On reversing the Simon-Lieb inequality in high-dimensional percolation

This paper establishes a partial reversal of the Simon-Lieb inequality for Bernoulli percolation in dimensions $d>6$, which leads to the uniform boundedness of Duminil-Copin and Tassion's quantity $\varphi_{p_c}(S)$ and provides a concise derivation of key near-critical estimates and sharp bounds on the critical one-arm probability.

The Gist

Imagine a vast, infinite city grid made of streets and intersections. This is our mathematical "city," called $\mathbb{Z}^d$. Now, imagine a heavy fog rolls in, and every street has a chance of being open or closed. If a street is open, you can walk on it; if it's closed, you can't. This is percolation: the study of how far you can walk from your starting point (the origin) before the closed streets block your path.

The paper focuses on what happens in very high dimensions (think of a city with 7, 8, or more directions to go, rather than just North, South, East, and West). In these high-dimensional cities, the rules of connectivity behave in a surprisingly simple, "average" way, similar to how a random walk (a drunkard's walk) behaves.

Here is the breakdown of the paper's discoveries using simple analogies:

1. The Old Rule: The "One-Way" Fence

For a long time, mathematicians had a powerful tool called the Simon-Lieb inequality. Think of this as a "One-Way Fence."

Imagine you are trying to get from your house (Point A) to a friend's house (Point B).

• The Old Rule: If you build a small fence around your house (a set $S$), the rule says: "The chance of getting to your friend is at most the chance of getting to the fence, plus the chance of jumping over the fence and then getting to your friend."
• The Problem: This rule is great for proving things are impossible or unlikely, but it's a "one-way" street. It tells you the probability is low, but it doesn't help you prove it's high enough. It's like saying, "You can't get there faster than this," but not helping you figure out if you can actually make the trip.

2. The New Discovery: The "Two-Way" Bridge

The authors of this paper discovered that in high-dimensional cities (dimensions greater than 6), this "One-Way Fence" rule can be partially reversed.

They proved a "Partially Reversed Simon-Lieb Inequality."

• The New Rule: They showed that the chance of getting from A to B is actually at least the chance of getting to the fence, PLUS a specific, calculated amount of "bonus" probability for crossing the fence.
• The Catch: To make this work, they had to be careful. When you cross the fence, you can't just assume the path is clear. You have to make sure you aren't walking through a "ghost cluster"—a tangled mess of streets you already explored that might block your new path.
• The Analogy: Imagine you are exploring a maze. The old rule said, "You can't get out faster than this." The new rule says, "If you step out of your current room, you have a guaranteed minimum chance of reaching the exit, provided you don't get stuck in the room you just left."

3. The Big Result: The "Crowded Party" is Under Control

The most famous application of their new rule concerns a quantity called $\phi_{pc}(S)$.

• What is it? Imagine a party at your house. You want to know how many people are standing right at the doorway, ready to leave your house and go into the neighborhood. This quantity measures the "expected number of pioneers" at the edge of any shape you draw in the city.
• The Old Mystery: In lower dimensions (like our 3D world), if you draw a huge, jagged, or weirdly shaped boundary, the number of people at the edge could theoretically explode to infinity. It was a mystery whether this number stayed manageable in high dimensions.
• The Paper's Claim: The authors proved that in high dimensions ($d > 6$), this number is always bounded. No matter how big or weird your shape is, the number of people at the edge never gets out of control. It stays within a fixed, safe limit.
• Why it matters: It's like discovering that no matter how chaotic a party gets, the number of people trying to leave through the door at any one moment never exceeds a specific number. This gives mathematicians a "safety net" to use in other complex calculations.

4. The "Sharp Length" and the "One-Arm"

Using this new "Two-Way Bridge" and the fact that the "party crowd" is under control, the authors solved two other puzzles:

• The Sharp Length ($L(p)$): As the fog gets thicker (approaching the critical point where the city stops being connected), the distance you can walk before hitting a wall grows. The paper proves exactly how fast this distance grows. It turns out it grows like the inverse of the square root of how close you are to the critical point. It's a precise recipe for how the city "breaks" as the fog rolls in.
• The One-Arm Probability: This asks: "What is the chance that you can walk from the center of the city to a circle of radius $n$?" The paper proves that in high dimensions, this chance drops off exactly like $1/n^2$. This confirms a decades-old prediction about how these high-dimensional cities behave.

Summary

In simple terms, this paper took a one-way traffic rule that mathematicians had used for decades and turned it into a two-way street for high-dimensional spaces. By doing so, they proved that the "edge" of any shape in these high-dimensional worlds is always well-behaved and predictable. This allowed them to quickly and cleanly solve several other long-standing puzzles about how these high-dimensional cities connect and disconnect.

Key Takeaway: In dimensions higher than 6, the chaotic randomness of percolation behaves with a surprising, orderly simplicity, and the authors found a new mathematical "bridge" to prove it.

Technical Summary

Title: On Reversing the Simon–Lieb Inequality in High-Dimensional Percolation

Problem Statement
The paper investigates Bernoulli percolation on the integer lattice $\mathbb{Z}^d$ in dimensions $d > 6$, a regime where mean-field behavior is predicted to occur. A central object in the analysis of such models is the restricted two-point function $\tau_{S,p}(x, y)$, representing the probability that $x$ and $y$ are connected by an open path entirely contained within a subset $S$. The study focuses on the behavior of these functions near the critical threshold $p_c$.

