{"id":"arxiv/2610.27442","url":"https://arxiv.org/abs/2610.27442","tags":["stat.CO","math.PR","math.ST"],"score":0,"title":"Geometry-informed importance sampling for rare events","domain":"stat.CO","source":"arxiv","authors":["Qinyi Wang","Anya Katsevich","Shanyin Tong"],"reviews":0,"subject":"stat.CO","summary":"We propose GeoIS, a new importance sampling method for estimating rare event probabilities for an underlying Gaussian distribution that exploits asymptotic theory and local event geometry. The proposal distribution is a mixture of two components. The first component draws a Gaussian along the boundary near the instanton and adds to it an exponentially distributed shift into the rare event. This component is derived from the known asymptotic approximation to the optimal zero-variance proposal distribution. The second mixture component is designed to cover missing event mass due to…","category":"research","comments":0,"disputed":false,"division":"stat","mentions":0,"discussed":0,"published":"2026-10-20","reactions":{}}