Gromov-Hausdorff-Weak Topology

• Gromov–Hausdorff–Weak topology is a framework for studying convergence in measured metric spaces by jointly controlling the convergence of supports and measures.
• It utilizes a localization-by-balls approach with an explicit GHv metric, ensuring detailed analysis for both compact and non-compact spaces.
• The topology underpins applications such as scaling limits of random trees and invariance principles for stochastic processes through rigorous probabilistic control.

The Gromov-Hausdorff-weak topology, also known as the Gromov–Hausdorff–vague or local Gromov–Hausdorff–Prohorov topology, is a fundamental structure for studying convergence of measured metric spaces that may be non-compact or carry stochastic processes. It rigorously merges geometric and measure-theoretic convergence by encoding joint convergence of supports and measures at all scales and is crucial for analyzing scaling limits of random spaces, random trees, and associated processes.

1. Metric Measure Space Framework

A metric boundedly finite measure space consists of a quadruple $(X, r, \rho, \mu)$ where $(X, r)$ is a complete, separable metric space, $\rho \in X$ is a distinguished root, and $\mu$ is a Borel measure finite on every bounded subset. Spaces are considered equivalent if there exists an isometry between the unions of the supports and roots that preserves measures and roots. Key subspaces include spaces whose metric is Heine–Borel ($\XHB$; all closed bounded sets are compact) and compact metric measure spaces ($\Xc$; compact supports and finite measures) (Athreya et al., 2014).

Isometry classes of such quadruples are the natural objects for topological study under Gromov-Hausdorff-weak convergence.

2. Definition and Metric of the Gromov-Hausdorff-Weak Topology

Consider two rooted boundedly-compact measured metric spaces, $\mathcal{X}_1 = (X_1, d_1, \rho_1, \mu_1)$ and $\mathcal{X}_2 = (X_2, d_2, \rho_2, \mu_2)$. For any radius $r>0$, let $\mathcal{X}_i^{(r)}$ denote the restriction to the closed ball $D(\rho_i, r)$ centered at the root, with inherited metric and measure.

The Gromov–Hausdorff–weak (GHv) distance is defined as: $$d_{\mathrm{GHv}} \bigl( \mathcal{X}_1, \mathcal{X}_2 \bigr) = \int_0^\infty e^{-r} \left(1 \wedge d_{\mathrm{sGHP}}(\mathcal{X}_1^{(r)}, \mathcal{X}_2^{(r)}) \right) \,dr,$$ where $d_{\mathrm{sGHP}}$ is a complete variant of the Gromov–Hausdorff–Prohorov metric for compact, pointed measured spaces. On compact spaces, $d_{\mathrm{GHP}}$ equates to the infimum, over all isometric embeddings into a common metric space, of the maximum among Hausdorff distance of supports, Prohorov distance of measures, and root distances (Athreya et al., 2014, Khezeli, 2018, Noda, 2024).

The topology is characterized by the convergence: for almost every $R > 0$, the restricted balls $\mathcal{X}_n^{(R)}$ converge to $\mathcal{X}^{(R)}$ in the Gromov–Hausdorff–Prohorov topology.

3. Relationship to Gromov-Weak and Gromov–Hausdorff–Prohorov Topologies

The Gromov–weak topology, induced by the Gromov–Prohorov metric and equivalent to the Gromov box metric, only requires weak convergence of measures associated to random finite samples (“sampling convergence”). It does not control the convergence of supports; “mass” may collapse onto vanishing subspaces (Löhr, 2011).

In contrast, the Gromov–Hausdorff–weak topology simultaneously requires:

• Weak convergence of the push-forward measures in a common ambient space,
• Hausdorff convergence of the supports, thereby prohibiting escape of mass to outliers and ensuring both geometric and measure-theoretic control (Athreya et al., 2014, Khezeli, 2018).

