| 2:02:36 AM PST - Tue, Nov 20th 2012 |
|
Thanks for your answer Edo.
In fact the modifications you propose did not work : I still see dispersion in the total time needed for the execution due to different total number of steps for convergence.
I think I solve my issue by specifying to take the scf threshold to 1e-06. I still need more runs to be sure.
Furthermore the outpout file are stored in 64-bit.
I might miss some point to understand how the convergence works : by default scf threshold is at 1e-04 and I see between runs only 4 significant digits after the decimal point during the iteration of ROHF. On the other part, the energy minimization has one criterium about total energy convergence which is by default 1e-06. I know there are others criteria, but with only those two, the last steps of convergence seems to me quite erratic.
To test it, I have taken the solution of the computation as the input for the benchmark : I would have expected all simulations to end at step 1. But over 7 runs, I saw 1 run taking 2 steps to converge.
Cheers,
Pierre-Antoine
|