Author Topic: aoforce step in NumForce calculations with reducible E representations  (Read 12406 times)

antti_karttunen

  • Sr. Member
  • ****
  • Posts: 227
  • Karma: +1/-0
Because aoforce does not support point groups having reducible E representations, we are performing NumForce calculations on such molecules. However, the last step in a NumForce calculation is the aoforce frequency analysis of the hessian generated by sammler. The results given by aoforce seem to be reasonable, so I suppose that the problems with aoforce and reducible E representations only affect actual analytic 2nd derivative calculations and not the frequency analysis part? I just wanted to check this, as the aoforce.out file contains the usual warning "It is STRONGLY RECOMMENDED to use a lower symmetry!"

turbomaster

  • Administrator
  • Full Member
  • *****
  • Posts: 63
  • Karma: +0/-0
The frequency analysis part of aoforce is not affected by the problems with reducible E representations. However, I wonder if it would not be more efficient to lower the symmetry of your system and do aoforce calculations instead of NumForce?

antti_karttunen

  • Sr. Member
  • ****
  • Posts: 227
  • Karma: +1/-0
Sorry, my original message was a bit confusing. Usually we just lower the symmetry for reducible E representations and then use aoforce. It is still more efficient than NumForce. But now we need MP2 frequencies and this is the reason for using NumForce. By the way, analytic RI-MP2 frequencies would be a nice feature in TURBOMOLE, especially with efficient parallellization  ;D
« Last Edit: March 28, 2007, 04:13:13 PM by antti_karttunen »

christof.haettig

  • Global Moderator
  • Sr. Member
  • *****
  • Posts: 291
  • Karma: +0/-0
    • Hattig's Group at the RUB
Well, in the recent versions (5.7...5.9) there is bug in the orbitals for reducible E
representations which also effects the (RI-)MP2 gradients. Thus, you should not use
these point groups for (RI-)MP2 geometry optimizations. NumForce should work
o.k. apart from the fact that the test of the gradient at the reference point might
be garbage.

Christof

Arrepiadd

  • Jr. Member
  • **
  • Posts: 12
  • Karma: +0/-0
Well, in the recent versions (5.7...5.9) there is bug in the orbitals for reducible E
representations which also effects the (RI-)MP2 gradients. Thus, you should not use
these point groups for (RI-)MP2 geometry optimizations.

I'm sorry to bring this old subject up but, does this still apply for optimizations, at the 5.10 version? I just did an RI-SCS-MP2 optimization in a Td molecule (and I defined that symmetry) so I was just wondering if the geometry is ok, or if I should just scrap it and start over with a lower symmetry... I find no information on the DOK file about this so I cannot be sure.

Thanks in advance!

PS - Since I'm bringing attention to it, is it ok then to do an analytical frequency calculation on such a job (MP2)? Because one of the previous posters mentioned needing RI-MP2 frequencies needing NumForce...
« Last Edit: March 18, 2008, 08:46:02 PM by Arrepiadd »

Arnim

  • Developers
  • Sr. Member
  • *
  • Posts: 253
  • Karma: +0/-0
Hi!

The bug with reducible E representations has been fixed in TM V5.9.1.

To the second point: there are no analytical 2. derivatives on MP2 level
implemented at the moment. At this level of theory, you have to use
NumForce.

Best,
Arnim
« Last Edit: March 20, 2008, 01:41:03 PM by christof.haettig »

Arrepiadd

  • Jr. Member
  • **
  • Posts: 12
  • Karma: +0/-0
Thank you for the information.

Best regards,
Daniel