Bohr Chaos in Dynamical Systems

• Bohr chaos is a topological property defined by the universal correlatability of every non-trivial bounded sequence with some observable.
• It is characterized by the interaction of specification, horseshoes, and hyperbolic structures, making it a stronger notion than classical positive entropy or Li–Yorke chaos.
• Distinct from uniquely ergodic or zero entropy systems, Bohr chaos highlights maximum arithmetic complexity and persistent non-cancellation in time averages.

Bohr chaos is a topological dynamical property that encapsulates a strong, universal non-orthogonality between bounded sequences and process observables, fundamentally differentiating it from classical and metric chaos. In a Bohr chaotic system, every “non-trivial” bounded real (or complex) sequence necessarily correlates—on average—with some observable along some orbit, precluding long-term cancellation. This universal correlatability is strictly stronger than positive entropy, horseshoe existence, or classical Li–Yorke chaos, and is in marked tension with orthogonality phenomena central to Sarnak’s Möbius disjointness conjecture (Fan et al., 2021, Tal, 2021, Kawaguchi, 11 Jan 2026). The structure and implications of Bohr chaos are deeply linked to specification, abundance of invariant measures, and hyperbolic/geometric features in dynamical systems.

1. Definition and Fundamental Characterization

Let $(X,T)$ denote a compact metric dynamical system, with $T:X\to X$ continuous and $C(X)$ the Banach space of continuous observables. A bounded sequence $(w_n)$ is called non-trivial if

$$\limsup_{N\to\infty}\frac{1}{N}\Bigl|\sum_{n=0}^{N-1} w_n\Bigr| > 0.$$

The sequence $(w_n)$ is said to be orthogonal to $(X,T)$ if, for every $f\in C(X)$ and every $x\in X$,

$$\lim_{N\to\infty}\frac{1}{N}\sum_{n=0}^{N-1} w_n f(T^n x) = 0.$$

A system is Bohr chaotic if there exists no non-trivial bounded sequence orthogonal to it; equivalently, for every non-trivial weight there is a pair $(f,x)$ (observable and basepoint) so that

$$\limsup_{N\to\infty} \Bigl|\frac{1}{N}\sum_{n=0}^{N-1} w_n f(T^n x)\Bigr| > 0.$$

This property demands, for every such sequence, the existence of persistent correlation with some dynamical observable. The notion is strictly topological: all systems topologically conjugate to a Bohr chaotic system inherit the property (Fan et al., 2021).

2. Structural Criteria and Obstructions

A central obstruction to Bohr chaos is unique ergodicity: if $(X,T)$ admits a unique invariant measure, then the Wiener–Wintner theorem ensures vanishing averages for all but countably many weights, so the system is not Bohr chaotic (Fan et al., 2021, Tal, 2021). More generally, any system with strictly fewer than continuum-many ergodic invariant measures cannot be Bohr chaotic. Zero topological entropy also precludes Bohr chaos since these systems are Furstenberg-disjoint from Bernoulli shifts and lack sufficient complexity to realize persistent correlation with arbitrary weights. Positive entropy alone does not suffice unless coupled with a richer invariant-measure structure (Tal, 2021).

The property is also invariant under topological extensions: any continuous surjective factor from a Bohr chaotic to another system inherits Bohr chaos (Fan et al., 2021). This underscores its status as a topological invariant, stricter than positive entropy.

3. Sufficient Conditions: Horseshoes, Specification, and Hyperbolicity

Systems with a horseshoe, specification, or certain hyperbolic structures always exhibit Bohr chaos. The existence of a (possibly high-order) horseshoe implies a subsystem topologically conjugate to a full symbolic shift; the latter is Bohr chaotic since, for any non-trivial sequence, one can code orbits to match its sign pattern, constructing correlated pairs ((Fan et al., 2021), Theorem 1.1).

The specification property, which ensures the ability to shadow arbitrarily long prescribed orbit segments with controlled errors, is a robust sufficient condition. In symbolic and invertible-dynamical settings, specification allows one to code any desired correlation structure, yielding a continuous factor map onto a full shift and thus guaranteeing Bohr chaos (Tal, 2021).

