Kaplan–Yorke Conjecture

• Kaplan–Yorke Conjecture is a framework that equates the information dimension of a chaotic attractor to its Lyapunov dimension derived from physical invariant measures.
• It employs mathematical tools such as Oseledets’ theorem and case studies (e.g., baker maps) to analyze conditions where the equality holds or fails, particularly in coupled versus uncoupled systems.
• The conjecture plays a fundamental role in multifractal analysis and ergodic theory, providing practical insights into predicting fractal dimensions in dissipative dynamical systems.

The Kaplan–Yorke conjecture postulates that for "typical" dissipative dynamical systems possessing a physical invariant measure, the information dimension of the measure coincides with the so-called Lyapunov (Kaplan–Yorke) dimension. This proposed equality, rooted in the structure of Lyapunov spectra, underlies the empirical observation that many chaotic attractors display fractal dimensions predictable from their local expansion and contraction rates. The conjecture is central to multifractal analysis, ergodic theory, and the theory of strange attractors.

1. Formal Statement and Mathematical Background

Let $F\colon M \subset \mathbb{R}^d \to M$ be a piecewise-$C^1$ map that admits an ergodic invariant (physical) measure $\mu$. By Oseledets’ theorem, for $\mu$-almost every point, the system exhibits Lyapunov exponents $$\lambda_1 \geq \lambda_2 \geq \cdots \geq \lambda_d$$. The Lyapunov (Kaplan–Yorke) dimension $D_{KY}(\mu)$ is defined by determining

$$j = \max\left\{ i : \sum_{k=1}^i \lambda_k(\mu) \geq 0 \right\},$$

with

$$D_{KY}(\mu) = j + \frac{\sum_{k=1}^j \lambda_k(\mu)}{|\lambda_{j+1}(\mu)|}.$$

The information (Rényi-information or correlation) dimension is

$$D_1(\mu) = \lim_{\varepsilon\to 0} \frac{\int_M \log\left[ \mu(B(x,\varepsilon)) \right]\, d\mu(x)}{\log \varepsilon}$$

(provided the limit exists). The Kaplan–Yorke conjecture asserts that for typical dynamical systems with a physical measure,

$D_1(\mu) = D_{KY}(\mu).$

For invertible maps in two dimensions, with Lyapunov exponents $C^1$0, $C^1$1, this reduces to $C^1$2 (Gröger et al., 2013, Hoover et al., 2019).

2. Lyapunov Spectrum, Additivity, and Mechanisms of Equality and Failure

Lyapunov exponents quantify the exponential rates of divergence or contraction of nearby trajectories along invariant directions. Information dimension, in contrast, measures the scaling of the probability measure’s mass under small balls. While $C^1$3 is additive under Cartesian products (i.e., for independent subsystems), $C^1$4 generally is not. Consequently, in product systems where the Lyapunov spectrum is not balanced across factors, $C^1$5 may fail to predict $C^1$6. This discrepancy is a key mechanism for the breakdown of the Kaplan–Yorke conjecture in uncoupled systems (Gröger et al., 2013).

3. Case Studies: Baker’s Maps, Coupling, and Dimension Mismatch

Uncoupled Systems: Skinny Baker’s Maps

On $C^1$7, the skinny baker’s map $C^1$8 defined by

$C^1$9

has an SRB measure of information dimension $\mu$0 and Lyapunov exponents $\mu$1. For two such systems with parameters $\mu$2, their uncoupled product map in four dimensions,

$\mu$3

carries a measure with

$\mu$4

but non-additive $\mu$5, leading to $\mu$6 unless $\mu$7 (Gröger et al., 2013).

Coupling and Prevalence

Introducing uni-directional coupling via a function $\mu$8 as

$\mu$9

generates a skew-product where $\mu$0 drives $\mu$1. For such systems:

• If $\mu$2, $\mu$3 for all $\mu$4 (dimension gap persists).
• If $\mu$5, the dimensions always coincide: $\mu$6.
• If $\mu$7, for a prevalent set of $\mu$8 in $\mu$9, $$\lambda_1 \geq \lambda_2 \geq \cdots \geq \lambda_d$$0 (dimension equality is restored “typically,” in the sense of prevalence).

These results depend crucially on the prevalence concept of Hunt–Sauer–Yorke, in which a property is typical if its failure set is “shy” (not merely meager or of measure zero in finite dimensions) (Gröger et al., 2013).

