Chapter 41B: Kronecker Canonical Form and Applications¶

Prerequisites: Regular Pencils (Ch41A) · Jordan Canonical Form (Ch12) · Minimal Polynomials

Chapter Outline: From Regular to Singular Pencils → Definition of the Kronecker Canonical Form (KCF) → Core Components: Regular Part (Finite and Infinite) + Singular Parts → Column and Row Minimal Indices → Rank and Defect Structure of Pencils → Deep Connections with Minimal Polynomials → Applications: Structure of Solutions to Singular Linear Systems, Controllability and Observability in Control Theory, and Algebraic Elimination

Extension: The Kronecker Canonical Form is the ultimate invariant under pencil similarity transformations; it covers not only eigenvalue information but also the "null-space flow" after complete operator decoupling, providing an algebraic panorama for understanding any singular evolutionary system.

In Ch41A, we dealt with regular matrix pencils having non-zero determinants. However, in more general scenarios—such as multi-input multi-output (MIMO) control systems or underdetermined differential systems—the characteristic polynomial may vanish identically. The Kronecker Canonical Form (KCF) provides a perfect decomposition for these most general singular pencils by introducing Minimal Indices. This chapter reveals the algebraic aesthetics hidden behind singular structures.

41B.1 Singular Pencils and the KCF¶

Definition 41B.1 (Singular Pencil)

A matrix pencil \(A - \lambda B\) is Singular if its characteristic polynomial \(p(\lambda) = \det(A - \lambda B) \equiv 0\) for all \(\lambda \in \mathbb{C}\), or if \(A\) and \(B\) are rectangular matrices.

Theorem 41B.1 (Kronecker Canonical Form)

For any matrix pencil \(A - \lambda B\), there exist non-singular matrices \(P\) and \(Q\) such that it is reduced to a block diagonal form containing: 1. Regular Part: Consists of blocks identical to the Weierstrass form (finite and infinite eigenvalues). 2. Right Singular Part: Blocks \(L_{\epsilon_i}\) determined by column minimal indices \(\epsilon_i\). 3. Left Singular Part: Blocks \(L_{\eta_j}^T\) determined by row minimal indices \(\eta_j\).

41B.2 Geometric Meaning of Minimal Indices¶

Definition 41B.2 (Minimal Index)

The index \(\epsilon_i\) represents the minimum degree of a polynomial vector \(\mathbf{x}(\lambda)\) such that \(A(\lambda)\mathbf{x}(\lambda) = \mathbf{0}\). This quantifies the dynamic dimension of the null space as a function of the frequency \(\lambda\).

41B.3 Control Theory Applications¶

Application: Controllability Subspaces

In the system \(\dot{x} = Ax + Bu\), the structure is entirely determined by the Kronecker structure of the pencil \([sI - A \ | \ -B]\). - Column minimal indices correspond to Controllability Indices. - They determine the shortest time steps or the dimension of control gains required to drive the state to a target.

Exercises¶

1. [Basics] Determine the singularity of \(A - \lambda B = \begin{pmatrix} 1 & \lambda \\ 0 & 0 \end{pmatrix}\).

Solution

Calculation: 1. This is a \(2 \times 2\) matrix. 2. Determinant: \(\det = 1 \cdot 0 - \lambda \cdot 0 = 0\). 3. Since the determinant is 0 for all \(\lambda\). Conclusion: This is a singular matrix pencil.

2. [Kronecker Block] Write the form of the block \(L_1\) corresponding to a column minimal index \(\epsilon=1\).

Solution

Construction: By definition, \(L_\epsilon\) is an \(\epsilon \times (\epsilon+1)\) pencil. For \(\epsilon=1\), it is \(1 \times 2\): \(L_1(\lambda) = \begin{pmatrix} 1 & -\lambda \end{pmatrix}\). Verification: Its null-space vector is \(\mathbf{x}(\lambda) = (\lambda, 1)^T\), which has degree exactly 1.

