Full classification of de Finetti type theorems for *-random variables in classical and free probability
Weihua Liu · 2025
Read Full classification of de Finetti type theorems for *-random variables in classical and free probability by Weihua Liu (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-2512-16170.txt.
Introduction
Weihua Liu classifies symmetry theorems in probability in a 2025 mathematics paper on arXiv. De Finetti-type theorems say that when a sequence of random variables has enough symmetry, it must arise from a simple random mixture. Versions exist in both classical probability and free probability, a noncommutative cousin used in random matrix theory.
Liu shows every nontrivial finite sequence carries a maximal symmetry of a special kind, described by a Woronowicz C star-algebra, the structure behind compact quantum groups. That yields a probability-based way to classify those quantum groups. He then classifies all de Finetti-type theorems for star-random variables under such symmetries. Only finitely many types exist, and they match the so-called easy quantum groups of Banica and Speicher. The result ties probability, algebra, and quantum symmetry into one tidy list.
Catalog record
| CREATOR | Weihua Liu |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 116 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2512-16170
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Full classification of de Finetti type theorems for *-random variables in classical and free probability Weihua Liu Abstract . Classical distributional symmetries can be described as invariance under the actions of semigroups (or groups) of matrix structures, and subsequently under the coactions of continuous functions on the matrix semigroups (or groups) generated by entry functions. By considering noncommutative entry functions on matrix structures, Woronowicz introduced corepresentations of compact quantum groups, namely Woronowicz’s C ∗ C^{*} -algebras (also known as compact matrix pseudogroups). We demonstrate that every nontrivial finite sequence of random variables admits a maximal distributional symmetry determined by a Woronowicz C ∗ C^{*} -algebra. This establishes a probabilistic framework for classifying compact quantum groups. Furthermore, we classify all de Finetti-type theorems for *-random variables that are invariant under distributional symmetries arising from compact matrix quantum groups in both classical and free probability settings. Our results show that only finitely many types of de Finetti theorems exist in these contexts, and the associated categories of (quantum) groups are the easy (quantum) groups introduced by Banica and Speicher. 1. Introduction The classical permutation groups S_{n} of n elements and orthogonal groups O_{n} on n -dimensional spaces have natural actions on n random variables x_{1},\cdots,x_{n} . For \sigma\in S_{n} , we can send x_{i} to x_{\sigma(i)} and for \sigma=(q_{i,j})_{i,j=1,\cdots,n}\in O_{n} , we can send x_{i} to \sum\limits_{k=1}^{n}q_{i,k}x_{k} . Random variables are said to be exchangeable (or orthogonal) if their joint distribution is invariant under the action of the corresponding group. For exchangeability, we have (x_{1},\cdots,x_{n})\stackrel{{\scriptstyle d}}{{=}}(x_{\sigma(1)},\cdots,x_{\sigma(n)}) for all \sigma\in S_{n} , and for orthogonality, we have (x_{1},\cdots,x_{n})\stackrel{{\scriptstyle d}}{{=}}(\sum\limits_{k=1}^{n}q_{1,k}x_{k},\cdots,\sum\limits_{k=1}^{n}q_{n,k}x_{k}) for all \sigma=(q_{i,j})_{i,j=1,\cdots,n}\in O_{n}. These two symmetries on finitely many random variables can be generalized to infinite sequences of random variables by assuming that all finite subsequences possess the given property. Note that exchangeability can also be expressed in terms of matrix operations as orthogonality by setting \sigma=(\delta_{\sigma(i),j})_{i,j=1,\cdots,n} . Based on this observation, we may introduce distributional symmetries for random variables beyond exchangeability and orthogonality by using matrix operations on elements from n\times n matrices. Given two n\times n matrices g=(g_{i,j}) and h=(h_{i,j}) over real numbers, if (x_{1},\cdots,x_{n})\stackrel{{\scriptstyle d}}{{=}}(\sum\limits_{k=1}^{n}g_{1,k}x_{k},\cdots,\sum\limits_{k=1}^{n}g_{n,k}x_{k}) and (x_{1},\cdots,x_{n})\stackrel{{\scriptstyle d}}{{=}}(\sum\limits_{k=1}^{n}h_{1,k}x_{k},\cdots,\sum\limits_{k=1}^{n}h_{n,k}x_{k}), then \begin{array}[]{rcl}(x_{1},\cdots,x_{n})&\stackrel{{\scriptstyle d}}{{=}}&(\sum\limits_{k=1}^{n}g_{1,k}x_{k},\cdots,\sum\limits_{k=1}^{n}g_{n,k}x_{k})\\ &\stackrel{{\scriptstyle d}}{{=}}&(\sum\limits_{l=1}^{n}h_{1,l}\sum\limits_{k=1}^{n}g_{l,k}x_{k},\cdots,\sum\limits_{l=1}^{n}h_{n,l}\sum\limits_{k=1}^{n}g_{l,k}x_{k})\\ &=&\left(\sum\limits_{k=1}^{n}(hg)_{1,k}x_{k},\cdots,\sum\limits_{k=1}^{n}(hg)_{n,k}x_{k}\right),\end{array} where (hg)_{i,j} is the (i,j) -th entry of hg . Thus, the joint distribution of (x_{1},\cdots,x_{n}) is invariant under the matrix operation of the semigroup generated by g and h . Two other important distributional symmetries, namely stationary and spreadability, can also be defined through matrix operations on semigroups of infinite-dimensional matrices generated by (\delta_{i,i+1})_{i,j\in N} and \{(\delta_{i,f(i)})_{i,j\in N}|f:\mathbb{N}\rightarrow\mathbb{N},f\text{ is strictly increasing }\} , respectively. The study of exchangeability dates back to 1930s, when de Finetti showed that infinite