Characteristic polynomials of non-Hermitian random band matrices near the threshold

This paper extends previous techniques to analyze the critical regime of non-Hermitian random band matrices where the bandwidth $W$ is proportional to the threshold $\sqrt{N}$, thereby characterizing the transition in the asymptotic behavior of the second correlation function of their characteristic polynomials.

The Gist

The Big Picture: The "Goldilocks" Zone of Chaos

Imagine you have a massive, chaotic crowd of people (the matrix) standing in a square grid. Each person is whispering a number to their neighbors. In mathematics, we call this a Random Band Matrix.

The "bandwidth" ($W$) is simply the rule of how far a person can whisper.

• Narrow Band ($W$ is small): You can only whisper to the person standing right next to you. The crowd is very local. Information gets stuck in small pockets.
• Wide Band ($W$ is huge): Everyone can shout to everyone else. The crowd is fully connected, and information spreads instantly everywhere.

For a long time, mathematicians knew what happened at the extremes:

1. Too Narrow: The system behaves like a collection of isolated islands (Poisson statistics).
2. Too Wide: The system behaves like a perfectly mixed soup (Ginibre ensemble), where everything is universal and predictable.

The Mystery: What happens right in the middle? The "Goldilocks" zone where the bandwidth is exactly proportional to the square root of the total number of people ($W \sim \sqrt{N}$). This is the critical threshold.

This paper, by Mariya and Tatyana Shcherbina, solves the mystery of what happens exactly at this tipping point.

---

The Analogy: The "Whispering Game"

Let's break down the technical concepts using a Whispering Game analogy.

1. The Players (The Matrix)

Imagine $N$ people in a line. They are playing a game where they pass a secret message (a complex number) to their neighbors.

• Non-Hermitian: This means the rules are slightly unfair. If Alice whispers to Bob, Bob doesn't necessarily whisper back with the same volume. It's a one-way street of influence, which makes the math much harder (like trying to predict the path of a leaf in a windy, chaotic river).
• Characteristic Polynomials: Think of this as the "fingerprint" of the whole crowd. It tells us about the collective behavior of all the whispers combined.

2. The Two Extremes (What we already knew)

• The "Island" Phase ($W \ll \sqrt{N}$): If the bandwidth is tiny, Alice only talks to Bob, Bob to Charlie, and so on. The message never travels far. The crowd acts like a bunch of disconnected groups. The math shows the behavior is "factorized" (simple and independent).
• The "Soup" Phase ($W \gg \sqrt{N}$): If the bandwidth is huge, Alice talks to everyone. The crowd becomes a single, giant, well-mixed entity. The math shows the behavior matches the famous Ginibre Ensemble (a standard model for random chaos).

3. The Critical Threshold ($W \sim \sqrt{N}$)

This is the edge of the cliff. The bandwidth is just right to let the message travel across the whole crowd, but not so fast that it becomes a perfect soup.

• The Problem: In this zone, the system is neither fully local nor fully global. It's a delicate dance. Previous studies (like their paper [21]) showed that a transition happens here, but they didn't know exactly what the dance looked like.
• The Goal: This paper asks: "If we zoom in on this exact tipping point, what is the precise mathematical shape of the crowd's behavior?"

---

The Method: The "Super-Transfer" Machine

To solve this, the authors use a technique called the SUSY Transfer Matrix method. Let's translate that into a story.

Imagine you are trying to predict the final state of a long line of dominoes falling.

• The Old Way: You try to calculate the fall of every single domino at once. Impossible for a million dominoes.
• The Transfer Matrix Way: You look at how Domino A affects Domino B, then B affects C. You build a "machine" (an operator) that takes the state of one domino and predicts the next. You chain these machines together.

The Authors' Innovation:
In this paper, they realized that at the critical threshold, this "machine" behaves like a specific type of diffusion process.

• Instead of the machine just jumping from one state to another, it starts to look like a smooth wave spreading out.
• They found that the behavior of this chaotic crowd is governed by a specific Differential Operator (a mathematical machine that describes how things change over space).

The Result (The "Aha!" Moment):
They discovered that at the critical threshold, the behavior of the crowd is described by a specific equation involving Legendre Polynomials (a type of curve used in physics to describe spherical shapes).

