betaFeedback

WorksPhysics and Astronomy

Multilevel Effects and the Limits of the Quantum Rabi Model in Ultrastrong Coupling system

Abstract

The spectrum of artificial atoms coupled to resonators in the ultrastrong coupling regime have been analyzed using the quantum Rabi model over a wide range of coupling strengths. However, the accuracy of models obtained by truncating multilevel artificial atoms to two levels, as well as the correspondence between effective parameters extracted from spectral fitting and the bare parameters of the circuit Hamiltonian, depends on the multilevel structure of the system. In this study, I analyze the spectrum of an ultrastrongly coupled circuit consisting of a multilevel superconducting artificial atom and a resonator using a circuit Hamiltonian, and investigate the effects of truncating higher atomic levels on the low-energy spectrum. Based on Löwdin partitioning, I evaluate the contributions of virtual transitions through higher atomic levels and show that diagonal energy corrections reduce the errors arising from simple two-level truncation over a wide parameter range. I also show that there are parameter regimes in which readjustment of the effective Rabi parameters successfully reproduces the multilevel spectrum, while systematic deviations remain in other regimes. These results demonstrate the necessity of separately evaluating two-level truncation, perturbative corrections, and effective-parameter fitting in ultrastrongly coupled systems, and provide guidance for quantitative spectral analysis of multilevel quantum circuits.

Introduction

In superconducting quantum circuits, the energy structure of artificial atoms can be designed with considerable flexibility, allowing the realization of anharmonicities and coupling strengths that differ from those of systems based on natural atoms. Owing to this high degree of design flexibility, various quantum phenomena have been reported in the ultrastrong coupling (USC) regime, including the verification of selection rules associated with parity symmetry [1], the observation of spectra corresponding to the simultaneous excitation of two atoms by a single photon [2], the realization of the deep-strong coupling regime [3, 4], and strong light–matter coupling using transmon artificial atoms with low anharmonicity [5, 6]. To extend these achievements to quantum computation and quantum error correction utilizing ultrastrong coupling [7, 8, 9, 10], the utilization of ground-state entanglement [11, 12], and quantum sensing [13], it is important to quantitatively understand the energy structure of these systems.

When the low-energy degrees of freedom of an artificial atom can be approximated by a two-level system, the coupled artificial atom–resonator system can be described over a wide range of coupling strengths by the following Rabi Hamiltonian [14, 15, 16]:

H^Rabi/ℏ=εσz+Δσx2+ωra^†a^+gσz(a^†+a^).(1)\hat{H}_{\mathrm{Rabi}}/\hbar= \frac{\varepsilon\sigma_z+\Delta\sigma_x}{2}+\omega_r\hat{a}^{\dagger}\hat{a}+g\sigma_z(\hat{a}^{\dagger}+\hat{a}). \tag*{(1)}

Here, ℏΔ\hbar\Delta is the minimum energy gap of the artificial atom, ℏε\hbar\varepsilon is the energy bias controlled by the external magnetic flux, ωr\omega_r is the angular frequency of the resonator, and gg is the coupling strength expressed in angular frequency units. The transition angular frequency of the artificial atom is given by ωq=ε2+Δ2\omega_q=\sqrt{\varepsilon^2+\Delta^2}.

However, even when the Rabi model accurately reproduces experimental spectra, its effective parameters do not necessarily coincide with the bare parameters defined by the multilevel circuit Hamiltonian. Indeed, previous studies have reported that readjustment of the parameters in the Rabi Hamiltonian is required to reproduce the spectra obtained from circuit models [17, 18].

Recently, Lamberto et al. proposed the Renormalized Quantum Rabi Model (RQRM), which perturbatively incorporates the contributions of higher atomic levels using the Schrieffer–Wolff transformation [19]. In this effective model, renormalization of the atomic transition frequency and coupling strength is accompanied by additional terms, including terms quadratic in the resonator operators. Numerical calculations for a one-dimensional double-well potential and a fluxonium–resonator system demonstrated that the RQRM improves the accuracy of the energy spectra compared with the conventional Rabi model. Furthermore, the importance of extending this approach to flux qubits with multidimensional potentials has been pointed out.

On the other hand, for actual multijunction superconducting circuits, it is important to systematically investigate, based on experimental results, the extent to which higher atomic levels affect the spectra and the parameter regions in which two-level truncation and perturbative effective models remain quantitatively accurate.

In addition, the validity of the two-level approximation in the ultrastrong coupling regime has also been discussed from the perspective of gauge invariance [20, 21]. Even when Hamiltonians are related by unitary transformations and are equivalent in the complete Hilbert space, simply truncating the matter subsystem to two levels generally breaks the equivalence between gauges. To address this issue, effective Hamiltonians that preserve consistency between gauges under the two-level approximation have been proposed. However, gauge consistency and quantitative accuracy in reproducing the spectra of multilevel systems must be examined separately. In particular, the contributions of virtual transitions through truncated higher atomic levels and the range over which these contributions can be incorporated perturbatively depend on the circuit representation and the parameter regime.

Circuit

The measured circuit is shown in Figure 1. An artificial atom is galvanically coupled to a lumped-element LC resonator through a Josephson junction. The resonator is capacitively coupled to a transmission line, and the transmission coefficient S21S_{21} was measured using a vector network analyzer (VNA).

