betaFeedback

WorksPhysics and Astronomy

Transient Synchronization and Entanglement through Shared Dissipation in Circuit QED

Abstract

We propose a superconducting circuit in which two resonators couple through DC-biased SQUIDs to a common lossy auxiliary mode. Adiabatic elimination yields collective damping and correlated noise with an overlap ξξ whose magnitude is tuned by the flux biases and whose sign follows the relative sign of the couplings. The full three-mode dynamics confirm this reduced description. For detuned resonators, we derive the exceptional-point condition, above which the mean amplitudes lock to a common frequency. For resonant modes, unequal collective decay produces transient synchronization, antiphase or in phase depending on the relative sign of the couplings, with an analytical synchronization time set by the collective decay splitting and the initial mean amplitudes. For initially separable squeezed states, the same relaxation converts local squeezing into transient entanglement, quantified by logarithmic negativity. We obtain closed-form expressions for the squeezing threshold above which entanglement appears and for the time at which it disappears. Thermal noise suppresses entanglement without changing the mean-amplitude synchronization time at fixed parameters. For realistic circuit parameters, these effects occur within a few hundred nanoseconds and are accessible through output-field measurements. A single tunable shared channel thus produces an exceptional point, transient synchronization, and transient entanglement, and provides a controllable source of correlated noise for superconducting circuits.

Introduction

In 1665, Christiaan Huygens observed antiphase synchronization between two pendulum clocks coupled through a common support [1]. Modern experiments with clocks and metronomes demonstrate how the coupling structure selects synchronized motion [2, 3]. In open quantum systems, a common reservoir provides a related mechanism: unequal damping of collective modes allows a slowly decaying contribution to dominate, establishing a fixed phase relation between oscillators [4, 5].

Reservoir engineering uses dissipation for quantum-state preparation [6, 7], including entanglement stabilization in superconducting circuits [8]. Bittner and Tyagi showed that correlated dissipative environments can select long-lived phase-synchronized motion through the symmetry of the dissipator [9]. A circuit realization must connect these effective correlations to tunable parameters and establish how private losses and thermal fluctuations affect both collective motion and quantum correlations.

A strongly damped auxiliary mode provides a shared reservoir and mediates dissipative interactions [10, 11]. Such interactions connect reservoir engineering to non-Hermitian spectral physics, including level attraction and exceptional points, where eigenvalues and eigenvectors coalesce [12, 13, 14]. Exceptional points have been observed in superconducting circuits, both in tunable resonators [15] and in a dissipative qubit [16]. For balanced losses and opposite detunings, dissipative coupling realizes anti-parity–time symmetry [17]. Exceptional points and synchronization are nevertheless distinct: selection of a slowly decaying collective mode does not require spectral coalescence.

Here, we propose a superconducting circuit that supports exceptional points, transient synchronization, and transient entanglement through tunable shared dissipation. Two system resonators couple to a lossy auxiliary resonator through DC-biased SQUIDs, while independent transmission lines provide private loss channels and readout ports. The design uses flux-tunable resonator couplings [18] and established circuit-QED elements [19, 20]. All couplings are static, with no direct coherent coupling between the system resonators. Eliminating the auxiliary mode yields collective damping and correlated input noise. We relate the normalized cross-decay coefficient ξ\xi to the circuit couplings and decay rates: its magnitude is set by the flux biases, while its sign follows the relative sign of the two couplings and determines which collective mode is protected from the shared channel. We check the reduced description against the full three-mode dynamics throughout.

For detuned resonators, we derive the exceptional-point condition at fixed private loss. Above the exceptional point, the mean amplitudes lock to a common frequency with a fixed relative phase. Comparison with the full three-mode spectrum shows that the finite response time of the auxiliary mode shifts the exceptional point by an amount that scales as 1/κb1/\kappa_b. For equal system frequencies, we obtain the time required to reach a prescribed antiphase tolerance; for couplings of opposite sign the same analysis gives in-phase alignment. This time is set by the collective decay splitting and the initial mean amplitudes, and numerical integration confirms the analytical prediction.

For initially separable squeezed states, we solve the reduced covariance dynamics and quantify entanglement using the logarithmic negativity [21, 22, 23]. Unequal collective relaxation converts local squeezing into transient intermode entanglement. We derive closed-form expressions for the squeezing threshold above which entanglement appears and for the time at which it disappears, and we examine the effects of shared coupling, thermal occupation, and the amount of initial squeezing. Entanglement and antiphase alignment can coexist, but they remain distinct: synchronization concerns decaying first moments, whereas entanglement concerns centered quantum fluctuations.

For realistic circuit parameters, these effects occur within a few hundred nanoseconds and persist at thermal occupations an order of magnitude higher than those of typical devices. We describe how the squeezed initial states can be prepared with a parametric pump, and how both the synchronization criterion and the covariance matrix are obtained from output-field quadrature measurements, including the surviving antiphase signal at the synchronization threshold. Because ∣ξ∣|\xi| can be calibrated from two ring-down rates, the circuit also provides a controllable source of correlated noise for studying correlated errors in superconducting processors [24, 25].

Section II presents the circuit, the reduced dynamics, and the physical parameters. Section III discusses the exceptional-point spectrum, synchronization, and entanglement. Section IV addresses state preparation, readout, and imperfections, and Section V summarizes the findings and discusses possible applications.

Model

Circuit model

We consider two superconducting resonators, a1a_1 and a2a_2, coupled to a common lossy auxiliary resonator bb, as shown in Figure 1. The system resonators are represented by the parallel LC circuits (L1,C1)(L_1,C_1) and (L2,C2)(L_2,C_2), and the auxiliary resonator by (Lb,Cb)(L_b,C_b). Each system resonator is coupled to bb through a DC-biased SQUID, whose flux bias Φj,dc\Phi_{j,\mathrm{dc}} sets the coupling strength gjg_j [18]. There is no direct coupling between a1a_1 and a2a_2. Each resonator is also coupled through a capacitor to an independent 50 Ω\Omega transmission line, which produces private energy decay of the system modes and strong damping of the auxiliary mode [19, 26]. When this damping is large, bb acts as a common reservoir for a1a_1 and a2a_2.

Circuit realizing a shared dissipation channel

Figure 1. Circuit realizing a shared dissipation channel. The system resonators a1a_1 and a2a_2 (blue) couple to the lossy auxiliary resonator bb (orange) through two DC-biased SQUIDs, whose fluxes Φ1,dc\Phi_{1,\mathrm{dc}} and Φ2,dc\Phi_{2,\mathrm{dc}} set g1g_1 and g2g_2. The capacitors Cγd,1C_{\gamma_{d,1}}, Cγd,2C_{\gamma_{d,2}}, and CκbC_{\kappa_b} couple the modes to independent 50 Ω\Omega lines, producing the energy-decay rates γd,1\gamma_{d,1}, γd,2\gamma_{d,2}, and κb\kappa_b. There is no direct coupling between the system resonators.

We work in the small-phase, linear regime of the SQUIDs and employ the rotating-wave approximation. Since the couplings are static, g1g_1 and g2g_2 are real. Their magnitudes are set by the flux biases, while their signs are fixed by the circuit geometry, i.e., by the polarity with which each coupler is connected to its resonator. We keep the signs general in Sec. II B, since the relative sign of g1g_1 and g2g_2 determines which collective mode is protected from the shared channel. The resonance frequencies include the local reactive shifts from the couplers and ports.

We use dimensionless units with the private energy-decay rate γd\gamma_d as the reference scale. Time is measured in units of γd−1\gamma_d^{-1}, so that τ=γdt\tau=\gamma_d t; frequencies, coupling strengths, and decay rates are measured in units of γd\gamma_d, and energies in units of ℏγd\hbar\gamma_d. We retain the same symbols for the scaled parameters, and an overdot denotes d/dτd/d\tau. Input fields are normalized with respect to τ\tau, while mode operators and quadrature variances are unchanged. Physical values are given in Sec. II C.

