🔢 mathematics

Non-perturbative finite connectivities in supercritical Bernoulli percolation on $\mathbb{Z}^d$ with $d\geq 3$

This paper resolves a longstanding problem in Bernoulli percolation by deriving sharp Ornstein--Zernike asymptotics for the probability of finite cluster connections between distant points throughout the entire supercritical regime on $\mathbb{Z}^d$ ( $d\geq 3$ ), utilizing a combination of a new volume bound and an adapted coarse-graining framework that eliminates the need for mass separation.

Original authors: Yacine Aoun, Kamil Khettabi

Published 2026-10-07

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Yacine Aoun, Kamil Khettabi

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a vast, infinite grid of points, like a three-dimensional lattice stretching forever in every direction. Now, picture that every connection between neighboring points has a chance of being open or closed, like a switch that is flipped randomly. If the chance of being open is high enough, these switches form vast, sprawling networks that can stretch across the entire grid. This is the world of percolation, a mathematical model used to understand how fluids flow through porous rock, how diseases spread through a population, or how electricity travels through a composite material. For decades, scientists have been able to predict the behavior of these networks when the connections are rare and the clusters are small and isolated. However, when the connections are frequent enough to form a giant, spanning network, the behavior of the smaller, finite clusters that get cut off from the main mass has remained a stubborn mystery. Understanding exactly how likely it is for two distant points to be connected by such a finite island of open paths is a fundamental question in statistical physics, one that has resisted a precise solution for a long time.

A team of researchers has now solved this longstanding problem for grids in three or more dimensions. They have derived a precise formula that describes the probability of two far-apart points being linked by a finite cluster when the system is in this highly connected, supercritical state. Their work confirms that this probability decays in a very specific way: it drops off exponentially as the distance between the points increases, but this drop is tempered by a predictable mathematical factor that depends on the geometry of the space. This result, known as Ornstein–Zernike asymptotics, had been proven for other scenarios, such as when connections are rare or in two-dimensional grids, but the three-dimensional case with abundant connections had eluded mathematicians until now. The authors achieved this by developing a new way to look at the structure of these finite clusters, proving that they behave in a surprisingly orderly fashion despite their random origins.

The core of the discovery lies in understanding the shape and size of these finite islands of connection. In a highly connected environment, one might expect these isolated clusters to be chaotic and unpredictable in their volume. However, the researchers proved that there is a strict limit on how large these clusters can grow relative to the distance between their endpoints. They showed that if a cluster connects two distant points, it is overwhelmingly unlikely to be excessively large; its volume is tightly controlled. This control is crucial because it prevents the cluster from becoming so dense or complex that it defies simple description. By establishing this limit, the authors were able to treat the cluster as a collection of simpler, irreducible pieces that can be analyzed one by one, rather than as a single, impenetrable blob.

To reach this conclusion, the team adapted a powerful framework originally designed for different physical systems. They broke down the complex path between two points into a sequence of smaller, self-contained segments. Think of these segments as the individual links in a chain, where each link is a small, irreducible cluster that cannot be split further without breaking the connection. The researchers demonstrated that these links behave like steps in a random walk, but with a very specific constraint: the steps are not arbitrary. They found that the "steps" taken by these clusters have a predictable average direction and size, and that the likelihood of taking a very large, unusual step drops off rapidly. This allowed them to model the entire connection as a journey made of many small, independent steps, a method that transforms a difficult geometric problem into a manageable statistical one.

A key insight in their work is that they did not need to assume a separation between different types of connections, a requirement that had complicated previous attempts to solve similar problems in other contexts. Instead, they showed that the natural limits on the size of the clusters were sufficient to ensure that the "steps" in their random walk model remained well-behaved. This meant that the mathematical machinery used to describe the decay of probability could be applied directly and rigorously. The result is a formula that not only predicts the exponential decay of the connection probability but also pinpoints the exact polynomial factor that modifies this decay. This factor depends on the dimension of the space and the direction in which the points are separated, providing a complete and sharp description of the phenomenon.

The implications of this work extend beyond the specific grid model studied. The methods developed by the authors offer a new toolkit for analyzing complex systems where randomness and connectivity interact. By proving that finite clusters in a supercritical regime follow these precise asymptotic rules, the researchers have closed a significant gap in our understanding of statistical mechanics. Their findings suggest that even in a regime where a giant connected component dominates the system, the smaller, finite components retain a hidden order that can be described with mathematical precision. This clarity allows physicists and mathematicians to better predict how correlations decay in materials and systems that are near, but not at, a critical transition point. The work stands as a definitive proof, resolving a question that has lingered for years and providing a solid foundation for future investigations into the behavior of random networks in high dimensions.

Technical Summary: Non-perturbative finite connectivities in supercritical Bernoulli percolation on $\mathbb{Z}^d$ with $d \ge 3$

Problem Statement
This work addresses a longstanding open problem in the theory of Bernoulli percolation: establishing sharp Ornstein–Zernike (OZ) asymptotics for the probability that two distant points are connected by a finite cluster in the supercritical regime ( $p > p_c$ ) on the lattice $\mathbb{Z}^d$ for dimensions $d \ge 3$ .

