Coprime Matrix-Fraction Descriptions

• Coprime matrix-fraction descriptions are factorizations of transfer matrices into minimal numerator and denominator matrices ensuring unique, robust system representations.
• They employ canonical forms, Bezout identities, and recursive pole-dislocation algorithms to guarantee minimality and coprimeness in system realizations.
• Applications include robust controller synthesis, network interpretations of LTI systems, and efficient numerical algorithms for time-delay and distributed parameter systems.

Coprime matrix-fraction descriptions (MFDs) are foundational representations in systems and control theory, particularly for the analysis and synthesis of linear systems with matrix-valued rational, polynomial, or meromorphic transfer functions. An MFD represents a transfer matrix as the product or quotient of two polynomial or analytic matrices, with "coprimeness" ensuring minimal and non-redundant factorization. These constructions are central for canonical forms, minimality, robust controller parameterizations, network representations, and numerical algorithms for state-space and time-delay systems.

1. Matrix-Fraction Descriptions: Definitions and Variants

A matrix-fraction description of a rational matrix $G(λ) \in \mathbb{C}(λ)^{m \times n}$ is a factorization into polynomial or analytic matrices:

• Right MFD: $G(λ) = N(λ) D(λ)^{-1}$, with $N(λ) \in \mathbb{C}[λ]^{m \times n}$ and $D(λ) \in \mathbb{C}[λ]^{n \times n}$ regular ($\det D(λ) \not\equiv 0$).
• Left MFD: $G(λ) = \tilde{D}(λ)^{-1} \tilde{N}(λ)$, with $\tilde{N}(λ) \in \mathbb{C}[λ]^{m \times n}$ and $\tilde{D}(λ) \in \mathbb{C}[λ]^{m \times m}$ regular.

Both forms arise naturally from polynomial matrix descriptions (PMDs) or analytic matrix descriptions (AMDs) of system transfer functions. In time-delay or time-invariant system analysis, the same framework generalizes with $A(λ)$, $B(λ)$, $C(λ)$, $D(λ)$ replaced by analytic (meromorphic or holomorphic) matrix-valued functions (Alam et al., 7 Dec 2025).

2. Coprimeness: Algebraic and System-Theoretic Interpretation

A pair $(N, D)$ is coprime if their only common right divisor is unimodular (i.e., invertible over the appropriate matrix ring). Equivalent characterizations include:

• Bezout Identity: There exist polynomial matrices $X(λ)$, $Y(λ)$ such that $X(λ)N(λ) + Y(λ)D(λ) = I_n$.
• Rank Condition: For all $λ$, $[\;N(λ)\;\; D(λ)\;]$ has full row rank $n$.
• Smith Form Criterion: The concatenated matrix $[N\;\; D]$ has unimodular invariant factors. These algebraic properties guarantee absence of pole-zero cancellations, uniqueness (up to unimodular transformations), and minimality of representation (Alam et al., 7 Dec 2025, Sabau et al., 2013, Varga, 2017).

In system-theoretic terms, for PMDs and AMDs, coprime MFDs correspond to irreducible (controllable and observable) system realizations. Minimal right coprime MFDs achieve the least possible degree in $D(λ)$, with the degree corresponding to the aggregate pole multiplicity of $G(λ)$ (Alam et al., 7 Dec 2025).

3. Construction of Coprime MFDs: Canonical Forms and Algorithms

Several methodologies exist for constructing coprime MFDs, leveraging algebraic, analytic, or computational techniques:

a) Direct Approach via Smith–McMillan Form

For $G(λ)$ of full normal rank $p$:

• Compute $$G(λ) \sim_{\rm loc} \mathrm{diag} \left(\frac{φ_1(λ)}{ψ_1(λ)}, \ldots, \frac{φ_p(λ)}{ψ_p(λ)}, 0_{m-p,n-p}\right)$$, where $\{φ_i, ψ_i\}$ are invariant zero/pole polynomials.
• Set $N_φ(λ) = \mathrm{diag}(φ_1, ..., φ_p) \oplus 0$, $D_ψ(λ) = \mathrm{diag}(ψ_1, ..., ψ_p) \oplus I$.
• Obtain $$G(λ) = E(λ) N_φ(λ) D_ψ(λ)^{-1} F(λ) = N(λ) D(λ)^{-1}$$ (Alam et al., 7 Dec 2025).

b) Block Decomposition via System Matrices

Given an analytic (or polynomial) system matrix

$$H(λ) = \begin{bmatrix} A(λ) & B(λ) \ -C(λ) & D(λ) \end{bmatrix},$$

define $W(λ) = [A(λ)\ \ B(λ)]$ and use its Smith form decomposition to extract a structure matrix $D_R(λ)$. This yields

$$G(λ) = N(λ) D(λ)^{-1}, \quad N(λ) = [-C(λ)\quad A(λ)] E(λ) D_R(λ).$$

This description systematically identifies the coprime denominator that accumulates all pole factors (Alam et al., 7 Dec 2025).

