{"id":"arxiv/2609.26501","url":"https://arxiv.org/abs/2609.26501","tags":["math.ST"],"score":0,"title":"Inference for H{ü}sler-Reiss block models","domain":"math.ST","source":"arxiv","authors":["Capel Alexandre","Demangeot Marine","Meyer Nicolas","Toulemonde Gwladys"],"reviews":0,"subject":"math.ST","summary":"Estimating the H{ü}sler-Reiss precision matrix is a fundamental problem for statistical inference in multivariate extremes. In high-dimensional settings, the number of unknown parameters grows quadratically with the dimension, making regularisation indispensable. Existing approaches regularise the estimation problem by exploiting sparsity. In this paper, we consider an alternative structural assumption, namely that the H{ü}sler-Reiss precision matrix is block-structured. To estimate such a matrix, we introduce a new regularisation framework based on a convex fusion penalty. By encouraging…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-09-22","reactions":{}}