Greetings!
I took the time to install the latest version.
I am also using Jozsef's basopt feautre, which is very good!
I am trying to optimize a 9-zeta basis set though, and while I like everything, I'm finding the AO-MO transformation to be extremely slow in MRCC compared to other programs. Often I use the CFOUR or MOLPRO interface for the integrals, and MRCC for the rest, but as you know, MOLPRO and CFOUR can only go up to 6-zeta (because they can't do k-functions).
Here is the speed for Li atom in aug-cc-pCV7Z(i), which only goes up to i-functions so that I can compare with MOLPRO:
Code:
************************ 2017-07-24 07:04:49 *************************
Executing integ...
....
Calculation of two-electron integrals...
1% done.
22% done.
55% done.
100% done.
CPU time [min]: 68.904 Wall time [min]: 7.297
************************ 2017-07-24 07:12:07 *************************
...
************************ 2017-07-24 07:13:06 *************************
Executing ovirt...
Mon Jul 24 10:35:41 EDT 2017
ovirt terminated normally
************************ 2017-07-24 10:35:42 *************************
This means it took 7 minutes to do the initial integrals, and 3 hours to do the AO to MO transformation.
In MOLPRO the integrals take 127 seconds in total !!!!
I noticed this difference in speed long ago, and I decided to just live with it, however for basis set optimization we have to run the calculations many times, and this 3 hours each time make it prohibitive (impossible to optimize the basis set).
I was doing basis set optimization with MOLPRO's BFGS routine written by Kirk Peterson until I got to 7-zeta which MOLPRO can't do because of the k-functions. 7-zeta and 8-zeta could be optimized "manually" by using MOLCAS or Gaussian on a grid of points, fitting a surface, and looking for the exponents that are "optimal". However for 9-zeta I'd rather use Jozsef's basopt routine because the number of exponents to tinker manually is six (3k, 2l, 1m) and for 8z it was just three (2l, 1m).
MOLPRO's fast integrals are done by SEWARD (same as MOLCAS) which can go up to angular momentum L=15, which is 15-zeta (whatever you call the functions after r-functions, because 's' comes after 'r' and s-functions are already L=0, so I guess we would call these t-functions). MOLPRO just can't do CISD on integrals beyond i-functions because there's a lot of hard-coded limits. Therefore, I wonder if it would be possible to run stand-alone MRCC with SEWARD integrals somehow? You might agree that running MRCC with MOLPRO or CFOUR integrals is possible, but not with the BASOPT utility.
With best wishes!!
Nike Dattani
