Problem with standalone MRCC for big basis sets

10 months 3 days ago #617 by Nike
Dear all,
I managed to compile MRCC with the new ovirt.f, which does not have the hard-coded truncation threshold of 10^-12, but instead truncates integrals that are smaller than itol, which I have set to 10^-18. I have not found this to help with the discrepancy between MRCC and MOLPRO/CFOUR.

I think this was expected, because in MOLPRO, changing itol from 10^-12 to 10^-18 made a difference which was orders of magnitude smaller than the the discrepancy between MRCC's correlation energies and the MOLPRO and CFOUR correlations energies (which agree with each other in all digits).

Most of the error goes away when we use CFOUR integrals, but I cannot use CFOUR for 7Z, 8Z and 9Z calculations. My next attempt is to use Psi4 to do the integrals and MRCC to do the correlation, But it has not yet been possible to compile Psi4 with higher than default angular momentum supported.

Therefore my preference would be to use standalone MRCC but based on the discrepancy with CFOUR and MOLPRO at 5Z level when using MRCC integrals, I cannot trust that the answers will be accurate when using standalone MRCC. I wonder if it could be possible to fix the problem with the integrals in standalone MRCC so that we don't have to rely on CFOUR and Psi4 for the non-relativistic integrals?

Perhaps the new ovirt.f works for you but not for me. Does it solve the discrepancy between CFOUR and MOLPRO described here ?

If so, I would like if I could have MRCC binaries that use the new ovirt.f, since when I compiled it, it doesn't seem the energies improved at all, but also increasing "maxit" in the MRCCCOMMON to 9999 didn't let me go beyond 50 iterations either, so maybe there's a problem with my compilation which you don't have in your compilation.

With best wishes !!!

Please Log in or Create an account to join the conversation.

10 months 1 day ago #620 by kallay
Dear Nike,
Please download and install the last program version, either the binary or the source. We have found a similar problem in the SCF code, which has been fixed in the last version. Probably it resolves your problem. I have just run a test for the Li atom with that basis set and reproduced Cfour's CCSD energy to pH accuracy.

Best regards,
Mihaly Kallay

Please Log in or Create an account to join the conversation.

Time to create page: 0.019 seconds
Powered by Kunena Forum