Id: arxiv/2611.22492 Url: https://arxiv.org/abs/2611.22492 Tags: math.DG Score: 0 Title: Equivariant Flow Matching on Homogeneous Spaces Domain: math.DG Source: arxiv Authors: Finn M. Sherry, Gijs Bellaard Reviews: 0 Subject: math.DG Category: research Comments: 0 Disputed: false Division: math Mentions: 0 Discussed: 0 Published: 2026-11-17 Page: https://delphi.science/r/arxiv/2611.22492 Flow matching enables the training of networks that can flow samples from an initial to a target distribution. The network is regressed against simple vector fields that are conditioned on end points in the target distribution. In the Euclidean case, these conditional vector fields induce flows along straight lines. Various generalisations have been proposed to perform flow matching on spaces where straight lines are not well-defined, such as Riemannian manifolds and Lie groups. Riemannian flow matching applies to many domains, but typically requires expensive simulation, while flow matching…