Author Topic: parallel geometry optimisation at DFT  (Read 5855 times)

jan

  • Jr. Member
  • **
  • Posts: 11
  • Karma: +0/-0
parallel geometry optimisation at DFT
« on: August 31, 2009, 04:28:18 pm »
I am running a geometry optimization at DFT level and I want to run an frequency calculation with NumForce afterwards. As recommended in the manual
I used the option "weight derivatives" in $dft to get reliable results. Starting the calculation is no problem and the first ridft-step is performed normally but
when turbomole call the rdgrad-module it crashes with the following message:

Do this calculation sequentially !
rdgrad ended abnormally
error in gradient step (1)

If I remove the keyword "weight derivatives" the calculation runs just fine. This is also the case if i switch to serial calculations, there the keyword seems to
make no trouble at all as the error message above idicates.
Am I doing something wrong or is it no possible to use "weight derivatives" in parallel mode?

Regards Jan

uwe

  • Global Moderator
  • Sr. Member
  • *****
  • Posts: 405
  • Karma: +0/-0
Re: parallel geometry optimisation at DFT
« Reply #1 on: September 17, 2009, 02:37:20 pm »
Hi,

weight derivatives are not parallelized. A solution would be to run the geometry optimization in parallel without weight derivatives, then turn to serial mode and add weight derivatives. A subsequent geometry optimization should do very few cycles as I expect the corrections from the weight derivatives to be quite small. The final NumForce should be run with the serial binaries and with the -mfile option anyway, this is much more efficient than using the parallel binaries.

Regards,

Uwe

Turboooo

  • Jr. Member
  • **
  • Posts: 19
  • Karma: +0/-0
Re: parallel geometry optimisation at DFT
« Reply #2 on: April 15, 2013, 12:28:42 pm »
Is it still current, that the "weight derivatives" can be used only serial or is it already in TURBOMOLE 6.4 SMP capable?