Re: Umbrella Sampling after multiple replicas metadynamics

From: Giacomo Fiorin (giacomo.fiorin_at_gmail.com)
Date: Fri Sep 28 2012 - 16:33:54 CDT

Hi Aron, no, the files are how they are supposed to be: each replica stores
its own stuff, and all energies and gradients are summed together at
runtime. For that you must continue to run with multiple replicas, if you
continue a single one by decoupling it from the others it's just not the
same simulation any more.

So yes, you can make a single-replica state file to continue from a certain
point, by summing together the data in those files just like you suggested.

Giacomo

On Fri, Sep 28, 2012 at 5:23 PM, Aron Broom <broomsday_at_gmail.com> wrote:

> I'm not sure anyone has tried this, but I suspect someone who works on the
> collective variables modules might have some input.
>
> Something I've done before which was very successful, was to use
> metadynamics to get a rough estimate of the Free energy surface, and then,
> restart the simulation, but with the hill height changed to 0 and a
> harmonic bias in order to do umbrella sampling. The value of this is that
> the umbrella sampling with WHAM can act to smooth out any lingering errors
> in the metadynamics and also provide a reliable estimate of the error.
>
> The above works great for a single metadynamics run. I've recently done
> some multiple replicas metadynamics in NAMD 2.9, and I would like to repeat
> this kind of process. I thought that because the replicas communicate, the
> final *.state file for the last replica to finish would essentially have
> all the data, and I could just use that to restart. It seems that isn't
> the case, and each replica's state file only holds the energies in it's
> grids for the biases it specifically applies? This seems at odds with how
> multiple replicas simulations run, but when I restart from one of these
> state files, I don't get back to what the final *.pmf looked like, and
> appear to only be getting a small portion.
>
> Is there a way to merge all these state files together? For instance,
> could I simply take the columns of the state file that correspond to the
> grid energies and just add all those values together, and hope that the
> resulting state file is what I'm looking for? And would one do that for
> both the energies and the gradients?
>
> I suppose this also applies to any time one might want to change the
> number of replicas being used after a restart?
>
> Or maybe I've just done something horribly wrong, and the final *.state
> file from any run really is the combination off all the hills for all the
> replicas?
>
> Thanks,
>
> ~Aron
>
> --
> Aron Broom M.Sc
> PhD Student
> Department of Chemistry
> University of Waterloo
>
>

This archive was generated by hypermail 2.1.6 : Mon Dec 31 2012 - 23:22:07 CST