Nonreciprocal Dispersive Coupling

• Nonreciprocal dispersive coupling is defined by direction-dependent interactions where frequency shifts, phase accumulations, and off-resonant processes vary with propagation direction.
• It is implemented through mechanisms such as asymmetric hopping in non-Hermitian chains, synthetic phonon modulation in waveguides, and engineered dissipative interactions in cavity and circuit QED.
• These methods enable practical applications like directional amplification, isolation in microwave and photonic circuits, and enhanced quantum sensing by co-designing coherent and dissipative pathways.

Nonreciprocal dispersive coupling denotes a class of interactions in which dispersive frequency shifts, phase accumulation, or off-resonant mode coupling depend on propagation direction or act asymmetrically on the coupled subsystems. In recent work, the term covers several technically distinct constructions: asymmetric real hopping in non-Hermitian lattices, phase-matched waveguide–resonator coupling generated by synthetic phonons, direction-selective electro-optic intermodal scattering, dissipatively engineered qubit–cavity interactions, and immittance-mediated couplings in circuit QED (Ryu, 2023, Peterson et al., 2017, Wang et al., 2023, Labarca et al., 2023). Across these settings, the common feature is that the interaction remains dispersive in the sense of being governed by frequency shifts, state-dependent phases, or off-resonant virtual processes, while reciprocity is broken either by asymmetric coherent paths, engineered dissipation, or both.

1. Definitions and formal scope

A standard scattering-theoretic definition identifies nonreciprocity with directional asymmetry, namely $S_{12} \neq S_{21}$; source and measurement points are then not interchangeable (Ryu, 2023). In tight-binding form, this appears as unequal off-diagonal couplings $t_l \neq t_r$. When the couplings are real and onsite gain/loss is absent, the non-Hermiticity arises solely from asymmetric dispersive hopping, so the model realizes nonreciprocal dispersive coupling without explicit gain or loss (Ryu, 2023).

In cavity and circuit QED, reciprocal dispersive coupling is conventionally written as

$$H_s = \omega \, a^\dagger a + \frac{\omega_q}{2}\sigma_z + \lambda \, \sigma_z a^\dagger a,$$

so that cavity and qubit frequency pulls are bidirectional. A nonreciprocal version replaces this symmetry by a dissipative jump operator of the form $a e^{i\theta\sigma_z}$, yielding a master equation in which the cavity imprints information on the qubit while the qubit does not back-act on the cavity dynamics in the same way (Xie et al., 3 Jun 2026). The experimentally studied qubit–cavity realization uses

$$\partial_t \hat{\rho} = -i \left[ \Delta_c \hat{a}^\dagger \hat{a} + \frac{\lambda}{2}\hat{\sigma}_z \hat{a}^\dagger \hat{a},\, \hat{\rho} \right] + \kappa\,\mathcal{D}[\hat{a}] \hat{\rho} + \Gamma\,\mathcal{D}\!\left[ e^{\frac{i\theta + \eta}{2}\hat{\sigma}_z}\hat{a} \right] \hat{\rho},$$

which compactly separates a reciprocal coherent dispersive term from a nonreciprocal dissipative dispersive-type term (Wang et al., 2023).

A more general circuit-theoretic formulation treats the coupler as a nonreciprocal linear environment specified by an immittance matrix $Y(\omega)$ or $Z(\omega)$. In that setting, effective dispersive exchange and decay between weakly anharmonic modes are extracted directly from the reciprocal and antisymmetric parts of the immittance. The phase of the effective coupling $J_{ij}=|J_{ij}|e^{i\theta_{ij}}$ is then controlled by the nonreciprocal part of the immittance, rather than by a direct Hamiltonian hopping term alone (Labarca et al., 2023).

2. Microscopic mechanisms

Several distinct microscopic routes generate nonreciprocal dispersive coupling. They differ in implementation, but each produces a direction-dependent dispersive response in an effective reduced model.