4. Illustration via Reversible Baker Maps: Empirical Approximations

Two detailed constructions—N2 and N3—provide explicit two-dimensional maps for analyzing the conjecture. The N2 map, a two-panel rotated baker map, and N3, a three-panel variation, both manifest dissipative fractal attractors.

For N2, computations yield:

Method	$$\lambda_1 \geq \lambda_2 \geq \cdots \geq \lambda_d$$1 (1D Projection)	$$\lambda_1 \geq \lambda_2 \geq \cdots \geq \lambda_d$$2 (2D)	Kaplan–Yorke $$\lambda_1 \geq \lambda_2 \geq \cdots \geq \lambda_d$$3
Area-wise	0.78969	1.78969
Point-wise	$$\lambda_1 \geq \lambda_2 \geq \cdots \geq \lambda_d$$4 0.7415	1.7415
Formula			$$\lambda_1 \geq \lambda_2 \geq \cdots \geq \lambda_d$$5

For N3, all three estimates coincide: $$\lambda_1 \geq \lambda_2 \geq \cdots \geq \lambda_d$$6 (1D), $$\lambda_1 \geq \lambda_2 \geq \cdots \geq \lambda_d$$7 (2D), $$\lambda_1 \geq \lambda_2 \geq \cdots \geq \lambda_d$$8 (Hoover et al., 2019).

In N2, exact self-similarity (“perfect scale model”) results in a breakdown of the conjectured equality, as the area-wise, point-wise, and Kaplan–Yorke estimates disagree. In contrast, N3’s broken scale invariance eliminates this discrepancy, and all approaches agree. This suggests that smoothness or the absence of exact self-similarity is essential for the validity of the Kaplan–Yorke formula.

5. Analytical and Computational Techniques

Dimension calculations employ several complementary methods:

• Potential-theoretic approach (Sauer–Yorke): The pointwise dimension $$\lambda_1 \geq \lambda_2 \geq \cdots \geq \lambda_d$$9 is the supremum of $D_{KY}(\mu)$0 such that the $D_{KY}(\mu)$1-potential $D_{KY}(\mu)$2 is finite.
• Block-triangular derivative analysis: Establishes that certain invariant exponents persist under coupling.
• Conjugacy analysis: For $D_{KY}(\mu)$3, the conjugacy $D_{KY}(\mu)$4 between uncoupled/coupled systems is bi-Lipschitz (dimensions match), while for $D_{KY}(\mu)$5 it is only Hölder (dimensions may differ).
• Random-walk equivalence: In one-dimensional projections of the baker maps, the evolution is exactly represented by a confined random walk, dramatically accelerating computation of invariant histograms and corresponding dimensions (Hoover et al., 2019).

These methodologies facilitate both theoretical analysis and high-precision empirical estimation of the dimension spectrum.

6. Open Problems and Implications

Several unresolved issues remain central to the Kaplan–Yorke framework:

• Characterizing the exceptional, non-prevalent coupling functions yielding persistent dimension gaps.
• Generalizing prevalence and the dimension equality to infinite-dimensional parameter families (e.g., PDEs, general diffeomorphism families).
• Systematically formulating a refined Kaplan–Yorke dimension for higher-dimensional skew-product systems (multiple driven blocks).
• Computing the complete dimensional spectrum $D_{KY}(\mu)$6, especially for $D_{KY}(\mu)$7.

A plausible implication is that robust dimension mismatch serves as a signature of one-way coupling in high-dimensional chaotic networks, while bi-directional or “sufficient” coupling generally restores the equality for prevalent parameter choices (Gröger et al., 2013). A refined conjectural formula, distinguishing between base and fiber Lyapunov exponents, may be necessary for such skew-product settings.

7. Summary and Broader Context

The Kaplan–Yorke conjecture formalizes the empirical connection between dynamical chaos and fractal geometry via Lyapunov spectra, providing a heuristic for the information dimension of attractors in “typical” dissipative systems. Rigorous studies confirm the conjecture’s broad validity for prevalent parameter choices, but also elucidate its limitations—especially in product systems or maps with exact self-similarity. The interplay between additivity, spectral structure, and system architecture is crucial. The conjecture remains fundamental in the theoretical and numerical analysis of non-equilibrium statistical mechanics, multifractal dynamics, and the theory of chaotic attractors (Gröger et al., 2013, Hoover et al., 2019).