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

Molecular Dynamics is characterized as an unbound problem, yet it possesses strong properties. The potential function defines the interactions between particles, indicating that they repel each other when too close and attract as they move further apart. A key feature of this function is its rapid convergence towards zero.

To optimize calculations in molecular dynamics, we implement a cutoff, limiting the potential function to a maximum radius. This approach ensures that we only consider particles within a specified distance from a central particle at any given moment.

In the context of parallelization, we utilize a domain decomposition strategy. Each computation node has a shared memory horizon, and each MPI process operates within a designated bluish area, surrounded by a ghost region. This ghost area contains copies of particles from neighboring MPI processes, facilitating the exchange of information necessary to update properties across processes.

Now, we will apply the molecular dynamics codes to perform physics simulations. For instance, we can observe a video illustrating an impact on Taluma, showcasing the impact event, the reflection of waves, and the nucleation of voids.