Herglotz-Nevanlinna Theorem Overview

• The Herglotz-Nevanlinna theorem is a precise representation of analytic functions mapping the upper half-plane to itself with nonnegative imaginary parts.
• It extends to multivariable, matrix, and operator-valued cases by enforcing strict measure, growth, and orthogonality conditions.
• Its applications span spectral theory, control systems, and passive network synthesis, linking analytic function theory with operator analysis.

A Herglotz-Nevanlinna theorem provides an exact integral representation for analytic functions mapping a complex domain—classically the upper half-plane—to itself or to its closure, with non-negative imaginary part. These functions and their representations are central in complex analysis, operator theory, control and systems theory, and mathematical physics, precisely characterizing all analytic self-maps with positivity constraints. The theorem extends robustly to functions of several complex variables, matrix-valued and operator-valued cases, and incorporates deep structural restrictions on the representing measure and the analytic boundary behavior of solutions.

1. The Classical One-Variable Herglotz-Nevanlinna Theorem

Let $H = \{ z \in \mathbb{C} : \Im z > 0 \}$ denote the upper half-plane. A function $f : H \rightarrow \mathbb{C}$ is a Herglotz (Nevanlinna–Pick) function if $f$ is analytic and $\Im f(z) \geq 0$ for all $z \in H$.

Theorem (Herglotz-Nevanlinna, one variable):

Suppose $f : H \rightarrow \mathbb{C}$ is analytic and $\Im f(z) \ge 0$. Then there exist unique real numbers $a \ge 0$, $b \in \mathbb{R}$, and a unique positive Borel measure $\mu$ on $f : H \rightarrow \mathbb{C}$0 satisfying

$f : H \rightarrow \mathbb{C}$1

such that for all $f : H \rightarrow \mathbb{C}$2,

$f : H \rightarrow \mathbb{C}$3

Uniqueness is absolute: $f : H \rightarrow \mathbb{C}$4 are determined by $f : H \rightarrow \mathbb{C}$5, with each component recovered from boundary limits and Stieltjes inversion. Conversely, any such $f : H \rightarrow \mathbb{C}$6 yields a Herglotz-Nevanlinna function (Agler et al., 2012, Hayashi et al., 10 Apr 2025, Luger et al., 2022).

Normalization and growth at infinity are fixed by the subtracted kernel $f : H \rightarrow \mathbb{C}$7, enforcing

$f : H \rightarrow \mathbb{C}$8

and ensuring the coefficients are determined by

$f : H \rightarrow \mathbb{C}$9

Atomic and absolutely continuous parts of the measure $f$0 are recovered from boundary behavior of $f$1 as $f$2 approaches the real axis or infinity, using nontangential or Stolz cone limits.

2. Multivariable Generalization: Integral Representations and Conditions

For the poly-upper-half-plane $f$3, the Loewner or Pick class comprises analytic $f$4 with $f$5. The multivariable Herglotz-Nevanlinna theorem provides an integral representation: $f$6 where $f$7, $f$8, and $f$9 is a positive Borel measure on $\Im f(z) \geq 0$0. The kernel $\Im f(z) \geq 0$1 is constructed as

$\Im f(z) \geq 0$2

A growth condition and a Nevanlinna orthogonality constraint on $\Im f(z) \geq 0$3 are required: $\Im f(z) \geq 0$4

$\Im f(z) \geq 0$5

with $\Im f(z) \geq 0$6 as in (Luger et al., 2017, Nedic et al., 2022, Nedic, 2020).

For $\Im f(z) \geq 0$7, the representation specializes accordingly and includes a further Nevanlinna orthogonality (moment vanishing) for the representing measure on $\Im f(z) \geq 0$8 (Luger et al., 2016). Uniqueness and atomicity restrictions become more rigid: nontrivial measures cannot be finite (unless identically zero), and all atoms vanish.

Reduction to the one-variable case is always possible by freezing $\Im f(z) \geq 0$9 variables, yielding a classical Herglotz-Nevanlinna function in the remaining coordinate.

3. Operator, Matrix, and Structured Resolvent Representations

The integral representation admits operator-theoretic and matrix generalization. For the operator-valued case, if $z \in H$0, analytic with $z \in H$1, there exist $z \in H$2, $z \in H$3, and a positive operator-valued Borel measure $z \in H$4 such that

$z \in H$5

Matrix-valued Herglotz-Nevanlinna functions $z \in H$6 admit partial fraction decompositions and have a matrix-valued Chebotarev form. Connections with matrix Hankel and Bezoutian moment criteria are exactly stated, generalizing classical spectral and positivity criteria (Zhan, 2020, Luger et al., 2022).

In the general multivariable setting, the Agler–Tully-Doyle–Young structured resolvent machinery yields canonical forms in terms of a densely-defined self-adjoint operator $z \in H$7, a positive decomposition $z \in H$8 of the identity on a Hilbert space $z \in H$9, and "structured pencils" $f : H \rightarrow \mathbb{C}$0. The integral-resolvent representation involves four types, classified by the structure and asymptotic growth of $f : H \rightarrow \mathbb{C}$1 at infinity (Agler et al., 2012):

• Type 1: $f : H \rightarrow \mathbb{C}$2
• Type 2: $f : H \rightarrow \mathbb{C}$3
• Type 3: $f : H \rightarrow \mathbb{C}$4
• Type 4: $f : H \rightarrow \mathbb{C}$5 as in Type 4, incorporating both constants $f : H \rightarrow \mathbb{C}$6, higher-degree polynomial normalization, and a more complex structured resolvent.

