Invariant measures with full support and approximation by zero-entropy systems in the C 0 C^{0} -Gromov–Hausdorff topology Jorge Crisóstomo, Richard Cubas Abstract . In this paper we prove that every homeomorphism of a compact metric space admitting an invariant probability measure with full support can be approximated in the C 0 C^{0} -Gromov–Hausdorff topology by homeomorphisms with zero topological entropy. The argument relies on the ergodic decomposition theorem and on the existence of points with dense positive orbit in the supports of suitable ergodic components. As a consequence, topological entropy is not stable under C 0 C^{0} -Gromov–Hausdorff perturbations within this class. We also show that if, in addition, the homeomorphism is topologically G ​ H GH -stable, then its periodic points are dense in the ambient space. Finally, by combining this framework with a previous result on transitive and topologically G ​ H GH -stable homeomorphisms, we deduce that every dynamics in this class admits an invariant measure with full support and therefore falls within the scope of the general approximation theorem by zero-entropy systems. 2025 Mathematics Subject Classification Primary 37D30 1. Introduction Topological stability is one of the central notions in Dynamical Systems for describing the qualitative persistence of a dynamics under perturbations. Roughly speaking, a topologically stable system preserves its dynamical behavior, at least at the level of semi-conjugacy or conjugacy, under sufficiently small perturbations when these are considered on the same phase space. In the classical C^{0} topology, this notion has been extensively studied and is closely related to fundamental properties such as expansiveness, shadowing, and topological conjugacy theory [ 10 , 11 , 6 , 8 ] . A more flexible framework was introduced by Arbieto and Morales in [ 1 ] , who defined a C^{0} -Gromov–Hausdorff distance between dynamical systems possibly defined on different compact metric spaces. This distance combines the uniform metric between maps with the Gromov–Hausdorff distance between metric spaces [ 4 , 2 ] , and allows one to compare dynamics even when the phase spaces do not coincide. Within this context, the authors also introduced the notion of GH topological stability, thereby extending the classical concept of topological stability to a more general setting [ 1 ] . From this viewpoint, several questions have arisen concerning the dynamical structure induced by the C^{0} -Gromov–Hausdorff topology. One of the most natural is to determine when a given dynamics can be approximated, in the GH^{0} sense, by simpler systems. In this direction, it was proved in [ 3 ] that every topologically transitive homeomorphism of a compact metric space can be approximated in the C^{0} -Gromov–Hausdorff topology by periodic dynamics. Later, Jung [ 5 ] obtained, under additional assumptions, converse results and related characterizations in terms of chain transitivity. Complementary results and a systematic exposition of this framework can also be found in the monograph of Lee and Morales [ 7 ] . The purpose of this paper is to revisit this approximation problem from an ergodic viewpoint. Instead of assuming topological transitivity as the main hypothesis, we assume that the homeomorphism admits an invariant probability measure with full support. This condition is natural from the standpoint of ergodic theory and, at the same time, sufficiently robust to produce finite approximating models [ 12 , 9 ] . In particular, we show that this hypothesis allows one to replace the existence of a globally dense orbit by a finite family of orbit blocks associated with suitably selected ergodic components. Our first main result states that every homeomorphism of a compact metric space admitting an invariant probability measure with full support can be approximated in the C^{0} -Gromov–Hausdorff topology by homeomorphisms with zero topological entropy. Theorem 1.1 . Let f:X\to X be a homeomorphism of a compact metric space. If f admits an invariant probability measure with full support, then for every \delta>0 there exist a compact