Ergodic Transport Theory, periodic maximizing probabilities and the twist condition

TL;DR

This paper analyzes ergodic transport problems focusing on cases where the maximizing probability is unique and periodic, establishing finiteness of optimal solutions under certain conditions.

Contribution

It provides a detailed analysis of unique periodic maximizing probabilities in ergodic transport, extending previous work with new conditions like the twist condition.

Findings

Under generic conditions, the set of optimal w_x is finite for all x.

The twist condition and uniqueness imply a finite structure of optimal solutions.

The results connect ergodic theory with transport and optimization principles.

Abstract

The present paper is a follow up of another one by A. O. Lopes, E. Oliveira and P. Thieullen which analyze ergodic transport problems. Our main focus will a more precise analysis of case where the maximizing probability is unique and is also a periodic orbit. Consider the shift T acting on the Bernoulli space \Sigma={1, 2, 3,.., d}^\mathbb{N} $ and $A:\Sigma \to \mathbb{R} a Holder potential. Denote m(A)=max_{\nu is an invariant probability for T} \int A(x) \; d\nu(x) and, \mu_{\infty,A}, any probability which attains the maximum value. We assume this probability is unique (a generic property). We denote \T the bilateral shift. For a given potential Holder A:\Sigma \to \mathbb{R}, we say that a Holder continuous function W: \hat{\Sigma} \to \mathbb{R} is a involution kernel for A, if there is a Holder function A^*:\Sigma \to \mathbb{R}, such that, A^*(w)= A\circ \T^{-1}(w,x)+ W \circ \T^{-1}(w,x) - W(w,x). We say that A^* is a dual potential of A. It is true that m(A)=m(A^*). We denote by V the calibrated subaction for A, and, V^* the one for A^*. We denote by I^* the deviation function for the family of Gibbs states for \beta A, when \beta \to \infty. For each x we get one (more than one) w_x such attains the supremum above. That is, solutions of V(x) = W(w_x,x) - V^* (w_x)- I^*(w_x). A pair of the form (x,w_x) is called an optimal pair. If \T is the shift acting on (x,w) \in {1, 2, 3,.., d}^\mathbb{Z}, then, the image by \T^{-1} of an optimal pair is also an optimal pair. Theorem - Generically, in the set of Holder potentials A that satisfy (i) the twist condition, (ii) uniqueness of maximizing probability which is supported in a periodic orbit, the set of possible optimal w_x, when x covers the all range of possible elements x in \in \Sigma, is finite.