Exceptional-Point Detector (EPD)

• Exceptional-Point Detector (EPD) is a sensor operating at a defective spectral degeneracy that exploits nonanalytic square-root splitting for high sensitivity.
• EPDs are implemented in diverse platforms—electronic, photonic, and optomechanical—by tuning system parameters to a second-order exceptional point in the spectral response.
• EPD designs face practical challenges such as precise tuning, stability management, and noise-limited sensitivity issues that demand innovative engineering solutions.

An Exceptional-Point Detector (EPD) is a detector or sensor operated at an exceptional point of degeneracy so that a minute perturbation lifts a defective modal degeneracy and produces an anomalously large change in resonance frequency, wavenumber, transmission peak, oscillation frequency, or scattering spectrum (Kazemi et al., 2019, Kononchuk et al., 2020, Corsaro et al., 30 Dec 2025). In the underlying literature, the same acronym often denotes the exceptional point of degeneracy itself; detector operation is therefore organized around creating, identifying, and perturbing that singular operating point (Kazemi et al., 2018). Across electronic, photonic, optomechanical, beam-wave, and scattering platforms, the dominant mechanism is the same: tune a system to a second-order defective degeneracy, then use the ensuing square-root spectral splitting as the sensing observable.

1. Concept, scope, and terminology

An EPD is a detector whose metrological resource is a non-Hermitian or Floquet-type degeneracy at which two or more eigenstates coalesce in both eigenvalue and eigenvector. This is stronger than an ordinary degeneracy, where eigenvalues may coincide while eigenvectors remain linearly independent (Kazemi et al., 2018). In practice, most realized or explicitly analyzed detector architectures in the cited literature are second-order: two states merge into a single Jordan-block state, and the perturbation response follows a square-root law (Kazemi et al., 2019, Nikzamir et al., 2021, Rouhi et al., 2021).

The term covers both fully developed sensors and more limited EP-identification platforms. Some systems are explicitly built to measure an external quantity, such as acceleration in a PT-symmetric optomechanical multilayer (Kononchuk et al., 2020), biosensing concentration in a time-modulated RF resonator (Kazemi et al., 2019), or capacitance perturbations in gyrator-coupled circuits (Nikzamir et al., 2021). Others are closer to detector back-ends or verification architectures: a lossy beamsplitter with photon-number-resolved post-selection realizes arbitrary-order EPs on demand (Quiroz-Juárez et al., 2019), a PT-symmetric waveguide array uses a direct intensity criterion to locate the symmetry-breaking point (Bahar et al., 1 Jul 2025), and periodic oscillator arrays infer EPD operation from synchronized saturation, phase progression, and length-independent oscillation frequency (Nikzamir et al., 2023, Nikzamir et al., 24 Jan 2025).

A notable feature of the field is architectural diversity. EPDs are induced by balanced gain and loss in PT-symmetric systems, by temporal periodicity in single resonators, by nonreciprocal gyrator coupling, by reciprocal four-element circuits with negative reactances, by space-time modulation of a single transmission line, and by radiative coupling in a sphere dimer whose materials are jointly tuned to enforce a real-frequency EPD (Kazemi et al., 2019, Rouhi et al., 2023, Rouhi et al., 2020, Corsaro et al., 30 Dec 2025). This breadth has shifted the subject from a narrow PT-symmetry program toward a broader detector design paradigm centered on defective spectral singularities.

2. Mathematical structure of EPD operation

The common mathematical feature is a defective operator. In linear time-periodic systems, the relevant object is the one-period monodromy or state-transition matrix $\Phi$, with Floquet eigenproblem

$\Phi \Psi(t)=\lambda \Psi(t),$

and, at the EPD, generalized eigenvectors satisfy

$$\left(\Phi-\lambda_e I\right)^p \Psi^{(p)}=0,\qquad p=1,2,\ldots,P$$

(Kazemi et al., 2018). In transfer-matrix or coupled-mode formulations, the same role is played by a transfer matrix, effective Hamiltonian, or reduced impedance matrix (Kononchuk et al., 2020, Rouhi et al., 2020, Corsaro et al., 30 Dec 2025).

