SCF convergence ferromagnetic MIL-127

wgtemmer

Dear CRYSTAL team,

I am trying to execute a spin polarized single point calculation on a metal-organic framework. The calculation I am trying to run is very similar to earlier work from your group (10.1021/acs.jpcc.9b06533).

I have tried many different settings and keywords, but I cannot obtain SCF convergence and I am hoping you can provide me some tips. Similar to the publication, I am using a 6-311G** basis set for the elements H, C, N, O and F. For the Fe atoms, I am using the POB-TZVP basis set.

I am trying to find the high spin state (all Fe atoms should have +5 spin).

Previous unsuccesful attempts include:

SPINLOCK 120 (all Fe atoms, instead of only those closes to F atoms)
ATOMSPIN +1 for all Fe atoms
UHF
Other functionals (PBE/PBE0/BLYP)
EIGSHIFT/LEVSHIFT
Various FMIXING values

I have attached the INPUT, fort.34 and BASISSETS.DAT files as a single .zip for the most promising set of input parameters, in which I directly modify the shell occupancy of 8 Fe atoms and 8 F atoms. Unfortunately, SCF convergence remains very slow (150+ SCF loops so far and still no convergence).

Are there other things I could try to obtain SCF convergence?

crystal23_mil127.zip

Best regards,
Wim Temmerman - Ghent University

GiacomoAmbrogio

Hi Wim,

Could you please provide the geometry of your system? An output file would also be very helpful.

In the meantime, please have a look at this thread.

wgtemmer

Dear Giacomo,

Thank you for your quick reply. I have converted the fort.34 file to a .cif file and have added the latest CRYSTAL output. This output is the result from 2 restarts, so this job has executed 250 (100 in this OUTPUT file, + 150 from previous jobs) SCF loops in total so far.

In case it's relevant, I am running this calculation on 2 nodes (256 cores in total). I will try to tighten the integral tolerances and increase the grid, to see if this helps.

mil127_structure_output.zip

Thanks in advance for your help,
Wim

GiacomoAmbrogio

As a general suggestion, I would recommend adding SMEAR, and possibly enabling DIIS, which can also be helpful. However, when using SPINLOCK, be careful, you should only turn DIIS on after SPINLOCK is disabled, so make sure to use THREDIIS instead.

You can also try increasing FMIXING to around 90 or 95, and using slightly higher TOLINTEG values.

Additionally, increasing SHRINK to 2 2 could be helpful. If the system goes through a metallic state, this adjustment can improve the Fermi surface (computational cost should not increase significantly).

The input file should be similar to this: INPUT.d12

As an additional note, targeting a high-spin state with all Fe atoms having +5 spin may not be physically reasonable for this type of MOF. Such a configuration could lead to severe instability in the electronic structure, which often manifests as SCF convergence issues. It might be worth checking whether a lower or mixed-spin configuration is more appropriate for the system.

wgtemmer

Alright, it sounds like the difficulties originate from the system itself, rather than the particular CRYSTAL input.

Thanks again for all the help!

Best regards,
Wim