Think of it like this:

• In the "Island" phase, the crowd is a jagged, broken line.
• In the "Soup" phase, the crowd is a perfectly flat table.
• At the Threshold: The crowd forms a smooth, rolling hill. The shape of this hill is exactly what the authors calculated.

---

Why Does This Matter?

You might ask, "Who cares about a math problem about whispering crowds?"

1. Physics of Materials: This math helps us understand how electricity moves through "thick wires" or disordered materials. The transition from "Island" to "Soup" is similar to the Anderson Transition, where a material suddenly changes from being an insulator (electricity stops) to a conductor (electricity flows).
2. Universality: The authors proved that this "rolling hill" shape is universal. It doesn't matter if the whispers are Gaussian, or if the crowd is sparse. As long as you are at the critical threshold, the math is the same. This is a huge deal in physics because it means we can use this simple model to predict complex real-world behaviors.
3. The "Edge" of Chaos: Understanding the exact moment a system tips from order to chaos is one of the holy grails of science. This paper gives us the precise blueprint for that tipping point in non-Hermitian systems (systems that don't follow standard energy conservation rules, like lasers or open quantum systems).

Summary in One Sentence

This paper uses advanced mathematical "machines" to prove that when a chaotic system is perfectly balanced between being isolated and being fully connected, its behavior transforms into a smooth, predictable wave pattern described by a specific set of curves, revealing a hidden order in the chaos.

Technical Summary

1. Problem Statement and Context

The paper investigates the local eigenvalue statistics of $N \times N$ non-Hermitian random band matrices (RBM) in the critical regime where the bandwidth $W$ is proportional to the threshold $\sqrt{N}$.

• Model: The matrices $H_N$ have independent complex Gaussian entries with zero mean and covariance $E\{H_{jk}\bar{H}_{jk}\} = J_{jk}$, where $J = (-W^2 \Delta + 1)^{-1}$ and $\Delta$ is the discrete Laplacian. The parameter $W$ represents the bandwidth.
• Background:
  • For Hermitian RBMs, it is well-established that a phase transition occurs at $W \sim \sqrt{N}$:
    • $W \gg \sqrt{N}$: Delocalized eigenvectors, universal Gaussian Unitary Ensemble (GUE) statistics.
    • $W \ll \sqrt{N}$: Localized eigenvectors, Poisson statistics.
  • For Non-Hermitian RBMs, the empirical spectral distribution converges to the Circular Law. However, local statistics were less understood.
• Previous Work ([21]): The authors previously studied the second correlation function of characteristic polynomials, $\Theta(z_1, z_2)$. They showed that for $W \gg \sqrt{N}$, the statistics match the Ginibre ensemble (universal), while for $1 \ll W \ll \sqrt{N}$, the statistics factorize (localized regime).
• The Gap: The behavior exactly at the critical threshold $W \sim \sqrt{N}$ (specifically $W = \kappa_* N^{1/2}$) remained an open problem. This paper aims to characterize the asymptotic local behavior of $\Theta(z_1, z_2)$ in this critical regime.

2. Methodology

The authors employ a rigorous Supersymmetry (SUSY) transfer matrix approach, extending techniques developed in their previous work [21]. The core of the methodology involves transforming the correlation function into an integral representation and analyzing the spectral properties of the associated transfer operator.

Key Technical Steps:

1. Integral Representation:
   The normalized correlation function $\tilde{\Theta}(z_1, z_2)$ is represented as a scalar product involving the $(N-1)$-th power of an integral operator $K_\zeta$:
   $$\tilde{\Theta}(z_1, z_2) = (K_\zeta^{N-1} g, g)$$
   Here, $K_\zeta$ acts on a space of functions defined over $2 \times 2$ complex matrices and the unitary group $U(2)$.

2. Concentration of Measure:
   The authors prove that the main contribution to the integral comes from a specific domain $\Omega_W$ where the matrix variables $Q$ are close to $u_* U$ (with $u_* = \sqrt{1-|z|^2}$). This allows them to restrict the integration to a small neighborhood of the saddle point.

