---
title: Phase-Compensated Non-Reciprocal Hopping
url: https://www.emergentmind.com/topics/phase-compensated-non-reciprocal-hopping
type: topic
---

# Phase-Compensated Non-Reciprocal Hopping

Phase-compensated non-reciprocal hopping denotes a class of engineered couplings in which directional transport is controlled primarily through phase structure rather than through unequal transmission magnitude alone. In the most direct realizations, complex hopping phases are arranged so that forward and backward paths interfere differently, often with nearly symmetric amplitudes but unequal transmitted phases, or with destructive cancellation in one direction and reinforcement in the other. The concept appears in several distinct but mathematically related settings: loop-based bosonic networks with synthetic flux [1612.01836], long-range phase-engineered microwave lattices implementing Gebhard–Ruckenstein hopping [1804.02936], spatiotemporally modulated and switched transmission-line devices that realize direction-dependent phase shifts [1803.06690; 2210.13243; 2411.18734], and non-Hermitian lattice models where non-reciprocal hopping can be compensated, reshaped, or effectively absorbed by similarity transformations, disorder, or pairing [2407.01372; 2505.13057; 2510.24851].

## 1. Conceptual definition and formal structure

At its most general, non-reciprocal hopping is a directional coupling between modes or sites for which the effective transfer from \(i\to j\) differs from that of \(j\to i\). In the phase-engineered realizations emphasized here, the asymmetry is encoded in complex phases of the couplings, synthetic gauge fluxes around loops, or direction-dependent Floquet phases. A compact representation is
\[
J_{ij}=|J_{ij}|e^{i\phi_{ij}},
\]
with non-reciprocity arising when the phase structure cannot be removed without changing boundary conditions or auxiliary channels.

A useful operational distinction emerges from the cited works. In the diamond network, non-reciprocity is engineered almost entirely by how the phases of the strong edge couplings are arranged around the loop, while the weak diagonal couplings provide the parametric mixing needed to break reciprocity but do not themselves require phase control [1612.01836]. In spatiotemporally modulated systems, the only contributor to nonreciprocity can be the nonreciprocal phase shift, namely equal transmitted amplitudes but different transmitted phases in opposite directions [2411.18734]. In time-modulated microwave phase shifters, the effective link can be written as
\[
t_{12}=|t|e^{-i\phi},\qquad t_{21}=|t|e^{+i\phi},
\]
so that the link itself behaves as a direction-dependent complex hopping element [2210.13243].

This suggests an *Editor's term*, “phase compensation,” for the design strategy in which auxiliary phases are chosen so that unwanted coherent contributions cancel while the desired directional pathway remains. In the diamond device, this means arranging the loop phase to produce destructive interference in one direction [1612.01836]. In switched transmission lines and temporal-loop phase shifters, it means canceling unwanted harmonics or reciprocal phase contributions so that the remaining effective link carries a controlled non-reciprocal phase [1803.06690; 2210.13243]. In non-Hermitian lattices, it can mean tuning additional terms so that the effective off-diagonal coupling is altered or the left and right Lyapunov exponents become equal [2407.01372; 2505.13057].

## 2. Loop-phase engineering in the four-mode diamond network

A concrete canonical example is the four-node diamond configuration for non-reciprocal transmission [1612.01836]. The system contains four electromagnetic modes, with modes 1 and 3 at frequency \(\omega\), modes 2 and 4 at frequency \(\Omega\), strong edge hoppings \(g,h,f,k\) between different frequencies, and weak parametric diagonal couplings \(\gamma\) between equal-frequency nodes. Under the rotating-wave approximation, the interaction Hamiltonian is
\[
\begin{aligned}
\mathbb{H}_i &=\hbar(g\hat{a}_1\hat{a}_2^\dagger+g^{*}\hat{a}_2\hat{a}_1^\dagger)
+\hbar(f\hat{a}_3\hat{a}_4^\dagger+f^{*}\hat{a}_4\hat{a}_3^\dagger)\\
&\quad+\hbar(h\hat{a}_2\hat{a}_3^\dagger+h^{*}\hat{a}_3\hat{a}_2^\dagger)
+\hbar(k\hat{a}_4\hat{a}_1^\dagger+k^{*}\hat{a}_1\hat{a}_4^\dagger)\\
&\quad+\hbar\gamma(\hat{a}_1\hat{a}_3+\hat{a}_1^\dagger\hat{a}_3^\dagger)
+\hbar\gamma(\hat{a}_2\hat{a}_4+\hat{a}_2^\dagger\hat{a}_4^\dagger).
\end{aligned}
\]
The edge couplings are complex,
\[
g=|g|e^{i\phi_g},\; h=|h|e^{i\phi_h},\; f=|f|e^{i\phi_f},\; k=|k|e^{i\phi_k},
\]
and the decisive control variable is the round-trip loop phase
\[
\theta=\phi_g+\phi_h+\phi_f+\phi_k.
\]

