[gmx-users] Re: gmx-users digest, Vol 1 #653 - 10 msgs

[Erik Lindahl]

> This is a particularly egregious problem with PME.  We published on 
> this in J. Comp. Chem. a couple of years ago.  It is the "flying ice 
> cube" effect--as the translational motion builds up the heat bath 
> keeps v squared the same so the temperature goes down, and if you let 
> it go you have an ice cube hurtling through space very fast.
>
> As Erik says, the solution is to keep resetting the center-of-mass 
> motion to zero.

I haven't tested running PME without resetting the c-o-m motion a lot, 
but this is probably the only area where problems are worse with PME 
than with cutoffs. PME calculates the forces by numerical 
differentiation of the interpolated potential, so the residual noise 
will be slightly higher than with the pairwise forces calculated for 
cutoffs. (action and reaction forces are not guaranteed to be exactly 
equal). nstcomm=1 should solve it in any case!

Cheers,

Erik