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

In this process, we begin by assigning unique identifiers to each voxel in the simulation. As we traverse through the voxels, we label the first voxel as 1. The second voxel, which is connected to the first, receives the same label. This continues sequentially until we reach the sixth voxel, which is isolated. At this point, we define a new connected component and increment the label for the next set of connected voxels.

As we progress to the next row, we find that the sixth voxel is connected to one of the previously labeled voxels. We can then assign it the same label as the first connected component. This process allows us to establish a global graph of connected components. However, not all voxels are yet connected, necessitating a final pass to consolidate the information and assign labels based on the minimum identifiers.

The objective is to achieve domain labeling, which is necessary for parallel processing between threads and MPI (Message Passing Interface). The domain labeling between threads is relatively straightforward, but we will focus on the more complex MPI domain labeling next.