Non-Hermitian Zero-Index Magneto-Optical Metawaveguide

• The paper shows that engineered non-Hermitian zero-index metawaveguides convert weak magneto-optical effects into giant directional asymmetry through exceptional point–mediated branch selection.
• It integrates generalized magneto-optical theory with multiport scattering concepts to achieve impedance-matched, near-zero effective index states using both EMNZ and INZ approaches.
• Experimental results confirm enhanced nonreciprocal phase shifts and losses, with measurements significantly outperforming conventional magneto-optical devices.

A non-Hermitian zero-index magneto-optical metawaveguide is a guided electromagnetic structure in which near-zero effective index, gyrotropic time-reversal breaking, and complex-spectrum non-Hermiticity are co-engineered so that phase advance, attenuation, and scattering are governed by branch topology and open-system modal selection rather than by ordinary weak magneto-optical perturbation alone. In the current literature, the topic is anchored by an experimentally realized a-Si/Ce:YIG metawaveguide in which a zero-index quasi-Dirac point and an exceptional-point-mediated multivalued complex eigenspace generate giant directional differences in phase and loss (Li et al., 7 Sep 2025). It is supported by Hermitian and weakly lossy gyromagnetic zero-index baselines, generalized magneto-optical waveguide theory, and reciprocal non-Hermitian zero-index scattering frameworks that provide the constitutive, modal, and multiport foundations of the field (Zhou et al., 2017, Yang et al., 2022, Zhou et al., 2022, Honda et al., 2023, Yan et al., 14 Jan 2025, Xu et al., 5 Jun 2026, Wang et al., 2020).

1. Constitutive and modal foundations

The constitutive starting point is the generalized waveguide relation

$$\mathbf{D} = \varepsilon \mathbf{E} + \xi \mathbf{H}, \qquad \mathbf{B} = \zeta \mathbf{E} + \mu \mathbf{H},$$

which accommodates both magneto-optical (MO) gyrotropy and magnetoelectric (ME) coupling in a single guided-wave formalism (Honda et al., 2023). In explicitly gyromagnetic realizations, the relevant tensor typically takes the form

$$\overset\leftrightarrow{\mu} = \begin{bmatrix} \mu & i\alpha & 0\ -i\alpha & \mu & 0\ 0 & 0 & \mu_0 \end{bmatrix},$$

or its effective-medium counterpart with off-diagonal parameter $\kappa$, as used for biased ferrites and YIG-based metamaterials (Zhou et al., 2017, Yang et al., 2022). In guided configurations, these off-diagonal terms break reciprocity and enter the propagation problem through terms proportional to the propagation constant, which is why forward and backward waves can acquire different phase and loss even in otherwise static structures (Honda et al., 2023).

Within this literature, “zero index” is not a single constitutive condition. One class consists of effective epsilon-and-mu-near-zero (EMNZ) media with $\Re(\varepsilon_{\text{eff}})\approx 0$ and $\Re(\mu_{\text{eff}})\approx 0$, as in magnetically tunable YIG-pillar metamaterials and the Ce:YIG bowtie metawaveguide (Yang et al., 2022, Li et al., 7 Sep 2025). A second class is the generalized gyrotropic zero-index medium associated with a $\Gamma$-point pseudospin-$1/2$ Dirac cone, where $\varepsilon_{ez}\to 0$ while $\det(\boldsymbol{\mu}_e)\to 0$ rather than $\boldsymbol{\mu}_e\to 0$ componentwise (Wang et al., 2020). A third class is modal index-near-zero (INZ) operation, defined by a guided branch with

$$\overset\leftrightarrow{\mu} = \begin{bmatrix} \mu & i\alpha & 0\ -i\alpha & \mu & 0\ 0 & 0 & \mu_0 \end{bmatrix},$$0

so that the effective phase index vanishes while the group velocity remains finite (Zhou et al., 2022). The field therefore uses both bulk effective-medium and guided-mode notions of “zero index,” depending on architecture.

