Re: Umbrella sampling-reliability of results

From: Aron Broom (broomsday_at_gmail.com)
Date: Fri May 04 2012 - 09:41:35 CDT

Parisa,

For US, one main thing you need to do is to make sure your samples are
independent. So initially you will have chosen to record values to the
.colvars.traj file at some arbitrary interval (say 1000 timesteps). If
your reaction coordinate has a decorrelation time that is longer than this,
then some of your measurements are not independent and you will not be able
to get an accurate estimate of the error through WHAM (the method you
likely are using to assemble all the windows). Basically if you construct
an autocorrelation plot (using xmgrace for instance), you need to see at
what timestep the correlation drops below 1/e (~0.37), if that ends up
being 10 measurements, then you need to trim your data down by keeping only
every tenth measure.

For both US and ABF one of the main problems will be slow degrees of
freedom that are not accounted for. I'm not sure what you did to decide
your ABF was converged, but you should plot the .colvars.traj for your ABF
run (all of it), and have a look at it. Initially it should appear stuck
in one region and then stuck somewhere else and so on, until it has
explored the range of the reaction coordinate you gave it. If it has truly
converged and there are no unaccounted for slow degrees of freedom, then in
the final parts of the simulation it should end up moving smoothly from the
beginning of your reaction coordinate to the end. If it still appears to
stick at certain points and then move on, you have some accounted for slow
degrees of freedom, and will need to be more careful in deciding that it is
converged.

~Aron

On Fri, May 4, 2012 at 10:20 AM, Parisa Akhski <
Parisa.Akhshi_at_chem.queensu.ca> wrote:

> **
>
> Dear NAMD experts,
>
> I have been trying to use Umbrella Sampling (US) method to get the PMF
> profile in a channel. I have already obtained the PMF profile using ABF
> method and know how the profile looks like.
>
> In US, I know that the first thing to check is the overlap between
> adjacent windows. Then, I followed all the instructions given in:
>
>
> http://www.ks.uiuc.edu/Research/namd/mailing_list/namd-l.2011-2012/2338.html
>
> and I got a close PMF profile to ABF. In ABF I could check the force and
> sampling during time from history files to check the convergence. In US
> also I can generate PMF profiles during simulation to compare them during
> time. I am wondering what else I should check here? That is, as long as the
> overlap is enough and the harmonic potential is enough to keep the ions at
> the center of bin and the PMF profiles are not changing anymore during
> time, I should get a good and reliable profile? Is there anything else I
> should check?
>
> Thanks in advance for your help,
>
> Parisa
>

-- 
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 - 23:21:30 CST