Openff-toolkit: Where should simtk.unit.Quantities be enforced?

[j-wags]

I'm going ahead with the idea of using simtk.unit.Quantity's wherever applicable, but I don't want to go overboard. Could I get a yes/no on the following cases?
1) require that distances are in a distance unit (eg conformer positions, virtualsite definitions in Angstrom/nanometer)
2) require that angles specify an angle unit (virtualsite in_plane_angle definition in degrees/radians)
3) require units of charge for electrostatics (eg virtualsite charge_increments in electron charges)
4) require units for epsilon/rmin_half/sigma when defining VdW parameters in a VirtualSite (different units, might be overboard if there's just one convention)

[davidlmobley]

👍 to all of these. Many codes use different units for length, energy, and angle, so anytime you have any of these involved there is potential for confusion or a need to consult the docs. That means that item 4 is especially a point of confusion as it combines length and energy.

(If only I had a dollar back for every time I had to go poking around in documentation to find out what units I needed to specify some quantity in. Please, no more.)

[jchodera]

I think we should definitely require users specify units of the correct dimension for any dimensioned quantities. This absolutely reduces critical and subtle failures.

Also, see the Mars Climate Orbiter.

[j-wags]

Ha! Ok, that settles it. unit.Quantity wrapping for all.

One odd thing is that OpenMM insists on getting epsilon in kJ/mol [1]. The per mole part becomes part of the unit. This would prevent someone from defining it in terms of a per-particle quantity (eg... electron*volt). On one hand, it might be good to leave this as a reminder for folks to distinguish between per-particle and per-mole quantities. On the other, forcing it to be a per-particle quantity might be cumbersome. For now I'll play it safe and check input for strict compatibility with kcal/mole.

Any thoughts?

[1] http://docs.openmm.org/7.0.0/api-c++/generated/OpenMM.NonbondedForce.html

[j-wags]

Unit type checks for all virtualsite parameters implemented in commit https://github.com/openforcefield/openforcefield/pull/119/commits/44cfcb8ac4c0dbb901172cf71387f2b21ab56ad8