Universality for fluctuations of counting statistics of random normal matrices

This paper establishes the universality of the asymptotic variance for the fluctuations of eigenvalue counting statistics in random normal matrices, demonstrating that for sets strictly inside the droplet the variance scales with the boundary length weighted by the potential's Laplacian, while for microscopic dilations of the droplet it generalizes to a formula involving the harmonic measure at infinity.

The Gist

The Big Picture: A Crowd of Repelling Particles

Imagine a giant, invisible dance floor (the complex plane). On this floor, we drop $n$ tiny, charged particles. These particles have a very specific personality: they hate being too close to each other. If you try to push two together, they push back hard. This is called "mutual repulsion."

However, there is also a "gravity" or a "potential field" (called $Q$) pulling them toward the center. The particles are stuck in a tug-of-war: they want to spread out to avoid each other, but they are pulled in by the external field.

In this paper, the authors study what happens when you have a huge number of these particles (as $n$ goes to infinity). They form a dense, shimmering cloud called the "Droplet."

The Question: How "Jittery" is the Crowd?

The researchers ask a simple question: If I draw a shape (a circle, a square, a weird blob) somewhere on this dance floor, how many particles will be inside it?

Let's call this number $N$.

• If the shape is in the middle of the crowd (the "bulk"), the particles are packed tight.
• If the shape is outside the crowd, it's empty.
• If the shape is right on the edge of the crowd, it's a mix.

The paper isn't just asking "How many?" It's asking: "How much does this number jump around?" (This is called the variance or fluctuation).

If you repeat the experiment 1,000 times, will the number of particles in your shape be exactly the same every time? No. It will wiggle. The paper calculates exactly how big that wiggle is.

The Two Main Discoveries

The paper finds two different rules for how much the crowd wiggles, depending on where your shape is.

1. The "Bulk" Rule (Inside the Crowd)

The Analogy: Imagine you are standing in the middle of a dense, packed concert crowd. You draw a small circle around yourself. Because the crowd is so tight, the only way the number of people inside your circle changes is if someone bumps into the edge of your circle. The people in the middle of the circle are too busy holding their ground to move in or out.

The Finding:
The amount of "jitter" (variance) depends almost entirely on the perimeter (the length of the boundary) of your shape.

• The Formula: The jitter is proportional to the length of the boundary $\times$ the square root of the total number of particles.
• The "Universality": This is the coolest part. It doesn't matter if the crowd is packed in a perfect circle, a square, or a weird blob. It doesn't matter if the "gravity" pulling them is strong in one spot and weak in another. As long as you are inside the crowd, the math is the same. The "jitter" is determined purely by the geometry of the boundary.

Simple Takeaway: Inside the crowd, the chaos is local. Only the fence matters, not the landscape behind it.

2. The "Edge" Rule (The Microscopic Boundary)

The Analogy: Now, imagine you are standing exactly on the edge of the dance floor, looking out into the empty void. You draw a shape that is microscopically larger or smaller than the crowd itself.

• If you make your shape slightly bigger, you might catch a few particles that are just barely hanging on the edge.
• If you make it slightly smaller, you might lose a few.

The Finding:
Here, the math gets more complex because the particles at the edge behave differently than those in the middle. They act like a "fuzzy" boundary.

• The authors found a new formula that involves something called Harmonic Measure.
• The Metaphor: Imagine the edge of the crowd is a coastline. If you drop a drop of ink at infinity (far away) and let it drift toward the crowd, the Harmonic Measure tells you how likely the ink is to hit a specific part of the coastline.
• The "jitter" at the edge depends on this "drift probability" combined with the shape of the edge.

Simple Takeaway: At the edge, the crowd is fuzzy. The amount of jitter depends on how "visible" that specific part of the edge is from the outside world.

Why Does This Matter? (The "So What?")

1. It's Universal: The authors proved that these rules apply to any smooth potential field $Q$. Whether you are modeling electrons in a superconductor, stars in a galaxy, or data points in a machine learning algorithm, if they repel each other like this, the "jitter" follows these exact laws.
2. It Connects to Physics: These particles are mathematically identical to 2D Coulomb gases (charged particles) and non-interacting Fermions (quantum particles like electrons). This helps physicists understand how quantum systems fluctuate.
3. It Solves a Puzzle: Before this, we only knew these rules for very simple, perfectly round shapes. This paper says, "It doesn't matter if your shape is a jagged rock or a perfect circle; the math works the same way."