The relevant transport from port 1 to port 3 has two coherent routes: a direct parametric path \(1\leftrightarrow 3\) via \(\gamma\), and indirect loop paths \(1\to2\to3\) and \(1\to4\to3\). The paper defines the non-reciprocity ratio
\[
R(\omega)=\frac{1}{2}\left[\left|\frac{S_{31}(\omega)}{S_{13}(\omega)}\right|^2+
\left|\frac{S_{13}(\omega)}{S_{31}(\omega)}\right|^2\right],
\]
and shows numerically that \(R(\omega)\) is a function of \(\theta\) only, is symmetric under \(\theta\to-\theta\), and is maximized at \(\theta=\pm\pi\) [1612.01836]. A convenient choice is
\[
\phi_g=\phi_h=\phi_f=\phi_k=\frac{\pi}{4},
\]
which gives \(\theta=\pi\).

The phase dependence has a direct interference interpretation. The paper describes a situation akin to
\[
t_{\text{forward}}\propto A_{\text{direct}}+A_{\text{loop}},\qquad
t_{\text{backward}}\propto A_{\text{direct}}+A_{\text{loop}}^{*},
\]
with \(A_{\text{loop}}\propto e^{i\theta}\). At \(\theta=\pi\), the loop contribution can cancel the direct term in one propagation direction while not canceling it in the other. The diagonal parametric links are essential, but their phase is stated to be irrelevant and may be dropped; all non-trivial gauge freedom resides on the four edge hoppings [1612.01836].

The network is described by an \(8\times8\) Langevin matrix and the scattering matrix
\[
[S(\omega)]=[I]-\sqrt{[\Gamma]}\left(i\omega[I]+[M]\right)^{-1}\sqrt{[\Gamma]}.
\]
Non-reciprocity appears as \(|S_{31}(\omega)|\neq|S_{13}(\omega)|\). With optimized intrinsic parameters, the reported non-reciprocity is \(R_{\max}\approx 12.39\,\text{dB}\) at a frequency slightly blue-detuned from \(\omega\) by about \(53\ \text{kHz}\). In the extrinsic pumped configuration, with \(\bar a_2=2.844\) and \(\bar a_4=0.4121\), the peak non-reciprocity exceeds \(130\,\text{dB}\). For directional amplification, the choice \(\bar a_2=0\), \(\bar a_4=10^2\) yields forward and backward gains of approximately \(\pm20\,\text{dB}\), isolation of order \(40\,\text{dB}\), and a \(-3\,\text{dB}\) bandwidth of approximately \(\pm1\,\text{MHz}\) about \(\omega\) [1612.01836].

## 3. Phase-only and phase-dominant non-reciprocity in modulated transmission systems

A second major lineage realizes phase-compensated non-reciprocal hopping through spatiotemporal modulation rather than through internal mode loops. In switched transmission lines, time-periodic switching of line segments produces non-reciprocity without magnetic bias, and the effective forward and reverse transmissions can be written as
\[
S_{21}(\omega)=|S_{21}(\omega)|e^{i\phi_{\text{fwd}}(\omega)},\qquad
S_{12}(\omega)=|S_{12}(\omega)|e^{i\phi_{\text{rev}}(\omega)},
\]
with \(|S_{21}|\approx|S_{12}|\) but \(\phi_{\text{fwd}}\neq\phi_{\text{rev}}\) in non-reciprocal phase-shifter operation [1803.06690]. The time-periodic scattering relation is
\[
b_p(\omega)=\sum_q\sum_{n=-\infty}^{\infty}S_{pq}^{(n)}(\omega)\,a_q(\omega+n\omega_m),
\]
and the design target is often a frequency-flat differential phase, not a large amplitude asymmetry [1803.06690].

