Complete Lyapunov Function in Dynamical Systems

• Complete Lyapunov functions are continuous scalar functions that remain constant on chain recurrent sets and strictly decrease on non-recurrent trajectories.
• The construction uses an integral approach to lift discrete-time Lyapunov functions to continuous-time systems, preserving key monotonicity and recurrence properties.
• Central to Conley theory, these functions decompose global dynamics into chain transitive components and gradient-like regions, aiding stability analysis.

A complete Lyapunov function is a continuous scalar function that encodes the global asymptotic structure of a dynamical system through its chain recurrent decomposition. For a semiflow $\phi_t:X\to X$ on a metric space $X$, it is a map $L:X\to[0,1]$ that is constant along orbits in the chain recurrent set $R_c(\phi)$, strictly decreasing along orbits outside $R_c(\phi)$, takes a nowhere dense set of values on $R_c(\phi)$, and assigns distinct level values to distinct chain transitive components. In this form, it is a central object in Conley theory: it separates recurrent dynamics from gradient-like dynamics and turns qualitative asymptotic behavior into the level-set structure of a single continuous function (Patrão, 2011).

1. Definition and defining properties

Let $\phi:[0,\infty)\times X\to X$, $(t,x)\mapsto \phi_t(x)$, be a semiflow on a metric space $X$. A complete Lyapunov function for $\phi_t$ is a continuous function

$X$0

satisfying three conditions (Patrão, 2011).

First, it is dynamically monotone in the precise Conley-theoretic sense. For each $X$1, the map

$X$2

is constant for $X$3. For each $X$4, the same map is decreasing; in the complete-Lyapunov-function setting this is understood as strict decrease outside the chain recurrent set (Patrão, 2011).

Second, the image of the chain recurrent set is thin: $X$5 is nowhere dense in $X$6. The closure of this image therefore has empty interior. This means that most values of $X$7 are not attained on chain recurrent points (Patrão, 2011).

Third, each recurrent level corresponds exactly to one chain transitive component. For every $X$8, the level set

$X$9

is a chain transitive component of $L:X\to[0,1]$0 (Patrão, 2011).

These conditions make the adjective “complete” substantive rather than decorative. A usual Lyapunov function distinguishes one attractor or one equilibrium from surrounding dynamics. A complete Lyapunov function performs this separation simultaneously for all chain recurrent components and places the non-chain-recurrent dynamics into strictly descending regions between recurrent plateaus (Patrão, 2011).

2. Semiflows, time-one maps, and the integral construction

The semiflow framework used in the existence theorem is the standard one. The map

$L:X\to[0,1]$1

satisfies $L:X\to[0,1]$2, the semigroup property

$L:X\to[0,1]$3

and continuity in both variables (Patrão, 2011).

A decisive simplification is to pass from continuous time to the time-one map

$L:X\to[0,1]$4

This is a discrete-time system obtained by sampling the semiflow at integer times. Patrão’s argument begins with a complete Lyapunov function

$L:X\to[0,1]$5

for the map $L:X\to[0,1]$6, and then lifts it to the full semiflow by defining

$L:X\to[0,1]$7

This is the central formula of Theorem 1.1 (Patrão, 2011).

The construction is designed to convert discrete-time monotonicity into continuous-time monotonicity. The integrand $L:X\to[0,1]$8 is continuous, so the integral is well defined, and $L:X\to[0,1]$9 is continuous as a function of $R_c(\phi)$0 by a standard compactness-and-uniform-continuity argument on $R_c(\phi)$1 (Patrão, 2011).

The monotonicity mechanism comes from the identity

$R_c(\phi)$2

By comparing the interval $R_c(\phi)$3 with $R_c(\phi)$4, and using the fact that $R_c(\phi)$5 decreases under the time-one map outside $R_c(\phi)$6, one obtains

$R_c(\phi)$7

with equality on the chain recurrent set and strict inequality outside it (Patrão, 2011).

This integral averaging does not discard the discrete-time information carried by $R_c(\phi)$8. It preserves the chain recurrent structure while producing a function that interacts correctly with all times $R_c(\phi)$9, not merely the integers.

3. Existence theorem on separable metric spaces

The main result is a transfer theorem from maps to semiflows. If $R_c(\phi)$0 is a complete Lyapunov function for the time-one map $R_c(\phi)$1, then

$R_c(\phi)$2

is a complete Lyapunov function for the semiflow $R_c(\phi)$3 (Patrão, 2011).

The existence statement follows immediately once one combines this with Hurley’s theorem for maps. Hurley proved that for any continuous map on a separable metric space there exists a complete Lyapunov function. Applying that theorem to $R_c(\phi)$4 yields $R_c(\phi)$5, and the integral construction then yields $R_c(\phi)$6 for the semiflow. Consequently, every semiflow on a separable metric space admits a complete Lyapunov function (Patrão, 2011).

A key structural input is the coincidence of chain recurrent sets: $R_c(\phi)$7 This allows the discrete-time complete Lyapunov function $R_c(\phi)$8, which is defined relative to $R_c(\phi)$9, to control the continuous-time chain recurrent set of the semiflow itself (Patrão, 2011).

The same transfer holds at the level of chain transitive components. The note uses Hurley’s results to identify the chain transitive components for $R_c(\phi)$0 and $R_c(\phi)$1, and then shows that the recurrent level sets of $R_c(\phi)$2 coincide with those of $R_c(\phi)$3. In particular,

$R_c(\phi)$4

so the nowhere-dense image property is preserved under the construction (Patrão, 2011).

