English

On the Convergence Rate of Sinkhorn's Algorithm

Optimization and Control 2025-04-08 v2 Analysis of PDEs Probability

Abstract

We study Sinkhorn's algorithm for solving the entropically regularized optimal transport problem. Its iterate πt\pi_{t} is shown to satisfy H(πtπ)+H(ππt)=O(t1)H(\pi_{t}|\pi_{*})+H(\pi_{*}|\pi_{t})=O(t^{-1}) where HH denotes relative entropy and π\pi_{*} the optimal coupling. This holds for a large class of cost functions and marginals, including quadratic cost with subgaussian marginals. We also obtain the rate O(t1)O(t^{-1}) for the dual suboptimality and O(t2)O(t^{-2}) for the marginal entropies. More precisely, we derive non-asymptotic bounds, and in contrast to previous results on linear convergence that are limited to bounded costs, our estimates do not deteriorate exponentially with the regularization parameter. We also obtain a stability result for π\pi_{*} as a function of the marginals, quantified in relative entropy.

Keywords

Cite

@article{arxiv.2212.06000,
  title  = {On the Convergence Rate of Sinkhorn's Algorithm},
  author = {Promit Ghosal and Marcel Nutz},
  journal= {arXiv preprint arXiv:2212.06000},
  year   = {2025}
}

Comments

Forthcoming in 'Mathematics of Operations Research'

R2 v1 2026-06-28T07:31:17.823Z