The extreme statistics of some noncolliding Brownian processes

Abstract: We consider certain noncolliding interacting particle systems driven by Brownian noise. A key example is drifted Brownian motions conditioned not to intersect and related models of eigenvalues of Hermitian random matrices. We establish limit theorems for the extremal particle. We find: (i) the scaling limit of the largest eigenvalue of Brownian motion over Hermitian, positive-definite matrices, (ii) Airy process limit for the largest eigenvalue of Dyson's Brownian motion for GUE started from generic initial conditions, and (iii) a Fredholm determinant formula for the running maximum of the top path among noncolliding Brownian bridges and, as a byproduct, a new formula for the law of largest eigenvalue in the Laguerre Orthogonal Ensemble as well as for a related point-to-line last passage percolation model.

• The paper presents a novel limiting distribution for the largest eigenvalue in structured+GUE perturbations using explicit Fredholm determinant formulas.
• It demonstrates Airy universality for Dyson's Brownian motion with diverse initial spectra through rigorous steepest descent analysis.
• It derives explicit determinant formulas for last passage percolation models and running maxima, unifying combinatorial and random matrix approaches.

Extreme Statistics for Noncolliding Brownian Processes

Introduction and Context

This work rigorously analyzes the extremal (largest) particle trajectories in a suite of noncolliding Brownian particle systems. These mathematical objects arise naturally in random matrix theory, stochastic growth models, and last passage percolation (LPP). The paper establishes several limit theorems and provides new Fredholm determinant formulas governing the law of the maximum in both finite and scaling limits, advancing the understanding of fluctuations at the spectral edge in several canonical models. The major results include (i) the identification of a novel limiting distribution for the largest eigenvalue in structured + GUE (Gaussian Unitary Ensemble) perturbation models, (ii) a demonstration of Airy process universality for extremal eigenvalues in Dyson's Brownian motion under broad initial data, and (iii) determinant expressions for full-space and point-to-line LPP problems, with further connections to the Laguerre Orthogonal Ensemble (LOE).

Main Results

1. Scaling Limit for the Largest Eigenvalue in Structured + GUE Perturbations

The first main theorem establishes the asymptotic law for the largest eigenvalue of a Hermitian random matrix of the form $H = H^{\mathrm{reg}} + H^{\mathrm{GUE}}$, where $H^{\mathrm{reg}}$ has a deterministic, equally spaced spectrum (an arithmetic progression). Upon suitable rescaling, the law of the largest eigenvalue converges, as the dimension $n\to\infty$, to a probability distribution characterized via a Fredholm determinant involving a novel double contour kernel $K_\Delta$, which depends on the step $\Delta$ of the arithmetic spectrum.

This limiting distribution does not coincide with previously known universality classes such as GUE Tracy-Widom or Gaussian, and the kernel is built explicitly out of Gamma functions and double integrals over carefully chosen contours. The rigorous derivation employs connections to determinantal point processes and the explicit analysis of the correlation kernel in the model.

A consequential corollary resolves the distribution of the largest logarithmically-transformed eigenvalue of Brownian motion on the symmetric space $GL(n,\mathbb{C})/U(n)$ (i.e., Hermitian positive-definite matrices), leveraging explicit SDEs for log-eigenvalues and previous characterization results.

2. Airy Universality for Dyson's Brownian Motion with Generic Initial Data

The analysis addresses Dyson's Brownian motion for GUE—$n$ Hermitian matrix-valued Brownian motion—started from an arbitrary Hermitian matrix $H_0$ with eigenvalues $\{\nu_j^n\}$. A significant theorem demonstrates that, provided the empirical spectral distributions are contained within a class $F(\alpha,\beta)$ meeting mild regularity and nondegeneracy requirements on eigenvalue spacings, the largest eigenvalue process, properly centered and scaled, converges in finite-dimensional distributions (f.d.d.) to the Airy process $H^{\mathrm{reg}}$0.

The Airy process, which arises as the top line of the Airy line ensemble with determinantal correlation kernel (extended Airy kernel), governs extremal fluctuations in numerous contexts, e.g., KPZ universality, random tilings, and stochastic growth models. The paper's proof leverages integrable probability techniques: it constructs the relevant determinantal kernels, performs rigorous steepest descent analysis, and exploits uniform contour decay to establish convergence of cumulants and Fredholm determinants. The result confirms the broad (edge) universality of the Airy process beyond the classical setting of i.i.d. or Wigner initial data, encompassing deterministic or highly inhomogeneous spectra with controlled crowding.

3. Fredholm Determinant Formulas for LPP Models and Running Maxima

A central technical contribution is the derivation of a Fredholm determinant formula for the point-to-line BLPP (Brownian Last Passage Percolation) model with drift, encompassing various boundary conditions, in terms of explicit, tractable contour integrals. The analysis extends to the problem of the running maximum of the largest eigenvalue in Hermitian Brownian motion with diagonal drift, unifying previous LPP representations and illuminating the Burke property in these systems.

A further corollary demonstrates that the distribution function of the largest eigenvalue of the LOE (Laguerre Orthogonal Ensemble) can be expressed as a Fredholm determinant with a particularly simple kernel, utilizing combinatorics of up/right paths in associated exponential LPP models. This yields new access to edge asymptotics in random matrix ensembles beyond GUE, and establishes a pathway for further generalization.

The methods developed rigorously connect finite-dimensional processes (e.g., Brownian bridges conditioned to avoid collisions) and their maximal statistics to integrable kernels and LPP interpretations, facilitating explicit calculation and asymptotic description.

Numerical and Structural Implications

The theorems yield explicit, computable distribution functions for the extremal statistics (in the $H^{\mathrm{reg}}$1 limit) of noncolliding Brownian systems under a broad array of initial and drift structures. In the case of the structured+GUE model, the emergence of a new distribution class for the spectral maximum is a strong, notable finding. Airy process universality under generic initial spectra reinforces and generalizes the familiar universality paradigm at the edge of large random matrices.

Moreover, the determinant formulas provide fully explicit representations, making precise numerical or theoretical investigation of extremal statistics possible, including studies of tail behavior, moderate deviations, and scaling transitions between classical ensembles and novel limiting distributions.

Theoretical Consequences and Future Directions

The results deepen the understanding of the interplay between noncolliding constraints, drift/boundary conditions, and extremal statistics in determinantal and integrable systems. The systematic use of Fredholm determinants in expressing the laws of maxima connects diverse areas: random matrix theory, SDEs on symmetric spaces, and LPP models.

Potential future developments include:

• Extending Airy process universality to more general non-Hermitian or non-self-dual dynamics, and quantifying the effect of heavy-tailed or clustered initial spectra.
• Investigating multi-point extrema and their joint laws, as well as the impact of different initial and boundary conditions.
• Leveraging the novel kernel for structured+GUE perturbations to explore crossover and critical regime behavior, and possible connections to physics, e.g., quantum chaos, harmonic oscillators with random perturbations.
• Applying the determinant expressions to study finer properties (e.g., large deviations, rare-event asymptotics) in LPP and stochastic growth.

Conclusion

This work presents a comprehensive and technically detailed study of the extremal statistics in noncolliding Brownian systems and related random matrix models. It establishes new limit theorems, identifies universality classes, and delivers explicit, analytically tractable formulas for the distribution of spectral maxima. The results reinforce the centrality of Airy process universality, clarify the impact of structured perturbations, and strengthen the connection between combinatorial, probabilistic, and random matrix-theoretic approaches to stochastic interacting particle systems.