The magneto-optical part of the concept is equally heterogeneous. In gyromagnetic photonic crystals, an unpaired Dirac cone at $$\overset\leftrightarrow{\mu} = \begin{bmatrix} \mu & i\alpha & 0\ -i\alpha & \mu & 0\ 0 & 0 & \mu_0 \end{bmatrix},$$1 can be produced by combining broken time-reversal symmetry with broken sublattice symmetry, leading to a near-zero-index gyromagnetic effective medium with a Voigt parameter near unity and unidirectional domain-wall states (Zhou et al., 2017). In waveguide-based metamaterials, the same gyrotropic physics is embedded in bounded structures that support only the relevant fundamental guided mode, which is the setting most directly associated with the term metawaveguide (Yang et al., 2022).

2. Zero-index magneto-optical architectures

The most explicit realization of a non-Hermitian zero-index magneto-optical metawaveguide is a one-dimensional array of amorphous-silicon bowtie resonators on a Ce:YIG film on an sGGG substrate. Its unit-cell parameters are

$$\overset\leftrightarrow{\mu} = \begin{bmatrix} \mu & i\alpha & 0\ -i\alpha & \mu & 0\ 0 & 0 & \mu_0 \end{bmatrix},$$2

and near the Brillouin-zone center it supports two nearly degenerate modes, described as a monopole-like lower-loss branch and a dipole-like higher-loss branch, yielding an impedance-matched EMNZ response with $$\overset\leftrightarrow{\mu} = \begin{bmatrix} \mu & i\alpha & 0\ -i\alpha & \mu & 0\ 0 & 0 & \mu_0 \end{bmatrix},$$3 and $$\overset\leftrightarrow{\mu} = \begin{bmatrix} \mu & i\alpha & 0\ -i\alpha & \mu & 0\ 0 & 0 & \mu_0 \end{bmatrix},$$4 near the zero-index wavelength (Li et al., 7 Sep 2025). The zero-index state was identified experimentally by an effectively infinite spatial wavelength in the interference image over the metawaveguide region and theoretically by the near-zero wavevector associated with the quasi-Dirac point.

Microwave embodiments use a different physical route. A magnetically tunable zero-index metamaterial was realized as a square array of YIG pillars sandwiched between two copper-clad laminates. The lattice constant is $$\overset\leftrightarrow{\mu} = \begin{bmatrix} \mu & i\alpha & 0\ -i\alpha & \mu & 0\ 0 & 0 & \mu_0 \end{bmatrix},$$5, the YIG pillar radius is $$\overset\leftrightarrow{\mu} = \begin{bmatrix} \mu & i\alpha & 0\ -i\alpha & \mu & 0\ 0 & 0 & \mu_0 \end{bmatrix},$$6, the pillar height is $$\overset\leftrightarrow{\mu} = \begin{bmatrix} \mu & i\alpha & 0\ -i\alpha & \mu & 0\ 0 & 0 & \mu_0 \end{bmatrix},$$7, and the plate spacing is $$\overset\leftrightarrow{\mu} = \begin{bmatrix} \mu & i\alpha & 0\ -i\alpha & \mu & 0\ 0 & 0 & \mu_0 \end{bmatrix},$$8. Under zero bias, the structure supports an accidental $$\overset\leftrightarrow{\mu} = \begin{bmatrix} \mu & i\alpha & 0\ -i\alpha & \mu & 0\ 0 & 0 & \mu_0 \end{bmatrix},$$9-point degeneracy between an electric monopole mode, a transverse magnetic dipole mode, and a longitudinal magnetic dipole mode, producing a Dirac-like cone at $\kappa$0 and an effective EMNZ state. Under $\kappa$1, the degeneracy is lifted and the system moves from a zero-index phase to single-negative MNG and ENG phases (Yang et al., 2022).

Several other geometries supply essential baselines. A 2D gyromagnetic photonic crystal with ferrite rods on a square lattice realizes an unpaired Dirac point at $\kappa$2 near $\kappa$3, and the resulting homogenized medium behaves as a near-zero-index magneto-optical continuum with $\kappa$4 and unidirectional interface states between oppositely biased domains (Zhou et al., 2017). Layered nonreciprocal heterostructures, such as PEC–dielectric–YIG–PEC in the microwave regime and PMC/dielectric/InSb heterostructures in the terahertz regime, instead realize broadband INZ behavior by shaping one-way surface-wave dispersion so that a guided branch crosses $\kappa$5 inside a complete one-way band (Zhou et al., 2022). These distinct realizations all satisfy the same architectural principle: zero-index behavior is generated by structured gyrotropic media rather than by ordinary weakly birefringent waveguide cores.

