When I run a PHVA of a large system, I see this error:
" You cannot request more PEs than atoms"
This does not go away whether I reduce the number of atoms for which I need a hessian or increase set_iter (which I think allows for more cpscf steps).
Any suggestions on how to get around this error? I see this reported as a bug in trac that is not yet resolved.
Could you post the input file? If it’s confidential, you may build a model system that can reproduce the error.
it’s ticket 2969, still unresolved
I ran you molecule, with a slightly modified $rem (only to make it run faster):
cpscf_nseg 8 ! maybe has to be equal to N_SOL?
This version runs to completion on 8 processors. Key part of output:
CPSCF will be done in 8 segments to save memory.
(This was set automatically but you could set with CPSCF_NSEG in $rem.) Note that I did have to increase SET_ITER; the segments took up to 36 iterations to converge the CPSCF equations.)
I think this is not so much a bug as a limitation of the current implementation: number of processors (PEs) cannot exceed number of atoms (even for full Hessian), and in partial Hessian the number of atoms is N_SOL.
Please try this with your real job and let us know how it turns out.
The CPSCF iterations are running, very slow with 8 cores, but no issues so far. I didn’t need to set CPSCF_NSEG. Will update if I see any other errors. Thank you, John!
I’ve updated the relevant bug ticket with this information, for what it’s worth.