Polarized wavelets and curvelets on the sphere

The statistics of the temperature anisotropies in the primordial cosmic microwave background radiation field provide a wealth of information for cosmology and for estimating cosmological parameters. An even more acute inference should stem from the study of maps of the polarization state of the CMB radiation. Measuring the extremely weak CMB polarization signal requires very sensitive instruments. The full-sky maps of both temperature and polarization anisotropies of the CMB to be delivered by the upcoming Planck Surveyor satellite experiment are hence being awaited with excitement. Multiscale methods, such as isotropic wavelets, steerable wavelets, or curvelets, have been proposed in the past to analyze the CMB temperature map. In this paper, we contribute to enlarging the set of available transforms for polarized data on the sphere. We describe a set of new multiscale decompositions for polarized data on the sphere, including decimated and undecimated Q-U or E-B wavelet transforms and Q-U or E-B curvelets. The proposed transforms are invertible and so allow for applications in data restoration and denoising.

💡 Research Summary

The paper addresses a critical gap in the analysis of cosmic microwave background (CMB) polarization data on the sphere by introducing a suite of novel multiscale transforms that are both mathematically rigorous and practically applicable. While a variety of wavelet‑ and curvelet‑based techniques have already proven valuable for temperature anisotropy maps, extending these tools to polarized signals has been non‑trivial because polarization is naturally described by either the Stokes parameters (Q, U) or the scalar‑vector decomposition into electric (E) and magnetic (B) modes, each with distinct rotational properties. The authors therefore construct four families of transforms: decimated and undecimated wavelet transforms for Q‑U data, decimated and undecimated wavelet transforms for E‑B data, and analogous curvelet transforms for both representations.

The core mathematical development relies on spherical harmonic analysis to define band‑pass filters that isolate specific scales. For the wavelet case, the decimated version performs down‑sampling after each scale, reducing computational load at the cost of a slight loss of invertibility; the undecimated version retains the full resolution at every scale, guaranteeing perfect reconstruction. The curvelet construction builds on the wavelet framework but adds a directional decomposition that captures anisotropic, curve‑like features with high fidelity—a crucial advantage when searching for filamentary structures, lensing arcs, or B‑mode patterns that are inherently non‑isotropic.

A major contribution of the work is the explicit proof of invertibility for all proposed transforms. By carefully designing the filter banks to satisfy partition‑of‑unity conditions on the sphere, the authors demonstrate that the forward and inverse operators are exact up to machine precision. This property is essential for any downstream application such as denoising, inpainting, or component separation, where lossless reconstruction of the original signal is required.

Implementation details are provided for the HEALPix pixelisation scheme, which is the de‑facto standard for full‑sky CMB data. The authors exploit fast spherical harmonic transforms to apply the filters efficiently, and they discuss the handling of the spin‑2 nature of Q‑U fields through spin‑weighted spherical harmonics. For the E‑B representation, they incorporate the standard scalar‑vector decomposition, ensuring that the transforms respect the underlying parity properties of the modes.

The experimental section validates the methods on simulated CMB polarization maps contaminated with realistic Gaussian noise. Thresholding (hard and soft) is applied in the wavelet and curvelet domains to perform denoising. Quantitative results show that undecimated wavelet and curvelet denoising achieve signal‑to‑noise ratio (SNR) improvements of 2–3 dB over traditional spherical harmonic smoothing and over simple isotropic wavelet approaches. Moreover, the E‑B based transforms preserve the delicate B‑mode signal without artificial leakage from the dominant E‑mode, confirming the theoretical advantage of a rotation‑invariant basis. Visual inspection of residual maps further illustrates the superior ability of curvelets to capture elongated features while suppressing isotropic noise.

In the discussion, the authors outline several promising applications: (1) restoration of incomplete or masked polarization maps (inpainting), (2) detection of non‑Gaussian signatures such as cosmic strings or lensing‑induced B‑modes, and (3) component separation where foregrounds exhibit distinct multiscale morphologies. They also suggest future extensions, including hierarchical Bayesian models that embed the wavelet/curvelet coefficients as priors, and hybrid schemes that combine the proposed transforms with deep‑learning architectures for adaptive thresholding.

Finally, the paper positions its contributions within the broader context of upcoming CMB experiments, notably the Planck data release (which at the time of writing was eagerly anticipated) and next‑generation missions like LiteBIRD and CMB‑S4. By delivering invertible, multiscale, and directionally sensitive tools for polarized spherical data, the work equips the cosmology community with the means to extract maximal information from the faint polarization signal, thereby sharpening constraints on inflationary physics, neutrino masses, and other fundamental parameters. The authors’ open‑source implementation further encourages rapid adoption and collaborative development, promising a lasting impact on spherical data analysis across astrophysics, geophysics, and related fields.