In a frame rotating at the auxiliary frequency ωb\omega_b, the Hamiltonian is

H=∑j=1,2[Δjaj†aj+gj(aj†b+b†aj)].(1)\mathcal{H}=\sum_{j=1,2}\left[\Delta_j a_j^\dagger a_j+g_j\left(a_j^\dagger b+b^\dagger a_j\right)\right]. \tag*{(1)}

where Δj=ωj−ωb\Delta_j=\omega_j-\omega_b. The Markov quantum Langevin equations are

a˙1=−(γd,12+iΔ1)a1−ig1b+γd,1a1,in,(2)\dot{a}_1=-\left(\frac{\gamma_{d,1}}{2}+i\Delta_1\right)a_1-ig_1b+\sqrt{\gamma_{d,1}}a_{1,\mathrm{in}}, \tag*{(2)}
a˙2=−(γd,22+iΔ2)a2−ig2b+γd,2a2,in,(3)\dot{a}_2=-\left(\frac{\gamma_{d,2}}{2}+i\Delta_2\right)a_2-ig_2b+\sqrt{\gamma_{d,2}}a_{2,\mathrm{in}}, \tag*{(3)}
b˙=−κb2b−i(g1a1+g2a2)+κbbin.(4)\dot{b}=-\frac{\kappa_b}{2}b-i(g_1a_1+g_2a_2)+\sqrt{\kappa_b}b_{\mathrm{in}}. \tag*{(4)}

The inputs are independent, zero-mean thermal fields. Writing f1=a1,inf_1=a_{1,\mathrm{in}}, f2=a2,inf_2=a_{2,\mathrm{in}}, and fb=binf_b=b_{\mathrm{in}}, their nonvanishing second moments are

⟨fμ(τ)fν†(τ′)⟩=δμν(nˉμ+1)δ(τ−τ′),(5)\langle f_\mu(\tau)f_\nu^\dagger(\tau')\rangle=\delta_{\mu\nu}(\bar{n}_\mu+1)\delta(\tau-\tau'), \tag*{(5)}
⟨fμ†(τ)fν(τ′)⟩=δμνnˉμδ(τ−τ′),(6)\langle f_\mu^\dagger(\tau)f_\nu(\tau')\rangle=\delta_{\mu\nu}\bar{n}_\mu\delta(\tau-\tau'), \tag*{(6)}

where μ,ν∈{1,2,b}\mu,\nu\in\{1,2,b\}. The occupations are taken constant over the relevant bandwidth, and all anomalous input correlations vanish.

The private lines also serve as readout ports. With the sign convention of Eqs. (2) and (3), the field leaving resonator jj is

aj,out=γd,jaj−aj,in.(7)a_{j,\mathrm{out}}=\sqrt{\gamma_{d,j}}a_j-a_{j,\mathrm{in}}. \tag*{(7)}

Since the inputs have zero mean, ⟨aj,out⟩=γd,j⟨aj⟩\langle a_{j,\mathrm{out}}\rangle=\sqrt{\gamma_{d,j}}\langle a_j\rangle, so averaged output traces give the mean amplitudes of both resonators directly. The fluctuations of aj,outa_{j,\mathrm{out}} carry the intracavity covariances. The quantities computed in Sec. III are therefore accessible through the private ports; the measurement is discussed in Sec. IV.

Effective dissipative coupling

We eliminate the auxiliary mode in the regime

κb2≫max⁡{g12+g22,γd,12,γd,22,∣Δ1∣,∣Δ2∣}.(8)\frac{\kappa_b}{2}\gg\max\left\{\sqrt{g_1^2+g_2^2},\frac{\gamma_{d,1}}{2},\frac{\gamma_{d,2}}{2},|\Delta_1|,|\Delta_2|\right\}. \tag*{(8)}

Setting b˙≃0\dot{b}\simeq0 in Eq. (4) gives

b≃−2iκb(g1a1+g2a2)+2κbbin.(9)b\simeq-\frac{2i}{\kappa_b}(g_1a_1+g_2a_2)+\frac{2}{\sqrt{\kappa_b}}b_{\mathrm{in}}. \tag*{(9)}

Keeping the detunings in the response of bb replaces 2/κb2/\kappa_b by 1/(κb/2−iΔj)1/(\kappa_b/2-i\Delta_j). The resulting corrections, of relative order 2∣Δj∣/κb2|\Delta_j|/\kappa_b, introduce frequency shifts and a coherent exchange between a1a_1 and a2a_2, and are neglected here. Substitution gives the reduced equations

a˙1=−(Γ12+iΔ1)a1−Γ122a2+F1,(10)\dot{a}_1=-\left(\frac{\Gamma_1}{2}+i\Delta_1\right)a_1-\frac{\Gamma_{12}}{2}a_2+F_1, \tag*{(10)}
a˙2=−(Γ22+iΔ2)a2−Γ122a1+F2,(11)\dot{a}_2=-\left(\frac{\Gamma_2}{2}+i\Delta_2\right)a_2-\frac{\Gamma_{12}}{2}a_1+F_2, \tag*{(11)}

where

γs,j=4gj2κb,Γj=γd,j+γs,j,(12)\gamma_{s,j}=\frac{4g_j^2}{\kappa_b},\qquad\Gamma_j=\gamma_{d,j}+\gamma_{s,j}, \tag*{(12)}
Γ12=4g1g2κb=σγs,1γs,2,σ=sgn⁡(g1g2),(13)\Gamma_{12}=\frac{4g_1g_2}{\kappa_b}=\sigma\sqrt{\gamma_{s,1}\gamma_{s,2}},\qquad\sigma=\operatorname{sgn}(g_1g_2), \tag*{(13)}

and the effective noise forces are

Fj=γd,jaj,in−2igjκbbin.(14)F_j=\sqrt{\gamma_{d,j}}a_{j,\mathrm{in}}-\frac{2ig_j}{\sqrt{\kappa_b}}b_{\mathrm{in}}. \tag*{(14)}

The local rates γs,j\gamma_{s,j} depend only on gj2g_j^2, whereas the cross-damping Γ12\Gamma_{12} and the shared part of the noise carry the relative sign of the couplings. These equations describe private damping of each mode together with a collective dissipator with jump operator (2/κb)(g1a1+g2a2)(2/\sqrt{\kappa_b})(g_1a_1+g_2a_2). The cross-damping and the correlated fluctuations thus originate from the same channel and must be retained together.

We characterize the shared channel by

ξ=Γ12Γ1Γ2=σγs,1γs,2(γd,1+γs,1)(γd,2+γs,2).(15)\xi=\frac{\Gamma_{12}}{\sqrt{\Gamma_1\Gamma_2}}=\sigma\sqrt{\frac{\gamma_{s,1}\gamma_{s,2}}{(\gamma_{d,1}+\gamma_{s,1})(\gamma_{d,2}+\gamma_{s,2})}}. \tag*{(15)}

For positive private decay rates and finite couplings, ∣ξ∣<1|\xi|<1: the magnitude of ξ\xi is set by the flux biases and its sign by the relative sign of g1g_1 and g2g_2. The normalized effective inputs ηj=Fj/Γj\eta_j=F_j/\sqrt{\Gamma_j} satisfy

[ηj(τ),ηk†(τ′)]=Mjkδ(τ−τ′),M=(1ξξ1),(16)[\eta_j(\tau),\eta_k^\dagger(\tau')]=M_{jk}\delta(\tau-\tau'),\qquad M=\begin{pmatrix}1&\xi\\ \xi&1\end{pmatrix}, \tag*{(16)}

and their cross-correlations are

⟨η1(τ)η2†(τ′)⟩=ξ(nˉb+1)δ(τ−τ′),(17)\langle\eta_1(\tau)\eta_2^\dagger(\tau')\rangle=\xi(\bar{n}_b+1)\delta(\tau-\tau'), \tag*{(17)}
⟨η1†(τ)η2(τ′)⟩=ξnˉbδ(τ−τ′).(18)\langle\eta_1^\dagger(\tau)\eta_2(\tau')\rangle=\xi\bar{n}_b\delta(\tau-\tau'). \tag*{(18)}

