Planar Site Percolation, End Structure, and the Benjamini-Schramm Conjecture

This paper resolves the extension of the Glazman–Harel–Zelesko theorem for planar site percolation under a countability hypothesis on end-equivalence classes while constructing a counterexample to the Benjamini–Schramm conjecture by demonstrating that planar graphs with minimum degree at least 7 can exhibit a uniqueness threshold strictly below $1-p_c$.

The Gist

Imagine you are standing in a vast, infinite city made of blocks (vertices) connected by streets (edges). This city is planar, meaning you can draw it on a flat piece of paper (or a sphere) without any streets crossing each other.

Now, imagine a random process where every block in this city is either Open (green, like a park) or Closed (gray, like a building) based on a coin flip. If the coin lands heads with probability $p$, the block becomes a park.

The big question mathematicians ask about this city is: How many giant, endless parks will form?

• If $p$ is very low, the parks are tiny islands.
• If $p$ is very high, the whole city becomes one giant park.
• But what happens in the middle? Is there a "sweet spot" where you have many different giant parks existing at the same time, forever separated from each other?

This paper, written by Zhongyang Li, tackles a famous 30-year-old mystery about this "sweet spot" in planar cities.

The Big Mystery: The Benjamini-Schramm Conjecture

In the 1990s, two mathematicians, Benjamini and Schramm, made a bold guess (a conjecture). They said:

"If your city is built such that every block has at least 7 neighbors (a very dense city), then there is a special range of probabilities where you will always have infinitely many giant parks."

They believed that in this dense, flat world, the "middle ground" is chaotic and full of infinite islands.

The Paper's Discovery: It Depends on the "Horizon"

Li's paper says: "The guess is mostly right, but not always."

The answer depends on how the city looks as you zoom out to infinity. In math, we call these distant points "Ends." Think of them as the different directions you can walk forever without hitting a wall.

Scenario A: The City with a Countable Horizon (The "Good" Case)

Imagine a city where, as you look toward the horizon, you can only see a countable number of distinct directions (like the points on a number line: 1, 2, 3...). Maybe the city looks like a tree with a few main branches stretching out forever.

• The Result: In this case, the Benjamini-Schramm guess is TRUE.
• The Analogy: If you have a finite or countable number of "directions" to infinity, the random parks will eventually split up. If you pick a probability $p$ in the middle range, you will almost certainly find infinitely many distinct, endless parks. They are like separate continents that never touch.

Scenario B: The City with an Uncountable Horizon (The "Bad" Case)

Now, imagine a city so complex that the horizon looks like a solid line or a cloud, with uncountably many directions (like every single point on a line segment). This is a much wilder, more crowded infinity.

• The Result: In this case, the Benjamini-Schramm guess is FALSE.
• The Analogy: Li constructed a specific, weird city with this "solid horizon." In this city, even though it's dense (every block has 7+ neighbors), there is a middle range where you do not get infinitely many parks. Instead, you might get exactly one giant park, or a few, but not an infinite army of them. The "infinity" of the city's shape forces the parks to merge into a single massive entity.

How Did They Solve It? (The Toolkit)

To prove this, Li invented a new way of looking at the city, called the FCA Framework:

1. Freudenthal Embedding (The Map): They took the infinite city and mapped it onto a sphere (like the Earth). This allowed them to treat the "ends" (the horizon) as actual points on the map.
2. Cycle Separation (The Fence): They asked: "Can a loop of streets separate two distant points?" If you can draw a fence (a cycle) that keeps two points apart, they are in different "neighborhoods" of infinity. If you can't, they are in the same neighborhood.
3. Alternating Arm Exploration (The Detective): This is the coolest part. Imagine you are trying to see if two distant parks can connect. You send out "arms" (paths) from a central point.
   • If you have a "1-arm" (a green path) and a "0-arm" (a gray path) alternating around a circle, they act like a fence.
   • Li proved that if you have enough of these alternating fences, they force the green parks to stay separate.
   • By counting how many "directions" (ends) exist, they could calculate exactly when these fences become strong enough to create infinite separation.

