[gmx-developers] adding angle dependent non-bonded interactions?

[Erik Lindahl]

Hi Julius,

>
> (1) non-bonded force computation.
> (2) bond list: for each atom, access to the atoms bonded to it.
>
> I assume the bond list doesn't change over time. Also, where are the
> parameters for the GROMACS force field stored?  Can one add new atom
> types? New interaction types?  Additional parameters? With this
> information I could implement the potential in an easily 
> user-modifiable
> way.
>

First thing to think about is how to represent the interactions, since 
that affects whether we are going to treat them as nonbonded (dynamic 
neighborlists) or bonded (fixed lists of interactions).

The problem is that angles involve three atoms, so the number of 
combinations will grow very rapidly. Even with a short cutoff each atom 
might have ~200 neighbors in range, meaning there are 20000 angle 
combinations for each atom in the system. With 10,000 atoms that would 
mean 200,000,000 entries in the neighborlist. 12 bytes per entry (3 
atoms*sizeof(int)) means you would need 2.5Gb of memory just for the 
neighborlist... and that's a small system with a short cutoff.

One obvious way to get around that is to only calculate the angle 
potential for a small number of triplets, but in the case it might be 
much more efficient to include them statically as a bonded interaction 
instead of creating huge 3d neighborlists.

Cheers,

Erik