Author Topic: aoforce + cosmo convergence  (Read 562 times)

martijn

  • Jr. Member
  • **
  • Posts: 21
  • Karma: +0/-0
aoforce + cosmo convergence
« on: December 07, 2017, 05:30:08 pm »
Hi,

When combining aoforce with cosmo in 7.2.1 I encounter the following situation that I have never observed for gas phase frequency calculations:

COSMO cphf update
   23       a      170        1.482021067048531D-05
            b      170        7.672115628665743D-06
 
 COSMO cphf update
   24       a      170        1.482182443756061D-05
            b      170        7.672115628665743D-06
 
 COSMO cphf update
   25       a      170        1.481440684000221D-05
            b      170        7.672115628665743D-06
 


 Warning! No convergence within  25 iterations.
 Unless you have specified a very low $escfiterlimit/$forceiterlimit,
 this is a reason to worry!



========================
 internal module stack:
------------------------
    force
========================

 CPKS-Iteration did not converge !
 force ended abnormally


Increasing the number of allowed iterations makes no difference whatsoever; the error/norm stays the same.

Does anybody have an idea what might be going on and if there's a parameter to play with in order to aid convergence? A NumForce + cosmo calculation on the same structure converges without any issues.

Thanks,

Martijn

uwe

  • Global Moderator
  • Sr. Member
  • *****
  • Posts: 420
  • Karma: +0/-0
Re: aoforce + cosmo convergence
« Reply #1 on: December 08, 2017, 10:10:53 am »
Hi Martijn,

the default convergence threshold is 10^-5, so you are more or less almost converged (1.48D-05). If the usual convergence enhancers like increasing DFT grid size (m3 to m4 or 4 or m5) and density convergence ($denconv 1d-7) do not help, you probably hit the accuracy limit of the COSMO cavity derivative (holes in the cavity, very high charges on some small cavity patches, ...).

A pragmatic solution would be to restart escf (since you are already near convergence) and change the convergence settings for the CPKS iterations to $rpaconv 4

Regards,

Uwe