🔢 mathematics

Edgeworth expansions of extreme eigenvalue distributions for random unitary ensembles

This paper establishes full Edgeworth expansions for the extreme eigenvalue distributions of Gaussian-type and Laguerre-type random unitary ensembles at their spectral edges, revealing universal correction structures expressed through finite sums of Airy and Bessel functions and their derivatives.

Original authors: Junwen Liu, Qianlu Mao, Lun Zhang

Published 2026-10-01

📖 4 min read🧠 Deep dive

CC0 1.0

Original authors: Junwen Liu, Qianlu Mao, Lun Zhang

Original paper dedicated to the public domain under CC0 1.0 (http://creativecommons.org/publicdomain/zero/1.0/). ✨ 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

In the vast landscape of mathematics, there is a branch dedicated to understanding the behavior of systems with countless moving parts, where individual actions are too chaotic to predict but the collective outcome follows strict laws. This is the realm of random matrix theory. Imagine a giant grid of numbers, where each entry is chosen at random according to a specific set of rules. When you look at the special values, called eigenvalues, that emerge from this grid, they do not scatter randomly. Instead, they arrange themselves with a surprising order, clustering in specific patterns that appear in everything from the energy levels of heavy atoms to the fluctuations in financial markets. For decades, mathematicians have known the broad strokes of these patterns: as the grid grows infinitely large, the distribution of these values settles into a few universal shapes.

However, knowing the final shape is only half the story. In the real world, systems are never infinite; they are finite, and the size of the system matters. Just as a small sample of a crowd might not perfectly reflect the demographics of a whole city, a finite matrix does not perfectly match the infinite ideal. The difference between the finite reality and the infinite ideal is where the most interesting details hide. For a long time, scientists could only describe the first, most obvious difference between the two. They knew how the pattern shifted slightly, but they lacked a complete map of the higher-order corrections—the subtle, layered adjustments that reveal the true structure of the system as it grows. Without this full picture, predictions for large but finite systems remain slightly blurry.

A team of researchers has now cleared away that blur. They have constructed a complete, high-resolution map of how these random systems behave at their very edges, where the eigenvalues are most sparse and the patterns are most delicate. The team focused on two major families of these random grids: one that models systems with a natural limit on their energy, and another that models systems where values cannot go below zero. By using a sophisticated method that treats the problem as a journey through a complex geometric space, they were able to peel back the layers of the infinite ideal to reveal the exact corrections needed for any finite size.

The researchers discovered that these corrections are not random noise. Instead, they follow a precise, universal structure. At the edge where the values fade away smoothly, the corrections are built from a specific type of wave function known as the Airy function, mixed with simple polynomial shapes. At the other edge, where the values hit a hard wall and cannot pass, the corrections are built from Bessel functions, another fundamental wave pattern found in nature. The team proved that these corrections can be written down as a full series, meaning one can calculate the behavior of the system to any desired level of precision, not just the first approximation.

What makes this discovery particularly powerful is that the researchers did not just find the first correction; they found the entire infinite sequence of them. They showed that every single term in this sequence is made of the same basic building blocks: combinations of these special wave functions and simple polynomials. This means that the "fuzziness" of a finite system is not chaotic; it is highly organized. The team also demonstrated how to translate these findings about the arrangement of numbers into predictions about the extreme values—the largest or smallest numbers in the grid. They found that the probability of finding these extreme values follows a predictable pattern, where the first correction term is directly linked to the slope of the main pattern.

This work resolves a long-standing question about whether these corrections could be fully described. Previous studies had managed to find the first step in the correction, but the path forward seemed blocked. The researchers showed that the path is open and that the structure of the corrections is universal, meaning it applies to a wide variety of systems as long as they share the same basic rules. Their findings confirm that the complex behavior of these random systems at their boundaries is governed by a deep, underlying simplicity. By providing the full expansion, they have given scientists a precise tool to calculate the behavior of large, finite systems with unprecedented accuracy, turning a vague approximation into a sharp, reliable prediction.

Technical Summary: Edgeworth Expansions of Extreme Eigenvalue Distributions for Random Unitary Ensembles

Problem Statement
In random matrix theory, the limiting distributions of extreme eigenvalues (the largest for Gaussian-type ensembles and the smallest for Laguerre-type ensembles) are well-established as universal laws: the Tracy-Widom distribution at the soft edge and a distribution related to the Bessel kernel at the hard edge. However, for finite sample sizes $n$ , understanding the rate of convergence and the structure of finite- $n$ corrections is essential for applied purposes. While isolated first-order correction terms have been studied for specific cases (e.g., the Gaussian Unitary Ensemble (GUE) and Laguerre Unitary Ensemble (LUE)), a general framework for full higher-order Edgeworth expansions for general Gaussian-type and Laguerre-type unitary ensembles has been lacking. Specifically, the existence of full asymptotic expansions and the universal algebraic structure of the correction terms for general polynomial potentials remained open questions.

