38.01.01 · dynamics / topological-dynamics

Dynamical Systems, Orbits, and Limit Sets

shipped3 tiersLean: none

Anchor (Master): Brin-Stuck 2002 *Introduction to Dynamical Systems* (Cambridge University Press) Ch. 1–2 (orbits, limit sets, conjugacy, the circle rotation, the doubling map, toral automorphisms, the horseshoe); Katok-Hasselblatt 1995 *Introduction to the Modern Theory of Dynamical Systems* (Cambridge) Ch. 1–2; Walters 1982 *An Introduction to Ergodic Theory* (Springer GTM 79) Ch. 5; Smale 1967 *Differentiable dynamical systems* (Bull. Amer. Math. Soc. 73) — the originating survey of the structural-stability programme and the horseshoe

Intuition Beginner

A dynamical system is a rule for how a point moves as time passes. You start somewhere, the rule tells you where you are one step later, then another step later, and so on. Two flavours of clock appear. Sometimes time ticks in whole steps: apply a map, apply it again, apply it again — this is a discrete system, like watching a population each year. Sometimes time flows continuously, like a river carrying a leaf along a smooth current; this is a flow. Either way, the object of study is not one trajectory but the whole pattern of trajectories filling the space.

The path a starting point traces out is its orbit. Some orbits are dull on purpose: a point that never moves is a fixed point, and a point that returns exactly to where it began after a fixed number of steps is periodic. Most orbits are more interesting. They wander, never quite repeating, and the real question is: where do they end up in the long run? That long-run destination is captured by the limit set — the collection of places the orbit keeps coming back near, no matter how long you wait.

Why bother with this language? Because it lets you ignore the impossible dream of an exact formula for every trajectory and instead ask the answerable question: what are the eventual shapes? A leaf in a river might spiral into a whirlpool, settle into a steady loop, or drift toward the bank. Naming these end-states — and recognising when two different-looking systems secretly have the same end-states — is the whole game.

Visual Beginner

Picture a square region of the plane with arrows showing where each point goes next. Drop a dot at a starting point and watch its orbit: a chain of dots hopping from place to place (discrete) or a smooth curve threading through the arrows (continuous). Three orbits are drawn. One spirals inward toward a single dot at the centre — a fixed point that attracts. One traces a closed loop, returning to itself forever — a periodic orbit. The third wanders across the square, never repeating, but its dots keep crowding toward the closed loop: the loop is its limit set.

Alongside sits a small table contrasting the two clocks and the basic orbit types.

feature	discrete system	continuous system
time	whole steps $0, 1, 2, \dots$	real numbers, flowing
rule	a map applied repeatedly	a flow $φ_{t}$
orbit	a sequence of points	a curve
simplest orbit	fixed point (stays put)	resting point (stays put)
repeating orbit	periodic point	closed loop
long-run target	limit set	limit set

Worked example Beginner

Take the doubling map on the circle, thought of as numbers from $0$ up to $1$ with $0$ and $1$ glued together. The rule is: double the number, and if the result is $1$ or more, subtract $1$ to stay on the circle. In symbols, the new value is $2 x$ reduced to its fractional part. Let us track a few orbits.

Step 1. Start at $x = 0$ . Doubling gives $0$ . It stays. So $0$ is a fixed point.

Step 2. Start at $x = 1/3$ . Doubling gives $2/3$ . Doubling again gives $4/3$ , and subtracting $1$ gives $1/3$ — back to the start after two steps. So $1/3$ is periodic with period $2$ , and its orbit is the two-point set ${1/3, 2/3}$ .

Step 3. Start at $x = 1/5$ . The orbit runs $1/5 \to 2/5 \to 4/5 \to 3/5 \to 1/5$ , a period- $4$ cycle.

Step 4. Start at an ordinary point like $x = 0.4 = 2/5$ . Wait — that landed in the period- $4$ cycle above. So pick $x = 0.1$ . Doubling repeatedly: $0.1 \to 0.2 \to 0.4 \to 0.8 \to 0.6 \to 0.2$ , and now it is trapped in ${0.2, 0.4, 0.8, 0.6}$ . The neat way to see the pattern is in binary: doubling just shifts the binary digits one place left and drops the part before the point. A point whose binary expansion never settles into a repeating block has an orbit that never repeats and visits the whole circle densely.

What this tells us: a rule as plain as "double it" already produces fixed points, short cycles, and orbits that fill the circle without ever repeating. The binary-shift viewpoint converts a question about a curve-bending map into a question about strings of digits — the first hint that very different-looking systems can be the same underneath.

Check your understanding Beginner

Formal definition Intermediate+

Let $X$ be a metric space 02.01.05. A dynamical system on $X$ is an action of a time monoid or group $T$ on $X$ by continuous maps. The two principal cases:

Discrete time. $T = Z_{\geq 0}$ (semigroup) or $T = Z$ (group). The action is generated by a single continuous map $f : X \to X$ ; the time- $n$ map is the iterate $f^{n} = f \circ \dots \circ f$ ( $n$ factors), with $f^{0} = id_{X}$ , and the semigroup law $f^{m + n} = f^{m} \circ f^{n}$ holds. When $f$ is a homeomorphism the action extends to $Z$ with $f^{- n} = (f^{- 1})^{n}$ .
Continuous time. $T = R_{\geq 0}$ or $T = R$ . The action is a flow: a continuous map $φ : T \times X \to X$ , written $φ^{t} (x) = φ (t, x)$ , satisfying $φ^{0} = id_{X}$ and $φ^{s + t} = φ^{s} \circ φ^{t}$ . This is exactly the one-parameter group structure of 02.12.02, here taken as the defining datum rather than constructed from a vector field.

In both cases write $T_{\geq 0}$ for the forward times. The forward orbit of $x$ is $O^{+} (x) = {f^{n} (x) : n \in T_{\geq 0}}$ (discrete) or ${φ^{t} (x) : t \geq 0}$ (continuous). When the time set is a group, the backward orbit $O^{-} (x)$ uses negative times and the full orbit is $O (x) = O^{+} (x) \cup O^{-} (x)$ .

