Scaling limits for the critical level-set percolation of the Gaussian free field on regular trees

Basic Information

• Paper ID: 2510.14786
• Title: Scaling limits for the critical level-set percolation of the Gaussian free field on regular trees
• Authors: Jiří Černý, Ramon Locher
• Classification: math.PR (Probability Theory)
• Publication Date: October 16, 2025
• Paper Link: https://arxiv.org/abs/2510.14786v1

Abstract

This paper continues the investigation of level-set percolation of the discrete Gaussian free field (GFF) at criticality on regular trees. The authors first derive precise asymptotic estimates for the probability that the critical level-set connected component containing the tree root reaches generation n, particularly proving that the one-arm exponent ρ = 1. Subsequently, they establish a Yaglom-type limit theorem for the GFF values in the nth generation within this connected component. Finally, they prove that under appropriate rescaling, this connected component converges to the Aldous continuum random tree as n → ∞ conditioned on reaching generation n.

Research Background and Motivation

Problem Background

Level-set percolation of the Gaussian free field, particularly on ℤᵈ, is an important model in percolation theory characterized by long-range dependence. Research in this area traces back to pioneering work in the 1980s and has regained attention in the past decade due to the discovery in RS13, which proved that the model undergoes a unique percolation phase transition at the critical threshold h* on ℤᵈ (d ≥ 3).

Research Motivation

1. Theoretical Completeness: Although scaling limit results for branching diffusions on bounded domains exist, analogous results for branching diffusions on general (unbounded) domains remain unavailable.
2. Technical Challenges: The branching processes appearing in this model possess an uncountable and unbounded type space, making previous results inapplicable.
3. Methodological Innovation: Adaptation and extension of strategies from Pow19 are necessary to address the fundamental challenges of unbounded type spaces.

Limitations of Existing Methods

• Classical Galton-Watson process theory primarily handles finite type spaces
• Existing scaling limit theory for branching processes cannot directly address unbounded type spaces
• Lack of deep understanding of precise asymptotic behavior at criticality

Core Contributions

1. Precise Asymptotic Estimates: Derivation of precise asymptotic behavior of one-arm probabilities, proving the one-arm exponent ρ = 1
2. Yaglom-type Limit Theorem: Establishment of limit distribution theory for conditional field values
3. Scaling Limit Convergence: Proof of convergence of critical connected components to the Aldous continuum random tree
4. Technical Methods: Development of new techniques for handling branching processes with unbounded type spaces

Methodology Details

Problem Definition

Study of critical level-set percolation properties of the Gaussian free field φ = (φᵥ)ᵥ∈T on a (d+1)-regular tree T, where:

• Superlevel set: E^h_φ := {v ∈ T : φᵥ ≥ h}
• Connected component containing the root: C^h_o := {v ∈ T : v and o are connected in E^h_φ}
• Critical height: h* = inf{h ∈ ℝ : P|C^h_o| = ∞ = 0}

Model Architecture

1. Gaussian Free Field Construction

The Gaussian free field is constructed recursively:

• φ_o := Y_o, where Y_o ~ N(0, σ²_ν), σ²_ν = d/(d-1)
• For v ≠ o: φ_v := d⁻¹φ_{p(v)} + Y_v, where Y_v ~ N(0, σ²_Y), σ²_Y = (d+1)/d

2. Operator-Theoretic Framework

Define the operator L_h acting on L²(ν):

L_h[f](x) = 1_{[h,∞)}(x) d ∫_{[h,∞)} f(y)ρ_Y(y - x/d) dy

where ρ_Y is the density function of Y. The critical value h* is determined by λ_{h*} = 1, where λ_h is the largest eigenvalue of L_h.

3. Spined Branching Process

Introduce spine construction to study conditional distributions:

• Under P^k_x, there are k spine lines passing through the branching process
• Under Q^k_x, vertices on spines have modified branching behavior
• Transition kernel: K(x,dy) = d(χ(y)/χ(x))ρ_Y(y-x/d)dy

Technical Innovations

1. Handling Unbounded Type Spaces

Through introduction of function families F̂ = {f_λ : λ ∈ [0,∞)}, where f_λ(x) = exp(-χ(x)/λ), uniform control over unbounded type spaces is achieved.

2. Uniform Convergence Proof

Proof of key uniform convergence results:

lim_{n→∞} u^λ_n/⟨χ, u^λ_n⟩ = χ in L²(ν), uniformly for λ ≥ 0

3. Martingale Methods

Construction of martingale S_n = χ(φ_) - Σ_{i≤Λ_n} χ(φ_) + Σ_{w∈Y(v_n)} χ(φ_w), establishing invariance principles.

