On 06/01/2025 19:41, Guberman-Pfeffer, Matthew via AMBER wrote:
> Hello Kiriko,
> I’ve performed exactly those simulations for a multi-heme named OmcS up to 340 K (but only published the 270 and 300 K results: https://pubs.acs.org/doi/full/10.1021/acs.jpcb.2c06822). I never saw this issue, but to Claro’s point, I only ran briefly on CPUs to equilibrate the density; everything else was on GPUs.
> Spikes in temperature cause me to think of spikes in velocities. I would be curious to know how you are building the topology with TLEaP? Are the necessary bonds between the heme and the protein explicitly defined (as they must be)?
> I’ve developed a program to, among other tasks, automatically prepare multi-heme proteins for AMBER simulations. Maybe it can be of use to you. The GitHub repo is here: https://github.com/Mag14011/BioDC.git
> Best regards,
> Matthew
Hi all,
one other thing I was finding together with the temperature "spikes" were similar spikes in some energy terms, most notably the BOND energy, when analyzing the trajectory with cpptraj. I tried to
chase this bug in different ways, but in the end I was unable to have a good idea of what was going on. Adrian Roitberg and David Case also tried to help, and provided some ideas and insights, but
unfortunately their suggestions also proved inconclusive.
The main point that we are sure about is that the problem only happens when using the MPI (CPU) version of PMEMD with the Monte Carlo barostat. Incidentally, this also means that implicit-solvent runs
are not affected. If one uses the Berendsen barostat the problem disappears, even though I could not find out where the bug is. Sander does not have this problem, and neither does the GPU version of
PMEMD, which would suggest that the bug is somewhere in the parallelization part of the code. Note that it will also have to be something that is triggered only when using the Monte Carlo barostat.
Then again, it is also possible that it is related to this paper
https://onlinelibrary.wiley.com/doi/10.1002/jcc.26798 If this is the case, then one could try rewriting PMEMD to use a soft cutoff.
To conclude, unless this is somehow addressed (which I don't see happening for the foreseeable future), the only safe thing is *avoid the use of PMEMD.MPI + MC barostat*. Either use sander, or the
Berendsen barostat, or run on GPUs.
Cheers,
Charo
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Mon Jan 06 2025 - 10:30:02 PST
