Re: abf integrate on two dimensional pmf

From: Ajasja Ljubetič (ajasja.ljubetic_at_gmail.com)
Date: Tue Oct 08 2013 - 03:20:52 CDT

Dear Nadeem,

How large are the differences? abf_integrate uses a Monte Carlo simulation
to estimate the integral of the gradient. As this process is
nondeterministic, it would explain small differences in obtained PMFs. You
could try increasing the number of steps and also to tweak other
options<http://www.ks.uiuc.edu/Research/namd/2.9/ug/node56.html#SECTION000133500000000000000>of
abf_integrate.

I would also look at the sampling histograms (data in *.count). Is the
logarithm of the number of samples (aka the residual free energy)
approximately flat?

Best regards,
Ajasja

On 8 October 2013 04:25, Nadeem A. Vellore <vnadeem_at_gmail.com> wrote:

>
> Dear NAMD users,
>
> I am trying to use the ABF with two variables (two-dimensional) for my
> system and I have some trouble in collating the data.
>
> 1. I ran the two variable combinations using discrete windows. (var-1 and
> 2 divided into 5 windows : 25 windows in total)
> 2. Using the normal combining method I created the *.grad and *.count file
> (with inputPrefix command in abf.in file and integrated by running for 0
> steps)
> 3. the abf_integrate code was then used for creating the PMF file using
> the default values
> ./abf_integrate comb_5ns.grad
> ./abf_integrate comb_6ns.grad
> This produced the *pmf, *histo, *est and *dev file.
>
> Everytime I re-run this on the same grad, I get a different PMF file. I am
> doing something wrong.
> Also, looking at the time-based PMF evolution, I see disconnection between
> the first and second timescale data (meaning 5th and 6th). This is
> meaningless at the moment, as I am not able to reproduce the PMF profile
> for a given traj.
>
> Can you please shed some light on this.
>
> Thanks
> Nadeem
>
>
>

This archive was generated by hypermail 2.1.6 : Wed Dec 31 2014 - 23:21:45 CST