Unbiasing of the harmonic restraint

This is my plan for implementing the harmonic restraint unbiasing in the analysis.

1. Parse solute-only trajectory and recompute the centroid distances and restraint energies.
2. After decorrelating, discard from the analysis all the frames with restraint energy greater than a threshold. The function will have an argument defining it, but what should the default value be? 2kT? 3kT?
3. Recompute the standard state correction by integrating a different potential that is the same as the restraint until it gets to the threshold, at which point it goes to infinity.
4. If the simulation used the anisotropic long-range dispersions correction, subtract the restraint energies to that state; otherwise, create a new state for MBAR that has the energy of the lambda=1 state minus the restraint energy.

I’d implement this in an external script for now (starting from this one, which is almost ready) since there are so many changes going on in the analysis, and I’d like to avoid re-doing the same work twice. Once the new analysis API will be merged into master, I’ll bring it into the analysis.

[jchodera]

@andrrizzi : I think you’re right that there’s enough value in simply unbiasing the harmonic restraint given a user-specified energy threshold that we should implement this for urgently-needed projects (Musashi, Merck) and then spec’ing out a more complete contact-based definition for analysis later.

[andrrizzi]

This was implemented in #915 .