Do not subtract tolerance from box size

[mattwthompson]

Description

Resolves #1106

Checklist

• Add tests
• Lint
• Update docstrings

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 93.62%. Comparing base (dec585e) to head (deb3e9b).
Report is 96 commits behind head on main.

Additional details and impacted files

[mattwthompson]

@lilyminium any idea if we still want this?

[mattwthompson]

@lilyminium @j-wags do we want this?

[lilyminium]

Yes please, I think it still makes sense, despite the original issue being solved.

Unless I'm misunderstanding what the box tolerance should be for, which is 100% a possibility

Edit: I've changed my opinion, please see below.

[Yoshanuikabundi]

When I wrote this code, I was concerned that packmol might not produce boxes that tile very well, and rather than introduce a second parameter I re-used the tolerance parameter to introduce a small buffer around the edge of the box. This only happens in the call to packmol - the actual returned box should be the requested size or density, just with that small empty buffer on one end of each dimension. In other words, packmol attempts to create a slightly overdense box, which is then placed in the corner of the returned box. After NVT equilibration, the box would be the requested uniform density. Part of my assumption was that PackMOL outputs are really only useful as starting points for equilibration.

[lilyminium]

Josh and I just had a quick call where we talked over the above, and we both agreed we'd prefer the option that works generally across systems. Since all the tests I ran to troubleshoot the original packing issue did not implement this fix, that means I have a weak preference to not merge this PR since we know things work without it. I'm happy to re-run tests with this fix to validate if anyone has a strong opinion the other way of course! (And this behaviour in Evaluator seems to work ok)

Edit to clear up my previous confusion over the tolerance parameter -- it defines the space between molecules, and as packmol packs the box fully (?) that means the buffer that Josh has added becomes the space between molecules when tiling.

[mattwthompson]

Thanks all, my preferences here are weak so I'll not go with this PR but make an attempt to document this behavior (whether or not it's different than Evaluator's code)

[mattwthompson]

I will say that it's a little odd that the density is lower because of the 10% scale-up in each linear dimension (33% scaleup in volume) and higher because of how the "tolerance" is subtracted from the edges of the box.