Re: How to use a proper force constant to decrease computational cost while keeping calculation precision?

From: Ana Celia Vila Verde (
Date: Wed Apr 20 2016 - 03:03:08 CDT

Hi Wei,

The article by Roux, B., The calculation of the potential of mean force
using computer simulations
/Computer Physics Communications, /*1995*/, 91/, 275-282
is very useful.

In general, running more windows with a slightly larger k is more
efficient than running fewer windows with lower k. Keep in mind that
there is no single value of k which is correct; many values of k,
coupled with different numbers of windows, will give you a good result,
provided that all windows overlap. Your k=35 might work with windows
that are 5, or even 10 degrees apart, for example. If you want to test,
and given that your system does not appear to be very complicated, you
can do a couple of other runs where you take the same k but use smaller
and larger windows, and see how your PMF is affected. If you reach a
point where two runs with different window sizes give the same PMF,
despite different overlap between windows, then you know you have enough

Oh, you should always use WHAM to build your PMF from umbrella sampling...

I hope it helps,


On 20/04/16 09:18, wliu wrote:
> Dear all,
> Recently, I am learning how to use umbrella sampling method to
> calculate the potential mean force of one torsion angle of a specific
> residue in a protein. I am curious about how to choose the force
> constant wisely.
> If k is too large, to fulfill the overlap, we have to employ more
> windows. On the contrary, the center value of torsion we set will be
> displaced largely. So, is there any quantitative criterion to judge if
> the value of k is reasonable (small enough and can guarantee the PMF
> calculation precision)?
> For example, if we set k=35, χ0=110, after a short time's MD (such as
> 6 ns), I got the distribution of χ, then I calculation the mean of χ,
> <χ>=104.12, standard deviation σχ=6.04. Then the displacement Δχ=5.88
> (<σχ). Thus, we consider k=35 to be confident (within 68% of Gaussian).
> Any suggestions or relevant literatures will be appreciated.
> Wei

This archive was generated by hypermail 2.1.6 : Tue Dec 27 2016 - 23:22:02 CST