Symmetric Linear Pencils Overview

• Symmetric linear pencils are matrix-valued functions A+λB defined with symmetric matrices, underpinning core aspects of spectral theory and algebraic geometry.
• Their classification via Segre symbols and canonical forms enables systematic analysis of eigenvalue behavior, structural equivalence, and linearizations.
• They have broad applications in optimization, control theory, and orthogonal polynomial recurrences, supporting efficient structure-preserving algorithms.

A symmetric linear pencil is a linear matrix-valued function $\lambda \mapsto A + \lambda B$ (or, in more general settings, $\lambda X + Y$) where $A$ and $B$ are symmetric (or Hermitian) matrices. The study of such pencils underlies core aspects of algebraic geometry, matrix theory, spectral analysis, convex optimization, and the spectral theory of operator pencils. Their classification, linearization, spectral behavior, and algebraic properties reveal deep connections between structure-preserving algorithms, projective geometry, and noncommutative convexity.

1. Fundamental Structures and Definitions

Let $S^n$ denote the space of $n \times n$ symmetric matrices over $\mathbb{C}$ or $\mathbb{R}$. A symmetric linear pencil $L = \operatorname{span}\{A, B\} \subset S^n$ is the set of matrices $\{A + \lambda B : \lambda \in \mathbb{C}\}$, or projectively, a line in the projective space $\mathbb{P}(S^n)$ (Fevola et al., 2020).

• Isomorphism classes: Two pencils $L = \operatorname{span}\{A, B\}$ and $L' = \operatorname{span}\{C, D\}$ are isomorphic if there exists $g \in GL(n)$ and $M \in GL(2)$ such that $(C, D) = (gAg^T, gBg^T)M$. This is the basis of the congruence action, crucial for understanding the geometric and spectral equivalence of pencils.

In the context of operator pencils and infinite-dimensional settings, symmetric linear pencils of the form $J_5 - \lambda J_3$, where $J_3$ is Jacobi and $J_5$ is symmetric five-diagonal, generalize classical eigenvalue problems and support higher-order recurrence theory (Zagorodnyuk, 2018, Zagorodnyuk, 2017).

2. Classification: Canonical Forms and Segre Symbols

The classification of symmetric linear pencils, especially two-dimensional subspaces ("pencils of quadrics"), falls under the Weierstrass–Segre theory (Fevola et al., 2020):

• Segre symbols: These encode the multiset of elementary divisors derived from $A - \lambda B$ over $\mathbb{C}[\lambda]$. For regular pencils (determinant a nonzero binary form), congruence classes are uniquely specified by these symbols, which correlate to configurations of $n$ points up to $PGL(2)$ transformations.
• Canonical forms: Explicit block-diagonal normal forms exist for each Segre symbol, and for $n = 2, 3$, a complete tabulation of types is available. For example, for $n = 2$, the diagonal and anti-diagonal types occur; for $n = 3$, five Segre symbols are realized.
• Orbit structure in projective geometry: In settings like $PG(\mathbb{F}^3 \otimes \mathbb{F}^3)$, symmetric pencils correspond to lines in the span of the Veronese variety and are classified by orbit and stabilizer types under $PGL(3, \mathbb{F})$, with explicit rank distributions and representative pencils (Lavrauw et al., 2017).

3. Structure-Preserving Linearizations and Block-Symmetric Forms

Preserving symmetry in linearizations of matrix polynomials is critical for spectrum and index recovery (Faßbender et al., 2016, Cachadina et al., 2017, Dopico et al., 2022, Bist et al., 2024). Several frameworks enable such structure:

• DL$(P)$ vector space: For $P(z)$ a square symmetric polynomial, all pencils in DL$(P)$ are block-symmetric, characterized by explicit block formulas indexed by the ansatz polynomial. Under the eigenvalue exclusion hypothesis, spectral data (eigenvalues, partial multiplicities, minimal indices) of $P$ can be recovered from any block-symmetric DL-pencil (Dopico et al., 2022).
• Block-Kronecker and block-minimal bases pencils: Families of symmetric block-Kronecker pencils provide a unifying method for companion and Fiedler-like forms, furnishing strong linearizations in both odd and even grade cases (with modified blocks for even degrees). Block-symmetry and explicit recovery maps for eigenvectors and minimal bases are systematically described (Faßbender et al., 2016, Cachadina et al., 2017).
• Double-ansatz method for systems: For transfer functions $G(\lambda)$ with regular, symmetric system polynomials, the unique block-symmetric double-ansatz pencil is the only structure-preserving linearization (modulo block-permutations). The block-symmetric subspace is singleton in this setting (Bist et al., 2024).

