Operator Norm Bounds for Multi-leg Matrix Tensors and applications to Random Matrix Theory

Benoît Collins, Wangjun Yuan · 2026

Read Operator Norm Bounds for Multi-leg Matrix Tensors and applications to Random Matrix Theory by Benoît Collins, Wangjun Yuan (2026) 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-2603-27659.txt.

Introduction

Collins and Yuan bound norms for multi-leg matrix tensors. They study extreme values of traced tensor products. A diagram method frames both upper and lower bounds. Bounds scale as powers of the factor dimension. Permutation matrices hit the lower bounds exactly. Proofs start with two factors and extend to any count. Open indices yield matrix bounds for all moments. One coefficient choice attains every lower bound. Random matrix uses follow from the bounds.

Catalog record

CREATORBenoît Collins, Wangjun Yuan
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE130313
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.27659

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Operator Norm Bounds for Multi-leg Matrix Tensors and applications to Random Matrix Theory Benoît Collins, Wangjun Yuan Abstract We investigate the extremal values of partial traces of matrix tensors under operator norm constraints. To evaluate these multi-linear quantities, we develop a comprehensive graphical formalism that encodes multi-leg partial traces, partial permutations, and their moments using colored directed graphs. With this graphical framework, we establish optimal, sharp bounds for the partial trace ( Tr σ 1 ⊗ … ⊗ Tr σ k ) ​ ( A 1 , … , A m ) (\mathrm{Tr}_{\sigma_{1}}\otimes\ldots\otimes\mathrm{Tr}_{\sigma_{k}})(A_{1},\ldots,A_{m}) over matrices bounded by ‖ A i ‖ ≤ 1 \|A_{i}\|\leq 1 . Specifically, we prove that this maximum evaluates exactly to N M ⁡ ( σ 1 , … , σ k ) N^{M(\sigma_{1},\ldots,\sigma_{k})} , where N N is the dimension and M M represents the maximal number of directed cycles in the associated graph across all possible internal vertex pairings. We further derive explicit operator norm estimates for matrices generated by partial traces of partial permutations. Finally, we apply these combinatorial bounds to multi-matrix random matrix theory. By examining models involving Ginibre ensembles, we extend concepts of asymptotic freeness to matrix coefficient algebras, establishing operator norm estimates that rigorously separate the asymptotic behavior of non-crossing and crossing pairings. Mathematics Subject Classification 2020 (MSC2020): Primary 15A60; Secondary 15B52, 46M05, 46L54, 47A30, 47C15, 47N30, 60B20. Keywords and phrases: Tensor monomials: Partial trace; Operator norm for tensors. 1 Introduction The integration of random unitary matrices with respect to the Haar measure plays a fundamental role in random matrix theory, operator algebras, and free probability. A central algebraic tool for evaluating these integrals is the Weingarten calculus [ 5 , 8 ] . For two m -tuples of matrices A_{1},\ldots,A_{m} and B_{1},\ldots,B_{m} in M_{N}(\mathbb{C}) , the expectation of their alternating product intertwined with a Haar-distributed unitary matrix U is given by a sum over permutations: \mathbb{E}\left[\mathrm{Tr}\left(A_{1}UB_{1}U^{*}\cdots A_{m}UB_{m}^{*}\right)\right]=\sum_{\begin{subarray}{c}\sigma,\tau,\rho\in\mathcal{P}([m])\\ \sigma\tau\rho=(1,\ldots,m)\end{subarray}}\mathrm{Tr}_{\sigma}(A_{1},\ldots,A_{m})\mathrm{Tr}_{\tau}(B_{1},\ldots,B_{m})\mathrm{Wg}(\rho,N), where \mathcal{P}([m]) is the set of all permutations on [m]=\{1,2,\ldots,m\} , and \mathrm{Tr}_{\sigma}(A_{1},\ldots,A_{m}) is defined as the product over the cycles of \sigma of the traces of the cyclically ordered products of the matrices. This precise algebraic framework has played a key role in establishing profound analytical results, including sharp operator norm bounds for non-commutative polynomials [ 2 ] and strong asymptotic freeness [ 1 ] . More recently, there has been a renewed interest in extending these models to the tensor setting, where the integral is taken over U=U_{1}\otimes\dots\otimes U_{k} , with each U_{i} being an independent Haar-distributed random matrix in M_{N}(\mathbb{C}) [ 4 ] . Such multi-fold tensor models arise naturally in various contexts, including the spectral distribution of large dimensional tensor products [ 7 ] . In this setting, the integrations yield combinations of fundamental multi-linear tensor invariants of the form: \displaystyle(\mathrm{Tr}_{\sigma_{1}}\otimes\ldots\otimes\mathrm{Tr}_{\sigma_{k}})(A_{1},\ldots,A_{m}), (1.1) where A_{1},\ldots,A_{m}\in M_{N}(\mathbb{C})^{\otimes k} . In this paper, we turn our attention to controlling the extremal values of these tensor invariants ( 1.1 ) under the constraint that the operator norms of the coefficient matrices satisfy \|A_{i}\|\leq 1 . If k=1 , the solution to this problem is trivial: the maximum is achieved when A_{i}=I_{N} for all i , yielding exactly