🔢 mathematics

Central limit theorems for non-linear functionals of Gaussian fields via Wiener chaos decomposition

This paper establishes a Central Limit Theorem for non-linear functionals of discrete Gaussian fields on the $d$ -dimensional lattice using Wiener chaos decomposition and the fourth moment theorem, demonstrating that even powers of the discrete Gaussian Free Field converge to Gaussian white noise while odd powers converge to a continuous Gaussian Free Field with explicit covariance.

Original authors: Fabio Coppini, Wioletta M. Ruszel

Published 2026-07-14

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Fabio Coppini, Wioletta M. Ruszel

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 you have a giant, invisible grid stretching out in every direction, like a 3D checkerboard made of pure math. On every single square of this grid sits a tiny, jittery number. These numbers are "Gaussian," which means they wiggle around zero in a very predictable, bell-curve way, but they are also connected to their neighbors. If one square jumps up, its neighbors are likely to jump up too, depending on how far apart they are.

Now, imagine you want to study what happens when you squint at this whole grid and look at the "average" behavior of these numbers, but with a twist: you decide to raise every single number to a power. Maybe you square them, cube them, or raise them to the 10th power. This turns your simple, wiggly numbers into something much more complex and non-linear.

The big question the authors, Fabio Coppini and Wioletta Ruszel, asked is: If we zoom out and look at this grid from far away, does this messy, powered-up chaos settle down into something smooth and predictable?

The Main Discovery: The Great Sorting Hat

The paper proves that yes, it does settle down, but the result depends entirely on whether you chose an even power or an odd power. It's like a magical sorting hat that treats even and odd numbers differently.

The Even Powers (The White Noise):
If you take an even power (like $x^2$ , $x^4$ , $x^6$ ), the chaos smooths out into what mathematicians call Gaussian white noise.
Think of white noise like static on an old TV screen. It's random, it's everywhere, and it has no memory of its past. If you look at two different spots on this "static," they don't care about each other at all. The paper shows that for even powers, the grid's behavior becomes this pure, unconnected static. This happens as long as the grid is big enough (specifically, in dimensions $d \ge 5$ for the field itself, or $d \ge 3$ if you look at the changes between neighbors).

The Odd Powers (The Ghostly Echo):
If you take an odd power (like $x^1$ , $x^3$ , $x^5$ ), the result is totally different. It doesn't become static. Instead, it turns into a continuous Gaussian Free Field.
Imagine this as a "ghostly echo" of the original grid. Unlike the static, this new field remembers its neighbors. If one part of the field moves, the parts nearby feel it. The paper proves that these odd powers converge to a smooth, connected field where the "distance" between points is measured by a specific mathematical rule called the Green's function.

How They Knew This Was True

The authors didn't just guess or run computer simulations; they built a rigorous mathematical proof. They used a powerful tool called Wiener chaos decomposition.

To understand this, imagine you have a complex song. You can break that song down into individual notes (frequencies). The authors did the same thing with their grid. They broke the complex "powered-up" numbers into a sum of simpler building blocks called Hermite polynomials.

They then used a famous rule called the Fourth Moment Theorem (discovered by Nualart and Peccati). Think of this theorem as a magic detector. It says: "If you have a bunch of these building blocks, and you check their 'fourth moment' (a specific way of measuring their shape), and it looks exactly like a perfect bell curve, then the whole thing is a bell curve."

By checking this condition for their grid, they proved mathematically that the limit is indeed a Gaussian distribution. They also proved that the "tightness" holds, which is a fancy way of saying the grid doesn't explode or behave wildly as it gets bigger; it stays well-behaved enough to have a limit.

What They Explicitly Ruled Out

The paper is very careful to say what doesn't happen.

No "One Size Fits All": They explicitly rule out the idea that all powers behave the same way. You cannot assume that raising the numbers to a power will always result in white noise. The odd powers are a distinct exception that creates a connected field, not static.
No "Infinite Susceptibility" for Even Powers: For the even powers to turn into white noise, the connections between the grid points (the covariance) must die off fast enough. If the grid is too "sticky" (infinite susceptibility), the theorem doesn't apply. The paper shows that for the standard Gaussian Free Field, this condition is met for even powers in high dimensions, but for odd powers, the connections are too strong, forcing the different limit.

The Bottom Line

The authors have proven that for a stationary grid of random numbers:

Even powers ( $x^{2p}$ ) $\rightarrow$ White Noise (Random, unconnected static).
Odd powers ( $x^{2p+1}$ ) $\rightarrow$ Gaussian Free Field (Smooth, connected field with long-range memory).

They didn't just suggest this; they proved it using the Fourth Moment Theorem and tightness arguments, showing that this behavior is a fundamental property of how these grids behave when you zoom out. It's a new way of looking at old problems, proving that the "parity" (even vs. odd) of your math operation completely changes the nature of the universe you see when you step back.

