# Re: ABF - sequence of MD runs

From: Jérôme Hénin (jhenin_at_ifr88.cnrs-mrs.fr)
Date: Tue Jan 05 2010 - 11:26:22 CST

Dear Krishnan,

First, to produce a combined PMF from separate simulations, the best
practice is to use produce an average of the gradient (.grad files)
from all simulations, weighted by their respective sample counts
(.count files). Then, integrate the averaged gradient numerically to
obtain the statistically-correct PMF.

The presence of the bias in a run should not have a deleterious
influence on the following one (provided that the system was not
driven far from equilibrium). The reason for this is that ABF applies
biases on some degrees of freedom (the reaction coordinates), but
computes averages at fixed values of the reaction coordinates:
therefore computation of these averages only relies on sampling other
(unbiased) degrees of freedom.

Best,
Jerome

2009/12/18 Krishnan, Marimuthu <krishnanm_at_ornl.gov>:
> Dear all,
>
> I have a naive question about ABF simulations using NAMD. I would appreciate if ABF/NAMD experts could help me understand it.
>
> I have performed a sequence of five 5-ns ABF runs. An equilibrated configuration sampled from a 300K-MD simulation was used as the starting structure for the first run.
> For the next four ABF runs, the starting structures and initial velocities were taken from the final structures/velocities obtained from the preceding runs (the PMF
>  calculated from the preceding 5-ns run will not be considered/included (for the averaging purposes) in the present run). At the end, I have a set of five PMF profiles.
> Is it meaningful to consider this set as an ensemble and to compute the average PMF profile out of it? Will the average PMF thus obtained be different (significantly or slightly) from that obtained from a set of 5 independent (each starting from MD-sampled random configurations and random velocities obtained from MB distribution) 5ns ABF runs? I am concerned about it for the following reason: if the starting structure of an ABF run was  taken from the preceding ABF run,  then the "influence" of the average force obtained from the preceding run would be present (implicitly) in the current ABF run. Is my understanding correct? Any expert advice would be greatly appreciated.
>
> cheers,
> Krishnan

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:53:39 CST