Planar percolation and the loop O(n) model

This paper proves that a broad class of planar site percolation processes satisfying tail triviality, positive association, and stochastic domination exhibits either zero or infinitely many infinite components, thereby resolving a 1996 conjecture by Benjamini and Schramm and confirming a key aspect of Nienhuis's 1982 phase diagram for the loop O(n) model on the hexagonal lattice.

The Gist

Imagine a vast, infinite city built on a flat sheet of paper (a planar graph). In this city, every building (vertex) can be either "open" (like a park) or "closed" (like a concrete wall). The rules of the city are random: sometimes a building is open, sometimes closed.

The big question mathematicians have been asking for decades is: If you keep opening buildings randomly, do you eventually get one giant, infinite park that stretches forever? Or do you get a few? Or maybe an infinite number of tiny, disconnected parks?

This paper, written by Glazman, Harel, and Zelesko, solves a major mystery about how these "parks" behave, especially when the city is built on a flat surface.

Here is the breakdown of their discovery using simple analogies.

1. The "All or Nothing" Rule

For a long time, mathematicians knew that on a perfect, repeating grid (like a chessboard), if you open buildings randomly, you either get zero infinite parks or exactly one giant infinite park. You can't get two giant parks that never touch, because they would eventually merge.

But what if the city isn't a perfect grid? What if it's a weird, irregular shape?

• The Old Fear: People worried that on a weird shape, you might get a "Goldilocks" scenario: exactly two or three infinite parks, but not a million.
• The New Discovery: The authors prove that this middle ground is impossible.
  • If you open buildings randomly (with a probability of 50% or less), you will never get a finite number of infinite parks (like 1, 2, or 100).
  • You will either get zero infinite parks, or infinitely many of them.
  • The Analogy: Imagine pouring water into a sponge. Either the water soaks into one giant connected blob, or it gets trapped in millions of tiny, disconnected bubbles. It never forms exactly three giant bubbles.

2. The "Mirror Image" Trick

How did they prove this? They used a clever trick involving a "mirror image."

• The Setup: Imagine you have a city where 50% of buildings are open (parks) and 50% are closed (walls).
• The Logic: If there were exactly one giant park, there would be exactly one giant wall surrounding it. But the authors show that if you have a "positive association" (meaning if one building is open, it makes its neighbors more likely to be open), the math forces a contradiction.
• The Result: If you have even a tiny chance of having one infinite park, the math forces you to have an infinite number of them. It's like a domino effect: if one infinite path exists, the rules of the game force the creation of infinite paths everywhere else.

3. The "Loop" Mystery (The Honeycomb Model)

The second half of the paper applies this "All or Nothing" rule to a famous physics model called the Loop O(n) model.

• The Scene: Imagine a honeycomb grid (like a beehive). Instead of open/closed buildings, you have loops (strings) that can form closed circles.
• The Question: Do these strings form just a few giant loops that go on forever? Or do they form a chaotic mess of infinite loops?
• The Conjecture: In 1982, a physicist named Nienhuis guessed that for certain settings, the honeycomb should be filled with infinitely many loops surrounding every single cell of the grid.
• The Proof: The authors used their "All or Nothing" rule to prove Nienhuis right! They showed that in a specific range of parameters, the honeycomb is indeed filled with an infinite number of loops.
• Why it matters: This confirms a 40-year-old guess about how matter behaves at the "critical point" (the exact moment a material changes state, like ice melting into water).

4. Why This Matters to the Real World

You might ask, "Who cares about infinite loops on a honeycomb?"

This isn't just about math puzzles. These models describe real-world phenomena:

• Magnetism: How tiny magnets align in a piece of iron.
• Fluids: How water flows through porous rock.
• Epidemics: How a disease spreads through a population.

The authors' result is powerful because it works even when the "city" (the graph) is messy, irregular, or random. They didn't need the city to be a perfect grid. They proved that nature has a fundamental rule: At the tipping point, things don't settle into a few big groups; they either stay small or explode into infinite complexity.