Schematic of the measured circuit, scanning electron microscope image, and flux-dependent resonator spectra

Figure 1. (a) Schematic of the measured circuit. A capacitively shunted flux qubit (artificial atom) is galvanically coupled to an LC resonator through the β\beta junction. The resonator is capacitively coupled to a transmission line for microwave measurements. (b) Scanning electron microscope (SEM) image of the artificial atom. The Josephson junctions are highlighted in red. An on-chip flux-bias line controls the external magnetic flux applied to the artificial atom. (c) Flux-dependent resonator spectrum measured through the microwave transmission coefficient S21S_{21} using a vector network analyzer. The white dotted line shows the fit obtained from Eq. (19), while the black dotted line shows the semiclassical result without the phenomenological frequency correction. (d) The two phenomenological frequency-shift contributions δω0/2π\delta\omega_0/2\pi and δω1/2π\delta\omega_1/2\pi used in panel (c). (e) High-resolution spectrum near half a flux quantum. The dotted lines indicate transition frequencies calculated from the circuit Hamiltonian in Eq. (5). The fitted circuit parameters are EJ/h=236E_J/h = 236 GHz, Ec/h=2.92E_c/h = 2.92 GHz, α=0.57\alpha= 0.57, β=1.37\beta= 1.37, γ=3.57\gamma= 3.57, Lr=1.59L_r = 1.59 nH, and Cr=414C_r = 414 fF.

To reduce the anharmonicity of the artificial atom, a capacitively shunted flux qubit is employed [23, 24]. The external magnetic flux is controlled by an on-chip flux-bias line.

I denote the Josephson energy of junction aa by EJE_J and assume that junction bb has the same size as junction aa. The potential energy of the artificial atom is expressed as

U^J=−EJ[βcos⁡φ^β+cos⁡φ^a+cos⁡(φex−φ^a−φ^α−φ^β)+αcos⁡φ^α]+ELφ^β2(2)\hat{U}_J = -E_J \left[\beta\cos\hat{\varphi}_\beta+ \cos\hat{\varphi}_a + \cos\left(\varphi_{\mathrm{ex}} - \hat{\varphi}_a - \hat{\varphi}_\alpha- \hat{\varphi}_\beta\right) + \alpha\cos\hat{\varphi}_\alpha\right] + E_L\hat{\varphi}_\beta^2 \tag*{(2)}

where φ^m\hat{\varphi}_m (m∈{a,α,β}m \in\{a,\alpha,\beta\}) represents the phase difference operator across each junction, and α\alpha and β\beta are the ratios of the Josephson energies of the corresponding junctions to that of junction aa. Here, EL=(Φ0/2π)2/(2Lr)E_L = (\Phi_0/2\pi)^2/(2L_r), where Φ0\Phi_0 is the magnetic flux quantum.

Using the charging energy EcE_c of junction aa, the Hamiltonian of the artificial atom is given by

H^A=4Ecn^TM−1n^+U^J,(3)\hat{H}_A = 4E_c\hat{\mathbf{n}}^{\mathsf{T}}M^{-1}\hat{\mathbf{n}} + \hat{U}_J, \tag*{(3)}

where MM is the normalized capacitance matrix (mass

In this study, starting from circuit parameters determined from experimental spectra, I analyze the effects of truncating higher atomic levels on the low-energy spectrum using Löwdin partitioning [22]. I compare a simple two-level model, diagonal energy corrections, and a second-order effective Hamiltonian including off-diagonal matrix elements. I further investigate the accuracy obtained by refitting the effective parameters of the Rabi model. By varying the circuit parameters and coupling strength, I systematically evaluate the errors of these approximations. In addition, I analyze the resonator spectrum over a wide range of magnetic flux biases using a semiclassical model based on Josephson inductance combined with a phenomenological frequency correction. matrix),

M=(2111γ+1111β+1)(4)M=\begin{pmatrix}2&1&1\\1&\gamma+1&1\\1&1&\beta+1\end{pmatrix} \tag*{(4)}

and n^\hat{n} is the vector of Cooper-pair number operators conjugate to the phase variables [25]. The parameter γ\gamma is a normalized parameter associated with the shunt capacitance. In the present circuit, γ=α\gamma=\alpha corresponds to the absence of an additional shunt capacitor.

The eigenstates ∣ψi⟩|\psi_i\rangle and eigenangular frequencies ωi\omega_i of the artificial atom are obtained numerically from H^A∣ψi⟩=ℏωi∣ψi⟩\hat{H}_A|\psi_i\rangle=\hbar\omega_i|\psi_i\rangle. The Hamiltonian of the coupled artificial atom–resonator system is then given by

H^AR/ℏ=∑iωi∣ψi⟩⟨ψi∣−∑i,jgij∣ψi⟩⟨ψj∣(a^†+a^)+ωra^†a^,(5)\hat{H}_{\mathrm{AR}}/\hbar=\sum_i\omega_i|\psi_i\rangle\langle\psi_i|-\sum_{i,j}g_{ij}|\psi_i\rangle\langle\psi_j|(\hat{a}^{\dagger}+\hat{a})+\omega_r\hat{a}^{\dagger}\hat{a}, \tag*{(5)}

where the coupling matrix elements are expressed as ℏgij=Izpf(Φ0/2π)⟨ψi∣φ^β∣ψj⟩\hbar g_{ij}=I_{\mathrm{zpf}}(\Phi_0/2\pi)\langle\psi_i|\hat{\varphi}_{\beta}|\psi_j\rangle with Izpf=ℏωr/2LrI_{\mathrm{zpf}}=\sqrt{\hbar\omega_r/2L_r} [25]. Here, IzpfI_{\mathrm{zpf}} is the zero-point fluctuation current of the resonator. In the following, the transition angular frequency from the ground state to the jjth excited state of the isolated artificial atom is denoted by ω0j(A)=ωj−ω0\omega_{0j}^{(A)}=\omega_j-\omega_0. I also define the normalized external magnetic flux as f=φex/(2π)f=\varphi_{\mathrm{ex}}/(2\pi).