metric space Y_{\delta} and a homeomorphism g_{\delta}:Y_{\delta}\to Y_{\delta} such that d_{GH^{0}}(f,g_{\delta})<\delta\qquad\text{and}\qquad h_{\topo}(g_{\delta})=0. The proof is based on the ergodic decomposition theorem. The full-support hypothesis makes it possible to select, at an arbitrarily fine spatial scale, a finite family of ergodic components whose supports cover the space up to a small error. On each of these supports we choose a point with dense positive orbit and, from sufficiently long segments of such orbits, we construct a finite system by closing each block into a cycle. In this way one obtains a finite dynamics which is GH^{0} -close to the original one. Since every continuous map on a finite space has zero topological entropy, the desired approximation follows. This result shows, in particular, that topological entropy is not stable under C^{0} -Gromov–Hausdorff perturbations, even within the class of homeomorphisms admitting invariant measures with full support. In this sense, the GH^{0} topology exhibits a markedly different behavior from that of the classical C^{0} topology. Our second main result concerns topologically GH -stable homeomorphisms. Using a periodic version of the previous construction, we prove that the combination of GH topological stability with a natural ergodic hypothesis imposes a strong periodic structure on the dynamics. Theorem 1.2 . Let f:X\to X be a topologically GH -stable homeomorphism of a compact metric space. If f admits an invariant probability measure with full support, then \overline{\Per(f)}=X. This result may be interpreted as a manifestation, in the GH^{0} context, of the general principle according to which dynamical stability forces recurrence and periodic approximation. As an immediate consequence, we obtain that minimal homeomorphisms cannot be topologically GH -stable under the hypotheses considered here. Moreover, we show that the presence of dense periodic points allows one to construct invariant measures with full support. By combining this observation with a previous result from [ 3 ] on the density of periodic points in the transitive and topologically GH -stable case, we deduce that every transitive and topologically GH -stable homeomorphism admits an invariant probability measure with full support. In particular, the general approximation theorem by zero-entropy systems also applies to this class. The conceptual novelty of the present work lies in the fact that the approximation is not derived from a global hypothesis of topological transitivity, but rather from a hypothesis of ergodic nature. More precisely, the argument shows that the existence of an invariant measure with full support suffices to reconstruct, at small scales, enough recurrence distributed throughout the whole space to produce finite approximating models. The paper is organized as follows. In Section 2 we collect the definitions and preliminary results concerning the C^{0} -Gromov–Hausdorff distance, topological entropy, invariant measures, and ergodic decomposition. In Section 3 we prove the main approximation theorem by zero-entropy systems and present a periodic version of the construction. In Section 4 we apply this scheme to the case of topologically GH -stable homeomorphisms, deduce the density of periodic points, and obtain an additional consequence for transitive topologically GH -stable homeomorphisms. Finally, in Section 5 we discuss the ergodic interpretation of the results and formulate some open questions. 2. Preliminaries In this section we fix notation and collect the concepts and results that will be used in the subsequent proofs. We restrict ourselves to the tools that are strictly necessary for the development of the paper. 