🔢 mathematics

Reachable and observable sets for switched systems via generalized Lyapunov equations: application to switched descriptor systems

This paper demonstrates that the solutions to the generalized Lyapunov equations proposed for model order reduction of switched descriptor systems effectively enclose the system's reachable and observable sets, thereby theoretically validating their suitability for balancing-based reduction.

Original authors: Mattia Manucci, Benjamin Unger

Published 2026-08-05

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Mattia Manucci, Benjamin Unger

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to steer a massive, complex machine—like a robot arm or a power grid—that doesn't just run smoothly all the time. Instead, it has different "modes" of operation, like shifting gears in a car or switching between walking and running. Sometimes, when it switches modes, the machine doesn't just change its speed; it might suddenly jerk, jump, or even produce a tiny, sharp "impulse" (like a sudden electric spark) before settling into the new rhythm. In the world of engineering, these are called switched systems. The challenge is that these machines are often described by equations that mix normal movement (differential equations) with rigid constraints (algebraic equations), making them incredibly hard to simulate on a computer. If the machine is too complex, the computer chokes trying to calculate every single detail.

To fix this, engineers use a trick called Model Order Reduction (MOR). Think of it like creating a "simplified sketch" of the machine. Instead of simulating every tiny bolt and wire, you want to find the most important parts that actually move and react to your controls, and ignore the rest. To know which parts are important, you need to map out the reachable set (all the places the machine can go if you push the buttons) and the observable set (all the parts you can "see" or measure from the outside). The big question is: how do you find these maps for a machine that jumps and sparks when it switches gears? This is the puzzle that Mattia Manucci and Benjamin Unger tackle in their recent work.

The Paper's Story: Mapping the Jumping Machine

In this paper, Manucci and Unger act like cartographers trying to draw a map of a very tricky territory: a machine that switches between different behaviors and occasionally jumps or sparks. Their goal is to prove that a specific mathematical tool, called Generalized Lyapunov Equations (GLEs), can successfully draw the boundaries of this territory.

Here is the core of their discovery: They show that if you solve these specific GLEs, the resulting mathematical "shapes" (called image sets) are guaranteed to enclose the actual reachable and observable sets of the system.

To understand this, imagine you are trying to find the exact area a dog can run in a park. The dog is fast, but it also has a leash that sometimes snaps tight and yanks it in a new direction (the jumps and impulses). Calculating the dog's exact path is a nightmare because of the sudden yanks. However, the authors prove that the GLEs act like a giant, slightly loose safety net. If you throw this net over the park, it will definitely cover every spot the dog could possibly reach, even if the net is a bit bigger than the exact area.

Why is this "bigger net" useful? Because in engineering, you don't always need the exact boundary to build a good simplified model. You just need to know that the important parts are inside the boundary you found. By proving that the GLE solutions always contain the true reachable and observable sets, the authors justify using these equations to create simplified models. If a part of the machine is inside the GLE net, it's worth keeping in your simplified sketch. If it's outside, you can safely ignore it.

How They Proved It

The authors didn't just guess; they built a rigorous bridge between the messy, real-world system and the cleaner mathematical tool.

Reformulating the Mess: First, they took the original system (which has jumps and impulses) and rewrote it into a slightly different version. They showed that the "input-to-output" behavior (what you put in and what you get out) stays exactly the same, even though the internal state looks different. This allowed them to treat the jumps as if they were just part of the input or output, rather than a chaotic internal event.
The "No-Jump" Comparison: They then compared this messy system to a "clean" version of the same machine that doesn't have the jumps. They proved that the reachable and observable sets of the messy, jumping machine are always subsets of the sets for the clean machine. In other words, the jumping machine can't go anywhere the clean machine couldn't go (if you account for the extra inputs).
The GLE Connection: For the "clean" machine (without jumps), it is already known that the solutions to the GLEs perfectly match the reachable and observable sets.
The Conclusion: Since the messy machine's territory is inside the clean machine's territory, and the clean machine's territory is inside the GLE net, the messy machine's territory must also be inside the GLE net.

What They Don't Claim

It is important to note what this paper does not say. The authors are not claiming that the GLE solutions give you the exact reachable set. They explicitly state that the GLE solutions provide a set that encloses or contains the true set. The GLE net might be slightly larger than the actual area the machine can reach, but it will never be too small (it won't miss any reachable spots).

Furthermore, they do not claim to have invented a new way to solve these equations or to have solved the problem for every possible type of machine in the universe. Their work is a theoretical proof that justifies using an existing method (the GLEs from their previous work) for a specific, complex class of systems (switched descriptor systems with jumps). They rely on mathematical proofs and logic rather than new experimental data or simulations in this specific paper (though they reference numerical experiments in their prior work).

Why This Matters

For a curious teenager, think of this as finding a reliable shortcut. If you want to build a video game simulation of a complex robot, you don't want your computer to crash. You need to simplify the robot's code. But if you simplify it too much, the robot might behave weirdly or break. This paper gives engineers a "safety guarantee." It says, "Hey, if you use these specific math equations to decide what to keep and what to throw away, you are guaranteed that your simplified robot will still be able to do everything the real robot can do. You won't accidentally delete a crucial part."

By proving that these Generalized Lyapunov Equations always cover the necessary ground, the authors provide a solid foundation for making complex, switching, jumping systems easier to simulate and control, without losing the essential behavior that makes them work.

