The g-Function: An ML-based Fragility Operator

Multi-hazard and multi-event fragility modeling has advanced rapidly in recent years, yet its widespread adoption remains limited by the computational burden of successive nonlinear simulations required to explicitly capture damage accumulation. Recent studies on earthquake–tsunami, fire-following-earthquake, and mainshock–aftershock sequences have revealed a recurring probabilistic structure governing joint fragility responses. This paper formalizes that structure through a unified g-function, a hazard-agnostic probabilistic operator that synthesizes multi-dimensional fragility surfaces directly from single-event fragility curves. Operating entirely in probability space, the g-function maps marginal exceedance probabilities to joint exceedance surfaces without direct dependence on hazard-specific intensity measures or physics. The formulation is demonstrated across three distinct hazard classes—earthquake–tsunami, fire-following-earthquake, and mainshock–aftershock—showing consistently high agreement with simulation-based benchmarks. The results establish the g-function as a general, scalable framework for multi-hazard fragility modeling, enabling efficient portfolio-level and community-scale resilience assessments without reliance on high-performance computing.