Re: An running issue in eABF-replica calculations

From: Jérôme Hénin (
Date: Mon Apr 12 2021 - 08:53:14 CDT

Hi René,

>> MW-eABF is not that different from ABF, it's just that the ABF quantities
>> (mainly the biasing force itself) pertain to the extended coordinate. Those are
>> shared exactly as in standard mwABF, because they are represented the same way
>> in the implementation.
>> The data for the *unbiased free energy estimators* is a separate question: it is
>> not shared in the current state of the code, and in a sense it does not need to
>> be shared during the simulation, as these estimators are really a matter of
>> post-processing. Anything that is done on the fly is just for convenience.

> 1. Does that mean that the output of multiple replicas simulations (using shared
> (e)ABF) MUST be merged in order for results to be valid

> since for a single replica the "actually" applied biasing force does is
> different from the corresponding content of the (zcount/zgrad) output files
> produced by this walker?
The zcount/zgrad data for each walker is still valid for the given replica. As the simulation converges they should converge to the same final values across replicas, even without sharing.

> 2. What is actually printed to the outputname.grad/count file (named the same
> way as in plain abf output) when using eABF, this is the gradient/counts of the
> fictious particle, correct?
That is correct.

>>>>> * Then latest literature would be a bit misleading when stating usage of
>>>>> MW-eABF.
>> Is there something else than the QM/MM tutorial that mentions it?

> It is stated in the paper about WtM-eABF Method ( Taming Rugged Free Energy
> Landscapes Using an Average Force [
> |;!!DZ3fjg!uPMvYCFbbdXOyaMgjQOMbuuCd681Djcttd8uytDXrz09dJFFUsowi6-jGHVG9H8jxg$ ] )
Ah, indeed.


This archive was generated by hypermail 2.1.6 : Fri Dec 31 2021 - 23:17:11 CST