Author Topic: Manually running Numforce SP gradient calculations?  (Read 6423 times)

mike

  • Jr. Member
  • **
  • Posts: 10
  • Karma: +0/-0
Manually running Numforce SP gradient calculations?
« on: November 18, 2008, 09:01:21 PM »
Dear all

Is it possible to pause (or stop and later restart) Numforce? I would like to run all the necessary single point gradient calculations manually and once they are all done, restart Numforce for creation and diagonalization of the hessian. Is this possible?

I would use the Numforce -mfile option, so using the serial binaries and spreading the independent gradient calculations over several processors but I can't figure how to get it to work with my queuing system.

Whats more, the number of processors I can specify for a single job is limited, but if I launch each gradient calculation as a separate job, subject to cluster demand, I can potentially have a much larger number of processors.

Your help would be much appreciated.

Mike

alex

  • Newbie
  • *
  • Posts: 3
  • Karma: +0/-0
Re: Manually running Numforce SP gradient calculations?
« Reply #1 on: November 20, 2008, 01:19:50 PM »
Hi Mike,

do I see this right:
Is your main motivation for doing this manually your queueing system/job script does not like NumForce?

Cheers

Alex

uwe

  • Global Moderator
  • Sr. Member
  • *****
  • Posts: 494
  • Karma: +0/-0
Re: Manually running Numforce SP gradient calculations?
« Reply #2 on: November 23, 2008, 08:37:12 PM »
Hi,

NumForce is restartable, it always checks for complete steps and skips them. (Hint: if one of the single-point jobs had problems, simply delete the corresponding files and restart NumForce).

To stop NumForce, just create a file called stop in the numforce directory of your input (simplest way is to enter touch stop there), and NumForce will stop at the next step.

Another feature is to call NumForce with the option -prep which just does part of the preparation and stops before starting the single point jobs.

Finally, -mfile expects a file which contains the names of the nodes where the single-point jobs shall run (each line should contain just one name). Most queuing systems generate such a file automatically - please ask your system administrator about it. If you run e.g. PBS, it would be:

NumForce [...] -mfile $PBS_NODEFILE

Regards,

Uwe