Fundamental Theorem of Dynamical Systems

• Fundamental Theorem of Dynamical Systems is a key result that decomposes a system into chain recurrent and gradient-like transient parts using complete Lyapunov functions.
• It establishes the Conley decomposition by defining chain recurrence through trapping regions and attractor–repeller pairs, aiding in the analysis of continuous, discrete, and hybrid systems.
• Alternative formulations like Sharkovskii’s theorem and categorical approaches extend its scope to periodic orbit ordering and structural symmetries in high-dimensional and hybrid dynamical systems.

Searching arXiv for recent and foundational papers on the Fundamental Theorem of Dynamical Systems and closely related formulations. The Fundamental Theorem of Dynamical Systems most commonly denotes the Conley–Franks theorem: a structural result asserting that a dynamical system decomposes into a chain recurrent part and a gradient-like transient part, and that this decomposition is encoded by a complete Lyapunov function (Lewis, 14 Aug 2025). In this formulation, the theorem separates recurrent dynamics from transient dynamics for continuous-time and discrete-time systems, and, in recent work, for flows, semiflows, and certain hybrid systems (Lewis, 14 Aug 2025). In parallel, the phrase is also used in one-dimensional dynamics for Sharkovskii’s theorem, which organizes the possible periods of interval maps and, in a recent infinite-dimensional extension, transfers this periodic-ordering principle to compact Banach-space maps that are topologically close to a one-dimensional model (Gierzkiewicz et al., 2024). These usages are distinct but related by a common theme: the extraction of universal dynamical structure from recurrence, order, and topological invariants.

1. Conley’s formulation and the recurrent–transient split

In the Conley framework, one works with a topological flow or semiflow $\Phi:T\times X\to X$ on a metric space $(X,d)$, with $T=\mathbb{R}$ or $T=\mathbb{Z}$ depending on whether time is continuous or discrete (Lewis, 14 Aug 2025). The theorem is built on the notion of chain recurrence. Using an error function $\epsilon\in C(X,(0,\infty))$, an $(\epsilon,T)$-chain from $x$ to $y$ is a finite sequence of points and times

$$(x_0=x,x_1,\dots,x_n=y),\qquad (t_0,\dots,t_{n-1}),$$

with $t_j\ge T$ and

$(X,d)$0

for all $(X,d)$1 (Lewis, 14 Aug 2025). A point is chain recurrent if such chains exist from the point back to itself for all admissible errors and time floors. The resulting set is the chain recurrent set, denoted $(X,d)$2.

The theorem asserts that the complement $(X,d)$3 is not arbitrary. It is exhausted by basins attached to trapping regions. If $(X,d)$4 denotes the collection of trapping regions and $(X,d)$5 the attracting set determined by $(X,d)$6, then

$(X,d)$7

and, for flows,

$(X,d)$8

where $(X,d)$9 is the attracting–repelling pair associated to $T=\mathbb{R}$0 (Lewis, 14 Aug 2025). This is the Conley decomposition: outside chain recurrence, the dynamics is gradient-like; inside it, the system decomposes into chain components.

A common misconception is that the theorem is primarily about asymptotic stability or hyperbolicity. In the Conley form it is not. Its hypotheses and conclusions are topological: it describes recurrence via chains, trapping regions, and attractor–repeller pairs, without assuming differentiability or hyperbolic structure (Lewis, 14 Aug 2025).

2. Trapping regions, attractors, repellers, and chain components

A trapping region $T=\mathbb{R}$1 is a nonempty set for which there exists $T=\mathbb{R}$2 such that

$T=\mathbb{R}$3

in a formulation that simultaneously covers continuous and discrete time (Lewis, 14 Aug 2025). The associated attracting set is

$T=\mathbb{R}$4

For flows, one also has the associated repelling set

$T=\mathbb{R}$5

If $T=\mathbb{R}$6 is open, then for flows the repelling set satisfies

$T=\mathbb{R}$7

(Lewis, 14 Aug 2025).

Chain recurrence induces an equivalence relation on $T=\mathbb{R}$8: two points are chain equivalent if each can be reached from the other by arbitrarily accurate chains. The equivalence classes are the chain components, and they are closed and invariant for flows, or forward-invariant for semiflows (Lewis, 14 Aug 2025). The theorem thus does more than identify a recurrent set; it canonically partitions that set into dynamically meaningful pieces.

Lewis’s unified treatment shows that the various chain formulations are equivalent after time normalization. If $T=\mathbb{R}$9 is invertible in the time domain, then chain equivalence, $T=\mathbb{Z}$0-chain equivalence, and exact $T=\mathbb{Z}$1-chain equivalence coincide. In particular,

$T=\mathbb{Z}$2

so for flows $T=\mathbb{Z}$3, and for discrete time the chain recurrent set of the semiflow agrees with that of the time-one map (Lewis, 14 Aug 2025). This equivalence is technically important because it reduces continuous-time questions to discrete-time ones without losing the recurrent structure.

