Composite Lyapunov Functions

• Composite Lyapunov-like functions are scalar functions that integrate multiple local Lyapunov candidates to certify stability, safety, and convergence across various subsystems.
• They are constructed using techniques such as summation, max/min operations, smooth bump patching, and convex combinations to handle nonlinearities, switching dynamics, and high-dimensional systems.
• These composite functions enable scalable analysis in control systems, singular perturbation problems, and hybrid models, providing formal guarantees even when a single global Lyapunov function is intractable.

A composite Lyapunov-like function is a scalar-valued function constructed from multiple constituent Lyapunov-type functions—often each tailored to specific local regions, subsystems, dynamics, or constraints—with the composite function serving as a global or semi-global certificate for qualitative properties such as stability, invariance, safety, convergence, or input-to-state robustness. This approach subsumes sum, max/min, convex combinations, or smooth patching of "local" Lyapunov candidates and systematically extends to control barrier functions (CBFs), singular perturbations, hybrid or switching systems, and distributed/dimension-reduced Lyapunov analyses.

1. Core Definitions and Canonical Constructions

The composite Lyapunov paradigm encompasses a range of constructions unified by the joint use of multiple Lyapunov-type functions. Formally, suppose one has candidate Lyapunov functions $V_i : \mathbb{R}^n \to \mathbb{R}_+$ (for $i = 1, \ldots, N$), each certifying some desirable property (e.g., local attractivity, safety, convergence within a subsystem, or dissipation in a reduced order model). A composite function may be defined via one of:

• Sum: $W(x) = \sum_{i=1}^N w_i V_i(x)$, where $w_i > 0$ are weights.
• Max/min, or polyhedral (piecewise-linear) forms: $W(x) = \max_{i} V_i(x)$ or $W(x) = \min_{Q} \max_{i \in Q} V_i(x)$, as in path-complete graph frameworks and sectional polyhedral Lyapunov construction (Angeli et al., 2016, Berger et al., 2022, Philippe et al., 2017).
• Partition-of-unity / smooth bump patching: Using functions $\sigma_i(x)$ forming a locally supported, $C^1$ partition of unity, $W(x) = \sum_i \sigma_i(x) V_i(x)$, or through explicit bump function transitions, e.g., $W(x) = (1-b(x)) V_1(x) + b(x) V_2(x)$ (Liu et al., 2 Oct 2025).
• Hybrid/convex composition: In singular perturbations or fixed-time stability, the composite candidate $i = 1, \ldots, N$0, with $i = 1, \ldots, N$1, fuses slow and boundary layer Lyapunov functions (Tang et al., 2024, Tang et al., 2024).
• Neural network sum-of-blocks: In distributed or compositional analysis, networks parameterize blockwise local Lyapunov functions with the overall form $i = 1, \ldots, N$2, where each $i = 1, \ldots, N$3 is a state partition (Grüne, 2020, Liu et al., 2024).

The resulting composite function is then required to satisfy suitable Lyapunov inequalities (classically, decrease along system trajectories) globally or over a specified domain.

2. Motivation and Significance in Control, Verification, and Stability

The composite Lyapunov approach is indispensable in settings where the synthesis of a single global Lyapunov function is either intractable (e.g., due to high-dimensionality, non-convexity, non-smoothness, or switching/hybrid dynamics) or fundamentally impossible due to topological obstructions. Key motivations include:

• Dimension reduction and scalability: Aggregating low-dimensional Lyapunov "partials" into a global certificate allows the analysis of high-dimensional nonlinear networks where global SOS or SDP-based optimization is computationally prohibitive (Wang et al., 25 Jun 2025).
• Control systems with safety constraints: Composite Lyapunov-like functions unify CLF (stabilization) and CBF (safety/invariance) properties through smooth patching, bump constructions, or log-sum-exp approximations, achieving formally verified safe stabilize regions that outperform traditional SOS designs (Liu et al., 2 Oct 2025).
• Temporal multiscale dynamics: Composite constructions underpin the main Lyapunov tool for singular perturbations, fixed-time convergence, and two-time-scale ISS, yielding fixed-time and robust stability certificates in multi-rate systems (Tang et al., 2024, Tang et al., 2024).
• Switching, hybrid, and piecewise-smooth systems: The max/min composition of Lyapunov pieces, with path-complete graph structures, effectively certifies stability for classes of switching and hybrid systems where no common polynomial Lyapunov function exists (Angeli et al., 2016, Philippe et al., 2017, Jongeneel et al., 23 Mar 2025, Berger et al., 2022).