Mechanism	Representative realization	Principal control parameter(s)
Asymmetric real hopping $t_l \neq t_r$	Directed 1D chain or ring (Ryu, 2023)	$t_l/t_r$, boundary condition $t_l \neq t_r$0, loss $t_l \neq t_r$1
Spatiotemporal modulation of couplers	Synthetic-phonon waveguide–resonator coupling (Peterson et al., 2017)	$t_l \neq t_r$2, $t_l \neq t_r$3, $t_l \neq t_r$4, $t_l \neq t_r$5, $t_l \neq t_r$6
Traveling-wave intermodal scattering	Thin-film LiNbO$t_l \neq t_r$7 EO mode conversion (Yim et al., 25 Feb 2026)	$t_l \neq t_r$8, $t_l \neq t_r$9, $$H_s = \omega \, a^\dagger a + \frac{\omega_q}{2}\sigma_z + \lambda \, \sigma_z a^\dagger a,$$0
Engineered nonlocal loss	Two-resonator TCMT model (Shen et al., 28 Sep 2025)	$$H_s = \omega \, a^\dagger a + \frac{\omega_q}{2}\sigma_z + \lambda \, \sigma_z a^\dagger a,$$1, $$H_s = \omega \, a^\dagger a + \frac{\omega_q}{2}\sigma_z + \lambda \, \sigma_z a^\dagger a,$$2, detuning
Virtual off-resonant mediation	Cavity–magnon nonlinear dispersive coupling (Zhang et al., 28 Apr 2026)	$$H_s = \omega \, a^\dagger a + \frac{\omega_q}{2}\sigma_z + \lambda \, \sigma_z a^\dagger a,$$3, $$H_s = \omega \, a^\dagger a + \frac{\omega_q}{2}\sigma_z + \lambda \, \sigma_z a^\dagger a,$$4, $$H_s = \omega \, a^\dagger a + \frac{\omega_q}{2}\sigma_z + \lambda \, \sigma_z a^\dagger a,$$5

In the non-Hermitian chain, the essential mechanism is purely real asymmetric hopping. The model with $$H_s = \omega \, a^\dagger a + \frac{\omega_q}{2}\sigma_z + \lambda \, \sigma_z a^\dagger a,$$6, $$H_s = \omega \, a^\dagger a + \frac{\omega_q}{2}\sigma_z + \lambda \, \sigma_z a^\dagger a,$$7 is unidirectional, and the authors explicitly interpret it as a master–slave structure with an “inherent source”; they further note that nonreciprocal models without gain and loss can be transformed to reciprocal models with gain and loss by similarity transformations (Ryu, 2023).

Synthetic-phonon implementations engineer nonreciprocal coupling by modulating the coupling rate at $$H_s = \omega \, a^\dagger a + \frac{\omega_q}{2}\sigma_z + \lambda \, \sigma_z a^\dagger a,$$8 spatially separated sites,

$$H_s = \omega \, a^\dagger a + \frac{\omega_q}{2}\sigma_z + \lambda \, \sigma_z a^\dagger a,$$9

so that coupling to a dark resonator state becomes phase matched in only one direction (Peterson et al., 2017). In an integrated lithium-niobate realization, a slow-wave radiofrequency transmission line with effective RF index $a e^{i\theta\sigma_z}$0 supplies the required momentum, and the interaction length is chosen so that the sinc momentum spectrum peaks for one propagation direction and approaches a null for the opposite one (Yim et al., 25 Feb 2026).

A different route is explicitly dissipative. In nonlocal loss engineering, a reciprocal coherent coupling $a e^{i\theta\sigma_z}$1 is supplemented by a state-dependent loss term that modifies only one off-diagonal element of the effective Hamiltonian,

$a e^{i\theta\sigma_z}$2

so the effective coupling acquires both dispersive and dissipative asymmetry (Shen et al., 28 Sep 2025). In cavity–magnon systems, by contrast, the nonreciprocal dispersive element is created through adiabatic elimination of a far-detuned signal cavity, which generates both an effective Kerr nonlinearity $a e^{i\theta\sigma_z}$3 and a nonlinear dispersive pump–magnon coupling $a e^{i\theta\sigma_z}$4 (Zhang et al., 28 Apr 2026).

3. Spectral structure and dynamical signatures

The spectral consequences of nonreciprocal dispersive coupling depend strongly on boundary conditions and on whether nonreciprocity enters eigenvalues, eigenvectors, or both. For the unidirectional chain with $a e^{i\theta\sigma_z}$5, periodic boundary conditions yield complex eigenenergies with orthogonal eigenstates, whereas open boundary conditions yield real eigenenergies with non-orthogonal eigenstates (Ryu, 2023). The two cases support distinct amplification mechanisms: in the ring,

$a e^{i\theta\sigma_z}$6

so amplification is exponential and eigenvalue-driven; in the open chain, the norm grows algebraically although all eigenenergies are real, and for $a e^{i\theta\sigma_z}$7 with an initial state at site 10 the amplitude obeys $a e^{i\theta\sigma_z}$8 (Ryu, 2023). This separation between amplification from complex eigenenergies and amplification from non-orthogonal eigenstates is a central dynamical signature of asymmetric dispersive hopping.

