[gmx-users] Advice needed on choosing a VdW cutoff

[Mark Abraham]

Steven Kirk wrote:
> Hello,
> 
> I would be very grateful for advice on the following system:
> 
> Consider a pair of spherical macromolecules of diameter ~2.5 nm, 
> arranged along an x-axis so that their centres of mass are 4nm apart, 
> then centered in a 14 x 10 x 10 nm periodic box, so that the minimum 
> distance between either macromolecule's centre of mass and its closest 
> simulation cell walls in the x,y, or z directions (standard orthogonal 
> Cartesian axis set) is 5 nm.

There was another brief thread on a pair of silica particles on this 
list some time last year with the author reporting that he didn't see 
attraction like he expected. I replied at the time pointing out that the 
construction of his system was such that the "other" particle in the box 
was as close as its periodic images so that there would be no net 
attraction, but you don't have that problem!

> The problem is that when I run this simulation, the expected drift of 
> the macromolecules towards each other does not occur. 

A time estimate would be useful here - does experiment have anything to 
say about it?

> Assuming that I 
> want to force every part of each macromolecule to 'see' every part of 
> the other, this would suggest a value of rvdw of > 6.5 nm, but I have 
> several worries about this:
> 
> 1. I have never seen a recommended rvdw in this forum over 1.4 nm, in 
> any model system

Usually it's a good idea to keep parameters like this near the values 
they held when the force field itself was parameterized, lest there be 
hidden dependencies. Also it's rare that you can afford to go much above 
  values like 1.4nm, and historically people used much smaller ones 
because that's what could be afforded then.

> 2. Should I use a standard, switched or other type of vdW cutoff?
> 3. Should I switch on long-range dispersion corrections (DispCorr = Ener) ?
> 
> My goal is to keep the periodic box for the advantages of PME, but 
> somehow reassure myself that the macromolecules can 'see' each other via 
> the vdW forces, so that they will drift together (the expected 
> behaviour) over the course of the simulation.
> 
> I have trawled the mailing lists for advice on this topic - the only 
> directly relevant post I could find involved the drifting apart of 
> membranes over the course of a simulation, and if I remember correctly, 
> the value of 'pme-order' was suggested as a culprit.

A version prior (3.2.1?) had a bug such that non-default pme-order was 
buggy unless the relevant source code was replaced. Use 3.3 and forget 
about that.

> I use the default value of 'pme-order' in my simulations.
> 
> Can anyone please advise me as to what to do next? I do not want to 
> abandon the investigation before I eliminate all possibilities that I am 
> doing something stupid with the configuration of the vdW treatment. 
> Maybe I do not have a long enough simulation, but the fact that I 
> actually see *repulsion* suggests something more fundamental is wrong.

Can you try a smaller system with an initial separation within a 
manageable vdW range to prove the concept? You could also try an 
implicit solvation model in some other MD suite, which greatly 
facilitates increasing rvdw.

Mark