Interpolating non-Hermitian universality classes A and AI$^\dagger$: eigenvalue density and transition regime

This paper utilizes the Kac-Rice formalism to derive the finite-size eigenvalue and eigenvector distributions for a Gaussian ensemble interpolating between non-Hermitian classes A and AI$^\dagger$, revealing that while bulk and fixed-parameter edge behaviors follow standard laws, a specific scaling of the interpolation parameter uncovers a new universal transitional regime in the edge eigenvalue density.

The Gist

Imagine you are running a massive experiment with thousands of spinning tops. In the world of mathematics and physics, these tops are represented by matrices (grids of numbers). Usually, scientists study two very different types of these tops:

1. The Chaotic Tops (Class A): These spin wildly with no rules. They represent systems where "time reversal symmetry" is broken (if you played a movie of them backward, they would look completely different).
2. The Symmetric Tops (Class AI†): These spin with a strict mirror rule. If you played the movie backward, they would look exactly the same.

For a long time, scientists knew how these two types behaved individually, but they didn't know what happened if you slowly turned a dial to change a Chaotic top into a Symmetric one. This paper builds that dial and describes exactly what happens as you turn it.

Here is a breakdown of their findings using simple analogies:

1. The "Dial" (The Interpolation)

The authors created a new mathematical model that acts like a dimmer switch.

• Setting 0: You get the Chaotic tops (Complex Ginibre matrices).
• Setting 1: You get the Symmetric tops (Complex Symmetric matrices).
• Settings in between: You get a mix of both.

They wanted to see how the "crowd" of numbers (eigenvalues) inside these matrices behaves as you slowly turn the dial from 0 to 1.

2. The "Party" in the Middle (The Bulk)

Imagine the numbers in the matrix are guests at a party.

• The Finding: No matter where you set the dial (whether the tops are mostly chaotic, mostly symmetric, or a perfect mix), the guests in the middle of the room always arrange themselves in a perfect circle.
• The Metaphor: It's like a dance floor where, regardless of the music genre, everyone in the center forms a perfect ring. The authors call this the "Circular Law." Their math proves that this ring shape is unshakeable, even as you change the rules of the game.

3. The "Edge" of the Room (The Transition)

The real magic happens at the edge of the party (the outer rim of the circle).

• The "Strong" Regime: If you keep the dial fixed at any number except the very end (1), the edge of the party looks exactly like the Chaotic tops. The symmetry doesn't change the edge behavior yet.
• The "Weak" Regime (The Discovery): The authors found a special, narrow window right before you hit the Symmetric setting. They had to turn the dial extremely close to 1 (specifically, scaling it with the size of the matrix) to see a new behavior.
• The Metaphor: Imagine you are walking toward a wall. For most of the walk, the wall looks like a brick wall (Chaotic). But right at the very last step, the wall suddenly starts to look like a mirror (Symmetric). The authors discovered the exact transition zone where the wall slowly morphs from bricks to glass. They derived a new formula that describes this smooth morphing process.

4. The "Universal" Guess

The authors did all their math using "Gaussian" matrices (a specific type of random number generator, like rolling perfect dice). However, they suspect this new "morphing" behavior is universal.

• The Analogy: It's like discovering that the way water flows around a rock is the same whether the water is fresh, salty, or slightly muddy. They believe their new formula for the edge transition works for any type of random matrix, not just the perfect dice they used. They ran computer simulations with "imperfect" dice (random numbers that aren't perfectly Gaussian) and found the results matched their theory perfectly.

Summary

In short, this paper:

1. Bridged the gap between two major classes of non-Hermitian random matrices.
2. Confirmed that the center of the matrix always follows a simple circular rule.
3. Discovered a new, smooth transition zone at the edge of the matrix that happens only when you are almost perfectly symmetric.
4. Proposed that this transition is a fundamental rule of nature for these types of systems, not just a quirk of the specific math they used.

They didn't just say "it changes"; they wrote down the exact mathematical recipe for how it changes, filling a gap in our understanding of how symmetry breaks in complex systems.

Technical Summary

Technical Summary: Interpolating non-Hermitian universality classes A and AI†

Problem Statement
The paper addresses the lack of analytical understanding regarding the transition between non-Hermitian universality classes, specifically between Class A (complex Ginibre matrices, lacking time-reversal symmetry) and Class AI† (complex symmetric matrices, possessing positive time-reversal symmetry). While the bulk spectral statistics for both classes are known to follow the circular law, their edge behaviors differ significantly. Previous works have established the eigenvalue density for Class A and, very recently, for Class AI†, but no analytical framework existed to describe the interpolation between these two regimes or to model the breaking of time-reversal symmetry in a continuous manner. The authors aim to construct a Gaussian ensemble that interpolates between these classes and derive the joint probability distribution function (JPDF) of an eigenvalue and its associated normalized right eigenvector, as well as the resulting eigenvalue density, at both finite matrix size $N$ and in the asymptotic limit.