Definition (fixed and periodic points). A point $x$ is fixed if $f (x) = x$ (resp. $φ^{t} (x) = x$ for all $t$ ). It is periodic if $f^{n} (x) = x$ for some integer $n \geq 1$ (resp. $φ^{T} (x) = x$ for some $T > 0$ ); the least such $n$ (resp. $T$ ) is the period, and the orbit of a periodic point is a periodic orbit.

Definition (omega- and alpha-limit sets). For a forward orbit, the omega-limit set of $x$ is $$ \omega(x) = { y \in X : f^{n_k}(x) \to y \text{ for some } n_k \to \infty } = \bigcap_{N \geq 0} \overline{{ f^n(x) : n \geq N }}, $$ with the analogous definition $ω (x) = ⋂_{s \geq 0} \overline{{φ^{t} (x) : t \geq s}}$ for a flow. When the time set is a group, the alpha-limit set $α (x)$ is defined identically with $n_{k} \to - \infty$ (resp. $t \to - \infty$ ): it is the omega-limit set of $x$ for the inverse system $f^{- 1}$ (resp. $φ^{- t}$ ).

Definition (invariant set). A set $A \subseteq X$ is forward-invariant if $f (A) \subseteq A$ (resp. $φ^{t} (A) \subseteq A$ for $t \geq 0$ ) and invariant if $f (A) = A$ (resp. $φ^{t} (A) = A$ for all $t$ ). An attractor is a closed invariant set $A$ with a neighbourhood $U \supseteq A$ such that $⋂_{n \geq 0} f^{n} (U) = A$ and $f^{n} (y) \to A$ for every $y \in U$ ; the largest such $U$ is the basin of $A$ .

Definition (non-wandering set). A point $x$ is wandering if it has a neighbourhood $U$ and a time $N$ with $f^{n} (U) \cap U = \emptyset$ for all $n \geq N$ (resp. $φ^{t} (U) \cap U = \emptyset$ for all $t \geq N$ ). The non-wandering set $Ω (f)$ (resp. $Ω (φ)$ ) is the set of points that are not wandering: those for which every neighbourhood returns to meet itself at arbitrarily large times.

Definition (topological conjugacy). Two discrete systems $f : X \to X$ and $g : Y \to Y$ are topologically conjugate if there is a homeomorphism $h : X \to Y$ with $h \circ f = g \circ h$ . The map $h$ is a conjugacy; it identifies the two systems orbit-by-orbit, since $h \circ f^{n} = g^{n} \circ h$ for every $n$ . For flows the corresponding notion is topological equivalence: a homeomorphism carrying orbits to orbits and preserving the direction of time (not necessarily the time parametrisation).

A sign / orientation convention. Throughout, omega refers to forward time and alpha to backward time, matching Birkhoff and Brin-Stuck; the flow law is $φ^{s + t} = φ^{s} \circ φ^{t}$ as in 02.12.02, so the omega-limit set of a flow is the forward accumulation set.

Counterexamples to common slips

An omega-limit set need not be a single orbit. For an irrational rotation of the circle, $ω (x)$ is the entire circle for every $x$ — the orbit is dense, so it accumulates everywhere.
A limit set can be empty. For the translation $f (x) = x + 1$ on $R$ , every forward orbit escapes to infinity; $ω (x) = \emptyset$ . The non-emptiness conclusions below require the forward orbit to lie in a compact set.
The non-wandering set is generally larger than the closure of the periodic points and larger than the union of limit sets. The three coincide for many hyperbolic systems but not in general; the inclusion $\overline{Per (f)} \subseteq Ω (f)$ can be strict.
Topological conjugacy is strictly finer than "same number of periodic points." Two maps can share period counts yet fail to be conjugate because conjugacy must match the topology of how orbits sit together, not merely tallies.

Key theorem with proof Intermediate+

Theorem (structure of the omega-limit set). Let $f : X \to X$ be a continuous map of a metric space, and let $x \in X$ have forward orbit contained in a compact set $K \subseteq X$ . Then $ω (x)$ is non-empty, compact, forward-invariant, and — when the orbit closure is connected, in particular for every flow — connected. If $f$ is a homeomorphism, $ω (x)$ is fully invariant: $f (ω (x)) = ω (x)$ . (See ^{[Brin-Stuck 2002 §1.6]}, ^{[Katok-Hasselblatt 1995 §1.5]}.)

Proof. Write $ω = ω (x)$ and $A_{N} = \overline{{f^{n} (x) : n \geq N}}$ , so $ω = ⋂_{N \geq 0} A_{N}$ .

Non-empty and compact. Each $A_{N}$ is a closed subset of the compact set $K$ (the orbit lies in $K$ , hence so does its closure), so each $A_{N}$ is compact and non-empty. The family ${A_{N}}$ is nested decreasing: $A_{N + 1} \subseteq A_{N}$ . A nested decreasing family of non-empty compact sets has non-empty intersection, so $ω \neq = \emptyset$ ; and $ω$ , an intersection of closed sets, is closed inside the compact $K$ , hence compact.

Characterisation by subsequences. A point $y$ lies in $ω$ if and only if there is a sequence $n_{k} \to \infty$ with $f^{n_{k}} (x) \to y$ . Indeed, $y \in A_{N}$ for every $N$ means that for each $N$ and each $ε > 0$ there is $n \geq N$ with $d (f^{n} (x), y) < ε$ ; choosing $ε = 1/ k$ and $N = k$ produces the required subsequence, and the converse is immediate.