Summary in One Sentence

The paper proves that for a massive crowd of repelling particles, the amount of random fluctuation in the number of particles inside any shape is determined almost entirely by the length of that shape's boundary, following a universal law that holds true regardless of the specific forces pulling the particles together.

Technical Summary

1. Problem Statement and Context

The paper investigates the statistical fluctuations of the number of eigenvalues in random normal matrix models. These models are defined by a probability measure on the space of $n \times n$ normal matrices $M$:
$$dP_n(M) = \frac{1}{Z_n} e^{-n \text{Tr} Q(M)} dM$$
where $Q: \mathbb{C} \to \mathbb{R}$ is a potential (external field). The eigenvalues $z_1, \dots, z_n$ form a Determinantal Point Process (DPP) characterized by a correlation kernel $K_n(z,w)$.

The primary object of study is the counting statistic $N_A^{(n)}$, which counts the number of eigenvalues falling within a Borel set $A \subset \mathbb{C}$. The authors focus on the asymptotic behavior of the number variance $\text{Var}(N_A^{(n)})$ as $n \to \infty$.

While the limiting density of eigenvalues is well-understood (concentrating on a compact set called the droplet $S_Q$), the fluctuations (variance) depend critically on the geometry of $A$ relative to $S_Q$. Previous results were limited to:

• Radially symmetric potentials ($Q(z) = Q(|z|)$) and concentric disks.
• Specific ensembles like the Ginibre ensemble ($Q(z)=|z|^2$) for general sets.
• Convex sets with $C^2$ boundaries under restrictive analytic conditions.

The paper aims to establish universality: a general limiting formula for the variance that holds for arbitrary smooth potentials $Q$ and arbitrary Borel sets $A$, covering both the bulk (inside the droplet) and the edge (microscopic dilation of the droplet boundary).

2. Methodology

The authors employ a combination of potential theory, asymptotic analysis of orthogonal polynomials, and geometric measure theory.

A. Correlation Kernel Analysis

The variance is expressed via the correlation kernel $K_n$:
$$\text{Var}(N_A^{(n)}) = \int_A \int_{A^c} |K_n(z,w)|^2 dA(z)dA(w)$$
The core of the proof involves analyzing the asymptotic behavior of $K_n(z,w)$ in two distinct regimes:

1. Bulk Regime: $z, w$ strictly inside the droplet $S_Q$. Here, the kernel behaves like a Gaussian.
2. Edge Regime: $z, w$ near the boundary $\partial S_Q$. The authors utilize and strengthen recent results by Hedenmalm and Wennman regarding the asymptotic expansion of orthogonal polynomials near the boundary. They derive precise asymptotic formulas for $K_n$ when $z$ and $w$ are close to the boundary but potentially distinct ($z \neq w$), involving the complementary error function ($\text{erfc}$) and conformal maps.

B. Approximation by Functions of Bounded Variation (BV)

To handle general Borel sets $A$ (whose indicator functions $1_A$ are not smooth), the authors approximate $1_A$ using functions of bounded variation.

• They utilize the De Giorgi structure theorem, which relates the total variation of an indicator function to the perimeter (Hausdorff measure of the measure-theoretic boundary $\partial^* A$).
• They prove a norm representation result (Proposition 5.1) showing that the variance integral converges to a weighted total variation norm involving $\sqrt{\Delta Q}$. This overcomes the difficulty that the kernel is not asymptotically radial for general potentials.

C. Harmonic Measure and Conformal Mapping

For the edge case, the geometry of the droplet is handled via a conformal map $\phi$ from the exterior of the droplet $S_Q^c$ to the exterior of the unit disk $\mathbb{D}^c$. The asymptotic behavior of the kernel near the edge is expressed in terms of $\phi$ and its derivative, linking the variance to the harmonic measure at infinity.

3. Key Contributions and Results

The paper establishes two main theorems providing universal limiting behaviors for the number variance.