3. Non-Hermitian branch topology and giant nonreciprocity

Non-Hermiticity enters these systems through complex eigenfrequencies or complex propagation constants generated by radiative loss, material absorption, and, in some extensions, gain. In the Ce:YIG metawaveguide, the spectrum contains a lower-loss branch and a higher-loss branch, with the higher-loss branch having approximately twice the radiation loss of the lower-loss branch. At zero magnetic field, the two branches are nearly degenerate in real frequency near $\kappa$6, but this alone does not create strong directional asymmetry. The essential step is a topological transition in the complex eigenspace under magnetic bias, organized around an exceptional point with residue $\kappa$7. The mapped eigenspace is multivalued, with two Riemann sheets and a square-root branch point; once the projected trajectory encloses the exceptional point, forward and backward waves are directed onto different photonic branches with largely distinct momenta and losses (Li et al., 7 Sep 2025).

This mechanism reframes magneto-optical nonreciprocity. In an ordinary magneto-optical waveguide, the off-diagonal tensor element of Ce:YIG gives only a small perturbation to the propagation constant or attenuation. In the non-Hermitian zero-index metawaveguide, that weak perturbation acts instead as a branch-selection trigger near a quasi-Dirac degeneracy. The relevant observables are

$\kappa$8

and both become large because the two directions no longer remain on the same band. The transition field is only about $\kappa$9, while the experiments were performed under $\Re(\varepsilon_{\text{eff}})\approx 0$0, sufficient to saturate the Ce:YIG (Li et al., 7 Sep 2025).

A related but distinct non-Hermitian route appears in the photonic pseudospin-$\Re(\varepsilon_{\text{eff}})\approx 0$1 magneto-optical effective medium derived from an unpaired $\Re(\varepsilon_{\text{eff}})\approx 0$2-point Dirac cone. There, a carefully chosen non-Hermitian permittivity perturbation produces a magneto-optical complex-conjugate metamaterial regime in which two linear bands coalesce into exceptional points at real frequency, a ring of EPs surrounds the original Dirac point, and an equivalent permeability $\Re(\varepsilon_{\text{eff}})\approx 0$3 becomes the complex conjugate of $\Re(\varepsilon_{\text{eff}})\approx 0$4 up to a real factor, making the effective refractive index $\Re(\varepsilon_{\text{eff}})\approx 0$5 real over a broad frequency range (Wang et al., 2020). In slab form, that same medium supports coherent perfect absorption and lasing, showing that non-Hermitian MO zero-index media are not restricted to passive transport.

Adjacent work on non-zero-index photon-magnon hybrids reinforces the same constitutive lesson. In a YIG-film–ISRR–microstrip structure, nonreciprocal negative refraction disappears when the imaginary part of the effective coupling is removed, indicating that the imaginary parts of $\Re(\varepsilon_{\text{eff}})\approx 0$6 and $\Re(\varepsilon_{\text{eff}})\approx 0$7 are not secondary corrections but control parameters for directional transport (Kim et al., 2024). Although that system is not a zero-index platform, it demonstrates that non-Hermitian metawaveguide design must treat loss engineering as a primary spectral degree of freedom.

4. Scattering matrices, exceptional points, and zero-index networks

The open-system side of the field is formulated most clearly through multiport scattering theory. In non-Hermitian zero-index materials connected to multiple open channels, the output amplitudes $\Re(\varepsilon_{\text{eff}})\approx 0$8 and input amplitudes $\Re(\varepsilon_{\text{eff}})\approx 0$9 satisfy

$\Re(\mu_{\text{eff}})\approx 0$0

and exceptional points are defined by coalescence of scattering eigenvalues and eigenvectors rather than by a closed-system Hamiltonian alone (Yan et al., 14 Jan 2025, Xu et al., 5 Jun 2026). In a three-channel ENZ-based architecture consisting of a central zero-index “CPU” connected to channels through air gaps, third-order lasing, reflecting, and absorbing EPs can all be realized. The absorbing third-order EP is especially notable because it is available in a purely lossy system and yields

$\Re(\mu_{\text{eff}})\approx 0$1

so that fixed-frequency output power responds ultrasensitively to perturbation (Yan et al., 14 Jan 2025).