2.1. The C^{0} -Gromov–Hausdorff distance Let (X,d) be a compact metric space. If A,B\subset X are nonempty subsets, the Hausdorff distance between A and B is defined by d_{H}(A,B)=\max\left\{\sup_{a\in A}d(a,B),\sup_{b\in B}d(A,b)\right\}, where d(a,B)=\inf_{b\in B}d(a,b),\qquad d(A,b)=\inf_{a\in A}d(a,b). For the basic theory of compact metric spaces and the Hausdorff distance, see [ 2 , 4 ] . We now recall the notion of a \Delta -isometry, which allows one to compare metric spaces that are not necessarily isometric. Definition 2.1 . Let (X,d^{X}) and (Y,d^{Y}) be compact metric spaces, and let \Delta>0 . A map, not necessarily continuous, i:X\to Y is called a \Delta -isometry if \max\left\{d_{H}(i(X),Y),\;\sup_{x,x^{\prime}\in X}\left|d^{Y}(i(x),i(x^{\prime}))-d^{X}(x,x^{\prime})\right|\right\}<\Delta. Based on this notion, the Gromov–Hausdorff distance between compact metric spaces is defined as follows [ 4 , 2 ] . Definition 2.2 . The Gromov–Hausdorff distance between two compact metric spaces X and Y is defined by d_{GH}(X,Y)=\inf\left\{\Delta>0:\;\exists\ \Delta\text{-isometries }i:X\to Y,\ j:Y\to X\right\}. If u,v:Z\to W are maps defined on the same compact metric space (Z,d^{Z}) and taking values in the same metric space (W,d^{W}) , we define their uniform distance by d_{C^{0}}(u,v)=\sup_{z\in Z}d^{W}\bigl(u(z),v(z)\bigr). When u and v are continuous, this coincides with the usual C^{0} distance. The following definition, introduced by Arbieto and Morales in [ 1 ] , combines the uniform distance between maps with the Gromov–Hausdorff distance between metric spaces. Definition 2.3 . Let f:X\to X and g:Y\to Y be continuous maps defined on compact metric spaces X and Y , respectively. The C^{0} -Gromov–Hausdorff distance between f and g is defined by d_{GH^{0}}(f,g)=\inf\Big\{\Delta>0:\;\exists\ \Delta\text{-isometries }i:X\to Y,\ j:Y\to X \hskip 68.28644pt\text{ such that }d_{C^{0}}(g\circ i,i\circ f)<\Delta\quad\text{and}\quad d_{C^{0}}(j\circ g,f\circ j)<\Delta\Big\}. This notion makes it possible to compare dynamical systems defined on different metric spaces and provides the natural framework for studying approximation phenomena in the dynamical Gromov–Hausdorff sense [ 1 , 7 ] . Definition 2.4 . Let \mathcal{C} be a family of continuous maps defined on compact metric spaces. We say that a continuous map f:X\to X is C^{0} -Gromov–Hausdorff approximated by elements of \mathcal{C} if there exists a sequence (g_{n})_{n\in\mathbb{N}}\subset\mathcal{C} such that \lim_{n\to\infty}d_{GH^{0}}(f,g_{n})=0. In particular, as we shall see later, the topology induced by d_{GH^{0}} allows one to approximate a dynamics defined on a compact space by dynamics defined on finite spaces. 2.2. Topological entropy Let f:X\to X be a continuous map defined on a compact metric space (X,d) . We recall the definition of topological entropy in terms of separated sets; see, for instance, [ 12 , 6 , 8 ] . Given n\in\mathbb{N} and \delta>0 , a finite subset E\subset X is said to be (n,\delta) -separated if for every x,y\in E , with x\neq y , one has \max_{0\leq j\leq n-1}d\big(f^{j}(x),f^{j}(y)\big)\geq\delta. We denote by s_{n}(f,\delta)=\sup\left\{\#E:\;E\subset X\text{ is an }(n,\delta)\text{-separated set}\right\}. Definition 2.5 . The topological entropy of f is defined by h_{\topo}(f)=\lim_{\delta\to 0}\left(\limsup_{n\to\infty}\frac{1}{n}\log s_{n}(f,\delta)\right). The only entropy property that we shall use explicitly is the following elementary fact. Lemma 2.6 . If Y is a finite set and g:Y\to Y is a map, then h_{\topo}(g)=0. Proof. Since Y is finite, for every n\in\mathbb{N} and every \delta>0 one has s_{n}(g,\delta)\leq\#Y. Therefore, \frac{1}{n}\log s_{n}(g,\delta)\leq\frac{1}{n}\log(\#Y). Taking the limit superior as n\to\infty , we obtain \limsup_{n\to\infty}\frac{1}{n}\log s_{n}(g,\delta)=0. Finally, letting \delta\to 0 , we conclude that h_{\topo}(g)=0. ∎ 2.3. Invariant measures and ergodic decomposition Let f:X\to X be a homeomorphism defined on a compact metric space X . We say that a Borel probability measure \mu on X is invariant under f if \mu(f^{-1}(E))=\mu(E) for every measurable set E\subset X . Definition 2.7 . The support of a Borel probability measure \mu on X is defined by \supp(\mu)=\left\{x\in X:\;\mu(U)>0\text{ for every open neighborhood }U\text{ of }x\right\}. We say that \mu has full support if \supp(\mu)=X. Definition 2.8 . An invariant measure \mu is said to be ergodic if for every measurable set A\subset X such that f^{-1}(A)=A , one has \mu(A)\in\{0,1\}. The basic theory of invariant