Full classification of de Finetti type theorems for *-random variables in classical and free probability

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 $C^$-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 $C^$-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.

💡 Research Summary

The paper provides a complete classification of de Finetti‑type theorems for *‑random variables in both classical and free probability. The authors begin by observing that classical distributional symmetries such as exchangeability (invariance under the permutation group Sₙ) and orthogonal invariance (invariance under the orthogonal group Oₙ) can be expressed as matrix actions on finite tuples of random variables. Extending this viewpoint to infinite sequences leads to the classical de Finetti, Ryll‑Nardzewski, and Freedman theorems.

In the non‑commutative setting, the usual exchangeability does not single out a universal independence relation (e.g., freeness). To capture richer symmetries, the paper adopts Woronowicz’s compact matrix quantum groups, i.e. C∗‑algebras generated by non‑commuting “entry functions” u_{ij} equipped with a comultiplication Δ(u_{ij}) = Σ_k u_{ik}⊗u_{kj}. The key observation is that every non‑trivial finite family of *‑random variables possesses a maximal distributional symmetry that is uniquely determined by a compact quantum group.

The classification problem is then reduced to identifying which compact (quantum) groups give rise to non‑trivial de Finetti theorems. The authors rely on the theory of “easy” groups and easy quantum groups introduced by Banica and Speicher. Easy groups are described by categories of set‑partitions; they interpolate between the permutation group Sₙ and the orthogonal group Oₙ. Their quantum analogues interpolate between the quantum permutation group A_s(n) and the quantum orthogonal group A_o(n). Prior work shows there are six classical easy groups and seven easy quantum groups in these intervals.

The main contribution of the paper is to show that, when one works with *‑random variables (i.e., possibly non‑self‑adjoint complex‑valued non‑commutative variables), the maximal symmetries that yield de Finetti theorems are precisely the “unitary” families of easy groups. In the free case the relevant quantum groups are

C(S⁺ₙ), C(O⁺ₙ), C(B⁺{s,n}), C(H⁺{s,n}), C(B⁺n), C(H⁺{m,n}) (m ≥ 3), C(H⁺_{0,n}), C(H′⁺_n), C(U⁺ₙ),

while in the classical case the corresponding groups are

C(Sₙ), C(Oₙ), C(B^{s}n), C(H^{s}n), C(B_n), C(H{m,n}) (m ≥ 3), C(H{0,n}), C(Uₙ).

For each of these groups the authors give an explicit description of the associated distributional property (e.g., “identically distributed shifted orthogonal elements”, “R‑diagonal elements”, “circular elements”, etc.). Theorem 1 states that for a finite tuple (x₁,…,xₙ) of *‑random variables which are freely (or classically) independent, the tuple is invariant under the coaction of the appropriate quantum (or classical) group if and only if the variables are identically distributed with the prescribed type of distribution.

Theorem 2 extends the result to infinite sequences. Given a W∗‑probability space (M, φ) generated by an infinite family (x_i) and assuming the joint distribution is invariant under a sequence of compact quantum groups A = (Aₙ)ₙ (chosen from the list above), there exists a φ‑preserving conditional expectation E : M → B onto a sub‑algebra B such that the sequence is conditionally free (or conditionally independent in the classical case) over B and each x_i has the same B‑valued distribution of the prescribed type. The proof uses Haar states on the quantum groups, averaging arguments, and the identification of the tail algebra as the fixed‑point algebra of the coaction.

The authors also introduce the notion of a distributional symmetry set DS(X, φ), the collection of all compact quantum group sequences under which a given infinite family X is invariant. They show that the maximal elements of DS are exactly the easy (quantum) groups listed above; any further symmetry would destroy the de Finetti representation.

In summary, the paper establishes that all de Finetti‑type theorems for *‑random variables in both classical and free probability are governed by the easy (quantum) groups lying between the permutation and unitary families. This yields a finite, explicit catalogue of possible symmetries and associated conditional independence structures, thereby completing the classification program initiated by Banica, Speicher, Curran, and others. The results deepen the connection between non‑commutative symmetries, partition categories, and probabilistic independence, and they open the way for applications in quantum information, random matrix theory, and non‑commutative harmonic analysis.