$\mathsf{GL}_N(\mathbb{C})$ Brownian motion and stochastic PDE on entire functions

This paper constructs the full edge scaling limit of singular values for $\mathsf{GL}_N(\mathbb{C})$ Brownian motion, demonstrating that the limiting paths satisfy an infinite system of interacting SDEs and that their rescaled reverse characteristic polynomials evolve according to a specific stochastic partial differential equation, while also establishing connections to universal limits of random matrix products and analogous results for Hua-Pickrell and Bessel models.

The Gist

Imagine you are watching a massive, chaotic dance floor. On this floor, there are thousands of dancers (representing numbers called "singular values") moving around, bumping into each other, and trying to avoid stepping on one another's toes. This dance is happening inside a giant, complex machine called the "General Linear Group" (GLN(C)), which is essentially a mathematical way of describing how matrices (grids of numbers) change over time.

This paper is about what happens when you zoom out so far that the individual dancers become invisible, and you only see the overall pattern of the crowd. The authors, Theodoros Assiotis and Zahra Sadat Mirsajjadi, figured out how to describe this infinite crowd using two different "languages": one that tracks the dancers' positions, and another that tracks the "shape" of the entire crowd.

Here is a breakdown of their discoveries using simple analogies:

1. The Dance of the Singular Values (The SDEs)

Imagine the dancers are trying to stay in a line, ordered from tallest to shortest. As they move, they are pushed by random gusts of wind (Brownian motion). However, they also have a strong social rule: they cannot cross each other. If two dancers get too close, a repulsive force pushes them apart.

• The Discovery: The authors proved that as the number of dancers grows to infinity, their movement settles into a predictable, yet random, pattern. They described this pattern using a massive system of equations (called Stochastic Differential Equations, or SDEs).
• The "Gibbs" Property: Think of this like a game of musical chairs with a twist. If you freeze the dance at any moment and look at a small group of dancers, their positions are determined by the "walls" created by the dancers immediately next to them. If you were to randomly reshuffle just that small group while keeping the neighbors fixed, they would settle into a specific, natural distribution. The authors showed that this "reshuffling" rule holds true even for the infinite crowd.

2. The Shape of the Crowd (The SPDE)

Instead of tracking every single dancer, imagine you are looking at the "shadow" or the "outline" cast by the entire crowd. In mathematics, this outline is called a "characteristic polynomial." It's a single, complex function that contains information about every single dancer.

• The Discovery: The authors found that this "shadow" doesn't just wiggle randomly; it evolves according to a specific, complex rule called a Stochastic Partial Differential Equation (SPDE).
• The Metaphor: Imagine the shadow is a piece of fabric being blown by the wind. The wind is random (noise), but the fabric also has a specific way of stretching and folding (drift). The authors wrote down the exact recipe for how this fabric moves.
• Why it's special: This equation is unique. It involves a "non-linear multiplicative noise," which is a fancy way of saying the randomness depends on the shape of the fabric itself. The paper claims this is the first time such an equation has been explicitly written down for this specific type of mathematical object.

3. The "Universal" Limit

The paper also connects this dance to other famous mathematical models.

• The Connection: If you start the dance with the dancers arranged in a very specific, perfect order (like a grid), the resulting pattern is the same as the pattern you get from multiplying many random matrices together. This suggests that this specific dance is a "universal" behavior that appears in many different random systems, much like how the number $\pi$ appears in circles, probability, and physics.
• The "Zeta" Functions: The authors also looked at two other types of dances (related to "Hua-Pickrell" and "Bessel" models). They showed that these dances eventually settle down into a stable, random shape known as a "stochastic zeta function." They even guessed (conjectured) how the individual dancers in these specific dances move, though they couldn't fully prove the rules for every single case yet.

4. The Secret Weapon: "Intertwiners"

How did they solve this? They used a powerful mathematical tool called "intertwiners."

• The Analogy: Imagine you have a set of Russian nesting dolls. Each doll represents a system with $N$ dancers. The authors found a magical key (the intertwiner) that allows you to translate the behavior of the $N$-dancer system directly into the behavior of the $(N+1)$-dancer system. Because this translation works perfectly for every size, they could mathematically "zoom out" to infinity and see the final, infinite pattern emerge clearly.

Summary

In short, this paper takes a chaotic, high-dimensional dance of numbers and proves that:

1. The dancers follow a specific set of random rules that keep them from colliding.
2. The overall "shape" of the crowd evolves according to a new, complex equation involving random noise.
3. This behavior is a universal pattern that appears in various random matrix systems, and the authors have provided the first clear mathematical description of how these infinite systems evolve over time.

They didn't just watch the dance; they wrote down the choreography for the infinite future.

Technical Summary

Technical Summary: GLN(C) Brownian Motion and SPDE on Entire Functions

Problem Statement
The paper addresses the construction and characterization of the full edge scaling limit of the singular values of multiplicative Brownian motion on the general linear group $GL_N(\mathbb{C})$. While the limiting paths for specific initial conditions (such as the identity matrix) have been studied in the context of products of random matrices and last passage percolation, this work aims to establish the limit starting from general initial conditions. Furthermore, the authors seek to describe the evolution of the associated rescaled reverse characteristic polynomials not merely as a collection of particle dynamics, but as a stochastic partial differential equation (SPDE) acting on the space of entire functions. The paper also extends these results to two other models: the Rider-Valkó model and the dynamical Cauchy model (Hua-Pickrell diffusions), whose stationary measures are given by stochastic zeta functions associated with the Bessel and Hua-Pickrell point processes, respectively.

