[gmx-users] g_rmsf over time windows

Tsjerk Wassenaar tsjerkw at gmail.com
Wed Oct 8 13:37:37 CEST 2008

Hi Giordano,

> I agree that 10 ns
> are a short sampling but the total simulation time is 15 ns + 2 ns I
> used to reach 298K;

Whether that's enough is largely dependent on the size of your protein
You should check whether you went beyond relaxation by inspecting the
cosine content of the first few eigenvectors.

> since the rmsd was almost flat in this 10 ns I was
> hoping that results were not influenced by relaxation.

This by itself does not mean anything. If it levels off at 0.1 nm,
okay, that's pretty good, but the further you go, the more
possibilities you have for getting a certain RMSD. It's a distance
measure. Being at 10 cm distance from you will make it easy to
localize me, but if I'm at 10 miles, you wouldn't now where to search.
I could well go 10 miles and then start following a circle, and you'd
argue, based on the distance, that I'd stopped moving!

> Maybe g_rmsf calculates fluctuations from the average structure in a certain
> time window and not from the reference or starting frame
> (and it is my fault not to have read the program help well). In this case my
> results would make sense.

That is indeed the case. And if your system had converged, there
wouldn't be a difference between the average structures from
subtrajectories A and B. The RMSF is only defined about the mean
structure. You don't want to go into non-central moments. If you have
the average structure for the whole trajectory, you can use it as the
reference to calculate the RMSF. I believe that g_rmsf has an option
to use deviations from the reference rather than from the calculated
average structure.

Hope it helps,


Tsjerk A. Wassenaar, Ph.D.
Junior UD (post-doc)
Biomolecular NMR, Bijvoet Center
Utrecht University
Padualaan 8
3584 CH Utrecht
The Netherlands
P: +31-30-2539931
F: +31-30-2537623

More information about the gromacs.org_gmx-users mailing list