Hiphive notes

[ajjackson]

I made a little proof-of-concept notebook to show that hiphive force constants can be used with Euphonic without involving Phonopy at all. This allows anharmonic renormalisation of frequencies by creating effective potentials corresponding to different temperatures.

Some aspects could use more work:

• I haven't quite figured out the cell ordering for non-diagonal supercells. This is sort of avoidable by giving Euphonic an oversized diagonal supercell (which doesn't really affect the underlying force constants) but it would be nice to handle arbitrary cells.
• It would be good to have an example including long-range dipole-dipole corrections. There are two steps to consider:
  • To create such a model hiphive needs training data with the electrostatic forces subtracted. Currently the suggested workflow uses Phonopy. I tried to do it with Euphonic's own Ewald sum mechanics but it was very slow; we may need to do some lower-level work to make this performant. Alternatively, we could look to import or implement the Ewald-sum forces in another package; it's not clear that this part needs to live in a Phonon package at all, and it could give a cleaner overall workflow.
  • Once such a model has been created we should be able to bring it into Euphonic fairly easily by adding the electrostatics information (i.e. Born charges and dielectric tensor). Unlike Phonopy/QE results we should not need to subtract the long-range terms as hiphive only deals directly with the short-ranged model.

But I think these notes are already a useful starting-point and allow existing models to be used.