Abstract
Zernike polynomials are a basis of orthogonal polynomials on the unit disk that provide a natural framework for representing smooth functions. They are widely utilized in fields such as optics and atmospheric sciences. This paper serves as a self-contained reference for Zernike polynomials, detailing algorithms for their evaluation and introducing novel numerical schemes for quadrature and interpolation. These schemes utilize a tensor product of equispaced nodes in the angular direction and the roots of specific Jacobi polynomials in the radial direction. Furthermore, the paper extends these properties to higher-dimensional unit balls. Numerical experiments demonstrate the efficiency and precision of the proposed methods.
Introduction
Zernike polynomials have long been the "gold standard" for representing smooth functions on a unit disk, particularly for describing lens aberrations and wavefront distortions. However, researchers frequently encounter practical hurdles when implementing them numerically. Existing literature is often fragmented, with conflicting notations and properties scattered across different disciplines. Moreover, standard numerical integration and interpolation methods are often not optimized for the disk's specific geometry, leading to inefficiencies or artifacts at the origin.
This paper addresses these gaps by providing a unified mathematical reference and introducing high-performance numerical schemes tailored to the geometry of the disk and its higher-dimensional counterparts.
Methodology
Numerical Framework
The authors propose a numerical framework based on a tensor product of nodes, specifically designed to handle the polar coordinate system of the disk:
- Angular Direction: The scheme employs equispaced nodes, which are the natural choice for periodic functions in the angular domain.
- Radial Direction: In the radial direction, the authors utilize the roots of shifted Jacobi polynomials, denoted as $\tilde{P}_n$. These polynomials are orthogonal on the interval $[0, 1]$ with respect to the weight function $w(x) = x$, which perfectly matches the $r \, dr$ area element of polar coordinates.
- Evaluation and Root Finding: To ensure numerical stability, the paper utilizes Kintner’s method, a recurrence relation that prevents the "blow-up" often seen with high-degree polynomials. For finding radial nodes, the authors describe an algorithm using the Prüfer transform, which converts the task of finding roots of oscillatory functions into solving a well-behaved differential equation.
Generalization to Higher Dimensions
A significant contribution of this work is the generalization of these 2D disk concepts to $(p+2)$-dimensional unit balls. This is achieved through the use of orthonormal surface harmonics, providing a robust framework for high-dimensional smooth function approximation.
Experimental Evaluation
To validate the proposed algorithms, the authors conducted numerical experiments using three synthetic test functions designed to challenge different aspects of the numerical schemes:
- Rational Function ($f_1$): $f_1(x, y) = [1 + 25(x^2 + y^2)]^{-1}$, which features a sharp peak at the origin.
- High-Frequency Bessel Function ($f_2$): $f_2(r, \theta) = J_{100}(150r) \cos(100\theta)$, testing the limits of oscillatory integration.
- Product of Legendre Polynomials ($f_3$): $f_3(r, \theta) = P_8(x)P_{12}(y)$, serving as a standard smooth test case.
Key Insights
The Radial Advantage
The proposed quadrature rule demonstrates a clear efficiency gain: it can integrate all Zernike polynomials up to order $2m-1$ using exactly $m$ radial nodes. In contrast, standard Gauss-Legendre nodes would require $m+1$ nodes. While this one-node difference may seem marginal in 2D, the advantage compounds significantly as the dimensionality of the problem increases.
Sparsity and Smoothness
For smooth functions, Zernike expansions are remarkably compact, requiring approximately half as many terms as standard Chebyshev-Fourier expansions to achieve equivalent precision. Furthermore, because the interpolation basis consists entirely of Zernike polynomials, the resulting approximation is guaranteed to be smooth across the entire disk, including the origin—a common failure point for other coordinate-based interpolation schemes.
Performance Results
The numerical results confirm the high precision of the scheme. For the rational test function ($f_1$), the quadrature reached a relative error of approximately $10^{-14}$, near machine epsilon, using a $40 \times 80$ grid. The scheme also proved robust against high-frequency oscillations; for the Bessel function ($f_2$), the error dropped to machine precision as soon as the number of nodes was sufficient to resolve the oscillations (approximately 20 radial nodes).
Conclusion and Future Work
This paper provides a comprehensive "one-stop-shop" for Zernike polynomials, offering both a theoretical reference and practical numerical tools. While the methods are highly efficient for smooth functions, their performance on functions with singularities or discontinuities remains an area for further investigation. Additionally, while the paper outlines the theory for $O(M^2 \log M)$ acceleration using Fast Multipole Methods (FMM) or Fast Fourier Transforms (FFT), the current experiments utilized the $O(M^3)$ version. Implementing and benchmarking these accelerated versions for very high-degree expansions represents a logical next step for future research.
Summary updated