Astrophysics informed Gaussian processes for gravitational-wave populations: Evidence for the onset of the pair-instability supernova mass gap

We analyze binary black hole (BBH) mergers from the latest Gravitational Wave Transient Catalog (GWTC-3) using a flexible, non-parametric framework to infer the underlying black hole mass distribution. Our model employs Gaussian Processes (GPs) with astrophysically motivated priors to represent the mass distribution, assuming both component masses are independently drawn from a common mass function. This approach enables us to capture complex features in the population without relying on rigid parametric forms. Motivated by predictions from binary stellar evolution, we focus on the presence of a mass gap in the range $40-120~ M_\odot$, attributed to the pair-instability supernova (PISN) and pulsational PISN (PPISN) processes. Using our GP-based model, we find for the first time, strong evidence for the onset of this mass gap, locating its lower edge at approximately $45-60 ~M_\odot$. We observe a suppression in the BBH merger rate when comparing the latest constraints against our GP model, by a factor of $10-60$ at $\sim60~M_\odot$, corresponding to the minimum of the mass gap. Additionally, we find evidence of a subpopulation of mergers populating the mass gap around $70 ~M_\odot$, which we argue is due to hierarchical mergers, as well as of a feature in the $40-50~ M_\odot$ range, albeit with low significance, which we attribute to the predicted PPISN build up. We discuss the astrophysical implications of this result in light of GW231123, a recently reported BBH merger with a total mass potentially above the PISN gap, suggesting the need for revised models of massive stellar evolution or alternative formation channels.