[Question] How to choose between SAMS & REMD

[yanze039]

Hi,

It's not a real issue, but I would like to explore more about the comparison between the two provided samplers in openmmtools, SAMS & REMD.

Practically, I can run SAMSSampler through a single replica to explore all thermodynamic states with desired probability distribution (no communication between replicas); while REMD requires the exchanges between multiple replicas (communication between replicas).

I am curious is there any efficiency difference between these two schemes , and which criteria should I rely on to pick a better one?

I am more interested in the sampling problem, so my context of "efficiency" would be the sampling capability, i.e. how efficient it can enhance sampling given the same resources (say, 8 gpu cards) and replicas (say, 8 replicas).

I would appreciate any resources, suggestions and discussion.

Thanks!