Noncolliding processes, matrix-valued processes and determinantal processes

A noncolliding diffusion process is a conditional process of $N$ independent one-dimensional diffusion processes such that the particles never collide with each other. This process realizes an interacting particle system with long-ranged strong repulsive forces acting between any pair of particles. When the individual diffusion process is a one-dimensional Brownian motion, the noncolliding process is equivalent in distribution with the eigenvalue process of an $N \times N$ Hermitian-matrix-valued process, which we call Dyson’s model. For any deterministic initial configuration of $N$ particles, distribution of particle positions of the noncolliding Brownian motion on the real line at any fixed time $t >0$ is a determinantal point process. We can prove that the process is determinantal in the sense that the multi-time correlation function for any chosen series of times, which determines joint distributions at these times, is also represented by a determinant. We study the asymptotic behavior of the system, when the number of Brownian motions $N$ in the system tends to infinity. This problem is concerned with the random matrix theory on the asymptotics of eigenvalue distributions, when the matrix size becomes infinity. In the present paper, we introduce a variety of noncolliding diffusion processes by generalizing the noncolliding Brownian motion, some of which are temporally inhomogeneous. We report the results of our research project to construct and study finite and infinite particle systems with long-ranged strong interactions realized by noncolliding processes.

💡 Research Summary

The paper investigates a class of interacting particle systems called non‑colliding diffusion processes, which are obtained by conditioning N independent one‑dimensional diffusion particles never to intersect. When the underlying diffusion is a standard Brownian motion, the resulting non‑colliding process coincides in law with the eigenvalue dynamics of an N × N Hermitian matrix‑valued process—Dyson’s model with β = 2. The authors first review elementary diffusion processes (Brownian motion, Brownian bridges, absorbing Brownian motions, Bessel processes, and generalized meanders) and describe their transition densities and Doob h‑transforms, establishing the necessary building blocks for later constructions.

A central tool is the Karlin‑McGregor formula, which expresses the transition density of N non‑colliding particles as a determinant of the single‑particle transition kernel. This determinant representation is the probabilistic analogue of the Slater determinant for free fermions and underlies the determinantal (Fermion) point‑process structure of the system. By invoking Bru’s theorem and its extensions, the authors relate the eigenvalue process of a Hermitian matrix‑valued diffusion to a system of stochastic differential equations, thereby recovering Dyson’s model from the matrix viewpoint.

The paper then proves that not only the spatial distribution at a fixed time but also the full multi‑time correlation functions are determinantal. Using the Eynard‑Mehta method, the authors construct a time‑dependent kernel K(t_i,x_i; t_j,x_j) such that any n‑point correlation function across arbitrary time slices is given by the determinant of the corresponding kernel matrix. This kernel involves the single‑particle transition density together with orthogonal polynomials (or Bessel functions) that evolve with time.

As N tends to infinity, the authors discuss the asymptotic behavior of the correlation kernel. In the bulk scaling limit the kernel converges to the sine kernel, while at the soft edge it converges to the Airy kernel, leading to the Tracy‑Widom distribution and Painlevé‑II equations for the largest eigenvalue fluctuations. These results connect the non‑colliding diffusion picture directly to classical results in random matrix theory.

A novel contribution of the work is the treatment of temporally inhomogeneous non‑colliding processes obtained by imposing the non‑collision condition only on a finite interval (0,T). In this setting the process loses its determinantal structure and instead becomes a Pfaffian point process, reminiscent of β = 1 and β = 4 ensembles. The authors relate these Pfaffian processes to Harish‑Chandra/Itzykson‑Zuber integrals, thereby extending the matrix‑model correspondence beyond the homogeneous case.

The final section surveys related topics not covered in detail, such as infinite‑particle limits, connections to multi‑matrix models, quantum transport, and extensions to general β‑ensembles. Overall, the paper provides a comprehensive framework that unifies non‑colliding diffusion processes, matrix‑valued stochastic dynamics, and determinantal/Pfaffian point‑process theory, offering deep insights into the interplay between probability, statistical physics, and random matrix theory.