Advice on frequency calculations on large solid systems -

PeterRemoto

Hey,
I was wondering if anyone has any advice with dealing with crystal systems with transiton metal complexes with a floating anion? Our objective is to calculate the low-frequency Raman modes of these structures and compare with experimental.
Using PBE functional with Grimme D3 dispersion correction and the pob-TZVP-rev2 basis set
So managed to optimise the structure - which took a while but was fine
908892_hicnis_PBE_D3_pob-TZVP-rev2_OPTGEOM_4_4.d12

Now I'm doing frequency calculations
908892_hicnis_PBE_D3_POB-TZVP-REV2_FREQ_4_4.d12
But now a bit stuck with the frequency calculations because it doesn't seem to converge for the INTCPHF structures
908892_hicnis_PBE_D3_POB-TZVP-REV2_FREQ_4_4-69914.out
Tried GUESSP with fort.20 (copied from fort.9) to try to speed up the some bits before I hit my cluster time limit. However, the calculation doesn't seem to properly get past the INTCPHF cycles - Electric field applied to X direction hits 200 cycles, which is what the MAXYCLES set, which is already pretty high - so I can't use fort.32 (copied from fort.31) for INTCPHF RESTART. I'm currently running a calculation with MAXCYCLES increased to 9999 but it doesn't seem to help much looking at the ongoing ouput file.

Not sure how else to troubleshoot this properly without using up all of our cluster allocation - so very open to advice and potential solutions to try out.

Happy to provide more input files if need be.

Thank you for your time,
Peter

PeterRemoto

Just an update on this - so I've tightened the TOLINTEG from 10 10 10 10 20 to 12 12 12 12 24 and it seemed to progressed the calculation beyond "Electric field applied to X direction" and now at XX direction. Will see how far it progresses and yet to see what happens when I use GUESSP and INTCHPF RESTART.

aerba

Hi,

Thanks for the update! I still hadn't had the time to look carefully into this. Now, by looking at your input file, I just noticed that you are using ANDERSON for the INTCPHF step instead of DIIS, which is the default option in CRYSTAL23. Can I ask you why you decided to switch to ANDERSON? Was DIIS not converging? If you instead used ANDERSON straight away, may I suggest to remove it and thus use DIIS and see how it behaves as a test?

PeterRemoto

Hi Erba,
Honestly, I have ANDERSON acceleration on because I inherited a file with these keywords written in when I was taught how to use CRYSTAL for frequency calculations - and worked pretty well so far with our other systems so I thought to leave it alone (though it was more for pharmacuetical systems). Thanks for the suggestion, I'll try it out if with the original TOLINTEG 10 10 10 10 20 and see what happens.

Thanks,
Peter