Figure 1(c) shows the resonator spectrum measured over one period of the magnetic flux bias. Except in the immediate vicinity of f=0.5f=0.5, the observed response predominantly corresponds to the resonator mode. Figure 1(e) shows the high-resolution spectrum near the flux bias where the transition frequency between the ground and first excited states of the artificial atom matches the resonator frequency. The fitting results obtained using Eq. (5) are indicated by dotted lines. The obtained circuit parameters are EJ/h=236E_J/h=236 GHz, Ec/h=2.92E_c/h=2.92 GHz, α=0.57\alpha=0.57, β=1.37\beta=1.37, γ=3.57\gamma=3.57, Lr=1.59L_r=1.59 nH, and Cr=414C_r=414 fF.

Two-level truncation

In this section, I evaluate the spectral errors arising from truncating the artificial atom to two levels within the same circuit representation and investigate the contributions of virtual transitions through higher atomic levels. Here, I distinguish between two-level truncation with fixed circuit parameters and refitting of the effective Rabi parameters.

Löwdin partitioning and second-order corrections

Let H^≡H^AR\hat{H}\equiv\hat{H}_{\mathrm{AR}}, I define the projection operators that retain the lowest two levels of the artificial atom as P^=(∣ψ0⟩⟨ψ0∣+∣ψ1⟩⟨ψ1∣)⊗I^r\hat{P}=(|\psi_0\rangle\langle\psi_0|+|\psi_1\rangle\langle\psi_1|)\otimes\hat{I}_r and Q^=I^−P^\hat{Q}=\hat{I}-\hat{P}, where I^r\hat{I}_r is the identity operator in the resonator Hilbert space. The Hamiltonian can be written in block form as

H^=(H^PPH^PQH^QPH^QQ),(6)\hat{H}=\begin{pmatrix}\hat{H}_{PP}&\hat{H}_{PQ}\\\hat{H}_{QP}&\hat{H}_{QQ}\end{pmatrix}, \tag*{(6)}

where H^PP=P^H^P^\hat{H}_{PP}=\hat{P}\hat{H}\hat{P}, H^PQ=P^H^Q^\hat{H}_{PQ}=\hat{P}\hat{H}\hat{Q}, H^QP=H^PQ†\hat{H}_{QP}=\hat{H}_{PQ}^{\dagger}, and H^QQ=Q^H^Q^\hat{H}_{QQ}=\hat{Q}\hat{H}\hat{Q}. From Eq. (5),

H^PQ=−ℏ∑i=01∑j≥2gij∣ψi⟩⟨ψj∣⊗(a^†+a^).(7)\hat{H}_{PQ}=-\hbar\sum_{i=0}^{1}\sum_{j\geq2}g_{ij}|\psi_i\rangle\langle\psi_j|\otimes(\hat{a}^{\dagger}+\hat{a}). \tag*{(7)}

By eliminating the QQ space through Löwdin partitioning, I obtain the energy-dependent effective Hamiltonian

H^eff(E)=H^PP+H^PQ(E−H^QQ)−1H^QP,(8)\hat{H}_{\mathrm{eff}}(E)=\hat{H}_{PP}+\hat{H}_{PQ}(E-\hat{H}_{QQ})^{-1}\hat{H}_{QP}, \tag*{(8)}

[22, 26, 27, 28]. When (E−H^QQ)−1(E-\hat{H}_{QQ})^{-1} exists, solving H^eff(E)∣χ⟩=E∣χ⟩\hat{H}_{\mathrm{eff}}(E)|\chi\rangle=E|\chi\rangle self-consistently with respect to EE is equivalent to solving the original eigenvalue problem for eigenstates with a nonzero projection onto the PP space.

In the following, I discuss the system using second-order approximations. First, I diagonalize the Hamiltonian in the PP space as H^PP∣un⟩=ϵn∣un⟩\hat{H}_{PP}|u_n\rangle=\epsilon_n|u_n\rangle. I define the matrix elements of the self-energy, including the internal interactions within the QQ space, as

Σnm(E)=⟨un∣H^PQ(E−H^QQ)−1H^QP∣um⟩.(9)\Sigma_{nm}(E)=\langle u_n|\hat{H}_{PQ}(E-\hat{H}_{QQ})^{-1}\hat{H}_{QP}|u_m\rangle. \tag*{(9)}

The diagonal correction evaluated at the unperturbed energy ϵn\epsilon_n is given by

Δϵn(2)=Σnn(ϵn)=⟨un∣H^PQ(ϵn−H^QQ)−1H^QP∣un⟩.(10)\Delta\epsilon_n^{(2)}=\Sigma_{nn}(\epsilon_n)=\langle u_n|\hat{H}_{PQ}(\epsilon_n-\hat{H}_{QQ})^{-1}\hat{H}_{QP}|u_n\rangle. \tag*{(10)}

Here, “second order” refers to the order in the coupling H^PQ+H^QP\hat{H}_{PQ}+\hat{H}_{QP} between the PP and QQ spaces, with H^PP\hat{H}_{PP} and H^QQ\hat{H}_{QQ} held fixed.

