Benjamini–Schramm Convergence

• Benjamini–Schramm convergence is a local convergence framework that analyzes how neighborhoods around random basepoints in large discrete or geometric structures become statistically independent of the global structure.
• It enables the study of spectral convergence, normalized invariants, and eigenfunction behavior across diverse settings such as graphs, Riemannian manifolds, and quantum graphs.
• The framework unifies probabilistic, combinatorial, and analytic methods to reveal universality phenomena in sparse structures with significant applications in random matrix theory and network analysis.

Benjamini–Schramm convergence is a general principle of local convergence for large discrete or geometric objects, typically formulated through the random-rooted local topology. It asserts that for sequences of graphs, Riemannian manifolds, homogeneous spaces, or combinatorial structures, the distribution of a bounded-radius neighborhood around a uniformly random basepoint becomes asymptotically independent of the global structure, converging to a unimodular random limit. The notion was first introduced in the setting of bounded-degree graphs by Benjamini and Schramm, and has since been extended to weighted graphs, quantum graphs, simplicial complexes, Riemannian manifolds, algebraic quotients, and beyond.

1. Origins and Core Definition

The classical definition for finite, bounded-degree graphs $G_n$ is: $(G_n)$ is Benjamini–Schramm (BS) convergent if, for every $r \ge 1$ and every finite rooted graph $(H,o)$, the probability that a random rooted $r$-ball in $G_n$ is isomorphic to $(H,o)$ tends to a limit as $n \to \infty$ (Abért et al., 2012, Abért et al., 2014, Bordenave, 11 Oct 2025, Georgakopoulos et al., 2015). This topology is defined via the local neighborhood metric on the space of isomorphism classes of locally finite rooted graphs, making the space compact and metrizable.

The root-based formulation extends naturally to other categories:

• Riemannian manifolds: $(M_n)$ BS-converges to $(X,g,o)$ if the empirical measures assigning mass $1/\text{Vol}(M_n)$ to each basepoint converge weakly to the Dirac mass on $(X,o)$, and for every $R>0$, the fraction of points with injectivity radius $<R$ in $M_n$ vanishes as $n\to\infty$ (Abert et al., 2018, Frączyk et al., 2 Feb 2026, Abert et al., 2018, Masson et al., 2016).
• Homogeneous spaces/modular quotients: Quotient spaces $H/\Gamma_n$ BS-converge to $H$ if for every radius $r>0$, the measure of points with injectivity radius $<r$ tends to zero (Mohammadi et al., 2021, Deitmar, 2024).

Key equivalent formulations:

• Weak convergence of empirical measures of rooted balls (in the graph, manifold, or metric space).
• Vanishing volume of "thin parts" (i.e., points with small injectivity radius) (Frączyk et al., 2 Feb 2026, Abert et al., 2018).
• Convergence of the associated "invariant random subgroup" measures to the trivial subgroup (Raimbault, 2022).

2. Main Examples and Limiting Objects

• Graphs: Towers of $d$-regular graphs with girth $\to\infty$ converge locally to the infinite $d$-regular tree $\mathbb{T}_d$ (Bordenave, 11 Oct 2025, Abért et al., 2014). Sparse Erdős–Rényi graphs $G(n,d/n)$ converge to a (Poisson) Galton–Watson tree (Adhikari et al., 5 Sep 2025, Bordenave, 11 Oct 2025). Uniform random trees converge to critical (or conditioned) Galton–Watson trees (Georgakopoulos et al., 2015).
• Manifolds and symmetric spaces: Towers of coverings of compact hyperbolic surfaces with injectivity radius tending to infinity BS-converge to the hyperbolic plane (Masson et al., 2016, Monk, 2020). Sequences of compact Riemannian symmetric space quotients converge to their universal cover when the thin part becomes negligible (Frączyk et al., 2 Feb 2026, Abert et al., 2018).
• Random translation surfaces: High-genus surfaces with area proportional to genus converge to the Poisson translation plane (Bowen et al., 7 Jan 2025).
• Quantum graphs: Sequences with bounded degree, lengths, and potentials have subsequential BS-limits; empirical spectral measures converge to the expected spectral measure of the limiting (possibly infinite) rooted quantum graph (Anantharaman et al., 2020).
• Simplicial complexes: Equivariant Benjamini–Schramm limits describe sofic random rooted simplicial $G$-complexes and control $\ell^2$-multiplicities (Kionke et al., 2019).

3. Spectral and Limit Theorems

BS-convergence dictates spectral properties in the large-scale limit. Central results include:

• Spectral convergence: The empirical spectral measures of normalized adjacency/Laplacian/local operators converge to the expected spectral measure of the limit (Bordenave, 11 Oct 2025, Andraus, 2017, Anantharaman et al., 2020, Csikvári et al., 2012, Masson et al., 2016, Deitmar, 2024, Deitmar, 2018). For locally symmetric spaces, BS-convergence is equivalent to Plancherel (weak spectral) convergence under uniform discreteness (Deitmar, 2018, Deitmar, 2024).
• Root moments and graph polynomials: For multiplicative graph polynomials of bounded exponential type (chromatic, Tutte, matching), the normalized root moments converge along any BS-convergent sequence (Csikvári et al., 2012, Abért et al., 2012, Abért et al., 2014). In particular, the normalized logarithm of the partition function/free energy converges to a harmonic function outside a bounded support disk.
• Eigenfunction statistics: In the manifold/Selberg trace formula setting, BS-convergence leads to convergence of spectral densities and quantum ergodicity in a fixed spectral window (Masson et al., 2016, Abert et al., 2018, Monk, 2020). The high-genus Weyl law and corresponding eigenfunction delocalization are consequences of this limit.