The local effective occupations are

nˉj,eff=γd,jnˉj+γs,jnˉbΓj,(19)\bar{n}_{j,\mathrm{eff}}=\frac{\gamma_{d,j}\bar{n}_j+\gamma_{s,j}\bar{n}_b}{\Gamma_j}, \tag*{(19)}

so that

⟨ηj†(τ)ηj(τ′)⟩=nˉj,effδ(τ−τ′),(20)\langle\eta_j^\dagger(\tau)\eta_j(\tau')\rangle=\bar{n}_{j,\mathrm{eff}}\delta(\tau-\tau'), \tag*{(20)}
⟨ηj(τ)ηj†(τ′)⟩=(nˉj,eff+1)δ(τ−τ′).(21)\langle\eta_j(\tau)\eta_j^\dagger(\tau')\rangle=(\bar{n}_{j,\mathrm{eff}}+1)\delta(\tau-\tau'). \tag*{(21)}

Thus ξ\xi measures the overlap of the effective input channels. For unequal reservoir occupations, it need not equal the cross-correlation normalized by the local thermal noise powers. The circuit realizes the correlated dissipation of Ref. [9] without a direct coherent coupling between the oscillators. The auxiliary mode couples only to the bright combination g1a1+g2a2g_1a_1+g_2a_2, while the orthogonal combination

aD=g2a1−g1a2g12+g22(22)a_D=\frac{g_2a_1-g_1a_2}{\sqrt{g_1^2+g_2^2}} \tag*{(22)}

is decoupled from the shared channel and is damped only by the private losses. For ∣g1∣=∣g2∣|g_1|=|g_2|, this dark mode is the antisymmetric combination a−=(a1−a2)/2a_-=(a_1-a_2)/\sqrt{2} when the couplings have equal signs (ξ>0\xi>0), and the symmetric combination a+=(a1+a2)/2a_+=(a_1+a_2)/\sqrt{2} when they have opposite signs (ξ<0\xi<0). Because a1a_1 and a2a_2 interact only through bb, the two cases are related by the local phase rotation a2→−a2a_2\to-a_2, which reverses the sign of g2g_2 and of ξ\xi. This transformation leaves the decay rates, the eigenvalue spectrum, and the entanglement unchanged, and only shifts the relative phase of the long-lived collective motion from π\pi to 00.

Hereafter we consider symmetric private decay and couplings of equal sign, γd,1=γd,2=1\gamma_{d,1}=\gamma_{d,2}=1 and g1=g2=gc>0g_1=g_2=g_c>0, so that 0≤ξ<10\leq\xi<1; the results for g1=−g2g_1=-g_2 follow from a2→−a2a_2\to-a_2. Then Γ1=Γ2=Γ\Gamma_1=\Gamma_2=\Gamma and Γ12=γs=ξΓ\Gamma_{12}=\gamma_s=\xi\Gamma. At fixed private decay, the coupling sweep is parameterized by

γs(ξ)=ξ1−ξ,Γ(ξ)=11−ξ,(23)\gamma_s(\xi)=\frac{\xi}{1-\xi},\qquad\Gamma(\xi)=\frac{1}{1-\xi}, \tag*{(23)}

and

gc(ξ)=12κbξ1−ξ.(24)g_c(\xi)=\frac{1}{2}\sqrt{\frac{\kappa_b\xi}{1-\xi}}. \tag*{(24)}

In this configuration the coupling condition in Eq. (8) reads 2gc/(κb/2)≪1\sqrt{2}g_c/(\kappa_b/2)\ll1, or equivalently ξ/(1−ξ)≪κb/2\xi/(1-\xi)\ll\kappa_b/2, so larger values of ξ\xi require a larger κb\kappa_b.

For a=(a1,a2)T\mathbf{a}=(a_1,a_2)^T, the reduced dynamics take the form

a˙=Wa+F,(25)\dot{\mathbf{a}}=W\mathbf{a}+\mathbf{F}, \tag*{(25)}

where F=(F1,F2)T\mathbf{F}=(F_1,F_2)^T and

W=−12(Γ+2iΔ1γsγsΓ+2iΔ2).(26)W=-\frac{1}{2}\begin{pmatrix}\Gamma+2i\Delta_1&\gamma_s\\ \gamma_s&\Gamma+2i\Delta_2\end{pmatrix}. \tag*{(26)}

The creation operators obey the complex-conjugate equations. Both the local damping and the cross-damping vary with ξ\xi. Reservoir occupations enter the noise correlations but not WW; consequently, its eigenvalues and the mean amplitudes are independent of these occupations for fixed rates and initial first moments.

Two detuning configurations are used below. Opposite detunings, Δ1=δ/2\Delta_1=\delta/2 and Δ2=−δ/2\Delta_2=-\delta/2, with δ=ω1−ω2\delta=\omega_1-\omega_2, are used to investigate exceptional points. Resonant modes, Δ1=Δ2=0\Delta_1=\Delta_2=0, are used for synchronization and entanglement; a common detuning would only add a common rotation in the reduced model while increasing the neglected corrections. Varying ξ\xi at fixed detunings and private decay assumes that the frequency shifts accompanying changes of the SQUID flux biases are compensated.

Physical parameters

To connect the scaled parameters to a device, we take resonance frequencies near 5 GHz, a private decay rate γd/2π=0.5\gamma_d/2\pi= 0.5 MHz, and an auxiliary decay rate κb/2π=100\kappa_b/2\pi= 100 MHz, so that κb=200\kappa_b = 200 in scaled units. The time unit is then 1/γd≃0.32 μs1/\gamma_d \simeq0.32\ \mu\mathrm{s}. For ξ=0.8\xi= 0.8, Eq. (24) gives gc/2π≃7.1g_c/2\pi\simeq7.1 MHz and 2gc/(κb/2)≃0.20\sqrt{2}g_c/(\kappa_b/2) \simeq0.20; for ξ=0.9\xi= 0.9, gc/2π≃10.6g_c/2\pi\simeq10.6 MHz and 2gc/(κb/2)≃0.30\sqrt{2}g_c/(\kappa_b/2) \simeq0.30. These couplings are well within the range of flux-tunable couplers between superconducting resonators [18]. All rates remain small compared with the resonance frequencies, consistent with the rotating-wave and Markov approximations. The private decay is dominated by the engineered port loss: an internal quality factor near 10610^6 contributes an internal decay rate of 2π×52\pi\times5 kHz, about 1% of γd\gamma_d.

Results

Exceptional point

For opposite detunings, Δ1=−Δ2=δ/2\Delta_1 = -\Delta_2 = \delta/2, the drift matrix of Eq. (26) can be written as W=−iHeffW = -i\mathcal{H}_{\mathrm{eff}}, with the effective non-Hermitian Hamiltonian

Heff=(δ/2−iγs/2iΓ−iγs/2−δ/2)−iΓ2I2.(27)\mathcal{H}_{\mathrm{eff}} = \begin{pmatrix} \delta/2-i\gamma_s/2 & i\Gamma\\ -i\gamma_s/2 & -\delta/2 \end{pmatrix} -\frac{i\Gamma}{2}\mathbb{I}_2. \tag*{(27)}

The coupling is purely imaginary, so Heff\mathcal{H}_{\mathrm{eff}} is anti-PT symmetric [17]. The eigenvalues of WW are

λ±=−Γ2±12γs2−δ2.(28)\lambda_{\pm}=-\frac{\Gamma}{2}\pm\frac{1}{2}\sqrt{\gamma_s^2-\delta^2}. \tag*{(28)}

For δ≠0\delta\ne0, the two eigenvalues and their eigenvectors coalesce at γs=∣δ∣\gamma_s = |\delta|, giving

ξEPred=∣δ∣1+∣δ∣.(29)\xi_{\mathrm{EP}}^{\mathrm{red}}=\frac{|\delta|}{1+|\delta|}. \tag*{(29)}

Below this point, the modes have equal decay rates and different oscillation frequencies. Above it, their frequencies coincide in the rotating frame while their decay rates split. The transition is driven entirely by the shared dissipation; no direct coherent coupling between the system resonators is needed. For the parameters of § 2 C, a mismatch δ=0.1\delta= 0.1 corresponds to δ/2π=50\delta/2\pi= 50 kHz, and the exceptional point is reached at gc/2π≃1.1g_c/2\pi\simeq1.1 MHz.

