[gmx-users] Reducing the dimensionality of my cross correlation matrix

[Bailey A.]

Dear gromacs users,

I posted a similar question the other day, but it was probably too long, so here it is more concisely:

I've taken the ascii output of g_covar and built the 3N x 3N covariance matrix in MATLAB from the 3N^2 x 3 matrix, which I've then normalized to get a 3N x 3N cross correlation matrix.

What I'd like is a N x N matrix, and so I've added together the three N x N parts of the 3N x 3N matrix and divided by 3 to get my N x N matrix.

However this doesn't look the same as something I'd previously calculated using Ran Friedman's amended g_covar script and I'm trying to understand the discrepancy.

Can anyone advise me on what the dimension reduction step should be? I can post my code if that helps.

Many thanks,

Alistair