MoSDeF Input Testing Matrix

[ahy3nz]

• We have a lot of structure/chemical formulas as reference systems
• We will soon have access to multiple general force fields (OPLS and GAFF) for atomtyping and parametrization

We need a somewhat-elegant and flexible way to test all these possible combinations. My first thought is just using it.product within mm_tests/mosdef_inputs to build + atomtype/parametrize all the different permutations and then shoot them all off to different engines

[mattwthompson]

What are the permutations, structures force fields engines?

[ahy3nz]

Yeah I was thinking something like that. I'm not sure what the actual usefulness is or if it'd just be something neat to try. In reality, I think basic unit tests that solely look at molecular models of a single function are probably the best at debugging force field translation

[mattwthompson]

I think each layer has a huge amount of utility

• Running over a bunch of structures gives us an idea of what chemistries give us problems (at each step in the process)

• Running over force fields gives us an idea of how much is covered

• Running over engines would let us diagnose issues with conversions or, potentially, errors in the engines themselves, particularly with exotic potentials/etc.

The last layer would be comparing the output energies (or other physical properties) to reference data. For example, we could probably write out a monumentally intricate molecule, type it in GAFFF, and run it in a bunch of engines, but are the results physical/reasonable/viablet/trustable? Probably outside the scope of this project but it's the sort of project I would like somebody to pick up. Basically I'm suggesting an updated version of this project (http://virtualchemistry.org/).