search

IRL2 / nanover-protocol

https://irl2.github.io/nanover-docs/#
Other
0 stars 1 forks source link

Cannot specify simulation parameters with ASE-OMM CLI #208

Open rhoslynroebuck opened 1 month ago

rhoslynroebuck commented 1 month ago

The ase-omm CLI command takes (requires?) as an input a NanoVer-OpenMM simulation xml, however, this runner does not actually read the Integrator field from this xml. This means that you can't specify dynamics parameters (e.g. temperature, timestep) from the command line and these parameters are actually set to some default values. It seems misleading to provide this CLI functionality if these parameters are ignored, and can't actually be changed from the default parameters using this interface (this can only be done from a python script).

It seems like the only use case would be if (a) we provide a way of writing a NanoVer-ASE-OMM xml that specifies parameters that are read the by the ase-omm runner AND (b) these parameters differ somehow from the available OpenMM parameters, otherwise you could just use the standard OpenMM runner.

Maybe we should consider deprecating this CLI for the moment and think about if it's worth doing the above.

If any of this information is incorrect or anything vital is missing, please do say!

Ragzouken commented 1 month ago

In a recent merge both nanover-ase-omm and nanover-omm-server were replaced to just tell you to use nanover-omni --omm and nanover-omni --ase-omm. There's no CLI arguments to do anything with the integrator here either though. I don't think the runners even offer any explicit options for this other than letting you access the integrator yourself to change values.

I agree that it probably doesn't make sense to support ase by CLI since it doesn't fully follow the input XMLs and the main motivation for ASE is having deeper access in python to dynamics (which you obviously don't have on the CLI). Only other thing I can think of is that ASE supports the boundary walls and regular OMM doesn't? Is that right?

The generic CLI even for nanover-omni will always be limited and maybe it's only really fit for demos vs. say launching it from a bespoke python script or jupyter notebook where you can set up exactly what you want. Maybe it would be good for us to figure out exactly what use cases for CLI we want to support vs what ought to be done in custom notebooks or setup scripts?