The ideal two-port gyrator in this framework has
\[
\mathbf{S}_{\text{gyr}}=
\begin{bmatrix}
0 & 1\\
-1 & 0
\end{bmatrix},
\]
which is a pure \(\pi\) non-reciprocal phase shift with unit transmission magnitude in both directions [1803.06690]. The corresponding effective hopping interpretation is explicit:
\[
t_{ij}(\omega)=|S_{ji}(\omega)|e^{i\phi_{ij}(\omega)},\qquad
\Delta\phi_{ij}(\omega)=\phi_{ij}(\omega)-\phi_{ji}(\omega).
\]
The paper presents switched transmission lines as implementing links whose magnitude of transmission can be nearly symmetric, but whose phase is direction-dependent and can be engineered to be broadband and approximately frequency-independent [1803.06690].

A closely related microwave realization is the magnet-free nonreciprocal phase shifter based on two temporal loops [2210.13243]. A time-modulated line section uses
\[
C_{\text{eq}}(t)=C_{\text{av}}+\delta\cos(\Omega t+\phi),
\]
which couples Floquet harmonics. For up-conversion, the converted wave picks up phase \(+\phi\); for down-conversion, it picks up phase \(-\phi\) [2210.13243]. By embedding four such sections into two temporal loops with modulation phases \(0,\pi,\phi,\phi+\pi\), and by arranging constructive and destructive interference of harmonics, the device yields
\[
\arg S_{21}=-\phi,\qquad \arg S_{12}=+\phi,
\]
so that the differential phase is
\[
\Delta\phi=\phi_{\text{backward}}-\phi_{\text{forward}}=2\phi.
\]
The reported insertion loss is less than \(1.2\,\text{dB}\), return loss exceeds \(28.1\,\text{dB}\) at port 1 and \(34.5\,\text{dB}\) at port 2, undesired harmonics are about \(31\,\text{dB}\) below the main harmonic, and the differential phase stays within about \(4^\circ\) of the design value over \(3\)–\(4\ \text{GHz}\), a \(28\%\) fractional bandwidth [2210.13243].

The mechanical spatiotemporal-modulation study takes the phase-only viewpoint even further [2411.18734]. It defines the reciprocity bias
\[
R=\lim_{T\to\infty}\frac{\int_0^T\big(x_n^F(\tau)-x_1^B(\tau)\big)^2\,d\tau}
{\int_0^T\big(x_n^F(\tau)\big)^2+\big(x_1^B(\tau)\big)^2\,d\tau},
\]
and constructs regimes where transmitted amplitudes or energies are equal but phases differ. The grounding stiffness is modulated as
\[
k_p(t)=k_{g,\mathrm{DC}}+k_{g,\mathrm{AC}}\cos(\omega_m t-\phi_p),\qquad \phi_p=(p-1)\phi.
\]
For weak modulation and one-sideband truncation, the response is decomposed into an envelope \(E_p(\tau)\) and carrier \(C_p(\tau)\), and nonreciprocal phase shifts are engineered by imposing equality of envelope magnitudes while allowing different envelope phases [2411.18734]. A particularly strong result is obtained for \(\phi=\pi\) and even \(n\): for any modulation depth \(K_m\) and any even number of units, \(|y_{n;q}^F|=|y_{1;q}^B|\) for all \(q\), while the phases differ so that
\[
E_n^F(\tau)=E_1^B(\tau\pm T_E/2).
\]
This is a pure envelope-level nonreciprocal phase shift with amplitude reciprocity preserved [2411.18734].

## 4. Phase-pattern synthesis in lattices and networks

Long-range microwave lattices provide a third archetype, in which the desired phase pattern is not merely a correction but the defining structure of the hopping Hamiltonian. In the superconducting implementation of Gebhard–Ruckenstein hopping, the effective cavity Hamiltonian is
\[
H=\sum_m \hbar\omega_m a_m^\dagger a_m+\sum_{m\ne n}\hbar J_{mn}a_m^\dagger a_n,
\]
with
\[
J_{mn}=\overline g_{m,n}e^{-i\theta_{m,n}}.
\]
The target Gebhard–Ruckenstein matrix is
\[
\eta_{m,n}=
\begin{cases}
\dfrac{i\pi\eta_0(-1)^{n-m}}{N\sin[\pi(n-m)/N]}, & n\ne m,\\[6pt]
0, & n=m,
\end{cases}
\]
so the physical modulation phases are chosen as
\[
\overline g_{m,n}=\frac{\pi\eta_0(-1)^{n-m}}{N\sin[\pi(n-m)/N]},\qquad
\theta_{m,n}=-\frac{\pi}{2},
\]
which makes the effective hopping purely imaginary and exactly equal to the Gebhard–Ruckenstein pattern [1804.02936].

