2025 workshop program
From LEGEND to Binary Black Hole: A Rare Event Journey Across Physics
The work
Abstract
From deep underground neutrino detectors to the far reaches of the universe, modern physics increasingly relies on extracting meaning from rare events. Whether searching for neutrinoless double-beta decay with the LEGEND experiment or reconstructing the formation history of black holes from gravitational-wave detections, we face a shared challenge: performing inference under extreme data scarcity and limited simulation budgets.
In this talk, I present a unified framework for rare event surrogate modeling, developed initially for optimizing the neutron shielding design of LEGEND using limited Monte Carlo simulations. The resulting method—RESuM—combines machine learning with probabilistic modeling to accelerate design optimization where signal rates are vanishingly low. Building on this foundation, I introduce RESOLVE, a rare event surrogate likelihood model for gravitational-wave paleontology. In this setting, population synthesis tools like COMPAS produce very few observable binary black hole mergers, making likelihood estimation computationally expensive and statistically fragile. RESOLVE integrates polynomial chaos expansion with Bayesian MCMC to construct a statistically sound surrogate likelihood, achieving proper coverage and enabling direct parameter estimation from LIGO observations.
By drawing a methodological throughline from LEGEND to LIGO, I demonstrate how rare event surrogate modeling provides a powerful, domain-agnostic approach to extracting physics from the scarcest of signals—helping us understand not only the structure of shielding materials but also the stellar pathways that shaped the universe.