To assess the elimination, we compare these eigenvalues with the two slow branches of the full three-mode drift matrix,

W3=(−1/2−iδ/20−igc0−1/2+iδ/2−igc−igc−igc−κb/2).(30)W_3 = \begin{pmatrix} -1/2-i\delta/2 & 0 & -ig_c \\ 0 & -1/2+i\delta/2 & -ig_c \\ -ig_c & -ig_c & -\kappa_b/2 \end{pmatrix}. \tag*{(30)}

with gcg_c given by Eq. (24), so that the same value of ξ\xi specifies corresponding full and reduced parameters. Figure 2 shows close agreement between the slow branches. The finite response time of the auxiliary mode shifts the coalescence point slightly, from ξEPred=0.0909\xi_{\mathrm{EP}}^{\mathrm{red}} = 0.0909 to ξEPfull=0.0905\xi_{\mathrm{EP}}^{\mathrm{full}} = 0.0905, a relative shift of 0.5%. This shift scales as 1/κb1/\kappa_b; halving κb\kappa_b to 100 doubles it. The rapidly decaying auxiliary branch is not shown.

Exceptional-point structure of the full and reduced models

Figure 2. Exceptional-point structure of the full and reduced models at fixed private decay. (a) Real and (b) imaginary parts of the two slow eigenvalues as functions of ξ\xi. (c) Eigenvalue splitting ∣λ+−λ−∣|\lambda_{+}-\lambda_{-}| near the coalescence points. Solid orange and dashed blue curves denote the full and reduced models, respectively. The vertical orange dash-dotted and blue dotted lines mark their respective exceptional points. Parameters are κb=200\kappa_b=200 and δ=0.1\delta=0.1 (Δ1=−Δ2=0.05\Delta_1=-\Delta_2=0.05). Eigenvalues are in units of γd\gamma_d, and gcg_c is varied according to Eq. (24).

Mean-amplitude synchronization

Above the exceptional point, the mean amplitudes lock to a common frequency. The slower eigenvector, belonging to λ+\lambda_+, satisfies

x2x1=−eiφ,sin⁡φ=∣δ∣γs.(31)\frac{x_2}{x_1}=-e^{i\varphi},\qquad\sin\varphi=\frac{|\delta|}{\gamma_s}. \tag*{(31)}

where xj=⟨aj⟩x_j=\langle a_j\rangle. The two resonators therefore oscillate with equal amplitudes and a fixed phase difference ∣arg⁡(x2/x1)∣=π−φ|\arg(x_2/x_1)|=\pi-\varphi, which approaches antiphase for γs≫∣δ∣\gamma_s\gg|\delta| and reduces to π/2\pi/2 at the exceptional point. Below the exceptional point, the two eigenmodes have different frequencies and the relative phase does not settle. For identical resonators, δ=0\delta=0, the locked state is exactly antiphase; we use this limit below because it admits closed-form expressions. For couplings of opposite sign, ξ<0\xi<0, the eigenvalues are unchanged and x2/x1=eiφx_2/x_1=e^{i\varphi}, so the locked state approaches in-phase motion instead.

Mean-amplitude synchronization

For resonant modes, Δ1=Δ2=0\Delta_1=\Delta_2=0, we introduce the collective modes a±=(a1±a2)/2a_{\pm}=(a_1\pm a_2)/\sqrt{2}. Their equations are

a˙±=−κ±2a±+F±,(32)\dot{a}_{\pm}=-\frac{\kappa_{\pm}}{2}a_{\pm}+F_{\pm}, \tag*{(32)}

where F±=(F1±F2)/2F_{\pm}=(F_1\pm F_2)/\sqrt{2} and

κ+=1+2γs,κ−=1.(33)\kappa_+=1+2\gamma_s,\qquad\kappa_-=1. \tag*{(33)}

The symmetric mode couples to the shared reservoir, whereas the shared noise and damping cancel from the antisymmetric mode. The latter is dark with respect to the shared channel but still undergoes private decay. This mechanism is the common-bath route to synchronization studied in Refs. [4, 5]; here it arises from an explicit circuit element whose strength is set by the SQUID flux bias.

Since the input fields have zero mean, the collective amplitudes x±=(x1±x2)/2x_{\pm}=(x_1\pm x_2)/\sqrt{2} obey

x±(τ)=x±(0)e−κ±τ/2.(34)x_{\pm}(\tau)=x_{\pm}(0)e^{-\kappa_{\pm}\tau/2}. \tag*{(34)}

For γs>0\gamma_s>0 and x−(0)≠0x_-(0)\ne0, the faster decay of the symmetric component leads to the antiphase relation x1≃−x2x_1\simeq-x_2. For couplings of opposite sign the roles of a±a_{\pm} are exchanged, the symmetric mode is dark, and the resonators synchronize in phase; the expressions below then hold with x1+x2x_1+x_2 and x1−x2x_1-x_2 interchanged, so that cot⁡(θ/2)\cot(\theta/2) is replaced by tan⁡(θ/2)\tan(\theta/2). We quantify the remaining symmetric component by

R(τ)=∣x1(τ)+x2(τ)x1(τ)−x2(τ)∣=∣α1+α2∣∣α1−α2∣e−γsτ,(35)R(\tau)=\left|\frac{x_1(\tau)+x_2(\tau)}{x_1(\tau)-x_2(\tau)}\right|=\frac{|\alpha_1+\alpha_2|}{|\alpha_1-\alpha_2|}e^{-\gamma_s\tau}, \tag*{(35)}

where αj=xj(0)\alpha_j=x_j(0). For a tolerance $0<\epsilon<1$, the synchronization time τsync\tau_{\mathrm{sync}} is the earliest time after which $R\le\epsilon remains satisfied. For nonzero initial symmetric and antisymmetric components,

τsync=max⁡{0,1γsln⁡(∣α1+α2∣ϵ∣α1−α2∣)}.(36)\tau_{\mathrm{sync}}=\max\left\{0,\frac{1}{\gamma_s}\ln\left(\frac{|\alpha_1+\alpha_2|}{\epsilon|\alpha_1-\alpha_2|}\right)\right\}. \tag*{(36)}

For α1=1\alpha_1=1 and α2=eiθ\alpha_2=e^{i\theta},

τsync(θ)=max⁡{0,1γsln⁡(∣cot⁡(θ/2)∣ϵ)}.(37)\tau_{\mathrm{sync}}(\theta)=\max\left\{0,\frac{1}{\gamma_s}\ln\left(\frac{|\cot(\theta/2)|}{\epsilon}\right)\right\}. \tag*{(37)}

An initially antisymmetric mean has τsync=0\tau_{\mathrm{sync}}=0, while an initially symmetric mean has no antisymmetric component and never reaches the antiphase condition. If both means vanish, the criterion is undefined.

This criterion describes antiphase alignment of decaying mean amplitudes rather than sustained oscillations, and ϵ\epsilon sets the allowed symmetric admixture. At fixed drift parameters and initial means, τsync\tau_{\mathrm{sync}} is independent of the bath occupations and of the initial squeezing.

Figure 3 illustrates the initial-phase dependence at ξ=0.8\xi=0.8. An initially symmetric mean remains symmetric and decays rapidly, whereas any nonzero antisymmetric component eventually dominates. Increasing the initial phase difference toward π\pi reduces the symmetric admixture and hence the synchronization time. Numerical integration agrees with Eq. (37). These trajectories are the same for coherent and displaced squeezed initial states with the same α1\alpha_1 and α2\alpha_2. For θ=5π/6\theta=5\pi/6 and ϵ=0.01\epsilon=0.01, τsync≃0.82\tau_{\mathrm{sync}}\simeq0.82, which corresponds to about 0.26 μs0.26\,\mu\mathrm{s} for the parameters of § 2 C.

