Gaussian free field convergence of the six-vertex model with $-1\leq\Delta\leq-\frac12$

The paper proves that the height function of the six-vertex model on $\mathbb{Z}^2$ with spectral parameter $\Delta \in [-1, -1/2]$ converges to a full-plane Gaussian free field in the scaling limit, a result that extends to anisotropic weights via a suitable lattice embedding.

The Gist

Imagine a giant, infinite checkerboard stretching out forever in every direction. On this board, we place little arrows on every edge, but with one strict rule: at every intersection (vertex), exactly two arrows must point in, and two must point out. This is the Six-Vertex Model.

Think of these arrows not just as directions, but as a secret code. If you follow the arrows, they trace out paths that look like a tangled web of rivers. Now, imagine that this web of rivers is actually the contour lines of a mountain range. Where the arrows flow, the "height" of the mountain changes. This is the Height Function.

For a long time, mathematicians knew that if you zoomed out far enough, this jagged, pixelated mountain range should smooth out into a perfectly random, wavy surface called a Gaussian Free Field (GFF). Think of the GFF as the mathematical equivalent of a calm, wind-swept ocean surface or a perfectly random static noise on an old TV screen. It's the "universal" shape that many different physical systems settle into when they are at a critical tipping point.

However, proving this for the Six-Vertex Model was like trying to prove that a specific, very stubborn type of knot untangles into a perfect circle. It was easy for simple knots (free fermions), but this model is a "genuinely interacting" knot—it's too complex to untangle with old tricks.

The Big Breakthrough

This paper, by Hugo Duminil-Copin and his team, finally proves that yes, this specific mountain range does smooth out into that perfect random ocean surface, but only under certain conditions (specifically, when the "weight" of the arrows, denoted by $\Delta$, is between -1 and -0.5).

Here is how they did it, explained through simple analogies:

1. The "Black Box" Ingredients

The authors didn't try to solve the whole puzzle at once. Instead, they built a machine using four special "ingredients" (black boxes) that they knew worked:

• Ingredient 1: Rotational Symmetry (The Spinning Top).
  Imagine spinning a top. No matter how you rotate it, it looks the same. The authors proved that if you look at the correlations (how the height at one point relates to another) from a distance, the pattern looks the same no matter which way you turn the board. It's perfectly round and symmetric.
• Ingredient 2: A Glimpse of Scale (The Zoom Lens).
  If you zoom in or out on a fractal, it usually looks the same. The authors couldn't prove the whole thing was scale-invariant, but they found a specific "observable" (a specific measurement) that did behave predictably when zoomed. This gave them a clue about the size of the waves in the final ocean.
• Ingredient 3: Regularity (The Smoothness Rule).
  They proved that the mountain range can't have jagged, impossible spikes. It has to be "smooth" in a statistical sense. This is like saying a real mountain can't have a cliff that goes straight up for a mile and then drops instantly; it has to follow certain natural laws. This allowed them to use powerful math tools to say, "If the pattern exists, it must be this specific type."
• Ingredient 4: Spectral Representation (The DNA of the System).
  This is the most technical part. Think of the system's behavior as a song. The "spectrum" is the list of all the musical notes (frequencies) that make up that song. The authors showed that the height function is made of a specific combination of these notes. By analyzing the "DNA" of the system (the eigenvalues of a transfer matrix), they could predict how the song sounds when played very slowly (the large-scale limit).

2. The Detective Work: The Dichotomy

The proof uses a clever "either/or" strategy, like a detective narrowing down suspects.

• Step 1: They looked at the two-point correlation (how two points relate). They showed that as you zoom out, the pattern must either look like the GFF ocean, or it must look like two different versions of the GFF ocean depending on how you zoom.
• Step 2: They proved that if the two-point pattern looks like the GFF, then all the complex multi-point patterns (involving 3, 4, or 100 points) must also look like the GFF.
• Step 3: They used the "DNA" (Ingredient 4) and the "Smoothness" (Ingredient 3) to show that the pattern is harmonic (it satisfies the same equations as a soap bubble). This forces the pattern to be exactly the GFF.
• Step 4: Finally, they used the "Zoom Lens" (Ingredient 2) to calculate exactly how big the waves are. They found a precise formula for the variance ($\sigma^2$) based on the arrow weights.