Forward invariance. Let $y \in ω$ with $f^{n_{k}} (x) \to y$ . By continuity of $f$ , $f^{n_{k} + 1} (x) = f (f^{n_{k}} (x)) \to f (y)$ , and $n_{k} + 1 \to \infty$ , so $f (y) \in ω$ . Thus $f (ω) \subseteq ω$ . If $f$ is a homeomorphism, applying the same argument to $f^{- 1}$ and using compactness shows $f (ω) = ω$ : given $y \in ω$ , the points $f^{n_{k} - 1} (x)$ lie in the compact $ω$ -approximating sets and have a convergent subsequence with limit $z \in ω$ satisfying $f (z) = y$ , so $y \in f (ω)$ .

Connectedness. Suppose, toward a contradiction, that $ω = P ⊔ Q$ with $P, Q$ non-empty, closed, and disjoint. Since $ω$ is compact, $P$ and $Q$ are compact and separated by a positive distance $3 δ = d (P, Q) > 0$ . Both $P$ and $Q$ are visited by the orbit at arbitrarily large times, so there are increasing times at which $f^{n} (x)$ lies within $δ$ of $P$ and later within $δ$ of $Q$ . In the flow case, the orbit $t \mapsto φ^{t} (x)$ is a continuous curve, so to pass from the $δ$ -neighbourhood of $P$ to the $δ$ -neighbourhood of $Q$ it must cross the closed middle band ${z : d (z, P) \geq δ and d (z, Q) \geq δ}$ , which is disjoint from $ω$ . This yields a sequence of crossing times $t_{k} \to \infty$ with $φ^{t_{k}} (x)$ in the compact middle band; a convergent subsequence gives a limit point of the orbit lying in the band, hence in $ω ∖ (P \cup Q) = \emptyset$ , a contradiction. So $ω$ is connected. (For a discrete map the orbit need not be connected, and $ω$ can be disconnected — for instance a period-two orbit; connectedness is the flow phenomenon, recorded here because every continuous-time limit set inherits it.)

This completes the proof. $□$

Bridge. The omega-limit-set structure theorem builds toward the entire classification programme of topological dynamics: once one knows that long-term behaviour is concentrated on a compact invariant set, the question of which compact invariant sets can occur becomes the organising problem of the field. The foundational reason the theorem holds is precompactness plus continuity — exactly the two hypotheses that the phase flow 02.12.02 supplies for flows of complete vector fields on compact manifolds, where every forward orbit is automatically precompact. This is exactly the input that the Poincaré-Bendixson theorem 02.12.10 sharpens in the planar case: there the connected compact invariant limit set is forced to be an equilibrium, a periodic orbit, or a graph of connections, and the present theorem is the dimension-free skeleton beneath that planar conclusion. The omega-limit set generalises the notion of a periodic orbit — a periodic orbit is its own omega-limit set — and the non-wandering set $Ω (f)$ collects all such recurrent behaviour into a single closed invariant set. The central insight is that topological conjugacy transports every object defined here, so the limit-set and non-wandering structure is an invariant of the system up to conjugacy, and the canonical examples below (rotation, doubling map, cat map, horseshoe) are distinguished precisely by these invariants.

The bridge is the recognition that recurrence — captured by $ω$ , $α$ , and $Ω$ — is the conjugacy-invariant residue of a dynamical system, and putting these together gives the working definition of "the same dynamics" used throughout the subject. This appears again in symbolic dynamics, where the doubling map and the shift turn out to be the same system viewed through a conjugacy.

Exercises Intermediate+

Exercise 3 (medium, symbolic).

Let $f : X \to X$ and $g : Y \to Y$ be conjugate via a homeomorphism $h : X \to Y$ with $h \circ f = g \circ h$ . Show that $h$ maps periodic points of $f$ bijectively onto periodic points of $g$ , preserving period.

Hint

Iterate the conjugacy relation to get $h \circ f^{n} = g^{n} \circ h$ , then apply it to a periodic point.

Answer

From $h \circ f = g \circ h$ , induction gives $h \circ f^{n} = g^{n} \circ h$ for every $n \geq 1$ : assuming $h \circ f^{n} = g^{n} \circ h$ , compose with $f$ on the right to get $h \circ f^{n + 1} = g^{n} \circ h \circ f = g^{n} \circ g \circ h = g^{n + 1} \circ h$ . If $f^{n} (x) = x$ , then $g^{n} (h (x)) = h (f^{n} (x)) = h (x)$ , so $h (x)$ is periodic for $g$ with period dividing $n$ . Conversely, applying the relation to $h^{- 1}$ (which conjugates $g$ to $f$ ) shows the period of $h (x)$ is no smaller than that of $x$ . Hence the least period is preserved, and $h$ restricts to a bijection $Per_{n} (f) \to Per_{n} (g)$ for each $n$ . Rubric: full credit for the iterated relation, both inclusions, and the period-preservation conclusion.

Exercise 5 (medium, short-answer).

Show that the non-wandering set $Ω (f)$ of a continuous map contains every periodic point and is closed.

Hint

For a periodic point, exhibit a neighbourhood that returns to itself. For closedness, show the wandering set is open.

Answer

Periodic points. If $f^{n} (p) = p$ , then for any neighbourhood $U$ of $p$ , continuity gives a neighbourhood $V \subseteq U$ with $f^{n} (V) \cap V ∋ p \neq = \emptyset$ , and the same holds for $f^{k n}$ at all $k$ . So no neighbourhood of $p$ eventually separates from itself; $p$ is non-wandering. Closed. The wandering set is open: if $x$ is wandering with witnessing neighbourhood $U$ and return-free time $N$ , then every point of $U$ is also wandering with the same $U$ and $N$ (the witnessing condition $f^{n} (U) \cap U = \emptyset$ for $n \geq N$ is shared by all of $U$ ). So the wandering set is a union of open sets, hence open, and its complement $Ω (f)$ is closed. Rubric: full credit for both the periodic-point containment and the open-complement argument.

Exercise 6 (medium, short-answer).