4. Homological and Cohomological Limits

BS-convergence provides a framework for the limiting behavior of global topological invariants:

• Volume-normalized Betti numbers: For compact or finite-volume Riemannian manifolds of nonpositive curvature, as long as thin parts are negligible, normalized Betti numbers converge, often to the $L^2$-Betti numbers of the universal cover (Abert et al., 2018, Frączyk et al., 2 Feb 2026, Cerbo et al., 2019). In higher rank, except for certain exceptional cases (notably $X=\mathbb{H}^3$), the only surviving Betti number in the large-volume limit is the middle degree, matching the $L^2$ invariant.
• Equivariant $\ell^2$-multiplicities: In the setting of random rooted simplicial $G$-complexes, equivariant BS-convergence ensures continuity of $\ell^2$-multiplicities for irreducible representations of $G$ (Kionke et al., 2019).

5. Probabilistic and Combinatorial Frameworks

BS-convergence is fundamental for the study of large sparse graphs, random structures, and statistical mechanics:

• Estimability: A function $p(G)$ (graph parameter) is estimable if it has a limit along any BS-convergent sequence. Parameters such as the normalized logarithm of the number of matchings, the matching ratio, and the independence ratio for claw-free graphs are estimable (Abért et al., 2014, Bordenave, 11 Oct 2025). However, for perfect matchings in $d$-regular bipartite graphs, the normalized logarithm fails to be estimable due to concentration phenomena.
• Random matrix theory: BS-convergence of weighted path graphs underlying tridiagonal random matrices leads to the Wigner semicircle and Marchenko–Pastur laws as limits of the spectral density (Andraus, 2017).
• Percolation and random processes: Convergence in BS sense allows for the understanding of local statistics and percolation thresholds in sparse structures (e.g. in unimodular random trees and hypergraphs) (Bordenave, 11 Oct 2025, Adhikari et al., 5 Sep 2025).
• Hypergraphs and higher-dimensional analogs: In random uniform hypergraph models, the BS limit is a block Galton–Watson tree with explicit spectral and combinatorial consequences (Adhikari et al., 5 Sep 2025).

6. Relationships, Extensions, and Structural Theorems

• Spectral vs. BS convergence equivalence: Under uniform discreteness, BS-convergence and weak spectral convergence of normalized traces of local, finite-propagation, and invariant operators are equivalent in homogeneous and non-homogeneous spaces (Deitmar, 2024, Deitmar, 2018). Absent uniform discreteness, BS-convergence is stronger; there can be spectrally convergent sequences failing BS-convergence.
• Compactification: Spaces of rooted graphs/complexes are compact if rooted balls (with structure, degree, or data bounds) admit a uniform metric; infinite-degree vertices or singularities are accommodated in generalized settings (Georgakopoulos et al., 2015, Anantharaman et al., 2020, Kionke et al., 2019).
• Equidistribution and quantum ergodicity: For sequences of manifold covers (e.g., congruence towers), BS-convergence controls pointwise equidistribution of Laplace eigenfunctions, with analogs to Berry's random wave conjecture at the local scale (Masson et al., 2016, Abert et al., 2018).
• Homogeneous spaces and IRS: Weak limits of invariant random subgroups (IRS) align with the BS-convergent subsequential law, encoding local geometric and group-theoretic information (Raimbault, 2022, Mohammadi et al., 2021, Frączyk et al., 2 Feb 2026).

7. Applications, Open Questions, and Future Directions

• Finiteness results: The negligible volume of thin parts in the BS limit underpins finiteness theorems for arithmetic reflection groups and congruence subgroups in arithmetic lattices (Raimbault, 2022, Frączyk et al., 2 Feb 2026).
• Generalized graph classes: Local limit theory via BS-convergence extends to subcritical random graphs and random graphs with excluded minors, with extensions to compactifications admitting vertices of infinite degree (Georgakopoulos et al., 2015).
• Statistical mechanics and partition functions: The method provides convergence of normalized logarithms of the chromatic, Tutte, and independence polynomials (free energy) in complex parameters in domains outside limiting support, facilitating analytic and algorithmic investigations (Abért et al., 2012, Csikvári et al., 2012).
• Translation surfaces and moduli: BS-convergence in high-genus translation surfaces reveals the Poisson translation plane as the universal local limit, with implications for the statistics of geodesics and cones (Bowen et al., 7 Jan 2025).
• Open problems: Questions remain regarding rates of convergence for deterministic treatments, convergence and structure of spectral measures on random or weighted hypergraphs, existence of singular continuous spectra, extension of the theory to directed or weighted graphs, and explicit characterization of limiting harmonic/free energy functions (Bordenave, 11 Oct 2025, Csikvári et al., 2012, Abert et al., 2018, Cerbo et al., 2019).

Benjamini–Schramm convergence unifies local-to-global phenomena in discrete, geometric, and spectral settings, offering a robust probabilistic and analytic framework for understanding the asymptotic geometry, topology, and analysis of large structures. It provides the technical foundation for universality results in combinatorics, Riemannian geometry, quantum chaos, and random matrix theory.