38.03.04 · dynamics / hyperbolicity-structural-stability

Shadowing and Structural Stability

shipped3 tiersLean: none

Anchor (Master): Katok-Hasselblatt 1995 *Introduction to the Modern Theory of Dynamical Systems* (Cambridge) Ch. 18 §18.1 (the Anosov shadowing theorem and closing lemma), §18.2 (structural stability of Anosov systems via shadowing), §18.5 ($\Omega$-stability and the structural-stability theorem); Pilyugin 1999 *Shadowing in Dynamical Systems* (Springer LNM 1706); Mañé 1988 *A proof of the $C^1$ stability conjecture* (Publ. IHÉS 66)

Intuition Beginner

Run a chaotic rule on a computer and a quiet disaster unfolds. Every step rounds the numbers a little, so the sequence of points the machine prints is not a real orbit of the rule — it is a fake orbit, one where each point is only approximately where the rule would have sent the previous one. In a stretching, chaotic system these tiny errors blow up fast, so after a few dozen steps the printed path has wandered far from the orbit you started with. It looks like the computation is worthless.

Shadowing is the rescue. It says: yes, your printed path is not the orbit you intended, but there is some genuine orbit of the rule that stays glued to your printed path the whole way — a real orbit shadowing the fake one, never straying more than a hair from it. So the picture on your screen is honest after all. It is not the orbit you asked for, but it is a true orbit of the very same rule, and for understanding the long-term shape of the dynamics that is exactly as good.

Why does this gift appear precisely in the most chaotic, stretching systems? Because the same uniform stretching that amplifies errors in one direction is uniform shrinking in another, and that balance lets a correcting true orbit be found and pinned down. The payoff is enormous: shadowing is the engine that proves a slightly bumped chaotic rule behaves exactly like the original, just relabelled — the property called structural stability.

Visual Beginner

Imagine a dotted path drawn across the page: these dots are the fake orbit, each one placed a small jump away from where the rule would carry the previous dot. Now draw a smooth solid curve that threads through a thin tube around the dotted path, staying inside the tube at every step. The solid curve is a real orbit of the rule; the dotted path is the pseudo-orbit it shadows. The tube has a fixed small width no matter how long the path runs.

A small table fixes the vocabulary.

name	what it is	everyday image
pseudo-orbit	a fake orbit with small jumps allowed	the computer's rounded printout
shadowing	a true orbit staying near the fake one	the solid curve in the tube
$δ$ (delta)	how big the jumps are allowed to be	how sloppy the printout is
$ε$ (epsilon)	how close the true orbit stays	the width of the tube

Worked example Beginner

Take the simplest stretching rule on the line: doubling, then wrapping back into the interval from $0$ to $1$ . So a point $x$ goes to the fractional part of $2 x$ . This rule is chaotic — it stretches distances by a factor of two each step.

Step 1. Build a pseudo-orbit. Start at $0.30$ . The rule sends it to $0.60$ , but say the machine rounds and records $0.61$ instead — a jump of $0.01$ . From $0.61$ the rule gives $0.22$ , but the machine records $0.23$ — again a jump of $0.01$ . Continue: the recorded fake orbit is $0.30, 0.61, 0.23, \dots$ , each step off by at most $δ = 0.01$ .

Step 2. Ask for a true orbit nearby. Because doubling spreads points out, a true orbit is pinned down by its future: to shadow the fake path we nudge the start slightly. Trying starts near $0.30$ and doubling exactly, one finds a genuine starting point — about $0.3017$ — whose exact doubling orbit stays within $ε = 0.02$ of every recorded point.

Step 3. Check the first few steps of the true orbit $0.3017 \to 0.6034 \to 0.2068 \to \dots$ against the fake $0.30 \to 0.61 \to 0.23 \to \dots$ : the gaps are $0.0017, 0.0066, 0.0232$ — staying within the tube.

What this tells us: a sloppy computed path with per-step error $0.01$ was shadowed by an exact orbit of the same doubling rule within $0.02$ . The small jump size $δ$ bought a small but fixed tube width $ε$ , and the fix held for the whole run rather than falling apart as the chaos amplified errors — the hallmark of shadowing in a stretching system.

Check your understanding Beginner

Exercise (easy, multiple choice).

A pseudo-orbit (a $δ$ -chain) of a rule $f$ is best described as a sequence of points in which:

A. every point is exactly the rule applied to the previous point. B. each point is within a small distance $δ$ of the rule applied to the previous point. C. the points repeat after finitely many steps. D. the distance between consecutive points shrinks to zero.

Hint

A pseudo-orbit is the "fake" orbit a computer prints: the rule is followed only approximately, with a small allowed slip at each step.

Answer

B. A $δ$ -pseudo-orbit is a sequence $x_{0}, x_{1}, x_{2}, \dots$ in which the rule is followed up to an error $δ$ : the distance from $f (x_{n})$ to the next listed point $x_{n + 1}$ is at most $δ$ . Option A describes a genuine orbit (no slip allowed); C describes a periodic orbit; D describes a different convergence property. The whole point of shadowing is that an approximate B-style path is tracked by a true A-style orbit.

Formal definition Intermediate+

Let $M$ be a smooth Riemannian manifold with distance $d$ , $f : M \to M$ a $C^{1}$ diffeomorphism, and $Λ \subseteq M$ a hyperbolic set with splitting $T_{Λ} M = E^{s} \oplus E^{u}$ and uniform constants $(C, λ)$ , $0 < λ < 1$ , as in 38.03.01. Throughout, $W_{ε}^{s} (x), W_{ε}^{u} (x)$ are the local stable/unstable discs of 38.03.03.

Definition ( $δ$ -pseudo-orbit). For $δ > 0$ , a (two-sided) $δ$ -pseudo-orbit or $δ$ -chain of $f$ is a sequence ${x_{n}}_{n \in Z}$ (or $n \geq 0$ ) in $M$ with $$ d\big(f(x_n),, x_{n+1}\big) \leq \delta \quad \text{for all } n. $$ A pseudo-orbit is periodic of period $p$ if $x_{n + p} = x_{n}$ for all $n$ . A finite segment with $d (f (x_{p - 1}), x_{0}) \leq δ$ is a $δ$ -cycle.

