[Paper Review] Selberg-type integrals and the variance conjecture for the operator norm

Beatrice-Helen Vritsiou|arXiv (Cornell University)|May 8, 2018

Spectral Theory in Mathematical Physics26 references3 citations

TL;DR

This paper verifies the generalized variance conjecture for the unit ball of the operator norm in classical matrix subspaces, including real symmetric and Hermitian matrices, by leveraging Selberg-type integrals and Weingarten calculus. It establishes that these bodies are in almost isotropic position despite not being isotropic, confirming the conjecture's validity with dimension-independent bounds on variance of the Euclidean norm.

ABSTRACT

The variance conjecture in Asymptotic Convex Geometry stipulates that the Euclidean norm of a random vector uniformly distributed in a (properly normalised) high-dimensional convex body $K\subset {\mathbb R}^n$ satisfies a Poincaré-type inequality, implying that its variance is much smaller than its expectation. We settle the conjecture for the cases when $K$ is the unit ball of the operator norm in classical subspaces of square matrices, which include the subspaces of self-adjoint matrices. Through the estimates we establish, we are also able to show that the unit ball of the operator norm in the subspace of real symmetric matrices or in the subspace of Hermitian matrices is not isotropic, yet is in almost isotropic position.

Motivation & Objective

• To verify the generalized variance conjecture for unit balls of the operator norm in classical matrix subspaces such as self-adjoint and Hermitian matrices.
• To investigate whether these operator norm bodies are isotropic or nearly isotropic, despite not satisfying the isotropic condition.
• To establish sharp estimates on the variance of the Euclidean norm on such convex bodies using advanced integral techniques.
• To connect the variance conjecture to the broader KLS conjecture via dimension-independent bounds.

Proposed method

• Utilizes Selberg-type integrals to compute moments of the operator norm on matrix spaces over ℝ, ℂ, and ℍ.
• Applies Weingarten calculus for unitary and orthogonal groups to evaluate integrals of products of matrix entries over the unit ball of the operator norm.
• Employs the covariance matrix and its largest singular value to quantify the variance of the squared Euclidean norm.
• Derives exact expressions for second moments of matrix entries and their products, enabling variance estimation.
• Compares results across different matrix types (real, complex, quaternionic) to confirm consistency in asymptotic behavior.
• Uses the equivalence between the variance conjecture and the KLS conjecture, as established by Eldan, to infer broader implications.

Experimental results

Research questions

• RQ1 Does the generalized variance conjecture hold for the unit ball of the operator norm in classical matrix subspaces such as self-adjoint or Hermitian matrices?
• RQ2 Are these operator norm bodies isotropic, or only nearly isotropic, in high dimensions?
• RQ3 Can Selberg-type integrals and Weingarten calculus be effectively used to compute the variance of the Euclidean norm on such convex bodies?
• RQ4 What is the precise dependence of the variance on the dimension and the structure of the matrix space?
• RQ5 How do the results for operator norm bodies relate to the broader KLS conjecture?

Key findings

• The generalized variance conjecture holds for the unit ball of the operator norm in the subspaces of real symmetric and Hermitian matrices, with a dimension-independent constant in the variance bound.
• The unit ball of the operator norm in the space of real symmetric matrices is not isotropic, but is in almost isotropic position, with the isotropic constant approaching 1 as dimension increases.
• For the complex Hermitian case, the variance of the squared Euclidean norm is bounded by a constant multiple of the largest eigenvalue of the covariance matrix, confirming the conjecture.
• Explicit computations show that the ratio of the fourth moment to the square of the second moment of the norm is asymptotically 1, consistent with the thin-shell phenomenon.
• The results confirm that the variance conjecture is valid for these matrix spaces, with the implied constant in the variance bound independent of dimension.
• The paper establishes that the operator norm bodies satisfy a Poincaré-type inequality with optimal constants, supporting their role in high-dimensional convex geometry.