The Gromov–Hausdorff–Prohorov metric metrizes the Gromov–Hausdorff–weak topology for compact measured metric spaces. For non-compact, pointed, boundedly-compact spaces, localizations by balls yield the Gromov–Hausdorff–weak topology (Noda, 2024).

4. Lower Mass-Bound Property and Characterizations

The discrepancy between Gromov–vague and Gromov–Hausdorff–vague convergence is captured exactly by the so-called lower mass-bound property: $$\mathrm{lm}_{R,\delta}(\mathcal{X}) = \inf\left\{ \mu(B(x, \delta)) : x \in \operatorname{supp}(\mu) \cap B(\rho, R) \right\}$$ for local bounds and

$$\mathrm{gml}_\delta(\mathcal{X}) = \inf_{R>0}\mathrm{lm}_{R,\delta}(\mathcal{X})$$

for global bounds.

A family $\mathcal{K}$ satisfies the property if for every $R>\delta>0$, $$\inf_{\mathcal{X} \in \mathcal{K}} \mathrm{lm}_{R,\delta}(\mathcal{X}) > 0$$. The main theorem asserts: Gromov–weak convergence plus the global lower mass-bound property is equivalent to Gromov–Hausdorff–weak convergence. Locally, Gromov–vague convergence plus a local lower mass-bound property equates to Gromov–Hausdorff–vague convergence (Athreya et al., 2014).

This property ensures tightness of mass and precludes “loss to infinitesimal outliers,” facilitating strong topological and probabilistic results.

5. Topological Properties: Polishness and Completeness

The GHv topology yields the following structural results:

• The space of isometry classes of compact measured metric spaces, under the Gromov–Hausdorff–weak metric, is Polish (complete, separable).
• The subspace of pointed, locally compact, Heine–Borel spaces, under the local version, is also Polish, with explicit (sketch) metrics.
• The coarse Gromov–vague topology is Polish on the space of boundedly-finite measure spaces, while its Heine–Borel and compact subspaces are Lusin but not Polish (Athreya et al., 2014, Noda, 2024).
• The GHv-topology on measure-marked spaces or other Gromov–Hausdorff-type topologies (e.g., with point-, closed-set-, or path-marks) fits into the general complete and separable “functor” framework (Khezeli, 2018, Noda, 2024).

This ensures that standard probabilistic tools—treating random metric measure spaces, tightness, and weak convergence—are applicable.

6. Applications and Examples

Convergence in the Gromov–Hausdorff–weak topology underpins invariance principles for random processes on spaces with complex geometry. For example:

• The scaling limit of uniform random labeled trees, when rescales and endowed with vertex measures, converges to the Brownian continuum random tree in GHv topology (Khezeli, 2018).
• For Galton–Watson trees conditioned to survive, after appropriate scaling and encoding by excursions, convergence in GHv yields the continuum Kallenberg–Kesten tree, providing a rigorous setting for pathwise invariance principles for random walks on such trees (Athreya et al., 2014).
• The “glue map” (coding excursions to measured real trees) is continuous in the GHv topology (Athreya et al., 2014, Löhr, 2011).

The general functorial framework in (Noda, 2024) extends to more complex structures: metric spaces with marked points, closed sets, stochastic processes, random fields, or laws thereof, all under a unifying GHv-style topology.

7. Generalizations and Unified Frameworks

Recent research provides comprehensive frameworks for metrizing GH-type topologies on extended classes of spaces. For instance, the construction of a general metric on rooted boundedly-compact spaces marked by various functorial structures, and the “localization by balls” approach, guarantee the applicability of the Gromov–Hausdorff–weak topology far beyond the classical case.

This includes, for example:

• Spaces with finite point or closed-set marks,
• Spaces marked by stochastic processes or random fields,
• Tightness and completeness criteria that are easily checked via local structure (Khezeli, 2018, Noda, 2024).

A plausible implication is that the GHv topology will continue to serve as the standard for rigorous convergence analysis in geometric probability, stochastic geometry, and related areas, especially in contexts where both the geometry and distributional structure of large random objects are significant.