1 Definition and basic concepts

The Weierstrass canonical form is a normal form for square matrix pencils that reveals their essential algebraic structure under invertible changes of basis. It is used to describe generalized eigenvalues, the behavior at infinity, and the separation between regular and singular components of a pencil. In practice, the form gives a precise way to compare pencils that differ only by equivalence transformations.

1.1 Matrix pencils

A matrix pencil is an expression of the form \(A-\lambda B\), where \(A\) and \(B\) are matrices of the same size and \(\lambda\) is a scalar parameter. Such objects arise naturally in generalized eigenvalue problems, where one seeks nonzero vectors \(x\) satisfying \[ (A-\lambda B)x=0. \] Unlike an ordinary eigenvalue problem, a pencil can encode both finite spectral data and a possible eigenvalue at infinity. This makes it especially useful for systems whose algebraic constraints affect the solution structure.

1.2 Equivalence transformations

Two pencils \(A-\lambda B\) and \(A'-\lambda B'\) are equivalent if there exist invertible matrices \(P\) and \(Q\) such that \[ A' = PAQ,\qquad B' = PBQ. \] These transformations correspond to independent changes of basis in the domain and codomain. They preserve the intrinsic structure of the pencil, including its eigenvalues, elementary divisors, and minimal indices. The Weierstrass form classifies pencils up to this relation.

1.3 Regular and singular pencils

A pencil is called regular if \(\det(A-\lambda B)\) is not the zero polynomial. In that case, the pencil has a well-defined finite spectral picture together with a possible contribution at infinity. If \(\det(A-\lambda B)\equiv 0\), the pencil is singular, meaning that its rank deficiency persists for all values of \(\lambda\). Singular pencils require additional invariants beyond eigenvalues, such as minimal indices, to describe their full structure.

1.4 Canonical representation

A canonical representation is a standardized matrix form that every object in a class can be reduced to by allowed transformations. For regular matrix pencils, the Weierstrass canonical form isolates the finite and infinite spectral parts into separate blocks. This decomposition makes the essential features immediately visible and provides a foundation for analysis and computation.

2 Statement of the Weierstrass canonical form

For a regular square pencil \(A-\lambda B\), there exist invertible matrices \(P\) and \(Q\) such that \[ P(A-\lambda B)Q \] is block diagonal with a part describing finite eigenvalues and a part describing the eigenvalue at infinity. The finite part is built from Jordan blocks associated with eigenvalues, while the infinite part is built from nilpotent blocks. This decomposition is unique up to permutation of blocks of the same type and size.

2.1 Regular part

The regular part of the canonical form consists of Jordan blocks for each finite eigenvalue. If \(\mu\) is an eigenvalue, the corresponding blocks record both its algebraic multiplicity and its chain structure. These blocks look like the familiar Jordan blocks from ordinary linear algebra, but they arise in the pencil setting through the pair \((A,B)\) rather than a single matrix.

2.2 Singular part

The singular part in the broader theory is associated with nonregular behavior, where the pencil fails to be invertible as a polynomial matrix. In the Weierstrass setting for regular pencils, the infinite spectral contribution plays an analogous role to a singular component. For genuinely singular pencils, the Weierstrass form is supplemented by the Kronecker canonical form, which includes additional blocks determined by minimal indices.

2.3 Block decomposition

The canonical form separates the pencil into blocks, each of which captures a specific invariant feature. This block decomposition is what makes the form so effective: local spectral information is isolated inside manageable submatrices. The result is a direct sum of canonical pieces that can be studied independently.

2.3.1 Finite eigenvalue blocks

Each finite eigenvalue \(\mu\) contributes one or more Jordan blocks of the form \[ J_k(\mu)= \begin{pmatrix} \mu & 1 & 0 & \cdots & 0\\ 0 & \mu & 1 & \ddots & \vdots\\ \vdots & \ddots & \ddots & \ddots & 0\\ 0 & \cdots & 0 & \mu & 1\\ 0 & \cdots & \cdots & 0 & \mu \end{pmatrix}, \] with sizes determined by the lengths of Jordan chains. These blocks encode both the eigenvalue and the degree of nontrivial coupling among generalized eigenvectors.

