Markov-Krein Correspondence Analysis

• Markov-Krein correspondence is a framework connecting positive measures and spectral data to characterize operator models and physical systems.
• It extends classical moment problems and string spectral theory to diverse fields through generalized resolvent formulas and operator perturbation techniques.
• Applications span random matrices, diffusion processes, and noncommutative function theory, enabling precise quantification of spectral fluctuations and operator extensions.

The Markov-Krein correspondence refers to a family of deep and precisely quantifiable relations between positive measures (or, more broadly, positive definite kernels or operator-theoretic objects) and the solution spaces or spectral data of certain operator models. Originating in classical analysis for problems such as the moment problem and spectral theory of canonical systems, the correspondence was generalized by Krein, Markov, and others to encompass a vast array of settings, including one-dimensional diffusions, operator extensions, noncommutative function theory, and random matrices. Central to this framework is the translation between the structure of a "string" (an object encoding physical, probabilistic, or operator-theoretic information) and a spectral measure, and, in many modern forms, the mapping between probability distributions and transformed measures via exponential, moment generating, or integral equations.

1. Classical Correspondence: Strings and Spectral Measures

In the canonical setting, a Krein string is a right-continuous, non-decreasing function $m$ defined on a half-line, possibly taking the value $+\infty$ and normalized such that $m(0-) = 0$. The central problem is to associate to $m$ a spectral measure $\sigma$ via the spectral theory of the generalized second-order operator (related to the Laplacian weighted by the string). Key relationships include the function $p(x)$ solving

$p(x) = 1 - \int_0^x p(x-y) \, dm(y),$

and the Laplace transform formula

$$f(x) = a + \int_0^\infty e^{-x\lambda} \, d\sigma(\lambda).$$

This structure encodes the "correspondence" between $m$ (physical data) and $\sigma$ (spectral data). Continuity of the correspondence, and its extension to classes of strings whose spectral functions exhibit only polynomial growth at infinity, is characterized via integrability conditions on the string and corresponding tightness conditions on sequences $(m_n)$ converging to $m$ (Kotani, 2013). For spectral measures with prescribed growth, one introduces a convex scale function $\varphi$ and places normalizing and integrability constraints: $$M(x) = \int_x^l (x-y) \, dm(y), \quad \int_{-\infty}^0 \varphi(M(x)) dx < \infty,$$ guaranteeing effective control over the correspondence.

2. Extensions to Operator Theory and Spectral Shift

The formulation of the Markov-Krein correspondence has been generalized to operator-theoretic settings, notably those involving spectral shift formulas and perturbations. Krein's formula for the difference between functions of two self-adjoint operators $H$ and $H_0$ is

$$\mathrm{Tr}\{\varphi(H) - \varphi(H_0)\} = \int_{a}^b \varphi'(x)\, \xi(x) dx,$$

where $\xi$ is the spectral shift function. In higher dimensions, for commuting tuples of operators $(H_1, H_2)$ and $(H^0_1, H^0_2)$ perturbed by Hilbert-Schmidt operators, a Stokes-like trace formula holds, involving a two-variable spectral shift measure $p$,

$$\mathrm{Tr}\{[H_1 - H^0_1] P_1(x,H_2) dx + [H_2 - H^0_2] P_2(H_1, y) dy\} = \iint_{[a,b]^2} \psi(x, y)\, p(dx, dy),$$

which encodes the multivariate spectral shift and directly generalizes the classical correspondence between moments and measures (Chattopadhyay et al., 2014).

3. Probabilistic and Non-Self-Adjoint Generalizations

In Markov process and diffusion theory, the correspondence is realized in terms of spectral representations of processes and the excursion theory at boundary points. If two minimal Markov semigroups are intertwined by an appropriate operator, their recurrent (e.g., reflecting) extensions share the same local time at a boundary point. When one semigroup is a quasi-diffusion, the Laplace exponent of the inverse local time admits a Stieltjes representation,

$$\Phi(q) = \int_0^\infty \frac{1}{\lambda + q}\, \mu(d\lambda),$$

where $\mu$ is the spectral measure of the string. Intertwining relationships extend the applicability to non-self-adjoint Markov semigroups by establishing the absolute continuity of their spectral measures with respect to $\mu$, thereby enlarging the scope of Krein's theory to more general settings and models such as reflected Laguerre semigroups and self-similar processes (Patie et al., 2017).

4. Analytical and Functional Models for Extensions

The Krein transform and its associated formalism provide a one-to-one correspondence between classes of linear relations—such as positive linear relations—and symmetric contractions. Given a linear relation $T$, the Krein transform is

$A(T) = 2(T + I)^{-1} - I,$

an involutive mapping. Positive relations correspond precisely to symmetric contractions. This explicit parameterization simplifies the construction of semi-bounded and self-adjoint extensions, such as Friedrichs and Krein–von Neumann extensions, with explicit formulas for positive extensions involving contraction operators in the deficiency spaces (Rios-Cangas, 2023, Emmel, 10 Jul 2024).

