🔢 mathematics

The $q<1$ Random-Cluster Model on Wired Trees: Uniqueness and Negative Dependence

This paper establishes the uniqueness of the supercritical wired Gibbs measure and proves negative dependence among branch-connectivity indicators for the random-cluster model with cluster weight $q<1$ on wired trees across all parameters $p$ .

Original authors: Heehyun Park

Published 2026-08-11

📖 3 min read🧠 Deep dive

CC BY 4.0

Original authors: Heehyun Park

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, invisible web of connections, like a giant spiderweb made of rubber bands and knots. In the world of probability and physics, scientists use a tool called the "random-cluster model" to study how things stick together or fall apart. Think of it as a game where you have a bunch of dots (vertices) and you randomly decide whether to snap a rubber band (an edge) between them or leave it broken. Sometimes, the dots just snap together on their own, like independent friends making plans. But in this specific game, there's a twist: the more groups of connected dots you have, the more "points" you get, or the fewer points you get, depending on a secret number called $q$ .

When $q$ is a big number (like 2 or 3), the game is friendly. If one part of the web gets connected, it makes it more likely for other parts to connect too. It's like a crowd at a concert: if one person starts dancing, everyone else is more likely to join in. This is called "positive dependence." But when $q$ is a tiny number, specifically between 0 and 1, the rules flip. The game becomes competitive. If one group of dots connects, it might actually make it less likely for another group to connect, as if they are fighting over a limited supply of energy. Scientists have long suspected that in this "tiny $q$ " world, everything pushes against each other, but proving exactly how they push has been a massive headache because the usual mathematical tools stop working.

This paper, written by Heehyun Park, dives deep into this tricky "tiny $q$ " world, but only on a very specific, perfectly symmetrical playground: an infinite tree where every branch splits into the same number of smaller branches. The author asks two big questions: First, if we force the very tips of this tree to be glued together (a condition called "wired"), is there only one way the whole tree can behave, or could it be chaotic with multiple possibilities? Second, does the "pushing apart" (negative dependence) actually happen between different branches of the tree?

The paper answers both questions with a resounding "yes." The author proves that on this infinite tree, there is exactly one unique way the system settles down, no matter how you look at it, provided the connection probability is high enough. Before this, scientists knew what happened when the connection probability was low (it was just random, independent snapping), but the high-probability zone was a mystery. Park shows that in this high-probability zone, the tree doesn't get confused; it snaps into a single, predictable state where a giant connected path stretches out to infinity.

Furthermore, the paper confirms the "pushing apart" theory. It proves that if you look at two different branches coming off the same spot, the chance that both branches are connected to the center is actually lower than if they were independent. They are negatively correlated. It's like a crowded room where if one person grabs a microphone, it becomes slightly less likely that a person in a different corner will also grab one, because the "energy" is being used up. The author doesn't just guess this; they build a rigorous mathematical proof using a clever "message-passing" system, showing that the branches effectively compete with each other. This work completes the map of how this specific model behaves, turning a foggy, uncertain region of physics into a clear, proven landscape.

Technical Summary: The $q < 1$ Random-Cluster Model on Wired Trees: Uniqueness and Negative Dependence

Problem Statement
The random-cluster model, a unifying framework for percolation and Potts models, exhibits qualitatively different behaviors depending on the cluster weight parameter $q$ . For $q \ge 1$ , the model satisfies the Fortuin-Kasteleyn-Ginibre (FKG) inequality, ensuring positive association and monotonicity of boundary conditions. However, for $0 < q < 1$ , the model is expected to exhibit negative dependence. While pairwise negative correlation is conjectured for general graphs, it remains an open problem. Specifically, on infinite trees with wired boundary conditions (where all boundary leaves are identified), the phase diagram and uniqueness of the Gibbs measure for $q < 1$ in the supercritical regime ( $p > p_c$ ) were not fully established. Previous work by Grimmett and Häggström addressed $q \ge 1$ , identifying uniqueness intervals and phase transitions, but the $q < 1$ regime required new tools due to the absence of FKG and stochastic domination.

Methodology
The paper employs a combination of message recursion analysis, fixed-point theory, and probabilistic coupling techniques tailored to the tree structure.

