Fast and Accurate Evaluation of Nonlocal Coulomb and Dipole-Dipole Interactions via the Nonuniform FFT

TL;DR

This paper introduces a fast, accurate algorithm for computing nonlocal Coulomb and dipole-dipole interactions using NUFFT, overcoming singularity issues in Fourier space and requiring only modest degrees of freedom for smooth, decaying densities.

Contribution

The paper develops a novel NUFFT-based method with spherical and polar discretizations to efficiently evaluate singular Fourier integrals for nonlocal interactions.

Findings

Achieves $O(N \log N)$ computational complexity.

Effectively cancels singularity at the origin in Fourier space.

Demonstrates high accuracy and efficiency through numerical examples.

Abstract

We present a fast and accurate algorithm for the evaluation of nonlocal (long-range) Coulomb and dipole-dipole interactions in free space. The governing potential is simply the convolution of an interaction kernel $U(\bx)$ and a density function $\rho(\bx)=|\psi(\bx)|^2$, for some complex-valued wave function $\psi(\bx)$, permitting the formal use of Fourier methods. These are hampered by the fact that the Fourier transform of the interaction kernel $\widehat{U}(\bk)$ has a singularity at the origin $\bk={\bf 0}$ in Fourier (phase) space. Thus, accuracy is lost when using a uniform Cartesian grid in $\bk$ which would otherwise permit the use of the FFT for evaluating the convolution. Here, we make use of a high-order discretization of the Fourier integral, accelerated by the nonuniform fast Fourier transform (NUFFT). By adopting spherical and polar phase-space discretizations in three and two dimensions, respectively, the singularity in $\hat{U}(\bk)$ at the origin is canceled, so that only a modest number of degrees of freedom are required to evaluate the Fourier integral, assuming that the density function $\rho(\bx)$ is smooth and decays sufficiently fast as $\bx \rightarrow \infty$. More precisely, the calculation requires $O(N\log N)$ operations, where $N$ is the total number of discretization points in the computational domain. Numerical examples are presented to demonstrate the performance of the algorithm.