Test and document external calculators

[ElliottKasoar]

Resolves #380

[harveydevereux]

I think this is a good addition for the docs, is there any reason why this would not work?

Obviously if the calc depends on something and you don't have it installed (lammps) it won't, which is fine.

But are there breaking code changes we or ASE might make, justifying testing subbing e.g. LennardJones into struct.calc et al.?

[ElliottKasoar]

I think this is a good addition for the docs, is there any reason why this would not work?

Obviously if the calc depends on something and you don't have it installed (lammps) it won't, which is fine.

But are there breaking code changes we or ASE might make, justifying testing subbing e.g. LennardJones into struct.calc et al.?

Yes it inevitably requires the user to set up the calculator correctly, which may well require other installations, and I'm not sure exactly how it would behave with the less native calculators e.g. CP2K requires running something like:

CP2K.command="env OMP_NUM_THREADS=2 mpiexec -np 4 cp2k_shell.psmp"

I knew that every calculation after SinglePoint was built in a way that it could (initially it had to) take SinglePoint.struct inputs, which meant it wouldn't try to reattach a calculator if it was already there.

The only thing I wasn't sure about was if anywhere we assume things like struct.calc.parameters having certain values, but I've always tried to make sure that sort of thing is optional.

I think in general the ASE calculator interface could change, but I don't think anything would break something like LennardJones that wouldn't break MACE etc., so perhaps it would be sensible to test for it.

[oerc0122]

Nice and simple, if it does the job, happy with it.