Re: Convergence of Sampling and Deciding on Simulation Length

From: Aron Broom (
Date: Tue Jul 16 2013 - 09:59:15 CDT

In the simplest case you could look at potential energy. For instance,
break your simulation into halves and ensure the mean and standard
deviation of the potential energy are equivalent within some tolerance.
You could be more precise by ensuring a the histograms matched also, so
that you were not only comparing the mean and standard deviation, but also
the distribution. If the halves aren't within tolerance, get rid of some
of the start of your simulation and repeat the analysis. Continue with
this pattern until you reach some minimum simulation length, at which
point, if it hasn't converged, you need to extend the simulation.

You can expand the above idea to any other variable also, such as separate
components of the potential energy, or some of the other variables you've


On Tue, Jul 16, 2013 at 7:08 AM, Aditya Ranganathan <>wrote:

> Hello All,
> With the ability to run simulations over long timescales and all the
> analysis with all the Essential Dynamics and Normal Mode Analysis tools
> that we have available to us now, I have a fundamental question to ask. How
> do we decide whether our simulation has run long enough for the PCA and EMD
> analysis to be of any relevance? For example, cosine content of a
> trajectory (Physical Rev. E. Vol 65, 031910) that can be calculated in
> Gromacs seems to be reported as one of the measures of understanding the
> contribution of diffusion to the protein motion during the simulation. Is
> there a way to calculate cosine content in VMD/NAMD? Also, are there other
> parameters which one should look at while trying to decide if the
> simulation has run long enough to give meaningful results and not noise.
> How do we check for convergence?
> Thanks and Regards
> Srivastav Ranganathan
> Research Scholar
> Department of Biosciences and Bioengineering,
> IIT Bombay
> Mumbai,
> India

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 - 23:21:26 CST