Summary in One Sentence

The authors proved that in random 2D systems, you can never have a "moderate" number of infinite structures; it's either none, or an endless ocean of them, a discovery that finally solves a 30-year-old mystery about how loops and magnets behave in nature.

Technical Summary

1. Problem Statement and Context

The paper addresses two fundamental problems in statistical physics and probability theory on planar graphs:

1. The Number of Infinite Components in Planar Percolation: A classical result by Burton and Keane (1989) states that for Bernoulli percolation on amenable transitive graphs, the number of infinite connected components ($N_\infty$) is almost surely either 0 or 1. However, for general planar graphs, it was an open question whether $N_\infty$ could be finite but greater than 1 (i.e., $1 < N_\infty < \infty$). Specifically, Benjamini and Schramm (1996) conjectured that for Bernoulli site percolation at the critical parameter $p=1/2$ on any planar graph, $N_\infty$ cannot be finite and non-zero.
2. The Phase Diagram of the Loop O(n) Model: The Loop O(n) model on the hexagonal lattice is a model of non-intersecting loops. Nienhuis (1982) conjectured a specific phase diagram involving a transition between exponential decay of loop lengths and the existence of macroscopic (infinite) loops. While parts of this diagram were known, a rigorous proof for a large region of parameters (specifically $n \in [1, 2]$ and $x \in [1/\sqrt{2}, 1]$) remained elusive, particularly because the model lacks the FKG (Fortuin-Kasteleyn-Ginibre) positive association property in its standard formulation.

2. Methodology

The authors develop a novel approach that combines stochastic domination, FKG inequalities, and planar topology to bypass the lack of symmetry and positive association in the target models.

A. General Percolation Framework (Theorem 1)

The core of the paper is Theorem 1, which establishes a dichotomy for a broad class of site percolation processes on infinite, locally finite planar graphs. The theorem applies to any random variable $\sigma$ (representing open/closed vertices) satisfying three conditions:

1. Tail Triviality: The probability of any tail event is 0 or 1.
2. Positive Association (FKG): Increasing events are positively correlated.
3. Stochastic Domination: The set of open vertices is stochastically dominated by the set of closed vertices (i.e., $\sigma \preceq 1-\sigma$).

Proof Strategy for Theorem 1:

• Assumption by Contradiction: Assume $1 \le N_\infty < \infty$.
• Active Points: Using tail triviality, the authors show that if there is a finite number of infinite components, there must exist an "active" accumulation point on the sphere (in the embedding) that is approached by an infinite component.
• Arm Events: They construct a sequence of nested domains $\Omega_n$ and partition the boundary $\partial \Omega_n$ into arcs. Using the FKG inequality and the stochastic domination condition, they prove that for any $k$, one can find $2k$ arcs on the boundary that simultaneously connect to infinity via both open and closed paths with probability approaching 1.
• Topological Contradiction: Planar topology dictates that if $2k$ alternating open/closed paths cross an annulus, there must be at least $k+1$ infinite components for the open set and $k+1$ for the closed set. By taking $k \to \infty$, this forces $N_\infty = \infty$, contradicting the assumption that $N_\infty$ is finite.
• Coupling: When $\sigma$ is not symmetric to $1-\sigma$ but is dominated by it, they use Strassen's theorem to construct a monotone coupling $(\sigma, \tau)$ where $\tau \sim 1-\sigma$ and $\sigma \le \tau$, extending the argument.

B. Application to Unimodular Graphs (Corollaries 1.1 & 1.2)

To apply Theorem 1 to random graphs, the authors must rule out the case $N_\infty = \infty$.

• They adapt the Burton-Keane argument (originally for transitive graphs) to unimodular invariantly amenable random graphs.
• They introduce an auxiliary Uniform Spanning Forest (USF) to create "trifurcation points" (vertices where an infinite cluster splits into three).
• Using the Mass Transport Principle (MTP), they show that the existence of infinitely many infinite components would imply a non-zero density of trifurcations, which contradicts the amenability of the graph.
• This extends the result to Divide and Color models, a generalization including Ising and fuzzy Potts models.

