Re: ABF question (sampling and colvars output file)

From: Jérôme Hénin (jhenin_at_ifr88.cnrs-mrs.fr)
Date: Mon Feb 07 2011 - 04:20:28 CST

Hi Parisa,

There is unfortunately not much I can say that hasn't been said before. The
most likely explanation is that some slow degrees of freedom that couple to
your RC are drifting (or have drifted at some point in the simulation). In
other words, maybe the RC is "not good enough" on this timescale. Maybe it
is not "local" enough: ideally the reference group used to measure the ion's
position should not move much with respect to the channel's walls. If the
channel has some flexibility, this can be a problem. Maybe you could use
residues that are closer to the pore, or even pore-lining, as reference
group.

One robust, but demanding criterion for convergence is that every region of
the RC range has been visited several times, and that the PMF has not
changed.

So far there is no PMF history file option, but I will think of
reintroducing it. To many users it is not necessary because long simulations
are often split into short ones to facilitate job management on clusters and
supercomputers. Also, the history can be reconstructed if the system forces
are written to the colvars trajectory file - I understand this is not the
most convenient way.

Cheers,
Jerome

On 6 February 2011 22:20, Parisa Akhski <Parisa.Akhshi_at_chem.queensu.ca>wrote:

>
> Dear NAMD experts,
>
> I am performing ABF calculations to capture the ion movement along a
> channel.
> I have improved the sampling in some of the windows by dividing the whole
> RC to several windows, however, there is still a big difference in the
> number of samples at two ends of the window in some cases even after a long
> simulation time (20 ns). I am wondering what may cause this and is there any
> other solution to improve and smoothen sampling? I know increasing the
> current number of windows may help, but any thing else I can try? The
> position of the ion is already within the each window.
>
> I have always seen people discuss this issue in the mailing list. I am
> wondering if there is any criteria to estimate the quality of the sampling?
> How do I know my sampling is good? If there is a huge difference in the
> number of samples in different areas (between 10k to 150k) or even more in
> one window but the force has converged (the pmf profile is not changing
> anymore), are the results acceptable?
>
> Also, in NAMD 2.6 the history file of the system is reported and there is
> always access to the force and free energy (PMF profile) of the system at
> different time steps for different bins, however, in NAMD 2.7 (colvars),
> only the final information of force and free energy is reported in .pmf file
> unless you just copy your data manually during the simulation. I am
> wondering if there is any way to have these data at different time steps.
> This is really necessary for my system.
>
> Many thanks,
> Parisa
>
>

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 05:23:33 CST