measures, the Krylov–Bogolyubov theorem, Birkhoff’s ergodic theorem, and ergodic decomposition may be found in [ 12 , 9 ] . We shall use the following ergodic decomposition theorem. Theorem 2.9 (Ergodic decomposition) . Let X be a compact metric space and let f:X\to X be a homeomorphism. If \mu is an f -invariant probability measure, then there exist a measurable set X_{0}\subset X with \mu(X_{0})=1 , a measurable partition \mathcal{P} of X_{0} , and a family of probability measures \{\mu_{P}\}_{P\in\mathcal{P}} such that: (i) for \hat{\mu} -almost every class P\in\mathcal{P} , one has \mu_{P}(P)=1 ; (ii) for every measurable set E\subset X , the map P\mapsto\mu_{P}(E) is measurable; (iii) for \hat{\mu} -almost every class P\in\mathcal{P} , the measure \mu_{P} is invariant and ergodic; (iv) for every measurable set E\subset X , \mu(E)=\int\mu_{P}(E)\,d\hat{\mu}(P). The following standard fact will be fundamental in the construction of the approximating systems. We include it for completeness, since it is precisely the point at which ergodicity is translated into topological transitivity on the support. Lemma 2.10 . Let f:X\to X be a continuous map on a compact metric space, and let \mu be an invariant ergodic probability measure. Then there exists a point \omega\in\supp(\mu) such that \overline{\mathcal{O}_{f}^{+}(\omega)}=\supp(\mu), where \mathcal{O}_{f}^{+}(\omega)=\{\omega,f(\omega),f^{2}(\omega),\dots\}. In particular, the restriction f|_{\supp(\mu)} is topologically transitive. Proof. Let \{U_{n}\}_{n\in\mathbb{N}} be a countable basis of relatively open subsets of \supp(\mu) . Since each U_{n} is a nonempty relatively open subset of \supp(\mu) , one has \mu(U_{n})>0\qquad\text{for every }n\in\mathbb{N}. By Birkhoff’s ergodic theorem applied to the indicator function \mathbf{1}_{U_{n}} , for each n\in\mathbb{N} there exists a set A_{n}\subset X with \mu(A_{n})=1 such that, for every x\in A_{n} , \lim_{N\to\infty}\frac{1}{N}\sum_{j=0}^{N-1}\mathbf{1}_{U_{n}}(f^{j}(x))=\int\mathbf{1}_{U_{n}}\,d\mu=\mu(U_{n})>0. Consequently, for every x\in A_{n} there exists some integer k\geq 0 such that f^{k}(x)\in U_{n} . In fact, the positive orbit of x enters U_{n} infinitely many times. Define A=\bigcap_{n\in\mathbb{N}}A_{n} . Then \mu(A)=1 . Since moreover \mu(\supp(\mu))=1 , there exists \omega\in A\cap\supp(\mu). Now let U\subset\supp(\mu) be any nonempty relatively open set. Then there exists n\in\mathbb{N} such that U_{n}\subset U\qquad\text{and}\qquad U_{n}\neq\emptyset. Since \omega\in A_{n} , there exists k\geq 0 such that f^{k}(\omega)\in U_{n}\subset U . Therefore, the positive orbit of \omega intersects every nonempty relatively open subset of \supp(\mu) . This implies that \overline{\mathcal{O}_{f}^{+}(\omega)}=\supp(\mu). The last assertion is immediate, since the existence of a dense positive orbit in \supp(\mu) means precisely that f|_{\supp(\mu)} is topologically transitive. ∎ 3. Approximation by zero-entropy systems In this section we prove the main result of the paper. The goal is to show that the existence of an invariant probability measure with full support makes it possible to construct, at every prescribed scale, a finite dynamical system that approximates the original homeomorphism in the C^{0} -Gromov–Hausdorff topology. Since every dynamics defined on a finite space has zero topological entropy, the desired approximation follows. Theorem 3.1 . Let f:X\to X be a homeomorphism of a compact metric space. If f admits an invariant probability measure with full support, then for every \delta>0 there exist a compact metric space Y_{\delta} and a homeomorphism g_{\delta}:Y_{\delta}\to Y_{\delta} such that d_{GH^{0}}(f,g_{\delta})<\delta\qquad\text{and}\qquad h_{\topo}(g_{\delta})=0. Proof. Let \mu be an f -invariant probability measure with full support. Fix \delta>0 . Since f is uniformly continuous and X is compact, there exists \beta>0 such that (3.1) \displaystyle d(x,y)<\beta\quad\Longrightarrow\quad d\bigl(f(x),f(y)\bigr)<\frac{\delta}{3}\qquad\text{for all }x,y\in X. Take 0<\alpha<\min\left\{\frac{\delta}{9},\frac{\beta}{2}\right\} . By compactness of X , there exists a