Definition ( $ε$ -shadowing). A point $y \in M$ $ε$ -shadows the pseudo-orbit ${x_{n}}$ if its genuine orbit stays $ε$ -close: $$ d\big(f^n(y),, x_n\big) \leq \varepsilon \quad \text{for all } n. $$ The system has the shadowing property on $Λ$ if for every $ε > 0$ there is $δ > 0$ such that every $δ$ -pseudo-orbit lying in a fixed neighbourhood of $Λ$ is $ε$ -shadowed by some $y$ .

Definition (locally maximal hyperbolic set). $Λ$ is locally maximal (or isolated) if there is an open $V \supseteq Λ$ with $Λ = ⋂_{n \in Z} f^{n} (V)$ . Anosov diffeomorphisms ( $Λ = M$ ) and the basic sets of an Axiom-A system 38.03.01 are locally maximal; local maximality is what makes the shadowing point lie in $Λ$ and be unique.

Sign / convention. Forward time contracts $E^{s}$ and expands $E^{u}$ , matching 38.03.01 and 38.03.03 and Katok-Hasselblatt ^{[Katok-Hasselblatt 1995 §18.1]}. The correction operator below is invertible because $I - A$ has the contracting block on $E^{s}$ inverted forward and the expanding block on $E^{u}$ inverted backward, the discrete analogue of the Perron forward/backward summation of 38.03.03.

Definition (topological conjugacy; structural stability). Diffeomorphisms $f, g : M \to M$ are topologically conjugate if there is a homeomorphism $h : M \to M$ with $h \circ f = g \circ h$ . $f$ is $C^{1}$ -structurally stable if there is a $C^{1}$ -neighbourhood $U$ of $f$ such that every $g \in U$ is topologically conjugate to $f$ .

Definition (Axiom A; strong transversality; no-cycle). $f$ satisfies Axiom A if $Ω (f)$ is hyperbolic with dense periodic points 38.03.01. It satisfies strong transversality if for all $x, y \in M$ the global stable and unstable manifolds meet transversally: $W^{s} (x) ⋔ W^{u} (y)$ . The basic sets satisfy the no-cycle condition if the relation $Ω_{i} \to Ω_{j}$ (when $W^{u} (Ω_{i}) \cap W^{s} (Ω_{j}) \neq = \emptyset$ ) generates no cycle among distinct basic sets.

Counterexamples to common slips

Shadowing is not interpolation. The shadowing orbit is a genuine orbit of $f$ ; it does not merely pass near the listed points at the listed times by some independent reparametrisation. The same $y$ , iterated by $f$ , must track $x_{n}$ at time $n$ for every $n$ simultaneously.
Hyperbolicity is essential, not decorative. An irrational rotation of the circle has no nonzero Lyapunov exponents and fails shadowing: a $δ$ -pseudo-orbit can accumulate a drift of order $N δ$ over $N$ steps that no true orbit corrects. Shadowing is a feature of hyperbolic, not merely recurrent, dynamics.
Local maximality is needed for the shadow to lie in $Λ$ and be unique. On a non-isolated hyperbolic set a pseudo-orbit may be shadowed only by an orbit leaving $Λ$ , and uniqueness can fail. The Anosov shadowing theorem asserts existence near $Λ$ and uniqueness within the locally maximal $Λ$ .
Structural stability is conjugacy, not closeness of orbits. A perturbation $g$ need not have orbits $C^{0}$ -close to those of $f$ pointwise in a naive sense; it has a homeomorphism $h$ carrying $f$ -orbits to $g$ -orbits. Two Anosov maps can have wildly different individual trajectories yet be conjugate.

Key theorem with proof Intermediate+

Theorem (Anosov shadowing lemma). Let $Λ$ be a locally maximal hyperbolic set for a $C^{1}$ diffeomorphism $f$ . There are a neighbourhood $V \supseteq Λ$ and $ε_{0} > 0$ such that for every $ε \in (0, ε_{0}]$ there is $δ > 0$ with the property: every $δ$ -pseudo-orbit ${x_{n}}_{n \in Z} \subseteq V$ is $ε$ -shadowed by a point $y$ , and if a pseudo-orbit is shadowed within $ε_{0}$ by two points of the locally maximal set then they coincide — the shadowing orbit in $Λ$ is unique. The same $δ$ -to- $ε$ assignment holds for one-sided and for periodic pseudo-orbits. (Katok-Hasselblatt ^{[Katok-Hasselblatt 1995 §18.1]}; Anosov ^{[Anosov 1967]}; Bowen ^{[Bowen 1975]}.)

Proof (hyperbolic fixed-point / Newton method). Work in adapted exponential charts at the points $x_{n}$ , identifying a neighbourhood of $x_{n}$ with a ball in $T_{x_{n}} M = E_{x_{n}}^{s} \oplus E_{x_{n}}^{u}$ , so that $∥ D f ∣_{E^{s}} ∥ \leq λ$ and $∥ D f^{- 1} ∣_{E^{u}} ∥ \leq λ$ uniformly. Seek the shadowing orbit as $y_{n} = exp_{x_{n}} (v_{n})$ for a bounded sequence of small correction vectors $v_{n} \in T_{x_{n}} M$ ; we want $f (y_{n}) = y_{n + 1}$ , i.e. the corrections satisfy the genuine-orbit equation.

The correction equation. In the charts, the orbit relation $f (y_{n}) = y_{n + 1}$ becomes, after writing $f (x_{n}) = exp_{x_{n + 1}} (ζ_{n})$ with $∥ ζ_{n} ∥ \leq δ$ (the pseudo-orbit error) and linearising, $$ v_{n+1} = A_n v_n + \zeta_n + R_n(v_n), \qquad A_n := Df_{x_n} : T_{x_n}M \to T_{x_{n+1}}M, $$ where $R_{n}$ is the higher-order remainder, $C^{1}$ -small: $∥ R_{n} (v) - R_{n} (v^{'}) ∥ \leq η ∥ v - v^{'} ∥$ on the $ε_{0}$ -ball with $η \to 0$ as $ε_{0} \to 0$ . Collecting the sequence $v = (v_{n})$ in the Banach space $X = ℓ^{\infty} (Z; T M ∣_{{x_{n}}})$ of bounded sections with the sup norm, the equation reads $(Lv)_{n} := v_{n + 1} - A_{n} v_{n} = ζ_{n} + R_{n} (v_{n})$ , i.e. $$ L v = \zeta + R(v). $$