4. Chain recurrence, decomposition, and Morse structure

The chain recurrent set $R_c(\phi)$5 consists of points that can be returned to by arbitrarily accurate and sufficiently long pseudo-orbits. In the semiflow setting, Conley’s theory and Hurley’s extensions yield a decomposition of the phase space into chain recurrent classes together with regions of gradient-like motion between them (Patrão, 2011).

A complete Lyapunov function gives a compact scalar encoding of this decomposition. On $R_c(\phi)$6,

$R_c(\phi)$7

while on $R_c(\phi)$8,

$R_c(\phi)$9

Hence chain recurrent points are exactly those on which $\phi:[0,\infty)\times X\to X$0 is constant along the forward orbit, and non-chain-recurrent points are exactly those on which $\phi:[0,\infty)\times X\to X$1 strictly decreases (Patrão, 2011).

For each recurrent value $\phi:[0,\infty)\times X\to X$2, the level set $\phi:[0,\infty)\times X\to X$3 is a chain transitive component. Distinct components therefore carry distinct scalar values. The data block explicitly states that this induces a Morse decomposition: a partially ordered collection of chain recurrent components with strictly descending $\phi:[0,\infty)\times X\to X$4-values outside those components (Patrão, 2011).

This decomposition is the main reason complete Lyapunov functions are stronger than ordinary Lyapunov functions. They do not merely detect stability near one invariant set; they organize the global asymptotic structure of the semiflow.

5. Technical setting and proof architecture

For the transfer theorem itself, $\phi:[0,\infty)\times X\to X$5 is only assumed to be a metric space and $\phi:[0,\infty)\times X\to X$6 a continuous semiflow. Separability is needed only when invoking Hurley’s existence theorem for the discrete-time map $\phi:[0,\infty)\times X\to X$7 (Patrão, 2011).

The proof proceeds through four technical steps. First, it establishes that the integral formula defines a continuous function $\phi:[0,\infty)\times X\to X$8. Second, it uses Hurley’s identity $\phi:[0,\infty)\times X\to X$9. Third, it proves monotonicity along semiflow orbits using the shift-of-interval computation and the discrete-time Lyapunov inequalities for $(t,x)\mapsto \phi_t(x)$0. Fourth, it identifies the recurrent level sets of $(t,x)\mapsto \phi_t(x)$1 with the chain transitive components of $(t,x)\mapsto \phi_t(x)$2 and transfers the nowhere-dense image property from $(t,x)\mapsto \phi_t(x)$3 to $(t,x)\mapsto \phi_t(x)$4 (Patrão, 2011).

The note also makes explicit that the extension from $(t,x)\mapsto \phi_t(x)$5 to $(t,x)\mapsto \phi_t(x)$6 is topological rather than measure-theoretic or variational. Once the discrete-time function exists, the semiflow construction is obtained by continuity, compactness of $(t,x)\mapsto \phi_t(x)$7, and the semigroup property (Patrão, 2011).

This topological character is useful in infinite-dimensional settings. The data block gives, as a typical suggestion, semiflows generated by ODEs or reaction–diffusion equations on separable Banach spaces: if a complete Lyapunov function is known for the time-one map, the integral formula yields one for the continuous-time semiflow (Patrão, 2011).

6. Historical context, later developments, and related terminology

The concept belongs to Conley’s theory. Conley proved existence for flows on compact metric spaces, Franks extended this to maps on compact metric spaces, and Hurley extended the theory first to locally compact spaces and then to arbitrary separable metric spaces. These existence results are described in the data block as instances of the “Fundamental Theorem of Dynamical Systems” (Patrão, 2011).

Later work sharpened the construction problem in directions that remain within the Conley-style notion of complete Lyapunov function. For autonomous ODEs, one paper proves that on any compact set contained in the complement of the chain-recurrent set, the orbital derivative of a complete Lyapunov function can be prescribed as any negative $(t,x)\mapsto \phi_t(x)$8-function, while retaining $(t,x)\mapsto \phi_t(x)$9 regularity of the complete Lyapunov function itself (Giesl et al., 2021). Another paper shows, in a very general abstract-dynamical-systems framework, that the existence of a non-coercive Lyapunov function is equivalent to integral uniform global asymptotic stability, which situates non-coercive global Lyapunov constructions within a broader stability theory (Mironchenko et al., 2018).

A distinct terminology must be separated from this Conley-theoretic usage. In switched-systems theory, “path-complete Lyapunov function” denotes a multiple-Lyapunov-function criterion indexed by a labeled directed graph that covers all switching words. That notion concerns arbitrary switching and common Lyapunov certificates, not chain recurrence, nowhere dense recurrent values, or chain transitive components (Angeli et al., 2016). Subsequent papers show that path-complete criteria induce common Lyapunov functions of min–max form and can be compared or refined through graph-theoretic constructions (Philippe et al., 2017, Jongeneel et al., 23 Mar 2025). The shared word “complete” is therefore terminologically adjacent but conceptually different.

In the Conley–Patrão sense, a complete Lyapunov function remains a scalar global organizer of recurrence and descent. Its recurrent plateaus identify the chain transitive components, and its strictly decreasing regions describe the gradient-like part of the dynamics. On separable metric spaces, the existence theorem for semiflows ensures that this organizing function is always available (Patrão, 2011).