The Hermitian second-order effective Hamiltonian including off-diagonal corrections is constructed as

(Heff(2))nm=ϵnδnm+12[Σnm(ϵn)+Σnm(ϵm)](11)\left(H_{\mathrm{eff}}^{(2)}\right)_{nm}=\epsilon_n\delta_{nm}+\frac{1}{2}\left[\Sigma_{nm}(\epsilon_n)+\Sigma_{nm}(\epsilon_m)\right] \tag*{(11)}

[28, 29]. Using H^QQ∣vk⟩=Ek(Q)∣vk⟩\hat{H}_{QQ}|v_k\rangle=E_k^{(Q)}|v_k\rangle and Wnk=⟨un∣H^PQ∣vk⟩W_{nk}=\langle u_n|\hat{H}_{PQ}|v_k\rangle, the correction term is expressed as

ΔHnm(2)=12∑kWnkWmk∗[1ϵn−Ek(Q)+1ϵm−Ek(Q)].(12)\Delta H_{nm}^{(2)}=\frac{1}{2}\sum_k W_{nk}W_{mk}^{*}\left[\frac{1}{\epsilon_n-E_k^{(Q)}}+\frac{1}{\epsilon_m-E_k^{(Q)}}\right]. \tag*{(12)}

For n=mn=m, this expression reduces to Eq. (10).

Equation (8) has the same projection form as the resolvent method used by Lamberto et al. to analyze higher-order corrections [19]. Both the Schrieffer–Wolff transformation and Löwdin partitioning aim to eliminate the degrees of freedom associated with higher atomic levels and incorporate their contributions into a low-energy effective Hamiltonian. However, the specific approximation methods differ. In the previous study, an effective Hamiltonian containing renormalized parameters and additional operator terms was constructed analytically through a perturbative expansion in the coupling to higher atomic levels. In contrast, in the present study, I divide the circuit Hamiltonian into a low-energy subspace PP and a higher-energy subspace QQ, and numerically evaluate the resolvent of the full H^QQ\hat{H}_{QQ}, including the internal interactions within the QQ space. This allows me to directly compare the accuracy of diagonal energy corrections for individual eigenstates with that of a second-order effective Hamiltonian including off-diagonal matrix elements.

In this paper, I refer to the method of calculating ϵn+Δϵn(2)\epsilon_n+\Delta\epsilon_n^{(2)} based on Eq. (10) as ”Löwdin diagonal,” and the method of rediagonalizing the second-order effective Hamiltonian constructed from Eq. (11) as ”Löwdin full.”

When the diagonal energy corrections Δϵn(2)\Delta\epsilon_n^{(2)} are used, the transition angular frequency for n>mn>m is given by

ω~nm=ωnm(P)+Δϵn(2)−Δϵm(2)ℏ,(13)\widetilde{\omega}_{nm}=\omega_{nm}^{(P)}+\frac{\Delta\epsilon_n^{(2)}-\Delta\epsilon_m^{(2)}}{\hbar}, \tag*{(13)}

where

ωnm(P)=ϵn−ϵmℏ.(14)\omega_{nm}^{(P)}=\frac{\epsilon_n-\epsilon_m}{\hbar}. \tag*{(14)}

When the energy ϵn\epsilon_n under consideration lies below all eigenvalues of the QQ-space Hamiltonian, Δϵn(2)≤0\Delta\epsilon_n^{(2)}\leq0. This means that virtual transitions through higher atomic levels lower the energy of the corresponding state. Furthermore, independently of perturbation theory, the eigenvalues obtained by projecting a given Hamiltonian onto a subspace provide upper bounds to the corresponding eigenvalues of the full Hamiltonian when both sets of eigenvalues are arranged in ascending order. However, because the correction to a transition frequency is determined by the difference between two energy corrections, its sign is not generally fixed. In Figure 2, simple two-level truncation tends to overestimate the transition frequencies. This indicates that the excited states under comparison experience larger downward energy shifts than the reference state.

Comparison of transition spectra

Figure 2. Comparison of transition spectra calculated using the full multilevel circuit Hamiltonian, the two-level truncated model, and the second-order Löwdin approximations. “Löwdin diagonal” denotes state-specific diagonal energy corrections evaluated using the full QQ-space resolvent Eq. (10), whereas “Löwdin full” denotes rediagonalization of the symmetrized second-order effective Hamiltonian including off-diagonal matrix elements Eq. (11). (a) Circuit parameters fitted to the experimental spectrum in Figure 1(e). (b) Same as (a), but with γ=α\gamma=\alpha, corresponding to the absence of the additional shunt capacitance. (c) Same as (a), but with all atom–resonator coupling matrix elements gijg_{ij} reduced by a factor of eight.

Comparison of transition spectra

Figure 2 shows a comparison of the transition spectra calculated using the approximate models and the full multilevel circuit Hamiltonian. The reference spectra were obtained by numerical diagonalization of Eq. (5). In the calculations, the Cooper-pair number for each junction was taken from −8-8 to 88, and 100 eigenstates of the artificial atom and 25 Fock states of the resonator were retained [25]. The 2-level model represents the result obtained by projecting the artificial atom onto its lowest two levels without changing the circuit parameters. I compare this result with the Löwdin diagonal and Löwdin full approximations defined in the previous subsection.