The linear operator $L$ is invertible. $L = S - A$ where $S$ is the shift $(S v)_{n} = v_{n + 1}$ and $A$ is the block cocycle $A_{n}$ . Split $v_{n} = v_{n}^{s} + v_{n}^{u}$ along $E^{s} \oplus E^{u}$ . On the stable block the relation $v_{n + 1}^{s} = A_{n}^{s} v_{n}^{s} + (source)$ is solved forward by the convergent Neumann-type sum $v_{n}^{s} = \sum_{k \leq n} (A_{n - 1}^{s} \dots A_{k}^{s}) (source_{k - 1})$ , convergent because $∥ A^{s} ∥ \leq λ < 1$ . On the unstable block the relation is solved *backward*, $v_{n}^{u} = - \sum_{k > n} (A_{n}^{u})^{- 1} \dots (source)$ , convergent because $∥ (A^{u})^{- 1} ∥ \leq λ$ . The two half-inverses assemble into a bounded inverse with $∥ L^{- 1} ∥ \leq \frac{1}{1 - λ} =: K$ , the discrete Perron summation of 38.03.03 applied to the variational cocycle.

Banach fixed point. The orbit equation $Lv = ζ + R (v)$ is equivalent to the fixed-point equation $$ v = \Phi(v) := L^{-1}\big(\zeta + R(v)\big). $$ For two sections $v, v^{'}$ , $∥Φ (v) - Φ (v^{'}) ∥ \leq K ∥ R (v) - R (v^{'}) ∥ \leq K η ∥ v - v^{'} ∥$ , a contraction once $ε_{0}$ is small enough that $K η < 1$ . The map $Φ$ carries the closed ball of radius $r = 2 K δ$ into itself: $∥Φ (v) ∥ \leq K (∥ ζ ∥ + ∥ R (v) ∥) \leq K (δ + η r) \leq 2 K δ = r$ for $δ, ε_{0}$ small. By the Banach fixed-point theorem there is a unique bounded $v^{*}$ with $∥ v^{*} ∥ \leq 2 K δ$ , giving a genuine orbit $y_{n} = exp_{x_{n}} (v_{n}^{*})$ with $d (f^{n} (y), x_{n}) = ∥ v_{n}^{*} ∥ \leq 2 K δ$ . Choosing $δ \leq ε / (2 K)$ yields $ε$ -shadowing.

Uniqueness on $Λ$ . Suppose $y, y^{'}$ both $ε_{0}$ -shadow the same pseudo-orbit and both orbits lie in $Λ$ . Then $d (f^{n} y, f^{n} y^{'}) \leq 2 ε_{0}$ for all $n \in Z$ , so by expansivity of the hyperbolic set $Λ$ (expansivity constant $> 2 ε_{0}$ ) the two orbits coincide: $y = y^{'}$ . Local maximality guarantees the constructed shadowing orbit, staying in $V$ with $⋂_{n} f^{n} (V) = Λ$ , actually lies in $Λ$ . $□$

Bridge. The shadowing lemma builds toward the entire structural and statistical theory of hyperbolic systems, and the foundational reason it holds is that the variational orbit operator $L = S - A$ is boundedly invertible exactly when the cocycle is hyperbolic — the contracting block inverted forward and the expanding block inverted backward, which is exactly the discrete Perron construction of 38.03.03 applied to errors instead of perturbations. This is the central insight: the same uniform splitting $E^{s} \oplus E^{u}$ that produced the invariant manifolds also makes a Newton step on the whole bi-infinite orbit converge, so an approximate orbit is corrected to a true one in a single bounded inverse. Shadowing generalises the closing-lemma estimate already glimpsed in 38.03.01 from near-return segments to arbitrary pseudo-orbits, and it appears again immediately in the structural-stability theorem, where the pseudo-orbit is a perturbed system's genuine orbit. Putting these together, hyperbolicity converts every kind of small dynamical error — round-off, perturbation, near-recurrence — into a controlled correction, and the bridge is that the orbit-correction operator $(I - A)^{- 1}$ is the single analytic object beneath shadowing, the closing lemma, and structural stability alike.

Exercises Intermediate+

Exercise 2 (easy, multiple choice).

Why does an irrational rotation of the circle fail to have the shadowing property?

A. It has no periodic points. B. It is not continuous. C. It has no hyperbolic splitting, so pseudo-orbit drift accumulates with no contracting/expanding direction to correct it. D. Its orbits are dense.

Hint

Shadowing relies on the orbit-correction operator $(I - A)^{- 1}$ being bounded, which needs $A$ hyperbolic.

Answer

C. A rotation has derivative of modulus one everywhere — no contraction, no expansion — so the operator $L = S - A$ is not boundedly invertible (it has spectrum on the unit circle), and a $δ$ -pseudo-orbit can drift by order $N δ$ over $N$ steps with no true orbit able to track it. Density of orbits (D) and absence of periodic points (A) are true of irrational rotations but are not the reason shadowing fails; the rotation is perfectly continuous (B). Rubric: full credit for C.

Exercise 3 (medium, symbolic).

For the linear hyperbolic model $f (z) = (λ z^{s}, λ^{- 1} z^{u})$ on $R^{s} \oplus R^{u}$ with $0 < λ < 1$ , solve the correction equation $v_{n + 1} = A v_{n} + ζ_{n}$ explicitly (with $A = diag (λ, λ^{- 1})$ ) for a bounded sequence $(v_{n})$ , and show the bound $∥ v ∥_{\infty} \leq \frac{1}{1 - λ} ∥ ζ ∥_{\infty}$ .

Hint

Solve the stable component $v^{s}$ forward and the unstable component $v^{u}$ backward (the only bounded choice), each a geometric series in $λ$ .