c) Recursive Pole-Dislocation Algorithms

For general rational matrices, the recursive pole-dislocation algorithm described in (Varga, 2017) uses descriptor realizations $(A - λE, B, C, D)$ and recursively transforms the realization via orthogonal (block Schur) forms and feedback updates. At each step, eigenvalues in a prescribed "bad" region ($Ω_b$) are shifted to a "good" region ($Ω_g$) by constructing elementary denominator factors, preserving minimality and coprimeness. The total McMillan degree of the denominator matches the aggregate multiplicity of poles in $Ω_b$, and the entire process is backward stable and efficient ($O(n^3)$ in state dimension).

Algorithmic Steps for Recursive Construction:

1. Transform to block Schur/Jordan form segregating "good" and "bad" poles.
2. For each "bad" block, apply feedback to relocate it to "good" region, updating the denominator factor.
3. Repeat until all poles are "good"; then $N(λ), M(λ)$ comprise a minimal coprime factorization (Varga, 2017).

4. Structural and Network Interpretations

Structured and sparse coprime MFDs, especially left coprime factorizations, have been shown to possess strong connections to the network structure of LTI systems. Specifically, in the Dynamical Structure Function (DSF) formalism, the sparsity pattern of coprime factors directly encodes allowable signal pathways and interconnections:

• Cascade, Ring, or Network Implementations: Sparsity in $M(s)$ and $N(s)$ allows interpretation as a network of interconnected LTI subsystems, with each nonzero entry mapping to a physical or logical connection (Sabau et al., 2013).
• DSF Correspondence: Every viable DSF pair translates to a stable left coprime factorization, and vice versa. The equivalence enables systematic design of distributed control architectures with prescribed topology.
• Uniqueness: Coprime factorizations are unique up to unimodular transformation (right multiplication by a constant invertible matrix). This parametric freedom allows for families of structured realizations consistent with different network topologies (Sabau et al., 2013).

5. Minimality, Irreducibility, and Canonical Forms

Minimal coprime MFDs are characterized by the least possible degree in the denominator:

• The degree $\ell$ of $\det D(λ)$ (for right MFD) is minimized and equals the sum of pole multiplicities of $G(λ)$.
• Irreducibility: The associated system AMD or PMD must be irreducible, i.e., both controllable and observable.
• Canonical Form: The Smith–McMillan canonical factorization yields diagonal numerator and denominator polynomials (invariant zeros and poles).
• Bezout Certificates: The coprimeness condition is constructively witnessed by polynomials $X(λ), Y(λ)$ solving the Bézout identity $X(λ)N(λ) + Y(λ)D(λ) = I_n$.

This structure underpins the use of MFDs in robust controller synthesis, disturbance decoupling, and fault detection, where minimal and irreducible forms guarantee no hidden local dynamics (Alam et al., 7 Dec 2025, Varga, 2017).

6. Extensions: Analytic Matrix Descriptions and Time-Delay Systems

The theoretical framework and results for PMDs generalize to analytic matrix descriptions (AMDs), extending the theory of MFDs to meromorphic (holomorphic) system matrices. Under these generalizations:

• Transfer functions with delays or distributed parameters can be analyzed.
• Coprime MFD construction (including minimal degree and structural results) carries over, facilitating analysis and synthesis of time-delay systems (Alam et al., 7 Dec 2025).

7. Illustrative Examples and Computational Aspects

Concrete construction is illustrated by explicit examples. For instance, for

$$G(λ) = \begin{bmatrix} \frac{λ+1}{λ} & \frac{1}{λ} \ \frac{2}{λ+2} & \frac{λ-1}{λ+2} \end{bmatrix},$$

the right-coprime MFD is obtained as

$$N(λ) = \operatorname{diag}(λ+1, λ-1), \quad D(λ) = \operatorname{diag}(λ, λ+2).$$

The Bézout identity is satisfied by appropriately constructed polynomial matrices $X(λ), Y(λ)$ (Alam et al., 7 Dec 2025).

From a computational perspective, modern factorization algorithms utilize backward-stable, orthogonal transformations (e.g., Householder, QZ algorithms), and operate at cubic computational complexity in system order. Recursive pole-dislocation achieves flexibility for improper or singular systems, generalizes to both continuous and discrete time, and accommodates arbitrary "good" regions for pole placement (Varga, 2017).

---

References:

For detailed theory, algorithms, and proofs see (Alam et al., 7 Dec 2025, Sabau et al., 2013, Varga, 2017). Classical background is given in works such as T. Kailath, “Linear Systems,” Prentice‐Hall, 1980; H. H. Rosenbrock, “State‐Space and Multivariable Theory,” Wiley, 1970; and A. I. Vardulakis, “Linear Multivariable Control,” Wiley, 1991.