EOM-CCSD converged to zero excitation energy


Just Got Here
Dear NWChem users and developers,
I encountered a problem with EOM-CCSD calculation. It converges to a zero excitation energy. Part of the output is:

Iteration    65
0.0000135820372 0.1526878036734 4.15485 25.8 26.6

Iteration    66
0.0000085819170 0.1526875872261 4.15484 27.1 27.9

Iteration    67
0.4469312094932 0.1273451531015 3.46524 27.8 28.6

Iteration    68
0.1396660064327 0.0206140009003 0.56094 28.9 29.7

Iteration    69
0.0826796488550 0.0045396039865 0.12353 25.3 26.1

Iteration    70
0.0321419413973 0.0011070725423 0.03012 26.8 27.6

Iteration    71
0.0157338758257 0.0002114210748 0.00575 28.1 28.9

Iteration    72
0.0054064264462 0.0000571089965 0.00155 28.7 29.6

Iteration    73
0.0016871976737 0.0000273435905 0.00074 25.3 26.1

Iteration    74
0.0013253745399 0.0000115606150 0.00031 26.8 27.7

Iteration    75
0.0005923666746 0.0000015037397 0.00004 27.7 28.6

Iteration    76
0.0002121482222 0.0000003632998 0.00001 29.1 29.9

Iteration    77
0.0001130179663 0.0000003071590 0.00001 25.7 26.6

Iteration    78
0.0000426245732 0.0000002084082 0.00001 27.0 27.9

Iteration    79
0.0000206317196 0.0000001147986 0.00000 27.4 28.2

Iteration    80
0.0000072627735 0.0000000563482 0.00000 29.4 30.3

Iteration    81
0.0000032597905 0.0000000327591 0.00000 24.9 25.8

Iteration    82
0.0000021872900 0.0000000187070 0.00000 26.9 27.7

Iteration    83
0.0000010843757 0.0000000051556 0.00000 28.2 29.0

Iteration    84
0.0000004116708 0.0000000011695 0.00000 29.0 29.9
EOMCCSD root nr. 1

The input is:

(geometry specification)
SCF
       THRESH 1.0e-10
TOL2E 1.0e-10
SINGLET
RHF
END
tce
       scf
creomsd(t)
freeze atomic
nroots 1
tilesize 20
THRESH 1.0d-6
2eorb
2emet 13
eomsol 2
end
TASK TCE ENERGY

Any ideas?

Clicked A Few Times
Hi,
You have several options:
1.) Please change the diis length (by default diis is set equal to 5). You may want to use longer diis cycle length (please use "diis 10" in the tce input group). One thing to remember is that eomsol 2 starts from the CIS vectors (in some cases this may be a crude approximation).
2.) If 1.) doesn't work please switch to the other type of the solver ("eomsol 1"). "eomsol 1" is more memory thirsty but uses different type of initial guesses.

Let me know how it works.
Best,
Karol

Just Got Here
Hi Karol,

Thank you for your reply. I tried both. In some cases "diis 10" with “eomsol 2" worked but generally it seems "diis 10" with "eomsol 1" is safer, though it is slower.

It helps a lot since such convergence problem occurs frequently. Many thanks!


Forum >> NWChem's corner >> NWChem functionality