Answer

Split $v_{n} = (v_{n}^{s}, v_{n}^{u})$ , $ζ_{n} = (ζ_{n}^{s}, ζ_{n}^{u})$ . Stable. $v_{n + 1}^{s} = λ v_{n}^{s} + ζ_{n}^{s}$ solved forward to stay bounded: $v_{n}^{s} = \sum_{k = 0}^{\infty} λ^{k} ζ_{n - 1 - k}^{s}$ , with $∥ v_{n}^{s} ∥ \leq \sum_{k \geq 0} λ^{k} ∥ ζ ∥_{\infty} = \frac{∥ ζ ∥ _{\infty}}{1 - λ}$ . Unstable. $v_{n + 1}^{u} = λ^{- 1} v_{n}^{u} + ζ_{n}^{u}$ , i.e. $v_{n}^{u} = λ (v_{n + 1}^{u} - ζ_{n}^{u})$ , solved backward: $v_{n}^{u} = - \sum_{k = 0}^{\infty} λ^{k + 1} ζ_{n + k}^{u}$ , with $∥ v_{n}^{u} ∥ \leq \sum_{k \geq 0} λ^{k + 1} ∥ ζ ∥_{\infty} = \frac{λ}{1 - λ} ∥ ζ ∥_{\infty} \leq \frac{∥ ζ ∥ _{\infty}}{1 - λ}$ . Both choices are the unique bounded solutions (any homogeneous term $λ^{n} c^{s}$ or $λ^{- n} c^{u}$ either decays harmlessly or blows up, and boundedness kills the growing one). Taking the max over components and $n$ gives $∥ v ∥_{\infty} \leq \frac{1}{1 - λ} ∥ ζ ∥_{\infty}$ , the operator bound $K = 1/ (1 - λ)$ . Rubric: full credit for the forward/backward split, both geometric series, and the bound.

Exercise 4 (medium, symbolic).

State the Anosov closing lemma as a corollary of shadowing and prove it: if $d (f^{p} (x), x) \leq δ$ for a point $x$ near a locally maximal hyperbolic set, there is a genuine periodic point $z$ of period $p$ with $d (f^{n} x, f^{n} z) \leq ε$ for $0 \leq n \leq p$ , provided $δ \leq δ (ε)$ .

Hint

The segment $x, f (x), \dots, f^{p - 1} (x)$ repeated periodically is a $δ$ -pseudo-orbit. Apply shadowing to the periodic pseudo-orbit; uniqueness forces the shadow to be periodic.

Answer

Form the periodic pseudo-orbit ${x_{n}}_{n \in Z}$ by setting $x_{n} = f^{n mod p} (x)$ and extending $p$ -periodically. The only step where the rule is not followed exactly is the wrap-around $x_{p - 1} = f^{p - 1} (x) \mapsto x_{p} = x$ , where $d (f (x_{p - 1}), x_{p}) = d (f^{p} (x), x) \leq δ$ ; all other steps are exact, so ${x_{n}}$ is a $δ$ -pseudo-orbit. By the shadowing lemma there is $y$ with $d (f^{n} y, x_{n}) \leq ε$ for all $n$ . Now ${x_{n + p}} = {x_{n}}$ as sequences, so $f^{p} (y)$ also $ε$ -shadows the same pseudo-orbit (shift-invariance of the chain). By the uniqueness clause on the locally maximal set, $f^{p} (y) = y$ : the shadow is periodic of period $p$ . Set $z = y$ ; then $d (f^{n} x, f^{n} z) = d (x_{n}, f^{n} z) \leq ε$ for $0 \leq n \leq p$ . Hence a near-return is shadowed by a true periodic orbit. Applying this to any non-wandering point recovers density of periodic points in $Ω (f)$ 38.03.01. Rubric: full credit for the periodic-pseudo-orbit construction, the uniqueness-forces-periodicity step, and the density corollary.

Exercise 5 (medium, short-answer).

Show that a locally maximal hyperbolic set is expansive, and explain why expansivity is exactly the ingredient that makes the shadowing orbit unique.

Hint

Two orbits staying $δ_{0}$ -close for all time put the separation vector in $W_{ε}^{s} \cap W_{ε}^{u} = {pt}$ (from 38.03.03).

Answer

Expansivity. Let $δ_{0}$ be smaller than the size of the local stable/unstable discs. If $x, y \in Λ$ satisfy $d (f^{n} x, f^{n} y) \leq δ_{0}$ for all $n \in Z$ , then forward-closeness gives $y \in W_{δ_{0}}^{s} (x)$ and backward-closeness gives $y \in W_{δ_{0}}^{u} (x)$ (the dynamical characterisation of the local manifolds, 38.03.03). These transverse discs of complementary dimension meet only at $x$ , so $y = x$ ; thus $Λ$ is expansive with constant $δ_{0}$ . Role in uniqueness. If two points $y, y^{'}$ both $ε$ -shadow one pseudo-orbit with $2 ε < δ_{0}$ , then $d (f^{n} y, f^{n} y^{'}) \leq d (f^{n} y, x_{n}) + d (x_{n}, f^{n} y^{'}) \leq 2 ε < δ_{0}$ for all $n$ , so expansivity forces $y = y^{'}$ . Expansivity is therefore precisely the statement that distinct orbits must eventually separate by $δ_{0}$ , which is incompatible with two orbits both hugging the same pseudo-orbit forever. Rubric: full credit for placing $y$ in both local manifolds and the $2 ε < δ_{0}$ uniqueness argument.

Exercise 6 (medium, short-answer).

Sketch how shadowing produces the conjugacy in the structural stability of an Anosov diffeomorphism: given $g$ $C^{1}$ -close to the Anosov map $f$ , define $h$ and show $h \circ f = g \circ h$ .

Hint

A $g$ -orbit is a $δ$ -pseudo-orbit for $f$ when $g$ is $C^{1}$ -close to $f$ . Shadow it by a unique $f$ -orbit; the assignment $x \mapsto$ (start of the shadowing $f$ -orbit) is $h$ .

