38.01.05 · dynamics / topological-dynamics

The Poincaré Rotation Number and Denjoy's Theorem

shipped3 tiersLean: none

Anchor (Master): Katok-Hasselblatt 1995 *Introduction to the Modern Theory of Dynamical Systems* (Cambridge) Ch. 11–12 (circle diffeomorphisms, rotation number, Denjoy theory, the bounded-variation argument, the counterexample, mode locking); Brin-Stuck 2002 *Introduction to Dynamical Systems* (Cambridge University Press) Ch. 4; de Melo–van Strien 1993 *One-Dimensional Dynamics* (Springer Ergebnisse 25) Ch. I (circle homeomorphisms and diffeomorphisms, Denjoy, rigidity); Herman 1979 *Sur la conjugaison différentiable des difféomorphismes du cercle à des rotations* (Publ. IHÉS 49) — the smooth-conjugacy and Arnold-tongue theory

Intuition Beginner

Take any way of stirring a circle that never tears it and never reverses its direction of travel — a smooth nudge that pushes every point a little forward, faster here and slower there. Even though different points move by different amounts, the whole motion has one honest average: over many steps, how much of a full turn does a point advance, on average, each step? That single average is the rotation number. It is the one number that survives all the local speeding-up and slowing-down and tells you the long-run rhythm of the motion.

If that average is a simple fraction, say one third, the motion locks into a repeating pattern: some point comes back exactly to where it started after three steps, and the system has a genuine cycle. If the average is not any simple fraction, no point ever returns exactly home — the motion has the same irrational rhythm as a plain steady rotation, and the deep question becomes whether it is merely a disguised steady rotation or something subtler.

Here is the surprise. If the nudging map is gentle enough — its speed does not change too wildly — then an irrational rhythm forces the motion to be a steady rotation in disguise: there is a way to relabel the circle that turns your map into a plain rotation. This is Denjoy's theorem. But if you let the speed change just a little too roughly, you can build a map with an irrational rhythm that is not a disguised rotation: it skips an entire arc forever, sprinkling visits only on a dust-like Cantor set. One number sets the rhythm; one degree of smoothness decides whether it is honest.

Visual Beginner

Picture a circle with a marked starting dot and a map that pushes each point forward by an uneven amount. Track one point for many steps, count its total advance in full turns, and divide by the number of steps. This ratio settles to a fixed value — the rotation number — no matter which point you started from. Two panels show the outcomes: a rational rotation number, where the tracked point falls into a short repeating cycle, and an irrational one, where the gentle map's spots fill the circle while the rough map's spots crowd onto a Cantor dust, leaving one arc forever empty.

The left circle is the rational case: a periodic cycle exists. The right circle contrasts the gentle case (orbit fills the circle) with the rough case (orbit lands only on a Cantor set, missing the shaded arc forever).

Worked example Beginner

We follow a map of the circle (positions from $0$ up to $1$ , with $0$ and $1$ the same point) and read off its average advance per step.

Step 1. First take the plainest map: rotate everything forward by exactly $1/3$ each step. The point $0$ goes $0 \to 1/3 \to 2/3 \to 0$ . After $3$ steps it has advanced exactly $1$ full turn, so its advance per step is $1 \div 3 = 1/3$ . The rotation number is $1/3$ , and there is a genuine $3$ -step cycle: the rotation number being a simple fraction with denominator $3$ matches a cycle of length $3$ .

Step 2. Now take an uneven map that still averages the same. Suppose it pushes the point at $0$ forward by $0.30$ , the next landing forward by $0.40$ , and the one after that forward by $0.30$ , so that after $3$ steps the point at $0$ has advanced $0.30 + 0.40 + 0.30 = 1.00$ , one full turn, and returns to $0$ . The per-step advances were uneven ( $0.30, 0.40, 0.30$ ) but the average is $1.00 \div 3 = 1/3$ again. Same rotation number, same $3$ -step cycle, even though the motion was not a steady rotation.

Step 3. Read the average for a long run. If after $100$ steps a tracked point has gone around the circle $33$ whole times plus a little, its average advance per step is about $33 \div 100 = 0.33$ . After $1000$ steps you would find about $333$ turns, giving $0.333$ , and the ratio keeps settling toward $1/3$ . That stable settling value is the rotation number, and it does not depend on which point you tracked.

What this tells us: the rotation number is a long-run average that ignores how unevenly the map moves individual points. When it is a simple fraction $p / q$ , a cycle of length $q$ must exist; when it is irrational, no exact cycle can exist, and the finer question of whether the motion is a disguised steady rotation is what Denjoy's theorem answers.

Check your understanding Beginner

Exercise (easy, multiple choice).

The rotation number of an orientation-preserving circle map measures:

A. The largest distance any single point is pushed in one step. B. The long-run average advance per step, in full turns, of a tracked point. C. The number of fixed points the map has. D. How fast the map is computed.

Hint

It is an average over many steps, and it comes out the same no matter which point you follow.

Answer

B. The rotation number is the limiting average advance per step, counted in full turns, of any tracked point; the limit exists and is the same for every starting point. The biggest one-step push (A) varies from point to point, the count of fixed points (C) is a different feature, and computation speed (D) is unrelated. Feedback-correct: the rotation number is exactly the long-run average rhythm. Feedback-wrong: A, C, and D describe other quantities, not the average advance.

Formal definition Intermediate+

Let $X = R / Z$ be the circle and let $π : R \to X$ be the covering projection $π (t) = t mod 1$ . Let $f : X \to X$ be an orientation-preserving homeomorphism. A lift of $f$ is a homeomorphism $F : R \to R$ with $π \circ F = f \circ π$ ; any two lifts differ by an integer, and orientation-preservation makes $F$ strictly increasing with $F (t + 1) = F (t) + 1$ for all $t$ . Thus the displacement $φ (t) = F (t) - t$ is continuous and $1$ -periodic, and $F^{n} (t + 1) = F^{n} (t) + 1$ for every $n \in Z$ .

