Shadowing Lemma in Dynamical Systems

• Shadowing lemma is a principle in dynamical systems that guarantees any approximate pseudo-orbit is closely tracked by an actual orbit under hyperbolic conditions.
• It extends to nonuniform, partial, infinite-dimensional, and group action contexts, providing rigorous frameworks for numerical validation and orbit determination.
• These methods underpin error estimates and computational strategies in chaotic dynamics, leveraging contraction mappings and explicit hyperbolicity constants.

The shadowing lemma is a fundamental result in the theory of dynamical systems, asserting that for broad classes of systems possessing some form of hyperbolicity, every approximate trajectory (“pseudo-orbit”) can be closely followed (“shadowed”) by a true orbit. This property bridges the gap between numerical or perturbed dynamical behavior and actual orbits, with far-reaching consequences for structural stability, orbit determination, and the rigorous analysis of chaos. The scope of the shadowing concept has significantly expanded over time, encompassing uniform and nonuniform hyperbolicity, partial and normal hyperbolicity, actions of groups, infinite-dimensional dynamics, and nonstandard contexts such as systems with grow-up or unbounded domains.

1. Classical Shadowing in Uniformly Hyperbolic Systems

In the setting of a compact manifold $M$ and a $C^r$ diffeomorphism $f : M \to M$, consider a compact $f$-invariant hyperbolic set $\Lambda \subset M$ characterized by a continuous splitting $T_x M = E^s_x \oplus E^u_x$ and uniform contraction/expansion constants: $$\|Df^n|_{E^s}\| \leq \lambda^n, \;\;\|Df^{-n}|_{E^u}\| \leq \lambda^n, \qquad \lambda \in (0,1), \; n\geq 0.$$ A finite or bi-infinite sequence $\{x_i\}$ is a $\delta$-pseudo-orbit if $d(f(x_i), x_{i+1}) < \delta$ for all $C^r$0. The classical shadowing lemma [Anosov–Bowen mechanism, (Thiam, 19 Apr 2026, Spoto et al., 2015)] asserts:

Shadowing Lemma: For each $C^r$1, there exists $C^r$2 such that any $C^r$3-pseudo-orbit $C^r$4 is $C^r$5-shadowed by a true orbit $C^r$6, i.e., $C^r$7 for all $C^r$8.

All estimates are explicit in terms of the contraction rate $C^r$9, the second-derivative bound, and injectivity radius, with the optimal linear tracking constant $f : M \to M$0 and error bounds $f : M \to M$1 (Thiam, 19 Apr 2026). Proof techniques utilize the “local product structure”, the bracket map, and fixed-point iterations for corrections in the stable/unstable directions.

2. Generalizations: Nonuniform, Partial, and Infinite-Dimensional Shadowing

2.1 Nonuniform and Generalized Shadowing

Mather’s characterization equates uniform hyperbolicity to invertibility of a sequence-space difference operator $f : M \to M$2 on weighted Banach spaces $f : M \to M$3; nonuniform hyperbolicity (e.g., all Lyapunov exponents nonzero almost everywhere) is equivalent to invertibility on the Fréchet space $f : M \to M$4 of subexponentially growing sequences (Dragicevic et al., 2011). Dragičević–Slijepčević extend the shadowing lemma as follows:

Generalized Shadowing Lemma: If $f : M \to M$5 is bijective with continuous inverse for every $f : M \to M$6 in an $f : M \to M$7-invariant set $f : M \to M$8, then $f : M \to M$9 is shadowable: for every $f$0 sufficiently small, every $f$1-pseudo-orbit in an appropriate neighborhood is $f$2-shadowed by a true orbit (Dragicevic et al., 2011).

This framework admits shadowing for sets exhibiting only pointwise invertibility of $f$3 (nonuniform hyperbolicity in measure), without uniform estimates, and can be adapted to regions failing standard hyperbolicity, such as certain twist maps and Hamiltonian systems. The proof is based on global approximate inverses in sequence space and a contraction mapping argument.

2.2 Partial and Central Shadowing

For partially hyperbolic, dynamically coherent diffeomorphisms (splitting $f$4), the central shadowing lemma asserts that pseudo-orbits can be shadowed by “central pseudo-trajectories” with jumps along the center foliation $f$5, with transverse and longitudinal errors both controlled by the pseudo-orbit size (Kryzhevich et al., 2011). This generalizes the classical lemma to cases where shadowing by actual orbits fails due to non-hyperbolic center behavior.

2.3 Shadowing in Infinite Dimensions

For Morse–Smale semigroups on separable Hilbert spaces (infinitely many degrees of freedom), shadowing still holds provided hyperbolicity and transversality conditions are met. Specifically (Arrieta et al., 12 Feb 2025):

• Lipschitz shadowing: On the global attractor, any $f$6-pseudo-orbit is $f$7-shadowed by a true orbit.
• Hölder shadowing: In an invariant neighborhood, the shadowing error is $f$8, with $f$9.

The proof relies on the uniform spectral gap of equilibria, reduction to non-autonomous cocycles, and contraction mapping in Banach (or weighted Banach) spaces.

2.4 Nonuniform Shadowing and Weighted Norms

For systems with unbounded phase space (“grow-up”), nonuniform shadowing properties are formulated where both the pseudo-orbit errors and required shadowing proximity depend on the boundary distance $\Lambda \subset M$0; in the original, non-compact domain, exponent shifts in ball geometry under Poincaré compactification must be accounted for (Osipov, 2014). Weighted shadowing replaces uniform proximity with control over a weighted functional; this is essential for flows where trajectories may escape to infinity.

