high performance computing on graphics processing units: hgpu.org

The PyGBe code solves the linearized Poisson-Boltzmann equation using a boundary-integral formulation. We use a boundary element method with a collocation approach, and solve it via a Krylov-subspace method. To do this efficiently, the matrix-vector multiplications in the Krylov iterations are accelerated with a treecode, achieving O(N log N) complexity. The code presents a Python environment for the user, while being efficient and fast. The core computational kernels are implemented in Cuda and interface with the user-visible code with PyCuda, for maximum ease-of-use combined with high performance on GPU hardware. This document provides background on the model and formulation of the numerical method, evidence of a validation exercise with well-known benchmarks – a spherical shell with a centered charge and one with an off-center charge- and a demonstration with a realistic biological geometry (lysozyme molecule).