Figure 2(a) shows the results obtained using the fitted parameters of the measured circuit. Simple two-level truncation produces large spectral deviations from the full multilevel circuit Hamiltonian. The Löwdin diagonal approximation improves the agreement in the low-energy region, whereas substantial deviations remain at higher energies. Even with the Löwdin full approximation, the agreement cannot be uniformly improved over the entire spectrum considered. Here, the coupling strength is g01/2π=5.45g_{01}/2\pi=5.45 GHz, and the atomic transition frequency is ω02(A)/2π=16.22\omega_{02}^{(A)}/2\pi=16.22 GHz at f=0.5f=0.5.

Figure 2(b) corresponds to the case where the additional shunt capacitor is removed by setting γ=α\gamma=\alpha. At f=0.5f=0.5, the atomic transition frequency is ω02(A)/2π=30.26\omega_{02}^{(\mathrm{A})}/2\pi=30.26 GHz, and the coupling strength is g01/2π=4.81g_{01}/2\pi=4.81 GHz. Although the energy separation between the lowest two atomic levels and the higher levels is large, substantial spectral errors still appear in the Löwdin full approximation. This indicates that the accuracy of the second-order approximation employed here cannot be determined solely from the bare atomic level separations.

Figure Figure 2(c) shows the results obtained using the same atomic parameters as in Figure 2(a), but with all coupling matrix elements reduced according to gij→gij/8g_{ij}\to g_{ij}/8. In this case, g01/2π=0.68g_{01}/2\pi=0.68 GHz at f=0.5f=0.5, and the system remains in the ultrastrong coupling regime. The atomic level structure is unchanged from that in panel (a), with ω02(A)/(2π)=16.22\omega_{02}^{(\mathrm{A})}/(2\pi)=16.22 GHz at f=0.5f=0.5. Reducing the coupling strength also decreases the errors arising from simple two-level truncation, and the spectra obtained with the Löwdin corrections agree well with those of the full multilevel circuit Hamiltonian. Therefore, the contributions of higher atomic levels are small in this regime and can be incorporated perturbatively.

Fitting effective Rabi parameters

Figure Figure 3 shows the results of fitting the spectra of the full multilevel circuit Hamiltonian with the Rabi Hamiltonian in two parameter regimes. Unlike simple two-level truncation, the effective parameters of the Rabi model are readjusted in this analysis. Furthermore, I introduce a phenomenological correction describing the magnetic-flux dependence of the resonator frequency ωr′(f)=ωr+Aε(f)\omega_r'(f)=\omega_r+A\varepsilon(f) following [17].

Comparison of transition spectra calculated using the full multilevel circuit Hamiltonian and the Rabi model with fitted effective parameters

Figure 3. Comparison of transition spectra calculated using the full multilevel circuit Hamiltonian and the Rabi model with fitted effective parameters. The fitted Rabi model also includes the phenomenological flux dependence of the resonator frequency. (a) Parameters corresponding to the measured circuit. (b) Modified circuit parameters with γ=α\gamma=\alpha.

For the measured circuit parameters shown in Figure 3(a), systematic deviations remain between the Rabi model and the full circuit Hamiltonian for the parameters and fitting range adopted in this study. The parameters obtained from the fitting are ε/[2π(f−0.5)]=1.58×103\varepsilon/[2\pi(f-0.5)]=1.58\times10^3 GHz, Δ/2π=0.22\Delta/2\pi=0.22 GHz, ωr/2π=5.30\omega_r/2\pi=5.30 GHz, g/2π=6.80g/2\pi=6.80 GHz, and A=2.48×10−4A=2.48\times10^{-4}. Here, ε/[2π(f−0.5)]\varepsilon/[2\pi(f-0.5)] represents the linear slope of ε/2π\varepsilon/2\pi with respect to the magnetic flux bias.

On the other hand, Figure 3(b) shows the case of γ=α\gamma=\alpha. Under this condition, the higher atomic levels are relatively well separated from the lowest two levels. Although some residual deviations remain, the modified Rabi model reproduces the spectrum within the displayed range with good accuracy. The obtained parameters are ε/[2π(f−0.5)]=1.44×103\varepsilon/[2\pi(f-0.5)]=1.44\times10^3 GHz, Δ/2π=5.01\Delta/2\pi=5.01 GHz, ωr/2π=5.23\omega_r/2\pi=5.23 GHz, g/2π=5.41g/2\pi=5.41 GHz, and A=5.52×10−4A=5.52\times10^{-4}. In both cases, the effective resonator frequency of the Rabi model is lower than the bare resonator frequency ωr/2π=6.21\omega_r/2\pi=6.21 GHz used in the simple two-level truncation. Furthermore, the effective coupling strength g/2π=6.80g/2\pi=6.80 GHz in Figure 3(a) is larger than g01/2π=5.45g_{01}/2\pi=5.45 GHz obtained from the circuit Hamiltonian. However, the Rabi-model parameter gg and the coupling matrix element g01g_{01} between specific atomic eigenstates are defined differently.

The bare parameters of the circuit Hamiltonian are defined within a representation that specifies how the system is partitioned into the artificial atom and the resonator. In contrast, the fitted parameters of the Rabi model are effective quantities introduced to reproduce the spectrum of a multilevel system using a reduced set of degrees of freedom, and they do not necessarily coincide with the bare parameters. When the representation and projection are fixed, the corrections arising from the elimination of higher atomic levels can be clearly defined. However, parameter differences obtained from different representations or fitting conditions cannot be regarded as unique, representation-independent renormalization quantities.