The detector principle follows from the local perturbation theory of a defective eigenvalue. For a second-order EPD, the perturbed eigenvalues obey a Puiseux expansion of the form

$$\lambda_p(\delta)\approx \lambda_e+(-1)^p\alpha_1\sqrt{\delta},$$

or, equivalently in frequency or wavenumber form,

$$\Delta f \propto \sqrt{\delta},\qquad \Delta \omega \propto \sqrt{\epsilon},\qquad \Delta \beta \propto \sqrt{\Delta}$$

(Kazemi et al., 2019, Kazemi et al., 2018, Rouhi et al., 2020). This nonanalytic response is the core transduction mechanism: ordinary resonant sensors typically shift linearly in the perturbation, whereas an EPD sensor exhibits fractional-power splitting.

The same structure generalizes to higher order. In the optomechanical PT-symmetric accelerometer, the perturbation theory is written as

$$\omega_n=\omega_{EP}+\sum_{m=1}^{\infty} c_m^{(n)}\epsilon^{m/N},$$

so an $N$-th order EP produces leading response $\Delta\omega\propto \epsilon^{1/N}$ (Kononchuk et al., 2020). In the post-selected lossy beamsplitter, an $N$-photon excitation yields an EP of order $N+1$ in the $\Phi \Psi(t)=\lambda \Psi(t),$0-photon manifold, providing a rare analytically solvable route to programmable high-order degeneracy (Quiroz-Juárez et al., 2019).

A second generic consequence is altered time evolution. At a second-order EPD, the defective Jordan structure produces an algebraic prefactor. In time-periodic resonators this appears as linear growth of state amplitude and quadratic growth of stored energy (Kazemi et al., 2019, Kazemi et al., 2018). In the reciprocal four-element LC circuit, the charge in the negative capacitor exhibits a linearly increasing envelope at the EPD, again reflecting double-pole dynamics rather than ordinary sinusoidal evolution (Rouhi et al., 2023).

3. Physical realizations and control parameters

The literature shows that EPD detectors are not tied to a single physical mechanism. They can be organized by how the degeneracy is created and by which parameter tunes the system onto the defective point.

Platform	EPD formation mechanism	Primary readout
Single LTP LC resonator	Periodic modulation of capacitance; tune $\Phi \Psi(t)=\lambda \Psi(t),$1	FFT peak splitting; linear voltage growth (Kazemi et al., 2019)
RF biosensor with IDC	Time-modulated capacitance in parallel with biosensing branch	Resonance shift of Floquet harmonic (Kazemi et al., 2019)
PT optomechanical accelerometer	Coupled defect modes with balanced gain/loss; acceleration detunes one cavity	Transmission-peak shift $\Phi \Psi(t)=\lambda \Psi(t),$2 (Kononchuk et al., 2020)
Gyrator-coupled LC circuits	Nonreciprocal coupling and negative reactances	Resonance or self-oscillation frequency (Nikzamir et al., 2021, Rouhi et al., 2021)
Space-time-modulated transmission line	Traveling modulation of per-unit-length capacitance	Wavenumber splitting $\Phi \Psi(t)=\lambda \Psi(t),$3 (Rouhi et al., 2020)
Sphere dimer scatterer	Coupled dipolar resonances; joint material tuning in retarded regime	Scattering-peak splitting or broadening (Corsaro et al., 30 Dec 2025)
Reciprocal four-element LC circuit	Shared capacitor and parallel inductor tuned to Jordan-form condition	Eigenvalue bifurcation near EPD (Rouhi et al., 2023)

Among these, the most consequential conceptual shift was the demonstration that a single passive linear time-periodic resonator can host a second-order EPD without requiring two coupled resonators and balanced gain/loss. In that architecture, the only principal tuning knob is the modulation frequency $\Phi \Psi(t)=\lambda \Psi(t),$4, and the defective point occurs at isolated Brillouin-zone center or edge conditions of the monodromy matrix (Kazemi et al., 2019). This differs sharply from PT-symmetric circuit sensors, where the EPD is typically contingent on simultaneous tuning of multiple active and passive parameters (Kazemi et al., 2019, Kononchuk et al., 2020).

Nonreciprocal gyrator-based circuits form a second major class. Two LC resonators, one implemented with negative reactances, can be tuned to a second-order EPD with purely real eigenfrequency in the lossless model; in hardware, unavoidable loss and active-element nonidealities usually drive the system into instability, which is then exploited as a self-oscillating sensing regime (Nikzamir et al., 2021, Rouhi et al., 2021). By contrast, a later reciprocal circuit shows that a second-order EPD with the same Jordan canonical form can be realized using only two LC loops with one shared capacitor, although exact operation again requires negative inductance and negative capacitance (Rouhi et al., 2023).

