Measures of Maximal Entropy

Updated 29 January 2026

MMEs are invariant measures that achieve the highest metric entropy, serving as a cornerstone in the thermodynamic formalism of dynamical systems.
Their existence, uniqueness, and finiteness depend on system regularity, expansiveness, and hyperbolicity as demonstrated via symbolic codings and Markov partitions.
Explicit constructions in suspension flows and partially hyperbolic maps reveal MMEs’ role in exposing ergodic properties, statistical universality, and complex entropy landscapes.

A measure of maximal entropy (MME) is a fundamental object in topological, symbolic, and smooth dynamical systems. For a continuous transformation $f:X \to X$ on a compact metric space, or for a continuous semiflow or flow, an MME is a probability measure that achieves the supremum of the metric (Kolmogorov–Sinai) entropy among all invariant measures. The theory of MMEs reveals subtle connections between entropy, hyperbolic structures, ergodic properties, and the geometry of invariant sets, and it is central to the thermodynamic formalism. The generic behavior, uniqueness, finiteness, and structural properties of MMEs depend acutely on the regularity, expansiveness, and hyperbolicity of the underlying system.

1. Foundational Principles and Definitions

Given a compact space $X$ and a continuous map $f:X\to X$ , the topological entropy $h_{\rm top}(f)$ measures the exponential growth of distinguishable orbit segments. For an $f$ -invariant Borel probability measure $\mu$ , the metric entropy $h_\mu(f)$ quantifies the exponential growth rate of distinguishable name sequences with respect to partitions. The variational principle asserts: $h_{\rm top}(f) = \sup_{\mu\in \mathcal{M}_f} h_\mu(f),$ where $\mathcal{M}_f$ denotes the simplex of invariant probability measures. An MME is any $\mu_*$ attaining this supremum: $h_{\mu_*}(f) = h_{\rm top}(f)$ .

For suspension semi-flows $\Phi^t$ over a base system $(X,f)$ with roof function $\phi:X\to(0,\infty)$ , the corresponding entropy and invariant measures are analyzed through the suspension space $X^\phi$ . The entropy of the time-one map is determined via Abramov's formula: $h_{\mu^\phi}(\Phi^1) = \frac{h_\mu(f)}{\int_X \phi\, d\mu}.$ The set of MMEs for $\Phi^t$ , denoted $\mathrm{MME}(\Phi)$ , corresponds to the lifts of MMEs for the base subject to maximizing $h_\mu(f)/\int \phi\,d\mu$ (Kucherenko et al., 2019).

2. Existence, Uniqueness, and Finiteness

Symbolic and Expansive Systems

In symbolic dynamics, such as subshifts of finite type or sofic shifts, the existence and uniqueness of MMEs are well understood: expansive systems with the specification property and with Hölder regularity (for potentials or roof functions) admit unique, fully supported MMEs (Kucherenko et al., 2019, Kucherenko et al., 2017). The uniqueness may fail if the specification property is absent or if the regularity of the roof is weakened; e.g., with merely continuous roofs, the set of MMEs can exhibit arbitrary finite, countable, or uncountable cardinality, even for systems orbit-equivalent to ones with unique MME (Kucherenko et al., 2019, Iommi et al., 2019).

Smooth and Partially Hyperbolic Contexts

For $C^\infty$ surface diffeomorphisms, Buzzi–Crovisier–Sarig established that, provided $h_{\rm top}(f) > 0$ , there are only finitely many ergodic MMEs; in the topologically transitive case, the MME is unique (Buzzi et al., 2018). These results extend, subject to uniform hyperbolicity or an entropy gap, to higher dimensional settings:

For open classes of partially hyperbolic diffeomorphisms with 1-dimensional center bundles and an entropy gap $h_u(f) > h_s(f)$ (unstable entropy exceeds stable), Mongez–Pacifico proved finiteness and robust upper semi-continuity of the number of MMEs (Mongez et al., 2024).
For non-singular $C^\infty$ flows on 3-manifolds with $h_{\rm top}(\varphi^t) > 0$ , the number of ergodic MMEs is finite (Zang, 27 Mar 2025).
For skew products with strong positive recurrence properties—a property implying a uniform hyperbolic block structure—similar finiteness and exponential mixing of MMEs hold (Marin et al., 12 Sep 2025).
In non-invertible and singular hyperbolic systems (e.g., dispersing billiards, certain non-uniformly expanding maps), criteria guarantee at most one MME per homoclinic class, generalizing the invertible theory (Lima et al., 2024).