The tent map $T (x) = 1 - ∣2 x - 1∣$ on $[0, 1]$ and the doubling map are related by a semiconjugacy. State what a semiconjugacy is (a conjugacy with $h$ only continuous and surjective, not necessarily injective) and explain why semiconjugacy still transports omega-limit sets forward: if $h \circ f = g \circ h$ then $h (ω_{f} (x)) \subseteq ω_{g} (h (x))$ .

Hint

Use the subsequence characterisation of the omega-limit set and continuity of $h$ .

Answer

A semiconjugacy from $f : X \to X$ to $g : Y \to Y$ is a continuous surjection $h : X \to Y$ with $h \circ f = g \circ h$ ; it differs from a conjugacy only in that $h$ need not be a homeomorphism. For the limit-set transport: let $y \in ω_{f} (x)$ , so $f^{n_{k}} (x) \to y$ for some $n_{k} \to \infty$ . By continuity, $h (f^{n_{k}} (x)) \to h (y)$ , and the intertwining gives $h (f^{n_{k}} (x)) = g^{n_{k}} (h (x))$ . Thus $g^{n_{k}} (h (x)) \to h (y)$ with $n_{k} \to \infty$ , which says exactly that $h (y) \in ω_{g} (h (x))$ . Hence $h (ω_{f} (x)) \subseteq ω_{g} (h (x))$ . (Equality can fail because $h$ may collapse distinct accumulation points.) Rubric: full credit for the definition and the continuity-plus-intertwining argument.

Exercise 7 (hard, short-answer).

Prove that the doubling map $E_{2} (x) = 2 x mod 1$ on the circle is topologically semiconjugate to the one-sided shift $σ$ on the space $Σ_{2}^{+} = {0, 1}^{N}$ of binary sequences, via the binary-coding map. Identify where injectivity fails.

Hint

Send a binary sequence $(a_{0} a_{1} a_{2} \dots)$ to the real number $\sum_{k \geq 0} a_{k} 2^{- (k + 1)}$ . Check that the shift corresponds to doubling, and find the two sequences coding the same dyadic rational.

Answer

Define $h : Σ_{2}^{+} \to S^{1}$ by $h (a_{0} a_{1} a_{2} \dots) = \sum_{k \geq 0} a_{k} 2^{- (k + 1)} mod 1$ . This is continuous (the series converges uniformly in the product topology) and surjective (every real in $[0, 1)$ has a binary expansion). The shift $σ (a_{0} a_{1} a_{2} \dots) = (a_{1} a_{2} a_{3} \dots)$ satisfies $$ h(\sigma(a)) = \sum_{k \geq 0} a_{k+1} 2^{-(k+1)} = 2 \sum_{k \geq 0} a_{k+1} 2^{-(k+2)} = 2\Big(h(a) - \tfrac{a_0}{2}\Big) \equiv 2, h(a) \pmod 1 = E_2(h(a)), $$ so $h \circ σ = E_{2} \circ h$ : $h$ is a semiconjugacy. Injectivity fails exactly at dyadic rationals, which have two binary expansions — for example $0.1000 \dots$ and $0.0111 \dots$ both code $1/2$ . On the dense $G_{δ}$ of non-dyadic points $h$ is injective, so the doubling map and the shift agree on a full-measure invariant set; this is the gateway to symbolic dynamics. Rubric: full credit for the coding map, the intertwining computation, and identifying dyadic rationals as the failure locus.

Exercise 8 (hard, short-answer).

Let $A = (2111)$ act on the torus $T^{2} = R^{2} / Z^{2}$ (the cat map). Show that the periodic points of the induced map $f_{A}$ are exactly the points with rational coordinates, and that they are dense.

Hint

A point $v + Z^{2}$ is periodic iff $A^{n} v \equiv v (mod Z^{2})$ for some $n$ . Consider points with denominator $q$ and count how many there are.

Answer

The matrix $A$ has integer entries and $det A = 1$ , so $A$ and $A^{- 1}$ both preserve $Z^{2}$ and $f_{A}$ is a homeomorphism of $T^{2}$ . Rational $\Rightarrow$ periodic. Fix $q \geq 1$ and consider the finite set $G_{q} = {(p_{1} / q, p_{2} / q) + Z^{2}}$ of points with denominator dividing $q$ ; there are $q^{2}$ of them. Since $A$ has integer entries, $f_{A}$ maps $G_{q}$ into itself, and being injective on the finite set $G_{q}$ it permutes $G_{q}$ . Every element of a finite permutation lies on a cycle, so each point of $G_{q}$ is periodic. Taking all $q$ shows every rational point is periodic. Periodic $\Rightarrow$ rational. If $A^{n} v \equiv v (mod Z^{2})$ then $(A^{n} - I) v \in Z^{2}$ ; since $det (A^{n} - I) \neq = 0$ (the eigenvalues of $A$ are $(3 \pm 5) /2 \neq = 1$ , so no eigenvalue of $A^{n}$ equals $1$ ), the matrix $A^{n} - I$ is invertible over $Q$ and $v = (A^{n} - I)^{- 1} w$ with $w \in Z^{2}$ has rational coordinates. Density. The rationals with denominator $q$ form a $1/ q$ -net of the torus, dense as $q \to \infty$ . Rubric: full credit for the permutation-of- $G_{q}$ argument, the invertibility of $A^{n} - I$ , and the density conclusion.

Advanced results Master

Theorem (orbit-closure decomposition and recurrence). Let $f : X \to X$ be a continuous map of a compact metric space. The forward orbit closure $\overline{O^{+} (x)}$ decomposes as $O^{+} (x) \cup ω (x)$ , with $ω (x)$ the closed invariant core. A point $x$ is (positively) recurrent if $x \in ω (x)$ . The recurrent points are dense in the support of any invariant Borel probability measure, and every non-empty closed invariant subset contains a minimal set — a closed invariant set with no proper non-empty closed invariant subset. (See ^{[Birkhoff 1927]}, ^{[Walters 1982 Ch. 5]}.)

