Embarrasingly parallel job

1

Hello,
I would like to run some frequency jobs in embarrassingly parallel mode. I have had success with this in the past, but I am now not sure what the environmental requirements for this are. The examples provided state that “This file will not work outside a cmake test environment”. What is this referring to? I can’t remember being concerned with this in past work.

Thanks
Fred

2

You’ve caught us in the stages of a transition in how we’re handling our capabilities for embarrassingly parallel computation. We’ll have much expanded support for this in the upcoming Psi 1.4. If you want to see the details, check out the preprint for the 1.4 paper. Presently, however, Psi has no built-in capabilities for this. The examples you are looking at will not work.

How soon do you need this, and how willing are you to modify and debug Python scripts? I see two solutions:

  1. If you’re willing to wait a bit, simply wait until the necessary feature comes onto the developmental version of Psi4 and use that. I say “a bit” because I don’t have a definitive estimate on when this will be. I expect to have one tomorrow afternoon.
  2. I have a Python script to do this for me that works using the latest Psi release, but it may need some modifications for your submission system. You’d be on your own for that part.
3

Thanks for your reply and thanks for the link to the preprint, it looks great.

If you are happy to share your python script, I would be very grateful.
Fred

4

One thing I should have asked - is this by finite difference of gradients or energies? The script I have (which is more complicated than I remembered…) is limited to finite difference of energies.