Distributed wave systems provide a third family. A single space-time-modulated transmission line exhibits an EPD by tuning $\Phi \Psi(t)=\lambda \Psi(t),$5, $\Phi \Psi(t)=\lambda \Psi(t),$6, or $\Phi \Psi(t)=\lambda \Psi(t),$7 in the traveling-wave modulation of the per-unit-length capacitance (Rouhi et al., 2020). Active radiating arrays and beam-wave oscillators use distributed gain and distributed extraction; in these systems the EPD is diagnosed through mode coalescence, threshold scaling, or saturation to a degenerate synchronized state rather than through a passive resonance spectrum (Mealy et al., 2021, Nikzamir et al., 24 Jan 2025, Mealy et al., 2020).

4. Detection observables and experimental signatures

The most direct EPD observable is branch coalescence in a reconstructed dispersion or resonance diagram. In the time-modulated single resonator, the detector was characterized by exciting the tank with an initial capacitor voltage, extracting resonance peaks from the Fourier transform of the natural response, and tracking the coalescence of two branches as $\Phi \Psi(t)=\lambda \Psi(t),$8 was swept (Kazemi et al., 2019). In the STM transmission line, the corresponding observable is wavenumber bifurcation in the Floquet-Bloch dispersion relation, with the first-order Puiseux approximation reproducing the split branches near the EPD (Rouhi et al., 2020).

A second class of observables is time-domain algebraic growth. The LTP resonator shows linear growth of the capacitor voltage exactly at the second-order EPD, in contrast to oscillatory decay away from it and exponential domination in the unstable regime (Kazemi et al., 2019). The earlier general theory for time-periodic resonators identifies the same signature as linear growth of current or voltage even in the absence of static gain (Kazemi et al., 2018). In the reciprocal four-element LC circuit, a linearly increasing charge envelope plays the same role (Rouhi et al., 2023).

Many photonic and scattering realizations rely instead on spectral lineshape observables. In the PT-symmetric optomechanical accelerometer, the scattering poles and transmission peaks inherit the square-root dependence of the underlying eigenfrequencies; after perturbation, one resonance becomes narrow and high-$\Phi \Psi(t)=\lambda \Psi(t),$9 while the other becomes overdamped (Kononchuk et al., 2020). In the sphere dimer, the practical signatures are two separated scattering peaks on one side of the EPD, a coalesced peak at the EPD, and a single broadened peak on the other side, depending on whether the perturbation mainly alters real or imaginary parts of the eigenfrequencies (Corsaro et al., 30 Dec 2025).

Several papers introduce observables specifically for detecting the presence or order of an EPD rather than sensing an external measurand. In the lossy beamsplitter, the post-selected no-loss survival probability obeys

$$\left(\Phi-\lambda_e I\right)^p \Psi^{(p)}=0,\qquad p=1,2,\ldots,P$$0

so the power-law exponent identifies an EP of order $$\left(\Phi-\lambda_e I\right)^p \Psi^{(p)}=0,\qquad p=1,2,\ldots,P$$1 in the $$\left(\Phi-\lambda_e I\right)^p \Psi^{(p)}=0,\qquad p=1,2,\ldots,P$$2-photon manifold (Quiroz-Juárez et al., 2019). In the PT-symmetric waveguide array, an intensity-only indicator

$$\left(\Phi-\lambda_e I\right)^p \Psi^{(p)}=0,\qquad p=1,2,\ldots,P$$3

locates the symmetry-breaking point directly from the propagation pattern, without diagonalizing the Hamiltonian (Bahar et al., 1 Jul 2025). In periodic active arrays, the coalescence parameter $$\left(\Phi-\lambda_e I\right)^p \Psi^{(p)}=0,\qquad p=1,2,\ldots,P$$4, phase progression between cells, uniform saturated gain, and oscillation frequency independent of array length serve as practical signatures of saturation at an EPD (Nikzamir et al., 2023, Nikzamir et al., 24 Jan 2025).

A further signature in beam-wave oscillators is macroscopic threshold scaling. In folded-waveguide BWOs with distributed power extraction, the starting current obeys

$$\left(\Phi-\lambda_e I\right)^p \Psi^{(p)}=0,\qquad p=1,2,\ldots,P$$5

and the nonvanishing asymptote $$\left(\Phi-\lambda_e I\right)^p \Psi^{(p)}=0,\qquad p=1,2,\ldots,P$$6 is taken as an operational fingerprint of EPD-induced degenerate synchronization (Mealy et al., 2021).

