Average Gromov hyperbolicity and the Parisi ansatz (Q2214071)

A metric space \((S,d)\) is called \(\delta\)-hyperbolic if it satisfies Gromov's four point condition \[ (x,y)_w \ge \min\{ (x,z)_w, (y,z)_w\} - \delta \] for any four points \(x, y, z, w \in S\), where \((x,y)_w\) denotes the Gromov product. The notion of \(\delta\)-hyperbolicity, introduced by Gromov in a group-theoretic context, has found applications in many areas of mathematics, science and engineering. Still, there is one aspect of Gromov hyperbolicity that is sometimes problematic when one ventures outside the domain of very regular spaces coming from group theory and Riemannian geometry. It is the fact that the four point condition is a worst-case condition: The space is not \(\delta\)-hyperbolic if there is even a single four-tuple \((x, y, z, w)\) for which the condition fails. For these reasons, one may naturally wonder whether the four point condition may be replaced by an averaged version. This has, indeed, been proposed recently in some physics papers, but these proposals have not been mathematically analyzed. The goal of the present paper is to fill this gap. The authors define a natural notion of average \(\delta\)-hyperbolicity. Recall that the classical notion of hyperbolic metric spaces is closely related to real trees. To this end the authors define the average \(\delta\)-tree-like property. The main result of the paper shows that for any \(\varepsilon > 0\) there exists some \(\delta > 0\) such that if \((S,d)\) is an average \(\delta\)-hyperbolic metric space, then it has an average \(\varepsilon\)-tree-like structure (see Theorem 2.4 and the preceding definitions for the precise statement). This result generalises to the setting of average hyperbolicity a classical result of \textit{É. Ghys} and \textit{P. de la Harpe} [``Espaces métriques hyperboliques'', in: Sur les groupes hyperboliques d'après Mikhael Gromov. Boston, MA: Birkhäuser. 27--45 (1990)]. Interestingly, unlike in the classical case, the result of the paper has no dependence on the size of \(S\). The main technical ingredient of the proof is a weighted version of Szemerédi's regularity lemma from graph theory. It is proved in Section 4 of the paper. Unfortunately, because of the regularity lemma the proof of the main theorem is not effective. Finding effective bounds on \(\delta\) in terms of \(\varepsilon\) is an interesting open problem. As an application, the authors show that hierarchically organized pure states can be constructed in any model of a spin glass that satisfies the Parisi ultrametricity ansatz.