A fundamental tool in this field is the Simon–Lieb inequality (a consequence of the van den Berg–Kesten or BK inequality), which provides an upper bound on the two-point function in a larger domain $\Lambda$ in terms of the function restricted to a smaller set $S$ and a sum over boundary edges. While this inequality is crucial for establishing exponential decay and the sharpness of the phase transition, it is generally an inequality. The authors address the question of whether this inequality can be "reversed" or made sharp in the mean-field regime ($d > 6$). Specifically, they seek to establish lower bounds that mirror the structure of the Simon–Lieb inequality, which would allow for a more precise structural analysis of connectivity events.

Methodology
The authors develop a novel structural approach to reverse the Simon–Lieb inequality by introducing specific geometric and probabilistic concepts tailored to high-dimensional percolation:

1. Partially Reversed Simon–Lieb Inequality: The core technical innovation is a new inequality (Theorem 1.2) that provides a lower bound for $\tau_{\Lambda,p}(o, x) - \tau_{S,p}(o, x)$. Unlike the standard Simon–Lieb inequality, the lower bound involves a sum over boundary edges $\{u, v\}$ where the connection from $v$ to $x$ is restricted to the complement of the cluster of $u$ (denoted $C(u; \Lambda)$). This formulation accounts for the non-Markovian nature of cluster exploration.
2. Regular and Escapable Points: To make the reversed inequality effective, the authors introduce the notions of "regular" and "escapable" points.
   • A point is regular if its cluster does not grow too densely (specifically, the volume of the cluster intersection with a box of radius $s$ is bounded by $s^4 (\log s)^7$).
   • A point is escapable if there exists a path from a neighbor outside $S$ to a point far from the cluster of the boundary point, avoiding the cluster itself.
     These concepts allow the authors to argue that, in high dimensions, the restriction to the complement of the cluster does not significantly diminish the connection probability.
3. Averaging and Induction: The proofs rely heavily on averaging arguments. The authors show that, on average, "pioneers" (boundary points connected to the origin) are regular and escapable. This allows them to bypass the need for pointwise lower bounds on the two-point function in the subcritical regime, which are generally unavailable. They utilize induction on scales and the "sharp length" $L(p)$ to propagate these estimates.
4. Comparison with Random Walks: The strategy draws inspiration from the random walk setting, where the Green's function satisfies an exact equality involving the first exit from a set. The authors adapt this intuition to percolation by constructing specific events (pivotal edges) that mimic the Markovian structure of random walks, leveraging the tree-like nature of clusters in $d > 6$.

Key Contributions and Results

• Partial Reversal of Simon–Lieb Inequality (Theorem 1.2): The paper proves that for $d \ge 2$, the Simon–Lieb inequality admits a partial reversal. The lower bound involves the two-point function restricted to the complement of the cluster of the boundary point.
• Uniform Boundedness of $\phi_{pc}(S)$ (Theorem 1.4): A major application is the proof that the quantity $\phi_{pc}(S)$, defined as the expected number of pioneers on the boundary of a set $S$ (weighted by connection probabilities), is uniformly bounded for all finite sets $S \subset \mathbb{Z}^d$ containing the origin, provided $d > 6$. This extends a known property of simple random walks to high-dimensional percolation.
• Near-Critical Estimates:
  • Sharp Length (Theorem 1.6): The authors derive sharp bounds on the sharp length $L(p)$, showing $L(p) \asymp (p_c - p)^{-1/2}$.
  • Two-Point Function (Theorem 1.8): They obtain precise near-critical bounds for the two-point function $\tau_p(0, x)$, establishing both upper and lower bounds of the form $(1 \vee |x|)^{-(d-2)} \exp(-c(p_c - p)^{1/2}|x|)$.
  • Correlation Lengths (Corollary 1.9): The results imply that various definitions of correlation length scale as $(p_c - p)^{-1/2}$.
• One-Arm Exponent (Theorem 1.11): Using the derived bounds, the paper provides a short, self-contained proof that the one-arm probability $\theta_n(p_c)$ decays as $n^{-2}$ in dimensions $d > 6$. This reproves a celebrated result of Kozma and Nachmias using a different, more direct route that avoids complex regular point analysis and entropic bounds.
• Corollary 1.5: The paper establishes a comparison result showing that $\phi_p(\Lambda)$ is bounded by a constant multiple of $\phi_p(S)$ for any $\Lambda \supset S$, highlighting a form of stability in the quantity $\phi_p$.

Significance and Claims
The authors claim that their work provides a "short and self-contained route" to several key results in high-dimensional percolation that were previously established using more intricate methods, such as the lace expansion or detailed regular point analysis.

• Structural Insight: The primary significance lies in demonstrating that the BK inequality, often viewed as a one-sided tool, can be effectively reversed in the mean-field regime. This suggests that connectivity events in high dimensions behave more like independent events (as in random walks) than in lower dimensions.
• Independence from Previous Heavy Machinery: The proofs of the near-critical estimates and the one-arm exponent rely primarily on classical results (Aizenman–Newman, Aizenman) and the new reversed inequality, rather than the heavy machinery of the lace expansion or the specific one-arm computations of Kozma and Nachmias.
• Robustness: The authors suggest that the techniques developed, particularly the use of regular and escapable points to handle cluster restrictions, are robust enough to be applied to other models, specifically mentioning the Ising model in dimensions $d > 4$ in forthcoming work.
• Optimality: The paper notes that the uniform boundedness of $\phi_{pc}(S)$ is optimal, as it is expected to fail in dimensions $2 \le d \le 6$.

In summary, the paper introduces a new structural tool—the partially reversed Simon–Lieb inequality—that simplifies the analysis of high-dimensional percolation, unifying the understanding of critical and near-critical behaviors through a framework that closely parallels the properties of simple random walks.