Computing with functions in spherical and polar geometries II. The disk

Heather Wilber, Alex Townsend, Grady B. Wright

Abstract

A collection of algorithms is described for numerically computing with smooth functions defined on the unit disk. Low rank approximations to functions in polar geometries are formed by synthesizing the disk analogue of the double Fourier sphere method with a structure-preserving variant of iterative Gaussian elimination that is shown to converge geometrically for certain analytic functions. This adaptive procedure is near-optimal in its sampling strategy, producing approximants that are stable for differentiation and facilitate the use of FFT-based algorithms in both variables. The low rank form of the approximants is especially useful for operations such as integration and differentiation, reducing them to essentially 1D procedures, and it is also exploited to formulate a new fast disk Poisson solver that computes low rank approximations to solutions. This work complements a companion paper (Part I) on computing with functions on the surface of the unit sphere.

Introduction

Numerical computation on a disk is inherently complicated by the coordinate singularity at the origin ($r=0$). In standard polar coordinates, the origin acts as a "pole" where all angular lines converge. This geometry imposes two significant constraints: functions must satisfy specific matching conditions at the origin to remain smooth, and standard grids tend to oversample the center, leading to computational inefficiency or stability issues in time-dependent simulations.

Existing numerical approaches often force a trade-off between the speed of Fast Fourier Transforms (FFT) and the preservation of regularity at the origin. This paper presents a framework that avoids these pitfalls by combining a "doubled" domain strategy with low-rank function approximation.

Methodology: Doubling and Structure Preservation

The authors' approach is built upon three core components designed to maintain efficiency and spectral accuracy:

  1. The Double Fourier Sphere (DFS) Analogue: The disk's domain is "doubled" to map a function $f(r, heta)$ onto a larger, periodic domain. This transformation effectively turns the disk into a rectangle where the origin is an interior point, allowing for the application of standard periodic numerical techniques.
  2. BMC-II Symmetry: To ensure the function remains physically meaningful and smooth at the pole, the method enforces Block-Mirror Centrosymmetric (BMC-II) symmetry. This constraint ensures the doubled function is consistent across the origin.
  3. Iterative Gaussian Elimination (GE): Functions are represented as low-rank approximants—sums of radial functions multiplied by angular functions. A structure-preserving variant of iterative GE adaptively selects the most important "slices" of the function, ensuring that every step of the approximation maintains the required BMC-II symmetry.
The Nobel medal (a) projected into polar coordinates (b) and then "doubled" using the DFS analogue (c) to create a periodic BMC-II structure.

This low-rank representation is a key solution to the oversampling problem. By sampling the function only at a sparse "skeleton" of points rather than a dense tensor-product grid, the method achieves near-optimal compression. For analytic functions, the approximation error converges geometrically as the rank increases.

Key Results and Applications

The low-rank framework enables significant performance gains in common calculus operations and differential equation solving:

  • 1D Efficiency for 2D Problems: Because the approximants are decomposed into 1D components, operations like integration and differentiation are reduced to a series of 1D procedures. These are extremely fast and leverage highly optimized software implementations.
  • Fast Poisson Solver: The authors developed a solver for the Poisson equation ($\nabla^2 u = f$) using the Alternating Direction Implicit (ADI) method. By exploiting the low-rank structure of the forcing function, the solver can find solutions on a $10,000 \times 10,000$ grid in under 5 seconds—up to five times faster than standard optimal-complexity solvers.
  • Software Integration: These algorithms are fully implemented in the open-source Chebfun package for MATLAB, specifically within the diskfun class.

Conclusion

This work provides a robust and efficient path for performing high-accuracy computations on the unit disk. By synthesizing domain doubling with structure-preserving low-rank approximations, the authors resolve the long-standing numerical difficulties associated with polar singularities. While the method is optimized for smooth functions, future work will explore extending these low-rank strategies to 3D ball geometries.

Summary updated

Abstract

A collection of algorithms is described for numerically computing with smooth functions defined on the unit disk. Low rank approximations to functions in polar geometries are formed by synthesizing the disk analogue of the double Fourier sphere method with a structure-preserving variant of iterative Gaussian elimination that is shown to converge geometrically for certain analytic functions. This adaptive procedure is near-optimal in its sampling strategy, producing approximants that are stable for differentiation and facilitate the use of FFT-based algorithms in both variables. The low rank form of the approximants is especially useful for operations such as integration and differentiation, reducing them to essentially 1D procedures, and it is also exploited to formulate a new fast disk Poisson solver that computes low rank approximations to solutions. This work complements a companion paper (Part I) on computing with functions on the surface of the unit sphere.
BibTeX
@article{wilber2017computing,
  author = {Heather Wilber and Alex Townsend and Grady B. Wright},
  title = {{Computing with functions in spherical and polar geometries II. The disk}},
  journal = {SIAM J. Sci. Comput.},
  year = {2017},
  doi = {10.1137/16m1070207},
}
Figure