Project 4–5: PFMMPACK
Principal Investigator: Eric Darve (Stanford)

(project duration: 2007-2009)

Many scientific and engineering problems of importance to the Army can be solved using integral equations. Such problems include material fracture behavior, microfluidic flows in sensing devices, electrostatic calculations in protein toxin simulations, and detection and imaging using acoustic signals.

The fast multipole method (FMM) reduces dramatically the computational cost and memory for simulating the dynamics of systems of particles and solving integral equations; however, the mathematics behind this method are extremely complex and its implementation is a real challenge, particularly on parallel systems.

The goal of this project is to develop FMMs for a variety of applications on sequential and parallel computers with an emphasis on computational efficiency, scalability, and error control.

PFMMPACK will provide an efficient, portable library to solve these equations, and it will be integrated easily into existing and new HPC applications. The package will include error estimation and is designed specifically to take advantage of existing and new architectures, including general purpose graphics processing units (GPGPUs).