This pattern is distinguished by an exactly linear band,
\[
E_\nu=\hbar\eta_0 k_\nu,
\]
with constant group velocity \(v_g=\eta_0\), so a localized excitation propagates chirally around the ring without group-velocity dispersion [1804.02936]. Once transmission lines are attached, the scattering matrix becomes
\[
S_{p,m}=\delta_{p,m}-\sqrt{\kappa_p\kappa_m}\,[\mathcal G^{-1}]_{m,p},
\]
where the internal matrix \(\mathcal G\) contains the complex hopping terms. For specific geometries, the reported optimal couplings include \(N=3\) with \(\kappa=2g\), \(N=5\) with ports at \(1,3,4\) and \(\kappa=4g\), and \(N=6\) with ports at \(1,3,5\) and \(\kappa\simeq4.328g\). For large \(N\), the forward transmission is near unity for \(-\pi g<\Delta\omega<\pi g\), giving an asymptotic bandwidth of \(2\pi g\) [1804.02936].

The same paper explicitly studies phase errors. For the \(N=6\) system, modifying two couplings as
\[
\overline g_{13}\to e^{i\theta_1}\overline g_{13},\qquad
\overline g_{35}\to e^{i\theta_2}\overline g_{35},
\]
still yields \(|S_{13}|^2\gtrsim0.93\) over a sizeable region corresponding to roughly \(30\%\) phase deviations in radians [1804.02936]. This does not imply insensitivity to phase; rather, it shows that moderate phase-compensation errors perturb but do not immediately destroy the targeted chiral transport.

A broader systems-level synthesis appears in switched transmission lines, where multi-section or multi-path architectures are used so that reciprocal phase contributions cancel while the non-reciprocal phase contributions add [1803.06690]. This suggests that in network design, compensation is often distributed: one part of the structure sets the synthetic flux, while another cancels reciprocal dispersion or sideband leakage.

## 5. Compensation by similarity transforms, disorder, and pairing

Not all phase-compensated non-reciprocal hopping is implemented by explicit pump-phase control. In one-dimensional non-reciprocal quasicrystals, the compensation can be formulated in terms of similarity transformations and asymmetric Lyapunov exponents [2407.01372]. The generic Hamiltonian is
\[
H=\sum_{j=1}^{L-1}\big(t_{l,j}c_j^\dagger c_{j+1}+t_{r,j}c_{j+1}^\dagger c_j\big)+\sum_{j=1}^{L}V_j c_j^\dagger c_j.
\]
For the non-reciprocal Aubry–André model,
\[
t_{l,j}=e^{g},\qquad t_{r,j}=e^{-g},\qquad V_j=2\lambda\cos(2\pi\alpha j+\theta).
\]
Under open boundary conditions, the non-reciprocal hopping can be removed by the similarity transformation
\[
S=\mathrm{diag}(e^{-g},e^{-2g},\dots,e^{-Lg}),\qquad H'=SHS^{-1},
\]
which maps the problem to a Hermitian Aubry–André chain. At the wavefunction level,
\[
\psi_j=e^{gj}\psi_j'.
\]
The left and right Lyapunov exponents are then
\[
\gamma_l(E)=\max\{\ln|\lambda|-g,0\},\qquad
\gamma_r(E)=\max\{\ln|\lambda|+g,0\},
\]
and the localization transition occurs at \(\lambda_c=e^{|g|}\) [2407.01372]. Here compensation is exact in the bulk, but boundary conditions retain the skin effect.

In more complicated models, such simple compensation fails globally but can still occur at special points where the left and right Lyapunov exponents become equal. In the non-reciprocal off-diagonal Aubry–André model, the difference \(\Delta_\gamma=\gamma_l-\gamma_r\) can vanish at a critical \(g_c\), yielding symmetric localization despite non-reciprocal hopping [2407.01372]. In the non-reciprocal mosaic model, \(\Delta_\gamma=0\) occurs at \(W=0\) and \(W=2\), again indicating effective compensation of directional asymmetry at the level of localization lengths [2407.01372].