Theorem 1.1: Bulk Universality

For a potential $Q$ that is $C^2$, real analytic on the interior of the droplet, and strictly subharmonic ($\Delta Q > 0$), and for any Borel set $A$ strictly inside the droplet ($A \Subset \mathring{S}_Q$):
$$\lim_{n \to \infty} \frac{1}{\sqrt{n}} \text{Var}(N_A^{(n)}) = \frac{1}{2\pi\sqrt{\pi}} \int_{\partial^* A} \sqrt{\Delta Q(z)} \, d\mathcal{H}^1(z)$$

• Significance: This generalizes previous results (e.g., for Ginibre or radial potentials) to arbitrary potentials and arbitrary sets (not just disks or convex sets). The variance scales as $\sqrt{n}$ and depends on the perimeter of $A$ weighted by the local density of states $\sqrt{\Delta Q}$.

Theorem 1.2: Edge Universality

Consider a microscopic dilation of the droplet boundary defined by a parameter $\delta \in \mathbb{R}$. Let $A_n(\delta)$ be the set obtained by expanding ($\delta > 0$) or contracting ($\delta < 0$) the droplet by a microscopic distance of order $1/\sqrt{n}$.
$$\lim_{n \to \infty} \frac{1}{\sqrt{n}} \text{Var}(N_{A_n(\delta)}^{(n)}) = f(\delta) \frac{1}{2\pi\sqrt{\pi}} \int_{\partial S_Q} \sqrt{\Delta Q(z)} \, d\omega_\infty^{S_Q^c}(z)$$
where:

• $f(\delta) = \frac{\sqrt{2\pi}}{4} \int_\delta^\infty \text{erfc}(t)\text{erfc}(-t) dt$.
• $d\omega_\infty^{S_Q^c}(z) = |\phi'(z)| d\mathcal{H}^1(z)$ is the harmonic measure at infinity associated with the exterior of the droplet.
• Significance: This generalizes results by Akemann, Byun, and Ebke (who required radial symmetry) to general non-symmetric potentials. It shows that the edge fluctuations are governed by the harmonic measure of the droplet's exterior, not just its Euclidean perimeter.

Technical Lemmas (Independent Interest)

The authors prove Lemma 1.3 and Lemma 1.4, which provide refined asymptotic estimates for the correlation kernel $K_n(z,w)$ near the boundary for $z \neq w$.

• These lemmas strengthen previous results by Hedenmalm and Wennman by allowing $z$ and $w$ to be distinct and providing uniform error bounds.
• They express the kernel in terms of the Szegő kernel and the complementary error function, bridging the gap between the bulk Gaussian behavior and the edge behavior.

4. Significance and Implications

1. Universality: The results demonstrate that the fluctuation statistics of random normal matrices are universal. The specific form of the potential $Q$ only enters the variance through the local density $\Delta Q$ and the geometry of the droplet (via harmonic measure), rather than through complex model-specific details.
2. Geometric Insight: The paper clarifies the role of the measure-theoretic boundary and harmonic measure in random matrix theory. It shows that for non-radial potentials, the "effective length" of the boundary relevant for fluctuations is weighted by the conformal geometry of the droplet's exterior.
3. Methodological Advancement: The use of weighted bounded variation norms to approximate indicator functions allows the authors to bypass the need for smooth boundaries or convexity assumptions, significantly broadening the class of admissible sets $A$.
4. Comparison with Hermitian Ensembles: The authors contrast their findings with the Gaussian Unitary Ensemble (GUE). In GUE, the variance in the bulk scales as $\log n$, whereas in random normal matrices, it scales as $\sqrt{n}$. Furthermore, the edge behavior in GUE is bounded (saturation), while in normal matrices, the edge variance scales with $\sqrt{n}$, indicating a fundamental difference in the rigidity of eigenvalues between 1D (Hermitian) and 2D (Normal) systems.

In summary, this paper provides a comprehensive and rigorous framework for understanding eigenvalue counting fluctuations in 2D Coulomb gases (random normal matrices) under general conditions, unifying bulk and edge behaviors through advanced asymptotic analysis of correlation kernels and geometric measure theory.