add τ in energy_hamiltonian required for some TermXc

[aouinaayoub]

if we use some meta-GGA functional, save and load the scfres with skip_hamiltonian=false (default), it causes energy_hamiltonian to error due to the missing kinetic energy density τ key, despite τ being available in the saved scfres

[mfherbst]

Thanks for observing this, very much agree something like this should be around.

The problem is this way you assume \tau to be always part of the jld, which may not be the case. I'd fence this off with some extra checks.

I think we have also a function to check whether \tau is required for the hamiltonian (see SCF).

[aouinaayoub]

Looking at how quantities are collected in scfres (https://github.com/JuliaMolSim/DFTK.jl/blob/master/src/scf/self_consistent_field.jl#L227), it seems τ will always be part of the jld, it's set to nothing when it's not needed.

[mfherbst]

I see, but that's actually not intended behaviour. The save_scfres has a flag save_ρ which controls whether ρ is saved (to be able to save disk space in case only high-level results such as the energy are needed). I'd argue the same flag should control whether τ is saved, too. So don't rely one a scfres[:τ] to be present.

I'd do something like

https://github.com/JuliaMolSim/DFTK.jl/blob/master/ext/DFTKJLD2Ext.jl#L105

also for τ and modify appropriately the check

https://github.com/JuliaMolSim/DFTK.jl/blob/master/ext/DFTKJLD2Ext.jl#L151

such that if τ is nothing, but the ham needs τ the ham is not built.

Probably we should also modify the behaviour of saving τ when save_ρ=false and document it appropriately.

[aouinaayoub]

I set τ to be retrieved from the dictionary or set to nothing, then check if a term in the basis needs τ before calling energy_hamiltonian.

[mfherbst]

There were some unrelated issues in the CI I had to fix plus I added the extra features I briefly mentioned on saving the scfres.