Definition (Poincaré rotation number). For a lift $F$ of an orientation-preserving circle homeomorphism $f$ , the rotation number of the lift is $$ \tilde\rho(F) = \lim_{n \to \infty} \frac{F^n(t) - t}{n}, $$ which exists and is independent of $t \in R$ . The rotation number of $f$ is $ρ (f) = \tilde{ρ} (F) mod 1 \in X$ ; changing the lift by $k \in Z$ changes $\tilde{ρ}$ by $k$ , so $ρ (f)$ is well-defined. For the pure rotation $R_{α} (x) = x + α$ with lift $t \mapsto t + α$ one has $ρ (R_{α}) = α mod 1$ , consistent with 38.01.04.

Definition (semiconjugacy and conjugacy). A continuous degree-one monotone map $h : X \to X$ (lifted to a non-decreasing $H : R \to R$ with $H (t + 1) = H (t) + 1$ ) is a semiconjugacy from $f$ to $g$ if $h \circ f = g \circ h$ ; it is a (topological) conjugacy if $h$ is in addition a homeomorphism. Conjugacy is the equivalence relation of 38.01.01 under which $ρ$ is invariant.

Definition (wandering interval). An open arc $I \subseteq X$ is a wandering interval for $f$ if the images $I, f (I), f^{2} (I), \dots$ are pairwise disjoint and $I$ is not contained in the basin of a periodic orbit (here, for irrational $ρ$ , simply pairwise disjoint forward images). A homeomorphism with irrational $ρ$ is minimal precisely when it has no wandering interval.

Definition (bounded variation; $C^{1 + bv}$ ). A $C^{1}$ diffeomorphism $f$ is $C^{1 + bv}$ if $lo g D f$ has bounded total variation $Var (lo g D f) = sup \sum_{i} ∣ lo g D f (x_{i + 1}) - lo g D f (x_{i}) ∣ < \infty$ over the circle. Every $C^{2}$ diffeomorphism is $C^{1 + bv}$ , since $Var (lo g D f) = \int_{X} ∣ D lo g D f ∣ \leq ∥ D^{2} f / D f ∥_{\infty} < \infty$ .

The rational case, made explicit. If $ρ (f) = p / q$ in lowest terms, then $f$ has a periodic orbit of period $q$ , and the periodic points are exactly the fixed points of $f^{q}$ ; on such an orbit, ordered around the circle, $f$ acts as the combinatorial rotation by $p / q$ . Between consecutive periodic points $f^{q}$ has no fixed point and orbits drift monotonically from one periodic point to the next, so non-periodic points are heteroclinic between periodic orbits. The map need not be minimal and need not be uniquely ergodic.

Counterexamples to common slips

The rotation number is conjugacy-invariant, not a metric quantity. Two diffeomorphisms can have wildly different displacement functions $φ$ yet the same $ρ$ ; $ρ$ depends only on the asymptotic order of the orbit, captured by the monotone semiconjugacy, not on local speeds.
Irrational $ρ$ does not by itself force minimality. The Denjoy counterexample has irrational $ρ$ yet a wandering interval and a Cantor minimal set strictly smaller than the circle. Minimality in the irrational case is a regularity phenomenon, supplied by Denjoy's theorem under $C^{1 + bv}$ .
Semiconjugacy is not conjugacy. In the Cantor-minimal-set case the monotone map $h$ collapses each complementary gap (wandering interval and its images) to a point, so $h$ is not injective; it conjugates only after the wandering intervals are removed. The map $h$ is a homeomorphism exactly when there are no wandering intervals.
$C^{1}$ is not enough. Denjoy's theorem fails at mere $C^{1}$ : the counterexample is a genuine $C^{1}$ diffeomorphism. The bounded-variation hypothesis on $lo g D f$ is the sharp dividing line, and it is what controls distortion along the first-return combinatorics.

Key theorem with proof Intermediate+

Theorem (existence and properties of the rotation number; Poincaré's classification). Let $f$ be an orientation-preserving circle homeomorphism with lift $F$ .

The limit $\tilde{ρ} (F) = lim_{n} (F^{n} (t) - t) / n$ exists and is independent of $t$ ; $ρ (f) = \tilde{ρ} (F) mod 1$ is a topological-conjugacy invariant, and $ρ$ depends continuously on $f$ in the $C^{0}$ topology.
$ρ (f) \in Q$ if and only if $f$ has a periodic orbit; if $ρ (f) = p / q$ in lowest terms then every periodic orbit has period $q$ .
If $ρ (f) = α \in / Q$ , then $f$ is semiconjugate to the rotation $R_{α}$ by a monotone degree-one map $h$ with $h \circ f = R_{α} \circ h$ . Either $f$ is minimal and $h$ is a conjugacy, or $f$ has a wandering interval and a unique minimal set $K$ which is a Cantor set, with $h$ collapsing the complementary gaps.

(See ^{[Poincaré 1885]}, ^{[Brin-Stuck 2002 §4.3]}, ^{[Katok-Hasselblatt 1995 §11.1]}.)

Proof.

