12 Jul 2026
Reading the Exceptional-Point Project I: Non-Hermitian Arrays from First Principles
This is Part I of a five-part guide to the paper Structured-to-generic crossover at arbitrary-order exceptional points in supersymmetric $\mathcal{PT}$-symmetric arrays.
The discussion assumes that the reader already knows the basic ideas of $\mathcal{PT}$ symmetry and supersymmetry. Here we build the remaining language from first principles:
| Part | Topic |
|---|---|
| I | non-Hermitian arrays |
| II | exceptional points |
| III | phase rigidity |
| IV | Petermann factor |
| V | notation and crossover formulas used in the paper |
The goal is not merely to define words. By the end of the series, a reader should be able to move from the array Hamiltonian
\[H_N(\Omega,\gamma)=\Omega J_x+i\gamma J_z\]to the crossover estimate
\[|\delta_c|\asymp \frac{2^{N-2}}{(N-1)!} \frac{|q|^{N/2}}{|\Omega|^{N-1}}\]without treating any symbol as mysterious.
1. Begin with Coupled Amplitudes
Consider $N$ resonators, waveguides, circuit nodes, or abstract basis states. Let
\[\mathbf a(t)= \begin{pmatrix} a_1(t)\\ a_2(t)\\ \vdots\\ a_N(t) \end{pmatrix}\]collect their complex amplitudes. In temporal coupled-mode theory their linear evolution is written as
\[i\frac{d\mathbf a}{dt}=H\mathbf a.\]This equation has the same mathematical form as the Schrodinger equation. The matrix $H$ tells us:
| Matrix element | Physical role |
|---|---|
| $H_{nn}$ | resonance frequency and gain or loss at site $n$ |
| $H_{nm}$, $n\neq m$ | coupling from site $m$ to site $n$ |
| eigenvalue $E_k$ | complex modal frequency |
| eigenvector $\psi_k$ | distribution of the mode over the array |
For a waveguide array, time is replaced by propagation distance:
\[i\frac{d\mathbf a}{dz}=H\mathbf a.\]The mathematics is unchanged. Only the units change: the entries of $H$ are angular frequencies in the temporal problem and inverse lengths in the spatial problem.
2. Hermitian and Non-Hermitian Matrices
The Hermitian conjugate of a matrix is
\[H^\dagger=(H^*)^{\mathsf T}.\]A Hamiltonian is Hermitian when
\[H=H^\dagger.\]Hermitian matrices have real eigenvalues and mutually orthogonal eigenvectors. They describe closed, conservative systems in elementary quantum mechanics.
A matrix is non-Hermitian when
\[H\neq H^\dagger.\]This does not mean that the model is unphysical. It usually means that the system exchanges energy with an environment. Loss, gain, radiation leakage, measurement back-action, and eliminated external channels can all produce an effective non-Hermitian Hamiltonian.
Important distinction: non-Hermitian does not mean arbitrary. A non-Hermitian matrix may still obey reciprocity, spatial symmetries, $\mathcal{PT}$ symmetry, or other constraints.
3. What a Complex Eigenvalue Means
Suppose
\[H\psi_k=E_k\psi_k, \qquad E_k=\omega_k+i\Gamma_k.\]The corresponding time dependence is
\[\mathbf a_k(t)=e^{-iE_kt}\psi_k =e^{-i\omega_kt}e^{\Gamma_kt}\psi_k.\]Therefore:
| Part of $E_k$ | Meaning |
|---|---|
| $\operatorname{Re}E_k=\omega_k$ | oscillation frequency |
| $\operatorname{Im}E_k=\Gamma_k<0$ | decay |
| $\operatorname{Im}E_k=\Gamma_k>0$ | growth |
This is the first reason complex spectra are natural in open systems.
