---
title: Shadowing in Dynamical Systems
url: https://www.emergentmind.com/topics/shadowing
type: topic
---

# Shadowing in Dynamical Systems

Shadowing

Shadowing is a foundational concept in the theory of dynamical systems, describing the phenomenon whereby approximate trajectories (pseudo-orbits) generated by numerical or perturbative processes can be closely tracked, or "shadowed," by exact trajectories of the system. Shadowing is central to understanding the reliability of numerical simulations, the structural stability of dynamical systems, the topology of attractors, and various forms of recurrence and mixing. Its many generalizations—limit shadowing, average shadowing, s-limit shadowing, inverse shadowing, and adaptations to non-autonomous, set-valued, or geometric contexts—provide tools for the analysis of both mathematical models and physical systems across dimensions and disciplines.

## 1. Classical Shadowing: Definitions and Mechanisms

Let $(X, d)$ be a compact metric space and $f:X \to X$ a continuous map. A $\delta$-pseudo-orbit is a sequence $(x_i)_{i \ge 0}$ with $d(f(x_i), x_{i+1}) < \delta$ for all $i$. The system $(X,f)$ has the (classical) shadowing property if for every $\varepsilon > 0$, there exists $\delta > 0$ so that any $\delta$-pseudo-orbit $(x_i)$ is $\varepsilon$-traced by some $y \in X$: $d(x_i, f^i(y)) < \varepsilon$ for all $i$ [1710.00313, 1701.05383, 1810.02262, 2504.00171].

The concept arises naturally in the numerical simulation of dynamical systems, where modeling or computational errors generate pseudo-orbits rather than true orbits. The shadowing property ensures that these pseudo-orbits can be interpreted as genuine dynamical behaviors up to controlled error.

A dynamic system with the shadowing property typically possesses hyperbolicity or other structures that ensure errors do not amplify in a way that prevents accurate retracing by a true trajectory. The equivalence of one-sided, two-sided, and backward shadowing in the presence of surjectivity and expansivity [2002.11199] is central for the topological characterization of systems with shadowing.

## 2. Generalizations: Limit, s-Limit, Average, and Inverse Shadowing

Beyond classical shadowing, several generalizations are crucial:

- **Limit Shadowing:** A (one-sided) sequence $(x_i)$ is a limit pseudo-orbit if $d(f(x_i), x_{i+1}) \to 0$ as $i \to \infty$. The system has the limit shadowing property if every limit pseudo-orbit is traced asymptotically: $d(x_i, f^i(y)) \to 0$ for some $y$ [1710.00313, 1701.05383]. Limit shadowing is generally weaker than classical shadowing, and neither property implies the other in full generality.

- **s-Limit Shadowing:** A system has s-limit shadowing if, for every $\varepsilon > 0$, there exists $\delta > 0$ such that all $\delta$-pseudo-orbits are $\varepsilon$-shadowed, and moreover, every asymptotic $\delta$-pseudo-orbit is asymptotically shadowed by the same point [1701.05383, 2111.09252].

- **Average Shadowing:** Average shadowing replaces uniform error bounds with demands on average error: a $\delta$-average pseudo-orbit satisfies $\frac{1}{n} \sum_{i=0}^{n-1} d(f(x_i), x_{i+1}) < \delta$ for large $n$; average shadowing requires the existence of $z$ so that $\limsup_{n \to \infty} \frac{1}{n} \sum_{i=0}^{n-1} d(f^i(z), x_i) < \varepsilon$ [1406.5822].

- **Inverse Shadowing:** Here, the role of exact and approximate trajectories is reversed. Given an exact orbit, the question is whether it can be closely approximated by trajectories of a nearby perturbed system or "method." Weaker forms such as ergodic inverse shadowing (EIS) and individual inverse shadowing (IIS) play a role in measure-theoretic and stability considerations [1907.06792].

- **h-Shadowing:** Requires that for each finite pseudo-orbit, a true orbit passes exactly through its final point (important in hyperbolic and symbolic dynamics) [2111.09252, 1907.02446].

These notions are often arranged in an implication hierarchy (ASP $\implies$ AASP $\implies$ average shadowing $\implies$ $d$-shadowing) [1406.5822, 2111.09252].

## 3. Shadowing and System Structure: Expansivity, Stability, and Genericity

A strong connection exists between shadowing and structural properties:

- **Expansivity:** Expansivity (existence of $r > 0$ such that distinct points eventually separate by at least $r$ under iteration) is intimately related to shadowing. For expansive homeomorphisms, shadowing, two-sided shadowing, s-limit shadowing, and unique shadowing are equivalent [2002.11199, 2504.00171]. Walters' theorem asserts that positively expansive maps with shadowing are topologically stable, with unique shadowing of pseudo-orbits.

- **Genericity:** In the space of continuous maps on a locally connected one-dimensional continuum ("graphite") or its subspace of surjections, the set of maps with the shadowing property is residual (i.e., generic in the sense of Baire category) [1810.02262]. This includes graphs, dendrites, Menger curves, and such Julia sets. It is an open question whether there exist locally connected one-dimensional continua where shadowing is non-generic.