Part 1 (existence via subadditivity). Set $a_{n} = max_{t} (F^{n} (t) - t)$ . Since $F^{n + m} (t) - t = (F^{m} (F^{n} t) - F^{n} t) + (F^{n} (t) - t)$ and $F^{m} (s) - s$ is $1$ -periodic in $s$ , taking maxima gives $a_{n + m} \leq a_{n} + a_{m}$ : the sequence $(a_{n})$ is subadditive, so $a_{n} / n \to in f_{n} a_{n} / n =: ℓ$ by Fekete's lemma. The displacement $φ = F - id$ is $1$ -periodic and continuous, hence bounded, so $max_{t} (F^{n} t - t) - min_{t} (F^{n} t - t) \leq 1$ (the values $F^{n} (t) - t$ over a period of length $1$ differ by less than the period because $F^{n}$ is increasing and commutes with $+ 1$ ). Therefore $(F^{n} (t) - t) / n \to ℓ$ for every $t$ , uniformly, giving existence and basepoint-independence. If $h$ conjugates $f$ to $g$ with lift $H$ , then $H F^{n} H^{- 1} = G^{n}$ and $H (s) - s$ is bounded, so $(G^{n} (s) - s) / n$ and $(F^{n} (t) - t) / n$ share the limit; thus $ρ$ is a conjugacy invariant. Continuity in $f$ follows because $ρ$ is squeezed: $⌊ F^{n} (0)⌋ / n \leq (F^{n} (0)) / n \leq ρ \leq (F^{n} (0) + 1) / n$ shows $∣ ρ (f) - p / q ∣ \leq 1/ q$ whenever $F^{q} (0) - p \in [0, 1)$ , and the same inequalities are stable under $C^{0}$ -small perturbations.

Part 2 (rationality $\Leftrightarrow$ periodic orbit). If $f^{q} (x_{0}) = x_{0}$ then a lift satisfies $F^{q} (t_{0}) = t_{0} + p$ for the representative $t_{0}$ of $x_{0}$ and some integer $p$ ; iterating, $F^{q k} (t_{0}) = t_{0} + k p$ , so $\tilde{ρ} (F) = lim_{k} (F^{q k} (t_{0}) - t_{0}) / (q k) = p / q$ , rational. Conversely suppose $\tilde{ρ} (F) = p / q$ in lowest terms; replace $F$ by $G = F^{q}$ , a lift of $f^{q}$ with $\tilde{ρ} (G) = q \cdot (p / q) = p$ , so $\tilde{ρ} (G) - p = 0$ . If $f^{q}$ had no fixed point then $g (t) := G (t) - t - p$ would be a continuous $1$ -periodic function vanishing nowhere, hence of constant sign, say $g \geq δ > 0$ ; then $G^{n} (t) - t - n p \geq n δ$ , forcing $\tilde{ρ} (G) - p \geq δ > 0$ , a contradiction (and symmetrically for $g \leq - δ$ ). So $f^{q}$ has a fixed point $x_{0}$ , a periodic point of $f$ of period dividing $q$ ; lowest terms of $p / q$ forces the period to be exactly $q$ , because a smaller period $q^{'}$ would give $ρ = p^{'} / q^{'}$ with $q^{'} < q$ .

Part 3 (semiconjugacy in the irrational case). Assume $α = ρ (f) \in / Q$ , so $f$ has no periodic points. Fix $t_{0} \in R$ and consider the countable orbit $Θ = {F^{n} (t_{0}) + m : n, m \in Z} \subseteq R$ . The key combinatorial fact is that the orbit of $t_{0}$ is ordered on the line in exactly the same cyclic order as the orbit of $t_{0}$ under the pure translation $t \mapsto t + α$ : for integers $n_{1}, n_{2}, m_{1}, m_{2}$ , $$ F^{n_1}(t_0) + m_1 < F^{n_2}(t_0) + m_2 \iff n_1\alpha + m_1 < n_2\alpha + m_2. $$ This order-isomorphism holds because $f$ has irrational rotation number and no periodic points, so the relative order of any two iterates can never reverse (a reversal would, by the intermediate value theorem applied to $F^{n} - id - m$ , produce a periodic point).

Define $H$ on the orbit $Θ$ by $H (F^{n} (t_{0}) + m) = n α + m + t_{0}^{'}$ for a fixed offset; $H$ is monotone there, hence extends to a non-decreasing surjection $H : R \to R$ by taking suprema on the closure and filling gaps by constancy. By construction $H \circ F = (\cdot + α) \circ H$ , which descends to $h \circ f = R_{α} \circ h$ . If the orbit closure is all of $X$ , the gaps are empty and $h$ is a homeomorphism: $f$ is conjugate to $R_{α}$ and is minimal. Otherwise the complement of the orbit closure is a countable union of disjoint open arcs permuted by $f$ with the dynamics of $R_{α}$ on the gaps; each is a wandering interval, $h$ collapses each to a point, and the orbit closure is a Cantor set $K$ — closed, invariant, with no isolated points (every point is a two-sided limit of the orbit) and nowhere dense (its complement is dense) — which is the unique minimal set. $□$

Bridge. The rotation number builds toward the entire classification of circle dynamics, and it appears again in 38.01.04 as the special value $ρ (R_{α}) = α$ that the rotation realises exactly. The foundational reason the construction works is that an orientation-preserving homeomorphism cannot reorder its own orbit, so the cyclic combinatorics of the orbit are forced to match those of a rigid rotation — this is exactly the order-isomorphism that produces the semiconjugacy $h$ , and it is dual to the way an invariant measure pins down the rotation in 38.01.04, since $h$ is nothing but the cumulative distribution function of the unique invariant measure. The central insight is that $ρ$ is the complete invariant of the combinatorial type while the dichotomy minimal-versus-Cantor is the residual geometric type that smoothness will decide; putting these together — existence from subadditivity, the rational case from the intermediate value theorem, the irrational case from the order-isomorphism — the bridge is that Poincaré reduces every circle homeomorphism to a rotation up to a monotone collapse, leaving exactly one question, the presence of wandering intervals, for Denjoy's theorem below to settle.

Exercises Intermediate+

Exercise 6 (medium, short-answer).