The type of representation corresponds directly to the order of growth at infinity:

• $f : H \rightarrow \mathbb{C}$7: Type 1,
• $f : H \rightarrow \mathbb{C}$8 but not $f : H \rightarrow \mathbb{C}$9: Type 2,
• $\Im f(z) \ge 0$0 etc.: Type 3,
• Most general: Type 4.

4. Characterization, Uniqueness, and Symmetry of Representing Measures

The representing measure $\Im f(z) \ge 0$1 in several variables (a Nevanlinna measure) is characterized by growth and (multi-)orthogonality, equivalently by a support property for its distributional Fourier transform: $\Im f(z) \ge 0$2 is supported in $\Im f(z) \ge 0$3 (Nedic et al., 2022).

Uniqueness is enforced via multidimensional versions of the Stieltjes inversion formula, extracting measure data from the boundary values of $\Im f(z) \ge 0$4 along admissible approach regions (e.g., non-tangential or Stolz approaches to the real axis for each variable) (Luger et al., 2016, Nedic et al., 2022).

The measure structure is rigid: nontrivial Nevanlinna measures in $\Im f(z) \ge 0$5 variables cannot be finite, have no atomic part, and their support is governed by the positivity properties of the associated kernel. Invariant under appropriate changes of variables (including Möbius automorphisms in the one-variable case), the measure transforms according to explicit weight and transfer rules (Hayashi et al., 10 Apr 2025).

When restricting the underlying function class to the unit polydisk, an analogous Korányi–Pukánszky representation exists. The representing measure on the distinguished boundary (torus) is subject to moment-vanishing (pluriharmonicity) constraints, and the Poisson–Szegő kernel is the relevant analytic object (Bhowmik et al., 30 May 2025).

5. Kernel Decomposition and Positivity Connections

The Herglotz-Nevanlinna theorem and its multivariable generalizations are closely linked to the positive definiteness of Nevanlinna kernels: $\Im f(z) \ge 0$6 (one variable), and in several variables

$\Im f(z) \ge 0$7

which are positive semi-definite if and only if $\Im f(z) \ge 0$8 or $\Im f(z) \ge 0$9 belongs to the Herglotz-Nevanlinna class. This equivalence is a pivotal technical component, enabling multiple dual characterizations, including via reproducing kernel Hilbert spaces and moment/matrix criteria (Nedic, 2020).

In higher dimensions, kernel-based decompositions remain central, with any Herglotz-Nevanlinna function admitting expression as a sum of a linear part and a Poisson-type function encoding the measure via a highly structured kernel integration.

6. Applications, Transformations, and Extensions

Herglotz-Nevanlinna representations are foundational in systems theory, passive network synthesis, control, and in spectral/resolvent theory of (possibly unbounded) self-adjoint operators. In applications, they provide structural constraints (e.g., sum rules, integral bounds) and facilitate inverse problems (e.g., recovery of measure from boundary data) (Luger et al., 2022).

The theorem is robust under Möbius transformation, with explicit rules for the adjustment of $a \ge 0$0 under precomposition with automorphisms. In fact, the full semigroup of complex Möbius endomorphisms acting on $a \ge 0$1 preserves the Pick class, with a complete description of how the measure transforms (Hayashi et al., 10 Apr 2025).

Matrix and operator-valued generalizations are central in multidimensional systems, operator monotone/convex functions, and spectral analysis. They admit partial fraction decompositions, moment problem solutions, and establish necessary Hankel-type positivity conditions (Zhan, 2020).

On the unit polydisc, superresolution results relate the uniqueness of Herglotz–Nevanlinna interpolants to extremal measures concentrated on real-algebraic sets and explicit stability bounds via Markov moment theory and volume estimates (Bhowmik et al., 30 May 2025).

7. Further Developments and Structural Properties

Recent work systematically analyzed the structure of Nevanlinna measures:

• Measures are classified via Fourier support, hyperplane supports, and extremality;
• Measures supported on hyperplanes must satisfy sharp positivity and polynomiality conditions, with explicit construction of extremal measures of the form $a \ge 0$2 for hyperplanes $a \ge 0$3 with normal in the non-negative orthant (Nedic et al., 2022);
• The singular/absolutely continuous decomposition of measures is analyzed under variable restriction, with pluriharmonic dependence of the parameter family;
• Quantitative lower and upper bounds on the measure of expanding/shrinking cubes in $a \ge 0$4 are obtained, further illustrating the nontriviality and growth of Nevanlinna measures in higher dimensions.

These developments provide a comprehensive framework for Herglotz–Nevanlinna theory across analytic, operator-theoretic, and multivariable domains, linking classical harmonic analysis with multidimensional positivity, spectral, and moment theory (Agler et al., 2012, Luger et al., 2016, Luger et al., 2017, Nedic, 2020, Nedic et al., 2022, Hayashi et al., 10 Apr 2025, Bhowmik et al., 30 May 2025, Zhan, 2020, Luger et al., 2022).