[gmx-users] is the tip4p MW velocity treated differently in gromacs 3 vs 4

[chris.neale at utoronto.ca]

Thanks Berk. I am only interested in these velocities because I am  
trying to figure out why I am getting some differing results recently  
(which I will post later).

I don't really care about the velocities output in the .gro this way,  
my interest was only based on the fact that I thought this might give  
me a clue as to what was going on. Since you point out that these  
massless velocities do not enter the dynamics then I guess something  
different is going on.

One clean and easy solution would be to simply never output velocities  
for a zero step mdrun .gro file in cases where they can not be  
determined accurately, and just put a note to that effect in the .log  
file.

Thanks for looking into this,
Chris.

-- original message --

Hi,

First let me state again that these velocities are irrelevant for the  
dynamics,
because the velocities only enter the equation through momentum and  
kinetic energy
which have a factor mass, which is zero for vsites.

Looking at the code there are two issues.

One is with continuation (unconstrained start in 3.3).
In 3.3 unconstrained start would only remove the constraining of the initial
coordinates, but not the construction of vsites, which makes their  
velocity incorrect.
In 4.0 this is fully correct (but you need to start with correct x and  
v for the vsites).

Without continuation the velocities of vsites at step 0 and incorrect  
in both 3.3 and 4.0.
In 3.3 and 4.0 the velocities of vsites are determined from the  
differences in positions,
but the positions at step -1 for constraining the initial coordinates  
are not stored.
A clean solution (and more accurate) would be to determine the vsite  
velocities
directly with the vsite weights from the velocities of the masses, but  
that would
increase the computational cost.

A much simpler solution would be to simply always set all vsite  
velocities to zero.

I'll see if a can come up with a simple and clean solution.

Berk