[gmx-users] COsine content of ED analysis

Berk Hess gmx3 at hotmail.com
Tue Sep 7 09:48:40 CEST 2004

>From: Chng Choon Peng <cpchng at bii.a-star.edu.sg>
>Reply-To: Discussion list for GROMACS users <gmx-users at gromacs.org>
>To: Discussion list for GROMACS users <gmx-users at gromacs.org>
>Subject: Re: [gmx-users] COsine content of ED analysis
>Date: Tue, 07 Sep 2004 08:31:12 +0800
>Hi Ram,
>   When you mean you've done an ED analysis, I take it as you've gotten the
>eigenvalues and eigenvectors from a covariance matrix (ie after running
>Then, you can use g_anaeig to get the principal components and then
>"g_analyze -f proj.xvg -cc" to compute the cosine content.
>A question for all:
>Actually, I'm simulating a very small protein to 5-10ns and the cosine
>content is still high when I use the average structure in the covariance
>matrix rather than the initial structure.
>So, I wonder if we really need long runs even for small proteins (< 100
>residues) in order for the protein to end up in a potential well rather 

Even for peptides you need microseconds or more.

The problem is that there is not one potential well, but there are many.
On a timescale of 5-10ns you might sample only one well.
But on longer timescales you will see jumping between many of these
wells. See for instance:
Hess, B. Convergence of sampling in protein simulations. Phys. Rev. E 
65:031910, 2002.


Play online games with your friends with MSN Messenger 

More information about the gromacs.org_gmx-users mailing list