Decay of connection probability in high-dimensional continuum percolation

Using the lace expansion combined with a recent deconvolution strategy and an $L^p$ induction argument, this paper proves that the critical two-point connection probability in high-dimensional random connection models decays as $|x|^{-(d-2)}$ and simplifies the existing proof for nearest-neighbour Bernoulli percolation on $\mathbb{Z}^d$ for dimensions $d \ge 11$.

The Gist

Imagine you are standing in a vast, foggy forest. Scattered randomly throughout this forest are trees. Sometimes, two trees are close enough that a vine connects them. Sometimes, they are too far apart, and no vine grows.

Now, imagine you are trying to figure out: If I stand at one tree, how likely am I to find a path of vines leading to a specific tree far away?

This is the core question of the paper you shared. The authors, Matthew Dickson and Yucheng Liu, are studying a mathematical model called the Random Connection Model. It's a way to describe how things connect in high-dimensional spaces (think of a forest that exists not just in 2D or 3D, but in 10, 20, or even 100 dimensions).

Here is the breakdown of their discovery, using simple analogies:

1. The Big Question: How fast does the connection fade?

In a low-dimensional forest (like our normal 3D world), if you look at a tree very far away, the chance of it being connected to you drops off very quickly. It's like trying to shout to someone across a huge canyon; the sound dies out fast.

However, in very high dimensions (which the authors focus on), the rules change. The forest is so "spread out" that the connections behave differently. The authors wanted to prove exactly how the connection probability fades as the distance gets huge.

They were looking for a specific mathematical "fingerprint" called the critical exponent $\eta$.

• If $\eta$ is large, the connection dies out very fast.
• If $\eta$ is zero, the connection dies out at a "standard" or "mean-field" rate, which is the simplest possible behavior for this type of system.

The Result: They proved that in high dimensions, $\eta = 0$.
In plain English: The connection probability fades away exactly as fast as the simplest, most basic physics would predict. There are no weird, hidden complexities slowing it down or speeding it up.

2. The Tool: The "Lace Expansion"

How did they prove this? They used a powerful mathematical tool called the Lace Expansion.

Think of the Lace Expansion like a way to untangle a giant, knotted ball of yarn.

• Imagine you want to know the path from Tree A to Tree Z. There are millions of possible paths (A→B→Z, A→C→D→Z, etc.).
• The "Lace Expansion" is a method that takes this messy, tangled web of possibilities and breaks it down into a neat, organized list of "loops" and "strands."
• It allows mathematicians to calculate the probability of connection by looking at these individual strands rather than the whole mess at once.

3. The New Twist: The "Deconvolution" Strategy

The authors didn't just use the old lace expansion; they upgraded it. They used a technique called deconvolution.

The Analogy:
Imagine you are trying to hear a specific instrument (the connection probability) in a loud orchestra. The music is a mix of many instruments (the complex math of the forest).

• Old method: You try to guess the instrument by listening to the whole song and subtracting the other sounds in your head. It's hard and error-prone.
• New method (Deconvolution): You use a special filter (a mathematical "noise-canceling headphone") that isolates the specific instrument you care about.

The authors used a "filter" (based on recent work by Liu and Slade) that allowed them to isolate the connection probability from the noise of the complex forest. This made the proof much cleaner and simpler than previous attempts.

4. The "Induction" Ladder

To prove their filter worked, they had to climb a ladder.

• They started by proving the connection was stable for short distances (the bottom rung).
• Then, they used a logical step (induction) to say: "If it works for distance $X$, it must also work for distance $X + \text{a little bit}$."
• They climbed this ladder all the way up to the very long distances.
• The Innovation: They used a specific type of math (called $L^p$ norms) that is like measuring the "volume" of the noise in different ways. This made it easier to prove that the noise wasn't getting too loud as they climbed the ladder.

5. Why Does This Matter?

You might ask, "Who cares about a forest in 100 dimensions?"

• Universality: This result shows that in high dimensions, nature simplifies. Complex systems (like percolation, which models how water flows through soil, or how diseases spread) start to behave in a predictable, "average" way.
• Simplifying the Future: The authors' method is much shorter and cleaner than the previous best proof (which took a famous mathematician, Hara, a long time to write). This means other scientists can now use this "shortcut" to solve even harder problems, like studying what happens at the very edge of a forest (the "incipient infinite cluster") or in a half-forest.

Summary

Dickson and Liu looked at a complex, high-dimensional forest. They wanted to know how fast the "connection signal" fades over long distances. Using a new, sharper version of a mathematical "untangling tool" (the Lace Expansion) and a "noise-canceling filter" (Deconvolution), they proved that the signal fades at the standard, predictable rate. This confirms that in high dimensions, the universe is surprisingly simple and orderly.

Technical Summary

1. Problem Statement

The paper investigates the Random Connection Model (RCM), a continuum percolation model on $\mathbb{R}^d$. In this model:

• Vertices are distributed according to a Poisson point process with intensity $\lambda$.
• Edges between vertices $x$ and $y$ are drawn independently with probability $\phi(x-y)$, where $\phi$ is an integrable, even adjacency function.

The primary focus is on the behavior of the critical two-point connection probability, denoted $\tau_{\lambda_c}(x) = P_{\lambda_c}(0 \leftrightarrow x)$, as the distance $|x| \to \infty$ in high dimensions ($d$ sufficiently large).

Specifically, the authors aim to prove that the critical exponent $\eta$, which characterizes the decay rate via the relation:
$$\tau_{\lambda_c}(x) \sim \frac{1}{|x|^{d-2+\eta}}$$
takes the mean-field value $\eta = 0$. This implies that the connection probability decays asymptotically as $|x|^{-(d-2)}$, potentially with anisotropic scaling determined by a covariance matrix.