In the full three-mode model with κb=200\kappa_b=200, τsync\tau_{\mathrm{sync}} is 3–4% shorter for the initial phases shown, because the exact decay rate of the symmetric mode slightly exceeds k+k_{+}. This difference scales as 1/κb1/\kappa_b.

Gaussian dynamics and entanglement

We introduce the quadratures

qj=aj+aj†2,pj=aj−aj†i2,(38)q_j=\frac{a_j+a_j^\dagger}{\sqrt{2}},\qquad p_j=\frac{a_j-a_j^\dagger}{i\sqrt{2}}, \tag*{(38)}

with [qj,pk]=iδjk[q_j,p_k]=i\delta_{jk}, so the vacuum variance is 1/21/2. For u=(q1,p1,q2,p2)Tu=(q_1,p_1,q_2,p_2)^{T}, the centered covariance matrix is

Vjk=12⟨δujδuk+δukδuj⟩,δuj=uj−⟨uj⟩.(39)V_{jk}=\frac{1}{2}\langle\delta u_j\delta u_k+\delta u_k\delta u_j\rangle,\qquad\delta u_j=u_j-\langle u_j\rangle. \tag*{(39)}

The system is initially prepared in a product of displaced squeezed states,

∣ψ(0)⟩=D1(α1)S1(s)∣0⟩⊗D2(α2)S2(s)∣0⟩,(40)|\psi(0)\rangle=D_1(\alpha_1)S_1(s)|0\rangle\otimes D_2(\alpha_2)S_2(s)|0\rangle, \tag*{(40)}

where Dj(αj)=exp⁡(αjaj†−αj∗aj)D_j(\alpha_j)=\exp(\alpha_j a_j^\dagger-\alpha_j^*a_j) and Sj(s)=exp⁡[s(aj2−aj†2)/2]S_j(s)=\exp[s(a_j^2-a_j^{\dagger2})/2], with equal real s≥0s\geq0 so that both modes are squeezed along the same quadrature. The initial covariance is

V(0)=12diag⁡(e−2s,e2s,e−2s,e2s),(41)V(0)=\frac{1}{2}\operatorname{diag}(e^{-2s},e^{2s},e^{-2s},e^{2s}), \tag*{(41)}

which does not depend on the displacements.

The linear dynamics with thermal Gaussian inputs preserve Gaussianity [23], and the covariance evolves as

V˙=AV+VAT+D,(42)\dot{V}=AV+VA^{T}+D, \tag*{(42)}

with

A=−12(Γ0γs00Γ0γsγs0Γ00γs0Γ).(43)A=-\frac{1}{2}\begin{pmatrix} \Gamma& 0 & \gamma_s & 0\\ 0 & \Gamma& 0 & \gamma_s\\ \gamma_s & 0 & \Gamma& 0\\ 0 & \gamma_s & 0 & \Gamma \end{pmatrix}. \tag*{(43)}
Initial-state dependence of antiphase synchronization in the reduced model

Figure 3. Initial-state dependence of antiphase synchronization in the reduced model. (a)–(g) Evolution of Re⁡[x1e−iωrτ]\operatorname{Re}[x_1e^{-i\omega_r\tau}] (solid orange) and Re⁡[x2e−iωrτ]\operatorname{Re}[x_2e^{-i\omega_r\tau}] (dashed blue) for α1=1\alpha_1=1 and α2=eiθ\alpha_2=e^{i\theta}, with θ=0\theta=0, π/6\pi/6, π/3\pi/3, π/2\pi/2, 2π/32\pi/3, 5π/65\pi/6, π\pi. The common reference rotation ωr=40\omega_r=40 is applied for visibility only and does not affect RR or τsync\tau_{\mathrm{sync}}. Dotted vertical lines mark the analytical synchronization times; triangles mark the numerically determined threshold crossings. Shading indicates R≤ϵR\leq\epsilon. (h) Analytical synchronization time versus the initial phase (solid curve), with numerical values (markers). Parameters are ξ=0.8\xi=0.8 (γs=4\gamma_s=4, Γ=5\Gamma=5), Δ1=Δ2=0\Delta_1=\Delta_2=0, and ϵ=0.01\epsilon=0.01. The reduced model describes the device of § 2 C with κb=200\kappa_b=200; at fixed ξ\xi, the curves do not depend on κb\kappa_b.

For equal input occupations nˉ1=nˉ2=nˉb=n\bar{n}_1=\bar{n}_2=\bar{n}_b=n, the diffusion matrix is D=−2vAD=-2vA, with v=n+1/2v=n+1/2. The stationary covariance is then vI4vI_4, and

V(τ)=vI4+eAτ[V(0)−vI4]eATτ.(44)V(\tau)=vI_4+e^{A\tau}[V(0)-vI_4]e^{A^{T}\tau}. \tag*{(44)}

Entanglement is quantified by the logarithmic negativity EN=ln⁡∥ρ12T2∥1E_N=\ln\lVert\rho_{12}^{T_2}\rVert_1 [22]. Partial transposition of mode 2 reverses p2p_2, V~=ΛVΛ\widetilde{V}=\Lambda V\Lambda with Λ=diag⁡(1,1,1,−1)\Lambda=\operatorname{diag}(1,1,1,-1). Writing VV in 2×22\times2 blocks B1B_1, B2B_2, and CC, and defining Σ~=det⁡B1+det⁡B2−2det⁡C\widetilde{\Sigma}=\det B_1+\det B_2-2\det C, the smallest symplectic eigenvalue of V~\widetilde{V} is

ν~−=Σ~−Σ~2−4det⁡V~2(45)\widetilde{\nu}_{-}=\sqrt{\frac{\widetilde{\Sigma}-\sqrt{\widetilde{\Sigma}^{2}-4\det\widetilde{V}}}{2}} \tag*{(45)}

and

EN=max⁡{0,−ln⁡(2ν~−)}.(46)E_N=\max\{0,-\ln(2\widetilde{\nu}_{-})\}. \tag*{(46)}

For two-mode Gaussian states, ν~−<1/2\widetilde{\nu}_{-}<1/2 is necessary and sufficient for entanglement [21].

The collective decay hierarchy that produces synchronization also converts local squeezing into intermode entanglement. To see this, we use the quadratures of the collective modes a±a_{\pm} introduced in Sec. III B,

q±=q1±q22,p±=p1±p22.(47)q_{\pm}=\frac{q_1\pm q_2}{\sqrt{2}},\qquad p_{\pm}=\frac{p_1\pm p_2}{\sqrt{2}}. \tag*{(47)}

In the physical basis, each quadrature is damped at the rate Γ\Gamma and coupled to the other resonator through the cross-damping γs\gamma_s. In the collective basis, the cross-damping adds to the symmetric rate and cancels from the antisymmetric one, so AA becomes diagonal with entries −κ±/2-\kappa_{\pm}/2, where κ±=Γ±γs\kappa_{\pm}=\Gamma\pm\gamma_s are the rates of Eq. (33). For equal occupations, the collective noises are uncorrelated. Since both resonators are squeezed identically, V(0)V(0) has the same diagonal form in the collective basis. The collective covariance therefore remains diagonal, with variances

vq,±(τ)=v+(e−2s2−v)e−κ±τ,(48)v_{q,\pm}(\tau)=v+\left(\frac{e^{-2s}}{2}-v\right)e^{-\kappa_{\pm}\tau}, \tag*{(48)}
vp,±(τ)=v+(e2s2−v)e−κ±τ.(49)v_{p,\pm}(\tau)=v+\left(\frac{e^{2s}}{2}-v\right)e^{-\kappa_{\pm}\tau}. \tag*{(49)}

Partial transposition interchanges p+p_{+} and p−p_{-}, so that

ν~−2=min⁡{vq,+vp,−,vq,−vp,+}.(50)\widetilde{\nu}_{-}^{2}=\min\{v_{q,+}v_{p,-},v_{q,-}v_{p,+}\}. \tag*{(50)}

Both products equal 1/41/4 at τ=0\tau=0. Entanglement develops when the anti-squeezed variance of the symmetric mode relaxes quickly while the squeezed variance of the antisymmetric mode degrades slowly, giving vq,−vp,+<1/4v_{q,-}v_{p,+}<1/4. Initial squeezing supplies the nonclassical resource, and the shared channel converts it into entanglement between the resonators. For couplings of opposite sign, a+a_{+} and a−a_{-} exchange roles and the relevant product becomes vq,+vp,−v_{q,+}v_{p,-}; since the two cases are related by the local rotation a2→−a2a_2\to-a_2, ENE_N is unchanged.