Answer

Since $f$ is Anosov, $M$ is locally maximal hyperbolic and shadowing holds globally. If $∥ f - g ∥_{C^{1}} \leq δ$ , then for any $x$ the genuine $g$ -orbit ${g^{n} (x)}_{n \in Z}$ satisfies $d (f (g^{n} x), g^{n + 1} x) = d (f (g^{n} x), g (g^{n} x)) \leq δ$ , so it is a $δ$ -pseudo-orbit of $f$ . By the shadowing lemma there is a unique $f$ -orbit $ε$ -shadowing it; let $h (x)$ be the time- $0$ point of that $f$ -orbit, so $d (f^{n} (h (x)), g^{n} (x)) \leq ε$ for all $n$ . Intertwining. The $g$ -orbit of $g (x)$ is the shift of the $g$ -orbit of $x$ , which is $ε$ -shadowed by the shift of the $f$ -orbit of $h (x)$ , i.e. by the $f$ -orbit of $f (h (x))$ . Uniqueness of the shadow gives $h (g (x)) = f (h (x))$ , i.e. $h \circ g = f \circ h$ (and symmetrically with $f, g$ swapped one builds the inverse). Homeomorphism. $h$ is continuous (the shadow depends continuously on the pseudo-orbit, from the fixed-point estimate), injective by expansivity, and surjective with continuous inverse built by shadowing $f$ -orbits with $g$ -orbits. Hence $h$ is a conjugacy and $f, g$ are topologically conjugate. Rubric: full credit for the pseudo-orbit observation, the uniqueness-driven intertwining, and the homeomorphism argument.

Exercise 7 (hard, short-answer).

Prove the converse direction of the expansive + shadowing $\Rightarrow$ topologically stable theorem (Walters): if $f$ is expansive and has the shadowing property, then $f$ is topologically stable — every $C^{0}$ -nearby homeomorphism $g$ admits a continuous surjection $h$ with $f \circ h = h \circ g$ and $h$ close to the identity.

Hint

A $g$ -orbit is a pseudo-orbit for $f$ ; shadow it by an $f$ -orbit; expansivity makes the shadow unique, hence $h$ well-defined and a semiconjugacy.

Answer

Let $ε$ be below half the expansivity constant of $f$ , and choose $δ$ from the shadowing property for this $ε$ . Suppose $g$ is a homeomorphism with $sup_{x} d (f (x), g (x)) \leq δ$ . For each $x$ , the $g$ -orbit ${g^{n} (x)}_{n \in Z}$ satisfies $d (f (g^{n} x), g^{n + 1} x) \leq δ$ , a $δ$ -pseudo-orbit of $f$ . Definition of $h$ . Shadowing gives a point $ε$ -tracking it; expansivity (the shadow distance $ε < \frac{1}{2}$ expansivity constant) makes this point unique, so $h (x) :=$ that point is well-defined, with $d (f^{n} (h (x)), g^{n} (x)) \leq ε$ for all $n$ . Continuity. If $x_{k} \to x$ then $g^{n} (x_{k}) \to g^{n} (x)$ for each $n$ ; the shadow points $h (x_{k})$ stay in a compact $ε$ -neighbourhood and any limit $ε$ -shadows the $g$ -orbit of $x$ , hence equals $h (x)$ by uniqueness — so $h (x_{k}) \to h (x)$ . Semiconjugacy. As in Exercise 6, the $g$ -orbit of $g (x)$ is shadowed both by the $f$ -orbit of $h (g (x))$ and by the $f$ -orbit of $f (h (x))$ ; uniqueness gives $h (g (x)) = f (h (x))$ , i.e. $h \circ g = f \circ h$ . Surjectivity. $h$ is $ε$ -close to the identity ( $d (h (x), x) = d (h (x), g^{0} x) \leq ε$ ), and a continuous map of a compact manifold $ε$ -close to the identity is surjective (degree/homotopy argument). Thus $f$ is topologically stable. This is the abstract skeleton of structural stability: expansivity + shadowing are the two properties of hyperbolic systems that force topological stability, and Anosov maps have both. Rubric: full credit for the well-definedness via expansivity, continuity, the semiconjugacy, and surjectivity.

Exercise 8 (hard, short-answer).

Explain the logical structure of the structural stability theorem (Robbin-Robinson-Mañé): state both directions, attribute each, and say precisely which uses shadowing and which is the hard converse.

Hint

Sufficiency (Axiom A + strong transversality $\Rightarrow$ structural stability) is Robbin ( $C^{2}$ ) and Robinson ( $C^{1}$ ), proved by shadowing-type constructions. Necessity is Mañé's $C^{1}$ stability conjecture.

Answer

Statement. A $C^{1}$ diffeomorphism $f$ of a closed manifold is $C^{1}$ -structurally stable if and only if it satisfies Axiom A and the strong transversality condition. Sufficiency ( $\Leftarrow$ ): Robbin ^{[Robbin 1971]} proved Axiom A + strong transversality $\Rightarrow$ $C^{2}$ -structural stability; Robinson ^{[Robinson 1976]} lowered the regularity to $C^{1}$ . The proof builds the conjugacy by shadowing on each basic set (as in Exercise 6) and then uses strong transversality to glue the pieces across the heteroclinic connections, the no-cycle condition guaranteeing the global homeomorphism is well-defined; this direction is the shadowing/invariant-manifold construction. Necessity ( $\Rightarrow$ ): Mañé ^{[Mañé 1988]} proved the $C^{1}$ stability conjecture — a $C^{1}$ -structurally-stable diffeomorphism must satisfy Axiom A (and strong transversality) — the hard converse, requiring the $C^{1}$ -perturbation technology (Pugh's closing lemma, Mañé's ergodic closing lemma, dominated splittings) to show that any obstruction to hyperbolicity can be destroyed by a $C^{1}$ -small perturbation, contradicting stability. Summary of which is which. The "easy", constructive direction is sufficiency and rests on shadowing; the "hard", destructive direction is necessity and rests on $C^{1}$ -perturbation closing lemmas. Together they characterise $C^{1}$ -structural stability exactly. The analogous $Ω$ -stability theorem characterises $Ω$ -stability by Axiom A + no-cycle (Smale's $Ω$ -stability, Palis-Smale; converse by Palis and Mañé). Rubric: full credit for both directions correctly attributed and for identifying shadowing with sufficiency and the closing-lemma technology with necessity.

Advanced results Master

Theorem (Anosov shadowing theorem, quantitative form). For a locally maximal hyperbolic set $Λ$ there are constants $L > 0$ and $δ_{0} > 0$ , depending only on $(C, λ)$ and the $C^{1}$ geometry, such that every $δ$ -pseudo-orbit with $δ \leq δ_{0}$ in a fixed neighbourhood of $Λ$ is $L δ$ -shadowed by a unique orbit of the locally maximal set; the shadowing point depends Lipschitz-continuously on the pseudo-orbit in the $ℓ^{\infty}$ metric. (Katok-Hasselblatt ^{[Katok-Hasselblatt 1995 §18.1]}.)