Methodology
The authors introduce a non-Hermitian Interpolating Ensemble (nHIE) defined by the random matrix $X_N = \sqrt{\frac{1+\tau}{2}}G_N + \sqrt{\frac{1-\tau}{2}}G_N^T$, where $G_N$ is a complex Ginibre matrix and $\tau \in [0,1]$ is an interpolation parameter. This is re-parameterized using a symmetry parameter $\sigma = \sqrt{1-\tau^2}$.

To derive the main results, the authors employ the Kac-Rice formalism recently developed by Fyodorov for non-Hermitian random matrices. This method allows for the calculation of the JPDF of an eigenvalue and its eigenvector by evaluating the expectation of an exponentiated trace involving Grassmann variables. The derivation involves:

1. Formulating the JPDF using the Kac-Rice formula with specific correlation structures for the matrix entries.
2. Performing ensemble averages over the Gaussian distribution of the matrix entries.
3. Utilizing Hubbard-Stratonovich transformations to decouple quartic Grassmann terms.
4. Reducing the resulting high-dimensional integrals to the evaluation of a Pfaffian of a $4N \times 4N$ matrix, which is then mapped to a $4 \times 4$ matrix determinant.
5. Integrating over the eigenvector degrees of freedom to obtain the marginal eigenvalue density.
6. Conducting asymptotic analysis as $N \to \infty$ for two distinct regimes: the bulk ($z = \sqrt{N}w$) and the edge ($|z| = \sqrt{N} + \eta$). Crucially, the edge analysis considers two scaling regimes for $\sigma$: fixed $\sigma$ (Strong Asymmetry) and $\sigma = 1 - \kappa N^{-1/2}$ (Weak Asymmetry).

Key Contributions and Results

1. Finite-N JPDF and Density: The authors derive an exact, closed-form expression for the JPDF of an eigenvalue $z$ and its normalized right eigenvector $v$ for the nHIE at finite $N$ (Theorem 1). From this, they derive the exact marginal density of eigenvalues $\rho^{(nHIE)}_N(z; \sigma)$ (Corollary 2). These formulas are valid for any $\sigma \in [0, 1]$ and reduce to the known results for Class A ($\sigma=0$) and Class AI† ($\sigma=1$).

2. Bulk Universality: In the bulk scaling limit ($N \to \infty$), the authors prove that the eigenvalue density converges to the standard circular law, $\frac{1}{\pi}\Theta[1-|w|^2]$, for all values of the symmetry parameter $\sigma$. This confirms that the interpolation does not alter the leading-order bulk statistics.

3. Strong Asymmetry Regime (Fixed $\sigma$): For fixed $\sigma \in [0, 1)$ as $N \to \infty$, the edge density is shown to follow the behavior associated with Class A (Ginibre matrices), given by $\frac{1}{2\pi}\text{erfc}(\sqrt{2}\eta)$. This indicates that unless the matrix is asymptotically symmetric, the system falls into the Ginibre universality class at the edge.

4. Weak Asymmetry Regime (Scaling $\sigma$): The paper identifies a transitional regime where the symmetry parameter scales as $\sigma = 1 - \kappa N^{-1/2}$. In this "weak asymmetry" regime, the authors derive a new edge density formula (Corollary 5) that smoothly interpolates between the edge density of Class AI† (at $\kappa=0$) and Class A (as $\kappa \to \infty$). This formula represents a previously unknown regime of weak asymmetry where time-reversal symmetry is broken but remains close to the symmetric limit.

5. Universality Conjecture: The authors conjecture that the derived edge density formulas for both the strong and weak asymmetry regimes hold universally for non-Gaussian matrices that satisfy the same correlation structure, provided they belong to the same universality classes. They provide numerical evidence supporting this conjecture by simulating matrices with Bernoulli and uniform entry distributions, observing strong agreement with the Gaussian-derived formulas.

Significance
The paper claims to provide the first analytical description of the transition between non-Hermitian universality classes A and AI†. It serves as the non-Hermitian extension of the seminal works by Mehta and Pandey, who studied the breaking of time-reversal symmetry in Hermitian matrices (interpolating between GOE and GUE). By establishing the existence of a "weak asymmetry" regime with a distinct edge density, the work fills a gap in the understanding of non-Hermitian spectral statistics. The results offer a unified framework to study the breaking of time-reversal symmetry in open quantum systems and provide a rigorous foundation for future studies on eigenvector non-orthogonality and other statistical properties across these classes.