A problem of file parsing for simulations with multiple coupled states

[wehs7661]

Hi alchemlyb developers,
I was running a Hamiltonian replica exchange simulation of 140 states, with the first 4 states being the coupled states (\lambda_{coul}, \lambda_{vdw}, \lambda_{restr}) = (0, 0, 0) since I need more data from physical states. However, I found that with the current GROMACS parsers, including extract_u_nk and extract_dHdl, the u_nk and dHdl returned only had 137 states instead of 140 states since the parses did not distinguish the states with the same lambda vectors. This will lead to wrong overlap matrix such that the probabilities of the same row do not sum up to 1. I also had a hard time converging when initializing pymbar.mbar.MBAR(), but I was not sure if convergence issue is related to the parsing issue. This might be an issue worthy to address. Thank you!

[mrshirts]

@wehs7661, can you upload a small data set (maybe just 20 lambda, for not too many steps) that has the same issue? It's much easier for developers to see the issues when they have data to look at.

[dotsdl]

Hi @wehs7661!

Hmmm, this is an interesting case; thank you for bringing it up!

I agree with @mrshirts that if you can package up a reduced set of this data and put it somewhere downloadable it would be useful to proceed. Expanded ensemble is an approach we definitely want to get right, and this is a case I haven't encountered yet.

Please also include a minimal working example of the code you are applying to this data to produce the problem you are seeing; that will help to make sure we're both approaching it with the same expectations of what "should work."

[wehs7661]

Hi @mrshirts and @dotsdl,
Thank you for your responses! Sorry that I didn't see the replies until this morning. I just ran a replica-exchange simulation of 28 states with 500 steps. I also wrote a jupyter notebook where the problem can be easily reproduced with some explanations of the problem. All the relevant files are compressed and can be downloaded here. Please let me know if you have problems accessing it or need more information. Hopefully this makes the problem more clear. Thank you!