5. Sensor classes and application domains

The clearest single-resonator EPD sensor is the linear time-periodic LC detector. A piecewise-constant capacitance alternates between $$\left(\Phi-\lambda_e I\right)^p \Psi^{(p)}=0,\qquad p=1,2,\ldots,P$$7 and $$\left(\Phi-\lambda_e I\right)^p \Psi^{(p)}=0,\qquad p=1,2,\ldots,P$$8, the modulation frequency is tuned to an EPD, and a small perturbation of one capacitance level produces the characteristic splitting

$$\left(\Phi-\lambda_e I\right)^p \Psi^{(p)}=0,\qquad p=1,2,\ldots,P$$9

(Kazemi et al., 2019). In the experimental implementation, the time-varying capacitance is synthesized by a fixed capacitor, an analog multiplier, and a two-level pump voltage. The detector observable is the frequency separation of FFT peaks of the capacitor voltage during the “ON” interval of a reset cycle, with experimentally resolvable differences down to $$\lambda_p(\delta)\approx \lambda_e+(-1)^p\alpha_1\sqrt{\delta},$$0 (Kazemi et al., 2019).

The biosensing extension of this platform adds an interdigitated biosensing capacitor $$\lambda_p(\delta)\approx \lambda_e+(-1)^p\alpha_1\sqrt{\delta},$$1 and conductance $$\lambda_p(\delta)\approx \lambda_e+(-1)^p\alpha_1\sqrt{\delta},$$2 in parallel with the time-modulated capacitor. Changes in analyte concentration modify the effective dielectric and conductive properties of the IDC, perturbing the Floquet operator and shifting the resonance of the sixth harmonic near $$\lambda_p(\delta)\approx \lambda_e+(-1)^p\alpha_1\sqrt{\delta},$$3 kHz. The detector is explicitly compared with a standard LC resonator and is reported to yield about a $$\lambda_p(\delta)\approx \lambda_e+(-1)^p\alpha_1\sqrt{\delta},$$4 larger relative resonance shift than a cited conventional RF biosensor at $$\lambda_p(\delta)\approx \lambda_e+(-1)^p\alpha_1\sqrt{\delta},$$5 (Kazemi et al., 2019). The same paper models both uniformly dissolved analytes, such as glucose, and thin adsorbed layers, such as a keratin-like film (Kazemi et al., 2019).

Optomechanical acceleration sensing is the most developed PT-symmetric EPD detector in the dataset. A one-dimensional silicon/air multilayer contains two balanced gain/loss defect cavities and a mechanically movable quarter-wave silicon layer $$\lambda_p(\delta)\approx \lambda_e+(-1)^p\alpha_1\sqrt{\delta},$$6; in-plane acceleration changes an adjacent air gap and detunes only one cavity. The resulting second-order EPD splitting obeys

$$\lambda_p(\delta)\approx \lambda_e+(-1)^p\alpha_1\sqrt{\delta},$$7

and the transmission-peak displacement follows $$\lambda_p(\delta)\approx \lambda_e+(-1)^p\alpha_1\sqrt{\delta},$$8, close to the ideal square-root law (Kononchuk et al., 2020). The same platform is described as suitable for low-frequency vibrations as well as acceleration.

Circuit implementations target physical, chemical, or biological perturbations mapped into capacitance or inductance changes. In the gyrator-based high-sensitivity oscillator, a small capacitance perturbation shifts the saturated oscillation frequency, and the measured spectral linewidth is about $$\lambda_p(\delta)\approx \lambda_e+(-1)^p\alpha_1\sqrt{\delta},$$9 Hz with a clean spectrum down to about $$\Delta f \propto \sqrt{\delta},\qquad \Delta \omega \propto \sqrt{\epsilon},\qquad \Delta \beta \propto \sqrt{\Delta}$$0 dB below the peak (Nikzamir et al., 2021). The related theoretical two-unstable-resonator circuit proposes liquid-level or liquid-content sensing by using one capacitor as a fluid-dependent transducer, with the EPD converting small capacitance changes into square-root resonance splitting (Rouhi et al., 2021).

