Quasi-Monte Carlo hyperinterpolation

TL;DR

This paper introduces QMC hyperinterpolation, a novel high-dimensional approximation method that replaces exact quadrature with quasi-Monte Carlo rules, achieving similar accuracy with reduced computational costs and enhanced robustness to noise.

Contribution

It proposes a new QMC hyperinterpolation scheme that relaxes quadrature exactness assumptions and provides algorithms for its construction, improving efficiency and robustness.

Findings

QMC hyperinterpolation attains accuracy comparable to traditional methods.

The method reduces computational costs in high-dimensional settings.

Lasso-based approach enhances robustness against sampling noise.

Abstract

This paper studies a generalization of hyperinterpolation over the high-dimensional unit cube. Hyperinterpolation of degree \( m \) serves as a discrete approximation of the \( L_2 \)-orthogonal projection of the same degree, using Fourier coefficients evaluated by a positive-weight quadrature rule that exactly integrates all polynomials of degree up to \( 2m \). Traditional hyperinterpolation methods often depend on exact quadrature assumptions, which can be impractical in high-dimensional contexts. We address the challenges and advancements in hyperinterpolation, bypassing the assumption of exactness for quadrature rules by replacing it with quasi-Monte Carlo (QMC) rules and propose a novel approximation scheme with an index set \( I \), which is referred to as QMC hyperinterpolation of range \( I \). In particular, we provide concrete construction algorithms for QMC hyperinterpolation with certain lattice rules. Consequently, we show that QMC hyperinterpolation achieves accuracy comparable to traditional hyperinterpolation while avoiding its higher computational costs. Furthermore, we introduce a Lasso-based approach to improve the robustness of QMC hyperinterpolation against noise from sampling processes. Numerical experiments validate the efficacy of our proposed methods.