[AMBER] Testing of the convergence by means of PCA

From: James Starlight <jmsstarlight.gmail.com>
Date: Fri, 22 Aug 2014 21:00:13 +0400

Dear Amber users!

I need to test the convergence of the set of the trajectories calculated
for the identical system using aMD with difference only in initial
velosities in each case.

For this task I've tried to compare eigenvectors from each trajectory. I
made 5 md 20-50ns calculations for my system changing only initial
velocity (assigned randomly during heating) and prolongation of the md run
in each case (from 20 to 50 ns). Than I've made PCA of each trajectory and
compare top 10 (with biggest variance) eigenvectors by means of the dot
product method with the eigenvectors from the first (called it reference
because it was the longer) trajectory. As the result I 've obtained ~ 0.8
overlap in top 5 modes (with biggest varience) . The only difference was in
the shortest (called it 2nd) trajectory where only fifth (not second as in
the rest cases) eigenvector had bigger overlap with first eigenvector from
reference trajectory ( in the rest cases the identical pairs of the
eigenvectors showed good correlation- in table it seems as the linnear
pattern of correlaiton).
 Does this indicate of good sampling and convergence in cases of all
trajectories (with the exeption of 2nd shortest trajectory)? How it could
be checked besides in terms of eigenvectors analysis?

Thanks for suggestions,

AMBER mailing list
Received on Fri Aug 22 2014 - 10:30:02 PDT
Custom Search