MD simulations can take days to compute, thus optimizing their performance is an important task. AutoPas is an open-source C++ library which tunes MD simulations and chooses the optimal algorithm for any given scenario. In the context of AutoPas, we implement the pairwise Verlet algorithm, which aims to improve on the existing Verlet lists neighbor identification algorithm. Other relevant components are added as well, such as a Structure of Arrays representation for the particle data. The new algorithm addresses the issue of scattered memory access and is designed to be well suited for shared memory parallelization. Our evaluation indicates that the pairwise Verlet algorithm performs well for a dense scenario or a larger interaction range between particles. However, other scenarios or other configurations such as the Structure of Arrays data layout appear to worsen its performance.

Keywords: Molecular Dynamics Simulation, Short-Range Forces, AutoPas