[Paper Review] A Note on Spherical Needlets

Minjie Fan|arXiv (Cornell University)|Aug 21, 2015

Electromagnetic Scattering and Analysis12 references3 citations

TL;DR

This paper introduces spherical needlets as a new class of spherical wavelets that offer superior spatial and frequency localization compared to traditional spherical harmonics, enabling sparse and accurate representation of functions with localized features. The authors present a Matlab package implementing spherical needlets and demonstrate through examples that thresholding needlet coefficients effectively removes Gibbs-type oscillations present in spherical harmonic reconstructions, achieving artifact-free, sparse approximations.

ABSTRACT

Compared with the traditional spherical harmonics, the spherical needlets are a new generation of spherical wavelets that possess several attractive properties. Their double localization in both spatial and frequency domains empowers them to easily and sparsely represent functions with small spatial scale features. This paper is divided into two parts. First, it reviews the spherical harmonics and discusses their limitations in representing functions with small spatial scale features. To overcome the limitations, it introduces the spherical needlets and their attractive properties. In the second part of the paper, a Matlab package for the spherical needlets is presented. The properties of the spherical needlets are demonstrated by several examples using the package.

Motivation & Objective

• To address the limitations of spherical harmonics in representing functions with small-scale, localized features due to global support and Gibbs phenomena.
• To introduce spherical needlets as a multiresolution alternative with superior spatial and frequency localization.
• To develop and release a practical Matlab package for computing spherical needlet transforms.
• To demonstrate through numerical examples that thresholding needlet coefficients yields sparse, artifact-free reconstructions compared to spherical harmonics.
• To provide a computational framework for researchers working with spherical data in geophysics, cosmology, and statistics.

Proposed method

• Construct spherical needlets using a Littlewood-Paley decomposition via a window function b(⋅) that is positive on (1/B, B) and satisfies a partition of unity condition.
• Discretize the sphere using exact cubature rules (Theorem 1) to define cubature points ξjk and weights λjk for each frequency scale j.
• Define spherical needlets as ψjk(x) = √λjk ∑_{l=⌈B^{j-1}⌉}^{⌊B^{j+1}⌋} b(l/B^j) ∑_m Ylm(ξjk) Y̅lm(x), leveraging the addition formula for efficient computation.
• Implement the forward and inverse spherical needlet transforms using fast algorithms, including FFT-based evaluation on HEALPix grids.
• Estimate needlet coefficients via inner products with spherical harmonics coefficients, enabling conversion between representations.
• Apply percentage-based hard thresholding to needlet coefficients (e.g., 95% of smallest magnitudes set to zero) to achieve sparse, artifact-free reconstructions.

Experimental results

Research questions

• RQ1 Can spherical needlets provide a more effective representation of spherical functions with localized features than spherical harmonics?
• RQ2 How does the sparsity of needlet coefficients compare to that of spherical harmonic coefficients for functions with small-scale features?
• RQ3 To what extent can thresholding needlet coefficients suppress Gibbs-type oscillations in function reconstruction?
• RQ4 Can the proposed Matlab package efficiently compute spherical needlet transforms for practical applications?
• RQ5 What is the relationship between the frequency localization of needlets and the spatial resolution of the reconstructed function?

Key findings

• Spherical needlets exhibit quasi-exponential decay in spatial domain, enabling strong localization around their centers, unlike globally supported spherical harmonics.
• The probability density functions of needlet coefficients at different scales (j=2,3,4) are spiky, indicating that only a few coefficients carry significant energy, confirming sparsity.
• After hard thresholding 95% of the smallest magnitude needlet coefficients at j=4, the reconstructed function shows no global oscillations, unlike the spherical harmonic reconstruction with l_max=16.
• The reconstructed function using thresholded needlet coefficients matches the original function well within its support, demonstrating that spatial localization preserves local accuracy.
• The inverse spherical needlet transform can be efficiently computed using FFT-based methods on HEALPix grids, enabling scalable implementation.
• The exact transformation between spherical harmonics and needlet coefficients ensures that unthresholded reconstructions are identical, but thresholding removes artifacts inherent in spherical harmonic expansions.