Message Recursion and Fixed Points: The author defines a "message" $f(u)$ at each vertex $u$ , representing a normalized ratio of partition functions conditioned on the vertex being connected or disconnected from the wired boundary. On a $\Delta$ -regular tree, these messages evolve via a scalar recursion $\Phi(x)$ raised to the power $d = \Delta - 1$ . The paper analyzes the fixed points of this map $\phi(x) = \Phi(x)^d$ .
- For $p \le p_c$ (where $p_c = q/(\Delta + q - 2)$ ), the map has a unique fixed point at $x=1$ , corresponding to zero connection probability to the wired boundary.
- For $p > p_c$ , $x=1$ becomes unstable, and a unique non-trivial attracting fixed point $x^* > 1$ emerges.
Weak Spatial Mixing via Amplification: To prove uniqueness in the supercritical regime without FKG, the paper establishes weak spatial mixing under "positive-density wired" boundary conditions.
- Amplification (Lemma 4.3): A key technical innovation replaces the stochastic domination arguments used for $q \ge 1$ . The author proves that a positive density of wired leaves on the boundary forces the messages at internal vertices to be uniformly bounded away from the unstable fixed point $1$.
- Contraction (Lemma 4.5): Once messages are bounded away from $1$, the contraction properties of the map $\phi$ around the stable fixed point $x^*$ ensure that the influence of the specific boundary configuration decays exponentially with distance.
Uniform Branch Survival: A critical bridge lemma (Lemma 5.7) demonstrates that for any infinite-volume wired Gibbs measure, the probability that a boundary vertex is connected to infinity through its exterior branch is uniformly bounded below by a constant $\theta > 0$ , regardless of the exterior configuration. This ensures that any wired DLR measure induces a "positive-density" boundary condition on distant balls, allowing the spatial mixing estimates to apply.
Negative Dependence via Exchangeability: For the negative dependence results, the paper analyzes the joint law of connectivity indicators for parallel branches.
- It establishes that the rank sequence of the connectivity vector is ultra-log-concave (ULC).
- Using Pemantle's criterion, this implies the Conditional Negative Association with positive external fields (CNA+) property for the connectivity indicators.
- A series-parallel argument (Lemma 6.3) shows that conditioning on terminal connection stochastically dominates conditioning on disconnection within each branch.
- Combining CNA+ for indicators with the stochastic ordering of branch configurations yields negative covariance for arbitrary increasing observables supported on disjoint branches.

Key Contributions and Results

Complete Wired DLR Phase Diagram ( $0 < q < 1$ ):
The paper proves that the wired DLR specification on the infinite $\Delta$ -regular tree admits a unique infinite-volume Gibbs measure for all $p \in (0, 1)$ .
- Subcritical/Critical ( $p \le p_c$ ): The unique measure is the product Bernoulli bond percolation with parameter $b_p = p / (p + q(1-p))$ . In this regime, there is no percolation.
- Supercritical ( $p > p_c$ ): The unique measure is the local limit of finite-volume measures with all-wired boundary conditions. This measure exhibits percolation ( $\mu^*(\rho \leftrightarrow \infty) > 0$ ).
- Significance: This resolves the supercritical uniqueness problem for $q < 1$ , a gap left by previous literature which relied on FKG. The proof shows that the "all-wired" limit is the only wired DLR state, even though the "free" limit (product measure) also exists and percolates in this regime, meaning strong uniqueness across boundary conventions fails.
Branch-Separated Negative Dependence:
The paper establishes that for the unique wired Gibbs measure, bounded increasing observables supported on disjoint collections of branches incident to a common vertex have non-positive covariance.
- Specifically, the vector of branch-connectivity indicators satisfies the CNA+ property.
- The covariance is strictly negative in the supercritical regime and zero in the subcritical/critical regime (where the measure is a product).
- Significance: This provides a rigorous form of negative dependence for the $q < 1$ model on trees, advancing beyond pairwise correlation to a structural property for disjoint branches, though it does not establish full negative association for arbitrary disjoint edge sets.

Significance and Claims
The paper claims to provide the first complete classification of the wired DLR phase diagram for the random-cluster model with $0 < q < 1$ on regular trees. It explicitly states that it does not prove the general Kahn-Neiman conjecture (that all $q < 1$ random-cluster measures are NA+), but rather establishes a "branch-separated" version of negative association.

The work is significant because it replaces the standard FKG-based tools (unavailable for $q < 1$ ) with a new "amplification" and "branch-survival" architecture. This allows for the proof of uniqueness in the supercritical regime and the derivation of negative dependence properties. The results unify the phase structure with the algebraic properties of the model (log-concavity of rank sequences) and clarify the behavior of the model on trees, which serves as a foundational setting for understanding phase transitions on sparse random graphs. The paper notes that while recent lattice results have improved stochastic comparisons, this work provides the exact wired-tree DLR classification.

More like this