Improving Posterior Predictive Checks for Gravitational Wave Population Analyses

Other Versions:

Abstract:: In population analyses of gravitational waves emitted from binary black holes (BBH), spin magnitude and tilt angle distributions provide critical insights regarding BBH evolutionary histories and formation channels. For this reason, developing reliable models of BBH spin component distributions continues to be an essential problem. However, the effects of spin magnitude and directions on gravitational wave signals are subdominant compared to the influence of the effective aligned spin and the effective precessing spin parameters. As a result, obtaining well-constrained measurements of these parameters is difficult, posing challenges when developing models of their astrophysical distributions and determining model accuracy. Posterior predictive checks (PPCs), a widely used model-checking method in gravitational wave science, especially fall short when applied using data with high uncertainties. In this project, we implement alternative data-level PPCs, partial PPCs, and split predictive checks. We aim to explore the efficacy of these methods by applying them to models of varying accuracy of simulated astrophysical populations with the same effective-spin distribution and different spin component distributions. Currently, we have successfully replicated previous results that demonstrate the inability of PPCs to demonstrate model misspecification when the data is highly uncertain.

Files in Document:

Other Files:

LIGO Document T2400245-v1