Quality over Quantity: Optimizing pulsar timing array analysis for stochastic and continuous gravitational wave signals

Lorenzo Speri, Nataliya K. Porayko, Mikel Falxa, Siyuan Chen, Jonathan R. Gair, Alberto Sesana, Stephen R. Taylor

Submitted on 6 November 2022


The search for gravitational waves using Pulsar Timing Arrays (PTAs) is a computationally expensive complex analysis that involves source-specific noise studies. As more pulsars are added to the arrays, this stage of PTA analysis will become increasingly challenging. Therefore, optimizing the number of included pulsars is crucial to reduce the computational burden of data analysis. Here, we present a suite of methods to rank pulsars for use within the scope of PTA analysis. First, we use the maximization of the signal-to-noise ratio as a proxy to select pulsars. With this method, we target the detection of stochastic and continuous gravitational wave signals. Next, we present a ranking that minimizes the coupling between spatial correlation signatures, namely monopolar, dipolar, and Hellings & Downs correlations. Finally, we also explore how to combine these two methods. We test these approaches against mock data using frequentist and Bayesian hypothesis testing. For equal-noise pulsars, we find that an optimal selection leads to an increase in the log-Bayes factor two times steeper than a random selection for the hypothesis test of a gravitational wave background versus a common uncorrelated red noise process. For the same test but for a realistic EPTA dataset, a subset of 25 pulsars selected out of 40 can provide a log-likelihood ratio that is 89% of the total, implying that an optimally selected subset of pulsars can yield results comparable to those obtained from the whole array. We expect these selection methods to play a crucial role in future PTA data combinations.


Subjects: Astrophysics - High Energy Astrophysical Phenomena; General Relativity and Quantum Cosmology


Distribution of angular separations {of 25 pulsars} selected with three selection methods, namely SNR$_\textrm{B}$-maximization, Coupling Matrix and Chimera. These methods have been applied to a dataset consisting of 200 pulsars with uniform sky distribution and equal noise properties. For reference, we also show a random selection of 25 pulsars.