Speaker
Description
Low-frequency correlated noise is a critical systematic for next-generation CMB experiments targeting the primordial B-mode signal. 1/f noise residuals are known to contaminate large angular scales and bias the estimation of the tensor-to-scalar ratio if they are not mitigated properly. In this context, we present BrahMap, a scalable and modular map-making framework designed for large-scale data reduction for current and future CMB polarization experiments. It features a user-friendly Python interface backed by performance-optimized C++ extensions with MPI+OpenMP parallelization. BrahMap implements generalized least squares (GLS) map-making with a large suite of noise covariance operators - from diagonal and circulant approximations to full Toeplitz operators - all accessible through a unified linear operator algebra interface. We explore several strategies for mitigating 1/f noise at the map-making level. First, we perform a systematic comparison of different circulant approximations of the exact noise covariance and assess their noise mitigation performance. Next, we implement superfast Toeplitz solvers to enable computing the exact GLS estimate using the full Toeplitz noise covariance. With realistic LiteBIRD simulations across a wide range of 1/f noise regimes, we demonstrate that the exact Toeplitz approach reduces the large-scale noise residual power by up to two orders of magnitude compared to circulant approximations, translating to an approximately four-fold reduction in the bias on the tensor-to-scalar ratio.