C. Application to Loop O(n) Model (Theorem 2)

The Loop O(n) model lacks FKG, so the authors cannot apply Theorem 1 directly.

• Edwards-Sokal Type Representation: They construct a "blocking vertex" percolation $\xi$ on the dual triangular lattice derived from a loop configuration $\omega$.
  • Vertices in loops are open with probability $(n-1)/n$.
  • Isolated vertices are open with probability $1-x^2$.
• Conditional FKG: While the annealed measure (averaged over loops) lacks FKG, the quenched measure (conditioned on the loop configuration) for the blocking vertices $\xi$ satisfies the conditions of Corollary 1.2 (specifically, it is a divide-and-color model with a finitary invariant partition).
• Non-Percolation of Blocking Vertices: For $n \in [1, 2]$ and $x \in [1/\sqrt{2}, 1]$, the parameters ensure the blocking vertices are stochastically dominated by $p=1/2$ percolation. Thus, $\xi$ does not percolate (no infinite clusters).
• Contradiction: If the loops did not surround every face infinitely often, the blocking vertices would form infinite clusters (via a domain wall argument). Since $\xi$ does not percolate, the loops must surround every face infinitely often.

3. Key Contributions and Results

1. Resolution of Conjecture 8 (Benjamini-Schramm, 1996):

   • Proved that for Bernoulli site percolation at $p=1/2$ on any infinite locally finite planar graph, the number of infinite components is either 0 or $\infty$. It is impossible to have a finite number $>0$.
   • This holds without requiring transitivity or specific embedding properties (e.g., no accumulation points).

2. Percolation Threshold Bound:

   • Established that for any invariantly amenable unimodular random rooted planar graph, the critical threshold $p_c \ge 1/2$.
   • This improves upon previous uniform lower bounds (e.g., Peled, 2020) and generalizes non-coexistence theorems.

3. Phase Diagram of Loop O(n) Model:

   • Proved that for $n \in [1, 2]$ and $x \in [1/\sqrt{2}, 1]$, every face of the hexagonal lattice is surrounded by infinitely many loops almost surely.
   • This confirms a significant portion of Nienhuis's 1982 conjecture regarding the existence of macroscopic loops.
   • The result holds in a region where the model is not known to have an FKG representation, marking a breakthrough in handling non-monotonic models.

4. Methodological Innovation:

   • Introduced a technique of applying positive association (FKG) to conditional (quenched) distributions rather than the full measure.
   • Demonstrated that stochastic domination by the complement ($1-\sigma$) combined with FKG is sufficient to force the $0$ or $\infty$ dichotomy in planar percolation, removing the need for symmetry assumptions.

4. Significance

• Theoretical Impact: The paper resolves a long-standing conjecture in percolation theory regarding the behavior of infinite components on general planar graphs. It bridges the gap between transitive lattice models and general unimodular random graphs.
• Model Generality: By decoupling the proof from the specific symmetries of the lattice and relying on topological and probabilistic properties (FKG + Domination), the method is robust and applicable to a wide class of models, including those without standard graphical representations.
• Loop O(n) Breakthrough: The confirmation of the macroscopic loop phase for $n \in [1, 2]$ is a major step toward understanding the conformal field theory limits of these models. It provides the first rigorous proof of macroscopic behavior in a large, non-perturbative region of the parameter space for the Loop O(n) model.
• Future Applications: The technique of using conditional FKG and stochastic domination is highlighted as a powerful tool that could be applied to other complex statistical mechanics models where global monotonicity fails but local conditional structures remain tractable.

In summary, this work provides a rigorous foundation for the phase transition behavior of planar percolation and loop models, settling key conjectures and introducing a versatile methodology for analyzing systems with complex dependencies.