Explain why an orientation-preserving circle homeomorphism cannot reverse the cyclic order of two distinct orbit points, and why this forbids periodic points when $ρ$ is irrational.

Hint

Use that lifts are strictly increasing; a reversal of order would force a coincidence, hence a periodic point.

Answer

A lift $F$ is strictly increasing, so it preserves the order of points on $R$ : $t < s \Rightarrow F (t) < F (s)$ , and inductively $F^{n} (t) < F^{n} (s)$ . Hence the cyclic order of orbit points on the circle is preserved by $f$ and never reverses. If two iterates $F^{n_{1}} (t_{0}) + m_{1}$ and $F^{n_{2}} (t_{0}) + m_{2}$ were to swap order relative to the corresponding rotation values $n_{1} α + m_{1}$ and $n_{2} α + m_{2}$ , the continuous function $F^{n_{2} - n_{1}} - id - (m_{1} - m_{2})$ would change sign and so vanish, producing a point with $F^{k} (t) = t + m$ , i.e. a periodic point. For irrational $α$ there are no periodic points (Part 2), so no such sign change occurs and the order is rigidly that of the rotation. Rubric: full credit for the monotonicity-preserves-order argument and the intermediate-value contradiction.

Exercise 7 (hard, symbolic).

Prove the easy half of Denjoy's distortion control: if $f$ is a $C^{1}$ circle diffeomorphism and $J$ is an interval such that $J, f (J), \dots, f^{n - 1} (J)$ are pairwise disjoint, show that $\sum_{k = 0}^{n - 1} lo g D f (x_{k}) - lo g D f (y_{k}) \leq Var (lo g D f)$ for any choice of points $x_{k}, y_{k} \in f^{k} (J)$ , where $Var$ is the total variation of $lo g D f$ over the circle.

Hint

The intervals $f^{k} (J)$ are disjoint, so the variation contributions of $lo g D f$ over them add up to at most the total variation over the whole circle.

Answer

For each $k$ , $x_{k}$ and $y_{k}$ lie in $f^{k} (J)$ , so $∣ lo g D f (x_{k}) - lo g D f (y_{k}) ∣ \leq Var_{f^{k} (J)} (lo g D f)$ , the total variation of $lo g D f$ restricted to the interval $f^{k} (J)$ . Summing over $k$ , $$ \sum_{k=0}^{n-1} |\log Df(x_k) - \log Df(y_k)| \le \sum_{k=0}^{n-1} \mathrm{Var}{f^k(J)}(\log Df). $$ Because the intervals $f^{0} (J), \dots, f^{n - 1} (J)$ are pairwise disjoint, the restricted variations over them sum to at most the total variation over the whole circle: variation is additive over disjoint subintervals and monotone under enlarging the domain, so $\sum_k \mathrm{Var}{f^k(J)}(\log Df) \le \mathrm{Var}(\log Df) $. C o mbinin g g i v es t h ec l aim . T hi s i se x a c tl y t h e b o u n d t ha t, a ppl i e d t o t h e f i r s t - r e t u r nin t er v a l s (w hi c ha r e d i s j o in t b y t h eco mbina t or i cso f i r r a t i o na l$ \rho $), k ee p s t h e d i s t or t i o n$ \prod_k Df / Df$ bounded and so excludes wandering intervals — the heart of Denjoy's theorem. Rubric: full credit for the per-interval variation bound and the disjointness-gives-additivity step.

Exercise 8 (hard, short-answer).

Sketch why the Denjoy counterexample requires the inserted intervals to have summable length, and what goes wrong (and right) at the boundary between $C^{1}$ and $C^{1 + bv}$ .

Hint

The inserted gaps $I_{n}$ replace an orbit; their total length must be finite to fit in the circle, and the derivative must match up $C^{1}$ across the now-Cantor minimal set.

Answer

To build the counterexample, fix an irrational $α$ and the rotation orbit ${n α}$ , and replace each orbit point by an inserted open interval $I_{n}$ of length $ℓ_{n} > 0$ with $\sum_{n} ℓ_{n} < \infty$ (summability is forced so the inserted intervals fit into a circle of finite length, leaving a Cantor set of positive-or-zero measure as the complement). One defines $f$ to map $I_{n}$ affinely-up-to-correction onto $I_{n + 1}$ , preserving the rotation order, and chooses the derivative on each $I_{n}$ so that the pieces glue to a global $C^{1}$ diffeomorphism: the derivative must tend to $1$ at the endpoints (the Cantor set), which is arranged by tapering $D f$ across $I_{n}$ with a bump of total variation comparable to $∣ ℓ_{n + 1} / ℓ_{n} - 1∣$ . Summing these variation contributions diverges — $lo g D f$ has unbounded total variation — which is precisely why the construction lives at $C^{1}$ but not $C^{1 + bv}$ : the orbit $I_{n}$ is a wandering interval, so $f$ is not minimal, its minimal set is the Cantor complement, and $f$ is only semiconjugate (not conjugate) to $R_{α}$ . Denjoy's theorem says the bounded-variation hypothesis bounds the accumulated distortion (Exercise 7), forbids the wandering interval, and forces conjugacy; the counterexample shows this hypothesis is sharp. Rubric: full credit for the summable-length necessity, the $C^{1}$ -gluing of the derivative, and the unbounded-variation explanation of why $C^{1 + bv}$ excludes it.

Advanced results Master

Theorem (Denjoy's theorem). Let $f$ be an orientation-preserving $C^{1}$ circle diffeomorphism whose derivative $lo g D f$ has bounded variation — in particular any $C^{2}$ diffeomorphism — and suppose $ρ (f) = α \in / Q$ . Then $f$ has no wandering interval; consequently $f$ is minimal and topologically conjugate to the rotation $R_{α}$ . (See ^{[Denjoy 1932]}, ^{[Katok-Hasselblatt 1995 §12.1]}.)

