[gmx-users] Simulation in parts

[Mark Abraham]

> Hello,
>      I was wondering if I could run a simulation in parts.

Absolutely... running calculations on large computing facilities requires
the ability to interact with a queueing system that will give you only
finite lengths of time, so "cut and paste" is a necessary attribute.

>  I would like
> to
> run a simulation for 100ps say, analyze it, and then sometime later
> continue
> on for another 100ps.  It seems like all I would need to do is take the
> *.gro, *.trr, and *.edr files output from the first md simulation, feed it
> back in to grompp and use mdrun.  I have tried this, and the output from
> the
> run seems okay, but I am not completely sure.  The output of grompp says;
> Velocities generated: ignoring velocities from input trajectory.  Will
> this
> cause a disjoint between the two simulations?

Yes - because your .mdp file is the same one you used when you started the
simulation, and when you did that you needed to generate velocities. There
are situations where you might want to regenerate velocities even when the
trajectory file has them, so grompp lets you do this. So set "gen_vel" to
"no".

A way to do a restart that avoids this problem is to use tpbconv -extend,
which means you don't supply an .mdp file.

>  Is there a way that I can
> have
> the velocities from the end of the first md simulation, input into the
> second?  Also, I used -time 100 and grompp says;  Reading frame    500
> time
> 100.000   Using frame at t = 100 ps
> Starting time for run is 0 ps
> Is there a way that I can have the starting time for the second run be
> 100ps
> so that the numbers match up in the *.edr files and the second run goes
> from
> 100ps to 200ps?

Yes, there's a run option that lets you specify the start time. It does
sound like spending an hour reading section 7.3 would be well spent for
you. :-)

The tpbconv approach is also superior here, because it will naturally
extend the simulation in time without you having to edit an .mdp file to
specify it, and possibly make mistakes, etc.

Mark