Generalize REUniverse

[theavey]

As implemented, this is really focused on parallel tempering-type replica exchange (where the temperatures differ between the replicas).

It would be good to ask someone who does non-temperature replica exchange to figure out how it would be best to do the indexing.
There might be some other parameter that might be natural to index by.

At the very least, possibly temps could be renamed to index or similar, and made optional.
Then, there could be a new argument index_type possibly that if it is 'temp', the replica's temperatures can be set by index.

[theavey]

Hey @dstelter92, you have any insight into this?

I have an class this is a set of replica trajectories. I mostly index them by temperature, which is natural for me, but do you have any thoughts on a more general way to index?

I guess the most general would just be an index, but that could take away by fun of being able to access one with REUniverse[temperature].

[dstelter92]

The way I did this in LAMMPS was to index by 'world' (essentially indexing by processor if 1 proc/world). Then have separate helper functions, world2temp, and temp2world. Then you can access REUniverse[temp2world[temperature]] or just REUniverse[world]. FWIW- this approach is much easier to parallelize.

[theavey]

So just essentially some arbitrary index that you keep track of using temp2world?

That seems very general: just use some index, but then define helper functions to go from some useful metric to the general index.

[dstelter92]

Yep. In the case of running the simulation, the index corresponds to the MPI world in the MPI universe. In your case for post-processing, it would be arbitrary.