2-Body molecular dynamics simulations do not always capture the correct qualitative changes to a system. In such cases, we need higher accuracy, which can be achieved by also considering 3-body interactions. To this end, we extend the molecular simulations library AutoPas with an implementation of Verlet lists that can handle 3-body interactions.
We consider three different approaches to iterate over all triplets in the force calculation step. The first uses 2-body neighbor lists and iterates over all pairs in them. The second approach intersects the lists of neighboring particles to obtain all mutual neighbors. Lastly, we save 3-body neighbor lists which contain all pairs of particles close enough for force calculation, which we can iterate over to find all triplets.
We find that the third approach is the fastest in medium to high-density simulations, while the first approach outperforms it in low-density scenarios. We also find that the 3-body neighbor lists take up much more memory than the 2-body neighbor lists, with the average neighbor list having a length that is proportional to the square of the average 2-body list length.
We keep both implementations for AutoTuning purposes and to provide a less memory-intensive solution.