The constant $L = 2 K = 2/ (1 - λ)$ (up to chart distortion) is read directly off the Banach inverse $∥ L^{- 1} ∥ \leq K$ of the Key theorem; the linearity $ε = L δ$ is the Newton-method signature, distinguishing genuine hyperbolic shadowing from the merely qualitative shadowing of some non-hyperbolic systems. Lipschitz dependence of the shadow on the chain is the perturbation estimate for fixed points of a uniform contraction, the same mechanism that gives continuous dependence of $W_{loc}^{s} (x)$ on $x$ in 38.03.03.

Theorem (shadowing characterises structural stability of Anosov systems). An Anosov diffeomorphism is structurally stable, and the conjugacy to any $C^{1}$ -small perturbation is the shadowing map $h (x) =$ the basepoint of the unique $f$ -orbit shadowing the $g$ -orbit of $x$ ; $h$ is a Hölder homeomorphism $C^{0}$ -close to the identity. (Anosov ^{[Anosov 1967]}; Katok-Hasselblatt ^{[Katok-Hasselblatt 1995 §18.2]}.)

That the perturbed system's orbits are themselves pseudo-orbits of the original is the whole content: structural stability becomes a corollary of shadowing rather than a separate theorem. The conjugacy is Hölder — generally not Lipschitz and never smooth unless $f, g$ are smoothly conjugate — because it is built from the stable/unstable foliations, which are only Hölder transversally 38.03.03; this is the source of the moduli of stability obstructing $C^{1}$ conjugacy.

Theorem ( $Ω$ -stability and the no-cycle condition). An Axiom-A diffeomorphism whose basic sets satisfy the no-cycle condition is $Ω$ -stable: for every $C^{1}$ -small $g$ there is a homeomorphism $Ω (f) \to Ω (g)$ conjugating $f ∣_{Ω (f)}$ to $g ∣_{Ω (g)}$ . The no-cycle condition is necessary. (Smale 1970; Palis 1970; Katok-Hasselblatt ^{[Katok-Hasselblatt 1995 §18.5]}.)

Shadowing on each basic set provides the conjugacy piecewise; the no-cycle (acyclicity of the phase diagram of basic sets, 38.03.01) is what lets the pieces be ordered and assembled without the heteroclinic connections forcing an inconsistency. Cycles support perturbations that create or destroy connecting orbits, breaking $Ω$ -conjugacy — the mechanism behind the examples separating $Ω$ -stability from structural stability.

Theorem (structural stability theorem: Robbin-Robinson-Mañé). A $C^{1}$ diffeomorphism of a closed manifold is $C^{1}$ -structurally stable if and only if it satisfies Axiom A and the strong transversality condition. (Robbin ^{[Robbin 1971]} $C^{2}$ ; Robinson ^{[Robinson 1976]} $C^{1}$ sufficiency; Mañé ^{[Mañé 1988]} $C^{1}$ necessity.)

Sufficiency is the shadowing/invariant-manifold construction extended across basic sets by strong transversality; necessity — the $C^{1}$ stability conjecture — is the deep converse, requiring Pugh's closing lemma and Mañé's dominated-splitting and ergodic-closing technology to show any failure of hyperbolicity is destroyed by a $C^{1}$ -small perturbation. The theorem closes Smale's programme: the structurally stable systems are exactly the uniformly hyperbolic ones with transverse invariant manifolds, no more and no less.

Synthesis. Shadowing is the organising mechanism, and the foundational reason the entire structural theory works is that a single bounded inverse $(I - A)^{- 1}$ of the variational cocycle corrects every species of small error at once: round-off into a true orbit (shadowing), near-return into a periodic orbit (the closing lemma), and a perturbed orbit into an original-system orbit (structural stability). This is exactly the discrete-time Perron forward/backward summation of 38.03.03 applied to orbit corrections rather than to manifold sections, and it is dual to the invariant-manifold construction — there one slaves the unstable coordinate to the stable one along a graph, here one inverts the same hyperbolic splitting along a whole bi-infinite sequence. The central insight is that hyperbolicity makes orbits rigid under error, so the spectral decomposition of 38.03.01 (which the closing lemma populated with dense periodic points) becomes structurally stable, and putting these together the Robbin-Robinson-Mañé theorem identifies structural stability with Axiom A plus transversality exactly. The bridge is that shadowing, expansivity, and the no-cycle condition are the three abstract distillates of uniform hyperbolicity, and structural stability is precisely their conjunction — the summit toward which 38.03.01 and 38.03.03 were always building.

Full proof set Master

Proposition (the shadowing constant is linear: $ε \leq L δ$ ). In the Key theorem's construction, the shadowing distance satisfies $|v^|\infty \leq \frac{|\zeta|\infty}{1 - \lambda - K\eta} $, w hi c h f or$ \eta \to 0 $g i v es$ \varepsilon \leq L\delta $w i t h$ L = 1/(1-\lambda)$ up to chart distortion.*

Proof. The fixed point $v^{*}$ satisfies $v^{*} = L^{- 1} (ζ + R (v^{*}))$ , so $∥ v^{*} ∥ \leq K ∥ ζ ∥ + K η ∥ v^{*} ∥$ , whence $(1 - K η) ∥ v^{*} ∥ \leq K ∥ ζ ∥$ and $∥ v^{*} ∥ \leq \frac{K}{1 - K η} ∥ ζ ∥ \leq \frac{K}{1 - K η} δ$ . As $ε_{0} \to 0$ the nonlinear Lipschitz constant $η \to 0$ , so the prefactor tends to $K = 1/ (1 - λ)$ , and translating from chart coordinates to Riemannian distance multiplies by the bounded chart-distortion constant, absorbed into $L$ . Hence $ε \leq L δ$ with $L$ depending only on $(C, λ)$ and the geometry. $□$

Proposition (uniqueness of the shadowing orbit on $Λ$ ). If $y, y^{'}$ each $ε$ -shadow the same pseudo-orbit ${x_{n}} \subseteq Λ$ with $2 ε$ below the expansivity constant $δ_{0}$ of $Λ$ , then $y = y^{'}$ .