3. Why This Matters

Before this paper, we knew that simple, non-interacting systems turned into the GFF. But the Six-Vertex Model is a "genuinely interacting" system—it's like a crowd of people where everyone is reacting to everyone else. Proving that this complex crowd still settles into the same universal shape as a simple crowd is a massive deal.

It confirms a deep prediction in physics: Universality. It means that the microscopic details (the specific rules of the arrows) don't matter as much as we thought. As long as you are at the critical tipping point, the macroscopic world looks the same: a beautiful, random, wavy Gaussian Free Field.

The Takeaway

The authors took a complex, interacting grid of arrows, used a mix of geometry, probability, and spectral analysis to show that it behaves like a perfect, random ocean surface. They didn't just guess; they built a rigorous mathematical bridge from the microscopic rules to the macroscopic reality, finally solving a problem that had been open for decades.

In short: They proved that a complex, tangled web of rules inevitably smooths out into the most beautiful, random wave pattern nature can create.

Technical Summary

Technical Summary: Gaussian Free Field Convergence of the Six-Vertex Model with $-1 \le \Delta \le -1/2$

1. Problem Statement

The paper addresses the scaling limit of the isotropic six-vertex model on the square lattice $\mathbb{Z}^2$ in the regime where the spectral parameter $\Delta \in [-1, -1/2]$. This corresponds to weights $a=b=1$ and $c \in [\sqrt{3}, 2]$.

The central question is whether the height function $h$ associated with the six-vertex configurations converges, as the mesh size $\delta \to 0$, to a Gaussian Free Field (GFF). While this convergence was previously established for the "free fermion" point ($\Delta=0, c=\sqrt{2}$) and for perturbations thereof, this regime represents a genuinely interacting system where no simple mapping to free fermions exists. The authors aim to prove that the height function converges to a properly scaled GFF, $\sigma \cdot \text{GFF}$, and to determine the explicit scaling factor $\sigma$ in terms of the model parameters.

2. Methodology

The proof synthesizes two major historical approaches in statistical mechanics: the transfer matrix formalism (spectral analysis) and discrete holomorphicity/percolation theory (geometric estimates). The authors avoid relying on the full integrability of the Bethe Ansatz for the entire proof, instead using it only for specific inputs (like free energy derivatives) while establishing the scaling limit via robust probabilistic estimates.

The methodology is structured into four main "ingredients" used as black boxes in the main proof:

A. Spectral Representation (Ingredient 4)

The authors utilize the transfer matrix formalism on a cylinder to derive a spectral representation for correlation functions.

• They express the two-point correlation function $\Phi_2$ as an integral against a spectral measure $\mu_L$ derived from the eigenvalues of the transfer matrix and the shift operator.
• By taking the infinite-volume and scaling limits, they show that any subsequential limit of the correlation function is determined by a limiting measure $\mu$.
• Crucially, they relate the harmonicity of the limit to the concentration of this measure on the set where the spectral parameters satisfy $|b| = a$.

B. Rotational Invariance (Ingredient 1)

Relying on prior work connecting the six-vertex model to the critical random-cluster model (via Baxter-Kelland-Wu correspondence), the authors invoke asymptotic rotational invariance of the $k$-point correlation functions.

• This property implies that any scaling limit must be invariant under rotations.
• Combined with the spectral representation, this forces the spectral measure to concentrate on specific sub-manifolds, effectively constraining the possible limiting behaviors.

C. Regularity and Qualitative Estimates (Ingredient 3)

To handle the lack of exact discrete holomorphicity in this interacting regime, the authors employ Russo-Seymour-Welsh (RSW) theory adapted to the six-vertex model via a spin representation.

