Numerical collapse in EOM-CCSD?


Click here for full thread
Clicked A Few Times
Hi,

I found an odd progression of energies during a NWChem 6.8 EOM-CCSD calculation of naphthalene in Cs symmetry, cc-pvtz basis, no frozen core. The following is from the a' irrep, and focusing on the 4.6 eV root that starts off as root 1.

...
 Iteration  38 using  134 trial vectors
   0.0000000240613   0.1694889939661    4.61203
...
 Iteration  39 using  135 trial vectors
   0.0000000240820   0.1694889939566    4.61203
...
 Iteration  40 using  136 trial vectors
   0.3118749370700   0.0423481583098    1.15235
   0.0000000240893   0.1694889939764    4.61203
...
...
 Iteration  65 using  193 trial vectors
   0.0000000393889  -0.0000000002695   -0.00000
   0.0000000327071  -0.0000000002695   -0.00000
   0.0000000098575   0.0000000000007    0.00000
   0.0000000238503   0.1694889939682    4.61203
   0.0000000568039   0.1944784791788    5.29203  1199.3    91.0
 --------------------------------------------------------------
 Iterations converged
 ------------------


The root at 4.61203 eV remains in the converged solution, but at iteration 40 shifts to the second root, as the first one plummets toward 0.00000 eV. This happens 2 more times, at iteration 43 (shifts to 3rd root) and iteration 47 (shifts to 4th root). Another program converges with 4.6120 eV as the first root (of course, thresholds may be different).

Any thoughts on whether this should be expected or reflects a problem in the input or code?

TCE input is vanilla,

tce
   scf
   ccsd
   nroots 5
   print time tile
end

task tce energy


Thanks; Chris