Problem with restarting CPKS calculation

Gryffindor

Hi,

I am encountering an issue while performing a CPKS calculation on a large system. Although the calculation proceeds without errors, it eventually hits the wall time limit. Unfortunately, no restart file (Fort.31) is generated, which prevents me from restarting the calculation.

I have attempted the calculation both with and without the DIIS keyword, but in both cases, the restart file is not written.

Thank You

The input and output files are attached below.
INPUT.d12 OUTPUT.d12

aerba

Hi,

You are running a quite large calculation here, with 800+ atoms/cell!

You are not getting the fort.31 unit to restart the CPKS step because your calculation did not make it to the CPKS step

Let me explain: before getting to the CPKS step, an SCF is performed. Your calculation stopped at cycle 14 of the SCF. By looking at your input/output file, I can suggest four things:

In the output, you can find this warning:

WARNING **** GENBUD **** COUL. BIPO BUFFER TOO SMALL - TO AVOID I/O SET BIPOSIZE = 14808875

which means that the program is swapping to disk. To avoid this and make things faster, you can insert the following keywords in the input file:

BIPOSIZE
14808875

I noticed you inserted the NODIIS keyword in the third block of the input. This is switching DIIS off for the SCF, not for the CPKS step. If this is what you want, ok then. Otherwise, I suggest you remove this keyword from there to keep DIIS active for the SCF.
If you want, you can restart the SCF by use of the GUESSP keyword, to start from the last density matrix of the previous incomplete job (through the fort.9 unit, to be renamed as fort.20).
remove the RESTART keyword from the CPKS block because, effectively, there is nothing to be restarted there.

If you do these four things, your input would look like:

ZIF-90_VOCs
EXTERNAL
CPKS
END
BASISSET
mybasis
DFT
HSE06
END
SCFDIR
TOLINTEG
7 7 7 7 14
TOLDEE
7
SHRINK
1 1
FMIXING
90
BIPOSIZE
14808875
GUESSP
END

Hope this helps,

Gryffindor

Hello Alessandro,

Thank you for your previous response. I could successfully complete most of my calculations with the large systems, except this one, where the fort.9 file is generated, but it's empty. Could you kindly help me understand what the problem is?

Thank you for all the support. Appreciate it!

Best regards,
Aparajita

INPUT.d12 OUTPUT.d12

aerba

Dear Aparajita,

It is strange that the fort.9 unit is empty. May it be a problem of the script in copying back that unit from the scratch folder to the working directory?

Anyway, you can try to trick CRYSTAL in this way:

By inspection of your output file:

CYC  13 ETOT(AU) -1.952535840054E+05 DETOT -1.58E-05

the energy difference is down to 10^-5 at cycle 13. So, you can try to run the calculation with a low energy convergence threshold of just 10^-4 with something like this:

TOLDEE
4

and the SCF would be considered as converged for the program. In this way the calculation is not killed and probably you will get your fort.9 unit,;

you can then use it to do a restart calculation with the GUESSP keyword (and by tightening the energy convergence threshold to something like TOLDEE 7, or 8):

GUESSP
TOLDEE
7

Hope this helps,

Gryffindor

Dear Alessandro,

I’m reaching out again regarding the CPKS step. After several attempts (and a fair bit of persistence!), I was finally able to complete the SCF calculations. However, the CPKS part continues to restart from the beginning, even when I provide the fort.9 (wavefunction) and fort.31 (restart) files as instructed.

As per your previous suggestions, I ensured the SCF was converged beforehand, but unfortunately, the CPKS has never been able to finish successfully on my side. Since I need these results quite urgently, I was wondering if you could try running it on your end to see if there's anything I'm missing?

Thank you for your patience.

Best regards,
Aparajita

OUTPUT.d12.d12
INPUT.d12.d12