While OZ asymptotics—characterizing exponential decay with a polynomial prefactor of order $\|x\|^{-(d-1)/2}$ —have been well-established for subcritical percolation and various spin models (e.g., Ising, FK percolation) in both subcritical and high-temperature regimes, the supercritical finite-connectivity case remained unresolved. Previous results in the supercritical phase were limited to:

Axial directions only (Braga, Procacci, Sanchis; Campanino, Gianfelice).
Directions sufficiently close to $p=1$ .
Point-to-plane asymptotics (Fribergh, Hammond).
A different asymptotic regime in $d=2$ (Campanino, Ioffe, Louidor).

The specific challenge in the supercritical regime is that the infinite cluster dominates the connectivity, and finite clusters are rare events. Proving OZ asymptotics requires controlling the geometry of these finite clusters without relying on a "separation of masses" between irreducible and unrestricted connections, a technique standard in subcritical analysis but unavailable here.

Methodology
The proof combines two primary ingredients: a priori volume control derived from renewal structures and an adaptation of the modern coarse-graining framework to finite supercritical connections.

A Priori Volume Control (Theorem 1.1):
The authors utilize the renewal structure developed in [2] to derive a bound on the volume of finite clusters connecting distant points. By decomposing the cluster into "strings" (irreducible components) and applying a tilted measure, they establish that the volume of a finite cluster connecting $0$ to $x$ is unlikely to exceed a linear scale $K\|x\|$ . Specifically, they prove that the probability of a connection with volume $|C_0| > K\|x\|$ decays with a stretched-exponential tail relative to the distance, preserving the exponential cost of the connection. This result relies on the fact that the "effective random-walk increments" of the cluster decomposition possess stretched-exponential tails.
Adapted Coarse-Graining and Irreducible Clusters:
The second ingredient adapts the coarse-graining and irreducible-cluster framework of [10] to the supercritical setting.
- Skeleton Construction: A skeleton (a directed tree) is associated with every finite cluster. The algorithm selects "boxes" along the cluster to define a backbone (trunk) and branches.
- Energy Estimates: The authors derive energy estimates for the skeleton, showing that the cost of disconnecting boxes from the cluster is sufficient to control the probability.
- Mass Gap and Cone-Points: A critical distinction from subcritical analysis is the lack of a natural separation of masses. The authors circumvent this by proving that, up to an exponential error, finite clusters possess a positive density of "cone-points" (vertices where the cluster is locally confined to a cone). This is achieved by showing that the a priori volume bound prevents the cluster from being too densely connected locally too often.
- Random Walk Representation: The cluster is decomposed into irreducible pieces (backward-irreducible, forward-irreducible, and irreducible). This decomposition allows the connectivity function $G_p(x)$ to be represented as a renewal equation involving a directed random walk with increments distributed according to the weights of these irreducible pieces.

Key Results

Theorem 1.1 (Volume Control): There exist constants $K, C, c > 0$ such that for any $x \in \mathbb{Z}^d \setminus \{0\}$ :
$P_p(0 \leftrightarrow x, K\|x\| < |C_0| < \infty) \le C \exp\left\{-\xi_p(x) - c\|x\|^{\frac{d-1}{2d}}\right\}.$
This establishes that the volume of the finite cluster is linearly bounded with high probability, with a stretched-exponential correction.
Theorem 1.2 (Sharp OZ Asymptotics): There exists a continuous function $\Psi_p: S^{d-1} \to (0, \infty)$ such that, uniformly in $x$ as $\|x\| \to \infty$ :
$P_p(0 \leftrightarrow x, |C_0| < \infty) = \frac{\Psi_p(x/\|x\|)}{\|x\|^{(d-1)/2}} e^{-\xi_p(x)} (1 + o_{\|x\|\to\infty}(1)).$
This confirms that the probability of finite connection decays exponentially with rate $\xi_p(x)$ and a polynomial prefactor of order $\|x\|^{-(d-1)/2}$ , consistent with the Ornstein–Zernike form.

Significance and Claims
The paper claims to solve the "last remaining asymptotics" for nearest-neighbor Bernoulli percolation on $\mathbb{Z}^d$ ( $d \ge 3$ ) by establishing OZ asymptotics in the entire supercritical regime, not just near $p=1$ or in specific directions.

The authors emphasize that their argument does not require the separation of masses between irreducible and unrestricted finite connections, which was a prerequisite in previous subcritical analyses. Instead, the combination of the volume bound (Theorem 1.1) and a modified coarse-graining construction provides the necessary stretched-exponential tails for the effective random-walk increments. This allows for the application of standard local central limit theorem techniques to derive the sharp asymptotics.

The work extends the modern Ornstein–Zernike theory to a regime where the infinite cluster exists, demonstrating that finite clusters still exhibit the same universal scaling behavior as in the subcritical phase, provided the volume is controlled. The authors note that their framework suggests potential extensions to general finite-range models, dependent percolation (like FK percolation), and low-temperature covariances in Ising and Potts models, though these are presented as directions for future research rather than results of the current paper.

More like this