3. Shadowing in Group Actions and Nonstandard Contexts

The development of shadowing for actions of finitely generated groups—for example, nilpotent, solvable, or free groups—shows that the existence of a single topologically Anosov (hyperbolic) element in a virtually nilpotent group promotes the shadowing property to the entire group action in the same region (Osipov et al., 2013). Expansivity and shadowing become strictly related: for surjective expansive maps, classical, two-sided, and limit shadowing all coincide, and the presence of positive expansivity ensures uniqueness of shadows (Good et al., 2020).

Nonabelian free groups and certain solvable groups (e.g., the Baumslag–Solitar group with parameter at the hyperbolic/elliptic threshold) are shown to lack shadowing, illustrating the nontrivial dependence of shadowing on group-theoretic and dynamical hyperbolicity properties.

4. Shadowing for Normally Hyperbolic Invariant Manifolds

For diffeomorphisms or flows possessing a normally hyperbolic invariant manifold (NHIM) $\Lambda \subset M$1 with transverse intersections of their stable and unstable manifolds, the shadowing lemma is formulated in terms of correct alignments of “windows” (local coordinate blocks) under the transition and inner dynamics. The sufficient conditions for shadowing an itinerary through a sequence of prescribed windows are entirely topological (correct alignment by the method of Conley or correctly aligned windows), and transcend geometric details (Delshams et al., 2012). This allows construction of orbits with controlled heteroclinic behavior, such as Arnold diffusion in near-integrable Hamiltonian systems.

5. Quantitative and Structural Implications

5.1 Regularity and Explicit Constants

In uniformly hyperbolic contexts, all shadowing bounds can be made explicit in terms of rates $\Lambda \subset M$2, derivative bounds, and geometric data (Thiam, 19 Apr 2026). Linear maps and the optimality of constants are established directly by geometric series and contraction properties.

5.2 Hölder and Finite-Interval Shadowing

The existence of Hölder shadowing estimates, such as $\Lambda \subset M$3-approximation up to timescales $\Lambda \subset M$4, implies structural stability when $\Lambda \subset M$5, $\Lambda \subset M$6 (Tikhomirov, 2011). Failure of these inequalities (as in conjectures of Hammel–Grebogi–Yorke) is incompatible with non-hyperbolic, non-structurally-stable dynamics.

5.3 Orbit Determination and Computational Horizons

The shadowing lemma provides the rigorous basis for Chaotic Orbit Determination: given a noisy pseudo-orbit, least squares fitting yields a shadowing orbit if the system is hyperbolic. However, numerical computation is limited by the rapid growth of condition numbers in the variational equations (the “computability horizon”), which in turn depends logarithmically on the inverse of roundoff and Lyapunov exponent (Spoto et al., 2015).

The asymptotic decrease of uncertainty is exponentially fast in the number of observations for hyperbolic orbits when only initial conditions are estimated; when dynamical parameters are included, only polynomial rate improvement is observed, contrary to naive expectations.

6. Shadowing for Linear, Infinite-Dimensional Semigroups

For the linear abstract Cauchy problem $\Lambda \subset M$7 on a Banach space, hyperbolicity (spectrum disjoint from the unit circle) is necessary and sufficient for the $\Lambda \subset M$8-shadowing property: every pseudo-orbit (piecewise-trajectories with small jumps) is shadowed by an exact solution with the discrepancy decaying to zero in norm (Lee et al., 2024). In finite-dimensional spaces, the converse holds; on the circle, pure rotations provide a counterexample. The results extend discrete-time theorems for linear homeomorphisms (Ombach’s theorem) to continuous-time and infinite dimensions.

7. Summary Table: Main Variants of Shadowing Lemmas

Context	Shadowing Property	Reference
Uniform hyperbolicity (compact $\Lambda \subset M$9)	Linear shadowing, explicit constants	(Thiam, 19 Apr 2026)
Nonuniform hyperbolicity (invertible $T_x M = E^s_x \oplus E^u_x$0)	Generalized shadowing, contraction in $T_x M = E^s_x \oplus E^u_x$1	(Dragicevic et al., 2011)
Partial hyperbolicity (center foliation $T_x M = E^s_x \oplus E^u_x$2)	Central shadowing (up to center leaves)	(Kryzhevich et al., 2011)
Infinite-dimensional Morse–Smale	Lipschitz/Hölder shadowing on attractor/neighborhood	(Arrieta et al., 12 Feb 2025)
Normally hyperbolic manifolds (NHIM)	Shadowing via aligned windows in ambient manifold	(Delshams et al., 2012)
Banach/linear semigroups	$T_x M = E^s_x \oplus E^u_x$3-shadowing iff hyperbolic generator	(Lee et al., 2024)
Group actions (nilpotent, solvable, free)	Shadowing ⇔ hyperbolic generators, obstructions	(Osipov et al., 2013)
Weighted/nonuniform (grow-up, compactification)	Variable, boundary-dependent shadowing	(Osipov, 2014)

Each context employs a tailored functional-analytic or geometric setup, often involving operators on sequence spaces, Banach fixed-point theorems, and explicit contraction mapping arguments.

---

The shadowing lemma and its extensions provide the main functional-analytic toolkit for transferring properties between numerical, experimental, or approximate trajectories and bona fide solutions of complex dynamical systems, under assumptions that precisely capture hyperbolicity, its nonuniform variants, or extensions to more general spaces and actions.