The mechanism is distortion control along the first-return combinatorics of the irrational rotation. Suppose a wandering interval $J$ existed: then $J, f (J), f^{2} (J), \dots$ are pairwise disjoint, so $\sum_{k} ∣ f^{k} (J) ∣ \leq 1$ and the lengths $∣ f^{k} (J) ∣ \to 0$ . Denjoy's estimate uses the continued-fraction denominators $q_{n}$ of $α$ (the closest-return times of 38.01.04) to select two long blocks of disjoint iterates straddling $J$ ; the bounded variation of $lo g D f$ caps the total distortion $\sum_{k} ∣ lo g D f^{q_{n}} (x_{k}) - lo g D f^{q_{n}} (y_{k}) ∣$ by $Var (lo g D f)$ uniformly in $n$ , so the derivatives of the return maps $f^{q_{n}}$ on the relevant intervals stay within a fixed ratio $C = e^{Var}$ of one another. This bounded distortion is incompatible with the lengths $∣ f^{k} (J) ∣$ shrinking while their order-combinatorics fold $J$ around the circle: comparing $J$ with its near-returns forces $\sum_{k} ∣ f^{k} (J) ∣ = \infty$ , contradicting disjointness inside a finite circle. Hence no wandering interval exists, the orbit is dense, and Part 3 of the Poincaré theorem upgrades the semiconjugacy $h$ to a conjugacy.

Theorem (the Denjoy counterexample). For every irrational $α$ there is an orientation-preserving $C^{1}$ circle diffeomorphism $f$ with $ρ (f) = α$ that is not minimal: $f$ has a wandering interval, its unique minimal set is a Cantor set $K ⊊ X$ , and $f$ is semiconjugate but not conjugate to $R_{α}$ . (See ^{[Denjoy 1932]}, ^{[de Melo-van Strien 1993 Ch. I]}.)

The construction blows up the orbit ${n α}$ of the rotation into inserted intervals $I_{n}$ of summable length $ℓ_{n}$ , defining $f$ to carry $I_{n}$ onto $I_{n + 1}$ in the rotation order and to fix the Cantor complement combinatorially. The derivative is tapered across each $I_{n}$ so the map is globally $C^{1}$ with derivative $1$ on $K$ ; the price is that $lo g D f$ accumulates variation $\sim \sum_{n} ∣ ℓ_{n + 1} / ℓ_{n} - 1∣$ , which is made to diverge, so $f$ is $C^{1}$ but never $C^{1 + bv}$ . The two theorems together pin the smoothness threshold exactly: Denjoy's conjugacy holds at $C^{1 + bv}$ and fails at $C^{1}$ .

Theorem (continuity, monotonicity, and mode locking). The map $f \mapsto ρ (f)$ is continuous in the $C^{0}$ topology and monotone along ordered one-parameter families: if $f_{ω} (x) = F_{ω} (x)$ with $F_{ω}$ pointwise non-decreasing in $ω$ , then $ω \mapsto ρ (f_{ω})$ is non-decreasing. At every rational value $p / q$ taken with a structurally stable (hyperbolic) periodic orbit, $ρ$ is locally constant — an Arnold tongue — so $ω \mapsto ρ (f_{ω})$ is a devil's staircase: continuous, non-decreasing, constant on a dense open set of $ω$ , yet attaining every irrational value on a Cantor set of parameters. (See ^{[Herman 1979]}, ^{[Katok-Hasselblatt 1995 §12.3]}.)

Mode locking is the parametric face of Part 2: a rational rotation number is carried by a periodic orbit, and a hyperbolic periodic orbit persists under perturbation, so $ρ$ cannot move off $p / q$ until the orbit is destroyed in a saddle-node bifurcation at the edge of the tongue. Irrational rotation numbers, by contrast, are unstable — any value can be perturbed away — so they occupy only a nowhere-dense Cantor set in parameter space, of positive Lebesgue measure when the partial quotients of $α$ are controlled (the Diophantine condition of Herman's theory). For a two-parameter family the tongues open from the $ω$ -axis at every rational with width growing like the strength of the nonlinearity, the Arnold-tongue picture that organises the transition from quasiperiodicity to phase locking.

Theorem (Herman–Yoccoz smooth linearisation; preview). If $f$ is a $C^{\infty}$ (or real-analytic) orientation-preserving circle diffeomorphism whose rotation number $α$ is Diophantine — there are $C, τ$ with $∣ α - p / q ∣ \geq C q^{- 2 - τ}$ for all $p / q$ — then the topological conjugacy of Denjoy's theorem is itself $C^{\infty}$ (resp. analytic): $f$ is smoothly conjugate to $R_{α}$ . For Liouville $α$ smooth conjugacy can fail even when topological conjugacy holds. (See ^{[Herman 1979]}.)

This sharpens Denjoy's topological conjugacy to a differentiable one under an arithmetic condition on $α$ , completing the regularity ladder: $C^{1 + bv}$ buys topological conjugacy unconditionally (Denjoy), while smooth conjugacy demands both extra smoothness of $f$ and Diophantine control of $ρ$ , with small divisors $e^{2 π i q α} - 1$ governing the linearised cohomological equation $h \circ f = R_{α} \circ h$ exactly as in the rotation eigenvalue computation of 38.01.04.

Synthesis. The rotation number is the single conjugacy invariant of circle homeomorphisms, and the foundational reason the whole theory coheres is that an orientation-preserving map cannot reorder its orbit, so its combinatorics are forced to be those of a rotation $R_{ρ}$ , recorded by the monotone semiconjugacy $h$ — which is exactly the cumulative distribution of the invariant measure of 38.01.04 read as a change of coordinate. The central insight is a clean factorisation of the classification into a combinatorial layer and a geometric layer: $ρ$ fixes the combinatorial type (rational $\Leftrightarrow$ periodic orbit, irrational $\Rightarrow$ rotation-ordered orbit), and the single remaining question — does a wandering interval exist — is the geometric residue that smoothness decides.