Methodology
The authors employ the Riemann-Hilbert (RH) approach to orthogonal polynomials, utilizing the Deift-Zhou nonlinear steepest descent method to perform asymptotic analysis. The core of the methodology involves:

RH Characterization: The correlation kernels $K_n(x,y)$ for the ensembles are expressed via the solution $Y(z)$ of a specific RH problem associated with orthogonal polynomials.
Steepest Descent Analysis: The RH problem is transformed through a series of steps (opening lenses, constructing global and local parametrices) to isolate the asymptotic behavior as $n \to \infty$ $n \to \infty$ .
- For Gaussian-type ensembles (soft edge), the local parametrix is constructed using the Airy function.
- For Laguerre-type ensembles (hard edge), the local parametrix near the origin utilizes the Bessel function, while the soft edge near the right endpoint uses the Airy function.
Refinement for Full Expansions: Unlike standard steepest descent which often yields only the leading order, the authors refine the analysis to derive full asymptotic expansions in powers of $n^{-2/3}$ (soft edge) and $n^{-1}$ (hard edge).
Wronskian Identity and Divisibility: A critical technical step involves proving a Wronskian identity for solutions of second-order ODEs (Proposition 3.2). This identity ensures that the singularities in the Christoffel-Darboux quotient cancel out, allowing the correction kernels to be expressed as finite sums of products of Airy (or Bessel) functions and their derivatives with polynomial coefficients.
Fredholm Determinant Lifting: The expansion of the correlation kernel is lifted to the distribution function of the extreme eigenvalue using Fredholm determinant theory. The authors utilize a finite-rank decomposition of the correction operators (Proposition 3.1), reducing the infinite-dimensional determinant expansion to finite-dimensional determinant calculations involving the resolvent of the limiting kernel.

Key Contributions and Results

Full Edgeworth Expansions: The paper establishes the existence of full asymptotic expansions for the correlation kernels and the extreme eigenvalue distribution functions for both Gaussian-type and Laguerre-type unitary ensembles under general one-cut regularity assumptions on the potential.
- Soft Edge (Gaussian-type): Theorem 2.2 provides an expansion for the rescaled kernel $\tilde{K}_n$ in powers of $n^{-2/3}$ . The coefficients are finite sums of products of Airy functions ( $Ai$ ) and their derivatives, with polynomial coefficients.
- Hard Edge (Laguerre-type): Theorem 2.5 provides an expansion for the rescaled kernel $\tilde{K}_n$ in powers of $n^{-1}$ . The coefficients involve Bessel functions ( $J_\alpha$ ) and their derivatives, again with polynomial coefficients.
Universal Structure of Correction Terms: A significant finding is the universal structure of the correction terms. The coefficients of the expansion are not arbitrary but are explicitly given by finite sums of special functions (Airy or Bessel) and their derivatives multiplied by polynomials.
Explicit First-Order Corrections: The paper derives explicit formulas for the first correction term ( $j=1$ $j = 1$ ) in the distribution functions:
- For the largest eigenvalue of Gaussian-type ensembles (Corollary 2.3), the first correction involves derivatives of the Tracy-Widom distribution function $F_{Ai}$ , specifically $F'_{Ai}$ and $F''_{Ai}$ .
- For the smallest eigenvalue of Laguerre-type ensembles (Corollary 2.6), the first correction involves the derivative of the Bessel distribution function $F_{Bes}$ .
Optimally Modified Scaling: The results imply an optimally modified hard-edge scaling for Laguerre-type ensembles (Equation 2.20), which eliminates the $O(n^{-1})$ correction term, improving the convergence rate to $O(n^{-2})$ .

Significance and Claims
The authors claim that their results resolve the conjecture (previously noted for GUE) that full asymptotic expansions exist at the soft edge, extending this to a general class of Gaussian-type potentials. Furthermore, they provide the first general formulation of full hard-edge expansions for Laguerre-type ensembles, going beyond the isolated first correction terms previously known.

The methodology is presented as a robust framework that reveals "universal structures" for correction terms. By demonstrating that the correction operators have finite rank, the paper shows that the complex problem of expanding Fredholm determinants reduces to finite-dimensional linear algebra, yielding explicit formulas. The authors suggest that this approach can be adapted to explore similar problems in random matrix theory and beyond, though they do not explicitly list specific future applications beyond the scope of the current ensembles.

The work unifies the treatment of soft and hard edges, providing a comprehensive theoretical foundation for finite-size corrections in random unitary ensembles, bridging the gap between leading-order universal limits and finite-sample behavior.

More like this