Re: ABF question

From: Jérôme Hénin (
Date: Wed Jun 23 2010 - 05:17:56 CDT

Dear Parisa,

On 22 June 2010 18:43, Parisa Akhski <> wrote:
> Dear Jerome and other ABF method users,
> I have a couple of questions using ABF method for an ion transport process:
> (1) I have generated a PMF profile and I want to know whether the results
> are reliable or not? One criteria I think is the smoothness of the sampling.
> Also, according to the tutorial, history file is useful for assessing the
> convergence of the free energy profile. I am assuming that I need to plot
> the free energy versus time step for all individual bins to see whether they
> all converge or not. Is it exactly what it means? Is the force convergence
> another necessity for reliable results?

Convergence of the force and the PMF are more or less equivalent.
Since the force is noisier than the PMF, it is easier to monitor the
PMF convergence. On the other hand, the force is easier to work with
for statistical analysis.

> (2) I am a little confused about the sampling in constrained and
> unconstrained MD simulations. Since in constrained MD we freeze the reaction
> coordinate at discrete values of reaction coordinates (RC) and we have to
> specify the exact positions for RC and the number of bins and also the total
> number of "run" in configuration file (for example 5000000), we are
> determining the exact number of samples in each bin (equal number of samples
> in all bins) . And because we determine the time step as well, we have the
> control on the time that the particle spends on each bin. In unconstrained
> simulation, however, we have no control on the particle positions and the
> ion moves freely and the sampling is random.

Indeed. You could call it "diffusive sampling".

> If these are true, can we
> conclude that in constrained MD the ion starts from the first bin and after
> spending the defined time, it moves to the next bin and...(the ion movement
> happens in just one direction  starting form one end and ending at the other
> end)

That is one possible schedule, but you could setup constrained
simulations differently. A common practice is to run each point
separately, for trivial parallelization.

> whereas, in unconstrained, it moves forth and back (without following
> just one direction)?

That is correct.

> Finally, can we say that the ns time scale in
> unconstrained MD (ABF) is just for the whole process not in each individual
> bin?


> I apologize for these simple questions but I am really confused!

No problem!


This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:54:16 CST