Parameter dependence of the resonator-frequency error

Figure Figure 4 shows the relative errors in the effective resonator frequency calculated using each approximate model as functions of the circuit parameter γ\gamma and the coupling scale cc. The parameter cc is a numerical scaling factor applied to all atom–resonator coupling matrix elements calculated for each γ\gamma, according to gij→cgijg_{ij}\to cg_{ij}. Here, c=1c=1 corresponds to the unscaled coupling at each value of γ\gamma, and the parameters of the measured circuit correspond to (c,γ)=(1,3.57)(c,\gamma)=(1,3.57).

Relative errors in the effective resonator frequency for the two-level, Löwdin diagonal, and Löwdin full approximations

Figure 4. Relative errors in the effective resonator frequency, identified with the ∣0⟩→∣2⟩|0\rangle\to|2\rangle transition frequency ω02\omega_{02} throughout the parameter range considered, with respect to the full multilevel circuit Hamiltonian. (a) Two-level truncated model. (b) State-specific diagonal energy corrections using the full QQ-space resolvent (Löwdin diagonal). (c) Rediagonalization of the symmetrized second-order effective Hamiltonian, including off-diagonal matrix elements (Löwdin full). The horizontal axis is the coupling scale cc, which uniformly changes all atom–resonator coupling matrix elements as gij→cgijg_{ij} \to c g_{ij}. The vertical axis is the circuit parameter γ\gamma. Relative errors are defined in Eq. (15), and values of 20% or more are saturated at the upper limit of the common color scale.

Within the parameter range considered in this study, the transition corresponding to the resonator mode can always be identified as the transition ω02\omega_{02} from the ground state to the second excited state of the coupled system. I therefore define the relative error in the effective resonator frequency ω02(M)\omega_{02}^{(M)} obtained from model MM as

ηM=100∣ω02(M)−ω02(full)ω02(full)∣[%],(15)\eta_M=100\left|\frac{\omega_{02}^{(M)}-\omega_{02}^{(\mathrm{full})}}{\omega_{02}^{(\mathrm{full})}}\right|[\%], \tag*{(15)}

where MM denotes one of the 2-level, Löwdin diagonal, or Löwdin full models, and ω02(full)\omega_{02}^{(\mathrm{full})} is the corresponding transition angular frequency obtained from the full multilevel circuit Hamiltonian. In Figure 4, values of ηM≥20%\eta_M \ge20\% are displayed using the same upper-limit color.

In the simple two-level truncation shown in Figure 4(a), the relative error in the resonator mode generally increases with increasing coupling scale cc. In contrast, the Löwdin diagonal approximation in Figure 4(b) substantially reduces the error over a wide parameter range. This result indicates that the energy corrections associated with virtual transitions through higher atomic levels account for a major part of the deviations in the resonator-mode transition frequency caused by simple two-level truncation.

In the Löwdin full approximation shown in Figure 4(c), large errors appear as the coupling scale increases, and the regions in which these errors occur vary nonmonotonically with γ\gamma. This result shows that rediagonalization of the energy-independent second-order effective Hamiltonian employed here does not necessarily guarantee improved accuracy over the entire parameter range investigated. The difference between the Löwdin diagonal and full approximations reflects the distinction between state-specific diagonal corrections and rediagonalization of the second-order matrix including off-diagonal corrections. However, this result does not imply that off-diagonal physical processes are unnecessary. In general, when ϵn\epsilon_n approaches an eigenvalue of H^QQ\hat{H}_{QQ}, the energy denominators in Eq. (9) can become small, potentially leading to large self-energy contributions. Furthermore, state mixing caused by rediagonalization of the correction matrix and the range of PP-space eigenstates retained in the calculation may also affect the accuracy of the approximation. These contributions cannot be uniquely separated based on Figure 4 alone. Therefore, the results presented here demonstrate the limitations of the second-order approximation employed in this study.

Interestingly, Figure 4(a) also shows parameter regions in which the relative error of the two-level truncation for the resonator mode decreases as γ\gamma increases. However, varying γ\gamma does not simply change a single measure of anharmonicity; it simultaneously modifies the energy-level separations, wave functions, and coupling matrix elements of the artificial atom. Moreover, corrections to transition frequencies depend on the difference between the energy corrections of the initial and final states. Therefore, this tendency cannot be attributed solely to changes in the shunt capacitance or anharmonicity, and the multilevel structure, including the coupling to higher atomic levels, must be taken into account.

Previous studies have discussed the classification of spectral regimes within the Rabi model, including the regime where the Jaynes–Cummings model is valid, the regime where counter-rotating terms can be treated perturbatively, and regimes beyond these approximations [30, 31]. In the present study, I have evaluated the accuracy of truncating a multilevel artificial atom to two levels and perturbatively incorporating the contributions of the eliminated atomic levels, thereby identifying a nonperturbative regime associated with the truncation of multilevel atomic degrees of freedom. These results are also consistent with previous work showing that incorporating the contributions of higher atomic levels into a low-energy effective Hamiltonian can improve the accuracy of the two-level approximation [19].

On the other hand, I find that the accuracy of the approximation can differ substantially between diagonal energy corrections applied to individual eigenstates and rediagonalization of a second-order effective Hamiltonian including off-diagonal matrix elements. This difference suggests that the treatment of energy dependence and state mixing within the low-energy subspace is important when incorporating the contributions of higher atomic levels.