finite set \{x_{1},\dots,x_{r}\}\subset X such that d_{H}(\{x_{1},\dots,x_{r}\},X)<\alpha. Let \{\mu_{P}\}_{P\in\mathcal{P}} be the ergodic decomposition of \mu as in Theorem 2.9 . Since \supp(\mu)=X , one has \mu\bigl(B(x_{i},\alpha)\bigr)>0\qquad\text{for every }i=1,\dots,r. Applying Theorem 2.9 , for each i\in\{1,\dots,r\} we obtain \mu\bigl(B(x_{i},\alpha)\bigr)=\int\mu_{P}\bigl(B(x_{i},\alpha)\bigr)\,d\hat{\mu}(P)>0. Since \mu_{P} is ergodic for \hat{\mu} -almost every class P\in\mathcal{P} , there exists a class P_{i}\in\mathcal{P} such that (3.2) \displaystyle\mu_{P_{i}}\bigl(B(x_{i},\alpha)\bigr)>0, and \mu_{P_{i}} is ergodic. Denote this measure by \mu_{i} . From ( 3.2 ), for every i=1,\ldots,r , one has (3.3) \displaystyle B(x_{i},\alpha)\cap\supp(\mu_{i})\neq\emptyset. By Lemma 2.10 , for each i there exists a point z_{i}\in\supp(\mu_{i}) such that \overline{\mathcal{O}_{f}^{+}(z_{i})}=\supp(\mu_{i}). Since B(x_{i},\alpha)\cap\supp(\mu_{i})\neq\emptyset and the positive orbit of z_{i} is dense in \supp(\mu_{i}) , there exists an integer \ell_{i}\geq 0 such that f^{\ell_{i}}(z_{i})\in B(x_{i},\alpha). Define a_{i}:=f^{\ell_{i}}(z_{i}). Since f^{\ell_{i}} is a homeomorphism and \supp(\mu_{i}) is f -invariant, one has (3.4) \overline{\mathcal{O}_{f}^{+}(a_{i})}=\overline{f^{\ell_{i}}\bigl(\mathcal{O}_{f}^{+}(z_{i})\bigr)}=f^{\ell_{i}}\bigl(\overline{\mathcal{O}_{f}^{+}(z_{i})}\bigr)=f^{\ell_{i}}\bigl(\supp(\mu_{i})\bigr)=\supp(\mu_{i}). In particular, a_{i}\in\supp(\mu_{i})\cap B(x_{i},\alpha). Since \supp(\mu_{i}) is f -invariant and a_{i}\in\supp(\mu_{i}) , one has f^{-1}(a_{i})\in\supp(\mu_{i}). By ( 3.4 ), since f^{-1}(a_{i})\in\supp(\mu_{i}) , there exists an integer N_{i}\geq 0 such that d\bigl(f^{N_{i}}(a_{i}),f^{-1}(a_{i})\bigr)<\beta. Applying f and using the choice of \beta in ( 3.1 ), we obtain d\bigl(f^{N_{i}+1}(a_{i}),a_{i}\bigr)=d\bigl(f(f^{N_{i}}(a_{i})),f(f^{-1}(a_{i}))\bigr)<\frac{\delta}{3}. We now define the finite set Y_{\delta}=\{(i,k):\,1\leq i\leq r,\ 0\leq k\leq N_{i}\}. On Y_{\delta} we consider the map q:Y_{\delta}\to X,\qquad q(i,k)=f^{k}(a_{i}). To avoid possible identifications between distinct pairs (i,k) and (j,\ell) having the same image in X , we equip Y_{\delta} with the metric \rho(u,v)=d\bigl(q(u),q(v)\bigr)+\alpha\,\eta(u,v),\qquad u,v\in Y_{\delta}, where \eta(u,v)=\begin{cases}0,&u=v,\\ 1,&u\neq v.\end{cases} Since d(q(u),q(v)) defines a pseudometric on Y_{\delta} and \alpha\,\eta(u,v) is a metric, the function \rho is a metric on Y_{\delta} . In particular, (Y_{\delta},\rho) is a compact metric space. We now define g_{\delta}:Y_{\delta}\to Y_{\delta} by g_{\delta}(i,k)=(i,k+1),\qquad 0\leq k\leq N_{i}-1, and g_{\delta}(i,N_{i})=(i,0). Clearly, g_{\delta} is a permutation of Y_{\delta} ; since Y_{\delta} is finite, it follows that g_{\delta} is a homeomorphism. Moreover, for every u=(i,k)\in Y_{\delta} one has (3.5) \displaystyle d\bigl(q(g_{\delta}(u)),f(q(u))\bigr)<\frac{\delta}{3}. Indeed, if 0\leq k\leq N_{i}-1 , then q(g_{\delta}(i,k))=q(i,k+1)=f^{k+1}(a_{i})=f(q(i,k)), whereas if k=N_{i} , then d\bigl(q(g_{\delta}(i,N_{i})),f(q(i,N_{i}))\bigr)=d\bigl(a_{i},f^{N_{i}+1}(a_{i})\bigr)<\frac{\delta}{3}. On the other hand, since \{a_{1},\dots,a_{r}\}\subset q(Y_{\delta}) , we have d_{H}(q(Y_{\delta}),X)\leq d_{H}(\{a_{1},\dots,a_{r}\},X). Moreover, d_{H}(\{a_{1},\dots,a_{r}\},X)\leq d_{H}(\{a_{1},\dots,a_{r}\},\{x_{1},\dots,x_{r}\})+d_{H}(\{x_{1},\dots,x_{r}\},X)<2\alpha, since a_{i}\in B(x_{i},\alpha) for every i . Therefore, (3.6) d_{H}(q(Y_{\delta}),X)<2\alpha<\delta. Furthermore, if u\neq v , then \rho(u,v)=d\bigl(q(u),q(v)\bigr)+\alpha. Therefore, for every u,v\in Y_{\delta} one has \bigl|d\bigl(q(u),q(v)\bigr)-\rho(u,v)\bigr|\leq\alpha . Hence, (3.7) \displaystyle\sup_{u,v\in Y_{\delta}}\bigl|d\bigl(q(u),q(v)\bigr)-\rho(u,v)\bigr|\leq\alpha<\delta. From ( 3.6 ) and ( 3.7 ) we conclude that q:(Y_{\delta},\rho)\to X is a \delta -isometry. Moreover, d_{C^{0}}(q\circ g_{\delta},f\circ q)<\frac{\delta}{3}<\delta. Since d_{H}(q(Y_{\delta}),X)<2\alpha , we construct a map j:X\to Y_{\delta} as follows: for each x\in X , choose a point j(x)\in