Technical Summary: Reachable and Observable Sets for Switched Systems via Generalized Lyapunov Equations

1. Problem Statement

The paper addresses the theoretical justification for using solutions to Generalized Lyapunov Equations (GLEs) in the context of Model Order Reduction (MOR) for switched differential-algebraic equations (DAEs). Specifically, it investigates switched systems of the form:
$\begin{cases} E_{q(t)}\dot{x}(t) = A_{q(t)}x(t) + B_{q(t)}u(t), \\ y(t) = C_{q(t)}x(t), \end{cases}$
where $q(t)$ is a switching signal, and the matrix $E_{q(t)}$ may be singular.

The core challenge lies in rigorously characterizing the reachable and observable sets of such systems. While recent work (Manucci & Unger, 2024) proposed using GLEs to derive a balancing-based MOR method for these systems, the theoretical link between the solutions of these GLEs and the actual reachable/observable subspaces of the switched DAE was not fully established. The paper aims to prove that the image sets of the GLE solutions always enclose the reachable and observable sets of the system, thereby validating their use as Gramians for MOR.

2. Methodology

2.1 Solution Space and Reformulation

The authors work within the space of piecewise-smooth distributions ( $\mathcal{D}^{\infty}_{pw}$ ) to handle the jumps and Dirac impulses that arise when switching between modes with different consistency spaces.

Using the Quasi-Weierstrass Form (QWF), the switched DAE is reformulated into an equivalent switched system of ordinary differential equations (ODEs) with state-dependent jumps and impulses at switching times. The reformulated system (8) is given by:
$\begin{cases} \dot{x}(t) = A^{\text{diff}}_{q_k}x(t) + B^{\text{diff}}_{q_k}u(t), & t \in (t_k, t_{k+1}), \\ x(t_k^+) = \Pi_{q_k}x(t_k^-) + \Pi_{q_k}B_{q_{k-1}}U_{q_{k-1}}(t_k^-), \\ y(t) = C^{\text{diff}}_{q_k}x(t) + D_{q_k}U_{q_k}(t), \\ y[t_k] = \text{impulsive terms depending on } x(t_k^-). \end{cases}$
Here, $\Pi_j$ represents the consistency projector, and the input $U$ includes derivatives of the control signal up to the nilpotency index of the system.

2.2 Characterization of Reachability and Observability

The paper rigorously defines the reachable and observable sets for this reformulated system:

Reachability: The authors define a "modified reachable set" that accounts for the input-dependent jumps. They prove (Lemma 3.3) that the reachable set can be computed recursively. Crucially, Theorem 3.4 demonstrates that the reachable set of the system with input-state-dependent jumps is contained within the reachable set of a simplified system with state-dependent jumps and an augmented input matrix.
Observability: Similarly, the unobservable set is characterized recursively (Lemma 4.3), accounting for impulsive output terms. Theorem 4.5 shows that the unobservable set of the original system is equivalent to that of a system without output impulses but with an augmented output matrix.

2.3 Connection to Generalized Lyapunov Equations

The authors introduce a further simplified switched linear system (35) where:

Jumps are purely state-dependent (no input-dependent jumps).
Inputs and outputs are augmented to capture the effects of the original system's impulses.
The projection matrices $\Pi_j$ are treated generally.

Theorem 5.1 establishes a critical inclusion relationship:

The reachable set $R$ of the system with jumps is a subset of the reachable set $\tilde{R}$ of the system without jumps (but with augmented matrices).
The observable set $O$ is a subset of the observable set $\tilde{O}$ of the jump-free system.

For the jump-free system with augmented matrices, it is a known result (cited from [13]) that the reachable and observable sets coincide with the image spaces of the solutions $P$ and $Q$ to the following Generalized Lyapunov Equations:
$\begin{aligned} AP + PA^T + \sum_{j=1}^M (F_j P F_j^T + \tilde{B}_j \tilde{B}_j^T) &= 0, \\ A^T Q + QA + \sum_{j=1}^M (F_j^T Q F_j + \tilde{C}_j^T \tilde{C}_j) &= 0. \end{aligned}$

3. Key Contributions and Results

Rigorous Characterization: The paper provides a formal definition and recursive characterization of reachable and observable sets for switched DAEs within the space of piecewise-smooth distributions.
Inclusion Theorems: The primary theoretical result is the proof that the reachable and observable sets of the complex switched DAE (with jumps and impulses) are subsets of the reachable and observable sets of a specific augmented switched ODE system.
Validation of GLEs: By combining the inclusion theorems with existing results on GLEs for switched ODEs, the authors prove that the solutions $P$ and $Q$ of the GLEs proposed in [12] satisfy:
$\text{Reachable Set} \subseteq \text{span}(P) \quad \text{and} \quad \text{Observable Set} \subseteq \text{span}(Q).$
This confirms that the image of the GLE solutions "encloses" the true system subspaces.

4. Significance and Claims

The paper claims to provide the theoretical foundation necessary to justify the use of GLE-based balancing for Model Order Reduction in switched DAEs.

Justification for MOR: The authors state that because the GLE solutions enclose the reachable and observable sets, they are suitable for use as Gramians in balancing-based MOR. This ensures that the reduced-order model preserves the input-output behavior of the original switched DAE system.
Scope of Claims: The authors are modest in their claims. They do not propose new numerical algorithms or experimental applications in this specific work. Instead, they clarify why a method proposed in their previous work ([12]) works. They explicitly note that the reformulation does not change the input-output map, which is the primary goal of MOR.
Limitations: The paper acknowledges that the reformulation involves augmented matrices that depend on adjacent modes, which could lead to a system with $M(M-1)^2$ modes. However, they note that the image sets of the GLE solutions remain consistent regardless of this specific formulation.

In summary, the paper bridges the gap between the abstract theory of switched DAEs (involving distributions and impulses) and the practical application of GLEs for model reduction, proving that the latter correctly captures the essential dynamics of the former.