The resolvent formalism is further generalized for symmetric relations of deficiency index $(1,1)$ to cover all regular, not necessarily self-adjoint, extensions. The extension resolvent admits formulas involving quasi-Herglotz functions (analytic functions with generalized integral representations) and kernel models in reproducing kernel Hilbert spaces,

$$(\widetilde{A} - w)^{-1}(f) = D_w(f) - \frac{f(w)}{g(w)} D_w(g),$$

where $D_w(f)$ is a difference quotient operator, and spectrum characterization reduces to analyzing the zeros of the defining function $g$ (Emmel, 10 Jul 2024).

5. Quantized and Noncommutative Analogues

The correspondence has been quantized, particularly in representation theory, where random signatures (encoding irreducible components in unitary group representations) generate families of discrete measures linked via interlacing conditions. Continuous limits yield continual Young diagrams, and there is a bijection between probability measures and diagram profiles encapsulated in the moment generating function identity,

$$\sum_{k \geq 0} m_k z^k = \exp\left(\sum_{k \geq 1} \frac{w_k}{k} z^k\right),$$

relating moments $(m_k)$ to cumulants $(w_k)$ (Goel et al., 2020). In operator-valued and $C^*$-algebraic contexts, generalized positive definite kernels $K(s, t)(a)$ are factored via Stinespring-type dilations,

$K(s, t)(a) = V(s)^* \pi(a) V(t),$

and kernel domination is characterized via operator Radon-Nikodym derivatives. When the associated representation $\pi$ is irreducible, domination reduces to scalar proportionality, directly paralleling the classical theory of pure states and extremal measures (Jorgensen et al., 27 May 2025).

6. Applications to Random Matrices and High-Temperature Asymptotics

In random matrix theory, the Markov-Krein correspondence characterizes spectral fluctuations of principal submatrices via Rayleigh measures and non-crossing partition combinatorics. If the empirical eigenvalue distribution of an $N \times N$ Hermitian matrix converges to $m$, the rescaled deviation for a principal submatrix converges in moments to the Rayleigh measure $T$ determined by the correspondence,

$M_k(T) = \sum_{p \in NC(k)} (k+1 - |p|) R_p(m),$

with $R_p(m)$ the free cumulants of $m$ (Fujie et al., 2021).

In high-temperature regimes for classical beta ensembles, the limiting spectral measure $\mu$ is the inverse Markov-Krein transform of a classical probability law: $$\int \frac{d\nu(x)}{(z-x)^c} = \exp\left(-c \int \log(z-x) d\mu(x)\right),$$ with $\nu$ corresponding to Gaussian, gamma, or beta distributions depending on the ensemble (Nakano et al., 2 Dec 2024). The mapping $\mu \mapsto \nu$ is inverted to characterize $\mu$ as $\mu = _c^{-1}(\nu)$. For dynamical ensembles, the measure-valued processes also evolve according to this relationship.

High-temperature matrix harmonic analysis exploits the Dirichlet process $D_{c\rho}$ to represent the Markov–Krein transformed measure as

$$\rho^{(c)} = \text{Law}\left(\int x\, D_{c\rho}(dx)\right),$$

establishing both existence and uniqueness of the correspondence via the random mean and connecting integrability conditions to measure properties (Zhang, 29 Aug 2025).

7. Spectral and Oscillatory Behavior in Canonical Systems

The interplay between the coefficients of canonical systems (e.g., Krein systems with oscillating potentials) and their spectral measures is governed by equivalence theorems—such as the continuous Nevai–Totik theorem—that connect decay rates of system coefficients to the decay properties of the Fourier transform of the Szegő function, entropy decay, and resonance solutions. For a Krein system,

$$P'(r,\lambda) = i\lambda P(r,\lambda) - a(r) P_*(r,\lambda), \quad P_*'(r,\lambda) = -a(r) P(r,\lambda),$$

the equivalence of rapid decay of $a$, absolute continuity and decay for the Fourier transform of the inverse Szegő function $\Pi$, entropy decay, and decaying solutions at special spectral parameters is established (Gubkin, 13 Sep 2024). These relations integrate the Markov-Krein correspondence with resonance and scattering theory in mathematical physics.

---

The Markov-Krein correspondence thus provides a unified framework for the analysis of spectral theory, operator extensions, random matrix ensembles, and noncommutative function theory, fusing classical measure-theoretic correspondences with modern analytic and probabilistic models. Its generalizations and applications continue to shape understanding in mathematical analysis, probability, operator theory, and mathematical physics.