Y_{\delta} such that (3.8) \displaystyle d\bigl(x,q(j(x))\bigr)<2\alpha. We claim that j is a \delta -isometry. Given x,x^{\prime}\in X , one has \bigl|d\bigl(q(j(x)),q(j(x^{\prime}))\bigr)-d(x,x^{\prime})\bigr|\leq d\bigl(x,q(j(x))\bigr)+d\bigl(x^{\prime},q(j(x^{\prime}))\bigr)<4\alpha. Moreover, by the definition of \rho , \bigl|\rho(j(x),j(x^{\prime}))-d\bigl(q(j(x)),q(j(x^{\prime}))\bigr)\bigr|\leq\alpha. Therefore, (3.9) \displaystyle\bigl|\rho(j(x),j(x^{\prime}))-d(x,x^{\prime})\bigr|<5\alpha<\delta, since \alpha<\delta/9 . On the other hand, for every y\in Y_{\delta} , \rho\bigl(j(q(y)),y\bigr)\leq d\bigl(q(j(q(y))),q(y)\bigr)+\alpha<3\alpha. Therefore, (3.10) \displaystyle d_{H}\bigl(j(X),Y_{\delta}\bigr)<3\alpha<\delta. From ( 3.9 ) and ( 3.10 ), we conclude that j:X\to(Y_{\delta},\rho) is a \delta -isometry. We now verify that d_{C^{0}}(j\circ f,g_{\delta}\circ j)<\delta . Indeed, let x\in X . By the definition of \rho , one has \rho\bigl(j(f(x)),g_{\delta}(j(x))\bigr)\leq d\bigl(q(j(f(x))),q(g_{\delta}(j(x)))\bigr)+\alpha. Moreover, \displaystyle d\bigl(q(j(f(x))),q(g_{\delta}(j(x)))\bigr) \displaystyle\leq d\bigl(q(j(f(x))),f(x)\bigr)+d\bigl(f(x),f(q(j(x)))\bigr) \displaystyle+d\bigl(f(q(j(x))),q(g_{\delta}(j(x)))\bigr). By the definition of j given in ( 3.8 ), the first term on the right-hand side is smaller than 2\alpha . For the second term, since d\bigl(x,q(j(x))\bigr)<2\alpha<\beta, the choice of \beta in ( 3.1 ) implies that d\bigl(f(x),f(q(j(x)))\bigr)<\frac{\delta}{3} . Finally, by ( 3.5 ), the third term is smaller than \frac{\delta}{3} . Consequently, \rho\bigl(j(f(x)),g_{\delta}(j(x))\bigr)<2\alpha+\frac{\delta}{3}+\frac{\delta}{3}+\alpha=3\alpha+\frac{2\delta}{3}<3\left(\frac{\delta}{9}\right)+\frac{2\delta}{3}=\delta. Therefore, d_{C^{0}}(j\circ f,g_{\delta}\circ j)<\delta. We have constructed \delta -isometries q:(Y_{\delta},\rho)\to X\qquad\text{and}\qquad j:X\to(Y_{\delta},\rho) such that d_{C^{0}}(q\circ g_{\delta},f\circ q)<\delta\qquad\text{and}\qquad d_{C^{0}}(j\circ f,g_{\delta}\circ j)<\delta. By the definition of the distance d_{GH^{0}} , it follows that d_{GH^{0}}(f,g_{\delta})<\delta. Finally, since Y_{\delta} is finite, Lemma 2.6 implies that h_{\topo}(g_{\delta})=0. This completes the proof. ∎ The previous construction also has an additional property that will be useful in the next section. Proposition 3.2 . Under the assumptions of Theorem 3.1 , the approximating homeomorphism g_{\delta} may be chosen in such a way that every point of Y_{\delta} is periodic. Proof. The construction carried out in the proof of Theorem 3.1 already has this property. Indeed, for each i=1,\dots,r , the subset \{(i,0),(i,1),\dots,(i,N_{i})\} is invariant under g_{\delta} , and the restriction of g_{\delta} to this set is a finite periodic orbit. Therefore, every point of Y_{\delta} is periodic for g_{\delta} . ∎ Remark 3.3 . The previous construction yields an approximating homeomorphism whose phase space is a finite union of periodic orbits. In particular, every point of Y_{\delta} is periodic. This property will be sufficient for the subsequent applications. 4. Consequences for topologically GH -stable systems In this section we analyze the dynamical consequences of the approximation constructed in the previous section when the additional hypothesis of GH topological stability is imposed. The central idea is that if a periodic system is sufficiently close, in the C^{0} -Gromov–Hausdorff topology, to a topologically GH -stable homeomorphism, then the semi-conjugacy provided by stability sends periodic points of the approximating system to periodic points of the original one. By combining this principle with the approximation by finite systems obtained in Theorem 3.1 , one deduces the density of periodic points. 