Hello Charo,
I can't speak directly to this issue, but I can confirm that the REMD
exchanges do occur a step later than expected/reported (e.g. Setting nstlim
100 means exchanges are at steps 101, 201, etc). This is probably a quirk
of leapfrog and they way kinetic energy is defined, but I could be wrong.
To be clear, I've never detected issues from this, but your reasoning
around RESPA sounds reasonable. Hopefully a REMD/SGLD expert can add more
background?
Slightly off-topic - is the benefit from RESPA particularly good? With mass
repartitioning on a GPU I've gotten the impression that the benefits are
quite marginal these days.
Cheers,
Brian
On Sat, Jan 30, 2021, 6:24 AM Charo del Genio <the.paraw.gmail.com> wrote:
> Dear all,
> yesterday I reported that, in RXSGLD runs, I was finding a serious
> discrepancy between the potential energy as computed by the REMD part of
> the engine and that computed by sander. I finally narrowed
> down the origin of the problem, and I am reasonably certain that it has to
> do with RESPA. In fact, if I stop using it (nrespa=1), then the discrepancy
> completely disappears. For sake of completeness,
> I note that the behaviour persists regardless of system size. For example,
> it even happens with pentapeptides. The discrepancy can be huge: in my
> system of interest, the "correct" potential energy is
> -2157 kcal/mol, whereas the replica exchange engine computes it as 13257
> kcal/mol (*positive*). Obviously, this also indicates that the exchanges
> happen or are rejected according to wrong energies,
> and, therefore, the final trajectories are more or less nonsense.
>
> I strongly suspect that this behaviour is due to the REMD engine computing
> the energies at step numbers that are not multiples of nrespa, thereby
> getting meaningless results.
>
> So, is this a bug, or is RESPA not supposed to be used in REMD runs?
>
>
> Cheers,
>
> Charo
>
>
>
>
> --
> Dr. Charo I. del Genio
> Senior Lecturer in Statistical Physics
>
> Applied Mathematics Research Centre (AMRC)
> Design Hub
> Coventry University Technology Park
> Coventry CV1 5FB
> UK
>
> https://charodelgenio.weebly.com
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Sat Jan 30 2021 - 07:30:03 PST