3. [Properties] Prove: If a matrix pencil is rectangular (\(m \neq n\)), it is necessarily singular.

Solution

Analysis: 1. A regular pencil requires \(A, B\) to be square with a non-vanishing characteristic polynomial. 2. For rectangular matrices, the traditional determinant criterion cannot be defined. 3. In Kronecker theory, a rectangular pencil must have non-zero left or right null-space polynomial vectors. Conclusion: Rectangular pencils always contain singular parts.

4. [Minimal Index] If a column minimal index is 0, what is the corresponding block?

Solution

Conclusion: The \(L_0\) block is a \(0 \times 1\) block (algebraically appearing as a column vector of zeros). Significance: This represents a constant right null-space vector that is independent of \(\lambda\).

5. [Dimension Formula] Let \(n_{reg}, n_{left}, n_{right}\) be the dimensions of the KCF components. Prove their sum equals \(n\).

Solution

Proof: Since the KCF is obtained via non-singular transformations \(P(A-\lambda B)Q\), the transformations preserve matrix dimensions. The sum of the number of columns across all diagonal blocks must equal the number of columns in the original matrix.

6. [Calculation] Find the minimal indices of \(\begin{pmatrix} 1 & \lambda & 0 \\ 0 & 0 & 1 \end{pmatrix}\).

Solution

Analysis: 1. The rows are independent, so the rank is 2. 2. With 3 columns, there must be 1 column minimal index. 3. Find \(\mathbf{x}(\lambda)\) such that \(M\mathbf{x} = 0\). 4. Let \(\mathbf{x} = (x_1, x_2, x_3)^T\). 5. Row 2 implies \(x_3 = 0\). Row 1 implies \(x_1 + \lambda x_2 = 0\). 6. Setting \(x_2 = 1\) gives \(x_1 = -\lambda\). 7. \(\mathbf{x}(\lambda) = (-\lambda, 1, 0)^T\). Conclusion: The maximum degree is 1, so the column minimal index \(\epsilon_1 = 1\).

7. [Weierstrass vs KCF] How does the KCF encompass the Weierstrass Canonical Form?

Solution

Relationship: KCF is a superset of the Weierstrass form. When the pencil is square and regular, the singular parts (\(L\) blocks) vanish, and the KCF reduces to the Weierstrass form. KCF completes the theory for rank-deficient square cases and all rectangular cases.

8. [Stability] Are dynamical systems described by singular pencils stable?

Solution

Analysis: 1. Singular parts imply directions of "free flow" (null-space vectors). 2. Solutions in these directions can have arbitrary time envelopes (due to the freedom in \(\mathbf{x}(\lambda)\)). Conclusion: They are generally unstable or lack well-defined causal evolutionary properties.

9. [Application] Why compute the KCF in control system design?

Solution

Reasoning: The KCF reveals which modes are controllable by inputs (corresponding to \(L\) blocks) and which are inherent to the system (regular part). This provides the foundational structural evidence for determining controllability or performing decoupling design.

10. [Numerical] Why is numerical computation of the KCF more challenging than Jordan form?

Solution

Challenges: 1. Beyond sensitivity to eigenvalues, KCF requires determining subtle changes in rank. 2. Perturbations can cause jumps in minimal indices (e.g., from 1 to 0). Countermeasure: Typically, the GUPTRI algorithm (based on unitary transformations and staircase decompositions) is used to approximate the Kronecker structure.

Chapter Summary¶

The Kronecker Canonical Form is the ultimate expression of operator interference theory:

Classification of Singularity: It proves that singularity is not disordered chaos but is characterized by definite "minimal indices" and highly symmetric structural blocks.
Dynamic Null Spaces: By quantifying the degree of polynomial vectors, KCF elevates the static kernel concept to a dynamic feature evolving with frequency, establishing a complete basis for describing generalized system evolution.
Deconstruction of Systems: As the algebraic bedrock of control theory, KCF enables the total separation of controllable, observable, singular, and regular components, representing the highest level of linear system structural analysis.