This is dual to the measure-theoretic picture of 38.01.04, where unique ergodicity removed the exceptional set; here bounded variation of $lo g D f$ plays the analogous role, removing wandering intervals by capping distortion, and putting these together the Poincaré dichotomy and the Denjoy threshold are one statement at two resolutions: minimality is generic among smooth enough maps and exceptional below the threshold. The bridge is that the irrational rotation of 38.01.04 is the universal model toward which every irrational-rotation-number diffeomorphism is driven — conjugate to it when $C^{1 + bv}$ , semiconjugate to it on a Cantor set when only $C^{1}$ — so the entire chapter's topological dynamics, 38.01.02's minimality and 38.01.04's equidistribution, generalises into the one-dimensional smooth theory exactly through this rotation number and its smoothness-graded refinements.

Full proof set Master

Proposition 1 (existence of the rotation number). For a lift $F$ of an orientation-preserving circle homeomorphism, $lim_{n} (F^{n} (t) - t) / n$ exists and is independent of $t$ .

Proof. Let $b_{n} = sup_{t} (F^{n} (t) - t)$ . For any $t$ , write $F^{n + m} (t) - t = [F^{m} (F^{n} t) - F^{n} t] + [F^{n} (t) - t]$ . The first bracket is the displacement of $F^{m}$ at the point $F^{n} (t)$ and is $1$ -periodic in its argument, so it is $\leq b_{m}$ ; the second is $\leq b_{n}$ . Taking the supremum over $t$ gives $b_{n + m} \leq b_{n} + b_{m}$ : the sequence $(b_{n})$ is subadditive, so by Fekete's lemma $b_{n} / n \to ℓ := in f_{n} b_{n} / n$ . Because $F^{n}$ is increasing and commutes with the unit translation, $sup_{t} (F^{n} t - t) - in f_{t} (F^{n} t - t) \leq 1$ , so $b_{n} - n (F^{n} (t) - t) / n \leq 1$ for every $t$ , i.e. $∣ (F^{n} (t) - t) / n - b_{n} / n ∣ \leq 1/ n$ . Hence $(F^{n} (t) - t) / n \to ℓ$ for every $t$ , and the limit is basepoint-independent. $□$

Proposition 2 (rational rotation number forces a periodic orbit). If $\tilde{ρ} (F) = p / q$ in lowest terms, then $f$ has a periodic point of period exactly $q$ .

Proof. Put $G = F^{q} - p$ , a lift-shift with $\tilde{ρ} (G) = q \tilde{ρ} (F) - p = 0$ . Consider $g (t) = G (t) - t$ , continuous and $1$ -periodic. If $g$ never vanished it would have constant sign by the intermediate value theorem; say $g \geq δ > 0$ . Then $G^{n} (t) - t = \sum_{k = 0}^{n - 1} g (G^{k} t) \geq n δ$ , so $\tilde{ρ} (G) \geq δ > 0$ , contradicting $\tilde{ρ} (G) = 0$ ; the case $g \leq - δ$ is symmetric. Hence $g (t_{0}) = 0$ for some $t_{0}$ , i.e. $F^{q} (t_{0}) = t_{0} + p$ , so $π (t_{0})$ is periodic of period dividing $q$ . If the period were $q^{'} < q$ then $F^{q^{'}} (t_{0}) = t_{0} + p^{'}$ for some integer $p^{'}$ , giving $\tilde{ρ} (F) = p^{'} / q^{'}$ with denominator $< q$ , contradicting lowest terms. So the period is $q$ . $□$

Proposition 3 (the orbit order-embeds into the rotation orbit when $ρ$ is irrational). If $ρ (f) = α \in / Q$ , then for all integers $n_{1}, n_{2}, m_{1}, m_{2}$ , $$ F^{n_1}(t_0) + m_1 < F^{n_2}(t_0) + m_2 \iff n_1\alpha + m_1 < n_2\alpha + m_2. $$

Proof. It suffices to treat $n_{1} = 0$ by applying $F^{- n_{1}}$ (an increasing bijection commuting with unit translation) to both sides, so consider $t_{0} + m_{1}$ versus $F^{n} (t_{0}) + m_{2}$ with $n = n_{2} - n_{1}$ , $m = m_{2} - m_{1}$ . Define $ψ (t) = F^{n} (t) - t - m$ , continuous and $1$ -periodic. Since $f$ has no periodic point (irrational $ρ$ , Proposition 2), $ψ$ never vanishes, so it has constant sign. Its sign is determined by the rotation comparison: $\tilde{ρ} (F^{n}) = n α$ , and integrating the sign of $ψ$ against the rotation shows $ψ > 0$ everywhere iff $n α - m > 0$ and $ψ < 0$ everywhere iff $n α - m < 0$ (a sign change in either would force a zero of $ψ$ or contradict the value of the rotation number). Evaluating at $t_{0}$ , $F^{n} (t_{0}) + m_{1} - (t_{0} + m_{2})$ has the same sign as $n α + m_{1} - m_{2}$ , which is the claimed equivalence. $□$

Proposition 4 (semiconjugacy to the rotation). If $ρ (f) = α \in / Q$ there is a non-decreasing degree-one continuous $h : X \to X$ with $h \circ f = R_{α} \circ h$ , a homeomorphism iff $f$ is minimal.