3. Operator Decomposition and Expansion:

   • The operator $K_\zeta$ is decomposed into a "radial" part (acting on positive matrices) and a "unitary" part (acting on $U(2)$).
   • The radial part is analyzed using a change of variables $R_i = u_*(I + W^{-1/2} \tilde{R}_i)$, leading to a quadratic approximation involving Hermite polynomials.
   • The unitary part is analyzed using the irreducible representations of $SU(2)$ (Wigner $D$-matrices or coefficients $t^{(\ell)}_{pk}$).

4. Spectral Analysis of the Transfer Operator:

   • Lemma 2.1 - 2.6: The authors establish that the operator $K_\zeta$ can be approximated by a block-diagonal operator in the basis of Hermite polynomials (radial) and spherical harmonics (unitary).
   • They prove that the eigenvalues of $K_\zeta$ cluster near the maximum eigenvalue $\lambda_{\max}$.
   • Crucially, they show that for the critical scaling $W \sim \sqrt{N}$ (where $\epsilon = \sqrt{W/N} \sim W^{-1/2}$), the operator $K_\zeta$ can be approximated by an operator of the form $I + N^{-1} A$, where $A$ is a differential operator.

5. Reduction to a Differential Operator:
   By analyzing the action of $K_\zeta$ on the subspace spanned by the leading eigenvectors, the discrete transfer matrix problem is reduced to the study of a continuous differential operator $A_0$ acting on functions of $z \in [-1, 1]$.

3. Key Contributions and Results

Main Theorem (Theorem 1.2)

The paper establishes the limiting behavior of the normalized second correlation function of characteristic polynomials in the critical regime $W = \kappa_* N^{1/2} (1+o(1))$:

$$\lim_{N \to \infty} \frac{\Theta(z_1, z_2)}{\Theta^{1/2}(z_1, z_1)\Theta^{1/2}(z_2, z_2)} = (e^{A_0} \mathbf{1}, \mathbf{1})$$

Where:

• $\mathbf{1}(z) \equiv 1$.
• $A_0$ is a second-order differential operator defined on $z \in [-1, 1]$:
  $$A_0 \phi(z) = \frac{1}{8(\kappa_* u_*)^2} \left( (1-z^2)\frac{d^2}{dz^2} - 2z\frac{d}{dz} \right) \phi(z) + 2z\phi(z)$$
• The boundary conditions are $|\phi(1)| < \infty$ and $|\phi(-1)| < \infty$.

Technical Lemmas

• Lemma 3.1 & 3.2: These lemmas rigorously justify the truncation of the infinite-dimensional operator $K_\zeta$ to a finite-dimensional subspace spanned by low-order Hermite and spherical harmonic modes. They prove that the error introduced by this truncation vanishes as $N \to \infty$.
• Lemma 2.7: Establishes the specific action of the operator on shifted eigenfunctions, linking the discrete matrix structure to the continuous differential operator $A_0$.

4. Significance and Implications

1. Completion of the Phase Transition Picture:
   This work fills the critical gap in the understanding of non-Hermitian RBMs. It confirms that the transition from Poisson-like (localized) to Ginibre-like (delocalized) statistics is continuous and governed by a specific differential operator at the critical point $W \sim \sqrt{N}$.

2. Universality Class:
   The result demonstrates that the critical statistics for non-Hermitian RBMs are distinct from both the fully delocalized (Ginibre) and fully localized (Poisson) regimes. The limiting object is the semigroup generated by the operator $A_0$, which resembles a diffusion process on the interval $[-1, 1]$ with a drift term.

3. Methodological Advancement:
   The paper successfully extends the SUSY transfer matrix technique to the critical scaling regime. The rigorous control of the remainder terms in the asymptotic expansion of the transfer operator (specifically handling the interplay between the bandwidth $W$ and the matrix size $N$) provides a robust framework for analyzing similar critical phenomena in random matrix theory.

4. Connection to Hermitian Case:
   The authors note that a similar result for Hermitian RBMs was obtained in [26, 27]. This work confirms that the non-Hermitian case exhibits a parallel critical behavior, reinforcing the universality of the $\sqrt{N}$ threshold across different symmetry classes, despite the different global spectral distributions (Circular Law vs. Wigner Semicircle).

In summary, the paper provides a rigorous mathematical derivation of the critical scaling limit for non-Hermitian random band matrices, identifying the specific differential operator that governs the correlation functions at the phase transition threshold.