Parallel Higher-order Boundary Integral Electrostatics Computation on Molecular Surfaces with Curved Triangulation

TL;DR

This paper introduces a parallel higher-order boundary integral method for solving the Poisson-Boltzmann equation on molecular surfaces, achieving improved accuracy and convergence crucial for biomolecular electrostatics.

Contribution

It presents a novel parallel higher-order boundary integral approach with curved triangulation, enhancing accuracy and efficiency over traditional flat-triangle methods.

Findings

Significantly improved accuracy over centroid collocation methods.

Enhanced convergence of the boundary integral Poisson-Boltzmann solver.

Effective parallelization using MPI accelerates computations.

Abstract

In this paper, we present a parallel higher-order boundary integral method to solve the linear Poisson-Boltzmann (PB) equation. In our method, a well-posed boundary integral formulation is used to ensure the fast convergence of Krylov subspace linear solver such as GMRES. The molecular surfaces are first discretized with flat triangles and then converted to curved triangles with the assistance of normal information at vertices. To maintain the desired accuracy, four-point Gauss-Radau quadratures are used on regular triangles and sixteen-point Gauss-Legendre quadratures together with regularization transformations are applied on singular triangles. To speed up our method, we take advantage of the embarrassingly parallel feature of boundary integral formulation, and parallelize the schemes with the message passing interface (MPI) implementation. Numerical tests show significantly improved accuracy and convergence of the proposed higher-order boundary integral Poisson-Boltzmann (HOBI-PB) solver compared with boundary integral PB solver using often-seen centroid collocation on flat triangles. The higher-order accuracy results achieved by present method are important to sensitive solvation analysis of biomolecules, particularly when accurate electrostatic surface potentials are critical in the molecular simulation. In addition, the higher-order boundary integral schemes presented here and their associated parallelization potentially can be applied to solving boundary integral equations in a general sense.