Add option for deterministic swap order (DEO) to ReplicaExchangeSampler

[pujaltes]

Description

This PR adds the option for deterministic_swap_order to ReplicaExchangeSampler for deterministic even/odd (DEO) swap alternation when swapping between pairs of neigbours (swap-neighbors). When enabled, the swap offset alternates deterministically between even and odd swaps which has been shown both empirically and theoretically to improve sampling efficiency across a variety of systems [1, 2]. Notably, DEO appears to be the default in the GROMACS/PLUMED replica-exchange implementation.

Note that I define a new private _deo_odd_offset attribute to track the even/odd swap order since tracking via _iteration could cause issues since this is not updated during the equilibration phase.

Todos

• Implement feature / fix bug
• Add tests
• Update documentation as needed
• Update changelog to summarize changes in behavior, enhancements, and bugfixes implemented in this PR

Status

• Ready to go

Changelog message

Add ``deterministic_swap_order`` parameter to ``ReplicaExchangeSampler`` for deterministic even/odd swap ordering (DEO).

[mikemhenry]

Thanks for this! I think all that is missing is some testing. Is there an analytical way we can test this?

[codecov]

Codecov Report

❌ Patch coverage is 78.94737% with 4 lines in your changes missing coverage. Please review.
✅ Project coverage is 85.26%. Comparing base (7b6e86c) to head (0e582c2).

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[pujaltes]

Hi Mike,

Thanks for reviewing the proposed code.

I drew inspiration from noegroup/reform and added some tests to empirically validate the deterministic (DEO) and random swapping schemes (SEO).

For DEO we just check that it alternates between even and odd pairs and for SEO we check that: (1) it isn't DEO; i.e. doesn't just alternate between even and odd swaps and (2) that it proposes swaps between even and odd states; i.e. doesn't exclusively propose even or odd swaps.

The SEO checks have ~0 probability of false failures (0.5^100), but I can remove them if you prefer and keep only the deterministic DEO tests.

I also ran yapf on test_sampling.py but it reformatted the entire file, so I excluded those changes to keep the diff focused. Let me know if you want me to run formatting before merging this.

Best,
Sebastian