2.3.2 Infinite eigenvalue blocks

The infinite eigenvalue is represented by nilpotent Jordan blocks, usually written in terms of a nilpotent matrix \(N\) with ones on the superdiagonal and zeros elsewhere. Such blocks correspond to the behavior of the pencil under inversion of the spectral parameter and describe how the system behaves near \(\lambda=\infty\). Their sizes measure the depth of the infinite chains.

3 Relation to generalized eigenvalue problems

The Weierstrass form is closely tied to generalized eigenvalue problems, where one studies solutions of \(Ax=\lambda Bx\). By transforming the pencil into canonical blocks, one can read off the spectrum and the chain structure directly. This makes the form a powerful theoretical and computational tool.

3.1 Finite eigenvalues

A finite eigenvalue \(\lambda_0\) is a scalar for which \(A-\lambda_0 B\) is singular. In the canonical form, each such value appears in one or more Jordan blocks. The total size of the blocks attached to \(\lambda_0\) equals its algebraic multiplicity in the characteristic polynomial of the regular part.

3.2 Eigenvalue at infinity

The eigenvalue at infinity is associated with singularity of the leading coefficient \(B\) when the pencil is viewed as a polynomial in \(\lambda\). It is detected by considering the reversed pencil \(\lambda A-B\). In the canonical form, infinity is represented by nilpotent structure, which captures the failure of the pencil to behave like an ordinary eigenvalue problem at large parameter values.

3.3 Algebraic multiplicity and Jordan chains

Algebraic multiplicity counts how often an eigenvalue occurs as a root of the characteristic polynomial. Jordan chains refine this count by describing how generalized eigenvectors link together. In a pencil, these chains are determined by repeated nullspace conditions involving powers of the shifted pencil, and they correspond exactly to the sizes of the Jordan blocks in the canonical form.

4 Construction of the canonical form

Constructing the Weierstrass canonical form involves reducing the pencil through equivalence transformations until its invariant structure becomes explicit. In theory, this reduction is exact; in applications, one often seeks a numerically stable approximation. The process typically begins by separating regular from singular behavior and then identifying eigenvalue blocks.

4.1 Reduction by equivalence

The first step is to apply invertible transformations to simplify the matrices simultaneously. Row and column operations adapted to the pair \((A,B)\) can isolate rank-deficient parts and reveal the regular core. Once a regular subpencil is identified, further reductions place it into Jordan-like blocks.

4.2 Computation of invariant subspaces

Invariant subspaces associated with a pencil generalize eigenspaces for ordinary matrices. Computing them involves solving nullspace problems for shifted pencils and their powers. These subspaces organize the generalized eigenvectors into chains and provide the subspace decomposition needed for canonical reduction.

4.3 Jordan block determination

The sizes of Jordan blocks are determined from the lengths of chains and the ranks of successive powers of the shifted pencil. In exact algebraic settings, this information can be derived from divisibility properties of the determinantal polynomials. Each block size records the degree of non-diagonalizability at a specific eigenvalue.

4.4 Numerical methods

In numerical linear algebra, exact canonical forms are often replaced by stable decompositions that approximate the same structure. Algorithms may use QZ-type methods, rank-revealing decompositions, or staircase reductions to expose finite and infinite eigenvalues. Because Jordan structure is sensitive to perturbations, numerical procedures usually focus on spectral separation and deflating subspaces rather than exact canonical blocks.

5 Invariants of a matrix pencil

The canonical form is valuable because it displays invariants that do not change under equivalence transformations. These invariants determine the equivalence class of the pencil and provide a complete structural description in the regular case. For singular pencils, additional combinatorial data are needed.

5.1 Elementary divisors

Elementary divisors are the powers of irreducible factors that appear in the invariant factor decomposition of the pencil. For a regular pencil over an algebraically closed field, they correspond to powers of linear factors associated with eigenvalues. Each elementary divisor matches a Jordan block size and therefore records local spectral multiplicity.

5.2 Minimal indices

Minimal indices measure the degrees of polynomial vectors in the left and right nullspaces of a singular pencil. They quantify the freedom in nonunique solutions and represent structural information that cannot be captured by eigenvalues alone. In the Kronecker theory of singular pencils, minimal indices are essential invariants.

5.3 Kronecker structure

The Kronecker structure combines elementary divisors and minimal indices into a full classification of matrix pencils. It extends the Weierstrass description from regular pencils to singular ones. The resulting canonical decomposition separates finite eigenvalues, infinite eigenvalues, and singular blocks in a unified framework.

