[gmx-users] autocorrelation functions from several simulations d

giordano mancini baloilgiullare at interfree.it
Wed May 20 15:38:45 CEST 2009

Hi all,

I have some doubts regarding the general use of acfs.
The molecule I am studying is formed by two polycyclic aromatic
sub-units joined by a 20-alkene "chain"; it is something like this:

   /--\/--\                       /--\/--\
A |  |   |-CH2-CH-CH2-CH-CH2(...)|  |   | B
   \--/\--/                       \--/\--/

however the subunits are different (compared to the drawing above and
one from the other) and contain heteroatoms. One of the subunits exist
either in a reduced or oxydated state in organic solution (I have used
DMSO and acetonitryle). In the reduced state the molecule forms hydrogen
bonds between the subunits and can fold with subunit A bonded to B,
while in the oxydated state these bonds do not exist and the subunits
are free to move. Of course, to fold up starting from an "open"
configuration the subunits must get close.
I have several simulations of this system in the oxydated state.
Starting from folded initial positions it eventually opens in all
simulations. To follow the unfold process I have extracted the distance
A-B (which I call Rab) from all simulations. When the H bonds exists Rab 
has small oscillations around a certain value while when it opens it can 
have any value up to the length of the molecule.
It could make sense to calculate the acf of Rab (with g_analyze) from 
each simulation
and use the autocorrelation time as an estimate of the unfold time?
I think that, upon oxydation, the system becomes disordered  (relative 
to Rab) and loose memory of past Rab values; the autocorrelation time 
should be shorter than that obtained for the reduced state.
Since I have 10 simulations what is the correct way to average results?
Calculate a mean autocorrelation time?
Doing this yields an average t0 of 155 ps. If I define an average "open"
Rab and "closed" Rab and the simply average the time to unfold i obtain
242+/-300 ps.
Finally, should I use, instead of Rab the difference Rab-Rab0 where the 
Rab0 is the mean or median value (for each simulation)?


More information about the gromacs.org_gmx-users mailing list