exchangeable sequences of random variables taking values in \{0,1\} are conditionally independent and identically distributed. This result was later extended to random variables taking values in compact Hausdorff spaces by Hewitt and Savage [ 13 ] . On the other hand, Freedman studied sequences of orthogonal random variables and showed that infinite sequences with this property are conditionally independent Gaussian Families [ 12 ] . For spreadability, Ryll-Nardzewski proved an extended de Finitte type theorem, infinite spreadable sequences of random variables are conditionally independent and identically distributed. Question 1.1 . Do we have de Finetti-type theorems for distributions other than Gaussian? Additionally, do we have de Finetti-type theorem for a given semigroup of (infinite-dimensional) matrices? Noncommutative probability as non-classical probability theory was introduced by von Neumann in the early thirties. Roughly speaking, in a very general setting of noncommutative probability, the set of random variables is replaced by a (noncommutative) algebra, referred to as a probability space, and the expectation is replaced by a linear functional. Free independence, introduced by Voiculescu to tackle the isomorphism problem of free group von Neumann algebras, serves as a quantum analogue to the independence relations in noncommutative probability. Notably, the only two unital universal independent relations for random variables, based on the natural requirements of associativity and a universal calculation rule for mixed moments, are classical independence and free independence [ 26 ] . Here, the universal independence relation means that the expectation functional on a probability space is completely determined by the expectation functionals on the sub-probability spaces that satisfy the given relation. The independence relation in noncommutative probability is more complicated than in classical probability. Indeed, if we do not require the universal relation to be unital, one additional type of independence emerges, known as Boolean independence [ 23 ] . If we do not require associativity, we will encounter infinitely many independence relations that satisfy a common weak condition known as conditional independence [ 14 ] . In addition, we are able to construct infinite exchangeable sequences of noncommutative random variables to satisfy infinitely many symmetric universal independence relation [ 14 ] . Therefore, in noncommutative probability, classical exchangeability cannot determine a universal independence relation conditionally as they do in the classical sense. Even if we strengthen exchangeability to orthogonality, we still do not obtain a de Finetti-type characterization for any universal independence relation. For example, an infinite sequence of q-Gaussian random variables is orthogonal but does not satisfy any common universal independence relation [ 6 ] . In noncommutative probability, a de Finetti-type theorem for exchangeability is as follows: if the probability space generated by exchangeable sequences of random variables behaves well, such as in a W*-probability space with a faithful normal state, then the random variables are conditionally independent [ 16 ] . To characterize a given universal independence relation, such as freeness, it requires more distributional symmetries than classical exchangeability. In 1987, Woronowicz introduced a notion of compact matrix quantum groups, which provide symmetries for noncommutative structures [ 36 ] . Given a Lie group G\subset M_{n}(\mathbb{C}) , by Weierstrass theorem, the algebra of continuous function C(G) on G is generated by \{u_{i,j}|i,j,=1,\cdots,n\} , where u_{i,j}(g)=g_{i,j} for g=(g_{i,j})\in M_{n}(\mathbb{C}) . The map \Delta(u_{i,j})=\sum\limits_{k=1}^{n}u_{i,k}\otimes u_{k,j} defines a homomorphism from C(G) to C(G)\otimes C(G) , which is called the comultiplication. By considering noncommutative C^{*} -algebras \mathcal{A} generated by elements \{u_{i,j}|i,j,=1,\cdots,n\} with well-defined comultiplication \Delta(u_{i,j})=\sum\limits_{k=1}^{n}u_{i,k}\otimes u_{k,j} from \mathcal{A} to \mathcal{A}\otimes_{\min}\mathcal{A} and some other conditions, one gets the notion of a compact matrix quantum group. In this framework, Wang introduced the quantum analogues \mathcal{A}_{s}(n) and \mathcal{A}_{o}(n) of permutation groups and orthogonal groups, respectively [ 34 ] . The distributional symmetries associated with classical Lie groups G\subset M_{n}(\mathbb{C}) can be extended via the entry functions {u_{i,j}\mid i,j=1,\ldots,n} . Consequently, the distributional symmetries generated by \mathcal{A}_{s}(n) and \mathcal{A}_{o}(n) are applied in noncommutative probability and are referred to as quantum exchangeability and quantum orthogonality. Köstler and Speicher found quantum exchangeability can be used to characterize freeness. In fact, they proved that an infinite sequence of noncommutative selfadjoint random variables in W^{*} -probability space with a faithful normal state is quantum exchangeable if and only if the random variables are identically distributed and free with respect to the conditional expectation onto their tail algebra [ 15 ] . Later, the free analogue of de Finetti theorem for non selfadjoint random variables is proved by Curran [ 8 ] and the free analogue of Freedman’s result associated with Wang’s quantum orthogonal groups is also proved by Curran [ 9 ] . Here comes a