6 Special cases

Several important special classes of pencils admit simplified interpretations of the Weierstrass form. These cases often occur in applications and allow one to infer structure more directly. Some special symmetries also constrain the allowable canonical blocks.

6.1 Nonsingular matrices

If \(B\) is invertible, then the generalized eigenvalue problem \(Ax=\lambda Bx\) reduces to the ordinary eigenvalue problem for \(B^{-1}A\). In that case, the pencil has no eigenvalue at infinity, and the Weierstrass form essentially coincides with the Jordan canonical form of \(B^{-1}A\). This is the simplest regular situation.

6.2 Hermitian and symmetric pencils

For Hermitian or symmetric pencils, additional structural relations hold between the matrices \(A\) and \(B\). These relations restrict the location and multiplicity of eigenvalues and often lead to more organized canonical patterns. In many theoretical settings, symmetry also improves interpretability and can support more stable computational methods.

6.3 Singular pencils

Singular pencils require the full machinery of generalized canonical classification. Their structure includes not only spectral data but also left and right minimal indices. These pencils model systems with algebraic dependencies or nonunique solution spaces, and their analysis is more intricate than that of regular pencils.

7 Applications

The Weierstrass canonical form is widely used whenever a problem is naturally expressed in terms of a matrix pencil rather than a single matrix. It is especially important when constraints, hidden variables, or differential terms produce generalized eigenvalue structures. Its applications range from pure linear algebra to systems theory.

7.1 Differential-algebraic equations

Differential-algebraic equations often lead to pencils of the form \(E\dot{x}=Ax\), where the matrix \(E\) may be singular. The Weierstrass form separates dynamic modes from algebraic constraints and helps identify consistent initial conditions. It also clarifies the index-related behavior of the system and the role of infinite eigenvalues.

7.2 Control theory

In control theory, matrix pencils arise in stability analysis, feedback design, and descriptor system models. The canonical form helps distinguish controllable and uncontrollable components and reveals modes that cannot be influenced in the usual way. It also supports the study of poles and structural decompositions of state-space models.

7.3 Structural analysis of linear systems

For linear systems with constraints or coupled equations, the canonical form provides a framework for identifying independent variables and redundant relations. It can be used to separate regular dynamics from algebraic dependencies and to diagnose degeneracies in the model. This structural viewpoint is important in modeling, simulation, and system reduction.

The Weierstrass canonical form belongs to a family of classification results that put algebraic objects into standard shapes. It is closely related to several other canonical forms from linear algebra and module theory. Each of these forms serves a different but complementary purpose.

8.1 Jordan canonical form

The Jordan canonical form classifies a single square matrix up to similarity. It organizes eigenvalues and Jordan chains into blocks, much like the regular part of the Weierstrass form. The key difference is that the Weierstrass form applies to a pair of matrices and therefore captures generalized eigenvalue behavior for pencils.

8.2 Kronecker canonical form

The Kronecker canonical form extends the Weierstrass form to all pencils, including singular ones. It adds blocks that encode minimal indices and other singular features. In effect, the Weierstrass form can be viewed as the regular portion of the broader Kronecker classification.

8.3 Smith normal form

The Smith normal form is a diagonal form for matrices over a principal ideal domain, including polynomial matrices. It reveals invariant factors and divisibility structure in a way that is closely connected to elementary divisors. For matrix pencils, Smith-type factorization ideas underlie the algebraic classification of spectral data.

9 Historical background

The development of the Weierstrass canonical form reflects the broader emergence of structural methods in linear algebra and function theory. Its importance grew as mathematicians sought invariant classifications for matrix equations and differential systems. Over time, it became a standard reference point in the study of pencils.

9.1 Karl Weierstrass and canonical classification

Karl Weierstrass is associated with early work on the classification of linear relations and polynomial matrices. The canonical form named after him reflects his influence on the invariant theory of algebraic objects. His ideas helped establish the principle that a suitable normal form can reveal the essential content of an equation or operator.

9.2 Later developments in matrix pencil theory

Later research extended the original classification into a systematic theory of matrix pencils, including singular cases and numerical algorithms. The introduction of the Kronecker canonical form and related decomposition methods broadened the scope of the subject. Modern applications in differential equations and control further strengthened the role of canonical pencil theory in applied mathematics.