In waveguide-resonator systems with synthetic phonons, the spectral signature is a directional sideband response at $a e^{i\theta\sigma_z}$9, because the products $$\partial_t \hat{\rho} = -i \left[ \Delta_c \hat{a}^\dagger \hat{a} + \frac{\lambda}{2}\hat{\sigma}_z \hat{a}^\dagger \hat{a},\, \hat{\rho} \right] + \kappa\,\mathcal{D}[\hat{a}] \hat{\rho} + \Gamma\,\mathcal{D}\!\left[ e^{\frac{i\theta + \eta}{2}\hat{\sigma}_z}\hat{a} \right] \hat{\rho},$$0 and $$\partial_t \hat{\rho} = -i \left[ \Delta_c \hat{a}^\dagger \hat{a} + \frac{\lambda}{2}\hat{\sigma}_z \hat{a}^\dagger \hat{a},\, \hat{\rho} \right] + \kappa\,\mathcal{D}[\hat{a}] \hat{\rho} + \Gamma\,\mathcal{D}\!\left[ e^{\frac{i\theta + \eta}{2}\hat{\sigma}_z}\hat{a} \right] \hat{\rho},$$1 entering $$\partial_t \hat{\rho} = -i \left[ \Delta_c \hat{a}^\dagger \hat{a} + \frac{\lambda}{2}\hat{\sigma}_z \hat{a}^\dagger \hat{a},\, \hat{\rho} \right] + \kappa\,\mathcal{D}[\hat{a}] \hat{\rho} + \Gamma\,\mathcal{D}\!\left[ e^{\frac{i\theta + \eta}{2}\hat{\sigma}_z}\hat{a} \right] \hat{\rho},$$2 and $$\partial_t \hat{\rho} = -i \left[ \Delta_c \hat{a}^\dagger \hat{a} + \frac{\lambda}{2}\hat{\sigma}_z \hat{a}^\dagger \hat{a},\, \hat{\rho} \right] + \kappa\,\mathcal{D}[\hat{a}] \hat{\rho} + \Gamma\,\mathcal{D}\!\left[ e^{\frac{i\theta + \eta}{2}\hat{\sigma}_z}\hat{a} \right] \hat{\rho},$$3 are generally unequal (Peterson et al., 2017). This enables isolation, gyration, and higher-order nonreciprocal filters with non-Lorentzian transfer functions. The same phase-matching logic appears in electro-optic intermodal scattering: the RF wavevector satisfies the intermodal condition only when optical and RF waves counter-propagate, while co-propagation is strongly phase mismatched. Experimentally, this produces a directional $$\partial_t \hat{\rho} = -i \left[ \Delta_c \hat{a}^\dagger \hat{a} + \frac{\lambda}{2}\hat{\sigma}_z \hat{a}^\dagger \hat{a},\, \hat{\rho} \right] + \kappa\,\mathcal{D}[\hat{a}] \hat{\rho} + \Gamma\,\mathcal{D}\!\left[ e^{\frac{i\theta + \eta}{2}\hat{\sigma}_z}\hat{a} \right] \hat{\rho},$$4 dB non-reciprocal scattering contrast on thin-film lithium niobate (Yim et al., 25 Feb 2026).

Loss-engineered systems exhibit a complementary scattering signature. At resonance, when $$\partial_t \hat{\rho} = -i \left[ \Delta_c \hat{a}^\dagger \hat{a} + \frac{\lambda}{2}\hat{\sigma}_z \hat{a}^\dagger \hat{a},\, \hat{\rho} \right] + \kappa\,\mathcal{D}[\hat{a}] \hat{\rho} + \Gamma\,\mathcal{D}\!\left[ e^{\frac{i\theta + \eta}{2}\hat{\sigma}_z}\hat{a} \right] \hat{\rho},$$5, the forward coupling can vanish,

$$\partial_t \hat{\rho} = -i \left[ \Delta_c \hat{a}^\dagger \hat{a} + \frac{\lambda}{2}\hat{\sigma}_z \hat{a}^\dagger \hat{a},\, \hat{\rho} \right] + \kappa\,\mathcal{D}[\hat{a}] \hat{\rho} + \Gamma\,\mathcal{D}\!\left[ e^{\frac{i\theta + \eta}{2}\hat{\sigma}_z}\hat{a} \right] \hat{\rho},$$6