The minimal sets are the irreducible carriers of recurrence: on a minimal set every orbit is dense, and a system is minimal exactly when $ω (x) = X$ for all $x$ . The irrational rotation is the model minimal system; a finite periodic orbit is the model minimal set in the discrete case. Zorn's lemma applied to the partially ordered family of non-empty closed invariant subsets, ordered by reverse inclusion, produces minimal sets inside any compact invariant set, since the intersection of a descending chain of non-empty compact invariant sets is non-empty, compact, and invariant.

Theorem (the doubling map is chaotic — conjugacy with the full shift). The doubling map $E_{2}$ on $S^{1}$ is, off the dyadic rationals, the binary-shift system. Consequently $E_{2}$ has periodic points of every period, dense periodic points, a dense orbit (topological transitivity), and sensitive dependence on initial conditions. The same holds for the full one-sided shift $σ$ on $Σ_{2}^{+}$ , which is the symbolic model.

The shift $σ$ on $Σ_{2}^{+}$ with the metric $d (a, b) = 2^{- m i n {k : a_{k} \neq = b_{k}}}$ has $2^{n}$ fixed points of $σ^{n}$ (the periodic sequences of period dividing $n$ ), a dense set of periodic points, a dense orbit obtained by concatenating all finite words, and expansiveness with constant $1$ . Devaney's three conditions for chaos — dense periodic points, topological transitivity, sensitive dependence — all hold, and sensitivity is in fact a consequence of the first two on an infinite metric space (Banks-Brooks-Cairns-Davis-Stacey 1992). The doubling map inherits each property through the semiconjugacy of Exercise 7, which is a conjugacy after removing the countable orbit of dyadic rationals.

Theorem (Smale horseshoe — preview). There is a diffeomorphism $f$ of the plane carrying a square $Q$ across itself in a horseshoe shape so that $f (Q) \cap Q$ is two horizontal strips. The maximal invariant set $Λ = ⋂_{n \in Z} f^{n} (Q)$ is a Cantor set, and $f ∣_{Λ}$ is topologically conjugate to the two-sided full shift $σ$ on $Σ_{2} = {0, 1}^{Z}$ . (Smale 1967 ^{[Smale 1967]}.)

The horseshoe is the geometric source of robust chaos: the conjugacy to the shift makes $Λ$ a structurally stable hyperbolic set, so the chaotic dynamics persists under $C^{1}$ perturbation. Coding a point of $Λ$ by recording, at each time $n \in Z$ , which of the two strips $f^{n}$ lands it in, produces the conjugating homeomorphism $Λ \to Σ_{2}$ . The horseshoe is the local model whenever a map has a transverse homoclinic point — the Smale-Birkhoff homoclinic theorem promotes any transverse homoclinic intersection to an embedded horseshoe, and hence to shift dynamics, which is why transverse homoclinic points are the universal signature of chaos in smooth systems.

Theorem (hyperbolic toral automorphism — the cat map). Let $A \in SL (2, Z)$ be hyperbolic (eigenvalues $∣ λ ∣ \neq = 1$ ). The induced automorphism $f_{A}$ of $T^{2}$ has dense periodic points (exactly the rational points), a dense orbit, positive topological entropy $lo g ∣ λ ∣$ where $λ$ is the expanding eigenvalue, and is structurally stable. Its non-wandering set is the whole torus.

The cat map $A = (2111)$ has eigenvalues $(3 \pm 5) /2$ , expanding and contracting along the two irrational eigendirections; the stable and unstable foliations are the lines of these slopes wrapped densely on the torus. The entropy $lo g \frac{3 + 5}{2}$ equals the growth rate $lim_{n} \frac{1}{n} lo g # Fix (f_{A}^{n})$ of periodic-point counts, since $# Fix (f_{A}^{n}) = ∣ det (A^{n} - I) ∣ = ∣ λ^{n} + λ^{- n} - 2∣$ , a manifestation of the Lefschetz fixed-point formula. The cat map is the model Anosov diffeomorphism and the simplest system that is simultaneously minimal-free, mixing, and chaotic.

Theorem (the logistic family and the period-doubling route). For the logistic map $f_{μ} (x) = μx (1 - x)$ on $[0, 1]$ with $0 \leq μ \leq 4$ , the dynamics interpolate from a single attracting fixed point ( $μ < 3$ ) through a cascade of period-doublings accumulating at $μ_{\infty} \approx 3.5699$ to a parameter regime of positive-entropy chaos; at $μ = 4$ the map is conjugate to the tent map and to the full one-sided shift, and the non-wandering set is all of $[0, 1]$ .

The period-doubling cascade exhibits the universal Feigenbaum constant $δ \approx 4.6692$ governing the geometric rate at which successive bifurcation parameters accumulate; the universality (Feigenbaum 1978, Coullet-Tresser 1978) is explained by a renormalisation fixed point in the space of unimodal maps. At $μ = 4$ the conjugacy $h (x) = sin^{2} (π x /2)$ carries the tent map to $f_{4}$ , and the tent map is in turn a factor of the shift, closing the loop: the logistic map at full parameter is the binary shift in disguise, the same identification that makes the doubling map chaotic.

Synthesis. The four canonical examples are not a list but a single structural story told four times. The central insight is that recurrence — encoded in orbits, omega- and alpha-limit sets, and the non-wandering set — is a topological-conjugacy invariant, and the entire content of a topological dynamical system is the conjugacy class of its recurrent part. This is exactly why the doubling map, the cat map, the horseshoe, and the logistic map at $μ = 4$ are all governed by the shift: each is conjugate (or semiconjugate after removing a thin exceptional set) to a shift space, and the shift is the universal model of chaotic recurrence. The bridge from the abstract limit-set theorem to these concrete systems is symbolic coding: partition the space, record which piece each iterate lands in, and the dynamics becomes a shift on sequences — the foundational reason that geometric chaos and combinatorial shift dynamics are the same subject. The negative theory and the positive theory are dual: a minimal system (irrational rotation) has every orbit dense and no proper recurrent substructure, while a hyperbolic system (cat map, horseshoe) has a rich lattice of periodic orbits dense in the non-wandering set; both are read off from the same limit-set and non-wandering-set invariants.

