An Icosahedron-based Method for Pixelizing the Celestial Sphere

Max TegmarkThe Astrophysical Journal 1996 · DOI

Abstract

For power spectrum estimation it's important that the pixelization of a CMB sky map be smooth and regular to high degree. With this criterion in mind the "COBE sky cube" was defined. This paper has as central theme to further improve on this elegant scheme which uses a cube as projective base—here an icosahedron is used in its place. Although the sky cube is excellent, a further reduction of 20 percent of the number of pixels can be obtained while the pixel distance is maintained, and without any degradation of accuracy for integration. The pixels are rounder in this scheme where they are hexagonal rather than square, and the faces are small in this implementation which simplifies area-equalization. The reason distortion is lessened is that the faces are smaller and therefore more flat. To use the method, you can get a FORTRAN code from the Internet.

Introduction: The Challenge of Spherical Pixelization

In cosmology, analyzing continuous signals like the Cosmic Microwave Background (CMB) requires "pixelizing" the celestial sphere—converting a continuous signal into a discrete map. This process is inherently difficult because a sphere cannot be tiled with perfectly regular shapes without introducing distortion.

By 1996, the standard approach was the COBE sky cube, which projected a cube onto a sphere. However, because a cube has only six large faces, pixels near the edges and corners become significantly distorted when projected onto the curved surface. For high-precision tasks like power spectrum estimation, researchers require pixels that are regular and represent equal areas. As datasets from missions like MAP and Planck grew, even minor inefficiencies in pixelization could lead to exponential increases in computational costs, which often scale as $O(N^3)$ where $N$ is the number of pixels.

Methodology: From Cubes to Icosahedrons

Max Tegmark proposed an improved scheme based on the icosahedron, a Platonic solid with 20 triangular faces. The increased number of faces means each face is smaller and "flatter" relative to the sphere, inherently reducing projection distortion.

Comparison of pixelization schemes: cube-based (left) vs. icosahedron-based (right).

The method involves three primary steps:

  1. Inscribe and Grid: A sphere is inscribed within an icosahedron, and each of its 20 triangular faces is tiled with a regular triangular grid.
  2. Radial Projection: These grid points are projected radially onto the sphere's surface.
  3. Area Equalization: Since radial projection does not preserve area, a coordinate transformation (Jacobian adjustment) is applied to shift points slightly, ensuring each pixel represents an equal area on the sky.

This process results in hexagonal pixels, which are more efficient for surface coverage than the square pixels produced by cube-based methods.

Key Results and Performance

The icosahedron method was evaluated using geometric benchmarks and window function analysis, comparing a 6252-point icosahedron grid against a 6144-point COBE sky cube.

Window function comparison between the 6144-point COBE sky cube and the 6252-point icosahedron grid.

Key findings include:

  • Reduced Distortion: The maximum distance from any point on the sphere to the nearest pixel center is reduced by approximately 10% compared to the cube-based method.
  • Enhanced Integration Accuracy: The icosahedron grid maintains integration accuracy up to roughly 10% higher multipole values ($\ell$) than a comparable cube-based grid.
  • Computational Efficiency: To match the accuracy of the COBE sky cube, the icosahedron method requires ~20% fewer pixels. This reduction is significant; a 12% improvement in resolution limits can allow for a 25% reduction in pixel count, effectively halving the CPU time required for $O(N^3)$ analysis.

Discussion and Legacy

While the icosahedron method represents a near-optimal solution to the "cubature problem" (the challenge of distributing $N$ points on a sphere), it is not mathematically perfect. However, Tegmark notes that further improvements using solids with more faces would likely yield diminishing returns relative to the increased complexity.

This work served as a critical milestone in CMB data processing. While later schemes like HEALPix introduced additional features like hierarchical structures and faster spherical harmonic transforms, the icosahedron-based approach demonstrated the profound impact that geometric optimization can have on large-scale cosmological computations.

Summary updated

Abstract

For power spectrum estimation it's important that the pixelization of a CMB sky map be smooth and regular to high degree. With this criterion in mind the ``COBE sky cube" was defined. This paper has as central theme to further improve on this elegant scheme which uses a cube as projective base - here an icosahedron is used in its place. Although the sky cube is excellent, a further reduction of 20 percent of the number of pixels can be obtained while the pixel distance is maintained, and without any degradation of accuracy for integration. The pixels are rounder in this scheme where they are hexagonal rather than square, and the faces are small in this implementation which simplifies area-equalization. The reason distortion is lessened is that the faces are smaller and therefore more flat. To use the method, you can get a FORTRAN code from the Internet.
BibTeX
@article{tegmark1996icosahedron,
  author = {Max Tegmark},
  title = {{An Icosahedron-based Method for Pixelizing the Celestial Sphere}},
  journal = {The Astrophysical Journal},
  year = {1996},
  doi = {10.1086/310310},
}
Figure