Approximately 95 pc. of the computing time in a simulation is expended for
the calculation of the pair forces, torques, energies etc. For systems
of more than particles the interaction is usually cut off at some
pair distance smaller than half the box side length: .
Even the scanning of all particle pairs for the condition
may become costly for large systems.
Verlet suggested that in such cases the possible interaction partners
of each particle are listed in an array, and only the particles in
that list are considered. As the molecules diffuse around, the lists
must be updated regularly by an occasional scan. Here is the
detailed procedure:
Figure 2.4:
Verlet neighbour lists:
.
All particles within a distance of particle are
entered in a list.
These lists of neighbours, , are stored in sequence, in one long
array. For each a pointer points to the start of
the list of its neighbours:
Figure 2.5:
Storage of Verlet lists
The first setup of the lists requires operations; every
10-20 steps the lists must be updated, again at the same cost.
However, in the time of diffusion between and ,
for each particle only the possible partners with indices
to
need be considered.
F. J. Vesely / University of Vienna