Hyperbolic sets provide another structural route: if a system admits an infinite chain-transitive hyperbolic set and has shadowing and expansiveness on its neighborhood, then it is Bohr chaotic (Kawaguchi, 11 Jan 2026). This encompasses classical Anosov and Smale systems, generic $C^1$-diffeomorphisms with hyperbolic invariant sets, and toral automorphisms of positive entropy.

Sufficient Condition	Result	Key Reference
Horseshoe	Bohr chaos	(Fan et al., 2021)
Specification	Bohr chaos	(Tal, 2021, Fan et al., 2021)
Hyperbolic set + shadowing	Bohr chaos	(Kawaguchi, 11 Jan 2026)
Toral affine map, $h_{\rm top}>0$	Bohr chaos	(Fan et al., 2021)

4. Comparative Landscape and Theoretical Relations

Bohr chaos is distinct from other topological chaos notions. While positive entropy, Li–Yorke chaos, and the presence of horseshoes are classical indicators of dynamical complexity, Bohr chaos encodes an arithmetic complexity—the absence of orthogonality to any bounded (non-trivial) weight sequence. This is diametrically opposed to properties studied in Sarnak's Möbius disjointness conjecture, where the Möbius function and, by extension, all bounded weakly almost periodic sequences are required to be orthogonal to zero-entropy systems (Fan et al., 2021).

Unlike metric or measure-theoretic chaos (e.g., K-systems, mixing), Bohr chaos is insensitive to invariant measures and is strictly topological; conversely, uniquely ergodic or “measure-simple” systems cannot exhibit Bohr chaos even if they have positive entropy.

Bohr chaos naturally links to the abundance of ergodic invariant measures and rich spectrum of dynamical eigenvalues (requiring essentially all unit-circle phases to appear as eigenvalues for some measure), further setting it apart from entropy or mixing-based chaos criteria (Tal, 2021).

5. Construction, Examples, and Minimality

Tal’s construction provides a minimal, non-uniquely ergodic, positive-entropy subshift that is Bohr chaotic (Tal, 2021). The approach uses recursively nested mixing SFTs, each designed to force all words of a given length while shrinking the subshift space, ultimately yielding a minimal intersection point with positive topological entropy and Bohr chaos (Theorem 3.3). Systems built via specification similarly code any prescribed orbit pattern, ensuring universal correlatability.

Generic examples beyond the symbolic category include:

• Piecewise monotonic interval maps with positive entropy (via the existence of SFT horseshoe subsystems),
• Toral automorphisms and affine maps with positive topological entropy,
• Generic $C^1$ or $C^{1+\alpha}$ diffeomorphisms on manifolds possessing Smale horseshoe structures or hyperbolic sets (Fan et al., 2021).

System Type	Bohr Chaos?	Mechanism or Obstruction
SFT with specification	Yes	Coding and shadowing
Minimal, uniquely ergodic	No	Lacks spectrum, invariant measures
Anosov diffeomorphism	Yes	Hyperbolicity + specification
Sturmian rotation	No	Zero entropy, unique ergodicity

6. Implications, Open Problems, and Directions

Bohr chaos introduces a new hierarchy of complexity that refines classical entropy- and mixing-based notions. It codifies a universal failure of cancellation for time-averages with bounded weights, thus serving as a marker for a maximal form of long-term unpredictability in the sense of dynamical averages.

Open problems include characterizing the precise interplay between invariant measure abundance and Bohr chaos, understanding the converse statements regarding joinings and continuous correlations, and extending the framework to systems lacking specification but realizing “semi-horseshoe” or partial coding structures (Tal, 2021). Other questions concern possible generalizations beyond bounded sequences and the implications for physical or arithmetic time series.

A plausible implication is that Bohr chaos, due to its stringent requirements, demarcates the “deeply nonintegrable” class of deterministic dynamics, distinguishing it from measure-theoretic randomness and other classical chaos. It also impinges on symbolic, smooth, and even group-automorphism systems, suggesting new routes for arithmetic complexity classification (Fan et al., 2021, Kawaguchi, 11 Jan 2026).