N^{\#\text{cycles}(\sigma_{1})} . However, evaluating these quantities becomes deeply combinatorial and highly non-trivial as soon as the number of legs k\geq 2 . There are at least three primary motivations to systematically delve into this operator norm analysis: 1. Operator Algebras and Free Probability: In the two-leg case ( k=2 ), this precise maximization problem emerged in the study of Hayes’ random matrix approach to the Peterson-Thom conjecture [ 9 ] . While estimates leveraging the complete positivity of partial traces allow one to conclude that the maximum behaves nicely if \sigma_{1} is non-crossing and \sigma_{2} is the full cycle, extending this to general permutations \sigma_{1},\sigma_{2} requires a fundamentally new understanding of matrix tensors. 2. Tensor Models and Invariant Theory: The theoretical study of matrix tensors has been significantly developed in recent years [ 4 , 3 ] , but not yet thoroughly investigated within an operator algebraic framework. Because quantities of the form ( 1.1 ) are the fundamental observables of matrix tensors in this theory, establishing sharp evaluations of these multi-leg traces is a vital preliminary step for future asymptotic analysis. 3. Quantum Information Theory: Bounds on these tensor invariants directly capture the presence and limitations of entanglement across multiple tensor subsystems [ 6 , 3 ] . Our results demonstrate that while entanglement effects actively manifest in these maximization problems, they remain quite tame and structurally bounded, guided by underlying symmetries and Schur-Weyl duality. To tackle this, we develop a comprehensive graphical calculus that translates the multilinear tensor structure into the analysis of colored directed graphs. Throughout this paper, for the sake of clarity, we carefully detail the proofs for the two-leg case ( k=2 ) before generalizing to an arbitrary number of legs. We also extend our framework beyond full permutations to arbitrary partial permutations, leading to explicit operator norm estimates for the resulting multi-leg matrices. Let us summarize the main results of this paper. Our analysis relies on encoding the multi-linear tensor structure into a colored directed graph G_{\sigma_{1},\ldots,\sigma_{k}} . In this graph, each matrix A_{i} is represented as a “rectangle” with k in-vertices and k out-vertices, and the permutations \sigma_{j} define the external directed edges between different rectangles. By introducing auxiliary internal edges (referred to as “blue edges”) that pair the in-vertices and out-vertices within each rectangle, we define our fundamental combinatorial invariant: let M(\sigma_{1},\ldots,\sigma_{k}) denote the maximal number of directed cycles in G_{\sigma_{1},\ldots,\sigma_{k}} over all possible valid connections of these internal blue edges. Using this graphical formalism, our first main result exactly evaluates the maximization problem for the scalar partial trace over the unit ball of the operator norm: Theorem (cf. Theorem 2 ) . For any integer k\geq 1 and permutations \sigma_{1},\ldots,\sigma_{k}\in\mathcal{P}([m]) , the maximum of the multi-leg partial trace over matrices bounded in operator norm is governed exactly by the cycle structure of the graph: \displaystyle\max_{\|A_{1}\|,\ldots,\|A_{m}\|\leq 1}\left|\left(\mathrm{Tr}_{\sigma_{1}}\otimes\ldots\otimes\mathrm{Tr}_{\sigma_{k}}\right)\left(A_{1},\ldots,A_{m}\right)\right|=N^{M(\sigma_{1},\ldots,\sigma_{k})}. We then extend this topological counting to the case where \sigma_{1},\ldots,\sigma_{k} are partial permutations . We denote by \mathcal{P}^{\prime}([m]) the set of partial permutations on [m] . In this regime, the partial trace does not contract all indices and consequently outputs a multi-leg matrix Y rather than a scalar. By lifting the graph structure to encode the moments \mathrm{Tr}((YY^{*})^{p}) , we obtain optimal, sharp estimates on the operator norm: Theorem (cf. Theorem 3 and Corollary 6 ) . For any partial permutations \sigma_{1},\ldots,\sigma_{k}\in\mathcal{P}^{\prime}([m]) , let Y be the matrix resulting from the partial trace evaluation. The operator norm of Y satisfies the sharp bound: \displaystyle\max_{\|A_{1}\|,\ldots,\|A_{m}\|\leq 1}\|Y\|=N^{M(\sigma_{1},\ldots,\sigma_{k})}. Finally, we apply these deterministic, combinatorial operator norm bounds to multi-matrix random matrix theory. By examining matrices conjugated by Ginibre ensembles (as a combinatorially tractable proxy for Haar unitaries), we establish operator norm estimates on the coefficient algebra that rigorously isolate topological contributions. Theorem (cf. Theorems 7 and 8 ) . In the context of the matrix coefficient algebra scaled with differing leg dimensions N^{d_{1}} and N^{d_{2}} , the partial trace limits sharply distinguish topological pairings of the Ginibre matrices. We establish explicit operator norm bounds demonstrating that crossing (non-planar) pairings are suppressed

Read the complete text