Re: [AMBER] T-REMD Reproducibility - Cluster v.s Desktop

From: Gustavo Seabra <gustavo.seabra.gmail.com>
Date: Fri, 25 Sep 2015 15:18:20 -0300

Hi Thiago,

Just to add my R$0,08 to what the others have already said here…

Remember that MD is a chaotic system. With 14 replicas, you have 14 chaotic systems exchanging information using random number-based MC exchanges to connect them. The fixed ig you use is just the seed for the random number generator, but there are many different factors that can influence the exact trajectory that each replica will follow.

In the end, as pointed out, what you can compare between MD runs are the final overall results, such as thermodynamic averages, populations, etc., and even this is assuming all your calculations are properly converged before you can gather the data.

So, in the end, yes, you should be expecting differences for your systems. Different runs, even on the same machine (unless using sander or pmemd.cuda, as Jason pointed out) will sample different regions of the phase space over time. If the calculations are converged you are assuming all regions have been properly sampled, but there’s no control over the order of the sampling. In a short amount of time, the regions sampled over the same time may be different for each calculation, and the calculations may be far from converged.

Gustavo Seabra



> Em 25/09/2015, à(s) 13:11, Thiago Lipinski Paes <thiagopaes.gmail.com> escreveu:
>
> Dear Amberists,
> I'm trying to analyze the behavior of a T-REMD simulation both on Cluster
> and Desktop, using MPI.
> I'm using the same .mdins (with a fixed ig=77777) to see if I can reproduce
> my Desktop results, but faster.
> I'm using 14 replicas (ng=14), with ng=70 (Cluster) and ng=28 (Desktop).
> The results are not the same.
> Can I say this is an effect of MC-Exchanges? Should I be expecting this
> kind of difference?
>
> Thanks in advance.
>
> thiago
>
> --
>
> Thiago Lipinski Paes
> Computer Engineer - Computer Science MSc - Computer Science PhD Candidate
> Laboratório de Bioinformática, Modelagem e Simulação de Biossistemas (LABIO)
> Pontifícia Universidade Católica do Rio Grande do Sul (PUCRS)
> Av. Ipiranga, 6681 - Partenon - Porto Alegre/RS - CEP: 90619-900
> Tel: +55 51 33203611 ext.8602
> _______________________________________________
> 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 Fri Sep 25 2015 - 11:30:03 PDT
Custom Search