🔬 mesoscale physics

Universal Eigenvector Statistics of Non-Hermitian Random Matrices

This paper establishes a universal duality between Hermitian level statistics and non-Hermitian eigenvector overlaps by deriving analytical expressions for these overlaps across ten symmetry classes using fermionic replica nonlinear $\sigma$ models, a result supported by numerical calculations in various physical models.

Original authors: Ze Chen, Zhenyu Xiao, Shinsei Ryu

Published 2026-09-21

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Ze Chen, Zhenyu Xiao, Shinsei Ryu

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

In the study of complex systems, from the flow of electricity in a wire to the behavior of light in a laser, scientists often rely on a powerful idea: that the chaotic details of a specific system can be understood by looking at the statistical patterns of a much larger, random collection of similar systems. This approach, known as random matrix theory, has long been the gold standard for understanding how energy levels in quantum systems are arranged. For decades, researchers focused almost exclusively on the "eigenvalues," which are the specific energy values a system can hold. By studying how these values are spaced apart, scientists could predict whether a system was behaving in an orderly, predictable way or in a chaotic, random fashion. This method has successfully explained everything from the vibrations of atomic nuclei to the transport of electrons in disordered materials.

However, energy levels are only half the story. Every energy value is associated with a specific "shape" or pattern of the system, known as an eigenvector. In the familiar world of standard quantum mechanics, these patterns are perfectly independent of one another, like distinct, non-overlapping shadows. But in a broader class of physical systems—those that lose energy to their surroundings or interact with an environment in a way that breaks perfect symmetry—these patterns become "non-orthogonal." This means they lean into one another, overlapping in a way that makes the system highly sensitive to tiny changes and prone to sudden, dramatic amplification. While scientists have known for some time that these overlaps exist and matter, they have lacked a complete map of how these overlaps behave across the vast landscape of different physical symmetries. Without this map, it has been difficult to predict how complex, open systems would respond to disturbances or how they would evolve over time.

A team of researchers at Princeton University has now filled this gap by uncovering a universal set of rules that govern these eigenvector overlaps. They discovered that just as the spacing of energy levels follows predictable statistical laws, the way these patterns overlap is also governed by a strict, universal order. The researchers found that the entire behavior of these overlaps can be traced back to the same underlying mathematical structures that dictate the spacing of energy levels in simpler, closed systems. In essence, they established a deep connection, or duality, between the geometry of the system's patterns and the statistics of its energy values. This means that by knowing how the energy levels are arranged in a related, simpler system, one can precisely calculate how the patterns of a complex, open system will overlap and interact.

To reach this conclusion, the team analyzed ten distinct categories of symmetry that describe how physical systems behave under time reversal and other transformations. They focused on the "bulk" of the system, where the energy levels are densely packed, as well as the "hard edge," a special region near zero energy where unique physical protections often occur. Using a sophisticated mathematical technique involving "replica" models—which allow physicists to calculate averages over many random possibilities—they derived exact formulas for how these overlaps behave. They found that in the main body of the spectrum, the way patterns overlap depends on just three fundamental types of symmetry, regardless of the specific details of the material. Near the zero-energy edge, the behavior becomes even more intricate, depending on the specific topological features of the system, such as the presence of protected zero-energy states.

The researchers did not stop at theory. They tested their predictions against a wide variety of physical models, including complex networks of interacting particles, disordered lattices that mimic impure metals, and models of fermions losing energy to their environment. In every case, from the simplest random matrices to these intricate physical simulations, the numerical data matched their theoretical formulas perfectly. The study confirmed that the normalized shape of these overlaps is universal: it does not depend on the specific material or the size of the system, but only on the fundamental symmetry class to which the system belongs. This universality holds true whether the system is a simple random matrix or a complex model of a dissipative quantum system.