Proof. For all $n \in Z$ , $d (f^{n} y, f^{n} y^{'}) \leq d (f^{n} y, x_{n}) + d (x_{n}, f^{n} y^{'}) \leq ε + ε = 2 ε < δ_{0}$ . A hyperbolic set is expansive with constant $δ_{0}$ (Exercise 5): any two orbits remaining $δ_{0}$ -close for all integer times coincide. Hence $y = y^{'}$ . The local maximality of $Λ$ ensures the shadowing orbit produced by the fixed point stays in $Λ$ , so the comparison takes place within the expansive set. $□$

Proposition (closing lemma from shadowing). Density of periodic points in $Ω (f)$ for an Axiom-A system follows from the periodic case of the shadowing lemma.

Proof. Let $x \in Ω (f)$ and $U$ a neighbourhood. Non-wandering gives $w$ near $x$ and $p \geq 1$ with $f^{p} (w)$ near $w$ , so $d (f^{p} (w), w) \leq δ$ . The $p$ -periodic extension of the segment $w, f (w), \dots, f^{p - 1} (w)$ is a $δ$ -pseudo-orbit (only the wrap step slips, by $\leq δ$ ). The shadowing lemma yields a shadow $y$ ; by the shift-invariance of a periodic chain and uniqueness on the locally maximal $Ω (f)$ , $f^{p} (y) = y$ , so $y$ is a genuine periodic point with $d (x, y) \leq d (x, w) + d (w, y) \leq d (x, w) + ε$ , inside $U$ for $δ, ε$ small. Hence $Per (f) \cap Ω (f)$ is dense in $Ω (f)$ . $□$

Proposition (the shadowing map is a conjugacy). For an Anosov $f$ and $C^{1}$ -close $g$ , the shadowing map $h$ of Exercise 6 is a homeomorphism with $h \circ g = f \circ h$ .

Proof. Well-defined and intertwining are Exercise 6. Continuity of $h$ . The shadow is the fixed point $v^{*} (ζ)$ of $Φ_{ζ} (v) = L^{- 1} (ζ + R (v))$ ; by the fixed-point perturbation estimate $∥ v^{*} (ζ) - v^{*} (ζ^{'}) ∥ \leq \frac{K}{1 - K η} ∥ ζ - ζ^{'} ∥$ , and the chain $ζ$ associated to $x$ depends continuously on $x$ (it is built from $g^{n} (x)$ ), so $h$ is continuous. Injectivity. If $h (x) = h (x^{'})$ then the $g$ -orbits of $x$ and $x^{'}$ are both $ε$ -shadowed by the same $f$ -orbit, so $d (g^{n} x, g^{n} x^{'}) \leq 2 ε$ for all $n$ ; since $g$ is also Anosov (hyperbolicity is $C^{1}$ -open, 38.03.01) and hence expansive with a constant $> 2 ε$ , $x = x^{'}$ . Surjectivity and continuous inverse. Reversing the roles of $f$ and $g$ — shadowing $f$ -orbits by $g$ -orbits — builds $h^{'}$ with $h^{'} \circ f = g \circ h^{'}$ ; uniqueness of shadows gives $h^{'} \circ h = id$ and $h \circ h^{'} = id$ , so $h$ is a homeomorphism. Hence $f$ and $g$ are topologically conjugate. $□$

Proposition (Hölder regularity of the conjugacy). The conjugacy $h$ between an Anosov $f$ and a $C^{1}$ -close $g$ is Hölder continuous.

Proof. $h$ carries $W_{f}^{s} (x)$ to $W_{g}^{s} (h (x))$ and $W_{f}^{u} (x)$ to $W_{g}^{u} (h (x))$ , because a forward-asymptotic pair for $g$ shadows a forward-asymptotic pair for $f$ . Along a single stable leaf the conjugacy is built from the leafwise holonomy, which contracts at the uniform rate $λ$ , while transverse to the leaves the dependence is controlled by the angle between $E^{s}$ and $E^{u}$ ; the standard estimate comparing the two exponential rates $λ^{s}, λ^{u}$ gives $d (h (x), h (x^{'})) \leq C d (x, x^{'})^{α}$ with Hölder exponent $α = lo g λ_{u}^{- 1} / lo g λ_{s}^{- 1}$ (the ratio of contraction to expansion rates). Smoothness fails in general because this exponent is typically $< 1$ and is a conjugacy invariant — the modulus of stability. $□$

Connections Master

Hyperbolic sets, Anosov and Axiom-A systems 38.03.01. This unit supplies the proofs the spectral-decomposition unit used as black boxes: the Anosov closing lemma that gives density of periodic points (Axiom A clause (ii)), the shadowing that makes basic sets topologically transitive, and the expansivity underlying isolation. Where 38.03.01 organised $Ω (f)$ into basic sets, this unit shows that organisation is robust — the no-cycle condition makes it $Ω$ -stable and strong transversality makes the whole system structurally stable.
The Hadamard-Perron stable and unstable manifold theorem 38.03.03. The orbit-correction operator $(I - A)^{- 1}$ proven invertible here is the same hyperbolic-splitting fixed point that built the invariant manifolds there: Perron's forward/backward summation of 38.03.03 is applied in this unit to orbit errors instead of to graph sections. Shadowing and the manifold theorem are dual readings of one estimate, and the structural-stability conjugacy can be built either way — by shadowing or by intersecting perturbed manifolds.
Lyapunov stability, direct method 02.12.08. Structural stability is the global, system-level analogue of the local equilibrium stability certified by a Lyapunov function: a Hurwitz linearisation makes one equilibrium robust, while a uniform hyperbolic splitting plus transversality makes the entire phase portrait robust up to conjugacy. The adapted metric that realises $C = 1$ in the shadowing estimate is again a quadratic Lyapunov form for the linearised cocycle.
Symbolic dynamics / subshifts of finite type 38.02.01. Shadowing is the mechanism Bowen used to build Markov partitions: a finite cover by rectangles produces admissible symbol sequences, and shadowing turns every admissible sequence into a genuine orbit, giving the semiconjugacy from a subshift of finite type onto a basic set. The specification property — orbit segments can be approximated and concatenated — is the strengthening of shadowing that drives the thermodynamic formalism.

Historical & philosophical context Master

The shadowing estimate originates with Dmitri Anosov, whose 1967 Steklov monograph on geodesic flows of negative curvature ^{[Anosov 1967]} proved that uniformly hyperbolic systems are structurally stable and, in the course of that proof, established the orbit-correction property now called shadowing. Rufus Bowen extracted shadowing as a named lemma in his 1975 Springer Lecture Notes ^{[Bowen 1975]}, where it became the tool for constructing Markov partitions and the symbolic coding of basic sets, the technical heart of the Sinai-Ruelle-Bowen thermodynamic formalism.

