Molcas Forum

om_imuo

Member

Registered: 2024-10-21

Posts: 2

CASPT2 job stuck

Dear users,

I am trying to run a CASPT2 calculation on a Dy-system using the reference CASSCF(9e, 14o) WF. My calculation does not seem to progress after this:

--------------------------------------------------------------------------------
 Estimated memory requirements:
  POLY3 :                   252747
  RHS:                    26162358
  SIGMA :                 70121552
  PRPCTL:                        0
 Available workspace:   1687176758

++ CASPT2 computation for group   1

********************************************************************************
 Multi-State initialization phase begins for group   1
--------------------------------------------------------------------------------
********************************************************************************
 Compute H0 matrices for state    1
--------------------------------------------------------------------------------
--------------------------------------------------------------------------------
 H0 matrices have been computed.

********************************************************************************
  CASPT2 EQUATION SOLUTION
--------------------------------------------------------------------------------

The job status is showing: CASPT2:Solve CASPT2 eqs for state   1

I have tried using ntasks-per-node=18 and 12G mem-per-cpu and MOLCAS_MEM=15000. The job did not progress for 10 hours and did not throw any error. Any idea why this is happening?

Thanks!

Ignacio

Administrator

From: Uppsala

Registered: 2015-11-03

Posts: 1,083

Re: CASPT2 job stuck

9 electrons in 14 orbitals should work fine with 1 process and 2-4 GB of ram. Perhaps you're overdoing it?

Also 15000 is more than 12G, and given that the actual usage is more than MOLCAS_MEM (as that only accounts for explicit memory allocations), you'll be using more memory than requested. And then, do you have 18*15000 = 270GB of ram available in the node?

om_imuo

Member

Registered: 2024-10-21

Posts: 2

Re: CASPT2 job stuck

Hi Ignacio, thanks for your response! I used 12G mem-per-cpu - so 18*12 = 216G as a whole, which is more than 15G (MOLCAS_MEM). I had an impression that MOLCAS_MEM allocate the total memory. Does it allocate memory only per cpu?

Also, I have another unrelated question. Is their a binary wavefunction file generated from a rassi-so calculation? I would like to use the spin-orbit coupled rassi wavefunctions (obtained by mixing some singlet and triplet casscf optimized states) for some kind of orbital analysis. Is there a way?

Many thanks!

Ignacio

Administrator

From: Uppsala

Registered: 2015-11-03

Posts: 1,083

Re: CASPT2 job stuck

MOLCAS_MEM (and almost everything else) is per process. Keep in mind that each process could potentially be run on a different node (computer), with separate disk etc.

RASSI should create a rassi.h5 file if you have HDF5 support. It may have the information you need.