Incipient infinite clusters and self-similarity for metric graph Gaussian free fields and loop soups

Zhenhao Cai,Jian Ding
2024-12-08
Abstract:In this paper, we establish the existence and equivalence of four types of incipient infinite clusters (IICs) for the critical Gaussian free field (GFF) level-set and the critical loop soup on the metric graph $\widetilde{\mathbb{Z}}^d$ for all $d\ge 3$ except the critical dimension $d=6$. These IICs are defined as four limiting conditional probabilities, involving different conditionings and various ways of taking limits: (1) conditioned on $\{0 \leftrightarrow{} \partial B(N)\}$ at criticality (where $\partial B(N)$ is the boundary of the box $B(N)$ centered at the origin $0$ with side length $2N$), and letting $N\to \infty$; (2) conditioned on $\{0\leftrightarrow{} \infty\}$ at super-criticality, and letting the parameter tend to the critical threshold; (3) conditioned on $\{0 \leftrightarrow{} x\}$ at criticality (where $x\in \mathbb{Z}^d$ is a lattice point), and letting $x\to \infty$; (4) conditioned on the event that the capacity of the critical cluster containing $0$ exceeds $T$, and letting $T\to \infty$. Our proof employs a robust framework of Basu and Sapozhinikov (2017) for constructing IICs as in (1) and (2) for Bernoulli percolation in low dimensions (i.e., $3\le d\le 5$), whereas a key hypothesis on the quasi-multiplicativity is proved in our companion paper. We further show that conditioned on $\{0 \leftrightarrow{} \partial B(N)\}$, the volume within $B(M)$ of the critical cluster containing $0$ is typically of order $M^{(\frac{d}{2}+1)\vee 4}$, regardless of how large $N$ is (provided with $N\gg M$). This phenomenon indicates that the critical cluster of GFF or loop soup exhibits self-similarity, which supports Werner's conjecture (2016) that such critical cluster has a scaling limit. Moreover, the exponent of the order $M^{(\frac{d}{2}+1)\vee 4}$ matches the conjectured fractal dimension of the scaling limit from Werner (2016).
Probability
What problem does this paper attempt to address?
The key problem that this paper attempts to solve lies in establishing and proving the existence and equivalence of four types of incipient infinite clusters (IICs) in the Gaussian Free Field (GFF) and loop soup models, especially on the metric graph \( \mathbb{Z}^d \) for all dimensions \( d \geq 3 \) and \( d \neq 6 \). Specifically: 1. **Define four types of IICs**: - Conditional on \( \{0 \leftrightarrow \partial B(N)\} \) at the critical point, as \( N \to \infty \). - Conditional on \( \{0 \leftrightarrow \infty\} \) in the super - critical state, with the parameter approaching the critical threshold. - Conditional on \( \{0 \leftrightarrow x\} \) at the critical point, as \( x \to \infty \). - Conditional on the capacity of the critical cluster containing the origin exceeding \( T \), as \( T \to \infty \). 2. **Proof framework**: - Use the robust framework proposed by Basu and Sapozhinikov (2017) to construct IICs in Bernoulli percolation in low dimensions (\( 3 \leq d \leq 5 \)). - Prove a key assumption - quasi - multiplicativity through the accompanying paper. 3. **Self - similarity**: - The paper further shows that under the condition \( \{0 \leftrightarrow \partial B(N)\} \), the volume of the critical cluster within \( B(M) \) is typically \( M^{(d/2 + 1)\vee 4} \), which indicates that the critical cluster has self - similarity and supports Werner (2016)'s conjecture that such a critical cluster has a scale limit. 4. **Main results**: - Theorem 1.1: For all \( d \geq 3 \) and \( d \neq 6 \), the above four types of IICs exist and are equivalent. - Theorem 1.2: Gives the exact order of the typical volume of the critical cluster, indicating its self - similarity. - Theorem 1.3: Shows the anti - concentration property of the volume of the critical cluster, implying that if there is a scale limit, it is random. Through these studies, the paper aims to gain a deep understanding of the structure of critical clusters and their scale - limit behavior in the Gaussian Free Field and loop soup models.