4. A Two-Site Example
The smallest useful example is
\[H_2= \frac{1}{2} \begin{pmatrix} i\gamma & \Omega\\ \Omega & -i\gamma \end{pmatrix}.\]Here $\Omega$ is the coupling scale and $\gamma$ is the gain-loss scale. The characteristic equation is
\[\det(EI-H_2)=0.\]Direct expansion gives
\[E^2-\frac{\Omega^2-\gamma^2}{4}=0.\]Define
\[q=\Omega^2-\gamma^2.\]Then
\[E_\pm=\pm\frac{\sqrt q}{2}.\]This single expression already displays three regimes:
| Condition | Spectrum |
|---|---|
| $q>0$ | two real eigenvalues |
| $q=0$ | both eigenvalues meet |
| $q<0$ | an imaginary-conjugate pair |
The paper uses the same quantity $q$ for every array size. It measures the signed distance from the exceptional-point boundary in squared-frequency units.
5. From Two Sites to an Array
For $N$ sites, the paper uses
\[H_N(\Omega,\gamma)=\Omega J_x+i\gamma J_z.\]The matrices $J_x$ and $J_z$ are the spin-$j$ angular-momentum matrices with
\[j=\frac{N-1}{2}.\]In the site basis, the nearest-neighbour coupling between sites $n$ and $n+1$ is
\[C_n=\frac{\Omega}{2}\sqrt{n(N-n)}, \qquad n=1,\ldots,N-1.\]The diagonal entry is proportional to
\[i\gamma m, \qquad m=j,j-1,\ldots,-j.\]Thus the array contains:
- a correlated nearest-neighbour coupling profile,
- a linear imaginary diagonal gradient,
- no long-range link in the ideal design.
The exact eigenvalues are
\[E_m=m\sqrt q, \qquad m=-j,-j+1,\ldots,j.\]The two-site result has therefore become an $N$-level ladder with common spacing $\sqrt q$.
6. Complex Symmetric Is Not Hermitian
The Hamiltonians used in the paper satisfy
\[H^{\mathsf T}=H,\]because the couplings are reciprocal. But generally
\[H^\dagger\neq H\]because of the imaginary gain-loss diagonal.
This is called a complex-symmetric matrix. It is important because its natural bilinear product is
\[\psi^{\mathsf T}\phi,\]whereas the ordinary positive norm is
\[\psi^\dagger\phi.\]The difference between these two products is the origin of phase rigidity and the Petermann factor, developed in Parts III and IV.
7. Right and Left Eigenvectors
For a general non-Hermitian matrix, right and left eigenvectors satisfy
\[H\psi_k^{R}=E_k\psi_k^{R},\]and
\[(\psi_k^{L})^\dagger H =E_k(\psi_k^{L})^\dagger.\]They are not generally related by ordinary Hermitian conjugation. Instead one uses biorthogonality:
\[(\psi_k^{L})^\dagger\psi_\ell^{R} =0, \qquad k\neq\ell.\]For a complex-symmetric matrix, the transpose structure makes the relevant self-overlap especially simple:
\[\psi_k^{\mathsf T}\psi_k.\]At an exceptional point this quantity can vanish even though $\psi_k^\dagger\psi_k$ is positive. That phenomenon is called self-orthogonality.
8. What to Remember Before Reading the Paper
| Symbol | Meaning |
|---|---|
| $N$ | number of sites or modes |
| $j=(N-1)/2$ | spin label used to construct the matrices |
| $\Omega$ | common coupling scale |
| $\gamma$ | gain-loss gradient scale |
| $q=\Omega^2-\gamma^2$ | structured distance from the exceptional point |
| $E_m=m\sqrt q$ | exact structured spectrum |
| $C_n$ | nearest-neighbour coupling at bond $n$ |
The essential lesson is:
A non-Hermitian array is a coupled-mode system whose matrix includes openness, gain, or loss. Its complex eigenvalues describe modal frequency and amplification or decay, while its nonorthogonal eigenvectors contain additional physics not visible in the spectrum alone.
Continue with Part II: Exceptional Points from First Principles.
Discussion