Josh Fass

>@maxentile : Would there be any issues if we did this (if we're not measuring work)? I don't think so! To be risk-averse, we erred on the side of including...

Quick question: Can you confirm if this error can also be triggered using the gaussian_work test system? https://github.com/choderalab/pymbar/blob/15f932a271343e611ed4be2d468c77b1d11cf01f/pymbar/testsystems/gaussian_work.py#L4 This would be helpful in investigating the bias part of this issue,...

>Only PR that affected torsions that I can think of is: >https://github.com/proteneer/timemachine/pull/626/files Checking if reverting that PR fixes this test.

Locally, this fails on master at a different assertion: https://github.com/proteneer/timemachine/blob/f195c00b4605173b9d970830025e452fb3e2f159/timemachine/md/enhanced.py#L348

Perhaps the context constructor should even assert something stronger, e.g. that the minimum box edge length >= cutoff / 2?

Checking in the context constructor made more sense when the box dimensions were constant for the lifetime of the context -- seems better for this check to be in the...

Clarification: is the main concern: 1. force singularities that occur at a single point (dU/dx = nans when theta = theta_0 = 0 for example), 2. energy discontinuities due to...

(Tangent: May be worth looking at numerical stability considerations for other angular distributions such as the Von Mises distribution, e.g. in Stan https://mc-stan.org/docs/2_22/functions-reference/von-mises-distribution.html or Mitsuba https://www.mitsuba-renderer.org/devblog/2012/07/numerically-stable-sampling-of-the-von-mises-fisher-distribution-on-s2-and-other-tricks/ )

>There are no discontinuities in the energy in any of the above functional forms. Oh! Shoot -- you're right. (Assumed there would be a discontinuity since `theta - theta_0` is...

For future reference, these were points of my confusion: * Forgot the domain of theta here is [0, pi] not [0, 2pi] (or arbitrary [x, x + 2pi]). It is...