Author Topic: parallel runs with PEECM method  (Read 7869 times)

sca

  • Newbie
  • *
  • Posts: 7
  • Karma: +0/-0
parallel runs with PEECM method
« on: June 23, 2008, 10:11:34 PM »
Hello,

       I am currently doing some calculations with the PEECM method. So far, I could successfully run few calculations with constrained optimization of a small cluster with the serial version of Turbomole 5.10. But when I tried to do the same calculation with the parallel version, I get some strange results. The first energy is the same with both the serial and parallel versions either I use ridft or dscf methods. But when I start the jobex run, the gradients are completely different. The parallel version gives very large gradients and the structures never converge! I also tried both parallel and serial calculations for simple systems (not embedded), where the results are same! Do any one has any experience or comments about this problem? Any help would be greatly appreciated!

christof.haettig

  • Global Moderator
  • Sr. Member
  • *****
  • Posts: 291
  • Karma: +0/-0
    • Hattig's Group at the RUB
Re: parallel runs with PEECM method
« Reply #1 on: June 24, 2008, 12:20:03 AM »
Hey,

the periodic point charge embedding is in TM V5.10 only implemented in the serial binaries, but (yet) not in the parallel version (at least as it concerns the gradients, I'm not sure about the energies). This should be fixed in the next version.
Christof

sca

  • Newbie
  • *
  • Posts: 7
  • Karma: +0/-0
Re: parallel runs with PEECM method
« Reply #2 on: June 24, 2008, 04:46:46 AM »
Thank you very much for the response! At least now I know that there is no mistake from my side, but it is quite disappointing to know that I can not use the parallel version for my computations!

christof.haettig

  • Global Moderator
  • Sr. Member
  • *****
  • Posts: 291
  • Karma: +0/-0
    • Hattig's Group at the RUB
Re: parallel runs with PEECM method
« Reply #3 on: June 24, 2008, 11:06:09 PM »
I cann't promise anything, but if it is important for you, you could try and ask COSMOLogic if they can help with a parallel binary, where this has been fixed.
Christof
PS: regards to Andreas Heyden.