Putting these together, the topological-dynamics framework — orbit, limit set, non-wandering set, conjugacy — classifies long-term behaviour into minimality at one extreme and shift-modelled chaos at the other, and identifies the structural-stability programme of Smale ^{[Smale 1967]} as the search for the systems whose conjugacy class is robust under perturbation. The phase flow 02.12.02 and the Poincaré-Bendixson dichotomy 02.12.10 are the continuous-time, low-dimensional special cases of this same architecture; the present unit is the dimension-free, discrete-and-continuous foundation on which symbolic dynamics, ergodic theory, and hyperbolicity are all built.

Full proof set Master

Proposition (omega-limit set is invariant for a homeomorphism). Let $f$ be a homeomorphism of a metric space and let $x$ have precompact forward orbit. Then $f (ω (x)) = ω (x)$ .

Proof. Forward invariance $f (ω (x)) \subseteq ω (x)$ was shown in the Key theorem: if $f^{n_{k}} (x) \to y$ then $f^{n_{k} + 1} (x) \to f (y) \in ω (x)$ . For the reverse inclusion, let $y \in ω (x)$ with $f^{n_{k}} (x) \to y$ and $n_{k} \to \infty$ . After discarding finitely many terms assume $n_{k} \geq 1$ . The points $f^{n_{k} - 1} (x)$ lie in the compact orbit closure $K$ , so a subsequence converges, say $f^{n_{k_{j}} - 1} (x) \to z$ ; since $n_{k_{j}} - 1 \to \infty$ , $z \in ω (x)$ . By continuity of $f$ , $f (z) = lim_{j} f (f^{n_{k_{j}} - 1} (x)) = lim_{j} f^{n_{k_{j}}} (x) = y$ . Hence $y = f (z) \in f (ω (x))$ , giving $ω (x) \subseteq f (ω (x))$ . The two inclusions yield equality. $□$

Proposition (topological conjugacy transports the non-wandering set). If $h : X \to Y$ is a topological conjugacy from $f$ to $g$ , then $h (Ω (f)) = Ω (g)$ .

Proof. It suffices to show $h (Ω (f)) \subseteq Ω (g)$ ; applying the same to $h^{- 1}$ (a conjugacy from $g$ to $f$ ) gives the reverse inclusion. Let $x \in Ω (f)$ and put $y = h (x)$ . Let $V$ be any neighbourhood of $y$ . Then $U = h^{- 1} (V)$ is a neighbourhood of $x$ (as $h$ is a homeomorphism). Because $x$ is non-wandering, for every $N$ there is $n \geq N$ with $f^{n} (U) \cap U \neq = \emptyset$ ; pick $u \in U$ with $f^{n} (u) \in U$ . Applying $h$ and using $h \circ f^{n} = g^{n} \circ h$ , the point $h (u) \in V$ satisfies $g^{n} (h (u)) = h (f^{n} (u)) \in h (U) = V$ . So $g^{n} (V) \cap V \neq = \emptyset$ for arbitrarily large $n$ , meaning $y \in Ω (g)$ . Thus $h (Ω (f)) \subseteq Ω (g)$ . $□$

Proposition (existence of minimal sets). Every non-empty compact invariant set $K$ for a homeomorphism $f$ of a metric space contains a minimal set.

Proof. Let $F$ be the family of non-empty closed (equivalently compact, since $K$ is compact) invariant subsets of $K$ , partially ordered by inclusion. $F$ is non-empty since $K \in F$ . Let $C \subseteq F$ be a chain (totally ordered subfamily). The intersection $M = ⋂ C$ is closed and invariant (an intersection of invariant sets is invariant); it is non-empty because $C$ is a chain of non-empty compact sets with the finite-intersection property, and a family of compact sets with the finite-intersection property has non-empty intersection. So $M \in F$ is a lower bound for $C$ . By Zorn's lemma (applied to $F$ ordered by reverse inclusion), $F$ has a minimal element $M_{0}$ under inclusion: a non-empty closed invariant set with no proper non-empty closed invariant subset. That is exactly a minimal set. $□$

Proposition (periodic-point count of a hyperbolic toral automorphism). For hyperbolic $A \in SL (2, Z)$ with eigenvalues $λ, λ^{- 1}$ , $∣ λ ∣ > 1$ , the number of fixed points of $f_{A}^{n}$ on $T^{2}$ is $# Fix (f_{A}^{n}) = ∣ det (A^{n} - I) ∣ = ∣ λ^{n} + λ^{- n} - 2∣$ , and the growth rate $lim_{n} \frac{1}{n} lo g # Fix (f_{A}^{n}) = lo g ∣ λ ∣$ .

Proof. A point $v + Z^{2}$ is fixed by $f_{A}^{n}$ iff $A^{n} v \equiv v (mod Z^{2})$ , i.e. $(A^{n} - I) v \in Z^{2}$ . Since $A$ is hyperbolic, no eigenvalue of $A^{n}$ equals $1$ , so $A^{n} - I$ is invertible over $R$ and the map $v \mapsto (A^{n} - I) v$ is a linear automorphism of $R^{2}$ taking $Z^{2}$ to the sublattice $(A^{n} - I) Z^{2}$ . The solutions $v mod Z^{2}$ correspond bijectively to $Z^{2} / (A^{n} - I) Z^{2}$ , a finite group of order $∣ det (A^{n} - I) ∣$ . Computing the determinant with eigenvalues $λ^{n}, λ^{- n}$ : $det (A^{n} - I) = (λ^{n} - 1) (λ^{- n} - 1) = λ^{n} λ^{- n} - λ^{n} - λ^{- n} + 1 = 2 - λ^{n} - λ^{- n}$ , whose absolute value is $∣ λ^{n} + λ^{- n} - 2∣$ . As $n \to \infty$ the term $λ^{n}$ dominates (for $∣ λ ∣ > 1$ ), so $\frac{1}{n} lo g ∣ λ^{n} + λ^{- n} - 2∣ \to lo g ∣ λ ∣$ . This limit is the topological entropy of $f_{A}$ , by the equality of entropy with periodic-orbit growth for hyperbolic toral automorphisms. $□$