A different form of compensation appears in the non-reciprocal Kitaev chain with engineered dissipation and pairing [2510.24851]. The effective directional couplings are
\[
w_{R/L}=w-i\frac{\Gamma_h}{2}e^{\mp i\theta_h},\qquad
\Delta_{R/L}=\Delta\mp i\frac{\Gamma_p}{2}e^{i\theta_p}.
\]
With \(\Gamma_h=2w\) and \(\theta_h=\pi/2\), the weak-pairing regime shows strongly directional dynamics. Pairing then competes with that directional bias. For coherent pairing, the critical point tends to \(\Delta_c=w\) as \(N\to\infty\), and for non-reciprocal pairing with \(\Gamma_p=2\Delta\), \(\theta_p=-\pi/2\), one finds
\[
\Delta_R=2\Delta,\qquad \Delta_L=0,
\]
and an \(N\)-fold exceptional point at \(\Delta=w\) where
\[
E_n(\Delta=w)=-4iw,\qquad \forall n.
\]
In the strong-pairing regime, the steady-state particle current under periodic boundary conditions vanishes as \(\Delta\to\infty\), even though the hopping remains non-reciprocal, while a finite pairing current remains [2510.24851]. The paper interprets this as a non-trivial breakdown or reshaping of non-reciprocity by pairing.

The disorder-driven non-Hermitian lattice study offers another mechanism: non-reciprocal hopping induces an inter-orbital coupling \(c\propto \sinh\kappa\), and disorder renormalizes diagonal and off-diagonal terms through the self-energy [2505.13057]. The effective \(2\times2\) Hamiltonian at \(\Gamma\) is
\[
H_{\text{EM}}(\Gamma)=
\begin{pmatrix}
\widetilde e_s & c+\Sigma_{\text{And}}^{12}\\
c+\Sigma_{\text{And}}^{21} & \widetilde e_p
\end{pmatrix},
\]
with exceptional-point condition
\[
(\widetilde e_s-\widetilde e_p)^2=4(c+\Sigma_{\text{And}}^{12})^2.
\]
The paper does not explicitly use the phrase “phase-compensated non-reciprocal hopping,” but states that if one were to introduce additional complex phases or gain/loss that generate an opposite off-diagonal term, one could partially compensate the impact of \(c\) on the exceptional-point condition [2505.13057]. That formulation makes compensation a spectral-design problem rather than only a transport-design problem.

## 6. Spectral, topological, and dynamical consequences

Phase-compensated non-reciprocal hopping is not a single phenomenon but a recurring mechanism that reorganizes interference, spectra, and dynamical phases.

In loop interferometers such as the diamond configuration, the primary consequence is directional scattering at identical input and output frequencies, with passive non-reciprocal transmission in the intrinsic setting and directional amplification in the extrinsic setting [1612.01836]. In switched transmission lines and temporal-loop phase shifters, the central outcome is a non-reciprocal phase shifter or gyrator: equal or nearly equal transmission magnitudes, but a direction-dependent phase that can be used as a broadband circuit primitive [1803.06690; 2210.13243]. In spatiotemporally modulated mechanical chains, the consequence can be a pure envelope-level time shift between forward and backward transmitted signals with matched amplitudes [2411.18734].

In long-range lattices such as Gebhard–Ruckenstein networks, the phase pattern fixes the entire band structure. The strictly linear dispersion produces constant unidirectional group velocity over a broad frequency range, which is why the resulting circulators can exhibit wide bandwidth [1804.02936]. In non-Hermitian quasicrystals, the same general theme appears as equality or inequality of left and right Lyapunov exponents, which tracks whether directional asymmetry is effectively compensated or instead manifests as a non-Hermitian skin effect [2407.01372].

The non-Hermitian many-body and topological examples show that phase-compensated non-reciprocal hopping also intersects with exceptional points and phase transitions. In multipopulation \(O(2)\)-symmetric systems, the linear coupling matrix \(J_{ab}\) is the analog of a non-Hermitian hopping matrix, and asymmetry \(J_{ab}\neq J_{ba}\) produces chiral phases, limit-cycle saddle-node bifurcations, Hopf bifurcations, and critical exceptional points in the linearized dynamics [2507.16763]. The phase-difference sector admits topological classification by winding numbers
\[
w_{ab}=\frac{1}{2\pi}\int_0^T\dot\phi_{ab}(t)\,dt,
\]
and the paper argues that tuning non-reciprocal couplings can move the system between static, chiral, quasi-periodic, and chaotic regimes [2507.16763]. This suggests that phase compensation, in the broader sense, can be understood as steering a non-reciprocal coupling network toward a desired orbit topology rather than merely canceling a transmission pathway.