The Markov partition and symbolic coding approach allows reduction of uniqueness/finiteness of MMEs for non-uniformly hyperbolic systems to Gurevich’s results for countable state Markov shifts: each irreducible component carries at most one MME (Buzzi et al., 2018, Lima et al., 2024).

Table: Cardinality of MMEs in Key Settings

System Class	Generic MME Structure	Reference
Expansive + Specification + Hölder (discrete-time)	Unique, fully supported	(Kucherenko et al., 2019)
Suspension flow with continuous roof	Any finite, countable, or uncountable number possible	(Kucherenko et al., 2019)
$C^\infty$ surface diffeomorphisms	Finitely many; unique if transitive	(Buzzi et al., 2018)
Certain partially hyperbolic diffeos ( $h_u > h_s$ )	Finitely many; upper semi-continuity under $C^1$ perturb	(Mongez et al., 2024)
Generic rational maps of degree $\geq 2$	Unique	(Hawkins et al., 2017)
Dyck shift, Heterochaos baker, etc.	Exactly two ergodic MMEs	(Takahasi et al., 2022)

3. Explicit Constructions and Universality Phenomena

A key theme is the universality of MME behavior in suspension and skew-product flows:

Suspension semi-flows: Given any closed invariant $Y \subset X$ with positive entropy, there exists a continuous roof $\rho$ such that the MMEs for the suspension consist of the lifts of the MMEs for $(Y, f|_Y)$ . For the full shift, by selecting subshifts with prescribed entropy-maximizing measures, one can realize the set of ergodic MMEs of the suspension semi-flow as having exactly any desired finite, countable, or uncountable cardinality (Kucherenko et al., 2019, Kucherenko et al., 2017, Iommi et al., 2019).
Dichotomy for topological skew-products: For partially hyperbolic systems with a 1D compact center direction virtually skew over a transitive Anosov, under a minimality condition, there is a dichotomy: either a unique non-hyperbolic MME or exactly two hyperbolic MMEs of opposite center exponent (Tahzibi et al., 2024, Ures et al., 2019, Buzzi et al., 2019).
Non-uniqueness in symbolic systems: For coded shifts with specific entropy characteristics, MMEs can be explicitly constructed and their computability investigated; cases such as the Dyck shift provide models with exactly two ergodic MMEs whose explicit symbolic representations are computable (Takahasi et al., 2022, Kucherenko et al., 22 Jan 2026).
Perturbative density results: Arbitrarily small perturbations of a suspension flow (in the $C^0$ topology) can yield a transition between unique and uncountably many MMEs, confirming the $C^0$ -density of both phenomena (Iommi et al., 2019).

4. Hyperbolic and Ergodic Properties

The ergodic and statistical structure of MMEs reflects deep connections with hyperbolicity and mixing:

Hyperbolic MMEs: In Axiom A and uniformly hyperbolic systems, all MMEs are hyperbolic (non-zero Lyapunov exponents). In partially hyperbolic, circle-bundle, or time-one maps of Anosov flows, either a unique MME with zero center exponent (non-hyperbolic) occurs, or exactly two MMEs with opposite-sign center exponents (Buzzi et al., 2019, Ures et al., 2019, Tahzibi et al., 2024).
SRB property: In certain partially hyperbolic or DA-diffeomorphism settings, rigidity phenomena ensure that MMEs coincide with SRB (Sinai–Ruelle–Bowen) measures if and only if the sum of positive Lyapunov exponents for all periodic orbits matches that of a linear model (Micena et al., 2024). In smooth volume-preserving settings, this can even imply smooth conjugacy to the linear model.
Mixing and statistical properties: For systems with strong positive recurrence, each ergodic MME exhibits exponential decay of correlations and satisfies advanced statistical properties such as large deviations and almost-sure invariance principles (Marin et al., 12 Sep 2025).
Homoclinic classes and uniqueness: In both invertible and non-invertible settings, the number of ergodic MMEs is controlled by the number of distinct hyperbolic homoclinic classes satisfying certain entropy bounds; within each class, at most one adapted hyperbolic MME may exist, and it is Bernoulli up to finite rotation (Buzzi et al., 2018, Lima et al., 2024, Ghezal, 15 Nov 2025).