- **Nonautonomous and Time-Varying Systems:** Shadowing extends to sequences of maps $\{f_n\}$ (nonautonomous dynamics). Contractive or expansive sequences yield (limit, s-limit, exponential) shadowing; these properties are inherited by products and iterates [2111.09252, 2403.15777]. In nonautonomous systems, under suitable contraction (e.g., $\prod_{i=0}^\infty\lambda_i = 0$), every pseudo-orbit is uniquely shadowed [2403.15777].

- **Stability and Structural Stability:** Shadowing is a necessary (and in some settings sufficient) condition for structural stability. Ergodic inverse shadowing implies weak continuity of invariant measures, and IIS characterizes $C^1$-structurally stable diffeomorphisms [1907.06792].

## 4. Shadowing in Extended and Specialized Contexts

Shadowing theory is robust under many standard constructions, but with notable exceptions and subtleties:

- **Hyperspaces and Symmetric Products:** The shadowing property is preserved under passage to the hyperspace $2^X$ (Hausdorff topology), but not in general to the space of subcontinua $\mathcal{C}(X)$; for instance, Anosov diffeomorphisms induce maps $C(f)$ on continua that fail to have shadowing, in contrast to dendrite monotone maps where shadowing equivalence holds [1907.02446, 2408.12688].

- **Multiple Maps and Set-Valued Systems:** For iterated function systems and multiple maps, shadowing and average shadowing are defined in the Hausdorff metric on nonempty compact subset space $\mathcal{K}(X)$. Their invariance properties and transitive behaviors diverge significantly from the single-map case [2311.02932].

- **Geometric and Physical Models:** Shadowing arises in physical applications. For instance, the shadowing function in geometric-optics treatments of rough-surface scattering quantifies the proportion of facets mutually visible to both light source and detector, directly entering the BRDF [2010.03927]. The effect of log-normal "shadowing" on signal propagation (distinct from dynamical systems shadowing) is central to physical-layer modeling in wireless networks, with direct statistical implications for packet delivery and protocol design [1004.0422].

- **Differential Equations with Grow-Up:** For noncompact systems, nonuniform and weighted shadowing analogues can be established via compactification and decompactification techniques (e.g., using Poincaré compactification) [1409.6980].

## 5. Quantitative and Algorithmic Aspects

Recent work addresses quantitative aspects and algorithms:

- **Shadowing Times and Data Assimilation:** In high-dimensional chaotic systems, shadowing times quantify how long model trajectories can remain statistically consistent with a given set of observations, providing a metric for model reliability [1909.04488]. The typical shadowing time decreases logarithmically with distance from the true trajectory in chaotic regimes.

- **Nonintrusive Shadowing Algorithms:** For sensitivity analysis in chaotic systems, nonintrusive shadowing techniques (e.g., NILSS) approximate derivatives of statistical quantities under parameter perturbation by computing so-called shadowing directions, rigorously capturing only the "shadowing contribution" to linear response. The residual "unstable" contribution, often small in strongly dissipative systems, must be corrected for in general [2003.09801].

- **Shadowing Maps and Hierarchies:** The concept of a shadowing map—providing a continuous, possibly equivariant section from pseudo-orbits to shadows—enables categorization of systems into a hierarchy based on the strength and invariance properties of the shadowing (e.g., shift-invariant, self-tuning, bracket-based structures). Uniqueness and further invariance reflect higher forms of hyperbolicity [2504.00171].

## 6. Entropy, Flexibility, and Local Shadowing

Shadowing imparts significant constraints on entropy and measure properties:

- **Entropy Flexibility:** In homeomorphisms with local or pointwise shadowing, invariant measures can be approximated arbitrarily closely by ergodic measures with higher entropy. Even pointwise-positive shadowing on a chain-recurrence class suffices to construct semi-horseshoes and guarantee positive entropy, generalizing classic results for expansive systems [2502.10752].

- **Local Shadowing:** The distinction between local (e.g., positively shadowable points or measures) and global shadowing is substantial: systems may have rich local shadowing without global shadowing, and such pointwise properties can suffice for positive entropy or complexity, even in the absence of global hyperbolicity or expansivity [2502.10752].

## 7. Open Problems and Future Directions

Unresolved questions span several dimensions:

- Existence and characterization of topological spaces (specifically higher-dimensional continua) for which shadowing fails to be generic or is not preserved under standard constructions.
- The relationship between average or s-limit shadowing and other mixing or sensitivity properties outside equicontinuous or expansive regimes.
- The sharpness of connections between shadowing (and its variants) and properties such as structural stability, entropy generation, and measure stability in more general (particularly non-hyperbolic or nonautonomous) settings.

The extensive network of implications, equivalences, and counterexamples demonstrates that shadowing remains a central organizing principle in the structure theory of dynamical systems, with ongoing developments in theory, computation, and application to physical and statistical modeling. 

**References:**  
[1710.00313], [1701.05383], [1810.02262], [2403.15777], [1406.5822], [2311.02932], [2111.09252], [2010.03927], [2002.11199], [1909.04488], [1907.06792], [2504.00171], [1907.02446], [1004.0422], [1409.6980], [2502.10752], [2003.09801], [2408.12688], [2208.13627].

Source: https://www.emergentmind.com/topics/shadowing