4.1. GH topological stability We recall the notion of GH topological stability introduced by Arbieto and Morales in [ 1 ] . This definition extends the classical notion of topological stability to the context in which the perturbed dynamics may be defined on different compact metric spaces. Definition 4.1 . Let f:X\to X be a homeomorphism of a compact metric space. We say that f is topologically GH -stable if for every \varepsilon>0 there exists \eta>0 such that, for every homeomorphism g:Y\to Y of a compact metric space satisfying d_{GH^{0}}(f,g)<\eta, there exists a continuous \varepsilon -isometry h:Y\to X such that f\circ h=h\circ g. The map h appearing in the previous definition provides a semi-conjugacy between g and f , and will be the fundamental tool for transferring periodicity from the approximating systems to the original system. 4.2. Density of periodic points We are now in a position to obtain the main application of the present work. Theorem 3.1 , together with Proposition 3.2 , produces periodic systems arbitrarily close to f in the GH^{0} topology. The GH topological stability then allows one to transfer periodicity from these approximating systems to the original dynamics. Theorem 4.2 . Let f:X\to X be a topologically GH -stable homeomorphism of a compact metric space. If f admits an invariant probability measure with full support, then \overline{\Per(f)}=X. Proof. Let \varepsilon>0 . Since f is topologically GH -stable, there exists \eta>0 such that for every homeomorphism g:Y\to Y with d_{GH^{0}}(f,g)<\eta, there exists a continuous \varepsilon -isometry h:Y\to X satisfying f\circ h=h\circ g. We now apply Proposition 3.2 at scale \eta . We obtain a compact metric space Y_{\eta} and a homeomorphism g_{\eta}:Y_{\eta}\to Y_{\eta} such that d_{GH^{0}}(f,g_{\eta})<\eta and, moreover, every point of Y_{\eta} is periodic for g_{\eta} . By the GH topological stability of f , there then exists a continuous \varepsilon -isometry h:Y_{\eta}\to X such that f\circ h=h\circ g_{\eta}. Let y\in Y_{\eta} . Since y is periodic for g_{\eta} , there exists an integer m\geq 1 such that g_{\eta}^{m}(y)=y. Applying h and using the semi-conjugacy relation, we obtain f^{m}(h(y))=h(g_{\eta}^{m}(y))=h(y). Therefore, h(y)\in\Per(f). Since y\in Y_{\eta} was arbitrary, we conclude that h(Y_{\eta})\subset\Per(f). On the other hand, since h is an \varepsilon -isometry, in particular one has d_{H}(h(Y_{\eta}),X)<\varepsilon. It follows that for every x\in X there exists p\in h(Y_{\eta})\subset\Per(f) such that d(x,p)<\varepsilon. Since \varepsilon>0 is arbitrary, we conclude that \overline{\Per(f)}=X. ∎ The previous theorem shows that the combination of GH topological stability with the existence of an invariant measure having full support imposes a strong periodic structure on the dynamics. In particular, on infinite compact spaces, this condition is incompatible with minimality. 4.3. A corollary for minimal homeomorphisms As an immediate consequence of Theorem 4.2 , we obtain the following obstruction to GH topological stability in the minimal setting. Corollary 4.3 . Let f:X\to X be a minimal homeomorphism of an infinite compact metric space. Then f cannot be topologically GH -stable. Proof. Every continuous homeomorphism of a compact metric space admits an invariant probability measure [ 12 ] . Let \mu be one such measure. Since \supp(\mu) is a nonempty compact f -invariant set, the minimality of f necessarily implies that \supp(\mu)=X. Suppose now, by contradiction, that f is topologically GH -stable. Then, by Theorem 4.2 , \overline{\Per(f)}=X. But a minimal homeomorphism on an infinite compact space cannot have periodic points. Indeed, if p were periodic, then its orbit would be finite, closed, and invariant; by minimality, that orbit would have to coincide with the whole space X , contradicting the fact that X is infinite. This contradiction proves that f cannot be topologically GH -stable. ∎ 4.4. Invariant measures with full support from dense periodic points The following lemma will be useful for relating our results to the transitive case previously treated in [ 3 ] . The statement is completely general and shows that the density of periodic points is sufficient to produce an invariant measure with full support. Lemma 4.4 . Let f:X\to X be a homeomorphism of a compact metric space. If \overline{\Per(f)}=X, then f admits an invariant probability measure with full support. Proof. Since X is compact metric, it is in particular separable. Because \overline{\Per(f)}=X, there