Wave and scattering systems extend the same logic beyond lumped circuits. In the space-time-modulated transmission line, tiny changes in any transmission-line or modulation parameter cause large shifts in the complex Bloch wavenumber, suggesting a highly sensitive transmission-line sensor (Rouhi et al., 2020). In the sphere dimer, geometric perturbation of the inter-sphere spacing causes square-root splitting of the real parts of scattering resonances, directly observable as splitting or broadening in scattering and absorption spectra (Corsaro et al., 30 Dec 2025). A plausible implication is that EPD detection need not be tied to resonators in the narrow circuit sense; it can also be realized as a scattering-resonance measurement in open electromagnetic systems.

6. Limitations, controversies, and adjacent alternatives

EPD detectors are extremely sensitive by construction, and the same feature creates their main engineering difficulties. Precise tuning is typically mandatory: time-modulated detectors must be biased very near the correct $$\Delta f \propto \sqrt{\delta},\qquad \Delta \omega \propto \sqrt{\epsilon},\qquad \Delta \beta \propto \sqrt{\Delta}$$1 and modulation depth (Kazemi et al., 2018), PT-symmetric designs require careful matching of gain, loss, and coupling (Kononchuk et al., 2020), and retarded sphere-dimer EPDs require joint material tuning rather than a simple conjugate-susceptibility rule (Corsaro et al., 30 Dec 2025). In several circuit platforms, the required negative reactances are synthesized by active circuits, so the nominally conservative model is only approximate in hardware (Nikzamir et al., 2021, Rouhi et al., 2023).

A second limitation is saturation or outright instability. In the time-modulated LC resonator, linear growth at the EPD forces the use of a reset waveform to reinitialize the capacitor and avoid multiplier saturation (Kazemi et al., 2019). In gyrator-based resonators, small losses and active negative-impedance realizations push the linear EPD off the real axis and naturally generate unstable modes; the practical device is therefore operated as a saturated oscillator rather than a passive linear resonator (Nikzamir et al., 2021, Rouhi et al., 2021). In active periodic arrays, saturation is not merely tolerated but used as the mechanism that lands the system on an EPD-compatible operating point (Nikzamir et al., 2023, Nikzamir et al., 24 Jan 2025).

The most persistent controversy concerns whether enhanced eigenvalue responsivity yields enhanced noise-limited sensitivity. In a Brillouin ring laser gyroscope biased near an EP, the signal enhancement factor

$$\Delta f \propto \sqrt{\delta},\qquad \Delta \omega \propto \sqrt{\epsilon},\qquad \Delta \beta \propto \sqrt{\Delta}$$2

is shown to be exactly compensated, in the fundamental-noise-limited regime, by linewidth broadening quantified by the Petermann factor

$$\Delta f \propto \sqrt{\delta},\qquad \Delta \omega \propto \sqrt{\epsilon},\qquad \Delta \beta \propto \sqrt{\Delta}$$3

so $$\Delta f \propto \sqrt{\delta},\qquad \Delta \omega \propto \sqrt{\epsilon},\qquad \Delta \beta \propto \sqrt{\Delta}$$4 and the fundamental signal-to-noise ratio does not improve (Wang et al., 2019). This result does not invalidate EPD detectors as transducers, but it sharply distinguishes enhanced responsivity from guaranteed metrological advantage.

That critique has already motivated adjacent alternatives. A recent chiral Hermitian cavity realizes a critical-point sensor with an EP-like square-root response in the spectral-extremum splitting, yet preserves orthogonal eigenvectors and keeps $$\Delta f \propto \sqrt{\delta},\qquad \Delta \omega \propto \sqrt{\epsilon},\qquad \Delta \beta \propto \sqrt{\Delta}$$5. It is therefore presented not as a true EP detector, but as an EP-like Hermitian critical-point detector that avoids the Petermann-factor divergence of non-Hermitian counterparts (Tang et al., 24 Jan 2026). This suggests that the enduring value of EPD research may lie less in any single platform than in the broader design principle it exposed: operation at a spectral critical point where a measurable quantity acquires a fractional-power perturbation law.

Within that broader perspective, the Exceptional-Point Detector is best understood as a family of devices whose defining trait is not a particular material system, but a deliberate biasing strategy: create a defective spectral degeneracy, encode the measurand as a perturbation of one control parameter, and read out the resulting nonanalytic splitting in frequency, wavenumber, transmission, intensity, or oscillation observables (Kazemi et al., 2019, Kononchuk et al., 2020, Rouhi et al., 2020).