Optimal filtering for CMB lensing reconstruction

TL;DR

This paper develops and compares optimized quadratic-estimator methods for CMB lensing reconstruction, demonstrating significant improvements in variance and robustness for upcoming high-resolution experiments.

Contribution

It introduces a second-stage filtering and analytic response model for lensing maps, enhancing the accuracy and efficiency of lensing power spectrum estimation.

Findings

Substantial variance reduction with optimal anisotropic filtering

Up to 30% variance improvement with additional filtering step

Analytic response model is accurate within a small correction

Abstract

Upcoming ground-based cosmic microwave background experiments will provide CMB maps with high sensitivity and resolution that can be used for high fidelity lensing reconstruction. However, the sky coverage will be incomplete and the noise highly anisotropic, so optimized estimators are required to extract the most information from the maps. We focus on quadratic-estimator based lensing reconstruction methods that are fast to implement, and compare new more-optimally filtered estimators with various estimators that have previously been used in the literature. Input CMB maps can be optimally inverse-signal-plus-noise filtered using conjugate gradient (or other) techniques to account for the noise anisotropy. However, lensing reconstructions from these filtered input maps have an anisotropic response to the lensing signal and are difficult to interpret directly. We describe a second-stage filtering of the lensing maps and analytic response model that can be used to construct lensing power spectrum estimates that account for the anisotropic response and noise inhomogeneity in an approximately optimal way while remaining fast to compute. We compare results for simulations of upcoming Simons Observatory and CMB Stage-4 experiments to show the robustness of the more optimal lensing reconstruction pipeline and quantify the improvement compared to less optimal estimators. We find a substantial improvement in reconstructed lensing power variance between optimal anisotropic and isotropic filtering of CMB maps, and up to 30% improvement in variance by using the additional filtering step on the reconstruction potential map. Our approximate analytic response model is unbiased to within a small percent-level additional Monte Carlo correction.