The disappearance of entanglement can be estimated in closed form. For κ+≫1\kappa_{+}\gg1, the symmetric mode reaches vp,+≃vv_{p,+}\simeq v quickly, and the condition becomes vq,−<1/(4v)v_{q,-}<1/(4v). This holds until

τd≃ln⁡v−e−2s/2v−1/(4v).(51)\tau_d\simeq\ln\frac{v-e^{-2s}/2}{v-1/(4v)}. \tag*{(51)}

Since τd\tau_d depends only on the antisymmetric mode, it is independent of ξ\xi once κ+\kappa_{+} is large. For n=0n=0 (v=1/2v=1/2), the denominator vanishes and the entanglement decays only asymptotically.

The dynamics contain neither parametric pair creation nor a squeezed reservoir. For coherent initial states (s=0s=0), passive linear evolution with thermal inputs keeps the Glauber PP function positive and cannot generate entanglement. For s>0s>0 the entanglement is transient, since private damping drives both collective modes to the same thermal covariance. The role of squeezing is also visible in the centered anomalous correlation m12=⟨δa1δa2⟩m_{12}=\langle\delta a_1\delta a_2\rangle, with δaj=aj−⟨aj⟩\delta a_j=a_j-\langle a_j\rangle:

m12(τ)=sinh⁡(2s)4[e−τ−e−(1+2γs)τ].(52)m_{12}(\tau)=\frac{\sinh(2s)}{4}\left[e^{-\tau}-e^{-(1+2\gamma_s)\tau}\right]. \tag*{(52)}

A nonzero m12m_{12} is not sufficient for entanglement, because thermal fluctuations can render the state separable while m12m_{12} remains finite.

The onset of entanglement also follows from Eq. (50). The product vq,−vp,+v_{q,-}v_{p,+} equals 1/41/4 at τ=0\tau=0, and it falls below 1/41/4 immediately afterward when its initial slope is negative. With x=e2sx=e^{2s}, this condition reads

2vx2−(1+κ+)x+2κ+v<0.(53)2vx^2-(1+\kappa_{+})x+2\kappa_{+}v<0. \tag*{(53)}

The smaller root of the left-hand side gives the threshold squeezing sths_{\mathrm{th}} above which entanglement appears. For ξ=0.8\xi=0.8, Eq. (53) gives sth=0.53s_{\mathrm{th}}=0.53, 1.06, and 2.13 dB for n=0.05n=0.05, 0.1, and 0.2, in agreement with the numerical thresholds to within 0.01 dB. The quadratic has no real roots when v>(1+κ+)/(4κ+)v>(1+\kappa_{+})/(4\sqrt{\kappa_{+}}), that is, n>0.33n>0.33 for ξ=0.8\xi=0.8. Numerically, we find no entanglement for n=0.5n=0.5 at any squeezing up to 17 dB.

Figure 4 shows ENE_N for initial squeezing of 6.5 dB (s=0.75s=0.75) and 3 dB (s=0.35s=0.35). At fixed n=0.1n=0.1 [Figure 4(a)], increasing ξ\xi separates the collective decay rates and enhances the peak entanglement: ξ=0\xi=0 and 0.4 give none, ξ=0.6\xi=0.6 gives a weak transient, and ξ=0.8\xi=0.8 and 0.9 give pronounced peaks. The ξ=0.8\xi=0.8 and 0.9 curves vanish at the same time, τ≃0.98\tau\simeq0.98, which coincides with Eq. (51) because both κ+=9\kappa_{+}=9 and κ−=19\kappa_{-}=19 are large compared with κ−=1\kappa_{-}=1.

Transient entanglement generated by collective dissipation

Figure 4. Transient entanglement generated by collective dissipation, obtained from Eq. (42) (solid lines). (a) Dependence on ξ=0,0.4,0.6,0.8,0.9\xi= 0, 0.4, 0.6, 0.8, 0.9 at fixed n=0.1n = 0.1, with the private decay held fixed. (b) Dependence on n=0,0.05,0.1,0.2,0.5n = 0, 0.05, 0.1, 0.2, 0.5 at fixed ξ=0.8\xi= 0.8. (c) Same as (b) for 3 dB of initial squeezing. Panels (a) and (b) use 6.5 dB (s=0.75s = 0.75) and panel (c) 3 dB (s=0.35s = 0.35) of squeezing, with identical squeezing orientations and Δ1=Δ2=0\Delta_1 = \Delta_2 = 0. Dashed black lines in (a) and (b) show the full three-mode model with κb=200\kappa_b = 200. In (b), the dotted line marks τsync\tau_{\mathrm{sync}} for α1=1\alpha_1 = 1, α2=ei5π/6\alpha_2 = e^{i5\pi/6}, and ϵ=0.01\epsilon= 0.01, and the shaded band marks the coexistence of R≤ϵR \leq\epsilon and EN>0E_N > 0 for n=0.1n = 0.1. Displacements affect the synchronization marker but not ENE_N.

At fixed ξ=0.8\xi=0.8 [Figure 4(b)], increasing nn raises the diffusion without changing the drift. Both the peak entanglement and its lifetime decrease, and no entanglement is obtained for n=0.5n=0.5. Equation (51) gives τd≃1.53\tau_d\simeq1.53, 0.98, and 0.54 for n=0.05n=0.05, 0.1, and 0.2, close to the numerical values 1.52, 0.98, and 0.52. For the parameters of § 2 C, τd≃0.98\tau_d\simeq0.98 corresponds to about 0.31 μs0.31\,\mu\mathrm{s}.

At 5 GHz, n=0.1n=0.1 corresponds to an effective temperature of about 100 mK. Superconducting resonators in dilution refrigerators typically thermalize near 40–60 mK, where n≲0.02n\lesssim0.02. The value n=0.1n=0.1 is therefore conservative, and Figure 4(b) shows that the entanglement survives at occupations an order of magnitude above the expected value; Eq. (53) sets the limit at n≃0.33n\simeq0.33 for ξ=0.8\xi=0.8. For n=0.02n=0.02, Eq. (51) gives τd≃2.3\tau_d\simeq2.3, or about 0.75 μs0.75\,\mu\mathrm{s}.

With 3 dB of initial squeezing [Figure 4(c)], the entanglement is weaker and shorter-lived, with a peak EN≃0.09E_N\simeq0.09 that vanishes at τ≃0.64\tau\simeq0.64 for n=0.1n=0.1. Since 3 dB lies above the threshold for n≤0.2n\leq0.2, entanglement persists for all occupations except n=0.5n=0.5, although it becomes weak for n=0.2n=0.2.

The full three-mode covariance dynamics with κb=200\kappa_b=200, including the auxiliary mode bb, reproduce these results [dashed lines in Figure 4(a) and Figure 4(b)]. For ξ=0.8\xi= 0.8 and 0.90.9 and all occupations shown, the peak ENE_N differs by less than 2% and the time at which entanglement vanishes agrees to within 0.01. The only visible difference is a short initial delay, during which the auxiliary mode builds up its correlations with the system modes.

The synchronization time is unaffected by nn, because the first moments do not depend on the occupations. For θ=5π/6\theta= 5\pi/6 and ϵ=0.01\epsilon= 0.01, τsync≃0.82\tau_{\mathrm{sync}} \simeq0.82, so at n=0.1n = 0.1 antiphase alignment and entanglement coexist over 0.82≲τ≲0.980.82 \lesssim\tau\lesssim0.98 [shaded in Figure 4(b)]. The width of this window depends on ϵ\epsilon and θ\theta. The two effects share their origin in the collective decay hierarchy but are distinct: synchronization concerns the first moments, whereas ENE_N concerns the centered fluctuations. Coherent initial states show the same synchronization while remaining separable.

