Massively Parallel Multigrid Algorithms

The modern supercomputer architecture continues to add more computational power while communication remains expensive. This necessitates a new breed of numerical algorithms that are designed to minimize communication from the start, even if requires more on-processor computation.

Instead of asking "this is a nice method in serial, how can we do this in a parallel setting?", we should be asking "given this parallel setting, how can we design an optimal method?" The goal of this project is to develop discretization strategies and multigrid algorithms that are specifically suited for the next generation of massively parallel computing machines.