2. Methodology

The proof relies on the Lace Expansion, a powerful perturbative technique originally developed for self-avoiding walks and later adapted for percolation. The methodology involves three main pillars:

A. The Lace Expansion Equation

The authors utilize the lace expansion to derive a convolution equation for the two-point function $\tau_\lambda$:
$$\tau_\lambda = (\phi + \Pi_\lambda) + \lambda(\phi + \Pi_\lambda) * \tau_\lambda$$
where $\Pi_\lambda$ is the "lace function" representing the sum of all lace diagrams. This is rearranged into a deconvolution form:
$$(\delta - J_\lambda) * (\lambda \tau_\lambda) = J_\lambda$$
where $J_\lambda = \lambda(\phi + \Pi_\lambda)$.

B. Deconvolution Strategy

To extract the asymptotic behavior of $\tau_{\lambda_c}(x)$, the authors apply a deconvolution theorem (from Liu and Slade [21]) on $\mathbb{R}^d$. This theorem states that if the Fourier transform of $J_\lambda$ satisfies specific conditions (infrared bounds and moment conditions), the inverse Fourier transform of the solution exhibits the desired $|x|^{-(d-2)}$ decay.

The core technical challenge is verifying the moment conditions for $J_\lambda$, specifically that the $(d-2)$-th moment of the lace function $\Pi_\lambda$ is finite (or "good").

C. Inductive $L^p$ Moment Estimation

The paper introduces a novel inductive strategy to prove the finiteness of moments of $\Pi_\lambda$:

1. Base Case: Establish that the 0-th and 2-nd moments of $\Pi_\lambda$ are bounded (using standard lace expansion convergence results).
2. Inductive Step: Prove that if the $\phi$-th moment is "good," then the $(\phi + 2)$-th moment is also "good."
3. Key Innovation: Unlike previous proofs (e.g., Hara [10]) that relied on strict $L^1$ decay hypotheses for the lace function, this paper utilizes $L^p$ norms (specifically an $L^p$ version of Hara's induction argument).
   • They define a moment as "good" if the weighted function $|x|^a \Pi_\lambda(x)$ belongs to specific $L^p \cap L^2 \cap L^\infty$ spaces.
   • This approach allows them to bypass difficult diagrammatic estimates (specifically the second diagrammatic estimate in Hara's work) by using simpler $L^p$ bounds and Fourier analysis.

3. Key Contributions

• Proof of Mean-Field Decay: The authors rigorously prove that for sufficiently large dimensions ($d \ge d_0 \ge 8$), the critical two-point function decays as $|x|^{-(d-2)}$ (i.e., $\eta=0$) for the Random Connection Model.
• Simplification of Existing Proofs: The method significantly simplifies the proof for nearest-neighbor Bernoulli percolation on $\mathbb{Z}^d$ (previously established by Hara in 2008 for $d \ge 11$). By using $L^p$ induction, they avoid the most technically demanding parts of Hara's original diagrammatic estimates.
• Generalization to Continuum: The results extend the mean-field behavior from lattice models to continuum models with general adjacency functions (including the Gilbert disk model and Gaussian models), provided the adjacency function satisfies specific integrability and infrared bound conditions.
• Anisotropic Decay: The result provides an explicit asymptotic formula involving a positive-definite diagonal matrix $\Sigma$, accounting for potential anisotropy in the decay:
  $$\tau_{\lambda_c}(x) \sim \frac{a_d}{\lambda_c \sqrt{\det \Sigma}} \frac{1}{(x \cdot \Sigma^{-1} x)^{(d-2)/2}}$$

4. Main Results

Theorem 1.3:
Let $\phi$ satisfy Assumption 1.1 (integrability, infrared bound, and specific decay properties). If the dimension $d$ is sufficiently large ($d \ge 8$), there exists a critical intensity $\lambda_c$ and a positive-definite matrix $\Sigma$ such that:
$$\tau_{\lambda_c}(x) \sim \frac{C}{(x \cdot \Sigma^{-1} x)^{(d-2)/2}} \quad \text{as } |x| \to \infty$$
where $C$ is a constant depending on $d$ and $\lambda_c$.

Proposition 1.6 (Inductive Step):
If the $\phi$-th moment of $\Pi_\lambda$ is "good" (in the defined $L^p$ sense), then the $(\phi+2)$-th moment is also "good." This proposition allows the induction to proceed from low moments up to the critical $(d-2)$-th moment required for the deconvolution theorem.

5. Significance and Implications

• Universality: The result confirms that the critical exponent $\eta=0$ is a universal feature of high-dimensional percolation, independent of the specific microscopic details of the connection function (as long as it is "spread-out" enough or the dimension is high enough).
• Applications to IIC: The precise asymptotic form of the two-point function is a prerequisite for constructing the Incipient Infinite Cluster (IIC) and studying percolation in restricted geometries (e.g., half-spaces or large tori).
• Methodological Advancement: The shift from $L^1$-based induction to $L^p$-based induction represents a significant technical refinement in the lace expansion literature. It demonstrates that estimating moments in $L^p$ spaces is often more tractable than proving pointwise decay, potentially opening the door to analyzing other complex models where diagrammatic estimates are difficult.
• Dimensional Threshold: While the optimal threshold for mean-field behavior is conjectured to be $d_c=6$, the paper establishes the result for $d \ge 8$. The authors note that the restriction to $d \ge 8$ is due to the use of a "square diagram" estimate to simplify the proof, not a fundamental limitation of the method.

In summary, this paper provides a streamlined, rigorous proof of mean-field behavior for continuum percolation in high dimensions, utilizing a novel $L^p$ induction strategy to simplify and generalize previous lattice-based results.