This suggests that the theorem is fundamentally about topological recurrence classes rather than about a preferred time formalism. The same recurrent/transient dichotomy survives passage between maps, flows, and semiflows when the chain relation is formulated correctly.

3. Complete Lyapunov functions

A complete Lyapunov function is a continuous map $T=\mathbb{Z}$4 satisfying four properties: it is nonincreasing along forward trajectories; it strictly decreases off the chain recurrent set; it is constant on chain recurrent trajectories; and on $T=\mathbb{Z}$5 it separates chain components, in the sense that two chain recurrent points have the same $T=\mathbb{Z}$6-value if and only if they belong to the same chain component (Lewis, 14 Aug 2025). For flows, the monotonicity sharpens: if $T=\mathbb{Z}$7, then $T=\mathbb{Z}$8 is strictly decreasing.

For discrete-time maps on separable metric spaces, Lewis gives an explicit construction. After building Lyapunov prefunctions $T=\mathbb{Z}$9 for a countable family of strong trapping regions, the global complete Lyapunov function is

$\epsilon\in C(X,(0,\infty))$0

This converges uniformly, is continuous, and yields the full Conley–Franks conclusion: $\epsilon\in C(X,(0,\infty))$1, strict decrease off $\epsilon\in C(X,(0,\infty))$2, constancy on chain components, and separation of distinct chain components (Lewis, 14 Aug 2025).

For continuous-time flows and semiflows, the same paper provides an explicit construction “for the first time” by integrating a complete Lyapunov function $\epsilon\in C(X,(0,\infty))$3 for the time-one map: $\epsilon\in C(X,(0,\infty))$4 The resulting $\epsilon\in C(X,(0,\infty))$5 is a complete Lyapunov function for the continuous-time system, and the set of values $\epsilon\in C(X,(0,\infty))$6 is closed and nowhere dense, inherited from the discrete-time construction (Lewis, 14 Aug 2025).

The importance of the complete Lyapunov function is not merely that it detects asymptotic descent. It gives a scalar realization of the Conley decomposition. Its level sets encode chain components and the partial order among them, thereby producing a Morse-type stratification of the dynamics (Lewis, 14 Aug 2025).

4. Extensions to semiflows and hybrid systems

The Conley theorem was originally formulated for continuous flows on compact metric spaces, and the Franks version covers continuous maps. Recent work extends this structure to hybrid dynamical systems, where trajectories combine continuous evolution and discrete resets (Kvalheim et al., 2020). In the model treated there, a topological hybrid system is a tuple

$\epsilon\in C(X,(0,\infty))$7

with flow set $\epsilon\in C(X,(0,\infty))$8, guard set $\epsilon\in C(X,(0,\infty))$9, local semiflow $(\epsilon,T)$0 on $(\epsilon,T)$1, and reset map $(\epsilon,T)$2 (Kvalheim et al., 2020).

The hybrid theory introduces hybrid $(\epsilon,T)$3-chains, a hybrid Conley relation, and the hybrid chain recurrent set $(\epsilon,T)$4 (Kvalheim et al., 2020). Under four hypotheses—determinism $(\epsilon,T)$5, every maximal execution infinite or Zeno, a trapping guard condition, and compactness of $(\epsilon,T)$6—there exists a hybrid complete Lyapunov function $(\epsilon,T)$7 such that $(\epsilon,T)$8 strictly decreases along flows outside $(\epsilon,T)$9, strictly decreases across resets outside $x$0, classifies chain-equivalent points in $x$1 by equality of values, and has $x$2 nowhere dense in $x$3 (Kvalheim et al., 2020).

The hybrid analogue of the decomposition theorem is

$x$4

together with an attractor–repeller characterization of chain equivalence (Kvalheim et al., 2020). The proof proceeds by embedding the hybrid system into a relaxed system, constructing a hybrid suspension carrying a continuous semiflow, and then pulling back the classical Conley theory from the suspension to the original hybrid dynamics (Kvalheim et al., 2020).

Concrete applications include Lagrangian hybrid systems with impacts and the bouncing-ball model. For the bouncing ball with restitution coefficient $x$5, the chain recurrent set on a compact energy sublevel reduces to $x$6, and an explicit complete Lyapunov function can be written as

$x$7

with a coefficient condition ensuring strict decrease across impacts (Kvalheim et al., 2020). This illustrates that Conley’s theorem is not confined to smooth autonomous systems; it persists in discontinuous event-driven dynamics when the topology of executions is controlled appropriately.

5. Sharkovskii’s theorem and the periodic-ordering sense of “fundamental theorem”

A second established usage of the phrase identifies the “Fundamental Theorem of Dynamical Systems” with Sharkovskii’s theorem for interval maps (Gierzkiewicz et al., 2024). For a continuous map $x$8, Sharkovskii introduced a total order $x$9 on $y$0,

$y$1

and proved that if $y$2 has a point of basic period $y$3, then for every $y$4 with $y$5, the map has a point of period $y$6 (Gierzkiewicz et al., 2024). In particular, period $y$7 implies points of every natural period.

