Re: fullSamples in ABF: applied at restart?

From: Jérôme Hénin (jhenin_at_ifr88.cnrs-mrs.fr)
Date: Mon Nov 21 2011 - 02:10:10 CST

Dear Neela,

fullSamples applies to the number of samples in each bin. That number
is saved and read from restart files, so when restarting from a
simulation with sufficient sampling, the bias is applied immediately.

Your second question is about choosing the value of fullSamples. The
bad news is that it is difficult to make an informed choice; the good
news is that the outcome is not very sensitive to this parameter, so
there is a range of safe values. The role of fullSamples is wait until
some of the noise in the measured force is averaged out before
applying the bias. In many cases, the system force contains a large
amplitude, high-frequency noise due to bond vibrations. Because of the
high frequency, this noise component is much reduced with values of
fullSample of 100 or 200 steps, i.e. a few periods of typical bond
vibrations.

The only reason I see to use larger values is to protect a "fragile"
system from distortion due to non-equilibrium forces. By fragile, I
mean a system that is likely to be pushed out of the most relevant
configurations by moderate non-equilibrium forces applied in the
initial regime of ABF. In such a case, larger values of fullSamples
can be used to deliberately slow down the process and keep the initial
regime even closer to equilibrium. The cost of this, of course, is
less effective exploration of configuration space.

Best,
Jerome

On 21 November 2011 05:19, Neelanjana Sengupta <senguptan_at_gmail.com> wrote:
>>
> In addition, I would like to know people's experience regarding accuracy of
> PMF profiles (if any) by increasing/decreasing the fullSamples steps. What
> if one uses a default of 100? (I understand it is situation dependent, but
> would still like to pose this question).
>
> Thanks,
> Neela
>
> On Mon, Nov 21, 2011 at 9:44 AM, Neelanjana Sengupta <senguptan_at_gmail.com>
> wrote:
>>
>> Dear NAMD experts,
>>
>> I need a clarification about an Adaptive Biasing Force (ABF) parameter in
>> colvars.
>>
>> Due to large fluctuations of the biasing force, the fullSamples option
>> (default 200 steps) is used to obtain accurate estimate of the reaction
>> coordinate before applying the bias.
>> However, is this applied only the very first time, or at each restart
>> step? (ie. if someone uses the .restart.colvar files, etc to continue
>> another calculation, will fullSamples be applied again)?
>>
>> Thanks and regards,
>> Neela
>
>
>
> --
> ~~~~~~~~~~~~~~~~~~~~
> Neelanjana Sengupta, PhD
> Physical Chemistry Division
> National Chemical Laboratory
> Dr. Homi Bhaba Road
> Pune 411008, India
> Phone: +91-20-2590 2087
> Fax: +91-20-2590 2636
> ~~~~~~~~~~~~~~~~~~~~
>

This archive was generated by hypermail 2.1.6 : Mon Dec 31 2012 - 23:20:59 CST