The most general theorem so far is the arbitrary-order result for configurable non-Hermitian zero-index networks. For an $\Re(\mu_{\text{eff}})\approx 0$2-port network, the scattering matrix can be written as

$\Re(\mu_{\text{eff}})\approx 0$3

a rank-1 perturbation of a diagonal matrix, and the maximum achievable scattering-EP order is $\Re(\mu_{\text{eff}})\approx 0$4. More precisely, if $\Re(\mu_{\text{eff}})\approx 0$5 channels are effective MNZ channels and $\Re(\mu_{\text{eff}})\approx 0$6 are ENZ channels, then the maximum order is

$\Re(\mu_{\text{eff}})\approx 0$7

Choosing $\Re(\mu_{\text{eff}})\approx 0$8 gives $\Re(\mu_{\text{eff}})\approx 0$9. All-ENZ networks support no EP and instead exhibit only a diabolic point, whereas all-MNZ networks are limited to order $\Gamma$0 because an in-phase dark state remains fixed (Xu et al., 5 Jun 2026).

These results matter for magneto-optical metawaveguides even though the original papers are not gyrotropic. They show that zero-index multiport architectures naturally compress the scattering problem into a low-dimensional algebraic form while retaining tunable open-system degeneracies. In the absorbing-EP setting, the measurable reflectance satisfies

$\Gamma$1

and at an $\Gamma$2-th order absorbing EP one obtains

$\Gamma$3

Conventional coherent perfect absorption gives $\Gamma$4; a second-order absorbing EP already outperforms it, and higher-order EPs further enhance the power-law response (Xu et al., 5 Jun 2026). The zero-index network literature therefore provides a rigorous scattering-theory backbone for future gyrotropic metawaveguides, while remaining explicitly reciprocal and non-magneto-optical in its present form.

A separate zero-index network development addresses deterministic routing rather than EP singularities. In a fully connected non-Hermitian zero-index network of nodes and channels, a target reciprocal scattering matrix is mapped algebraically to physical parameters through

$\Gamma$5

This inverse-design framework enables arbitrary reflectionless routing for nearly any desired symmetric scattering response, implemented in a rectangular waveguide near TE$\Gamma$6 cutoff with photonic doping (Wang et al., 26 Dec 2025). Because the theory is reciprocal and built for symmetric $\Gamma$7, it is not yet a magneto-optical design method, but it establishes an analytical route from desired scattering response to zero-index metawaveguide parameters.

5. Experimental observables and demonstrated functions

The strongest optical metrics have been obtained in the Ce:YIG metawaveguide. In a straight $\Gamma$8-long sample, forward and backward transmission diverge near $\Gamma$9, giving $1/2$0 for that device length. Using lengths from $1/2$1 to $1/2$2, the nonreciprocal loss scales linearly with length, yielding $1/2$3 at $1/2$4. In a $1/2$5-long metawaveguide-inserted microring resonator, the measured nonreciprocal phase shift peaks at

$1/2$6

near $1/2$7, and the nonreciprocal loss peaks at

$1/2$8

near $1/2$9. The inferred figure of merit is $\varepsilon_{ez}\to 0$0, based on a total propagation loss of $\varepsilon_{ez}\to 0$1. The measured phase shift is $\varepsilon_{ez}\to 0$2 larger than that of a reported Si/Ce:YIG waveguide with $\varepsilon_{ez}\to 0$3, and the largest enhancement coincides with the experimentally observed zero-index state characterized by effectively infinite spatial wavelength (Li et al., 7 Sep 2025).

