Re: Non-equilibrium effects in ABF runs

From: Aron Broom (
Date: Tue Aug 13 2013 - 15:59:16 CDT

In terms of the way the ABF method works, the second of your descriptions
is correct, that is, once 1000 samples are reached, the bias will always be
applied when it re-enters bin A. Note of course that the bias itself will
still be updated by additional sampling when it re-enters.

In terms of equilibration, the key thing to note with ABF is that the end
goal of the bias it to make all values of your reaction coordinate equally
likely. Equilibration then, can be seen when your system no longer appears
to prefer any region of your reaction coordinate over another. Of course
it may not move smoothly along, due to unbiased degrees of freedom, and so
as you alluded to, you might need to calculate the autocorrelation of the
reaction coordinate exploration in order to get an estimate of how large a
window in time you will want to look at in order to judge the equilibration.


On Tue, Aug 13, 2013 at 12:38 PM, Matthew Ralph Adendorff <>wrote:

> Good day,
> I have a question regarding the fullSamples parameter for Adaptive Biasing
> Force runs. I am looking to run ABF on some DNA duplexes in explicit
> solvent and so am concerned about non-equilibrium effects in the runs. When
> the system enters a bin, the fullSamples parameter sets how many counts
> must be reached before the bias is applied to that bin. Let's say we have
> set fullSamples to 1000 (very small for this type of system I know). If the
> system were to move out of Bin A, after 1000 counts, into an adjacent Bin B
> under the influence of the bias and then move back into Bin A (perhaps
> through natural diffusion of the reaction coordinate), would another 1000
> counts need to be achieved before more bias is applied? If this is the
> case, perhaps a blocking method could be used to determine the best value
> of fullSamples to ensure that the system is in fact equilibrated before
> more bias is added. If not, is there an effective way, perhaps through
> checking autocorrelation functions ove!
> r time, to ensure that non-equilibrium effects have been accounted for in
> sampling along the reaction coordinate? I understand how to check when a
> trajectory has equilibrated, however, I am unsure as to whether a biased
> trajectory can be treated in a similar fashion as forces are continually
> being added to the simulation. Any advice/referral to extra reading would
> be greatly appreciated.
> Best regards,
> Matthew

Aron Broom M.Sc
PhD Student
Department of Chemistry
University of Waterloo

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