The structural-stability theorem was assembled over two decades. Joel Robbin proved in 1971 ^{[Robbin 1971]} that Axiom A together with strong transversality implies $C^{2}$ -structural stability; Clark Robinson lowered the regularity to $C^{1}$ in 1976 ^{[Robinson 1976]}, completing the sufficiency direction along the lines Smale had conjectured. The converse — the $C^{1}$ stability conjecture, that structural stability forces Axiom A and strong transversality — resisted until Ricardo Mañé's 1988 proof in the Publications de l'IHÉS ^{[Mañé 1988]}, which deployed dominated splittings and an ergodic closing lemma to destroy any non-hyperbolic behaviour by a $C^{1}$ -small perturbation. Katok and Hasselblatt's 1995 treatise ^{[Katok-Hasselblatt 1995]} gives the canonical modern account of shadowing and the structural-stability theorem, and Pilyugin's 1999 monograph surveys the shadowing theory as a subject in its own right.

Bibliography Master

@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{Anosov1967,
  author  = {Anosov, Dmitri V.},
  title   = {Geodesic flows on closed {R}iemannian manifolds of negative curvature},
  journal = {Proceedings of the Steklov Institute of Mathematics},
  volume  = {90},
  year    = {1967}
}

@book{Bowen1975,
  author    = {Bowen, Rufus},
  title     = {Equilibrium States and the Ergodic Theory of {A}nosov Diffeomorphisms},
  publisher = {Springer-Verlag},
  series    = {Lecture Notes in Mathematics},
  volume    = {470},
  year      = {1975}
}

@article{Robbin1971,
  author  = {Robbin, Joel W.},
  title   = {A structural stability theorem},
  journal = {Annals of Mathematics},
  volume  = {94},
  year    = {1971},
  pages   = {447--493}
}

@article{Robinson1976,
  author  = {Robinson, Clark},
  title   = {Structural stability of {$C^1$} diffeomorphisms},
  journal = {Journal of Differential Equations},
  volume  = {22},
  year    = {1976},
  pages   = {28--73}
}

@article{Mane1988,
  author  = {Ma{\~n}{\'e}, Ricardo},
  title   = {A proof of the {$C^1$} stability conjecture},
  journal = {Publications Math\'ematiques de l'IH\'ES},
  volume  = {66},
  year    = {1988},
  pages   = {161--210}
}

@book{Pilyugin1999,
  author    = {Pilyugin, Sergei Yu.},
  title     = {Shadowing in Dynamical Systems},
  publisher = {Springer-Verlag},
  series    = {Lecture Notes in Mathematics},
  volume    = {1706},
  year      = {1999}
}

@book{Robinson1999,
  author    = {Robinson, Clark},
  title     = {Dynamical Systems: Stability, Symbolic Dynamics, and Chaos},
  publisher = {CRC Press},
  edition   = {2},
  year      = {1999}
}

Prerequisites

38.03.01
38.03.03

Tier anchors

beginner: Strogatz 1994 *Nonlinear Dynamics and Chaos* (Addison-Wesley) Ch. 9–12 (sensitive dependence, numerical orbits of chaotic maps); 3Blue1Brown *Chaos* and the Lorenz-attractor visual essays for the round-off / true-orbit imagery
intermediate: Brin-Stuck 2002 *Introduction to Dynamical Systems* (Cambridge University Press) §5.3–§5.5 (the shadowing lemma and structural stability of Anosov diffeomorphisms); Robinson 1999 *Dynamical Systems: Stability, Symbolic Dynamics, and Chaos* (CRC) Ch. 5 §5.7, §8.2
master: Katok-Hasselblatt 1995 *Introduction to the Modern Theory of Dynamical Systems* (Cambridge) Ch. 18 §18.1 (the Anosov shadowing theorem and closing lemma), §18.2 (structural stability of Anosov systems via shadowing), §18.5 ($\Omega$-stability and the structural-stability theorem); Pilyugin 1999 *Shadowing in Dynamical Systems* (Springer LNM 1706); Mañé 1988 *A proof of the $C^1$ stability conjecture* (Publ. IHÉS 66)

References

Katok, Hasselblatt 1995 — Introduction to the Modern Theory of Dynamical Systems · Ch. 18 §18.1 (the Anosov shadowing theorem: every $\delta$-pseudo-orbit in a neighbourhood of a hyperbolic set is $\varepsilon$-shadowed by a true orbit, uniquely on the locally maximal set; the closing lemma), §18.2 (structural stability of Anosov diffeomorphisms, the conjugacy from shadowing), §18.5 ($\Omega$-stability, the no-cycle condition); Theorem 18.1.2, Theorem 18.1.3, Theorem 18.2.3; Encyclopedia of Mathematics and its Applications 54, Cambridge University Press 1995
Anosov 1967 — Geodesic flows on closed Riemannian manifolds of negative curvature · Proceedings of the Steklov Institute of Mathematics 90 (1967); the originating shadowing (specification-of-orbits) estimate for uniformly hyperbolic systems and the structural stability of Anosov flows and diffeomorphisms
Bowen 1975 — Equilibrium States and the Ergodic Theory of Anosov Diffeomorphisms · Springer Lecture Notes in Mathematics 470; the shadowing lemma stated and used to build Markov partitions, the specification property, and the symbolic coding of basic sets
Robbin 1971 — A structural stability theorem · Annals of Mathematics 94 (1971) 447–493; the first proof that Axiom A plus strong transversality implies $C^2$ structural stability
Robinson 1976 — Structural stability of C^1 diffeomorphisms · Journal of Differential Equations 22 (1976) 28–73; the $C^1$ sufficiency: Axiom A and strong transversality imply $C^1$ structural stability
Mañé 1988 — A proof of the C^1 stability conjecture · Publications Mathématiques de l'IHÉS 66 (1988) 161–210; the converse: a $C^1$-structurally-stable diffeomorphism satisfies Axiom A and strong transversality, completing the structural-stability theorem

Estimated time

beginner: 18m
intermediate: 45m
master: 88m