The Takeaway

• For "Simple" Infinite Cities: If the horizon is simple (countable), the universe is chaotic and full of infinite islands in the middle range.
• For "Complex" Infinite Cities: If the horizon is wild (uncountable), the universe can be more orderly, allowing for a single giant island even in the middle range.

Why does this matter?
This resolves a 30-year-old debate. It shows that the geometry of infinity matters. You can't just look at how many neighbors a block has; you have to look at the shape of the entire universe. It's a reminder that in the world of probability and infinite structures, "infinity" comes in different flavors, and they behave very differently.

In short: The paper proves that while dense, flat cities usually have infinite islands, there is a hidden, complex type of city where the islands merge into one, breaking the old rule.

Technical Summary

1. Problem Statement and Context

The paper addresses the fundamental problem of uniqueness of infinite clusters in Bernoulli site percolation on infinite, connected, locally finite planar graphs.

• Definitions:
  • $p_c(G)$: The critical threshold where an infinite open cluster first appears.
  • $p_u(G)$: The uniqueness threshold where the number of infinite open clusters becomes exactly one.
  • Coexistence Interval: The range $(p_c, p_u)$ where multiple infinite clusters may exist.
• The Benjamini–Schramm Conjecture (Conjecture 7 in [4]):
  For a planar graph $G$ with minimum vertex degree $\delta(G) \geq 7$, it was conjectured that:
  1. $p_c(G) < 1/2$.
  2. For every $p \in (p_c(G), 1 - p_c(G))$, there are almost surely infinitely many infinite open clusters.
     This implies the inequality $p_u(G) \geq 1 - p_c(G)$.
• Previous Progress:
  • Glazman, Harel, and Zelesko [15] proved that for $p \in (p_c, 1/2]$, there are infinitely many infinite clusters (the "lower half" of the coexistence regime).
  • The "upper half" ($p \in (1/2, 1-p_c)$) remained open, particularly for graphs with complex end structures.

The Core Question: Does the inequality $p_u(G) \geq 1 - p_c(G)$ hold for all locally finite planar graphs with $\delta(G) \geq 7$, or does the structure of the graph's "ends" (infinite directions) play a decisive role?

---

2. Methodology: The FCA Framework

The author introduces a novel framework called FCA (Freudenthal–Cutset–Arms) to analyze percolation on general planar graphs without relying on transitivity or symmetry.

A. Freudenthal Embedding and End-Equivalence

• Embedding: The paper utilizes a well-separated Freudenthal embedding $\phi: G \hookrightarrow S^2$. This maps the graph to the sphere such that the "ends" of the graph (infinite paths going to infinity) correspond bijectively to a set of accumulation points $A \subset S^2$.
• Cycle-Separation Equivalence: Two ends (accumulation points) are defined as equivalent if no cycle in $G$ separates them on the sphere. This partitions the set of ends $A$ into equivalence classes $\mathcal{F}(G)$.
• Directional Thresholds: For each equivalence class $F \in \mathcal{F}(G)$, a directional critical threshold $p_{c,F}(G)$ is defined, representing the probability required to connect to that specific "direction" at infinity.

B. Cutset Characterization

• The paper employs a cutset-based differential inequality approach. It characterizes $p_{c,F}(G)$ using the $\phi$-functional, which relates connectivity probabilities to the existence of finite cutsets separating a vertex from a specific end class.
• This allows for analytic control over connectivity without assuming the graph is transitive.

C. Alternating-Arm Exploration

• The core geometric argument involves constructing alternating-arm events.
• The author defines an $F$-adapted boundary cycle $\partial_F S$ around a finite set $S$ relative to an end class $F$.
• By exploring the complement of a growing cluster, the method forces the existence of disjoint "corridors" (arms) connecting the boundary to the end $F$.
• Topological Separation: On the sphere $S^2$, these disjoint arms separate the complement into distinct regions. If the arms alternate between open (1) and closed (0) states, the topological constraints force the existence of multiple disjoint infinite clusters in the unexplored regions.

---

3. Key Contributions and Results

The paper provides a complete resolution to the extension problem posed by Glazman–Harel–Zelesko, distinguishing between graphs with countable and uncountable end-equivalence classes.