Technical Summary: Central Limit Theorems for Non-linear Functionals of Gaussian Fields via Wiener Chaos Decomposition

Problem Statement
The paper addresses the asymptotic behavior of non-linear functionals of stationary discrete Gaussian fields defined on the $d$ -dimensional lattice $\mathbb{Z}^d$ . Specifically, it investigates the convergence of the field $\Phi_N(j) = H(X_j)$ , where $X$ is a centered stationary Gaussian field with covariance $\rho$ , and $H: \mathbb{R} \to \mathbb{R}$ is a non-linear function with finite variance. The study focuses on the action of this field on test functions $f$ over a domain $D = [-1/2, 1/2]^d$ , defined as $\langle \Phi_N, f \rangle = N^{-d} \sum_{j \in B_N} H(X_j) f(j/N)$ . The central question is to establish conditions under which the rescaled field $N^{d/2}\Phi_N$ converges in distribution to a Gaussian limit (specifically Gaussian white noise or a continuous Gaussian Free Field) within the space of distributions $H^{-\alpha}(D)$ .

Methodology
The authors employ the Wiener chaos decomposition and the Fourth Moment Theorem (Nualart-Peccati, 2005) as their primary analytical tools.

Hermite Expansion: The function $H$ is decomposed into Hermite polynomials $H = \sum_{q=m}^\infty c_q H_q$ , where $m$ is the Hermite rank. This allows the field $\Phi_N$ to be expressed as a sum of orthogonal components $S_{N,q}(f)$ , each corresponding to the $q$ -th Wiener chaos.
Isomorphism and Representation: The stationary Gaussian field $X$ is represented via an isonormal Gaussian process $W$ on a Hilbert space $\mathcal{H}$ . This facilitates the expression of the chaos components as multiple Wiener-Itô integrals.
Fourth Moment Theorem: Instead of using the classical method of moments (which relies on Feynman diagrams and combinatorial estimates), the authors utilize the Nualart-Peccati characterization. This theorem states that a sequence of random variables in a fixed Wiener chaos converges to a Gaussian distribution if and only if their fourth moments converge to 3 (or equivalently, if the contraction norms of the kernels vanish).
Tightness: The authors establish tightness in the negative Sobolev space $H^{-\alpha}(D)$ by bounding the $L^2$ norm of the process using the spectral properties of the Laplacian and the orthogonality of the Hermite expansion.

Key Contributions and Results

Main Theorem (Theorem 1): The paper proves a Central Limit Theorem (CLT) for the field $\Phi_N$ as a distribution. Under the condition that $\sum_{u \in \mathbb{Z}^d} |\rho(u)|^m < \infty$ , the rescaled field $N^{d/2}\Phi_N$ converges in law in $H^{-\alpha}(D)$ (for $\alpha > d/2$ ) to a centered Gaussian white noise $W$ scaled by a constant $C_m$ . This result generalizes the classical Breuer-Major theorem, which typically addresses the convergence of finite-dimensional distributions of sums, to the convergence of the field itself as a random distribution.
Proof Technique Innovation: The authors explicitly note that their approach avoids the combinatorial estimates and Feynman diagram calculations typical of original Breuer-Major proofs. Instead, they rely on the Fourth Moment Theorem to directly express finite-dimensional distributions as Itô integrals, a method described as "new" for discrete lattice configurations.
Tightness Result (Lemma 4): A tightness result for the sequence $\Phi_N$ in $H^{-\alpha}(D)$ is provided, which the authors claim is missing in existing literature for this specific setting.
Application to Gaussian Free Field (GFF):
- Even Powers: For $H(x) = x^{2p}$ , the Hermite rank is $m \ge 2$ . The paper shows that for the discrete GFF (where covariance decays as $|u|^{-(d-2)}$ ), even powers converge to Gaussian white noise provided $d \ge 5$ (or $d \ge 3$ for the gradient of the GFF).
- Odd Powers: For $H(x) = x^{2p+1}$ , the Hermite rank is $m=1$ . The summability condition $\sum |\rho(u)| < \infty$ fails for the GFF in dimensions $d \ge 3$ . Consequently, the limit is not white noise. Instead, Theorem 3 establishes that the centered field converges to a continuous Gaussian Free Field with an explicit covariance structure derived from the continuous Green's function.

Significance and Claims
The paper positions itself as a bridge between the well-developed theory of non-linear functionals of Gaussian vectors (Breuer-Major theorems) and the study of Gaussian fields on lattices.

Novelty: The authors claim that while the tools (Wiener chaos, Fourth Moment Theorem) are standard in stochastic analysis, their application to discrete lattice configurations to prove field convergence appears to be new.
Generalization: The result generalizes the original Breuer-Major theorem by addressing the convergence of the field $\Phi_N$ as a distribution rather than just the finite-dimensional distributions of partial sums.
Modesty: The authors acknowledge that the literature on Breuer-Major theorems is vast and that their work does not introduce new probabilistic tools but rather applies existing ones in a new context. They explicitly refrain from extending the tightness result to more general Besov spaces due to technical complexity, noting that their current Hilbert space approach is sufficient for the stated results.
Applications: The paper highlights the specific application to powers of the GFF, distinguishing the behavior of even and odd powers, and notes that the convergence of even powers to white noise is a direct consequence of their main theorem, whereas the odd power case requires a distinct analysis leading to a continuous GFF limit.

The Main Discovery: The Great Sorting Hat

How They Knew This Was True

What They Explicitly Ruled Out

The Bottom Line

Technical Summary: Central Limit Theorems for Non-linear Functionals of Gaussian Fields via Wiener Chaos Decomposition

More like this