The Maximum Entropy of a Metric Space

We define a one-parameter family of entropies, each assigning a real number to any probability measure on a compact metric space (or, more generally, a compact Hausdorff space with a notion of similarity between points). These generalize the Shannon and Rényi entropies of information theory. We prove that on any space X, there is a single probability measure maximizing all these entropies simultaneously. Moreover, all the entropies have the same maximum value: the maximum entropy of X. As X is scaled up, the maximum entropy grows, and its asymptotics determine geometric information about X, including the volume and dimension. And the large-scale limit of the maximizing measure itself provides an answer to the question: what is the canonical measure on a metric space? Primarily, we work not with entropy itself but its exponential, which in its finite form is already in use as a measure of biodiversity. Our main theorem was first proved in the finite case by Leinster and Meckes.

Ergodic optimization in dynamical systems

Ergodic optimization is the study of problems relating to maximizing orbits and invariant measures, and maximum ergodic averages. An orbit of a dynamical system is called$f$-maximizing if the time average of the real-valued function$f$along the orbit is larger than along all other orbits, and an invariant probability measure is called$f$-maximizing if it gives$f$a larger space average than any other invariant probability measure. In this paper, we consider the main strands of ergodic optimization, beginning with an influential model problem, and the interpretation of ergodic optimization as the zero temperature limit of thermodynamic formalism. We describe typical properties of maximizing measures for various spaces of functions, the key tool of adding a coboundary so as to reveal properties of these measures, as well as certain classes of functions where the maximizing measure is known to be Sturmian.

Lyapunov-maximizing measures for pairs of weighted shift operators

Motivated by recent investigations of ergodic optimization for matrix cocycles, we study the measures of the maximum top Lyapunov exponent for pairs of bounded weighted shift operators on a separable Hilbert space. We prove that, for generic pairs of weighted shift operators, the Lyapunov-maximizing measure is unique, and show that there exist pairs of operators whose unique Lyapunov-maximizing measure takes any prescribed value less than $\log 2$ for its metric entropy. We also show that, in contrast to the matrix case, the Lyapunov-maximizing measures of pairs of bounded operators are, in general, not characterized by their supports: we construct explicitly a pair of operators and a pair of ergodic measures on the 2-shift with identical supports, such that one of the two measures is Lyapunov-maximizing for the pair of operators and the other measure is not. Our proofs make use of the Ornstein $\overline{d}$-metric to estimate differences in the top Lyapunov exponent of a pair of weighted shift operators as the underlying measure is varied.

Maximizing entropy measures for random dynamical systems

We prove the existence of relative maximal entropy measures for certain random dynamical systems of the type [Formula: see text], where [Formula: see text] is an invertibe map preserving an ergodic measure [Formula: see text] and [Formula: see text] is a local diffeomorphism of a compact Riemannian manifold exhibiting some non-uniform expansion. As a consequence of our proofs, we obtain an integral formula for the relative topological entropy as the integral of the logarithm of the topological degree of [Formula: see text] with respect to [Formula: see text]. When [Formula: see text] is topologically exact and the supremum of the topological degree of [Formula: see text] is finite, the maximizing measure is unique and positive on open sets.

On the existence of maximizing measures for irreducible countable Markov shifts: a dynamical proof

We prove that if ${\Sigma }_{\mathbf{A} } ( \mathbb{N} )$ is an irreducible Markov shift space over $ \mathbb{N} $ and $f: {\Sigma }_{\mathbf{A} } ( \mathbb{N} )\rightarrow \mathbb{R} $ is coercive with bounded variation then there exists a maximizing probability measure for $f$, whose support lies on a Markov subshift over a finite alphabet. Furthermore, the support of any maximizing measure is contained in this same compact subshift. To the best of our knowledge, this is the first proof beyond the finitely primitive case in the general irreducible non-compact setting. It is also noteworthy that our technique works for the full shift over positive real sequences.

Maximizing measures for partially hyperbolic systems with compact center leaves

We obtain the following dichotomy for accessible partially hyperbolic diffeomorphisms of three-dimensional manifolds having compact center leaves: either there is a unique entropy-maximizing measure, this measure has the Bernoulli property and its center Lyapunov exponent is 0, or there are a finite number of entropy-maximizing measures, all of them with non-zero center Lyapunov exponents (at least one with a negative exponent and one with a positive exponent), that are finite extensions of a Bernoulli system. In the first case of the dichotomy, we obtain that the system is topologically conjugated to a rotation extension of a hyperbolic system. This implies that the second case of the dichotomy holds for an open and dense set of diffeomorphisms in the hypothesis of our result. As a consequence, we obtain an open set of topologically mixing diffeomorphisms having more than one entropy-maximizing measure.

Flattening functions on flowers

Let T be an orientation-preserving Lipschitz expanding map of the circle ${\mathbb T}$. A pre-image selector is a map $\tau : {\mathbb T} \to {\mathbb T}$ with finitely many discontinuities, each of which is a jump discontinuity, and such that τ(x)∈T−1(x) for all $x\in {\mathbb T}$. The closure of the image of a pre-image selector is called a flower and a flower with p connected components is called a p-flower. We say that a real-valued Lipschitz function can be Lipschitz flattened on a flower whenever it is Lipschitz cohomologous to a constant on that flower. The space of Lipschitz functions which can be flattened on a given p-flower is shown to be of codimension p in the space of all Lipschitz functions, and the linear constraints determining this subspace are derived explicitly. If a Lipschitz function f has a maximizing measure S which is Sturmian (i.e. is carried by a 1-flower), it is shown that f can be Lipschitz flattened on some 1-flower carrying S.

A LARGE DEVIATION PRINCIPLE FOR THE EQUILIBRIUM STATES OF HÖLDER POTENTIALS: THE ZERO TEMPERATURE CASE

Consider a α-Hölder function A : Σ → ℝ and assume that it admits a unique maximizing measure μmax. For each β, we denote μβ, the unique equilibrium measure associated to βA. We show that (μβ) satisfies a Large Deviation Principle, that is, for any cylinder C of Σ, [Formula: see text] where [Formula: see text] where V(x) is any strict subaction of A.