Searcharxiv⌕ Search
Home/Archive/arXiv

arXiv · 2610.14562

Quantitative Stability and Sample Complexity for Regularized Gromov-Wasserstein Distance

Pengtao Li·Xiaohui Chen
View original source ↗Read source document ↗Export metadata ↓

Abstract

The Gromov-Wasserstein (GW) distance provides a principled framework for comparing two probability distributions supported on different metric spaces by matching their individual intrinsic geometric structures. In practice, the GW distance is estimated from finite samples and often incorporates regularization for statistical and computational tractability. This motivates a quantitative understanding of the stability and estimation error of regularized GW distances. In this work, we study a general family of divergence regularized GW distances, including entropy and Tsallis regularizations over the full range of powers, from a nonasymptotic perspective. Specifically, we establish Lipschitz continuity of the value of regularized GW and H{ö}lder continuity of its optimal coupling under Wasserstein perturbations of the input probability measures. Moreover, we derive the parametric rate of convergence for the empirical regularized (2,2)-GW distance on compactly supported Euclidean spaces.

Explore related subjects

math.STstat.TH

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Pengtao Li, Xiaohui Chen. 2026-10-10. Quantitative Stability and Sample Complexity for Regularized Gromov-Wasserstein Distance. https://arxiv.org/abs/2610.14562

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Sparse estimation for the drift of high-dimensional Ornstein--Uhlenbeck processes with i.i.d. paths

We study sparsity-regularized maximum likelihood estimation for the drift parameter of high-dimensional non-stationary Ornstein--Uhlenbeck processes given repeated measurements of i.i.d. paths. In particular, we show that Lasso and Slope estimators can achieve the minimax optimal rate of convergence. We exhibit numerical experiments for sparse estimation methods and show their performance.

math.ST↗
arXiv2026

Minimax Wasserstein Testing under Differential Privacy

We study Wasserstein-$p$ goodness-of-fit and two-sample testing under differential privacy. For general compact metric spaces, we establish upper and lower bounds on the minimax separation radii in terms of covering numbers, matching up to logarithmic factors. On the hypercube $[0,1]^d$, these bounds match up to constant factors for two-sample testing in all dimensions and for goodness-of-fit testing whenever $d\neq 4p$, thereby closing logarithmic gaps in existing non-private results. At the critical case $(d,p)=(4,1)$, we further show that a logarithmic factor is unavoidable by establishing matching minimax bounds for non-private Wasserstein-$1$ goodness-of-fit testing under a bounded-density assumption. Finally, for unbounded domains, we establish minimax-optimal separation rates up to logarithmic factors under an exponential tail condition.

math.ST↗
arXiv2026

The Harmonic Entropy Estimator: Minimax Optimality and Semiparametric Efficiency for Infinite Alphabets

Shannon entropy estimation for discrete distributions is a classical nonparametric problem whose statistical limits are well characterized under finite support. Whether minimax and efficiency guarantees extend to infinite-support distributions remains open, in part because the expectation and variance of logarithm-transformed multinomial counts lack closed forms, forcing analyses to rely on bounding arguments rather than exact moment expressions. We address this obstacle with an algebraic framework built on an exact identity relating harmonic-transformed counts to their log-probabilities, yielding the harmonic entropy estimator. With this framework, we obtain two main results characterizing the statistical limits of this problem for distributions with sufficiently fast-decaying tails. First, for the class of distributions with at least quadratically decaying tails ($p_j \lesssim j^{-2}$), we establish that the $L_2$-minimax convergence rate is $1/n$, a rate anchored by estimator variance for both finite- and infinite-support distributions. Second, under the stronger condition $p_j = o(j^{-2})$, we demonstrate that the harmonic estimator is semiparametrically efficient: $\sqrt{n}[\hat{H}(\mathbf{X}_n) - H(X)]$ converges weakly to a normal distribution with variance achieving the semiparametric efficiency bound, $\mathrm{Var}[\log p(X)]$. Furthermore, the harmonic estimator is universally less biased than the plug-in estimator by circumventing the approximation error inherent to expectations of logarithm-transformed counts. More broadly, the harmonic approach may provide a foundation for developing new theory in entropy-based estimation, such as inference for mutual information under similar nonparametric regimes.

math.ST↗