• Spin Representation: The height function modulo 4 is encoded into a pair of spin configurations on dual sublattices, satisfying the Fortuin-Kasteleyn-Ginibre (FKG) inequality for $c \ge 1$.
• Circuit Estimates: They establish uniform lower bounds on the probability of alternating circuits (loops of $\omega^+$ and $\omega^-$) in annuli.
• Regularity: These estimates provide bounds on the correlation functions (decay rates) and ensure precompactness of the family of correlation functions, allowing the extraction of convergent subsequences.
• Mixing: They prove that correlations decay sufficiently fast to allow for the identification of limiting structures.

D. Scale Invariance via Free Energy (Ingredient 2)

To uniquely identify the variance $\sigma^2$ of the limiting GFF, the authors connect the scaling limit to the free energy of the model.

• They use the fact that the free energy $f(s)$ is twice differentiable at zero slope (a result derived from Bethe Ansatz techniques in prior literature).
• They establish a correspondence between the second derivative $f''(0)$ and the variance of the GFF limit, effectively using a large deviation principle to link the macroscopic free energy to the microscopic fluctuations.

3. Key Contributions and Results

Main Theorem (Theorem 2.8)

The primary result is the proof that for the isotropic six-vertex model with $a=b=1$ and $c \in [\sqrt{3}, 2]$ (i.e., $\Delta \in [-1, -1/2]$), the scaled height function $h^{(\delta)}$ converges in law to a Gaussian Free Field with variance $\sigma^2$:
$$\sigma^2 = \frac{2}{\arccos \Delta} = \frac{1}{\arcsin(c/2)}$$
The convergence is established in three modes:

1. Multi-point correlation functions: Uniform convergence on compact sets.
2. Finite-dimensional marginals: Weak convergence to a Gaussian distribution.
3. Distributional convergence: Convergence in negative regularity Hölder spaces ($C^\alpha$ for $\alpha \in (-1, 0)$) and Sobolev/Besov spaces.

Extension to Anisotropic Case (Theorem 3.3)

The result is extended to the anisotropic case ($a \neq b$) by embedding the lattice into a rotated geometry ($L_\theta \mathbb{Z}^2$). While full Hölder convergence is not proven for the anisotropic case due to technical difficulties in establishing RSW estimates without rotational symmetry, the convergence of correlation functions and finite-dimensional marginals is established.

Dichotomy and Uniqueness

A significant technical contribution is the proof of a dichotomy for the two-point function. The authors show that any subsequential limit must either be a multiple of the GFF or a mixture of two distinct GFFs. By combining this with the spectral representation and rotational invariance, they prove that the limit is unique and corresponds to a single GFF with the specific variance derived from the free energy.

Critical Exponents

As a direct application, the paper derives the existence and values of critical exponents for the associated random-cluster model (FK percolation) with cluster weight $q \in [1, 4]$. This includes the one-arm, two-arm, and energy exponents, which were previously unknown for $q=3$ and $q=4$ (corresponding to 3-state and 4-state Potts models).

4. Significance

• Beyond Free Fermions: This is the first rigorous proof of GFF convergence for a genuinely interacting integrable model across a substantial range of parameters. It moves beyond the "free fermion" point ($\Delta=0$) where exact solvability via free fermions was available.
• New Proof Paradigm: The paper introduces a novel framework that bypasses the need for exact discrete holomorphic observables (which do not exist in this regime). Instead, it uses a priori estimates (RSW theory) to extract subsequential limits and then uses spectral analysis and rotational invariance to identify the limit. This "soft" approach combined with "hard" spectral inputs is a major methodological advance.
• Universality: The results confirm the universality conjecture for 2D statistical mechanics models at continuous phase transitions, showing that the six-vertex model belongs to the $c=1$ universality class (Free Boson/GFF) in this regime.
• Impact on Potts Models: By establishing the link to the random-cluster model, the results provide the first rigorous derivation of critical exponents for the 3-state and 4-state Potts models in the continuum limit.

In summary, the paper provides a rigorous bridge between the microscopic interacting lattice model and the macroscopic Gaussian Free Field, utilizing a sophisticated blend of spectral theory, percolation estimates, and conformal invariance principles to solve a long-standing problem in statistical mechanics.