Re: [AMBER] Found bug in sander in restarting SGLD replica exchange runs

From: Carlos Simmerling <>
Date: Tue, 2 Feb 2021 20:03:28 -0500

I'm copying Xiongwu's reply (he's the RXSGLD developer) here so it can be
in the archive.

---------- Forwarded message ---------
From: Wu, Xiongwu (NIH/NHLBI) [E] <>
Date: Tue, Feb 2, 2021 at 7:53 PM
Subject: RE: problem in RXSGLD in Amber
To: Carlos Simmerling <>, <>

Hi Charo,

Thank you for pointing out the bug and submit the fix patch. I am not in
that email list so have to reply to you directly.

Your patch is a good fix of the bug and we will patch into next release.

Please feel free to let me know if you have any questions.

Best regards,


On Sun, Jan 31, 2021 at 7:36 AM Charo del Genio <> wrote:

> Dear all,
> in reference to my previous message, I think I did find the bug,
> in the end. What seems to happen is that, upon restarting the RXSGLD run,
> sander reads the final *guiding* temperatures from the
> replica records in the restart files, and overwrites temp0 with them. In
> other words, it treats it at a T-REMD rather than a RXSGLD. In turn, this
> causes the error down the line.
> Upon detecting that the run is a RXSGLD, and not a T-REMD, the correct
> behaviour for sander should be to overwrite tempsg, rather than temp0, with
> the final temperatures found in the restart files,
> since these are the final SGLD guiding temperatures for each replica,
> rather than their actual temperatures, which, for RXSGLD runs, are normally
> all the same and specified by temp0 in the input files.
> It is highly likely that this is due to the sander command-line option
> (-rem 1) being the same for both T-REMD and RXSGLD, together with the
> omission of checks in the code to distinguish between the
> two cases. If I may give a suggestion, a new command-line option should be
> introduced specifically for RXSGLD (for example -rem 6, which I believe is
> currently unused), because the differences in
> behaviour with T-REMD are only few, but they are subtle and substantial.
> This way, one would avoid similar problems in the future.
> Before a fix, a very cumbersome workaround is to make sure the restart
> files are ASCII (ntxo=1), then, at the end of each run, edit them and
> remove the last number in the second row of each file,
> keeping track of them. Then, edit all the input files and replace by hand
> the value of tempsg in each replica with the value of its current stage,
> which is the last number in the second row of the
> corresponding restart file (the one that was removed). Then, restart the
> job. Obviously, if one has more than just a few replicas, this can become a
> bit confusing.
> So, is there any way I can help in patching this?
> 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
> _______________________________________________
> AMBER mailing list
AMBER mailing list
Received on Tue Feb 02 2021 - 17:30:03 PST
Custom Search