Matrix-Valued Spectral Measure

• Matrix-valued spectral measures are Borel measures that assign Hermitian matrix values, extending classical spectral theory to non-scalar settings.
• They facilitate advanced analysis in moment problems, sum rules, orthogonal polynomials, and perturbation theory, particularly in random matrix ensembles.
• These measures underpin applications in operator theory, quantum mechanics, multivariate signal processing, and spectral geometry through rigorous mathematical frameworks.

A matrix-valued spectral measure is a Borel measure on the real line (or, more generally, a locally compact space) that takes values in the space of complex or Hermitian matrices, and provides a rigorous mathematical framework for extending classical spectral theory, sum rules, moment problems, and orthogonal polynomial constructions to non-scalar, operator- or matrix-valued settings. Such measures play a central role in the modern spectral analysis of self-adjoint operators, non-commutative probability, perturbation theory, and structured random matrix ensembles.

1. Definition and Basic Properties

Let $p\geq 1$ be an integer. A $p\times p$ matrix-valued spectral measure $\Sigma$ on $\mathbb R$ is a set function

$$A \subset \mathbb R \mapsto \Sigma(A) = (\Sigma_{ij}(A))_{1 \leq i, j \leq p}$$

such that:

• Each $\Sigma_{ij}$ is a complex Borel measure;
• For every Borel set $A \subset \mathbb R$, the matrix $\Sigma(A)$ is Hermitian: $\Sigma(A) = \Sigma(A)^*$;
• (Positive semi-definiteness) $\Sigma(A)\succeq 0$, i.e., $x^* \Sigma(A) x \geq 0$ for all $x \in \mathbb{C}^p$;
• In many applications, normalization holds: $\Sigma(\mathbb{R})=I_p$.

The matrix-valued analog of the Riesz-Markov-Kakutani representation theorem applies in this context, and such measures naturally arise from spectral decompositions of self-adjoint operators projected onto finite-dimensional subspaces (Gamboa et al., 2016, Liaw et al., 2018, Abbott et al., 2 Aug 2025).

2. Construction from Operators and Spectral Decomposition

Given a self-adjoint operator $A$ on a Hilbert space $\mathcal{H}$ and a $d$-dimensional subspace—often represented via an operator $B: \mathbb C^d \to \mathcal H$—the spectral measure associated with $A$ is

$\mathbf M(E) = B^* \mathbb{E}^A(E) B,$

where $\mathbb{E}^A(E)$ is the spectral projector of $A$ for $E\subset\mathbb R$, yielding a $d \times d$ positive semidefinite matrix-valued measure (Liaw et al., 2018). This construction extends to finite-rank perturbations, block Jacobi matrices, polynomials in Jacobi operators, and random matrix ensembles (Roman et al., 1 Sep 2025, Gamboa et al., 2016).

A fundamental result is that any Hermitian matrix $A\in \mathfrak{H}_n$ admits a spectral representation:

$A = \sum_{j=1}^\ell \lambda_j E_{\lambda_j}$

with $E_{\lambda_j}$ orthogonal projectors. When extended to non-commutative exponentials, Katsnelson established that $e^{tA+B}$ admits representation as a (matrix-valued) Laplace transform:

$e^{tA+B} = \int e^{\lambda t} M(d\lambda)$

with $M$ a matrix Borel measure supported on $\operatorname{conv}(\sigma(A))$ (Katsnelson, 2016).

3. Moment Problems and Matrix-Valued Measures

The moment problem for matrix-valued measures asks: given a sequence of Hermitian (usually positive definite) moment matrices $(C_m)_{m=0}^{2n}$, does there exist a positive semidefinite matrix-valued measure $\rho(\lambda)$ on $\mathbb R$ with

$$C_m = \int \lambda^m d\rho(\lambda), \quad m = 0,\dots,2n?$$

The matrix Hamburger problem generalizes the classical scalar case. Kovalishina's theorems provide a complete description of the non-degenerate truncated matrix Hamburger problem in terms of Weyl-matrix-balls, J-positivity, and explicit Mӧbius parameterizations of admissible Stieltjes transforms (Abbott et al., 2 Aug 2025). Moment-positivity (block Hankel matrix $H_n \succeq 0$) is necessary; positive extendability is required for sufficiency.

Matrix-valued moment formulations are central in applications such as extracting spectral densities from quantum field theory correlators, with rigorous pointwise bounds given in terms of SDP-expressible constraints (Abbott et al., 2 Aug 2025).

4. Sum Rules and Large Deviations

Matrix-valued spectral measures admit sum rules relating spectral and recursion data, now in a non-scalar setting. Gamboa, Nagel, and Rouault (Gamboa et al., 2016) established such rules for Hermitian matrix measures. For a reference measure $\Sigma_{\rm ref}$ and an arbitrary normalized $\Sigma$ (possibly with outliers),

$$\mathcal K(\Sigma_{\rm ref}\|\Sigma) + \sum \mathcal F^\pm(\lambda_i^\pm) = \sum_{k=1}^\infty \left\{\tfrac12 \operatorname{tr}(B_k^2) + \operatorname{tr} G(A_kA_k^*)\right\},$$

where $\mathcal K$ is the reversed matrix Kullback-Leibler divergence, $\mathcal F^\pm$ rates for outlier eigenvalues, and $\{A_k, B_k\}$ are block recursion coefficients. This sum rule is derived by equating large deviations rate functions for both spectral measures and the tridiagonal coefficient representations in unitary-invariant ensembles (Hermite, Marchenko–Pastur, etc).

