MCLR divergence problem with McKinley and Alaska module

Posted by Lukas F. Pasteka on February 16, 2010 at 12:59:24:

Hello,

while optimizing geometries with CASSCF and consequently doing the frequency analysis with McKinley, usually for state-averaged cases (but not necessarily), I ran into a problem with MCLR divergence several times. Corresponding error in the output file is:

"Warning perturbation number X will diverge
No convergence for perturbation no: X. Increase Iter."

I found out that if occupancy of an orbital is close to 2 and the energy is close to near inactive orbital, I should change the active space to eliminate this. However, this is not the case.

I also know that increasing the iterations in MCLR module is possible via including NoMCLR keyword in McKinley and adding MCLR section with the Iter keyword set to higher number. But this also doesn't help.

And to my knowledge there is no such option how to do similar precaution if MCLR invoked by Alaska causes the trouble, because in Alaska, the is no NoMCLR keyword. Or is there a way how to do it anyway?

What are the other possible solutions to this problem?

And I have yet another question. When doing the frequency analysis for the excited states, how should one properly construct the McKinley input? It seems that RlxRoot keyword in RACCSF module is not sufficient in specifying the higher root for optimization if MCLR is invoked. Should I just include the SALA keyword in MCLR or is there anything else needed? And how is the CASPT2 case different?

Any help is highly appreciated.

Lukas

Follow Ups:

How to define Proccessor number in Molcas 7.6 Saumik Sen 13:08:27 2/23/12 (0)
Re: MCLR divergence problem with McKinley and Alaska module Roland Lindh 08:47:41 2/17/10 (1)
- Re: MCLR divergence problem with McKinley and Alaska module Lukas F. Pasteka 12:53:08 3/02/10 (0)

Post a Followup

Name:
E-Mail:

Subject:

if B is 1s22s22p1, what is Li?

Passfield:

Comments:
: Hello, : while optimizing geometries with CASSCF and consequently doing the frequency analysis with McKinley, usually for state-averaged cases (but not necessarily), I ran into a problem with MCLR divergence several times. Corresponding error in the output file is: : "Warning perturbation number X will diverge : No convergence for perturbation no: X. Increase Iter." : I found out that if occupancy of an orbital is close to 2 and the energy is close to near inactive orbital, I should change the active space to eliminate this. However, this is not the case. : I also know that increasing the iterations in MCLR module is possible via including NoMCLR keyword in McKinley and adding MCLR section with the Iter keyword set to higher number. But this also doesn't help. : And to my knowledge there is no such option how to do similar precaution if MCLR invoked by Alaska causes the trouble, because in Alaska, the is no NoMCLR keyword. Or is there a way how to do it anyway? : What are the other possible solutions to this problem? : : And I have yet another question. When doing the frequency analysis for the excited states, how should one properly construct the McKinley input? It seems that RlxRoot keyword in RACCSF module is not sufficient in specifying the higher root for optimization if MCLR is invoked. Should I just include the SALA keyword in MCLR or is there anything else needed? And how is the CASPT2 case different? : Any help is highly appreciated. : Lukas

[ Follow Ups ] [ Post Followup ] [ Molcas user's WWWBoard ]