Re: [AMBER] power failure + resume

From: Jason Swails <>
Date: Sun, 8 Nov 2015 11:25:31 -0500

On Sun, Nov 8, 2015 at 6:40 AM, shahab shariati <>

> Dear Jason,
> In future MD simulations, I will consider all points you mentioned.
> Now, in my case, I want to resume sander after power failure.
> > so your restart file is unlikely to correspond to the output at time step
> > 2445000​. It could belong to any step between 2445000 and 2447500.
> The step 2445000 is corresponded to 978. For more assurance, I want to get
> a rst file in frame 975 (before step 978). Then resume sander using this
> new rst file.
> How to get a rst file from a mdcrd at the special frame?

​As Bill mentioned, the trajectory does not have velocities, so you cannot
do that. However, you are putting way too much emphasis on getting a
"true" restart. It doesn't matter how careful you are, it is impossible to
get exactly the same trajectory as you would have gotten had the power not
failed in your case.​ The precision issue is one of them (you can avoid
this with ntxo=2, which writes a full-precision restart in NetCDF format).
Another is that Langevin dynamics depends on a random number stream, and
there is no way of "restarting" the random number generator to pick up
where the last simulation left off.

The good news is that this does not matter *at all*. You are using
Langevin dynamics, so the trajectory is stochastic and destroys temporal
information, anyway. You will not be able to tell any difference between
1000 frames taken from a single Langevin dynamics trajectory run for 1 ns
or 1000 frames taken by 2 legs of an interrupted Langevin that was run out
to just over 1 ns.

On top of that, you have probably already spent more time trying to figure
out how to save the part of the simulation you ran before the power failure
than it would take to just re-run the 1 ns of dynamics...


Jason M. Swails
Rutgers University
Postdoctoral Researcher
AMBER mailing list
Received on Sun Nov 08 2015 - 08:30:02 PST
Custom Search