Searching for structure in the binary black hole spin distribution

The spins of black holes in merging binaries can reveal information related to the formation and evolution of these systems. Combining events to infer the astrophysical distribution of black hole spins allows us to determine the relative contribution from different formation scenarios to the population. Many previous works have modeled spin population distributions using low-dimensional models with statistical or astrophysical motivations. While these are valuable approaches when the observed population is small, they make strong assumptions about the shape of the underlying distribution and are highly susceptible to biases due to mismodeling. The results obtained with such parametric models are valid only if the allowed shape of the distribution is well motivated (i.e., for astrophysical reasons). Unless the allowed shape of the distribution is well motivated (i.e., for astrophysical reasons), results obtained with such models, thus, may exhibit systematic biases with respect to the true underlying astrophysical distribution, along with resulting uncertainties not being reflective of our true uncertainty in the astrophysical distribution. In this work, we relax these prior assumptions and model the spin distributions using a more data-driven approach, modeling these distributions with flexible cubic spline interpolants in order to allow for capturing structures that the parametric models cannot. We find that adding this flexibility to the model substantially increases the uncertainty in the inferred distributions but find a general trend for lower support at high spin magnitude and a spin tilt distribution consistent with isotropic orientations. We infer that 62%--87% of black holes have spin magnitudes less than $a=0.5$ and 27%--50% (90% credible levels) of black holes exhibit negative ${\ensuremath{\chi}}_{\mathrm{eff}}$. Using the inferred ${\ensuremath{\chi}}_{\mathrm{eff}}$ distribution, we place a conservative upper limit of 37% for the contribution of hierarchical mergers to the astrophysical binary black hole population. Additionally, we find that artifacts from unconverged Monte Carlo integrals in the likelihood can manifest as spurious peaks and structures in inferred distributions, mandating the use of a sufficient number of samples when using Monte Carlo integration for population inference.