Experimental Setup

Theoretical Verification Framework

This is purely theoretical research verified through rigorous mathematical proofs:

1. Asymptotic Analysis

• Spectral theory analysis of operator L_h properties
• Establishment of recurrence relations and integral equations
• Application of martingale convergence theorems

2. Convergence of Probability Measures

• Use of Gromov-Hausdorff topology
• Establishment of compactness and convergence
• Application of continuous mapping theorem

Evaluation Metrics

• Asymptotic equivalence: f(n) ~ g(n) as n → ∞
• L²(ν) norm convergence
• Distributional convergence (Skorokhod topology)
• Gromov-Hausdorff convergence

Experimental Results

Main Results

Theorem 2.1 (One-arm Probability)

For each x ≥ h*, as n → ∞:

P_x[N^+_n ≠ ∅] = C₁χ(x)n⁻¹(1 + o(1))
P[N^+_n ≠ ∅] = C₁⟨1,χ⟩n⁻¹(1 + o(1))

where C₁ = 2d/((d-1)⟨χ²,χ⟩), proving the one-arm exponent ρ = 1.

Theorem 2.2 (Yaglom Limit)

For f ∈ L²(ν), let Z be an exponential random variable with mean 1, then:

lim_{n→∞} Z^{f,x}_n = lim_{n→∞} Z^f_n = C₁⁻¹⟨χ,f⟩Z in distribution

Theorem 2.3 (Scaling Limit)

For each x ≥ h*, as n → ∞:

(T_{n,x}, d_{n,x}) → (T_e, d_e)

converges in distribution in the Gromov-Hausdorff topology, where (T_e, d_e) is the conditioned Brownian continuum random tree.

Technical Results

1. Uniformity Estimates

Proof of key uniformity results:

lim_{n→∞} 1/n (1/a^λ_n - 1/a^λ_0) = C₁⁻¹, uniformly for λ ≥ 0

2. Martingale Convergence

Establishment of invariance principle for martingale S_n with variance σ² = ⟨χ,V⟩/⟨χ,1⟩.

3. Geometric Convergence

Proof of convergence of distance matrices, establishing connections between height processes and martingale processes.

Related Work

Historical Development

• 1980s: Pioneering work by Bricmont, Lebowitz, Maes, and others
• 2010s: Rodriguez-Sznitman's revival of interest
• Recent: Progress by Drewitz, Prévost, Rodriguez, and others in various dimensions

Branching Process Theory

• Classical Theory: Foundational work by Kolmogorov and Yaglom
• Multi-type Processes: Extensions by Mode, Miermont, and others
• Continuous Limits: Continuum random tree theory by Aldous and Le Gall

Technical Comparison

Main distinctions of this work from existing literature:

1. New methods for handling unbounded type spaces
2. Establishment of uniform convergence
3. Precise analysis at criticality

Conclusions and Discussion

Main Conclusions

1. Critical Exponents: Determination of one-arm exponent ρ = 1, complementing known δ = 2 and β = 1
2. Limit Distributions: Establishment of complete Yaglom-type limit theory
3. Geometric Convergence: Proof of convergence to the Aldous continuum random tree

Limitations

1. Domain Restrictions: Results limited to regular trees; generalization to arbitrary graphs remains open
2. Technical Complexity: Handling of unbounded type spaces increases proof complexity
3. Computational Implementation: Numerical verification of theoretical results is challenging

Future Directions

1. Generalization to more general graph structures
2. Investigation of scaling limits in non-critical regimes
3. Development of corresponding numerical methods and simulation techniques

In-Depth Evaluation

Strengths

1. Theoretical Depth: Resolution of fundamental technical challenges in branching processes with unbounded type spaces
2. Methodological Innovation: Development of new techniques for handling uniform convergence
3. Result Completeness: Establishment of complete theory from probability estimates to geometric convergence
4. Proof Rigor: Rigorous mathematical arguments with precise technical treatment

Weaknesses

1. Application Scope: Results are primarily theoretical with limited practical applications
2. Computational Complexity: Involved operators and eigenfunction calculations are complex
3. Generalization Difficulty: Extension to more general settings faces technical obstacles

Impact

1. Theoretical Contribution: Provides important technical tools for branching process theory
2. Methodological Value: Methods for handling unbounded type spaces have universal significance
3. Interdisciplinary Connection: Bridges multiple branches of probability theory, analysis, and geometry

Applicable Scenarios

1. Branching process research in probability theory
2. Percolation models in statistical physics
3. Random geometry and random graph theory
4. Related mathematical physics problems

References

This paper cites important literature from probability theory, branching process theory, and random geometry, including:

• Aldous's pioneering work on continuum random trees
• Powell's recent results on branching diffusions on bounded domains
• Rodriguez-Sznitman's foundational research on Gaussian free field percolation
• Harris-Roberts's technical developments on multi-spine branching processes