The paper “Sharkovskii theorem for infinite dimensional dynamical systems” develops an infinite-dimensional extension. Let

$y$8

with $y$9 a Banach space, and let

$$(x_0=x,x_1,\dots,x_n=y),\qquad (t_0,\dots,t_{n-1}),$$0

where $$(x_0=x,x_1,\dots,x_n=y),\qquad (t_0,\dots,t_{n-1}),$$1 is closed, convex, and bounded. If a continuous compact map $$(x_0=x,x_1,\dots,x_n=y),\qquad (t_0,\dots,t_{n-1}),$$2 admits an $$(x_0=x,x_1,\dots,x_n=y),\qquad (t_0,\dots,t_{n-1}),$$3-periodic orbit with disjoint intervals $$(x_0=x,x_1,\dots,x_n=y),\qquad (t_0,\dots,t_{n-1}),$$4 satisfying

$$(x_0=x,x_1,\dots,x_n=y),\qquad (t_0,\dots,t_{n-1}),$$5

then for every $$(x_0=x,x_1,\dots,x_n=y),\qquad (t_0,\dots,t_{n-1}),$$6 with $$(x_0=x,x_1,\dots,x_n=y),\qquad (t_0,\dots,t_{n-1}),$$7, the map $$(x_0=x,x_1,\dots,x_n=y),\qquad (t_0,\dots,t_{n-1}),$$8 has a point of least period $$(x_0=x,x_1,\dots,x_n=y),\qquad (t_0,\dots,t_{n-1}),$$9 (Gierzkiewicz et al., 2024). The geometric interpretation is “one unstable direction + contraction in other directions”: ordering is carried by the scalar coordinate $t_j\ge T$0, while the remaining finite-dimensional and infinite-dimensional directions are trapped in bounded tails.

The proof transfers one-dimensional Sharkovskii itinerary data to infinite dimensions via interval coverings, h-sets with one exit, covering relations, and Leray–Schauder degree (Gierzkiewicz et al., 2024). The DDE application uses the delayed Rössler system

$t_j\ge T$1

with phase space $t_j\ge T$2, and rigorously verifies the hypotheses for

$t_j\ge T$3

The conclusion is that the delayed Rössler system has periodic orbits of all natural periods $t_j\ge T$4 (Gierzkiewicz et al., 2024).

The theorem is purely topological: it guarantees existence of periodic orbits, but not density of periodic points, multiplicity, stability, or hyperbolicity (Gierzkiewicz et al., 2024). It also does not directly assert Li–Yorke chaos for the infinite-dimensional semiflow, even though period $t_j\ge T$5 yields all periods in the Sharkovskii sense.

6. Categorical and discrete-time reformulations

A more recent line of work gives a categorical “fundamental theorem” for discrete dynamical systems through the notion of a cycle set (Carranza et al., 5 Jun 2025). A discrete dynamical system is a pair $t_j\ge T$6, viewed as a functor from the monoid category $t_j\ge T$7 to $t_j\ge T$8; its state space is the digraph with vertices $t_j\ge T$9 and edges $(X,d)$00 (Carranza et al., 5 Jun 2025). The key construction is the attractor functor

$(X,d)$01

where

$(X,d)$02

so $(X,d)$03 records directed cycles of all lengths together with their rotation and degeneracy structure (Carranza et al., 5 Jun 2025).

Within this framework, the principal decomposition result concerns semi-direct products. For a system $(X,d)$04, the set of $(X,d)$05-cycles decomposes as

$(X,d)$06

as $(X,d)$07-sets, where $(X,d)$08 ranges over periodic forcing classes in the “driver” system $(X,d)$09 (Carranza et al., 5 Jun 2025). Dynamically, each periodic orbit in $(X,d)$10 induces a time-periodic forcing of the subsystem on $(X,d)$11, and the global periodic attractors split as the disjoint union of the periodic attractors of the driven subsystems.

This categorical theory intersects the Conley picture in the discrete metric setting. There, for $(X,d)$12, $(X,d)$13-chains coincide with true orbit segments, so chain recurrence reduces to periodicity; the chain recurrent classes are precisely directed cycles, and the cycle set captures the full chain recurrent structure (Carranza et al., 5 Jun 2025). In general topological settings, however, cycle sets capture only the periodic part, whereas Conley’s theorem also encompasses nonperiodic chain recurrent phenomena (Carranza et al., 5 Jun 2025).

A useful editorial distinction is therefore the following. In the Conley sense, the Fundamental Theorem of Dynamical Systems is a theorem about all recurrence, organized by complete Lyapunov functions and attractor–repeller pairs. In the Sharkovskii and categorical senses, the phrase singles out universal organization principles for periodic recurrence. The recent literature shows that both viewpoints remain active, and both now extend well beyond their original settings—to semiflows, hybrid systems, Banach-space maps, delay equations, and categorical models of modular discrete dynamics (Lewis, 14 Aug 2025).