{"id":"arxiv/2608.28135","url":"https://arxiv.org/abs/2608.28135","tags":["math.NA"],"score":0,"title":"Stochastic trace estimation for positive trace-class operators","domain":"math.NA","source":"arxiv","authors":["Zvonimir Bujanović","Luka Grubišić","Daniel Kressner","Hrvoje Olić"],"reviews":0,"subject":"math.NA","summary":"Implicit trace estimation aims to approximate the trace of a matrix or linear operator accessible only through matrix-vector or operator-vector products. In the matrix setting, the Girard-Hutchinson estimator typically requires $\\mathcal{O}(\\varepsilon^{-2})$ products to achieve accuracy $\\varepsilon$, while the variance-reduced Hutch++ algorithm reduces this sample complexity to $\\mathcal{O}(\\varepsilon^{-1})$ for positive semidefinite matrices. We develop infinite-dimensional analogues of these estimators for positive trace-class operators on separable Hilbert spaces. The idealized…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-08-28","reactions":{}}