Fast adaptive computation of neighboring atoms

The main cost of a molecular dynamics or Monte Carlo simulation is the computation of the current potential energy or forces resulting from the interaction of atoms composing the molecular system. When a distance cut-off is used to speed up this computation, a fast method is needed to determine pairs of neighboring atoms. We have recently introduced an adaptive torsion-angle quasi-statics simulation algorithm, which enables users to finely trade between precision and computational cost, while providing some precision guarantees. In that algorithm, proximity queries are adaptively performed using hierarchies of oriented bounding boxes. In this paper, we show that using axis-aligned bounding boxes results in faster proximity queries. We thus introduce a semi-adaptive method to determine pairs of neighboring atoms, where all bounding boxes in the hierarchy are updated, but where interaction lists are adaptively updated during the simulation. The new method allows us to perform proximity queries about two orders of magnitude faster than the previous approach.