One of the most striking findings is how these overlaps behave when energy levels get very close to each other. In the bulk of the spectrum, as two energy levels approach one another, the average overlap between their patterns grows dramatically, diverging in a specific way that depends on the symmetry class. This mathematical divergence signals that the system is becoming extremely sensitive to perturbations, a phenomenon that is crucial for understanding how lasers amplify light or how quantum systems respond to noise. Similarly, near the zero-energy edge, the presence of protected states causes the overlaps to spike in a predictable manner, revealing how the system's topology influences its stability. The researchers showed that these divergences are not random accidents but are direct consequences of the underlying symmetry, providing a clear, quantitative link between the system's structure and its dynamical response.

This work effectively extends the reach of random matrix theory from the study of energy levels to the study of the system's shapes and patterns. By proving that the statistics of eigenvector overlaps are determined by the same principles that govern energy level spacing, the researchers have provided a new, powerful tool for analyzing open quantum systems. Their results suggest that the geometry of these systems is not a chaotic mess but is instead organized by a hidden, universal order. This insight allows scientists to predict the behavior of complex, open systems with a level of precision that was previously impossible, offering a new benchmark for identifying quantum chaos and understanding the dynamics of systems that exchange energy with their surroundings. The study confirms that even in the most complex and dissipative environments, the fundamental laws of symmetry impose a clear, predictable structure on how the system's components interact and overlap.

Technical Summary: Universal Eigenvector Statistics of Non-Hermitian Random Matrices

Problem Statement
Random Matrix Theory (RMT) has long provided a statistical framework for the eigenvalues of complex quantum systems, establishing that chaotic spectra exhibit universal random-matrix statistics. While this correspondence is well-understood for Hermitian systems, its extension to non-Hermitian systems has primarily focused on eigenvalue distributions. However, non-Hermitian systems exhibit a distinct feature: eigenvector nonorthogonality. The left and right eigenvectors of a non-Hermitian matrix are generally distinct, leading to nontrivial biorthogonal overlaps. These overlaps govern critical physical phenomena, including spectral sensitivity to perturbations (particularly near exceptional points), transient amplification, relaxation dynamics, and spontaneous emission noise.

Despite previous work establishing overlap statistics for specific ensembles like the Ginibre ensemble, a comprehensive understanding of how symmetry organizes eigenvector statistics across the full landscape of non-Hermitian random matrices was lacking. Specifically, it was unclear whether the established duality between non-Hermitian level statistics and Hermitian spectral correlations could be extended to eigenvector overlaps across the ten symmetry classes of the Altland–Zirnbauer0 (AZ0) classification.

Methodology
The authors employ the fermionic replica nonlinear $\sigma$ model technique to derive analytical expressions for eigenvector overlaps in the limit of large matrix size ( $N \to \infty$ ).

Replica Construction: The authors construct $n$ $n$ -replica partition functions ( $Z_{1,n}$ $Z_{1, n}$ and $Z_{2,n}$ $Z_{2, n}$ ) to generate the density-weighted diagonal ( $O_1$ $O_{1}$ ) and off-diagonal ( $O_2$ $O_{2}$ ) overlaps.
- For diagonal overlaps, a one-point partition function with a source parameter $\eta$ is introduced. The second derivative with respect to $\eta$ extracts the trace of the product of projectors, $\text{Tr}(\Pi_a \Pi_a^\dagger)$ .
- For off-diagonal overlaps, a two-copy partition function with mixed resolvents is utilized to generate correlations between distinct eigenvalues.
Hermitization and Duality: A key methodological step involves "Hermitization," where the non-Hermitian matrix $H$ is embedded into a larger Hermitian matrix $H_z$ . This maps the non-Hermitian problem to a Hermitian chiral class (e.g., AIII, CI, DIII).
Saddle-Point Analysis: In the large- $N$ limit, the partition functions are evaluated at the saddle point. The source derivatives are expressed in terms of angular integrals over the symmetry groups (e.g., $U(n)$ , $Sp(n)$ , $O(n)$ ) that determine the eigenvalue statistics.
Poisson Transforms: The authors relate the overlap statistics to the Poisson transforms ( $P_1$ and $P_2$ ) of the known eigenvalue correlation functions of the corresponding Hermitian partner classes. This establishes a direct link: the same Hermitian spectral correlations that determine non-Hermitian level statistics also determine the eigenvector overlaps.