Proof. By Proposition 3 the map $Θ \to R$ , $F^{n} (t_{0}) + m \mapsto n α + m + c$ (fixed constant $c$ ), is strictly order-preserving on the orbit $Θ$ . Extend it to $H : R \to R$ by $H (t) = sup {n α + m + c : F^{n} (t_{0}) + m \leq t}$ . Then $H$ is non-decreasing, satisfies $H (t + 1) = H (t) + 1$ (the supremand set shifts by adding $1$ to $m$ ), and is continuous because its image is dense (the rotation orbit ${n α + m}$ is dense in $R$ ), so $H$ has no jumps. From $F (Θ) \subseteq Θ$ shifting indices by $1$ , $H (F (t)) = H (t) + α$ , which descends to $h \circ f = R_{α} \circ h$ . The map $H$ is constant exactly on the closures of the complementary gaps of $\overline{Θ}$ ; these gaps are empty iff $\overline{Θ} = R$ iff $f$ is minimal, in which case $H$ is strictly increasing and $h$ is a homeomorphism. $□$

Proposition 5 (Denjoy bounded-distortion bound). Let $f$ be a $C^{1}$ circle diffeomorphism with $V := Var (lo g D f) < \infty$ , and let $J, f (J), \dots, f^{n - 1} (J)$ be pairwise disjoint. Then for all $x, y \in J$ , $$ e^{-V} \le \frac{Df^n(x)}{Df^n(y)} \le e^{V}. $$

Proof. By the chain rule $lo g D f^{n} (x) - lo g D f^{n} (y) = \sum_{k = 0}^{n - 1} (lo g D f (f^{k} x) - lo g D f (f^{k} y))$ , with $f^{k} x, f^{k} y \in f^{k} (J)$ . Each term is bounded in absolute value by $Var_{f^{k} (J)} (lo g D f)$ , the variation of $lo g D f$ over the interval $f^{k} (J)$ . Since the intervals $f^{k} (J)$ , $0 \leq k \leq n - 1$ , are pairwise disjoint and variation is additive over disjoint subintervals and monotone in the domain, $\sum_{k = 0}^{n - 1} Var_{f^{k} (J)} (lo g D f) \leq Var (lo g D f) = V$ . Hence $∣ lo g D f^{n} (x) - lo g D f^{n} (y) ∣ \leq V$ , and exponentiating gives the stated two-sided bound. This uniform-in- $n$ control of the return-map distortion is the engine that, applied along the closest-return blocks of an assumed wandering interval, contradicts the summability $\sum_{k} ∣ f^{k} (J) ∣ \leq 1$ and so proves Denjoy's theorem. $□$

Connections Master

Circle rotations and unique ergodicity 38.01.04. The rotation $R_{α}$ is the universal target of this unit: $ρ (R_{α}) = α$ , and every orientation-preserving circle homeomorphism with that rotation number is semiconjugate to $R_{α}$ , conjugate to it when minimal. The unique invariant measure constructed there is exactly the object whose cumulative distribution function is the semiconjugacy $h$ here, and the small divisors $e^{2 π i q α} - 1$ governing smooth linearisation are the rotation eigenvalues computed in that unit.
Minimality and recurrence 38.01.02. Poincaré's classification feeds directly into the minimal-set theory of that unit: in the irrational case either the whole circle is minimal (Denjoy, $C^{1 + bv}$ ) or the unique minimal set is a Cantor set (the counterexample), and this Cantor minimal set is the model totally-disconnected minimal system promised there. The almost-periodic points and syndetic returns of that unit are realised by every orbit of a Denjoy-minimal diffeomorphism.
Dynamical systems, orbits, and limit sets 38.01.01. The conjugacy invariance of $ρ$ is an instance of the conjugacy framework of that unit, and the wandering interval is the one-dimensional incarnation of wandering points and the complement of the non-wandering set $Ω (f)$ defined there; the Cantor minimal set is the non-wandering set in the counterexample.

Historical & philosophical context Master

Henri Poincaré introduced the rotation number and proved its existence, rationality dichotomy, and the classification of the irrational case in the third part of his 1885 memoir Sur les courbes définies par les équations différentielles ^{[Poincaré 1885]}, where circle homeomorphisms arise as the return maps of flows on the torus. Poincaré already saw that an irrational rotation number forces the orbit to be ordered like a rigid rotation and conjectured that the exceptional Cantor case could occur, but he could not decide whether smoothness excluded it. The question stood for nearly half a century until Arnaud Denjoy's 1932 Journal de Mathématiques paper ^{[Denjoy 1932]} proved that a $C^{1}$ diffeomorphism with derivative of bounded variation — in particular any $C^{2}$ diffeomorphism — has no wandering interval and is therefore conjugate to the rotation, and in the same paper constructed the $C^{1}$ counterexample showing the hypothesis sharp.

The differentiable theory was completed by Michael Herman in his 1979 IHÉS memoir ^{[Herman 1979]}, which proved that for Diophantine rotation numbers a smooth diffeomorphism is smoothly conjugate to the rotation, organising the result through the small-divisor analysis later refined by Jean-Christophe Yoccoz; the Arnold-tongue and mode-locking picture, originating in Vladimir Arnold's study of the standard family, sits within this theory. Welington de Melo and Sebastian van Strien's One-Dimensional Dynamics ^{[de Melo-van Strien 1993]} gives the modern cross-ratio proof of Denjoy's theorem and its extension to maps with critical points, and Anatole Katok and Boris Hasselblatt's Introduction to the Modern Theory of Dynamical Systems ^{[Katok-Hasselblatt 1995]} presents the rotation number, Poincaré's classification, and Denjoy's theorem and counterexample as the foundational chapter of circle dynamics.

