Abstract
We study a two-dimensional massless field in a box with potential $V(\nabla \phi(\cdot))$ and zero boundary conditions, where $V$ is any symmetric and uniformly convex function. While previous work by Naddaf-Spencer and Miller established that macroscopic averages of this field converge to a continuum Gaussian free field, this paper focuses on the microscopic extremal behavior. We prove that the distribution of the local marginal $\phi(x)$ for any $x$ in the bulk possesses a Gaussian tail. Furthermore, we characterize the leading order of the maximum and the dimension of high points, generalizing results previously established for the discrete Gaussian free field (DGFF) to a broader class of Ginzburg-Landau fields.
Introduction
In the study of random surfaces and statistical mechanics, the Discrete Gaussian Free Field (DGFF) serves as a canonical model. Its extremal properties—specifically the behavior of its maximum and the distribution of its "high points"—are well-understood. However, the DGFF is a specific case where the energy potential is strictly quadratic.
This research extends these findings to Ginzburg-Landau fields, which are lattice gradient Gibbs measures defined by a more general potential $V$. Here, $V$ is required only to be symmetric and uniformly convex. While it has been known that these fields behave like Gaussian fields on a macroscopic scale, the local, microscopic behavior—particularly at the extremes—remained an open question. This paper bridges that gap, demonstrating that the log-correlated structure of these fields drives a universal behavior regardless of the specific shape of the potential.
The Ginzburg-Landau Model
The field is defined on a two-dimensional lattice $D_N = [-N, N]^2 \cap \mathbb{Z}^2$ with zero boundary conditions. The probability measure is proportional to $\exp(-\sum V(\nabla \phi))$, where the sum is over nearest-neighbor bonds. The uniform convexity of $V$ ensures the stability of the field, but the non-quadratic nature of the potential introduces significant dependencies that are absent in the standard DGFF.
A critical concept in this analysis is the effective stiffness, denoted as $g(V)$. This constant represents the variance of the Gaussian field that the Ginzburg-Landau field converges to macroscopically. One of the primary contributions of this work is showing that this same constant $g$ governs the microscopic extremes.
Multiscale Analysis and Decoupling
The primary challenge in analyzing the maximum of these fields is their log-correlated nature: the height at any given point is strongly influenced by the heights of surrounding points across multiple spatial scales. To address this, the authors utilize a multiscale decomposition.
Harmonic Averages and Coupling
The field is analyzed through averages over concentric circles, which allows for the tracking of height accumulation from the boundary to the bulk. In the Gaussian case, increments between these scales are independent. For Ginzburg-Landau fields, this independence is lost. The authors employ "Approximate Harmonic Coupling," a technique that shows these increments decouple asymptotically. This allows the height growth to be modeled as an approximate random walk, making the application of probabilistic tools like the first and second moment methods feasible.
Tail Estimates
The authors prove that the local marginal distribution $\phi(x)$ in the bulk has a Gaussian tail. Specifically, for a point $v$, the probability of exceeding a height $u$ is bounded by: $$P(\phi(v) \ge u) \le \exp\left( -\frac{u^2}{2g \log(\text{dist}(v, \partial D_N))} + \dots \right)$$ This result is pivotal, as it suggests that even with a non-quadratic potential, the "energy cost" of high peaks eventually scales quadratically, mirroring the Gaussian case.
Main Results
1. Law of Large Numbers for the Maximum
The study establishes that the maximum value of the field, when normalized by the logarithm of the lattice size $N$, converges in $L^2$ to a constant determined by the effective stiffness: $$\frac{\max_{v \in D_N} \phi(v)}{\log N} \xrightarrow{N \to \infty} 2\sqrt{g}$$ This confirms that the leading-order behavior of the maximum is identical to that of a DGFF with variance $g$.
2. Fractal Dimension of High Points
The authors also characterize the set of "high points" $H_N(\eta)$, defined as the set of vertices where the field exceeds a threshold $\eta \cdot 2\sqrt{g} \log N$ for $\eta \in (0, 1)$. They show that the size of this set scales as $N^{2(1-\eta^2)}$. This matches the fractal dimension observed in the DGFF, further reinforcing the universality of log-correlated fields.
Discussion and Future Directions
The findings suggest a deep universality: the extremal behavior of 2D random surfaces is primarily determined by the log-correlation of the field rather than the specific local interactions defined by the potential $V$.
However, several questions remain. While the existence of the effective stiffness $g$ is proven, calculating its value for a specific non-quadratic $V$ remains a difficult problem in homogenization theory. Additionally, while this paper identifies the leading-order term of the maximum, the sub-leading order (the "log-log" correction) is yet to be rigorously established for general Ginzburg-Landau fields. Finally, extending these multiscale techniques to other models, such as dimer models or fields with non-convex potentials, represents the next frontier in the study of random surfaces.
Summary updated