choderalab / chiron

Differentiable Markov Chain Monte Carlo
https://github.com/choderalab/chiron/wiki
MIT License
14 stars 1 forks source link

Multistate sampling #8

Closed wiederm closed 4 months ago

wiederm commented 8 months ago

Description

Simulate multiple (thermodynamic) states. This base implementation provides multistate sampling, assuming a single replicate and sampler state per thermodynamic state. It also implements these methods for sequential computation. This implementation uses code from openmmtools where possible.

Todos

Status

codecov-commenter commented 7 months ago

Codecov Report

Merging #8 (bb58931) into main (58db5df) will increase coverage by 0.99%. The diff coverage is 88.77%.

Additional details and impacted files
chrisiacovella commented 7 months ago

I think we're also going to need a Monte Carlo barostat soon. We can start off with just an atomic-scaling barostat, but that should be enough to get us going. The easiest test system will be an ideal gas (noninteracting atoms) test system.

I put together the atomic-scaling barostat in PR #14 that is more or less ready to go ( I still need to add in a CI test for an ideal gas...a test of TraPPE UA methane converged to the anticipated density, but that is too time intensive for CI).

chrisiacovella commented 5 months ago

I did a few commits where I made the multistate code support accepting a list of neighbor/pair lists (this required some changes to the neighbor/pair list logic itself, as they weren't properly supporting systems without box_vectors set). I'll note that the harmonic oscillator tests don't actually use the neighbor/pair list at all (not supported in the potential, for obviously reasons), so I think we might still need some additional functionality tests to make sure these are used correctly.

chrisiacovella commented 5 months ago

I will create a separate Example of the multistate harmonic oscillator (basically a standalone copy of the code in test_multistate). This can explore the number of steps to converge to the analytical values of the free energy to make sure the the number of steps actually are sufficient.

wiederm commented 4 months ago

Thank you for taking this over @chrisiacovella ! Great work!

I agree that we can drop 3.9