On recurrence and entropy in hyperspace of continua in dimension one

We show that if $G$ is a topological graph, and $f$ is continuous map, then the induced map $\tilde{f}$ acting on the hyperspace $C(G)$ of all connected subsets of $G$ by natural formula $\tilde{f}(C)=f(C)$ carries the same entropy as $f$. This is well known that it does not hold on the larger hyperspace of all compact subsets. Also negative examples were given for the hyperspace $C(X)$ on some continua $X$, including dendrites. Our work extends previous positive results obtained first for much simpler case of compact interval by completely different tools.

💡 Research Summary

The paper investigates the relationship between a continuous self‑map f on a compact metric space that is a topological graph G and the induced map \tilde f acting on the hyperspace C(G) of all non‑empty connected subsets of G. The main theorem asserts that the topological entropy of f coincides with that of \tilde f, i.e. h_top(f)=h_top(\tilde f). This result extends earlier positive findings for intervals and trees to the broader class of one‑dimensional continua that may contain cycles (loops).

The authors begin by recalling the necessary background on topological graphs, dynamical systems, and hyperspaces equipped with the Hausdorff metric. They note that while the induced map on the full hyperspace 2^X (all compact subsets) can dramatically increase entropy—indeed, any map with positive entropy yields infinite entropy on 2^X—restricting to the connected‑set hyperspace C(X) often preserves entropy, as known for intervals and trees. However, counter‑examples exist for certain dendrites, showing that the property is not universal even in dimension one.

A central part of the work is a detailed analysis of the recurrent set Rec(\tilde f). Using Blokh’s theory of ω‑limit sets on graphs, the authors classify all possible maximal ω‑limit sets: periodic orbits, cycles of graphs, basic sets, circumferential sets, and solenoidal sets. They prove (Theorem 4.6) that any non‑degenerate recurrent continuum in C(G) is either periodic or, in the exceptional case, belongs to a minimal set generated by an irrational rotation on a loop. Lemma 3.1–3.3 establish that a recurrent continuum cannot intersect the ω‑limit set of any of its points unless it is contained in a common invariant cycle. This structural description shows that the dynamics of \tilde f on its recurrent part is essentially a finite union of periodic subsystems together with at most one rotational subsystem.

Having identified the recurrent structure, the authors invoke the variational principle for topological entropy: h_top(f)=sup_{μ∈M_f} h_μ(f), where M_f denotes the set of f‑invariant Borel probability measures. They construct an “almost conjugacy” (Definition 2.3) between the restriction of \tilde f to Rec(\tilde f) and the original map f. The map ϕ:C(G)→G sending a connected set to a distinguished point (e.g., a point of minimal distance to a fixed reference) is surjective, has connected fibers, and each fiber contains uniformly boundedly many points. By a result of Bowden, such an almost conjugacy guarantees equality of topological entropies for the two systems. Consequently, any invariant measure for \tilde f projects to an invariant measure for f with the same metric entropy, and conversely any invariant measure for f lifts to an invariant measure for \tilde f. This bi‑directional correspondence shows that entropy cannot increase under the passage from f to \tilde f.

The paper also discusses why the theorem is essentially one‑dimensional. In dimensions two and higher, simple examples exist where a map with zero entropy induces a map on 2^X with infinite entropy. Moreover, even within one‑dimensional continua, dendrites can exhibit entropy growth on C(X), as shown in earlier works. The presence of loops in graphs introduces complications such as non‑unique minimal arcs and irrational rotations, but the authors’ analysis shows that these do not lead to entropy inflation beyond that of f.

In the final sections, the authors assemble the pieces: the characterization of recurrent continua, the almost conjugacy, and the variational principle, to prove Theorem 1.1. They conclude that for any continuous graph map, the induced map on the hyperspace of connected subsets preserves topological entropy. This unifies and extends previous results, providing a comprehensive understanding of how collective dynamics on connected subsets reflect the complexity of the underlying system. The paper suggests future directions, such as exploring weighted or directed graphs, non‑continuous maps, or higher‑dimensional analogues, where entropy behavior may differ.