Methodology
The core methodology relies on the "method of intertwiners" introduced by Borodin and Olshanski, a formalism that constructs an abstract Feller-Markov process on the boundary of a projective system from a tower of consistent finite-dimensional processes.

1. Consistency Relations: The authors first establish that the finite-dimensional dynamics of the singular values (and related models) satisfy specific consistency relations under the "corner" projection (mapping $(N+1) \times (N+1)$ matrices to $N \times N$ matrices). This allows the application of the Borodin-Olshanski framework to construct infinite-dimensional limiting processes on spaces of entire functions (specifically the Laguerre-Pólya class).
2. Gibbs Resampling: A key step involves proving that the limiting paths satisfy a Gibbs resampling property with respect to exponential Brownian bridges. This property, combined with the non-intersection of paths (established via Lyapunov functions), ensures the well-posedness of the infinite-dimensional system.
3. From Particles to SPDEs: The authors derive the SPDEs for the limiting characteristic polynomials by applying Itô's formula to the finite-dimensional reverse characteristic polynomials and taking the $N \to \infty$ limit. They utilize residue calculus and the Stieltjes transform (logarithmic derivative of the polynomial) to bridge the gap between the SPDE for the entire function and the infinite system of interacting SDEs for the zeros (singular values).
4. Convergence to Equilibrium: The long-time behavior is analyzed by studying the drift terms of the limiting processes, showing convergence to the stochastic zeta functions associated with the respective point processes.

Key Contributions and Results

• Construction of the Limiting Process for $GL_N(\mathbb{C})$:
  The paper constructs a unique Feller diffusion process $(X_t^{GL})_{t \ge 0}$ on the space $\Upsilon_+$ (a space of sequences and parameters related to entire functions). It proves that the rescaled singular values of $GL_N(\mathbb{C})$ Brownian motion converge in distribution to this process.

  • Infinite SDE System: The projection of this process onto the singular values satisfies an infinite system of SDEs with singular log-interaction:
    $$dx_i(t) = x_i(t)dw_i(t) + \frac{\theta}{2}x_i(t)dt + \sum_{j \neq i} \frac{x_i(t)x_j(t)}{x_i(t) - x_j(t)}dt$$
  • Gibbs Property: The limiting line ensemble possesses a Gibbs resampling property with respect to exponential Brownian bridges.
  • SPDE for Characteristic Polynomials: The rescaled reverse characteristic polynomial $g_t(z)$ converges to a Markov process solving the SPDE:
    $$dg_t(z) = dM_t(z; g) + \left[ \frac{\theta}{2} z \partial_z g_t(z) - \frac{z^2}{2} \partial^2_z g_t(z) \right] dt$$
    where the martingale term $M_t$ has a specific non-linear covariation structure dependent on $g_t$ and its derivatives.

• Extension to Stochastic Zeta Functions:
  The authors prove analogous results for two other models:

  1. Rider-Valkó Model: The limiting rescaled reverse characteristic polynomial converges to a process solving an SPDE that drives the system toward the Bessel stochastic zeta function $B_\nu$.
  2. Dynamical Cauchy Model (Hua-Pickrell): The limiting process converges to a solution of an SPDE with linear drift and multiplicative noise, driving the system toward the Hua-Pickrell stochastic zeta function $HP_s$. For $s=0$, this provides the first natural dynamical model for the stochastic zeta function of the sine point process.

• Equations for Zeros:
  By analyzing the logarithmic derivative of the limiting entire functions, the authors derive equations for the reciprocals of the zeros. For the Hua-Pickrell model, they derive a system of SDEs for the zeros under a specific assumption regarding their ordering and non-explosion. They propose a conjecture (Conjecture 1.9) that these zeros satisfy a specific infinite system of SDEs involving a principal value sum, which would characterize the dynamics of the Hua-Pickrell stochastic zeta function.

Significance and Claims
The paper claims to provide the first explicit examples in the literature of stochastic evolutions associated with the Laguerre-Pólya class of entire functions that arise as scaling limits of random matrix models.

• Novelty of SPDEs: The derived SPDEs feature a non-linear multiplicative noise term and a linear drift, a structure not previously seen in the stochastic PDE literature for such objects. The covariation structure of the noise is noted to be reminiscent of correlation kernels from random matrix theory.
• Universality: The work connects the edge scaling limit of $GL_N(\mathbb{C})$ Brownian motion (from general initial conditions) to the universal critical stochastic process arising in products of random matrices.
• Integrability: The results highlight a "hidden integrability" in these models, manifested through the consistency relations that allow the application of the Borodin-Olshanski formalism.
• Open Problems: The authors modestly note that while they establish the existence of the limiting processes and their SPDE descriptions, the uniqueness of the solutions to the infinite SDE systems (specifically regarding the "rigid-path" property) remains an open problem. They also note that while the convergence to equilibrium is proven, the description of the limiting entire function for complex parameters in the Hua-Pickrell model is less explicit than for real parameters.

In summary, the paper establishes a rigorous link between finite-dimensional random matrix dynamics and infinite-dimensional stochastic evolutions on spaces of entire functions, providing new SPDE descriptions for these limits and connecting them to stochastic zeta functions.