Efficient Dealiased Convolutions without Padding

TL;DR

This paper introduces a new method for computing dealiased linear convolutions efficiently without zero-padding, reducing memory and computation costs, especially for higher-dimensional and higher-order convolutions, by leveraging FFT libraries.

Contribution

The authors develop algorithms for dealiased convolutions that eliminate the need for zero-padding, enabling more efficient higher-dimensional and higher-order convolution computations.

Findings

Memory requirements are comparable to zero-padding methods.

Significant reduction in computation time for higher-dimensional convolutions.

Implementation available in FFTW++ for practical use.

Abstract

Algorithms are developed for calculating dealiased linear convolution sums without the expense of conventional zero-padding or phase-shift techniques. For one-dimensional in-place convolutions, the memory requirements are identical with the zero-padding technique, with the important distinction that the additional work memory need not be contiguous with the input data. This decoupling of data and work arrays dramatically reduces the memory and computation time required to evaluate higher-dimensional in-place convolutions. The technique also allows one to dealias the higher-order convolutions that arise from Fourier transforming cubic and higher powers. Implicitly dealiased convolutions can be built on top of state-of-the-art fast Fourier transform libraries: vectorized multidimensional implementations for the complex and centered Hermitian (pseudospectral) cases have been implemented in the open-source software FFTW++.