1. Authors: Giuseppe Puglisi, Davide Poletti, Giulio Fabbian, Carlo Baccigalupi, Luca Heltai, Radek Stompor
Title: Iterative map-making with two-level preconditioning for polarized Cosmic Microwave Background data sets
Abstract: An estimation of the sky signal from streams of Time Ordered Data (TOD) acquired by Cosmic Microwave Background (CMB) experiments is one of the most important steps in the context of CMB data analysis referred to as the map-making problem. The continuously growing CMB data sets render the CMB map-making problem more challenging in terms of computational cost and memory in particular in the context of ground based experiments. In this context, we study a novel class of the Preconditioned Conjugate Gradient (PCG) solvers which invoke two-level preconditioners. We compare them against PCG solvers commonly used in the map-making context considering their precision and time-to-solution. We compare these new methods on realistic, simulated data sets reflecting the characteristics of current and forthcoming CMB ground-based experiment. We develop an embarrassingly parallel implementation of the approach where each processor performs a sequential map-making for a subset of the TOD. We find that considering the map level residuals the new class of solvers permits achieving tolerance of up to 3 orders of magnitude better than the standard approach, where the residual level often saturates before convergence is reached. This corresponds to an important improvement in the precision of recovered power spectra in particular on the largest angular scales. The new method also typically requires fewer iterations to reach a required precision and thus shorter runtimes for a single map-making solution. However, the construction of an appropriate two-level preconditioner can be as costly as a single standard map-making run. Nevertheless, if the same problem needs to be solved multiple times, e.g., as in Monte Carlo simulations, this cost has to be incurred only once, and the method should be competitive not only as far as its precision but also its performance is concerned.
2. Authors: Jan Papež, Laura Grigori, Radek Stompor
Title: Solving linear equations with messenger-field and conjugate gradients techniques – an application to CMB data analysis
Abstract: We discuss linear system solvers invoking a messenger-field and compare them with (preconditioned) conjugate gradients approaches. We show that the messenger-field techniques correspond to fixed point iterations of an appropriately preconditioned initial system of linear equations. We then argue that a conjugate gradient solver applied to the same preconditioned system, or equivalently a preconditioned conjugate gradient solver using the same preconditioner and applied to the original system, will in general ensure at least a comparable and typically better performance in terms of the number of iterations to convergence and time-to-solution. We illustrate our conclusions on two common examples drawn from the Cosmic Microwave Background data analysis: Wiener filtering and map-making. In addition, and contrary to the standard lore in the CMB field, we show that the performance of the preconditioned conjugate gradient solver can depend importantly on the starting vector. This observation seems of particular importance in the cases of map-making of high signal-to-noise sky maps and therefore should be of relevance for the next generation of CMB experiments.
3. Authors: Jan Papež, Laura Grigori, Radek Stompor
Title: Accelerating linear system solvers for time domain component separation of cosmic microwave background data.
Abstract: Component separation is one of the key stages of any modern, cosmic microwave background (CMB) data analysis pipeline. It is an inherently non-linear procedure and typically involves a series of sequential solutions of linear systems with similar, albeit not identical system matrices, derived for different data models of the same data set. Sequences of this kind arise for instance in the maximization of the data likelihood with respect to foreground parameters or sampling of their posterior distribution. However, they are also common in many other contexts. In this work we consider solving the component separation problem directly in the measurement (time) domain, which can have a number of important advantageous over the more standard pixel-based methods, in particular if non-negligible time-domain noise correlations are present as it is commonly the case. The time-domain based approach implies, however, significant computational effort due to the need to manipulate the full volume of time-domain data set. To address this challenge, we propose and study efficient solvers adapted to solving time-domain-based, component separation systems and their sequences and which are capable of capitalizing on information derived from the previous solutions. This is achieved either via adapting the initial guess of the subsequent system or through a so-called subspace recycling, which allows to construct progressively more efficient, two-level preconditioners. We report an overall speed-up over solving the systems independently of a factor of nearly 7, or 5, in the worked examples inspired respectively by the likelihood maximization and likelihood sampling procedures we consider in this work.