On-the-fly clustering for exascale molecular dynamics simulations.

We propose a new on-the-fly approach for developing an algorithm that detects particle aggregates or, more broadly, regions of interest within molecular dynamics simulations. This algorithm should be versatile, capable of identifying any binarizable zone of interest. It must also be robust and scalable to function effectively with billions of atoms. Additionally, it should be simple and low-cost relative to the overall simulation expenses. This will enable physicists to obtain accurate measurements, including quantification of bars and qualification of data. To achieve these objectives, we have selected connected component analysis as our methodology.