Proposition (sensitive dependence from transitivity and dense periodic points). Let $f$ be a continuous map of an infinite metric space $X$ with a dense orbit and dense periodic points. Then $f$ has sensitive dependence on initial conditions: there is $δ > 0$ such that for every $x$ and every neighbourhood $U$ of $x$ there is $y \in U$ and $n$ with $d (f^{n} (x), f^{n} (y)) > δ$ . (Banks-Brooks-Cairns-Davis-Stacey 1992.)

Proof. Since $X$ is infinite and the periodic points are dense, there exist at least two distinct periodic orbits; let $8 δ$ be a positive lower bound such that every point of $X$ is at distance $\geq 4 δ$ from one of two fixed periodic orbits $P, Q$ — concretely, pick periodic orbits $P, Q$ with $d (P, Q) =: 8 δ > 0$ , so any $x$ has $d (x, P) \geq 4 δ$ or $d (x, Q) \geq 4 δ$ . Fix $x$ and a neighbourhood $U$ . By density of periodic points choose a periodic point $p \in U$ of some period $m$ . Choose a periodic orbit $W \in {P, Q}$ with $d (x, W) \geq 4 δ$ , and let $w \in W$ . By transitivity there is a point $z \in U \cap V$ , where $V = ⋂_{i = 0}^{m} f^{- i} (B (f^{i} (w), δ))$ is the open set of points shadowing $w$ 's orbit for $m$ steps to within $δ$ . Now estimate at the time $n = m ⌈ k ⌉$ realising the return of $p$ : using the triangle inequality across $f^{n} (p) = p$ , $f^{n} (z)$ near $f^{n} (w) \in W$ , and $f^{n} (x)$ , at least one of $d (f^{n} (x), f^{n} (p))$ and $d (f^{n} (x), f^{n} (z))$ exceeds $δ$ — otherwise $d (p, W) \leq d (f^{n} (p), f^{n} (x)) + d (f^{n} (x), f^{n} (z)) + d (f^{n} (z), W) < 3 δ$ , contradicting $d (x, W) \geq 4 δ$ propagated through the chosen times. Both $p$ and $z$ lie in $U$ , so sensitivity holds with constant $δ$ . $□$

Connections Master

Phase flow / one-parameter group 02.12.02. A flow is the continuous-time instance of a dynamical system as defined here: the one-parameter group axioms $φ^{0} = id$ , $φ^{s + t} = φ^{s} \circ φ^{t}$ are the action axioms with time monoid $T = R$ . The omega-limit set, invariant sets, and non-wandering set of this unit specialise directly to the flow of a complete vector field, and the precompactness hypothesis that drives the limit-set structure theorem is automatic for flows on compact manifolds. The present unit abstracts the flow datum away from its differential-equation origin and treats it on the same footing as an iterated map.
Poincaré-Bendixson theorem 02.12.10. The planar Poincaré-Bendixson dichotomy is the two-dimensional sharpening of the limit-set structure theorem proved here: where the present theorem gives only "non-empty, compact, connected, invariant," the planar Jordan-curve input forces the limit set to be an equilibrium, a periodic orbit, or a graph of connections. This unit supplies the dimension-free skeleton; Poincaré-Bendixson supplies the planar flesh, and the contrast with the cat map and horseshoe shows why the higher-dimensional and discrete cases admit chaotic limit sets that the plane forbids.
Phase space, vector field, integral curve 02.12.01. The orbit of a point under a flow is the image of its integral curve, and a fixed point of the dynamical system is an equilibrium of the generating vector field. The phase-space formalism gives the geometric meaning to the abstract orbit and limit-set definitions; the recurrence structure studied here is the long-time behaviour of the integral curves introduced there.
Metric space 02.01.05. The entire topological-dynamics apparatus — convergence of subsequences defining limit sets, the shift metric $d (a, b) = 2^{- m i n {k : a_{k} \neq = b_{k}}}$ on sequence space, compactness arguments for non-emptiness of limit sets, sensitive dependence measured by a distance $δ$ — is metric-space theory. The compactness and completeness facts that make limit sets non-empty and minimal sets exist are exactly the metric-space results imported here.

Historical & philosophical context Master

The qualitative study of dynamical systems originates with Henri Poincaré's four-part memoir Sur les courbes définies par une équation différentielle (1881–1886), which replaced the search for closed-form solutions with the study of the topological structure of orbits — phase portraits, periodic orbits, and the index theory of equilibria ^{[Smale 1967]}. George David Birkhoff systematised the recurrence theory in his 1927 American Mathematical Society Colloquium volume Dynamical Systems ^{[Birkhoff 1927]}, where the non-wandering set, minimal sets, and recurrent points first appear as the organising invariants of a continuous flow or map; Birkhoff's recurrence theorem and his pointwise ergodic theorem (1931) are the analytic counterparts of the topological structure recorded in this unit.