Resonator spectrum over a wide flux range

So far, I have analyzed the effects of truncating higher atomic levels on the low-energy spectrum near half a flux quantum. On the other hand, the resonator spectrum shown in Figure 1(c) varies over a wide range of magnetic flux biases. Except in the immediate vicinity of f=0.5f = 0.5, this response predominantly corresponds to the resonator mode and reflects the magnetic-flux dependence of the effective inductance of the artificial atom, which shares a superconducting loop with the resonator.

In principle, the blocks of the circuit Hamiltonian can be constructed at each magnetic flux bias, and the corrections obtained from Löwdin partitioning can be evaluated in the same manner as in the previous section. However, over a wide range of magnetic flux biases, the shape of the potential changes substantially from a double-well to a single-well structure. With the expansion in a harmonic-oscillator or plane-wave basis employed in this study, a large Hilbert space is required to achieve sufficient convergence of the eigenvalues over the entire range, resulting in increased computational cost [25].

I therefore introduce a phenomenological frequency correction into a semiclassical model based on Josephson inductance. This approach provides an efficient description of the experimental spectrum over a wide magnetic flux range and serves as a means of estimating the initial parameters for fitting the multilevel circuit Hamiltonian. Furthermore, independently of the role of the additional phenomenological coefficients, examining whether the spectra in different magnetic flux regions can be described using a common set of circuit parameters provides supplementary information for assessing the consistency of the parameters.

Except in the immediate vicinity of f=0.5f = 0.5, when the probe angular frequency ωp\omega_{\mathrm{p}} is sufficiently lower than the excitation frequencies of the artificial atom, I can consider a response in which atomic excitations are suppressed. In the following, I further adopt a semiclassical approximation in which the phases across the Josephson junctions are treated near the minima of the potential, and describe the response of the artificial atom in terms of Josephson inductance [32, 33].

Defining LJ=(Φ0/2π)2/EJL_{\mathrm{J}} = (\Phi_0/2\pi)^2/E_{\mathrm{J}}, the inductance of junction aa is given by LJ/cos⁡φaL_{\mathrm{J}}/\cos\varphi_a. The effective inductance seen across the β\beta junction is evaluated using the following expression:

LA=LJβcos⁡φβ+2cos⁡φa+αcos⁡φαcos⁡φacos⁡φα.(16)L_{\mathrm{A}} = \frac{L_{\mathrm{J}}}{\beta\cos\varphi_\beta+ \frac{2\cos\varphi_a + \alpha\cos\varphi_\alpha}{\cos\varphi_a \cos\varphi_\alpha}}. \tag*{(16)}

To provide a physical motivation for the spectral corrections arising from higher atomic levels, I expand (10) in terms of the eigenstates of the QQ space:

Δϵn(2)=−∑k∣⟨vk∣H^QP∣un⟩∣2Ek(Q)−ϵn.(17)\Delta\epsilon_n^{(2)} = -\sum_k \frac{\left|\langle v_k|\hat{H}_{QP}|u_n\rangle\right|^2}{E_k^{(Q)} - \epsilon_n}. \tag*{(17)}

The denominator represents the energy difference between the state of interest in the PP space and an intermediate state in the QQ space. When Ek(Q)>ϵnE_k^{(Q)} > \epsilon_n, this coupling produces a negative energy correction due to level repulsion. This is consistent with the concept of virtual transitions in the dispersive regime [34].

In this section, rather than rigorously deriving the following correction function from (17), I introduce a phenomenological function to describe the downward shift and magnetic-flux dependence of the measured resonator spectrum:

δω(f)=∑i=01δωi(f),\delta\omega(f) = \sum_{i=0}^{1} \delta\omega_i(f),
δωi(f)2π=−Ai2+Bi21+Ci2(f−0.5)2.(18)\frac{\delta\omega_i(f)}{2\pi} = -\sqrt{A_i^2 + \frac{B_i^2}{1 + C_i^2(f - 0.5)^2}}. \tag*{(18)}

Here, AiA_i and BiB_i have dimensions of ordinary frequency, whereas CiC_i is dimensionless. The corrected resonator angular frequency is defined as

ωres(ph)(f)=1LA(f)Cr+δω(f).(19)\omega_{\mathrm{res}}^{(\mathrm{ph})}(f) = \frac{1}{\sqrt{L_{\mathrm{A}}(f)C_{\mathrm{r}}}} + \delta\omega(f). \tag*{(19)}

The coefficients obtained by fitting (18) are A0=17.94A_0 = 17.94 MHz, A1=0.63A_1 = 0.63 MHz, B0=48.51B_0 = 48.51 MHz, B1=12.79B_1 = 12.79 MHz, C0=6.04C_0 = 6.04, and C1=52.59C_1 = 52.59. Since C1>C0C_1 > C_0, the contribution with i=1i = 1 exhibits a magnetic-flux dependence that is more localized near half a flux quantum than that with i=0i = 0.

The white dotted line in Figure 1(c) shows the fitting result including the phenomenological correction, whereas the black dotted line shows the semiclassical result without this correction. Figure 1(d) shows the two correction terms, one of which exhibits a strong magnetic-flux dependence near f=0.5f = 0.5, while the other exhibits a weaker magnetic-flux dependence over the entire range. The circuit parameters other than the phenomenological correction coefficients are the same as those used in the analysis of Figure 1(e).