yielding perfect isolation in one direction. The same framework yields a relative 3dB bandwidth of $$\partial_t \hat{\rho} = -i \left[ \Delta_c \hat{a}^\dagger \hat{a} + \frac{\lambda}{2}\hat{\sigma}_z \hat{a}^\dagger \hat{a},\, \hat{\rho} \right] + \kappa\,\mathcal{D}[\hat{a}] \hat{\rho} + \Gamma\,\mathcal{D}\!\left[ e^{\frac{i\theta + \eta}{2}\hat{\sigma}_z}\hat{a} \right] \hat{\rho},$$7 and high isolation ($$\partial_t \hat{\rho} = -i \left[ \Delta_c \hat{a}^\dagger \hat{a} + \frac{\lambda}{2}\hat{\sigma}_z \hat{a}^\dagger \hat{a},\, \hat{\rho} \right] + \kappa\,\mathcal{D}[\hat{a}] \hat{\rho} + \Gamma\,\mathcal{D}\!\left[ e^{\frac{i\theta + \eta}{2}\hat{\sigma}_z}\hat{a} \right] \hat{\rho},$$8 dB) over a wide power range in the example given (Shen et al., 28 Sep 2025). Here the nonreciprocal dispersive effect appears through a complex off-diagonal term generated by dissipation, rather than through a purely coherent synthetic gauge field.

4. Circuit-QED realizations and quantum sensing

The most direct quantum realization is the transmon–cavity experiment on a ferrite-loaded waveguide junction. There the measured qubit–cavity dynamics show asymmetric frequency pulls and photon shot-noise dephasing, and the degree of nonreciprocity is tuned in situ by the magnetic-field bias of the ferrite component (Wang et al., 2023). In the effective model, the cavity conditioned on qubit state $$\partial_t \hat{\rho} = -i \left[ \Delta_c \hat{a}^\dagger \hat{a} + \frac{\lambda}{2}\hat{\sigma}_z \hat{a}^\dagger \hat{a},\, \hat{\rho} \right] + \kappa\,\mathcal{D}[\hat{a}] \hat{\rho} + \Gamma\,\mathcal{D}\!\left[ e^{\frac{i\theta + \eta}{2}\hat{\sigma}_z}\hat{a} \right] \hat{\rho},$$9 has complex frequency

$Y(\omega)$0

so the coherent cavity pull is set by $Y(\omega)$1, while qubit-state-dependent damping is set by $Y(\omega)$2 (Wang et al., 2023). The qubit coherence obeys

$Y(\omega)$3

which makes explicit that photon loss events produce qubit phase kicks and dephasing that are not reciprocally encoded in the cavity dynamics (Wang et al., 2023).

The circuit-theory generalization replaces the specific ferrite device by an arbitrary nonreciprocal linear environment described by $Y(\omega)$4 or $Y(\omega)$5. The resulting dispersive Lindblad master equation,

$Y(\omega)$6

contains dressed qubit frequencies, anharmonicities, exchange couplings $Y(\omega)$7, cross-Kerr terms $Y(\omega)$8, drive amplitudes, and collective decay rates $Y(\omega)$9, all written in terms of the immittance of the coupler (Labarca et al., 2023). In particular, the phase of the effective hopping is fixed by the antisymmetric immittance contribution, so nonreciprocal dispersive coupling becomes a circuit-synthesis problem rather than a device-specific perturbation (Labarca et al., 2023).

A sensing application follows from the dissipative one-way qubit–cavity coupling. For cavity photon-number estimation, nonreciprocal dispersive coupling yields higher precision than reciprocal dispersive coupling, and the advantage becomes more pronounced as photon number increases; for direct measurement of the single-photon driving strength, no superiority is found; but when the driving-strength information is first converted into cavity photon number, the nonreciprocal scheme again outperforms the reciprocal one, with the advantage becoming increasingly significant at larger driving strength (Xie et al., 3 Jun 2026). In that analysis, the nonreciprocal master equation is built from the jump operator $Z(\omega)$0, so the sensing gain is directly tied to the same one-way dispersive structure identified experimentally (Xie et al., 3 Jun 2026).

5. Nonlinear, photonic, and material platforms

Nonreciprocal dispersive coupling is not restricted to qubit–cavity systems. In a hybrid system of two microwave cavities and one YIG sphere, a far-detuned signal cavity is adiabatically eliminated, producing the effective Hamiltonian

$Z(\omega)$1

with

$Z(\omega)$2

Here the sign of the dispersive Kerr term $Z(\omega)$3 is controlled by the sign of $Z(\omega)$4, and the unconventional magnon blockade condition

$Z(\omega)$5

is satisfied only for one sign of $Z(\omega)$6 at fixed $Z(\omega)$7. Numerically, one sign of $Z(\omega)$8 produces no deep dip in $Z(\omega)$9, while the opposite sign yields a deep antibunching dip with $J_{ij}=|J_{ij}|e^{i\theta_{ij}}$0 at the same $J_{ij}=|J_{ij}|e^{i\theta_{ij}}$1 (Zhang et al., 28 Apr 2026). In this usage, nonreciprocity is realized in parameter space through a sign-tunable dispersive nonlinearity rather than through spatially asymmetric ports.

