On the Kolmogorov-Feller weak law of large numbers for Fréchet mean on non-compact symmetric spaces
Jongmin Lee, Sungkyu Jung · 2025
Read On the Kolmogorov-Feller weak law of large numbers for Fréchet mean on non-compact symmetric spaces by Jongmin Lee, Sungkyu Jung (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-02074.txt.
Introduction
Jongmin Lee and Sungkyu Jung prove a weak law of large numbers for Frechet means on non-compact symmetric spaces. A Frechet mean generalizes an average to curved spaces by minimizing squared distances. The Kolmogorov-Feller weak law governs when sample means settle near the true mean. Their data need not share one distribution, extending past identical sampling results.
Non-compact symmetric spaces include key examples like positive definite matrices, central in statistics and engineering. Product spaces of such examples are covered as well. Independent but non-identical data are handled throughout the proof. The result strengthens limit theory for data living on curved domains. It supports inference methods that average matrix valued observations.
Catalog record
| CREATOR | Jongmin Lee, Sungkyu Jung |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 34 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2509-02074
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
On the Kolmogorov-Feller weak law of large numbers for Fréchet mean on non-compact symmetric spaces Jongmin Lee, Sungkyu Jung Abstract We establish the Kolmogorov-Feller weak law of large numbers for Fréchet mean on non-compact symmetric spaces under certain regularity conditions. Our results accommodate non-identically distributed random variables and are accompanied by an illustrative example that demonstrates their applicability to the space of symmetric positive-definite matrices. keywords [class=MSC] 60F05 , 62G35 keywords Fréchet mean , Law of large numbers , Probability on manifolds , Robust statistics † † orcid: 0000-0003-1723-4615 † † orcid: 0000-0002-6023-8956 and ?? Department of Statistics, Pusan National University, ?? ?? Department of Statistics and Institute for Data Innovation in Science, Seoul National University, ?? 1 Introduction For a sequence of independent and identically distributed real-valued random variables, (X_{i})_{i=1}^{\infty} , the celebrated Kolmogorov-Feller weak law of large numbers (see, e.g., ( Feller 1971 , pp. 234-236) , ( Naderi et al. 2019 ) ) says that \mbox{if}\penalty\ n\mathbb{P}(|X_{1}|>n)\underset{n\to\infty}{\to}0,\penalty\ \mbox{then}\penalty\ \frac{\sum_{i=1}^{n}X_{i}-n\mathbb{E}[X_{1}1_{|X_{1}|\leq n}]}{n}\overset{\mathbb{P}}{\to}0\penalty\ \mbox{as}\penalty\ n\to\infty. (1) The result in ( 1 ) can be viewed as an extension of the weak law of large numbers for random variables whose distributions are possibly too heavy-tailed to admit a finite expectation (i.e., \mathbb{E}|X_{1}|=\infty ), yet still exhibits moderate tail decay in the sense that n\mathbb{P}(|X_{1}|>n)\underset{n\to\infty}{\to}0 , which is called the Kolmogorov-Feller tail condition. The Kolmogorov-Feller weak law of large numbers has been extended to various settings. For example, Stoica and Li 2010 investigates the case where (X_{i})_{i=1}^{\infty} is assumed to be exchangeable, and a conditional version of the extended Kolmogorov-Feller weak law of large numbers is established by Yuan and Hu 2015 . Moreover, Naderi et al. 2019 studies a version in which a certain type of dependence is imposed on (X_{i})_{i=1}^{\infty} . While these studies provide solid foundations in the context of the real line, their extension to more general spaces is not straightforward. Beyond the real line, given a metric space (M,d) and an n -tuple of observations (x_{1},x_{2},\ldots x_{n})\in M^{n} , it is natural to consider the sample Fréchet mean for central tendency, defined as \mbox{argmin}_{m\in M}\sum_{i=1}^{n}d^{2}(x_{i},m), which generalizes the Euclidean sample mean. While theoretical properties for the sample Fréchet mean, such as the law of large numbers and central limit theorem, have been established in Riemannian manifolds or general metric spaces (see, e.g., Ziezold 1977 ; Sturm 2003 ; Bhattacharya and Patrangenaru 2003 ; Barden, Le and Owen 2013 ; Schötz 2022 ; Evans and Jaffe 2024 ; Schötz 2025 ), their validity under heavy-tailed distributions that may lack a finite first moment remain unexamined thus far. To address these gaps, we generalize the Kolmogorov-Feller weak law of large numbers to a specific type of metric spaces. The obstacles for the generalization of ( 1 ) are two-fold: (i) the possibility of non-uniqueness of Fréchet means (notably, compact manifolds such as S^{k} may suffer from the non-uniqueness of Fréchet