Convergence rate of entropy-regularized multi-marginal optimal transport costs

Luca Nenna,Paul Pegon

DOI: https://doi.org/10.4153/S0008414X24000257

2024-04-09

Abstract:We investigate the convergence rate of multi-marginal optimal transport costs that are regularized with the Boltzmann-Shannon entropy, as the noise parameter $\varepsilon$ tends to $0$. We establish lower and upper bounds on the difference with the unregularized cost of the form $C\varepsilon\log(1/\varepsilon)+O(\varepsilon)$ for some explicit dimensional constants $C$ depending on the marginals and on the ground cost, but not on the optimal transport plans themselves. Upper bounds are obtained for Lipschitz costs or locally semi-concave costs for a finer estimate, and lower bounds for $\mathscr{C}^2$ costs satisfying some signature condition on the mixed second derivatives that may include degenerate costs, thus generalizing results previously in the two marginals case and for non-degenerate costs. We obtain in particular matching bounds in some typical situations where the optimal plan is deterministic.

Optimization and Control,Analysis of PDEs,Functional Analysis

What problem does this paper attempt to address?

This paper discusses the convergence rate issue of entropy regularized multi-marginal optimal transport (MOT) problem. In probability measure theory and optimization theory, multi-marginal optimal transport is a problem of finding the optimal way to map multiple probability distributions from one space to another, while entropy regularization is a method to simplify calculations and improve algorithm stability by introducing an entropy term.

As the noise parameter ε tends to 0, the paper studies the difference between the cost of entropy regularized multi-marginal optimal transport (MOTε) and the cost of unregularized transport (MOT 0). The authors establish upper and lower bound estimates, which depend on the local properties of the cost function (such as Lipschitz continuity or local convexity) and the characteristic conditions of the mixed second derivatives of the cost function. For certain specific cases, they provide matching upper and lower bounds that do not depend on the specific form of the optimal transport plan but rather on the dimensions of the marginal distributions.

In the case of two marginals, existing literature provides upper bounds on the convergence rate, but these bounds may depend on the quantization dimension of the optimal solution. The paper extends these techniques to handle degenerate cost functions and provides smaller upper bounds that only depend on the dimensions of the marginal distributions rather than the characteristics of the optimal solution. In addition, they establish lower bounds by utilizing the signature conditions of the mixed second derivatives of the cost function to estimate integrals, thereby improving the results of previous work.

The structure of the paper includes an introduction, preliminary knowledge, upper bound theorem, lower bound theorem, and some examples that demonstrate how to obtain matching boundaries in different scenarios. These results are important for understanding the behavior of entropy regularization in multi-marginal optimal transport and for numerical approximation in practical applications.