Structure Optimization employing rijk (HF) and rimp2

[Stefan]

Hello,

I want to optimize a structure employing the rijk-approximation for the HF-part as well as the rimp2 approach. Employing jobex does not work. I receive the following error message:

Program dscf can not use the RI-J approximation.
Please either use the ridft program or delete the option $rij
from your control file.

I used the following command to start the calculation:

nohup jobex -level mp2 -ri -c 5 -gcart 4 &

The following seems to work:
1. starting ridft
2. starting rimp2
3. starting relax
4. restart with 1.
Which flag must be used to employ the jobex script?

Regards, Sefan

[uwe]

Hi,

unfortunately jobex is not able to differ the RI of RI-MP2 from the RI of RI-DFT (which is RI-J).

That is a bit unfortunate, but quite easy to circumvent:

Code Select Expand

export dscf=ridft
jobex -ri -level mp2 [...]
unset dscf

should work.

The unset dscf line is necessary to avoid problems with non-RI-MP2 calculations in the same shell after the job, since a usual Hartree-Fock geometry optimization would also use ridft instead of dscf otherwise.

The same trick can also be used for RI-CC2 or RI-MP2 calculations with the ricc2 program (jobex -level cc2) together with RI-JK.

Regards,

Uwe

[Stefan]

Thank you.

Regards, Stefan

[Arnim]

Hi,

here is a little outlook for this problem.

In the next version it will be possible to perfom RI-SCF based geometry optimizations 
with the jobex-script, routinely. Using the rijk-approximation for the HF-part and the CPHF-part
in MP2 or CC2 gradient calculations. The ridft and ricc2 programs will be used and everything
can be prepared in define.

Regards,
Arnim

[christof.haettig]

Hi,

the optimization 'by hand'  or the trick described by Uwe are work arounds for the time being, but note:
The gradients which rimp2 and/or ricc2 compute will in that case be those for conventional (i.e. non-RI) SCF+
RI-MP2 energy and do not fit 100% to the computed total energy. Probably in most cases, this will not be
a serious problem, but you should be aware that you can't use tight convergence threshold for jobex in that
case.
           Christof

[Stefan]

Thank you for the further informations.

Regards, Stefan

[christof.haettig]

The above discussion is outdated since several years.

To optimize geometries at the RIJK-HG + RI-MP2 (or RI-CC2) level use the following command:
       jobex -level cc2 -rijk