Experimental considerations

Preparation of the squeezed initial state

The entanglement results of Sec. III C require each system resonator to be squeezed before τ=0\tau= 0. This can be done with a flux-pumped SQUID in each system resonator, as in flux-driven Josephson parametric amplifiers [27]. Modulating its flux at 2ωj2\omega_j generates the squeezing Hamiltonian i(ϵj/2)(aj†2−aj2)i(\epsilon_j/2)(a_j^{\dagger2} - a_j^2), and a pump pulse of duration TpT_p produces s=ϵjTps = \epsilon_j T_p in the absence of loss. The pump is switched off before τ=0\tau= 0. Afterward the flux through this SQUID is static, so it only renormalizes ωj\omega_j, and the dynamics remain those of the time-independent model of § 2.

In the steady state, intracavity squeezing generated by a parametric drive on a damped mode is limited to 3 dB. A short pump pulse with ϵj≫γd\epsilon_j \gg\gamma_d avoids this limit, because the private loss acts only weakly during the pulse. Intracavity squeezing beyond 3 dB has also been demonstrated by reservoir engineering [28]. For the parameters of § 2 C, a pump rate ϵj/2π=5\epsilon_j/2\pi= 5 MHz reaches 6.5 dB in about 26 ns, including the private loss during the pulse. This is an order of magnitude shorter than 1/γd≃0.32 μs1/\gamma_d \simeq0.32\ \mu\mathrm{s}, and the resulting state has a purity of about 0.95. A pump rate of 3 MHz requires about 45 ns.

During the preparation, the couplers should be biased at a weakly coupled point, so that the shared channel does not act on the resonators while they are being squeezed. Stepping the flux biases to the operating point then starts the dynamics at τ=0\tau= 0. The step switches between two static configurations, and the auxiliary mode relaxes to its thermal state within 2/κb≃32/\kappa_b \simeq3 ns. This is the initial condition used in the full three-mode calculations of Sec. III.

Readout

Both quantities studied in Sec. III are obtained from the fields leaving the private ports, Eq. (7). Averaged heterodyne records give ⟨aj,out⟩=γd,jxj\langle a_{j,\mathrm{out}}\rangle= \sqrt{\gamma_{d,j}}x_j, and hence x1±x2x_1 \pm x_2 and R(τ)R(\tau).

For equal private decay rates, RR is independent of γd\gamma_d once the gain and phase of the two detection chains are calibrated against each other.

At the synchronization threshold the antiphase signal is still large. The antisymmetric amplitude has decayed only by the factor e−κ−τsync/2e^{-\kappa_- \tau_{\mathrm{sync}}/2}, which equals 0.66 for θ=5π/6\theta= 5\pi/6. The demanding part of the measurement is resolving the symmetric admixture at the level R≃ϵ=0.01R \simeq\epsilon= 0.01, which requires averaging over many repetitions. Since the resonators return to equilibrium within a few microseconds, 10510^5 repetitions take less than one second.

The covariance matrix, and hence ENE_N, follows from the second moments of the amplified output records. The added noise of the amplification chain is subtracted using reference measurements with the resonators in their thermal state [29, 30]. Near-quantum-limited amplifiers [31] reduce the number of repetitions required. Reconstructing the intracavity covariance V(τ)V(\tau) requires temporal mode matching: the output records are integrated with filter functions matched to the resonator modes, and the measured moments are related to the intracavity moments through Eq. (7) [30]. Because V(τ)V(\tau) evolves on the timescale 1/κ+1/\kappa_+, the filter windows must be short compared with it, or the measured output correlations can be compared directly with the model.

The dynamics last about 0.3 μs0.3\ \mu\mathrm{s}, and their fastest timescale is the decay of the symmetric mode, at rate κ++γd/2π≃4.5\kappa_+ + \gamma_d/2\pi\simeq4.5 MHz for ξ=0.8\xi= 0.8. A detection bandwidth of several megahertz is therefore sufficient. This is well within the bandwidth of current readout chains, including traveling-wave parametric amplifiers [31]. The strongly damped auxiliary port does not need to be measured.

Imperfections

Coupling asymmetry. For equal private decay and g1≠g2g_1 \ne g_2, the combination g2a1−g1a2g_2a_1 - g_1a_2 remains an exact eigenmode of the reduced drift matrix, decaying at the private rate. The long-time state is then still antiphase, but with amplitude ratio ∣x2/x1∣=g1/g2\lvert x_2/x_1\rvert= g_1/g_2. Unequal private decay rates, γd,1≠γd,2\gamma_{d,1} \ne\gamma_{d,2}, mix this mode with the bright mode; the admixture is small when ∣γd,1−γd,2∣≪γs,1+γs,2\lvert\gamma_{d,1} - \gamma_{d,2}\rvert\ll\gamma_{s,1} + \gamma_{s,2}.

Residual detuning. A residual frequency mismatch δ\delta above the exceptional point shifts the locked relative phase from π\pi to π−arcsin⁡(∣δ∣/γs)\pi- \arcsin(\lvert\delta\rvert/\gamma_s), according to Eq. (31). For ξ=0.8\xi= 0.8 (γs=4\gamma_s = 4) and δ=0.1\delta= 0.1, the shift is 0.025 rad.

Flux noise. Away from zero flux, the SQUID inductance depends on Φj,dc\Phi_{j,\mathrm{dc}} to first order, so flux noise modulates both gjg_j and the resonator frequencies. Since the relevant dynamics last a few hundred nanoseconds, slow flux drift mainly produces shot-to-shot variation of ϵ\epsilon and of the detunings. This variation can be reduced by operating close to zero flux, where the inductance is first-order insensitive.

Conclusion

We have proposed a superconducting circuit in which two resonators share a dissipation channel through a strongly damped auxiliary resonator, with the couplings set by static flux biases of two DC SQUIDs. Adiabatic elimination of the auxiliary mode yields cross-damping and correlated noise that originate from the same collective dissipator. The magnitude of the shared-channel parameter ξ\xi is tuned continuously within 0≤∣ξ∣<10 \leq|\xi| < 1 by the coupler biases, while its sign is fixed by the relative sign of the two couplings. The circuit thus realizes the correlated dissipation of Ref. [9] without a direct coherent coupling between the resonators.

The shared channel produces three effects. For detuned resonators, the eigenvalues coalesce at an exceptional point ξEP=∣δ∣/(1+∣δ∣)\xi_{\mathrm{EP}} = |\delta|/(1 + |\delta|), above which the mean amplitudes lock to a common frequency with a fixed relative phase. For identical resonators, the collective mode that couples to the auxiliary resonator decays faster than the dark mode, which drives the mean amplitudes into antiphase alignment for couplings of equal sign and into in-phase alignment for couplings of opposite sign. The synchronization time is set by the collective decay splitting and the initial mean amplitudes. When the resonators are initially squeezed, the same decay hierarchy converts local squeezing into transient entanglement. We obtained closed-form expressions for the squeezing threshold above which entanglement appears and for the time at which it disappears. Synchronization and entanglement share this origin but remain distinct: the former concerns first moments and also occurs for coherent states, whereas the latter requires initial squeezing and is limited by thermal noise. The full three-mode dynamics confirm the reduced description to within a few percent for κb=200\kappa_b = 200. For a private decay rate γd/2π=0.5\gamma_d/2\pi= 0.5 MHz, these effects occur within about 0.3 μs0.3~\mu\mathrm{s}, persist at thermal occupations an order of magnitude higher than those of typical devices, and are accessible through the output fields of the private ports. Beyond realizing the Bittner–Tyagi model, the circuit is useful in its own right because the noise correlation is set by the coupler fluxes and, for identical modes, can be calibrated from the two collective ring-down rates, ∣ξ∣=(k+−k−)/(k++k−)|\xi| = (k_+ - k_-)/(k_+ + k_-). It can therefore serve as a controllable source of correlated noise, a test bed for the correlated errors that limit error correction in superconducting processors [24, 25]. The dark mode, which decays only through the private losses, is a bosonic analogue of a decoherence-free subspace [32] whose protection can be switched with the flux. Since the entanglement leaks out through the two independent lines, the device can also serve as a source of entangled microwave fields that requires no two-mode pump [33].

