Re: [AMBER] duplicate simulations problem

From: David A Case <>
Date: Wed, 24 Nov 2021 15:34:23 -0500

On Thu, Nov 25, 2021, Sadaf Rani wrote:

>I am running these simulations on
>GPU. I am using the same combination as you mentioned .e.g. ntt=3 and
>ig=-1. Does it mean that although the simulation has varying results the
>convergence should be similar? Could you please give your comments on
>it how should I proceed with the duplicate run of simulation? Either I need
>to give the same seed number?

If you want to check that two simulations give identical results, then, yes,
you need explicitly set a seed (via ig) and keep it the same for the
multiple runs. In fact, this is what we do in "stress tests" of GPU
hardware, to make sure that the expected identical results are indeed found.

To run replicated simulations to test convergence, stick with the default ig
value of -1. Then the differences between (among) multiple runs is one
measure of reproducibility, akin to re-running an experiment to look for
noise in the outputs.

I don't know anything about your system, but 20 ns of production is fairly
short by modern standards. You should think about switching to barostat=2,
which will be both faster and more in accord with a true canonical ensemble.


AMBER mailing list
Received on Wed Nov 24 2021 - 13:00:02 PST
Custom Search