Re: Simulated Annealing with many frequent restarts

From: Francesco Pietra (
Date: Fri May 01 2015 - 01:21:28 CDT

1) I was just wondering whether to post a similar question, on a broader
basis. In general, for a classical MD, restarting on a large cluster means
(correct me if I am wrong) restarting with a different distribution of the
system along the various nodes. Should the aim be of investigating how the
system develops, is there a real continuity from one run to another one?

2) A non secondary problem is that using many nodes makes the queue longer,
with certain clusters too long. So that many tens of restarting runs become

3) It has become common to have "accelerators" on the nodes, GPUs of Phi.
I suspect this exacerbates the problems at (1).


francesco pietra

On Fri, May 1, 2015 at 1:51 AM, Phillips, Carolyn L. <>

> I apologize if this has been answered before but searching the user list +
> manual has not answered this for me.
> I want to do a simulated annealing of a system. It would seems this can
> be done by temperature reassignment. However, I also am working with a
> wallclock limit that means I need to restart my simulation with fairly high
> frequency. What I cannot figure out is whether NAMD can correctly return
> to the specified temperature schedule when restarted automatically. For
> example, if the temperature is ramping from 300K to 350K by incrementing by
> 10K every 1000 time steps, and the simulation is restarted after 20001 time
> steps, will the temperature of the system revert to 300K or be at 320K? If
> I use “restartsave yes” does that do anything?
> Cheers,
> Carolyn

This archive was generated by hypermail 2.1.6 : Tue Dec 27 2016 - 23:21:06 CST