Minimum supercell size for periodic calculations

[Eric Patterson]

Hi all,

I am a complete beginner with CP2K and would like to perform CELL_OPT followed by VIBRATIONAL_ANALYSIS on periodic solid molecular systems (mostly organics). Are there general guidelines for how big a supercell must be for results to be considered reliable, either in terms of absolute dimensions or # of atoms/molecules? 

By way of examples, if I study glycine, there are only two molecules in the unit cell, which I assume is much too small. If I use a 2x2x2 supercell, there are 16 molecules (160 atoms) and the cell dimensions are between 10 and 12 Angstroms in all dimensions. If I chose benzoic acid, the c vector of the unit cell is already greater than 20 Angstrom, so I am considering a 2x2x1 supercell (16 molecules or 240 atoms, cell dimensions of 10 or 11 Angstroms in a and b, 21 in c). Are these reasonable choices?

I plan to use PBE with GPW, MOLOPT basis sets, and NLCC potentials if that matters.

Any guidance is appreciated.

Thanks,

Eric

[Xavier Bidault]

Hi Eric,

You can have a look at these papers and relevant details in their supplementary information.

https://doi.org/10.1063/5.0041511

https://doi.org/10.1039/D2RA05062E

As a matter of fact for epsilon-CL20, I could have just used a 2x1x1 supercell, making every dimension larger than 10 A, but an attempt with a 2x2x2 supercell showed that 2x1x1 was not enough (unconverged low frequency range of the vDOS). My guess was because the crystal is not orthorhombic and that I had to use a larger supercell to allow for more vibrational freedom.

You can also see my concerns about Grimme D2 vs D3(BJ). But this may only concern nitramines.

Yes, NLCC GTH PP with Molopt-DZVP in CP2K should do the job. But no less then 600 Ry for the energy cutoff.

Best,

Xavier

--
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+uns...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/724a26c2-31a5-4240-b1c1-5740a26b0d18n%40googlegroups.com.

[Eric Patterson]

Hello Xavier,

It was actually your paper that gave me the idea that something like 16 molecules and/or ~10 Angstroms per side cell might be a good general guideline. I appear to be getting sufficient results with this approach, but your caution about CL20 is instructive.

Thank you!

Best,

Eric

To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/CAD0N%2BNWnPQapVc0jeYMrEXoYVqe_Ghw62UN665Buq_9nQYZKWg%40mail.gmail.com.