Searching beyond the fiducial stochastic gravitational wave background in pulsar timing array data using likelihood reweighting

Abstract: Since the recent announcements of evidence for a stochastic gravitational wave background from several pulsar timing array collaborations, much effort has been devoted to explore features beyond the fiducial Hellings-Downs background including those arising in modified gravity theories and deterministic gravitational wave signals. Inspired by previous studies, we propose a method to efficiently screen these models using likelihood reweighting based on the fiducial model. In order to alleviate the well-known unstable weight estimates in vanilla importance sampling, we implement reweighting for the second time making use of the kernel density estimation of the previously reweighted samples. We tested this method by analyzing three simulated datasets with an injected sinusoid signal applied to all pulsars. It is found that likelihood reweighting not only gives results compatible with those from full Bayesian analyses when the signal is subdominant, but is also able to recover the signal posterior to a reasonable accuracy in the presence of a rather strong signal. Given samples from the fiducial model, this method could bring an at least $\mathcal{O}(10)$-time speedup in analyzing new models.