Across the cited literature, a consistent misconception is that non-reciprocity must primarily mean unequal forward and backward amplitudes. Several of these works show otherwise. Switched transmission lines and temporal-loop devices realize non-reciprocity as a phase-only or phase-dominant effect [1803.06690; 2210.13243]. The mechanical spatiotemporal system explicitly constructs states in which transmitted amplitudes are equal while transmitted phases differ [2411.18734]. Conversely, the diamond network shows that very large amplitude asymmetry can still be driven by phase design, because the amplitude contrast itself comes from phase-controlled interference rather than from a static imbalance alone [1612.01836].

## 7. Physical implementations, design rules, and limitations

The implementation platforms described in these works are diverse but share a common design logic. In superconducting microwave circuits, edge hoppings can be realized as parametric frequency-conversion couplings with drive-controlled phases, while diagonal parametric couplings are driven at \(\omega+\Omega\) and need no precise phase control [1612.01836]. The same circuit-QED ecosystem supports all-to-all long-range couplings through Josephson-ring mediators, where the drive phases directly set the effective hopping phases needed for Gebhard–Ruckenstein transport [1804.02936].

In radio-frequency and microwave engineering, switched transmission lines provide broadband, lossless and compact non-reciprocity, including non-reciprocal phase shifters, ultra-broadband gyrators and isolators, frequency-conversion isolators, and circulators, with a \(25\ \text{GHz}\) circulator demonstrated in \(45\ \text{nm}\) SOI CMOS technology [1803.06690]. The time-modulated two-loop phase shifter provides a reconfigurable, IC-compatible phase-only non-reciprocal element based on varactors, splitters, and line delays [2210.13243]. In mechanical metamaterials, the control knobs are the modulation phase increment \(\phi\), modulation frequency \(\Omega_m\), coupling stiffness \(K_c\), and modulation depth \(K_m\), which together can be tuned to satisfy amplitude-equality constraints while leaving a phase difference [2411.18734].

Several concrete design rules recur. One is to identify the loop or modulation phase that controls reciprocity and tune it to the appropriate critical value, such as \(\theta=\pi\) in the diamond network [1612.01836]. Another is to separate the resource that carries synthetic flux from the resource that provides gain, parametric mixing, or sideband conversion. The diamond device separates edge-phase control from diagonal parametric coupling [1612.01836]; the temporal-loop phase shifter separates modulation phase control from harmonic suppression by interference [2210.13243]; switched transmission lines separate non-reciprocal timing from reciprocal dispersion, which can then be compensated by multi-section design [1803.06690].

The limitations are equally consistent. The diamond network is highly sensitive to the loop phase \(\theta\), the parametric strength \(\gamma\), and the quality factors \(Q_n\); deviations from \(\theta=\pi\) quickly reduce isolation [1612.01836]. In mechanical phase-only transmission, the envelope formalism with one-sideband truncation is accurate only for short systems, weak modulation, and \(\Omega_m<\Omega_f\); when \(\Omega_m>\Omega_f\), the envelope fails to represent the displacement maxima [2411.18734]. In switched transmission lines and time-modulated phase shifters, switch resistance, parasitic capacitance, finite rise time, and line dispersion all distort the intended non-reciprocal phase [1803.06690; 2210.13243]. In non-Hermitian lattices, exact compensation by similarity transformation is special to models with uniform asymmetry; once non-reciprocity is bond-dependent or entangled with quasiperiodic modulation, compensation generally survives only as an effective or spectral notion, not as an exact local gauge removal [2407.01372].

A plausible implication is that “phase-compensated non-reciprocal hopping” is best regarded not as a single device class but as a unifying design principle. It encompasses synthetic-flux interferometers, phase-only non-reciprocal links, exact long-range phase patterns that linearize bands, and non-Hermitian compensation conditions that equalize localization or relocate exceptional points. What ties these together is the controlled use of phase to bias directionality while suppressing, redirecting, or reinterpreting unwanted reciprocal and non-reciprocal contributions.

Source: https://www.emergentmind.com/topics/phase-compensated-non-reciprocal-hopping