5. Methodological Advances and Proof Techniques

The development of the general theory relies on several high-level methodologies:

Symbolic Codings: For non-uniformly hyperbolic systems, Markov partitions and countable state Markov shifts provide a universal symbolic coding, enabling transfer of ergodic and thermodynamic formalism. Irreducibility decomposes the measure space into homoclinic classes, each potentially supporting a unique large-entropy measure (Buzzi et al., 2018, Lima et al., 2024).
Uniform Pesin Blocks and Pliss-type Lemmas: For partially hyperbolic systems—with dominated splitting and uniform entropy gaps—construction of uniform Pesin blocks enables control of local entropy and uniform sizes of unstable/stable manifolds, facilitating finiteness arguments and perturbative stability (Mongez et al., 2024, Mongez et al., 24 Feb 2025).
Spectral Decomposition: Reduction to finitely many (measured) homoclinic classes allows application of symbolic and per-class uniqueness results, critical for establishing finiteness even in non-invertible or surface endomorphisms (Buzzi et al., 2018, Ghezal, 15 Nov 2025).
Thermodynamic Constructions: For symbolic suspension flows, convex analysis for pressure functions and the construction of roof functions tailored to support specified equilibrium states enable precise engineering of the MME structure (Kucherenko et al., 2019, Kucherenko et al., 2017, Iommi et al., 2019).

6. Structural Universality, Perturbations, and Open Directions

The emerging picture from modern research is that in the topological and partially hyperbolic categories, the behavior of MMEs is structurally universal and decoupled from naive orbit equivalence. For continuous roof functions in suspension flows, or low-regularity perturbations, finiteness, countability, and uncountability of MMEs are open to engineering via the base system or via perturbed pressure functions (Kucherenko et al., 2019, Iommi et al., 2019). This universality underscores that uniqueness, cardinalities, and even the “approachability” of MMEs are fundamentally determined by subtle regularity or hyperbolicity conditions, not just orbit structure.

Broader Implications

Rigidity and invariant geometries: In partially hyperbolic and DA-diffeomorphism settings, periodic orbit exponent spectra dictate whether MMEs coincide with SRB and physical measures, leading to dynamical rigidity and, in some cases, smooth conjugacy with the linear model (Micena et al., 2024).
Computation and algorithmic aspects: In symbolic dynamics, the computability of MMEs depends sharply on the entropy characteristics (concatenation versus residual) and on the explicit computability of the Vere–Jones parameter (Kucherenko et al., 22 Jan 2026).
Statistical universality: For systems satisfying strong positive recurrence, not only is the number of MMEs finite, but each such measure is exponentially mixing, reinforcing the statistical universality of the dynamical system (Marin et al., 12 Sep 2025).
Extensions and Open Problems: Ongoing research explores MMEs in systems with higher-dimensional center bundles, robustness of universality phenomena, and the statistical properties and uniqueness criteria for non-uniformly hyperbolic systems with singularities or non-invertibility (Lima et al., 2024).

7. Key Examples and Applications

Specific constructions illustrate the scope and subtlety of MMEs:

Suspension flows with prescribed MME cardinality: By constructing roof functions over full shifts using prescribed subshifts $Y$ , one engineers MMEs of any desired finite, countable, or uncountable cardinality (Kucherenko et al., 2019, Kucherenko et al., 2017, Iommi et al., 2019).
Dichotomies in partially hyperbolic systems: Time-one maps of Anosov flows, circle bundle extensions over Anosov maps, and non-accessible topological skew products exhibit a dichotomy: either exactly two hyperbolic MMEs of opposite sign or a unique non-hyperbolic MME (Buzzi et al., 2019, Ures et al., 2019, Tahzibi et al., 2024).
Heterochaos baker and Dyck shift: These systems are models with two ergodic MMEs and provide insight into the breakdown of entropy-approachability and the complexity of the entropy landscape in higher-dimensional piecewise-affine systems (Takahasi et al., 2022).
Surface endomorphisms with entropy gap: Finite-entropy criteria control the number of MMEs, even in non-invertible local diffeomorphisms with degree-induced lower entropy bounds (Ghezal, 15 Nov 2025).
Symbolic systems with computable MMEs: For coded shifts with computable Vere–Jones parameters, explicit and algorithmically accessible MMEs are constructed for a wide array of symbolic spaces, including $S$ -gap and $\beta$ -shifts. Non-computability can arise when residual complexity dominates, even when the measure itself is unique (Kucherenko et al., 22 Jan 2026).

The theory of measures of maximal entropy reveals a landscape controlled by entropy, symbolic coding, hyperbolic geometry, and regularity. Finiteness, uniqueness, computability, and statistical properties of MMEs serve as organizing principles in the thermodynamic approach to dynamical systems, with ongoing research probing the boundaries of universality, rigidity, and computability in increasingly broad dynamical contexts.