A Flexible Noise Model for Pulsar Timing Array Analyses

As pulsar timing array datasets get larger and more sensitive, it's crucial to improve how we model and understand noise in our observations. Traditionally, we've used simple mathematical models for the power spectral density of the various kinds of noise in our data: either a two-parameter power-law, or a fully free spectrum model where each frequency bin is fit independently. A power-law works reasonably well, but there are physical reasons one might expect non-power-law behavior. However, a free spectrum model offers maximum flexibility while it likely also over-fits the data. In our work, we introduce a new model that smoothly fills the gap between these two approaches. By combining several adjustable components with suitable shrinkage priors, our model can adapt to the data, turning certain features on or off as needed. This flexibility allows us to better represent the actual noise without unnecessary complications. We will give examples of where this is useful for both individual pulsar analyses, and for a full analysis of an array of pulsars.