I am trying to understand the stability of the hypothetical MOF crystal. According to my analysis, there is only one imaginary frequency. In DFT calculations, this only one negative frequency can be mitigated by enhancing the k-point set and cut-off radii of the reciprocal space in some cases.
As far as I know, supercell method (finite differences method) is used in GULP. I used the supercell option and gfnff_kspace keyword, nothing changed and I received momery error, respectively. When I use the supercell and kpoints options, the same memory error arise.
Do you mind if I asked the possible ways to calculate more accurate phonon frequencies? Relevant files can be seen below.
If there is an imaginary mode with DFT then unless this is an artefact of insufficient parameters (such as a real space grid) then changing the phonon k points won’t solve this (it’s not clear whether you are referring to electronic or phonon k points).
For most methods, including the one that you are using, GULP uses analytical 2nd derivatives and not finite differences (though you can force this if you really want for testing purposes, though it’s not recommended).
It’s not necessary for you to build a supercell to compute the phonons away from gamma since GULP will automatically do this if you set up k points. In your case this results in a 3x2x2 supercell which is smaller than your supercell and is determined on the maximum real space range of interaction. This is only used here because k points are not directly implemented yet for this method & so the supercell approach is automatically used.
Running out of memory is obviously to do with your machine not having enough for the calculation you’ve asked for. Remember with your 4x4x3 supercell this means you have 6768 coordinates (3N) and so the matrix will require ~360 MB at gamma, though 2.5 copies may be required and so this should only be about 1 GB of memory in total. If you run out of memory then try quitting other applications, use a computer with more memory and/or run in parallel.
None of the above has anything to do with the accuracy of the phonon frequencies, which depends on the force field or level of QM theory that you are using.
Hope this helps,
Julian
This post has been removed by the moderator for containing incorrect information. It asserted that GULP uses finite differences (not true - the second derivatives are analytical) and suggested that 0.01 Angstrom was a good starting point for finite differencing (it’s too large for accurate 2nd derivatives). See the earlier post where the correct response to this post was given.