[gmx-users] Solvation box size

[Mark Abraham]

David Mobley wrote:
> Bob,
> 
> Lots of people have different opinions about this. My personal opinion is
> that they should be more like the whole cutoff away from the box edge,
> although the answer probably depends on the properties you're trying to look
> at. One factor is that water properties are somewhat different near
> proteins, etc. (i.e. there are solvation shells which extend out a ways
> before you really get to  bulk water). So if an atom in one box can "feel"
> the altered water properties near the image of the protein in the next box,
> it will give you slightly different answers.

The water properties will be more perturbed still if you have so little 
solvation still that one water molecule feels the long-range influence 
of multiple copies of the solute.

Hunenberger and McCammon have a technique for assessing the extent to 
which periodicity is perturbing the PMF of biomolecular systems - see J. 
Chem. Phys. B. 104:3668 (2000) and related papers. In particular they 
opine that "For this specific system [polyalanine octapeptide with 
charged termini], we would recommend the inclusion of at least three 
solvation layers (about 0.85 nm) between the peptide termini and the 
nearest unit-cell walls. In fact, such a requirement is not 
unrealistically drastic for the simulation of typical biomolecules."

> Again, it depends on what you're looking at. I'm doing free energy
> calculations where I disappear a ligand in a protein binding site (which is
> buried) and it's sufficient for my purposes to use a box size which is only
> about 5A bigger in each direction than the largest protein dimension, even
> though my cutoffs are more like 9A. If there is a "conventional" wisdom on
> this it is probably that you should use slightly larger box sizes than 5A
> extra in each direction, but for what I'm doing, it seems not to make a
> difference. This is already a fairly large system, too, so that probably
> helps. I'm being more careful when it comes to small molecules.
> 
> Sorry I can't be more definitive. And other people may disagree with me.

Yes, it isn't the sort of question that untrained humans' judgement is 
any good at answering. It's a complex many-body through-space 
interaction whose macroscopic effects may only be seen over long 
simulation times (much longer than typical non-REMD simulations). You 
need to look at your system, at *recent* literature, your computer 
hardware limitations and come up with something defensible.

Mark