3. Principal Analytical Methodologies

The rigorous construction and validation of composite Lyapunov-like functions rely on several analytical and algorithmic mechanisms:

• Differential/difference inequalities and small-gain reasoning: Composite decrease is established via a pair (or network) of inequalities with interconnection/coupling terms managed through small-gain or residual bounds, enabling strict decay and convergence rate theorems (Saoud, 9 Oct 2025, Wang et al., 25 Jun 2025, Grüne, 2020).
• Patching and smoothing: Non-smooth, max-type compositions are regularized using log-sum-exp (softmax) relaxations and smooth bump patching, ensuring differentiability needed for controller synthesis and formal verification over compact state sets (Liu et al., 2 Oct 2025).
• Graph-theoretic aggregation: Path-complete graph Lyapunov frameworks encode contracts between Lyapunov pieces through labeled edges, leading to composite functions of the form $i = 1, \ldots, N$4. Observer constructions and partial orderings facilitate comparison and systematic refinement (Angeli et al., 2016, Philippe et al., 2017, Jongeneel et al., 23 Mar 2025).
• Optimization-based synthesis: Distributed or piecewise Lyapunov candidates are constructed via parallel SOS programs or counterexample-guided refinement; for polyhedral Lyapunov functions, sound and terminating learning/verification alternations rely on cutting-plane arguments (Berger et al., 2022, Wang et al., 25 Jun 2025).
• Neural-parameterized compositionality: Deep networks parameterize blockwise Lyapunov pieces, reconstructing global Lyapunov functions of the compositional form and achieving favorable complexity scaling relative to system dimension (Liu et al., 2024, Grüne, 2020).

4. Benchmarks, Case Studies, and Formal Guarantees

Multiple composite Lyapunov-like certification techniques have demonstrated empirical and formal advances across various benchmark systems:

• Control-affine systems with safety constraints: Softmax-patched CLBFs verified via δ-complete SMT yield safe stabilization regions up to 50% larger than the best SOS-based designs for power converters and nonlinear 2D systems (Liu et al., 2 Oct 2025).
• High-dimensional oscillator networks: Distributed composite Lyapunov functions composed from partial certificates accurately capture the non-convex ROA of van der Pol/Ising oscillator networks (Wang et al., 25 Jun 2025).
• Fixed-time and singular perturbation systems: Composite Lyapunov constructions in two-scale ODE systems provide explicit bounds on settling time (independent of initial conditions), robustness to disturbances, and encompass both asymptotic and fixed-time attractivity (Tang et al., 2024, Tang et al., 2024).
• Hybrid and piecewise-linear systems: For hybrid linear systems, the synthesis of max-type polyhedral Lyapunov functions via counterexample refinement is NP-hard, but practical and terminating algorithms exist and are empirically effective for multi-mode planar and rotational systems (Berger et al., 2022).
• Switching and path-complete graphs: Observer-induced composite Lyapunov forms of min-of-max structure rigorously guarantee invariance and stability for a range of path-complete criteria, with comparison algorithms based on linear programming (Angeli et al., 2016, Philippe et al., 2017).

5. Advanced Topics: Formal Comparison and Refinement

The analysis and design of composite Lyapunov-like frameworks are enhanced by methods for comparison, ordering, and refinement:

• Partial orderings: For path-complete Lyapunov graphs, preorders are characterized through combinatorial and algebraic simulation relations. Explicit linear program-based certificates enable comparison of the conservativeness of two distinct Lyapunov certificates (Philippe et al., 2017).
• Composition lifts: Path-complete criteria can be systematically enriched via composition lifting, expanding the library of Lyapunov pieces under functional composition with subsystem maps. While certain composition-based dominance conjectures are false, iterative refinement of path-complete graphs using lifts yields more powerful certificates (Jongeneel et al., 23 Mar 2025).
• Complexity and Solvability: For polyhedral (max-type) Lyapunov synthesis, the existence problem is NP-hard, and achieving robust termination requires integrating cutting-plane logic to ensure finite convergence at polynomial cost in dimension and number of pieces (Berger et al., 2022).

6. Future Directions and Open Challenges

Composite Lyapunov-like paradigms are continuously evolving to address the scalability, robustness, and generality demands of contemporary control and verification:

• Input constraints and robustness: Many existing composite Lyapunov frameworks handle unconstrained inputs; extending these to input-bounded control problems remains an open direction (Liu et al., 2 Oct 2025).
• Scalable neural and distributed methods: The expressiveness and scalability of neural-parameterized blockwise Lyapunov construction promise further inroads for ultra-high-dimensional systems, provided rigorous verification can keep pace (Liu et al., 2024, Grüne, 2020).
• Formal refinement and regulation of graph structures: The integration of graph-theoretic composition lifts and preordering precepts is critical for organizing large Lyapunov libraries and transitioning between candidate structures with provable guarantees (Jongeneel et al., 23 Mar 2025, Philippe et al., 2017).
• Synthetic and safety-aware hybridizations: Patching techniques and formal SMT-based relaxations continue to expand the tractable class of safety-constrained and hybrid systems amenable to composite Lyapunov certification (Liu et al., 2 Oct 2025).

In summary, composite Lyapunov-like functions—through mechanisms such as summation, max-type aggregation, smooth patching, and distributed synthesis—enable rigorous, scalable, and formally verifiable certificates in problems ranging from high-dimensional nonlinear networks to switched, hybrid, and safety-critical systems. Their ongoing development is closely tied to advances in optimization, algorithmic verification, and dynamical systems theory.