Moment Dilations and Functional Calculus for Random Operators
James Tian · 2025
Read Moment Dilations and Functional Calculus for Random Operators by James Tian (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-2509-06309.txt.
Introduction
Random tuples of operators hide a tidy moment structure, reveals James Tian in a 2025 paper. Averaging word moments over chance produces a positive kernel invisible to pointwise bounds. A single shift-positivity condition controls that kernel, and when it holds, the kernel admits a Cuntz dilation, realizing all mean-square effects inside one canonical isometric model.
Consequences include a mean-square free von Neumann inequality and a free functional calculus for random tuples. Tian then defines a ladder of higher-order defects of the kernel and proves their positivity matches a nested chain of projections inside one dilation. The decomposition splits the tuple into dissipative and unitary parts, Wold-style, with a curvature invariant measuring what never dissipates.
Catalog record
| CREATOR | James Tian |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 33 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2509-06309
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Moment Dilations and Functional Calculus for Random Operators James Tian Abstract . We develop a dilation theory for tuples of random operators on a Hilbert space. Their joint distribution defines a moment kernel by expectation of operator products. We prove that this kernel admits a dilation to a Cuntz family of isometries precisely when the associated shifted kernel is dominated in the positive-definite order. As consequences, we establish a mean-square version of von Neumann’s inequality and construct a free functional calculus for random operators, extending from polynomials to the free disk algebra and the free Hardy algebra. These results extend classical dilation theory into a probabilistic setting and provide new tools for analyzing noncommutative random systems. 2020 Mathematics Subject Classification Primary: 47A20; secondary: 46L53, 47A60, 47A63, 47B80 1. Introduction Dilation theory is one of the central tools in operator theory. The classical Sz.-Nagy theorem [ SNFBK10 ] shows that every contraction on a Hilbert space can be dilated to a unitary operator, and many important consequences follow from this. Later, Frazho, Bunce, and Popescu developed multivariable versions, proving that tuples of operators satisfying certain contractive conditions can be dilated to the universal Cuntz family of isometries with orthogonal ranges [ Fra82 , Bun84 , Pop89 ] . These results give a way to study complicated operators by embedding them into larger spaces where the structure is simpler and more rigid. Another perspective on dilation comes from the theory of positive definite kernels [ Aro50 , Sti55 , Arv10 , SNFBK10 , Ped58 , Sza21 ] . Given a family of operators, one can associate kernels that record their algebraic relations. Positivity of these kernels is closely tied to the existence of dilations, and this connection goes back to early work of Aronszajn and Pedrick on reproducing kernel Hilbert spaces. In this paper we consider the case where the operators themselves are random. Suppose A=\left(A_{1},\dots,A_{d}\right) is a d -tuple of random operators on a Hilbert space H . For each pair of words \alpha,\beta in the free semigroup \mathbb{F}^{+}_{d} , we form the moment kernel K\left(\alpha,\beta\right)=\mathbb{E}[A^{\alpha}\left(A^{\beta}\right)^{*}] . This kernel captures the averaged behavior of the random tuple. It is always positive definite, but the natural question is: when does K come from compressing a universal Cuntz family, as in the deterministic dilation theorems? Our first main result (Theorem 1 ) answers this question. We show that K admits such a dilation if and only if it dominates a shifted version