Progress

• Make reproducing case for reported issue
• Ensure case reproduces for openff-1.2.0, but not openff-1.0.0 or openff-1.1.0
• Make openff-1.2.1 candidate (both constrained and unconstrained), with changes described below
• Ensure candidate FF does not reproduce error
• Follow Standard Release Process

Problem Description

On Aug 11, 2020:

Christopher Bayly 11:09
We have just uncovered a really bad symptom with Parsley 1.2 when running MD on a molecule containing a a propyne substituent on a phenyl, as in c1ccccc1C#CC. The symptom is that the MD blows up and returns a NaN, but it only happens when Hydrogen Mass Repartitionlng (HMR) is turned on. We looked into it and the 4 fs timestep with HMR is the cause: the methyl bond with the triple-bond carbon hits a resonant frequency and ultimately explodes causing the NaN. This should not happen and I think I can see why: I think the force constants for the C#C triple bond and the attached methyl C-CH3 do not make sense. For the latter (single bond), bond type b24 has a super-high force constant 1.2977e+03 that you would expect for a triple bond, not a single bond. Conversely the triple bond with bond type b27 has a weak force constant 8.23e+02 that you would expect for a single bond not a triple bond. You can check this against the other "single bond" and "double/triple bond" force constants in the same section of the offxml. The smoking gun is that when we use GAFF2 which has more-or-less expected k's (high for the triple bond, regular low for the single), there is no issue with MD using HMR. (edited)

(gif from Gaetano Calabro)

Proposed resolution

Per decision in FF-release meeting on September 3, 2020

Decisions:

We will modify 1.2.0 by hand to assign b24’s k to 800, b27’s k to 1000 → Rename to b24-MANand b27-MAN → test against propyne containing molecule (jax 15 beta-secretase case from DH) → Release 1.2.1