Microwave experiments demonstrate the same interplay of zero index and magnetic tuning in a different regime. In the YIG-pillar DCZIM, magnetic bias toggles the medium between an unbiased EMNZ state at $\varepsilon_{ez}\to 0$4 and a biased state in which the effective index changes from $\varepsilon_{ez}\to 0$5 to $\varepsilon_{ez}\to 0$6 and the impedance changes from $\varepsilon_{ez}\to 0$7 to $\varepsilon_{ez}\to 0$8. An S-shaped supercoupler with $\varepsilon_{ez}\to 0$9 YIG pillars and two sharp $\det(\boldsymbol{\mu}_e)\to 0$0 bends exhibits tunable supercoupling with a low intrinsic loss of $\det(\boldsymbol{\mu}_e)\to 0$1 and an extinction ratio of up to $\det(\boldsymbol{\mu}_e)\to 0$2 at $\det(\boldsymbol{\mu}_e)\to 0$3 (Yang et al., 2022). The experiment shows that a magnetically biased zero-index metawaveguide can be switched between a phase-uniform transmission state and a stop-band state without changing physical geometry.

Layered nonreciprocal heterostructures extend the metawaveguide functionality to buffer-like devices. In microwave and terahertz PEC/PMC–dielectric–YIG/InSb stacks, the zero-index operating point is a guided mode with $\det(\boldsymbol{\mu}_e)\to 0$4 and $\det(\boldsymbol{\mu}_e)\to 0$5, which enables zero-phase-shift optical buffers. Finite-element simulations report transmission efficiencies more than $\det(\boldsymbol{\mu}_e)\to 0$6, and in one tapered microwave design the phase shift in the ultrathin section can be less than $\det(\boldsymbol{\mu}_e)\to 0$7 (Zhou et al., 2022). These devices do not explicitly exploit exceptional points, but they show that zero-index nonreciprocal guided transport is compatible with strong confinement, slow propagation, and finite throughput.

6. Interpretation, misconceptions, and extensions

A recurrent misconception is that zero index in magneto-optical metawaveguides always means an isotropic double-zero medium. The literature does not support that simplification. Some platforms are EMNZ, some are ENZ with a determinant-zero gyrotropic permeability tensor, and some are defined only at the modal level by $\det(\boldsymbol{\mu}_e)\to 0$8 with finite $\det(\boldsymbol{\mu}_e)\to 0$9 (Yang et al., 2022, Wang et al., 2020, Zhou et al., 2022). This distinction matters because impedance matching, interface physics, and the meaning of near-uniform phase depend on which zero-index mechanism is actually present.

A second misconception is that non-Hermiticity is synonymous with optical gain. In the zero-index EP literature, purely lossy systems can host absorbing exceptional points of second, third, and arbitrary higher order, and these points can outperform conventional coherent perfect absorption in perturbative sensitivity without requiring amplification (Yan et al., 14 Jan 2025, Xu et al., 5 Jun 2026). Conversely, gain can be introduced to realize lasing EPs or complex-conjugate magneto-optical metamaterials, but it is not a prerequisite for non-Hermitian behavior (Wang et al., 2020).

A third misconception is that giant nonreciprocity must be limited by the bare gyrotropic coefficient of the material. The experimentally realized Ce:YIG metawaveguide shows a different regime: a weak magneto-optical response becomes a trigger for inter-sheet branch selection in a non-Hermitian zero-index eigenspace, so the observed nonreciprocity is governed by the engineered separation between photonic branches rather than by the bare off-diagonal tensor element alone (Li et al., 7 Sep 2025). This is the central conceptual shift introduced by exceptional-point-mediated zero-index magneto-optical metawaveguides.

A plausible extension, suggested by the current division of the literature, is to merge the tensor-waveguide formalism of MO and ME media with the reciprocal scattering-theory and inverse-design frameworks developed for non-Hermitian zero-index networks. The present network theories target symmetric reciprocal $\boldsymbol{\mu}_e\to 0$0-matrices, whereas a genuinely magneto-optical metawaveguide would require nonsymmetric scattering, $\boldsymbol{\mu}_e\to 0$1, and gyrotropic constitutive tensors as primary design variables (Honda et al., 2023, Wang et al., 26 Dec 2025, Xu et al., 5 Jun 2026). This suggests a next stage in which magnetic bias becomes an additional tuning parameter for exceptional-point order, directional absorption, or nonreciprocal routing. The existing literature therefore defines the topic not as a single device class but as a convergence zone: zero-index mode engineering supplies the near-degenerate phase landscape, magneto-optics supplies directionality, and non-Hermitian spectral topology converts small directional bias into large measurable asymmetry.