K_{\Sigma} in the sense of positive definite kernels. Equivalently, the difference K-K_{\Sigma} must itself be positive definite. This gives a precise condition for when the random tuple has a dilation to a Cuntz family acting on a larger Hilbert space. Once the dilation is established, further consequences follow. We obtain a mean-square version of von Neumann’s inequality, which controls random polynomials in A by the corresponding universal model. We then build a mean-square free functional calculus. Starting from noncommutative polynomials, the construction extends first to the free disk algebra and then, using radial limits, to the free Hardy algebra [ Pop91 , DP98 , Pop10 ] . The result is a functional calculus for random operator tuples that behaves in close analogy with the deterministic theory. The point of this work is that dilation methods, originally developed for single operators and later generalized to noncommuting tuples, also extend naturally to random operators once one considers their moment kernels. The theorems here provide the basic tools: a dilation theorem, an inequality of von Neumann type, and a functional calculus. There is a formal resemblance to operator-valued free probability [ Voi85 , Voi91 , Spe98 , NS06 ] , where distributions are described by operator-valued moment functionals. Our construction differs in that we work with kernels defined by classical expectation over random operators, rather than with conditional expectations on noncommutative probability spaces. The analogy highlights the common role of positivity of moment data in both settings, even though the underlying notions of probability are distinct. In Section 2 , we introduce the moment kernel K for a d -tuple of random operators and prove our main dilation result (Theorem 1 ), which establishes the equivalence between the kernel inequality K_{\Sigma}\leq K and the existence of a dilation involving a Cuntz family. We then derive a mean-square von Neumann inequality (Corollary 4 ) for noncommutative polynomials as a direct consequence. Section 3 uses this inequality to construct the mean-square functional calculus. We first define the map for the free disk algebra \mathcal{A}_{d} (Theorem 8 ) and then, using radial limits, extend it to the free Hardy algebra F^{\infty}_{d} (Theorem 10 ). 2. Dilation of Free Moment Kernels The starting point of our analysis is the moment kernel associated with a tuple of random operators. In classical dilation theory, one studies a fixed operator tuple and encodes its algebraic relations through powers and products. In the random setting, the natural object is obtained by averaging such products, leading to a positive definite kernel on the free semigroup. This kernel captures the mean-square behavior of the random tuple and provides the right entry point for dilation. The goal of this section is to characterize precisely when such kernels admit dilations to Cuntz families of isometries, and to establish the equivalence between this dilation property and a positivity condition on a shifted version of the kernel. Fix d\in\mathbb{N} . Let \mathbb{F}{}^{+}_{d} be the free semigroup on generators \left\{1,\dots,d\right\} , with neutral element \emptyset . For a word \alpha=i_{1}\dots i_{k}\in\mathbb{F}^{+}_{d} , write \tilde{\alpha}=i_{k}\cdots i_{1} for the reversed word. Given any d -tuple of operators S=\left(S_{1},\dots,S_{d}\right) on a Hilbert space H , define S^{\alpha}\coloneqq S_{i_{1}}\cdots S_{i_{k}},\quad S^{\emptyset}\coloneqq I_{H}. Then for every \alpha\in\mathbb{F}^{+}_{d} , \left(S^{\alpha}\right)^{*}=\left(S^{*}\right)^{\tilde{\alpha}}, where S^{*}=\left(S^{*}_{1},\dots,S^{*}_{d}\right) . Let \left(\Omega,\mathbb{P}\right) be a probability space and A=\left(A_{1},\dots,A_{d}\right) a d -tuple of (strongly measurable) random operators A_{i}\colon\Omega\rightarrow\mathcal{L}\left(H\right) on a Hilbert