[gmx-users] nstlist with vdw only

[Erik Lindahl]

Hi,

On Aug 21, 2007, at 8:08 AM, Nicolas Schmidt wrote:

> Would't this mean, that molecules could suddenly appear and lead to  
> a discontinuity in the LJ-potential?
> Or in what way should I modify the r_vdw according to what rlist? I  
> wanna end with a cut-off of the LJ-potentail around 1.75nm.
>

Certainly - but there will be a discountinuity even with nstlist=1  
unless you use switch or shift modifications to the interaction.

For perfect energy conservation you should pick rlist slightly larger  
than rvdw (depends on your charge group size and nstlist, but start  
with 0.2-0.3 nm), and apply switch or shift.

However, there will be a bit of performance price to pay for that.

Personally, I think energy conservation is overrated. There's nothing  
wrong with it, but energy conservation per se doesn't mean anything.  
You could use the potential V=sin(r) for Lennard-Jones interaction  
and still get conservation if it is switched smoothly to zero at the  
cutoff distance.

The goal in most applications is to improve the accuracy of free  
energy, diffusion coefficients, or other observables, and then it is  
sometimes better to focus on the sampling and let thermostats take  
care of the drifts in energy.


Cheers,

Erik