Key Contributions and Results

Symmetry Classification: The work covers the ten AZ0 symmetry classes, which are defined by transposition-based time-reversal symmetry ( $\text{TRS}^\dagger$ ), particle-hole symmetry (PHS), and sublattice symmetry (SLS). These classes are distinguished by the presence or absence of these symmetries and their signs.
Bulk Statistics (Spectral Interior):
- Diagonal Overlaps ( $O_1$ ): The leading-order bulk diagonal overlap is found to be $\pi O_1(z) = \frac{\beta}{2}(1 - |z|^2/N)$ , where $\beta$ is determined solely by $\text{TRS}^\dagger$ ( $\beta=1, 2, 4$ for classes AI $^\dagger$ , A, and AII $^\dagger$ , respectively). This amplitude differs by symmetry class but is not universal in absolute magnitude without normalization.
- Off-Diagonal Overlaps ( $O_2$ ): The normalized off-diagonal overlap, $O_2(z_1, z_2)/O_1(z)$ , exhibits three universal forms determined exclusively by $\text{TRS}^\dagger$ . These forms are expressed in terms of the Poisson transform $P_2$ of the Hermitian level correlation function:
  $\frac{2\pi O_2(z_1, z_2)}{O_1(z)} = -1 + \frac{d}{d|\omega|^2} \left[ |\omega|^2 P_2(|\omega|^2) \right]$
  where $\omega = z_1 - z_2$ . The three families correspond to the Hermitian classes A, AI, and AII.
- Asymptotics: At small separations ( $|\omega| \to 0$ ), the average off-diagonal overlap at fixed eigenvalues diverges as $|\omega|^{-2}$ (or logarithmically modified for AI $^\dagger$ ), reflecting the scaling of spectral projectors near coalescence. At large separations, the overlap decays as $|\omega|^{-4}$ .
Hard-Edge Statistics (Near the Origin):
- For the seven classes with PHS or SLS, statistics near the origin depend on the full symmetry class and the topological sector (characterized by an index $\nu$ ).
- The diagonal overlap is given by $\pi O_1(z) = \frac{\beta}{4}[1 + P_1(|z|^2)]$ , where $P_1$ is the Poisson transform of the Hermitian hard-edge spectral density.
- Zero Modes: Protected zero modes (present in chiral and certain non-chiral classes) contribute a $1/|z|^2$ term to the overlap density, leading to specific divergences in the average overlap at fixed eigenvalues depending on the topological sector $\nu$ .
Numerical Verification:
- The analytical predictions were verified numerically using Gaussian ensembles for all ten AZ0 classes.
- Universality was further tested in physical models: the non-Hermitian Sachdev–Ye–Kitaev (SYK) model, the non-Hermitian Anderson model, and a dissipative fermion model on a bipartite lattice. In all cases, after appropriate spectral unfolding and normalization, the numerical overlap profiles matched the theoretical predictions.

Significance and Claims
The paper claims to establish a duality between Hermitian level statistics and non-Hermitian eigenvector overlaps. The central finding is that the same Hermitian spectral correlations determine both the eigenvalue statistics and the eigenvector nonorthogonality of non-Hermitian random matrices in the large- $N$ limit.

The authors position this work as an extension of the random-matrix description of quantum chaos from purely spectral properties to eigenvector geometry. By providing analytical benchmarks for eigenvector overlaps across ten symmetry classes, the work offers new tools for identifying ergodic behavior and characterizing non-Hermitian dynamics in open quantum systems, optical systems, and dissipative many-body systems. The results suggest that the relationship between level statistics and eigenvector correlations may be a robust feature of quantum chaos, potentially surviving beyond the strict random-matrix scaling limit, though the paper notes that investigating this in localized regimes remains an open question.

More like this