space H , and assume \mathbb{E}\left[\left\|A^{\alpha}\right\|^{2}\right]<\infty,\quad\alpha\in\mathbb{F}^{+}_{d}. Define the moment kernel K\colon\mathbb{F}^{+}_{d}\times\mathbb{F}^{+}_{d}\rightarrow\mathcal{L}\left(H\right) by (2.1) K\left(\alpha,\beta\right)\coloneqq\mathbb{E}\left[A^{\alpha}\left(A^{\beta}\right)^{*}\right],\quad\alpha,\beta\in\mathbb{F}^{+}_{d}. Then K is positive definite (p.d.): for every finite family \left\{\left(\alpha_{j},u_{j}\right)\right\}^{N}_{j=1}\subset\mathbb{F}^{+}_{d}\times H , \displaystyle\sum\nolimits^{N}_{i,j=1}\left\langle u_{i},K\left(\alpha_{i},\alpha_{j}\right)u_{j}\right\rangle_{H} \displaystyle=\sum\nolimits^{N}_{i,j=1}\mathbb{E}\left[\left\langle u_{i},A^{\alpha_{i}}\left(A^{\alpha_{j}}\right)^{*}u_{j}\right\rangle_{H}\right] \displaystyle=\mathbb{E}\left[\left\|\sum\nolimits_{i}\left(A^{\alpha_{i}}\right)^{*}u_{i}\right\|^{2}_{H}\right]\geq 0. Define the kernel K_{\Sigma}\left(\alpha,\beta\right)\coloneqq\sum^{d}_{i=1}K\left(\alpha i,\beta i\right),\quad\alpha,\beta\in\mathbb{F}^{+}_{d}. Theorem 1 (Free moment dilation) . For the moment kernel K in ( 2.1 ), the following are equivalent: (1) There exists a Hilbert space \mathcal{K} , a Cuntz family S=\left(S_{1},\dots,S_{d}\right) on \mathcal{K} (i.e., S^{*}_{i}S_{j}=\delta_{i,j}I_{\mathcal{K}} and \sum^{d}_{i=1}S_{i}S^{*}_{i}=I_{\mathcal{K}} ), a projection P\in\mathcal{L}\left(\mathcal{K}\right) , and an isometry W\colon H\rightarrow\mathcal{K} such that (2.2) K\left(\alpha,\beta\right)=W^{*}S^{\alpha}P\left(S^{\beta}\right)^{*}W,\quad\alpha,\beta\in\mathbb{F}^{+}_{d}; and, (2.3) \sum\nolimits^{d}_{i=1}S_{i}PS^{*}_{i}\leq P\;\text{on }M\coloneqq\overline{span}\left\{\left(S^{\gamma}\right)^{*}WH:\gamma\in\mathbb{F}^{+}_{d}\right\} (2) The kernel K_{\Sigma} is dominated by K in the p.d. order: (2.4) K_{\Sigma}\leq K. (That is, K-K_{\Sigma} is p.d.) Moreover, K_{\Sigma}=K if and only if (2.5) K\left(\alpha,\beta\right)=W^{*}S^{\alpha}\left(S^{\beta}\right)^{*}W,\quad\alpha,\beta\in\mathbb{F}^{+}_{d}. Proof. ( 1 ) \Rightarrow ( 2 ). Fix a finite family \left\{\left(\alpha_{j},u_{j}\right)\right\}^{N}_{j=1}\subset\mathbb{F}^{+}_{d}\times H and set y\coloneqq\sum\nolimits^{N}_{j=1}\left(S^{\alpha_{j}}\right)^{*}Wu_{j}\in M. Then (using S^{\alpha i}=S^{\alpha}S_{i} and ( 2.2 )) \displaystyle\sum\nolimits^{N}_{j,k=1}\left\langle u_{j},K_{\Sigma}\left(\alpha_{j},\alpha_{k}\right)u_{k}\right\rangle_{H} \displaystyle= \displaystyle\sum\nolimits^{d}_{i=1}\sum\nolimits^{N}_{j,k=1}\left\langle u_{j},K\left(\alpha_{j}i,\alpha_{k}i\right)u_{k}\right\rangle_{H} \displaystyle= \displaystyle\sum\nolimits^{d}_{i=1}\sum\nolimits^{N}_{j,k=1}\left\langle\left(S^{\alpha_{j}}\right)^{*}Wu_{j},\left(S_{i}PS^{*}_{i}\right)\left(S^{\alpha_{k}}\right)^{*}Wu_{k}\right\rangle_{\mathcal{K}} \displaystyle= \displaystyle\left\langle y,\left(\sum\nolimits^{d}_{i=1}S_{i}PS^{*}_{i}\right)y\right\rangle_{\mathcal{K}}. Similarly, using ( 2.2 ), \sum\nolimits^{N}_{j,k=1}\left\langle u_{j},K\left(\alpha_{j},\alpha_{k}\right)u_{k}\right\rangle_{H}=\sum\nolimits^{N}_{j,k=1}\left\langle\left(S^{\alpha_{j}}\right)^{*}Wu_{j},P\left(S^{\alpha_{k}}\right)^{*}Wu_{k}\right\rangle_{\mathcal{K}}=\left\langle y,Py\right\rangle_{\mathcal{K}}. By assumption ( 2.3 ), we have \left\langle y,\left(\sum\nolimits^{d}_{i=1}S_{i}PS^{*}_{i}\right)y\right\rangle_{\mathcal{K}}\leq\left\langle y,Py\right\rangle_{\mathcal{K}}, which is exactly ( 2.4 ). ( 2 ) \Rightarrow ( 1 ). Since K is p.d., it admits a Kolmogorov/GNS factorization: there exists a Hilbert space H^{\prime} and operators V_{\alpha}\colon H\rightarrow H^{\prime} such that K\left(\alpha,\beta\right)=V^{*}_{\alpha}V_{\beta},\quad\alpha,\beta\in\mathbb{F}^{+}_{d}, and \mathscr{D}\coloneqq span\left\{V_{\alpha}u:\alpha\in\mathbb{F}^{+}_{d},u\in H\right\} is dense in H^{\prime} . Concretely, one may take H^{\prime} to be the reproducing kernel