[gmx-users] microcanonical

Mark Abraham Mark.Abraham at anu.edu.au
Thu May 18 10:15:21 CEST 2006

karamyog singh wrote:
> I exactly did that. I initially thought that since I have placed the 
> atoms randomly so it is probable that atoms experience large forces and 
> since I am not allowing any temperature scaling etc. , I am getting 
> large KE. Therefore I first minimized my configuration. I got a frozen 
> structure by doing pressure and temp. scaling. Then I removed 
> everything. After that what happens is, the whole crystal starts 
> breaking up. First one atom gains velocity and moves then another atom 
> and then another. After tht the whole system just breaks.

Ah... your .mdp file generated velocities... you don't want to carefully 
equilibrate something, and then put near-random numbers in place of the 
velocities. You just want to turn off the regulation and set gen_vel = no.

> Any idea why? Is it because the no. of steps for my initial md run is 
> small . Shall I stabilize my configuration for a longer time?
> The cut off that I have taken is the value of 'r' at which the minima 
> for LJ exists. Is it ok or shall I increase this too?

This is certain to be too small. With correctly modeled LJ, you'd expect 
to see a peak in the radial distribution function corresponding to that 
minimum, since all the atoms will be jostling to get there with respect 
to all other atoms - impossible of course, but they'll do it on average 
for their closest neighbours. Here you won't necessarily see that, 
because your effective potential function for all pairs of atoms drops 
from infinity at zero distance to the distance of the minimum value, and 
then to zero.

Elaborating, as soon as an atom moves to a distance greater than the 
distance of the minimum, they don't interact. The other nearby 
interactions will become more repulsive (if the gas is dense) and 
effectively correct for it, but your effective potential function is not 
LJ any more. Further you say your gas is dilute, so it seems unlikely 
that you have a good physical model... atoms will collide, bounce past 
the minimum distance and have no interactions with anything until they 
collide with another atom. This is not true for an LJ system.

I'd say you want a cut-off such that the potential at that cutoff is 
approximately zero...


More information about the gromacs.org_gmx-users mailing list