exists a dense sequence \{p_{n}\}_{n\in\mathbb{N}}\subset\Per(f). For each n\in\mathbb{N} , let \pi_{n}\geq 1 denote the period of p_{n} , and define the periodic measure associated with its orbit by \mu_{n}=\frac{1}{\pi_{n}}\sum_{k=0}^{\pi_{n}-1}\delta_{f^{k}(p_{n})}. Each \mu_{n} is a Borel probability measure invariant under f . Now consider the measure \mu=\sum_{n=1}^{\infty}2^{-n}\mu_{n}. Since \sum_{n=1}^{\infty}2^{-n}=1, the measure \mu is a Borel probability measure. Moreover, by linearity of the pushforward operation and the invariance of each \mu_{n} , we obtain f_{*}\mu=\sum_{n=1}^{\infty}2^{-n}f_{*}\mu_{n}=\sum_{n=1}^{\infty}2^{-n}\mu_{n}=\mu. Therefore, \mu is invariant under f . Finally, let us show that \supp(\mu)=X . Let U\subset X be a nonempty open set. Since the sequence \{p_{n}\}_{n\in\mathbb{N}} is dense, there exists n_{0}\in\mathbb{N} such that p_{n_{0}}\in U. Since p_{n_{0}} belongs to the periodic orbit supporting \mu_{n_{0}} , it follows that \mu_{n_{0}}(U)>0. Consequently, \mu(U)\geq 2^{-n_{0}}\mu_{n_{0}}(U)>0. We have proved that every nonempty open set has positive measure with respect to \mu , which is equivalent to \supp(\mu)=X. ∎ The usefulness of the previous lemma is that it converts density-of-periodic-points results into statements of ergodic character. Proposition 4.5 . Let f:X\to X be a topologically transitive and topologically GH -stable homeomorphism of a compact metric space. Then f admits an invariant probability measure with full support. Proof. By [ 3 , Theorem 3.1.4] , one has \overline{\Per(f)}=X. Applying Lemma 4.4 , we conclude that f admits an invariant probability measure with full support. ∎ The previous proposition shows that, within the class of transitive and topologically GH -stable homeomorphisms, the ergodic hypothesis of Theorem 3.1 is automatic. Corollary 4.6 . Let f:X\to X be a topologically transitive and topologically GH -stable homeomorphism of a compact metric space. Then for every \delta>0 there exist a compact metric space Y_{\delta} and a homeomorphism g_{\delta}:Y_{\delta}\to Y_{\delta} such that d_{GH^{0}}(f,g_{\delta})<\delta\qquad\text{and}\qquad h_{\topo}(g_{\delta})=0. Proof. By Proposition 4.5 , the homeomorphism f admits an invariant probability measure with full support. The result then follows immediately from Theorem 3.1 . ∎ 5. Ergodic interpretation and final remarks Theorem 3.1 shows that the existence of an invariant measure with full support is sufficient to construct finite approximations of the dynamics in the C^{0} -Gromov–Hausdorff topology. In this sense, the result may be interpreted as a principle of global discretization: although the system need not be transitive, the ergodic information distributed throughout the whole space makes it possible to produce finite approximating models. From the entropy viewpoint, it also follows that topological entropy is not stable under C^{0} -Gromov–Hausdorff perturbations within the class under consideration, since every dynamics admitting an invariant measure with full support can be approximated by systems with zero topological entropy. On the other hand, Theorem 4.2 shows that, under GH topological stability, this periodic approximation translates into density of periodic points for the original dynamics. This highlights a close relationship between stability, finite approximation, and periodic recurrence. Finally, Proposition 4.5 shows that, in the class of transitive and topologically GH -stable homeomorphisms, the hypothesis of the existence of an invariant measure with full support arises naturally. In this way, the ergodic approach developed here recovers and extends previous results obtained in the transitive setting. We conclude with a few questions that arise naturally from this work. Question 5.1 . Can Theorem 3.1 be extended to the case of homeomorphisms admitting invariant measures whose support is not full? Question 5.2 . What can be said about the behavior of measures of maximal entropy under C^{0} -Gromov–Hausdorff perturbations? Question 5.3 . Is it possible to formulate a more directly ergodic version of a closing lemma in the GH^{0} topology?