Singular Perturbation Lyapunov Function

• Singular Perturbation Lyapunov Function is a composite strategy that constructs individual Lyapunov functions for slow and fast subsystems, ensuring overall stability and convergence.
• The methodology employs convex combinations and structured quadratic forms to manage interconnection bounds and guarantee fixed-time or input-to-state stability.
• Applications span deterministic, stochastic hybrid, and switched systems, demonstrating robust performance despite complex time-scale separations.

A Singular Perturbation Lyapunov Function is a composite Lyapunov candidate used to verify stability, robustness, and convergence properties in dynamical systems exhibiting multiple distinct time scales, typically parameterized by a small positive parameter $\varepsilon$. Such systems naturally arise in various engineering, physics, and control contexts and are formally expressed as two (or more) coupled subsystems: the so-called slow and fast (boundary-layer) dynamics. The singular perturbation Lyapunov framework leverages appropriately constructed Lyapunov functions for each time scale and combines them—via convex combinations or structured quadratic forms—to obtain system-level stability guarantees, including asymptotic, fixed-time, or input-to-state stability, even under disturbances, stochastic effects, or hybrid (continuous/discrete) switching.

1. Mathematical Structure of Singularly Perturbed Systems

Singularly perturbed models partition the state vector $$y=(x, z) \in \mathbb{R}^{n_1} \times \mathbb{R}^{n_2}$$ into “slow” states $x$ and “fast” states $z$, and feature dynamics of the prototypical form

$$\begin{aligned} \dot{x} &= f(x, z, u),\ \varepsilon\,\dot{z} &= g(x, z, u), \end{aligned}$$

where $u$ is an input and $\varepsilon > 0$ captures the time-scale separation. As $\varepsilon \to 0$, $z$ converges rapidly to a quasi–steady-state manifold $z = h(x)$, yielding two limiting subsystems:

• The reduced (slow) system: $$y=(x, z) \in \mathbb{R}^{n_1} \times \mathbb{R}^{n_2}$$0,
• The boundary-layer (fast) system: $$y=(x, z) \in \mathbb{R}^{n_1} \times \mathbb{R}^{n_2}$$1 with $$y=(x, z) \in \mathbb{R}^{n_1} \times \mathbb{R}^{n_2}$$2 and rescaled time $$y=(x, z) \in \mathbb{R}^{n_1} \times \mathbb{R}^{n_2}$$3.

Extensions to hybrid and stochastic settings include flow and jump sets, outer-semicontinuous and convex-valued right-hand sides, and random or measure-driven events, as formalized in (Poveda, 2023). The composite flow map is frequently written as $$y=(x, z) \in \mathbb{R}^{n_1} \times \mathbb{R}^{n_2}$$4.

2. Construction of Composite Lyapunov Functions

Standard methodology requires establishing individual Lyapunov (or Foster–Lyapunov) functions for the slow and fast subsystems:

• Fast subsystem: $$y=(x, z) \in \mathbb{R}^{n_1} \times \mathbb{R}^{n_2}$$5, positive definite in $$y=(x, z) \in \mathbb{R}^{n_1} \times \mathbb{R}^{n_2}$$6 (relative to the manifold $$y=(x, z) \in \mathbb{R}^{n_1} \times \mathbb{R}^{n_2}$$7), satisfying

$$y=(x, z) \in \mathbb{R}^{n_1} \times \mathbb{R}^{n_2}$$8

for some $$y=(x, z) \in \mathbb{R}^{n_1} \times \mathbb{R}^{n_2}$$9, and appropriate positive-definiteness and decrease-along-jumps conditions.

• Slow subsystem: $x$0, positive definite with respect to a compact set $x$1, satisfying

$x$2

where $x$3 is the reduced slow dynamics and $x$4 a target or neighborhood set.

The composite Lyapunov function is then constructed as

$x$5

or, equivalently, as a convex combination, with $x$6 determined according to the coupling strengths between subsystems. In the context of fixed-time or input-to-state stability, additional “mixed-power” or “half-power” transforms (e.g., $x$7) are employed to facilitate bounding the interconnection terms (Tang et al., 2024, Tang et al., 2024).

3. Sufficient Conditions and Main Theorems

Stability and convergence of the overall system via the composite Lyapunov approach hinge on several key sufficient conditions:

• Positive-definiteness and dissipativity of $x$8 and $x$9 individually (with possible power-type or squared dissipation, depending on the setting).
• Explicit interconnection bounds on cross-terms resulting from the slow–fast coupling, typically formulated as:

$z$0

with a requirement that at least one of the self-dissipation parameters (e.g., $z$1 or $z$2) be sufficiently negative.

• A “smallness” condition for $z$3: For fixed convex weight $z$4, there exists $z$5 such that, for all $z$6, the weighted quadratic form in the composite Lyapunov time-derivative is negative definite. For stochastic hybrid systems, analogous average or probabilistic decrease conditions on jumps are imposed (Poveda, 2023).
• In ISS or fixed-time ISS settings, the composite Lyapunov function $z$7 must satisfy

$z$8

with exponents $z$9, $$\begin{aligned} \dot{x} &= f(x, z, u),\ \varepsilon\,\dot{z} &= g(x, z, u), \end{aligned}$$0, and $$\begin{aligned} \dot{x} &= f(x, z, u),\ \varepsilon\,\dot{z} &= g(x, z, u), \end{aligned}$$1 (Tang et al., 2024, Tang et al., 2024). This structure guarantees uniform global fixed-time convergence to the equilibrium (or, in the ISS case, to a disturbance-dependent ball).

