-
Notifications
You must be signed in to change notification settings - Fork 22
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Packmol interface: can't pack mnsol benchmark inputs with assumed evaluator defaults #1062
Comments
I'll have a full list of solvent/solute combos a bit later, my tests are running super super slow, so erm.. maybe in a day or so. |
Part of the issue here might just be that the target density is too high - n-butanol has a density of 0.8 g/ml! Reducing the density does fix at least this case. Were people manually setting the density for each solvent type with evaluator? |
Unfortunately I don't know past what's in the paper. My experience with Packmol in other projects had me often packing at ~20% less density and running equilibration for longer. Packing has been such a headache for me at ~5% under the target density that I'd avoid packing above the target density whenever possible |
Thanks, it would be great to hear from the folks that do use evaluator for non-water mixtures how they do things (maybe they just set the value to 0.8 too). Do you think there's scope for changing the default value @mattwthompson ? I.e. |
Short answer: yeah, no thought went into that default value and something more like 0.8 g/cc or lower is easy to defend. Longer answer: will sync up with scientists for more feedback here - stay tuned & thanks for your patience! |
I don't have a diagnosis of why the behavior is different, but I have consensus on lowering the default value. For a study you might benefit from using a different density for each system, but I'll probably just set it to a default value around 0.7-0.8 g/cc. Do any of your systems fail to pack at that density, or possibly for funkier reasons @IAlibay? |
Thanks, I'll have to check a bit later this week (for this current set). That being said, I am still waiting on a set of necessary solvents from the OpenFF science team, until I have those my development efforts are stalled. Will get back as soon as I have the full details. |
I'm going to be updating the default value of this argument to somewhere between 0.6 and 0.8 g/cc, which I expect to handle most issues you've been running into. When you get back to this (on your own timeline - I'm not a stakeholder here) could you update your scripts to use a lower target mass density and see if that fixes your issues? This behavior won't land until 0.4.0 or 0.4.1, so I have to ask you to manually update things until then. |
Is there any chance solvent packing tests can be implemented at the Interchange level? It feels like doing this downstream is incorrect. |
Probably - I think the unit tests are relatively strong but regression/application tests for packing something at the scale of a realistic simulation. I think I can distill your issues down to packing not converging at 0.95 g/cc? |
i think it's more "it doesn't match evaluator" (which may not be a bug!) |
Maybe to add to this - it looks like somewhere we're going to have to add a test with a list of solvents that we expect to be able to pack properly in a way that doesn't crash (for the use of the OpenFF science folks needing to run those systems for benchmarks). I can do that downstream but, especially given the context, it feels like this should be an interchange test not a downstream one. |
I think that's fair - perhaps we can spin up some extra resources to run a larger set (tens?) of solvents at realistic sizes (~5 nm cubes) every once in a while, like a weekly cron job or every time something is committed to the For now I have only partial solutions (#1067 and #1068 should get us most of the way) with hopes of unblocking that work whenever you end up getting back to it. (No time pressure from me personally on this!) |
Description
I'm trying to test if we can pack mnsol (from the 2.0 sage benchmark) with Interchange.
My rough understanding (from the paper and some limited looking at the evaluator v3.11 code) is that it was possible to pack all the systems with 2k molecules, a cubic box and a 0.95 * unit.gram/unit.milliliter target density.
However, I can't seem to be able to make this work using Interchange.
The main questions here are:
Reproduction
Here's a good example case:
Output
Software versions
Interchange v0.3.29
The text was updated successfully, but these errors were encountered: