DFT single point crashes for large systems

[Addiw7]

Dear Dalton Users,

I am performing single-point CAM-B3LYP-D3BJ/aug-cc-pVDZ calculations for large system (421 atoms, 1336 electrons) and I can't see a good
reason my calculations suddenly crash. I attach both outputs. These calculations were restarted using SIRIUS.RST file created after one
SCF optimization step.

This is not a memory issue as used resources are much less than available.

Best wishes,
Dawid

[magnus]

I suggest that you first update to the latest release (https://gitlab.com/dalton/dalton/tree/release/2016). I assume you had to change some parameters, e.g., in DALTON/include/maxorb.h? If that is the case, you should also provide those changes.

[Addiw7]

Yes, I had to change some parameters in maxorb.h and infvar.h.

In maxorb.h I set


PARAMETER (MXSHEL = 8000, MXPRIM = 80000 )
PARAMETER (MXCORB = 10000, MXORBT = MXCORB*(MXCORB + 1)/2 )
PARAMETER (MAXOCC = 8000)
PARAMETER (MXCORB_CC = 3000 )

and in infvar.h

PARAMETER ( MAXWOP = 5000000 ) which is ten times larger than original value.

and in mxcent.h
PARAMETER (MXCENT_QM = 800, MXCENT = 800, MXCOOR = 3*MXCENT)

[magnus]

This may not be relevant anymore but there are two possible issues with your setup:

• MXQM3 in include/qm3.h should have the same value as MXCENT_QM in include/mxcent.h
• using a lower threshold for numerical linear dependency

I was able to get past the first iteration by fixing these two points though I was also using a newer version of Dalton.

[Addiw7]

Thank you for time to check that. I will also give it a try after I recompile Dalton 2016.2 that I got.

By the way, what was the wall time for first iteration and how many processors did you use?

[magnus]

Unfortunately I deleted the files after verification and never checked the timings. I believe it was several hours using probably 12 nodes with 24 MPI processes per node.