4. Grassmannian Stratification and Algebraic Geometry

The moduli of symmetric linear pencils are stratified in the Grassmannian $\operatorname{Gr}(2, S^n)$ by Segre symbols (Fevola et al., 2020):

• Strata $Gr_\sigma$: Each symbol $\sigma$ defines a locally closed stratum whose closure is an irreducible algebraic subvariety, and the closure relations follow a partial order coincident with Jordan decomposition closure relations.
• Codimension computations: The codimension of a stratum is computed via the sum over conjugate partitions of Segre blocks.
• Detection algorithms: Segre symbols may be computed algorithmically via Jordan canonical form of $AB^{-1}$ or Smith normal form of $A - \lambda B$; similarly, Grassmannian strata can be cut out via Plücker coordinates or Stiefel minors.

5. Spectral Theory, Eigenvalue Behavior, and Parameter Dependence

Symmetric definite pencils $A(x) - \lambda B(x)$ with parametric dependence admit detailed spectral analysis (Dieci et al., 2021):

• Smoothness and coalescence: Away from eigenvalue crossings, eigenvalues and eigenvectors are as smooth as $A$ and $B$. At generic crossing points (conical intersections), smoothness drops in a manner predicted by codimension and analytic structure.
• Block-diagonalization and monodromy: Spectral projectors are smooth except at crossings. At conical intersections, continuation around a loop accumulates sign-flips in eigenvector frames, enabling robust detection of such singularities.
• Random ensemble results: In SG$^+$ pencils (structured random models), the expected number of conical intersections grows like a power law in $n$, with exponent sensitive to matrix bandwidth.

6. Applications: Optimization, Convexity, and Matrix Inequalities

Hermitian linear pencils $L(x) = A_0 + \sum_{i=1}^g A_i x_i$ are central to semidefinite optimization, matrix convexity, and real algebraic geometry (Volčič, 2024):

• Free spectrahedra: The matricial positivity domain $\mathcal{D}_L$ is matrix convex, and every convex free semialgebraic set arises as a free spectrahedron for some Hermitian pencil (Helton–McCullough theorem).
• Detection and representation algorithms: Deciding whether a positivity domain is a free spectrahedron and constructing the representing pencil reduces to noncommutative realization theory, block-triangular decompositions, and semidefinite programming.
• Positivstellensatz and eigenvalue optimization: Certificates for positivity on $\mathcal{D}_L$ are sums of squares plus LMI terms. Eigenvalue optimization subject to pencil constraints is strongly dual via SDP representations.
• Broader impacts: These pencils underlie LMIs for control, operator system state spaces, hyperbolic polynomial determinantal representations, and matrix relaxation hierarchies in quantum information and real algebraic geometry.

7. Jacobi-Type and Infinite-Dimensional Pencils

In spectral and difference/differential equation theory, symmetric pencils of the form $J_5 - \lambda J_3$ generalize classical orthogonal polynomial recurrence and provide a framework for matrix-valued orthogonality and higher-order difference (and differential) operators (Zagorodnyuk, 2018, Zagorodnyuk, 2017):

• Fourth-order difference equations: Solution spaces are built from associated and shifted polynomials, with explicit orthogonality relations and recurrence formulas.
• Spectral measures and inverse problems: Existence and uniqueness of spectral functions are characterized via operator representations, and inverse spectral problems are solved through integral models and moment problems.
• Matrix orthogonality and perturbation theory: Perturbed classical polynomial systems (e.g., Jacobi) lead to families satisfying fourth-order differential equations, summarizing bispectral duality and block-Jacobi realizations.

Table: Canonical Types of Block-Symmetric Linearizations (finite-degree polynomials)

Family	Pencil Structure	Symmetry Condition
DL$(P)$ (Dopico et al., 2022)	$L(z) = zL_1 + L_0$ (see block formula)	$P^T(z) = P(z)$ (coeff.)
Block-Kronecker (odd) (Faßbender et al., 2016)	$$\begin{bmatrix} \lambda B + A & o\,L_s(\lambda)^T \ L_s(\lambda) & 0 \end{bmatrix}$$	Diagonal/off-diagonal symmetry
Block-Kronecker (even) (Faßbender et al., 2016)	$$\begin{bmatrix} o\,\widehat{L}_t(\lambda) & \lambda B+A \ \lambda B+A & \widehat{L}_t(\lambda)^T \end{bmatrix}$$	Block-symmetry in each $M_{ij}$
Double-ansatz (Bist et al., 2024)	Unique block-symmetric (cubic-quadratic, etc.)	$A_i^T = A_i, D_j^T = D_j$
Four canonical Fiedler-like (Cachadina et al., 2017)	Permutationally block-congruent to Kronecker form	Block-symmetric and AS condition

Each listed family supports structure-preserving recovery of all spectral quantities under mild regularity and exclusion hypotheses.

---

In summary, symmetric linear pencils serve as foundational constructs for the geometry, spectral theory, and optimization of matrix systems. Their classification, structure-preserving linearizations, and rich spectral behavior underpin significant advances in both theoretical frameworks and computational methodologies across algebra, geometry, and applied mathematics.