We present a method that assesses the theoretical detection limit of a Bayesian Markov chain Monte Carlo search for a periodic gravitational wave signal emitted by a neutron star. Inverse probability yields an upper limit estimate for the strength wh
en a signal could not be detected in an observed data set. The proposed method is based on Bayesian model comparison that automatically quantifies Occam s Razor. It limits the complexity of a model by favoring the most parsimonious model that explains the data. By comparing the model with a signal from a pulsar to the null model that assumes solely noise, we derive the detection probability and an estimate for the upper limit that a search, for example, for a narrow-band emission for SN1987a, might yield on data at the sensitivity of LIGO data for an observation time of one year.
We estimate the probability of detecting a gravitational wave signal from coalescing compact binaries in simulated data from a ground-based interferometer detector of gravitational radiation using Bayesian model selection. The simulated waveform of t
he chirp signal is assumed to be a spin-less Post-Newtonian (PN) waveform of a given expansion order, while the searching template is assumed to be either of the same Post-Newtonian family as the simulated signal or one level below its Post-Newtonian expansion order. Within the Bayesian framework, and by applying a reversible jump Markov chain Monte Carlo simulation algorithm, we compare PN1.5 vs. PN2.0 and PN3.0 vs. PN3.5 wave forms by deriving the detection probabilities, the statistical uncertainties due to noise as a function of the SNR, and the posterior distributions of the parameters. Our analysis indicates that the detection probabilities are not compromised when simplified models are used for the comparison, while the accuracies in the determination of the parameters characterizing these signals can be significantly worsened, no matter what the considered Post-Newtonian order expansion comparison is.