A distinct photonic realization uses electromagnetically induced transparency in a hot rubidium vapor with a moiré photonic lattice. In the forward, co-propagating configuration, Doppler shifts cancel in the two-photon resonance, the real part of the susceptibility is strongly modulated by the coupling field, and the probe can undergo transverse localization. In the backward configuration, the two-photon detuning becomes $J_{ij}=|J_{ij}|e^{i\theta_{ij}}$2, EIT is destroyed, the refractive-index modulation is nearly uniform, and the probe beam broadens and is strongly attenuated (Liang et al., 2024). The measured localization factor is approximately $J_{ij}=|J_{ij}|e^{i\theta_{ij}}$3, and the isolation ratio increases from about $J_{ij}=|J_{ij}|e^{i\theta_{ij}}$4 dB to $J_{ij}=|J_{ij}|e^{i\theta_{ij}}$5 dB as the probe power increases from $J_{ij}=|J_{ij}|e^{i\theta_{ij}}$6 to $J_{ij}=|J_{ij}|e^{i\theta_{ij}}$7 (Liang et al., 2024). This platform emphasizes that nonreciprocal dispersive coupling can act on spatial mode structure, not only on discrete cavity frequencies.

A further microwave-circuit manifestation arises in passive Hall-based devices. There, capacitive coupling to a two-dimensional Hall material produces a multiport admittance with complex frequency-dependent $J_{ij}=|J_{ij}|e^{i\theta_{ij}}$8 and $J_{ij}=|J_{ij}|e^{i\theta_{ij}}$9, and in the quantum Hall limit the effective circuit becomes an ideal circulator plus dispersive stubs. Including the full AC response reveals counterpropagating features that could be exploited to dynamically switch the non-reciprocity of the device (Bosco, 31 Aug 2025). This indicates that nonreciprocal dispersive coupling can be encoded directly in the intrinsic conductivity tensor of a material platform.

6. Conceptual distinctions and recurring misconceptions

A persistent ambiguity concerns the distinction between strict electromagnetic nonreciprocity and nonsymmetric effective coupling. In detuned photonic directional couplers, the overlap formula

$t_l \neq t_r$0

generically yields $t_l \neq t_r$1, so the reduced coupled-mode Hamiltonian is nonsymmetric. However, the underlying Maxwell problem remains reciprocal; what is broken is the symmetry of the effective tight-binding model, not Lorentz reciprocity in the full scattering sense (Vicencio et al., 2024). This distinction is important because several works on nonreciprocal dispersive coupling operate at the level of projected or effective mode dynamics.

A second misconception is that nonreciprocity must imply asymmetric transmission. A dual-resonator microwave system coupled only through a shared transmission line provides a counterexample: the transmission amplitudes remain reciprocal, $t_l \neq t_r$2, while reflection and absorption are strongly asymmetric, with $t_l \neq t_r$3 and $t_l \neq t_r$4. The same traveling-wave-induced indirect coupling produces an EIT-like peak at $t_l \neq t_r$5 mm and near-zero reflection with almost perfect absorption in one direction near $t_l \neq t_r$6 mm, with the reflectionless point tied to a non-Hermitian scattering-matrix condition (Kim et al., 2023). Nonreciprocal dispersive coupling can therefore manifest at the level of frequency pulls, linewidths, reflection zeros, or absorption asymmetry even when forward and backward transmission remain equal.

A third distinction concerns coherent versus dissipative origin. In the unidirectional non-Hermitian chain, asymmetric real hopping alone produces amplification without external gain; in nonlocal loss engineering, the asymmetry originates in dissipation but the effective off-diagonal term has both dispersive and dissipative components (Ryu, 2023, Shen et al., 28 Sep 2025). Taken together, these results suggest a common design pattern: nonreciprocal dispersive coupling emerges whenever off-resonant phase accumulation, virtual-mode elimination, or state-dependent loss becomes direction selective. A plausible implication is that future architectures will increasingly co-design coherent paths and dissipative channels, rather than treating dispersion and loss as separate resources.