Special cases:

• Semicircle reference ($\Sigma_{\rm SC}$): recovers the Damanik-Killip-Simon matrix extension of the Killip–Simon rule.
• Marchenko-Pastur ($\Sigma_{\rm MP}$): yields analogous rate-functional identities for the LUE ensemble.

5. Orthogonal Polynomials and Random Walks

Matrix-valued orthogonal polynomials (MVOPs) extend scalar orthogonal polynomials, admitting matrix-valued measures as weight functions:

$\int P_n(t) dW(t) P_k(t)^* = \delta_{nk} I_m,$

where $W(t)$ is a positive semidefinite $m\times m$ weight matrix, possibly supported on discrete spectra of a random walk transition matrix. The block spectral theorem applies to a class of Markov chains with polynomial or block-tridiagonal generators, generalizing the Karlin–McGregor framework to higher dimensions (Roman et al., 1 Sep 2025).

Explicit forms for MVOPs and their matrix-valued orthogonality relations allow expressing $n$-step transition probabilities and spectral projections for random walks with complex, possibly non-commuting transition structures.

6. Perturbation Theory and Vector Mutual Singularity

In the context of finite-rank perturbations of self-adjoint operators, the framework of matrix-valued spectral measures is essential. For $A_\Gamma = A + B\Gamma B^*$ (with $B$ mapping into a finite-dimensional subspace and $\Gamma$ Hermitian), the spectral measures $\mathbf{M}$ and $\mathbf{M}_\Gamma$ may have overlapping scalar supports. However, Liaw and Treil introduced the notion of vector mutual singularity:

$$\mathbf M \perp_v \mathbf N \Longleftrightarrow \exists \text{ measurable orthogonal projection field } \Pi(x) : \Pi(x) \mathbf M(dx) \Pi(x) = 0,\, (I-\Pi(x)) \mathbf N(dx) (I-\Pi(x)) = 0 \text{ a.e. } x.$$

This property generalizes the mutual singularity of scalar measures and restores the correct structure in higher-rank settings. The Muckenhoupt $A_2$ matrix condition for Cauchy transforms is necessary for two-weight norm estimates, ensuring the validity of singular/mutually singular decompositions (Liaw et al., 2018).

Spectral representation results and generalized Aronszajn–Donoghue theorems rely on this matrix measure perspective.

7. Applications and Further Developments

Matrix-valued spectral measures are fundamental to:

• Functional calculi for operator-valued functions, especially for non-commuting or perturbed operators [$e^{tA+B}$, (Katsnelson, 2016)]
• Quantum statistical mechanics and lattice field theory, via nonparametric spectral density estimation and analytic continuation (Abbott et al., 2 Aug 2025)
• Multivariate time series and signal processing (matrix-valued power spectral densities) (Ning et al., 2014)
• Spectral geometry and direct integral decompositions in periodic operator theory (Kutsenko, 2012)
• Extremal convex geometry in function theory on planar domains, Schur–Agler class decomposition, and operator dilation problems (Ball et al., 2011)

Metrics on the space of matrix-valued spectral measures (e.g., Wasserstein-type, dual test-function frameworks) have been developed and proven to metrize the weak-* topology, enabling stable comparisons of measures in high-dimensional or non-commutative settings.

Recent advances include explicit hypergeometric function transforms associated with $2\times2$ matrix-valued orthogonal polynomials and continuous/discrete spectra with matrix multiplicities (Groenevelt et al., 2012).

---

Table: Key Settings and Matrix-Valued Spectral Measure Objects

Context	Measure Type	Core Reference(s)
Operator spectral theorem	Psd. Hermitian	(Gamboa et al., 2016, Liaw et al., 2018)
Laplace transform of $e^{tA+B}$	General matrix	(Katsnelson, 2016)
Orthogonal polynomials & MOPRL	Rank-one weight	(Roman et al., 1 Sep 2025, Groenevelt et al., 2012)
Moment problems, LQCD correlators	Psd. Hermitian	(Abbott et al., 2 Aug 2025)
Sum rules, random matrices	Normalized psd.	(Gamboa et al., 2016)
Direct-integral spectra	Psd. Hermitian	(Kutsenko, 2012)
Dilation, Schur/Agler classes	Normalized psd.	(Ball et al., 2011)

---

Matrix-valued spectral measures unify and generalize diverse strands in contemporary mathematical physics, operator theory, random matrix theory, and noncommutative probability, providing both explicit and abstract tools for spectral analysis in multivariate, block, or operator-valued settings.