The abstract reformulation of a dynamical system as an action of a group or semigroup, with topological conjugacy as the equivalence relation, crystallised in the mid-twentieth century. Stephen Smale's 1967 survey Differentiable dynamical systems (Bulletin of the American Mathematical Society 73) ^{[Smale 1967]} introduced the horseshoe, the Axiom-A decomposition of the non-wandering set into basic sets, and the structural-stability programme — the search for systems whose conjugacy class is stable under perturbation. Michael Brin and Garrett Stuck's Introduction to Dynamical Systems (Cambridge University Press, 2002) ^{[Brin-Stuck 2002]} and Anatole Katok and Boris Hasselblatt's Introduction to the Modern Theory of Dynamical Systems (Cambridge, 1995) ^{[Katok-Hasselblatt 1995]} are the modern canonical treatments, opening with exactly the orbit / limit-set / conjugacy framework of this unit before specialising to symbolic, smooth, and ergodic theory. Robert Devaney's An Introduction to Chaotic Dynamical Systems (1989) ^{[Devaney 1989]} gave the influential definition of chaos as the conjunction of dense periodic points, topological transitivity, and sensitive dependence; the Banks-Brooks-Cairns-Davis-Stacey theorem (1992) later showed sensitivity is redundant, a logical economy proved in the present unit's Full proof set.

Bibliography Master

@book{BrinStuck2002,
  author    = {Brin, Michael and Stuck, Garrett},
  title     = {Introduction to Dynamical Systems},
  publisher = {Cambridge University Press},
  year      = {2002}
}

@book{KatokHasselblatt1995,
  author    = {Katok, Anatole and Hasselblatt, Boris},
  title     = {Introduction to the Modern Theory of Dynamical Systems},
  publisher = {Cambridge University Press},
  series    = {Encyclopedia of Mathematics and its Applications},
  volume    = {54},
  year      = {1995}
}

@article{Smale1967,
  author  = {Smale, Stephen},
  title   = {Differentiable dynamical systems},
  journal = {Bulletin of the American Mathematical Society},
  volume  = {73},
  year    = {1967},
  pages   = {747--817}
}

@book{Birkhoff1927,
  author    = {Birkhoff, George D.},
  title     = {Dynamical Systems},
  publisher = {American Mathematical Society},
  series    = {Colloquium Publications},
  volume    = {9},
  year      = {1927}
}

@book{Devaney1989,
  author    = {Devaney, Robert L.},
  title     = {An Introduction to Chaotic Dynamical Systems},
  publisher = {Addison-Wesley},
  edition   = {2},
  year      = {1989}
}

@book{Walters1982,
  author    = {Walters, Peter},
  title     = {An Introduction to Ergodic Theory},
  publisher = {Springer-Verlag},
  series    = {Graduate Texts in Mathematics},
  volume    = {79},
  year      = {1982}
}

@article{BanksBrooks1992,
  author  = {Banks, John and Brooks, Jeff and Cairns, Grant and Davis, Gary and Stacey, Peter},
  title   = {On {D}evaney's definition of chaos},
  journal = {American Mathematical Monthly},
  volume  = {99},
  year    = {1992},
  pages   = {332--334}
}

@article{Feigenbaum1978,
  author  = {Feigenbaum, Mitchell J.},
  title   = {Quantitative universality for a class of nonlinear transformations},
  journal = {Journal of Statistical Physics},
  volume  = {19},
  year    = {1978},
  pages   = {25--52}
}

Prerequisites

02.12.02
02.01.05
02.12.01

Tier anchors

beginner: Strogatz 1994 *Nonlinear Dynamics and Chaos* (Addison-Wesley) Ch. 1–2 and Ch. 10 (the iterated-map and flow pictures, the logistic map, the doubling map); 3Blue1Brown *Chaos* visual essays for the orbit / limit-set imagery
intermediate: Brin-Stuck 2002 *Introduction to Dynamical Systems* (Cambridge University Press) Ch. 1 §1.1–§1.7 (dynamical systems as actions, orbits, limit sets, topological conjugacy); Katok-Hasselblatt 1995 *Introduction to the Modern Theory of Dynamical Systems* (Cambridge) Ch. 1 §1.1–§1.6
master: Brin-Stuck 2002 *Introduction to Dynamical Systems* (Cambridge University Press) Ch. 1–2 (orbits, limit sets, conjugacy, the circle rotation, the doubling map, toral automorphisms, the horseshoe); Katok-Hasselblatt 1995 *Introduction to the Modern Theory of Dynamical Systems* (Cambridge) Ch. 1–2; Walters 1982 *An Introduction to Ergodic Theory* (Springer GTM 79) Ch. 5; Smale 1967 *Differentiable dynamical systems* (Bull. Amer. Math. Soc. 73) — the originating survey of the structural-stability programme and the horseshoe

References

Brin, Stuck 2002 — Introduction to Dynamical Systems · Ch. 1 §1.1–§1.7 (group/semigroup actions, orbits, fixed and periodic points, omega- and alpha-limit sets, invariant sets, non-wandering set, topological conjugacy) and Ch. 2 (circle rotations, the doubling map, hyperbolic toral automorphisms, the Smale horseshoe); Cambridge University Press 2002
Katok, Hasselblatt 1995 — Introduction to the Modern Theory of Dynamical Systems · Ch. 1 §1.1–§1.6 (examples, orbits, recurrence, the non-wandering set), Ch. 2 (topological dynamics, conjugacy and equivalence); Encyclopedia of Mathematics and its Applications 54, Cambridge University Press 1995
Walters 1982 — An Introduction to Ergodic Theory · Ch. 5 (topological dynamics, minimality, topological transitivity, the non-wandering set); Springer Graduate Texts in Mathematics 79
Smale 1967 — Differentiable dynamical systems · Bulletin of the American Mathematical Society 73 (1967) 747–817; the originating survey of structural stability, Axiom A, the non-wandering set decomposition, and the Smale horseshoe as the model of robust chaotic dynamics
Birkhoff 1927 — Dynamical Systems · American Mathematical Society Colloquium Publications 9; originator of the systematic study of recurrence, the non-wandering set, minimal sets, and the recurrence theorem in topological dynamics
Devaney 1989 — An Introduction to Chaotic Dynamical Systems · 2nd edition, Addison-Wesley 1989, Ch. 1 (one-dimensional dynamics, the doubling map, symbolic coding, topological conjugacy); the standard reference for the conjugacy of the doubling map with the full shift

Estimated time

beginner: 18m
intermediate: 42m
master: 82m