0.2366537
0.8651500 0.2789119 -0.3784022 0.6088734
0.3619440 0.3668663 -0.2692137 -0.3917793
0.1547400 0.2790591 0.2917542 -0.8700229
0.0654290 0.1380481 0.5436698 0.4200505
0.0229000 0.0341950 0.2628308 0.5799497
carbon p
0.0080150 1.0
carbon p
0.0028052 1.0
carbon d
1.9000000 0.0987312 -0.1455013
0.6650000 0.4529661 -0.5076809
0.2327500 0.4362457 -0.1015630
0.0814630 0.2719250 0.9251922
carbon d
0.028512 1.0
end


SCF
SINGLET
RHF
END
TASK SCF ENERGY
TCE
CCSD
END
TASK TCE ENERGY

Mark A. Watson

Clicked A Few Times
Please get rid of nosegment inthe input
Hi Mark,
please get rid of nosegment option in the basis set group.

Without this option I am getting finite CCSD energy
  16   0.0000001551310  -0.3464721739550     1.1     1.2     0.4
17 0.0000000695991 -0.3464721762496 1.1 1.2 0.4
-----------------------------------------------------------------
Iterations converged
CCSD correlation energy / hartree = -0.346472176249560
CCSD total energy / hartree = -78.411268418161650

The energy is different from the one prodiced by MOLPRO
but I assumed (following your input) that all electron are correlated.
I guess that the MOLPRO by default keeps all core electron frozen.
In TCE by defult all electrons are correlated.

Please give a try and let me know how it works.

Best,
Karol

P.S. For RHF reference please use the following form of the TCE group

SCF
SINGLET
RHF
END

TCE
CCSD
2eorb
2emet 13
tilesize 15
END

TASK TCE ENERGY

Clicked A Few Times
no "nosegment" & all core electrons frozen & problem solved
After freezing core electrons (and getting rid of nosegment)
TCE gives:

  17   0.0000000679101  -0.3297237707525     0.9     1.0     0.4
-----------------------------------------------------------------
Iterations converge
CCSD correlation energy / hartree = -0.329723770752516
CCSD total energy / hartree = -78.394520012664600


it looks like difference between TCE and Molpro CCSD energies is on the order of 2.2*10^(-8)

Best,
Karol

Just Got Here
Thanks Karol! This worked for me. And yes, the other trivial issue was related to the frozen core (or not).


Forum >> NWChem's corner >> Running NWChem