Re: [AMBER] FW: FW: How to use "divergence" command

From: Daniel Roe <>
Date: Wed, 9 Dec 2015 09:51:15 -0700

On Wed, Dec 9, 2015 at 6:28 AM, Jason Swails <> wrote:
> kldiv is not a command. It is an option for the kde command, and it
> allows you to calculate the K-L divergence for a *continuous* distribution
> function constructed from Gaussian kernels centered at each datapoint.
> Note that in this case, the *second* formula in the Wikipedia article
> applies, as this is now a continuous distribution rather than a discrete
> one.

Just to expand on this a bit, 'kldiv' allows you to calculate the
time-dependent K-L divergence of two distributions, i.e. it's a way to
track how well those distributions overlap over time. As two
independent simulations converge, the overlap between probability
distributions for various properties will overlap more and more and
thus the K-L divergence will become smaller. Since K-L divergence is
undefined if one distribution has population in a bin and the other
does not, using a Gaussian kernel density estimator is a particularly
nice way to do this since it essentially reduces the chance that both
bins will have absolutely no population. Further discussion and
examples of usage can be found in these publications:

and the supporting information therein. Hope this helps,


> HTH,
> Jason
> --
> Jason M. Swails
> BioMaPS,
> Rutgers University
> Postdoctoral Researcher
> _______________________________________________
> AMBER mailing list

Daniel R. Roe, PhD
Department of Medicinal Chemistry
University of Utah
30 South 2000 East, Room 307
Salt Lake City, UT 84112-5820
(801) 587-9652
(801) 585-6208 (Fax)
AMBER mailing list
Received on Wed Dec 09 2015 - 09:00:03 PST
Custom Search