Data availability

The code and data that support the findings of this article will be made openly available upon publication.

Acknowledgments

This work was supported by the National Science Foundation under Grant No. CHE-2404788 and the Robert A. Welch Foundation (Grant No. E-1337).

Author contributions

Areeda Ayoub conceived the circuit-based approach to engineering thermal correlations, developed the model, performed the analytical and numerical calculations, prepared the figures, and wrote the initial manuscript. Eric Bittner provided scientific guidance, helped refine the research direction, and supervised the project. Both authors contributed to interpreting the results and reviewing and editing the manuscript.

Declaration of AI assistance

During this work, the authors used ChatGPT (OpenAI) and Claude (Anthropic) to assist with discussions of the theoretical model, checking analytical derivations, developing and reviewing numerical and plotting code, verifying references, and drafting and editing the manuscript. All AI-assisted material was critically reviewed and verified by the authors, who take full responsibility for the accuracy and integrity of the reported results and the final manuscript.

Conflicts of Interest

The authors declare no competing financial or non-financial interests.

References

  1. [1]M. Bennett, M. F. Schatz, H. Rockwood, and K. Wiesenfeld, Proc. R. Soc. Lond. A 458, 563 (2002).
  2. [2]J. Pantalone, Am. J. Phys. 70, 999 (2002).
  3. [3]J. Peña Ramirez, L. A. Olvera, H. Nijmeijer, and J. Alvarez, Sci. Rep. 6, 23580 (2016).
  4. [4]G. L. Giorgi, F. Galve, G. Manzano, P. Colet, and R. Zambrini, Phys. Rev. A 85, 052101 (2012).
  5. [5]G. Manzano, F. Galve, G. L. Giorgi, E. Hernández-García, and R. Zambrini, Sci. Rep. 3, 1439 (2013).
  6. [6]J. F. Poyatos, J. I. Cirac, and P. Zoller, Phys. Rev. Lett. 77, 4728 (1996).
  7. [7]F. Verstraete, M. M. Wolf, and J. I. Cirac, Nat. Phys. 5, 633 (2009).
  8. [8]S. Shankar, M. Hartridge, Z. Leghtas, K. M. Sliwa, A. Narla, U. Vool, S. M. Girvin, L. Frunzio, M. Mirrahimi, and M. H. Devoret, Nat. 504, 419 (2013).
  9. [9]E. R. Bittner and B. Tyagi, J. Chem. Phys. 162, 104116 (2025).
  10. [10]A. Metelmann and A. A. Clerk, Phys. Rev. X 5, 021025 (2015).
  11. [11]W. Yu, J. Wang, H. Y. Yuan, and J. Xiao, Phys. Rev. Lett. 123, 227201 (2019).
  12. [12]M. Harder, Y. Yang, B. M. Yao, C. H. Yu, J. W. Rao, Y. S. Gui, R. L. Stamps, and C.-M. Hu, Phys. Rev. Lett. 121, 137203 (2018).
  13. [13]M.-A. Miri and A. Alù, Sci. 363, eaar7709 (2019).
  14. [14]Y. Ashida, Z. Gong, and M. Ueda, Adv. Phys. 69, 249 (2020).
  15. [15]M. Partanen, J. Goetz, K. Y. Tan, K. Kohvakka, V. Sevriuk, R. E. Lake, R. Kokkonie mi, J. Ikonen, D. Hazra, A. Mäkinen, E. Hyyp pä, L. Grönberg, V. Vesterinen, M. Silveri, and M. Möttönen, Phys. Rev. B 100, 134505 (2019).
  16. [16]M. Naghi loo, M. Abbasi, Y. N. Jog le kar, and K. W. Murch, Nat. Phys. 15, 1232 (2019).
  17. [17]P. Peng, W. Cao, C. Shen, W. Qu, J. Wen, L. Jiang, and Y. Xiao, Nat. Phys. 12, 1139 (2016).
  18. [18]B. Peropadre, D. Zueco, F. Wulschner, F. Deppe, A. Marx, R. Gross, and J. J. García-Ripoll, Phys. Rev. B 87, 134504 (2013).
  19. [19]M. Cattaneo and G. S. Paraoanu, Adv. Quantum Technol. 4, 2100054 (2021).
  20. [20]A. Blais, A. L. Grimsmo, S. M. Girvin, and A. Wallraff, Rev. Mod. Phys. 93, 025005 (2021).
  21. [21]R. Simon, Phys. Rev. Lett. 84, 2726 (2000).
  22. [22]G. Vidal and R. F. Werner, Phys. Rev. A 65, 032314 (2002).
  23. [23]C. Weedbrook, S. Pirandola, R. García-Patrón, N. J. C. F. T. C. Ralph, J. H. Shapiro, and S. Lloyd, Rev. Mod. Phys. 84, 621 (2012).
  24. [24]C. D. Wilson, S. Abdullah, N. A. Kurinsky, C. Stanford, L. Cardani, G. D’Imperio, C. Tomei, L. Faoro, L. B. Ioffe, C. H. Liu, A. Ompreczak, B. G. Christensen, J. L. DuBois, and R. McDermott, Nature 594, 369 (2021).
  25. [25]M. McEwen, L. Faoro, K. Arya, A. Dunsworth, T. Huang, S. Kim, B. Burkett, A. Fowler, F. Arute, J. C. Bardin, A. Bengtsson, A. Bilmes, B. B. Buckley, N. Bushnell, Z. Chen, R. Collins, S. Demura, A. R. Derk, C. Erickson, M. Giustina, S. D. Harrington, S. Hong, E. Jeffrey, J. Kelly, P. V. Klimov, F. Kostr it sa, P. Laptev, A. Locharla, X. Mi, K. C. Miao, S. Montazeri, J. Mutus, O. Naaman, M. Neeley, C. Neill, A. Opre mack, C. Quintana, N. Redd, P. Roushan, D. Sank, K. J. Satzinger, V. Shvarts, T. White, Z. J. Yao, P. Yeh, J. Yoo, Y. Chen, V. Smelyansky, J. M. Martinis, H. Neven, A. Megrant, L. Ioffe, and R. Barends, Nat. Phys. 18, 107 (2022).
  26. [26]P. Krantz, M. Kjaergaard, F. Yan, T. P. Orlando, S. Gustavsson, and W. D. Oliver, Appl. Phys. Rev. 6, 021318 (2019).
  27. [27]T. Yamamoto, K. Inomata, M. Watanabe, K. Matsuba, T. Miyazaki, W. D. Oliver, Y. Nakamura, and J. S. Tsai, Appl. Phys. Lett. 93, 042510 (2008).
  28. [28]R. Dassonneville, R. Assouly, T. Peronnin, A. A. Clerk, A. Bienfait, and B. Huard, PRX Quantum 2, 020323 (2021).
  29. [29]C. Eichler, D. Bozyigit, C. Lang, M. Baur, L. Steffen, J. M. Fink, S. Filipp, and A. Wallraff, Phys. Rev. Lett. 107, 113601 (2011).
  30. [30]C. Eichler, D. Bozyigit, and A. Wallraff, Phys. Rev. A 86, 032106 (2012).
  31. [31]C. Macklin, K. O’Brien, D. Hover, M. E. Schwartz, A. Bolkhovsky, X. Zhang, W. D. Oliver, and I. Siddiqi, Sci. 350, 307 (2015).
  32. [32]D. A. Lidar, I. L. Chuang, and K. B. Whaley, Phys. Rev. Lett. 81, 2594 (1998).
  33. [33]E. Fink, F. N. Roch, F. Mallet, M. H. Devoret, and B. Huard, Phys. Rev. Lett. 109, 183901 (2012).

Paper details

vlnc:W_ummj7dwv
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

ChatGPT (OpenAI), Claude (Anthropic)

View statement
Looking for related papers…
Contents