Bibliography Master

@article{Poincare1885,
  author  = {Poincar\'e, Henri},
  title   = {Sur les courbes d\'efinies par les \'equations diff\'erentielles, III},
  journal = {Journal de Math\'ematiques Pures et Appliqu\'ees},
  series  = {(4)},
  volume  = {1},
  year    = {1885},
  pages   = {167--244}
}

@article{Denjoy1932,
  author  = {Denjoy, Arnaud},
  title   = {Sur les courbes d\'efinies par les \'equations diff\'erentielles \`a la surface du tore},
  journal = {Journal de Math\'ematiques Pures et Appliqu\'ees},
  series  = {(9)},
  volume  = {11},
  year    = {1932},
  pages   = {333--375}
}

@article{Herman1979,
  author  = {Herman, Michael R.},
  title   = {Sur la conjugaison diff\'erentiable des diff\'eomorphismes du cercle \`a des rotations},
  journal = {Publications Math\'ematiques de l'IH\'ES},
  volume  = {49},
  year    = {1979},
  pages   = {5--233}
}

@book{deMelovanStrien1993,
  author    = {de Melo, Welington and van Strien, Sebastian},
  title     = {One-Dimensional Dynamics},
  publisher = {Springer-Verlag},
  series    = {Ergebnisse der Mathematik und ihrer Grenzgebiete},
  volume    = {25},
  year      = {1993}
}

@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{Yoccoz1984,
  author  = {Yoccoz, Jean-Christophe},
  title   = {Conjugaison diff\'erentiable des diff\'eomorphismes du cercle dont le nombre de rotation v\'erifie une condition diophantienne},
  journal = {Annales Scientifiques de l'\'Ecole Normale Sup\'erieure},
  series  = {(4)},
  volume  = {17},
  year    = {1984},
  pages   = {333--359}
}

Prerequisites

38.01.04
38.01.02

Tier anchors

beginner: Strogatz 1994 *Nonlinear Dynamics and Chaos* (Addison-Wesley) Ch. 8 (circle maps, the average winding rate, phase locking and Arnold tongues); 3Blue1Brown — the winding-number and irrational-rotation visualisations for the average-rotation-per-step imagery
intermediate: Brin-Stuck 2002 *Introduction to Dynamical Systems* (Cambridge University Press) Ch. 4 §4.3–§4.4 (lifts, the rotation number, the rational/periodic-orbit equivalence, Poincaré's classification, Denjoy's theorem and counterexample); Katok-Hasselblatt 1995 *Introduction to the Modern Theory of Dynamical Systems* (Cambridge) §11.1–§11.2 and §12.1–§12.3
master: Katok-Hasselblatt 1995 *Introduction to the Modern Theory of Dynamical Systems* (Cambridge) Ch. 11–12 (circle diffeomorphisms, rotation number, Denjoy theory, the bounded-variation argument, the counterexample, mode locking); Brin-Stuck 2002 *Introduction to Dynamical Systems* (Cambridge University Press) Ch. 4; de Melo–van Strien 1993 *One-Dimensional Dynamics* (Springer Ergebnisse 25) Ch. I (circle homeomorphisms and diffeomorphisms, Denjoy, rigidity); Herman 1979 *Sur la conjugaison différentiable des difféomorphismes du cercle à des rotations* (Publ. IHÉS 49) — the smooth-conjugacy and Arnold-tongue theory

References

Brin, Stuck 2002 — Introduction to Dynamical Systems · Ch. 4 §4.3–§4.4 (the lift $F$ of an orientation-preserving circle homeomorphism; the Poincaré rotation number $\rho(f)=\lim_n (F^n(x)-x)/n$, existence and independence of $x$, conjugacy invariance and continuity; $\rho$ rational iff a periodic orbit exists; Poincaré's classification of the irrational case as minimal-or-Cantor; Denjoy's theorem for $C^2$ diffeomorphisms and the $C^1$ counterexample); Cambridge University Press 2002
Katok, Hasselblatt 1995 — Introduction to the Modern Theory of Dynamical Systems · §11.1–§11.2 (rotation number, its properties, the dichotomy via the semiconjugacy to $R_\rho$) and §12.1–§12.3 (Denjoy's theorem via bounded variation of $\log Df$, wandering intervals, the Denjoy construction); Encyclopedia of Mathematics and its Applications 54, Cambridge University Press 1995
Poincaré 1885 — Sur les courbes définies par les équations différentielles, III · Journal de Mathématiques Pures et Appliquées (4) 1 (1885), 167–244 (the rotation number $\rho$ of a flow on the torus / a circle homeomorphism, its existence and rationality dichotomy, the classification of the irrational case as conjugate-or-semiconjugate to a rotation); the founding memoir
Denjoy 1932 — Sur les courbes définies par les équations différentielles à la surface du tore · Journal de Mathématiques Pures et Appliquées (9) 11 (1932), 333–375 (the $C^2$ — more precisely $C^1$ with derivative of bounded variation — theorem that an orientation-preserving circle diffeomorphism with irrational rotation number is topologically conjugate to the rotation $R_\rho$, together with the $C^1$ counterexample possessing a wandering interval and a Cantor minimal set)
de Melo, van Strien 1993 — One-Dimensional Dynamics · Springer Ergebnisse der Mathematik 25, Ch. I (circle homeomorphisms and diffeomorphisms; the rotation number; Poincaré's and Denjoy's theorems with the bounded-distortion / cross-ratio proof; an introduction to the rigidity and renormalisation theory of critical circle maps)
Herman 1979 — Sur la conjugaison différentiable des difféomorphismes du cercle à des rotations · Publications Mathématiques de l'IHÉS 49 (1979), 5–233 (the smooth/analytic linearisation theory: for Diophantine $\rho$ a smooth diffeomorphism is smoothly conjugate to $R_\rho$; the local and global Arnold-tongue and mode-locking picture, sharpening Denjoy's topological conjugacy to differentiable conjugacy under arithmetic conditions on $\rho$)

Estimated time

beginner: 18m
intermediate: 50m
master: 90m