[gmx-users] Fwd: Question on Gromacs: rlist, rvdw, rcoulomb

[Mark Abraham]

> I am a user of gromacs. I am quite confused when I came across the
> relation of rlist and rvdw / rcoulomb in vdw and coulomb calculation.
>
> In cut-off, it says rvdw / rcoulomb should be greater than rlist, while
> in switch/shift, it says rvdw should be smaller than rlist...
>
> I guess in cut-off, it's to get all the updated atoms in the calculation
> of vdw/ coulomb by setting rvdw/ rcoulomb greater, but why it's the
> opposite in switch...
>
> Could you help me to understand how it's done? Thanks a lot. 

They're different algorithms, and they work as described. rlist controls 
the size of the neighbour lists, which list all the atom pairs whose 
interactions the nonbonded kernels compute each MD step. rvdw/rcoulomb 
control where the interaction is judged to be small enough that 
particles at greater distances can be neglected. The only basis 
governing their relative size is the algorithm you want to implement. 
With twin-range neighbourlists, the interactions between rvdw and 
rcoulomb and rlist are computed whenever the neighbour lists are 
constructed, and re-used for nstlist steps (see manual 3.4.2). With 
switch/shift interactions, GROMACS generates the potential in tabular 
form, so that if rlist > rvdw, then some interactions will be computed 
at distances such that the resulting potential will be known to be zero 
(see manual 4.1.5). In both cases there's a trade-off between the cost 
of re-constructing the neighbour lists, the cost of evaluating the 
resulting interactions and the accuracy limitations imposed by the 
underlying assumptions.

Mark