Paper detail

Applying Bayesian Inference to the first International Pulsar Timing Array data challenge

This is a very brief summary of the techniques I used to analyze the IPTA challenge 1 data sets. I tried many things, and more failed than succeeded, but in the end I found two approaches that appear to work based on tests done using the open data sets. One approach works directly with the time domain data, and the other works with a specially constructed Fourier transform of the data. The raw data was run through TEMPO2 to produce reduced timing residuals for the analysis. Standard Markov Chain Monte Carlo techniques were used to produce samples from the posterior distribution function for the model parameters. The model parameters include the gravitational wave amplitude and spectral slope, and the white noise amplitude for each pulsar in the array. While red timing noise was only included in Dataset 3, I found that it was necessary to include effective red noise in all the analyses to account for some of the spurious effects introduced by the TEMPO2 timing fit. This added an additional amplitude and slope parameter for each pulsar, so my overall model for the 36 pulsars residuals has 110 parameters. As an alternative to using an effective red noise model, I also tried to simultaneously re-fit the timing model model while looking for the gravitational wave signal, but for reasons that are not yet clear, this approach was not very successful. I comment briefly on ways in which the algorithms could be improved. My best estimates for the gravitational wave amplitudes in the three closed (blind) data sets are: (1) $A=(7.3\pm 1.0)\times 10^{-15}$; (2) $A=(5.7\pm 0.6)\times 10^{-14}$; and (3) $A=(4.6\pm 1.3)\times 10^{-15}$.