means), and (ii) the difficulty for evaluation of \mathbb{E}[X_{1}1_{|X_{1}|\leq n}] in the right-hand side of ( 1 ). The solution for (i) is to restrict the domain to symmetric spaces of non-compact type, where the sample Fréchet mean is known to be unique, albeit the population Fréchet mean may not exist. The remedy for (ii) is to impose a symmetry condition on the underlying probability distribution ( \mathbb{P}_{X} ), as defined precisely in the next section. A class of metric spaces known as symmetric spaces of non-compact type (hereafter referred to as non-compact symmetric spaces) encompasses a broad class of metric spaces commonly encountered in statistical data analysis. Representative examples of such spaces include Euclidean space \mathbb{R}^{k} , hyperbolic space \mathbb{H}^{k} , the space of symmetric positive-definite matrices (\text{Sym}^{+}(k),\penalty\ d_{\text{\tiny AI}}) , equipped with the affine-invariant metric,and their product spaces. Given a symmetric distribution \mathbb{P}_{X} on a non-compact symmetric space (M,d) , let \mu\in M be a center of symmetry of \mathbb{P}_{X} , which we take as the reference point. To generalize ( 1 ) to our setting, the strategy is to replace |X_{1}| with d(X_{1},\mu) , and the Euclidean sample mean \sum_{i=1}^{n}X_{i}/n with the sample Fréchet mean. We emphasize that our main results, which are entirely new in the fields of probability on manifolds and geometric statistics, provide sufficient conditions under which the weak law of large numbers holds, even for highly heavy-tailed distributions. The remainder of this article is organized as follows. The next section presents the necessary preliminaries and the main results, accompanied by illustrative examples. Section 3 provides auxiliary results and proofs, some of which are deferred to the appendix for better readability. 2 Main results A simply connected and geodesically complete Riemannian manifold is said to be a Hadamard manifold if it has non-positive curvatures everywhere. A connected, complete and smooth Riemannian manifold (M,d) is said to be a symmetric space if for any \nu\in M , there exists a unique isometry s_{\nu} (called geodesic symmetry) that fixes \nu and reverses all geodesics through \nu ; that is, for each x\in M , s_{\nu}(x)=\mbox{Exp}_{\nu}\{-\mbox{Log}_{\nu}(x)\} , provided it is well-defined. It is known that any non-compact symmetric space is considered as simply connected and has non-positive curvatures everywhere ( Helgason 1979 ) ; thus, it is Hadamard. Moreover, for M that is Hadamard and for each p\in M , the Riemannian logarithmic map at p is a diffeomorphism between M and T_{p}M\cong\mathbb{R}^{k} by the Cartan-Hadamard theorem. The notion of geodesic symmetry of a probability distribution \mathbb{P}_{X} is defined as follows. Definition 1 . Given a symmetric space (M,d) and a probability distribution \mathbb{P}_{X} on the space, if there exists a \mu\in M such that the distribution of s_{\mu}(X) , \mathbb{P}_{s_{\mu}(X)} , equals \mathbb{P}_{X} , then \mathbb{P}_{X} is said to be geodesically symmetric about \mu . A natural question arising from the definition is that for a given distribution \mathbb{P}_{X} , if the distribution is geodesically symmetric about \mu_{1},\mu_{2}\in M , then is it true that \mu_{1}=\mu_{2} ? The answer to the question is ‘yes’ under the following non-compactness condition. Proposition 1 . Let (M,d) be a non-compact symmetric space. Suppose that \mathbb{P}_{X} is geodesically symmetric about both \mu_{1} and \mu_{2}\in M . Then, \mu_{1}=\mu_{2} . Fig 1: An illustration of the transvection from x to s_{\mu_{2}}\circ s_{\mu_{1}}(x) . Proof of Proposition 1 . Suppose, for contradiction, that \mu_{1}\neq\mu_{2} . Let T=s_{\mu_{1}}\circ s_{\mu_{2}} , and \gamma be the geodesic from \mu_{1} to \mu_{2} . Denote \ell=2d(\mu_{1},\mu_{2})>0 . Then, note that T is a translation-type isometry of length \ell along the direction of \gamma , more precisely, referred to as a transvection. (For an illustration of the transvection, see Figure 1 .) It can be shown that for any x\in M , d(x,s_{\mu_{2}}\circ s_{\mu_{1}}(x))\geq 2d(\mu_{1},\mu_{2})=\ell by an application of ( Sturm 2003 , Corollary 2.5) . It implies that for any x\in M and for any m\in\mathbb{N} , d(x,T^{m}(x))\geq\ell m . Thus, the isometry T , when repeatedly applied, pushes any point to infinity. For given bounded sets K and A , define a constant C_{K,A}=\sup_{k\in K,a\in A}d(k,a)\in[0,\infty) . There exists an integer m such that \ell m>C_{K,A} . For any k\in K and for any a\in A , by triangle inequality, d(k,T^{m}(a))\geq d(a,T^{m}(a))-d(k,a)\geq\ell m-C_{K,A}>0, which implies that K\cap T^{m}(A)=\phi . That is, for any bounded sets