Ultrasoft Pseudopotentials

[Technici4n]

Ultrasoft (US) pseudopotentials would be a good feature to have. They are quite standard: example quite a few pseudopotentials in SSSP are ultrasoft, for example, and they can also pave the way to PAW (#987) in the future.

Here is a roadmap with rough milestones we discussed with Michael:

• Check that the LOBPCG eigensolver works with generalized eigenvalue problems.
• Parse the US pseudopotential files, and implement the local term, the nonlocal term, and the augmentation charges.
• Check the total energy against Quantum ESPRESSO, but without doing a full SCF. In practice this means: running an SCF with a NC pseudo, then using that density to evaluate the energy non self-consistently using an US pseudo. Do the same with both DFTK and QE and check that the results match.
• Check that the potential/hamiltonian is also correct. Could simply use finite differences against the total energy.
• Validate a full SCF, and bands.
• Also support US for various kinds of post-processing such as forces and stresses.

To be clear, I don't think anyone is working on this at the moment, but we took a bit of time to write down this "roadmap" to have something more concrete than "just implement ultrasoft".

[antoine-levitt]

I tested LOBPCG with generalized eigenproblems so that should be good. In principle it should not be very hard, it's just a matter of plugging the right formulas at the right places.