4. Applications: Stochastic Hybrid, Fixed-Time, and Input-to-State Stability

The singular perturbation Lyapunov method has been extended beyond classical deterministic ODEs to address:

• Stochastic hybrid dynamical systems, with both continuous flows (constrained differential inclusions) and jumps (constrained difference inclusions subject to random inputs), as in (Poveda, 2023). Here, composite Foster–Lyapunov and Lagrange–Foster functions yield criteria for uniform global asymptotic stability in probability (UGASp) of the compact set $$\begin{aligned} \dot{x} &= f(x, z, u),\ \varepsilon\,\dot{z} &= g(x, z, u), \end{aligned}$$2.
• Fixed-time and input-to-state stability (ISS) for nonlinear singularly perturbed systems, where solutions must converge in a time independent of initial state for vanishing disturbance, with explicit uniform upper bounds for settling time given by

$$\begin{aligned} \dot{x} &= f(x, z, u),\ \varepsilon\,\dot{z} &= g(x, z, u), \end{aligned}$$3

(Tang et al., 2024, Tang et al., 2024). This is achieved by imposing two-term power-type dissipation inequalities on $$\begin{aligned} \dot{x} &= f(x, z, u),\ \varepsilon\,\dot{z} &= g(x, z, u), \end{aligned}$$4 and $$\begin{aligned} \dot{x} &= f(x, z, u),\ \varepsilon\,\dot{z} &= g(x, z, u), \end{aligned}$$5 and explicitly controlling the coupling via the matrix $$\begin{aligned} \dot{x} &= f(x, z, u),\ \varepsilon\,\dot{z} &= g(x, z, u), \end{aligned}$$6 in the time-derivative of $$\begin{aligned} \dot{x} &= f(x, z, u),\ \varepsilon\,\dot{z} &= g(x, z, u), \end{aligned}$$7.

• Hybrid specializations such as Markov jump-linear systems and deterministic switched systems—with the theory recovering classical results (e.g., Kokotović–O’Malley, Saberi & Kokotović) for ODEs and extending to nonsmooth cases (Poveda, 2023).

5. Verification Procedure and Illustrative Examples

Application of the singular perturbation Lyapunov framework typically involves the following steps (see (Tang et al., 2024, Tang et al., 2024)):

1. Model Reduction: Write the original system, identify the quasi-steady-state manifold $$\begin{aligned} \dot{x} &= f(x, z, u),\ \varepsilon\,\dot{z} &= g(x, z, u), \end{aligned}$$8, and split into reduced and boundary-layer subsystems.
2. Lyapunov Construction: Find candidate $$\begin{aligned} \dot{x} &= f(x, z, u),\ \varepsilon\,\dot{z} &= g(x, z, u), \end{aligned}$$9, $u$0 satisfying regularity and dissipation inequalities (with or without explicit ISS/fixed-time extensions).
3. Interconnection Bounds: Bound $u$1, $u$2 in quadratic form and determine the corresponding constants $u$3.
4. Parameter Selection: Choose convex weight $u$4 (or $u$5) to satisfy self-dissipation requirements; compute $u$6 to guarantee negative definiteness.
5. Composite Lyapunov Derivative and Stability Assessment: Check that the required decrease condition (possibly stochastic/ISS/fixed-time) is met.
6. Explicit Convergence Bound: If applicable, read off explicit settling time from the derived dissipation rates.

Illustrative examples provided in (Tang et al., 2024, Tang et al., 2024) include scalar systems with nonlinearities, quadratic cost optimization with singular perturbations, and high-order interconnected systems, demonstrating uniform convergence times and quantitative robustness to bounded disturbance.

6. Specializations, Broader Connections, and Methodological Guidelines

The singular perturbation Lyapunov approach unifies several stability frameworks:

• Quadratic and nonsmooth Lyapunov theory: Classical quadratic Lyapunov functions are a special case for ODEs with linear/quadratic dynamics (Poveda, 2023).
• Composite method and power-type dissipation: Extension to fixed-time and ISS contexts utilizes two-term dissipation and convex combinations, requiring careful cross-term management (Tang et al., 2024, Tang et al., 2024).
• Guidelines: Researchers are advised to establish strong self-dissipation for individual subsystems, control cross-coupling via quadratic forms, judiciously select $u$7 and $u$8, and apply the composite Lyapunov dissipation to deduce global—or probabilistic—stability with explicit convergence rate or time bounds.

A plausible implication is that advances in singular perturbation Lyapunov theory continue to broaden the admissible classes of systems (including stochastic, hybrid, nonlinear, and fixed-time stable) and enable rigorous, rate-optimal stabilization and control techniques in multi-scale dynamical settings.