Result 1: The Countable Case (Theorem 1.7 (ii))

• Hypothesis: The set of end-equivalence classes $\mathcal{F}(G)$ is countable. (This includes all graphs with countably many ends and all properly embedded planar graphs).
• Main Theorem:
  $$p_c(G) = \inf_{F \in \mathcal{F}(G)} p_{c,F}(G)$$
  Consequently, for every $p \in (p_c(G), 1 - p_c(G))$, there are almost surely infinitely many infinite open clusters.
• Implication: For this broad class of graphs, the Benjamini–Schramm conjecture holds true: $p_u(G) \geq 1 - p_c(G)$. The non-uniqueness regime extends fully from $p_c$ to $1-p_c$.

Result 2: The Uncountable Case and Counterexample (Theorem 1.7 (iii))

• Hypothesis: The set of end-equivalence classes $\mathcal{F}(G)$ is uncountable.
• Counterexample Construction: The author constructs an explicit infinite, connected, locally finite planar graph $G$ with minimum degree $\delta(G) \geq 7$ (specifically, a "two-sided $M$-adic tree of strips with triangulated faces") such that:
  1. $p_c(G) < 1/2$.
  2. There exists a parameter $p \in (1/2, 1 - p_c(G))$ where the number of infinite clusters is finite (specifically, exactly 1 with positive probability).
• Conclusion: For this graph, $p_u(G) < 1 - p_c(G)$.
• Significance: This disproves the Benjamini–Schramm Conjecture 7 in full generality. The conjecture fails when the end structure is sufficiently complex (uncountable equivalence classes).

Result 3: Necessity of Local Finiteness

• The paper also provides an example of a planar graph that is not locally finite where the coexistence predictions fail even more drastically, confirming that the "locally finite" assumption is indispensable for such theorems.

---

4. Technical Highlights of the Proofs

1. Reduction to Directional Thresholds:
   The proof for the countable case relies on showing that if $p > p_c$, then $p$ must exceed $p_{c,F}$ for at least one class $F$. The countability allows a union bound argument to show that infinitely many clusters must form across the different directions.

2. The "Uncountable" Gap:
   In the uncountable case, the infimum of directional thresholds $\inf p_{c,F}$ can be strictly less than the global $p_c$. The constructed counterexample exploits this gap. The graph is designed so that while global connectivity exists, the "corridors" to specific uncountably many ends are blocked in a way that prevents the formation of infinitely many clusters simultaneously in the upper half of the interval.

3. The $M$-adic Tree Construction:
   The counterexample (Construction 8.1) is a sophisticated planar graph built from a rooted $B$-ary tree ($B=M^d$).

   • Vertices are grouped into blocks based on $M$-adic addresses.
   • Horizontal edges and fan triangulations create a "strip" structure.
   • Two copies are glued vertically.
   • Key Property: The geometry ensures that for a specific $p$, the probability of connecting to "deep" ends decays exponentially, but the sheer number of ends (uncountable) allows for global percolation while restricting the number of infinite clusters to be finite.

---

5. Significance and Impact

• Resolution of a 30-Year Open Problem: The paper settles the question of whether planarity and high minimum degree ($\geq 7$) are sufficient to guarantee infinitely many infinite clusters throughout the entire coexistence interval. The answer is no in full generality, but yes under a natural countability hypothesis on the end structure.
• Refinement of the Benjamini–Schramm Conjecture: It clarifies that the conjecture requires an additional topological constraint (countability of end-equivalence classes) to hold.
• New Methodological Tools: The FCA framework (Freudenthal embedding + Cutset analysis + Alternating arms) provides a powerful toolkit for studying percolation on non-transitive, non-amenable planar graphs. It bridges topological graph theory (ends, embeddings) with probabilistic methods (cutsets, differential inequalities).
• Universality and Rigidity: The work draws a parallel between percolation thresholds and conformal geometry (He–Schramm theorem), suggesting that the "countability" of the boundary (ends) is a critical rigidity factor in both fields.

In summary, Zhongyang Li's paper demonstrates that the topology of the "ends" of a planar graph is the decisive factor in determining the uniqueness threshold, providing a definitive counterexample to the general form of the Benjamini–Schramm conjecture while establishing a robust positive result for the vast majority of "well-behaved" planar graphs.