2021-05-14 OpenMM/OpenFF check in

JW – Sorry for not getting SMIRNOFF spec changes from PE+JC email thread distilled+proposed. This is still on my to-do list.

MT – I’m implementing the splitting of nonbondedforce into two CustomNonbondedForce and two exceptions

(General) Is PME cutoff important? Are there cases where 1-4 distances could span the cutoff so they’d get mangled with exceptions?

• MT – so cutoffs and 1-4s don't interact with each other in the vast majority of cases.

• PE – If you’re looking to separate vdW and Electrostatics energies, then having two CustomNonbondedForces, then this is a fine choice

•  

MT – If I ALSO wanted to get 1-4 pair energy for each component, what’s the best way to go? Could I have two exception forces?

• PE – I think that would work, but I’m not 100% sure. Feel free to test it out and tell me if anything comes up. There may be some interaction between PME and 1-4 interactions, but I think it could be handled right.

• MT – So this would be a NonbondedForce, CustomNonbondedForce, and two BondForces

MT – I’ve been working on energy equivalence between engines

JW – Thanks for feedback on PME questions. I’m inclined to agree with you re: giving some ground on how much we specify nonbonded, and how much we leave up to the engines

MT – How self-consistent are MD engines?

PE – A few factors:

• Things will always be limited by machine precision.

• Last-digit differences may arise based on the order of adding numbers, which can be effectively random even on the same machine, since the order of operations may affect rounding.

• Insufficient sampling of energy landscape will lead to differences in outcome

MT – Given a single-point energy evaluation, how reproducible should results be?

• PE – You’ll get basically the same result, maybe a digit or two less than machine precision.