Conclusion

In this study, I have analyzed the spectra of a multilevel superconducting artificial atom coupled to a resonator in the ultrastrong coupling regime, both experimentally and theoretically, and investigated the effects of truncating higher atomic levels on the low-energy spectrum. Using a multilevel circuit Hamiltonian consistent with the measured spectrum as a reference, I compared simple two-level truncation with corrections based on Löwdin partitioning. I showed that energy shifts arising from virtual transitions through higher atomic levels are an important source of the spectral deviations between the two-level model and the multilevel circuit model.

Furthermore, I systematically varied the coupling strength and circuit parameters and evaluated the relative error in ω02\omega_{02}, corresponding to the resonator mode. Even in regions where simple two-level truncation produces large errors, diagonal energy corrections based on Löwdin partitioning substantially reduce the errors over a wide parameter range. On the other hand, rediagonalization of the second-order effective Hamiltonian including off-diagonal matrix elements does not necessarily improve the accuracy. This result indicates the range of applicability of the second-order approximation employed here, rather than a limitation of exact Löwdin partitioning. The results also suggest that the accuracy of these approximations cannot be determined solely from the atomic level separations or anharmonicity, but also depends on the coupling matrix elements to higher atomic levels and the differences between the energy corrections of the states involved in the transitions.

When the effective parameters of the Rabi model are refitted, the spectrum of the multilevel circuit Hamiltonian can be reproduced with good accuracy under certain conditions, whereas systematic residual deviations remain for parameters corresponding to the measured circuit. This indicates not only that the bare circuit parameters and effective Rabi parameters cannot be directly identified with each other, but also that some multilevel effects cannot be absorbed solely by readjusting constant parameters. Furthermore, the experimental resonator spectrum over a wide range of magnetic flux biases was reproduced by combining a semiclassical model based on Josephson inductance with a phenomenological frequency correction.

These results demonstrate the importance of separately evaluating the accuracy of two-level truncation, the perturbative contributions of virtual transitions through higher atomic levels, and refitting to an effective Rabi model in the quantitative modeling of ultrastrongly coupled quantum circuits. These comparisons provide guidance for understanding the conditions under which low-dimensional effective models are valid and their limitations in ultrastrongly coupled systems containing multilevel artificial atoms.

References

Acknowledgments

I thank H. Mukai, R. Stassi, S. Masuda, K. Mizuno, and T. Aoki for helpful discussions and comments. This work was supported by JSPS KAKENHI (Grant Nos. JP22K21294 and JP23K13048), project JPNP16007, and the MATSUO FOUNDATION.

References

  1. [1]S.-P. Wang, A. Ridolfo, T. Li, S. Savasta, F. Nori, Y. Nakamura, and J. Q. You, Probing the symmetry breaking of a light–matter system by an ancillary qubit, Nature Communications 14, 4397 (2023).DOI
  2. [2]A. Tomonaga, R. Stassi, H. Mukai, F. Nori, F. Yoshihara, and J. S. Tsai, Spectral properties of two superconducting artificial atoms coupled to a resonator in the ultrastrong coupling regime, 16, 5294.arxiv.org/abs/2307.15437
  3. [3]F. Yoshihara, T. Fuse, S. Ashhab, K. Kakuyanagi, S. Saito, and K. Semba, Superconducting qubit–oscillator circuit beyond the ultrastrong-coupling regime, Nature Physics 13, 44 (2017).DOI
  4. [4]P. Forn-Díaz, J. J. García-Ripoll, B. Peropadre, J.-L. Orgiazzi, M. A. Tarasov, R. Beljansky, C. M. Wilson, and A. Lupascu, Ultrastrong coupling of a single artificial atom to an electromagnetic continuum in the nonperturbative regime, Nature Physics 13, 39 (2017).
  5. [5]S. J. Bosman, M. F. Gely, V. Singh, A. Bruno, D. Bothner, and G. A. Steele, Multi-mode ultra-strong coupling in circuit quantum electrodynamics, 3, 46.
  6. [6]Y. Ye, J. B. Kline, A. Yen, G. Cunningham, M. Tan, A. Zang, M. Gingras, B. M. Niedzielski, H. Stickler, K. Serniak, M. E. Schwartz, and K. P. O’Brien, Near-ultrastrong nonlinear light-matter coupling in superconducting circuits, 16, 3799.
  7. [7]T. H. Kyaw, D. A. Herrera-Martí, E. Solano, G. Romero, and L.-C. Kwek, Creation of quantum error correcting codes in the ultrastrong coupling regime, Physical Review B 91, 064503 (2015).DOI
  8. [8]Y. Wang, J. Zhang, C. Wu, J. Q. You, and G. Romero, Holonomic quantum computation in the ultrastrong-coupling regime of circuit QED, Physical Review A 94, 012328 (2016).DOI
  9. [9]R. Stassi, M. Cirio, and F. Nori, Scalable quantum computer with superconducting circuits in the ultrastrong coupling regime, npj Quantum Information 6, 1 (2020).
  10. [10]Y.-H. Chen, W. Qin, R. Stassi, X. Wang, and F. Nori, Fast bosonic-code holonomic quantum computation with ultrastrong light-matter coupling, Physical Review Research 3, 033275 (2021).DOI
  11. [11]T. Jaako, Z.-L. Xiang, J. J. Garcia-Ripoll, and P. Rabl, Ultrastrong-coupling phenomena beyond the dicke model, 94, 033850.
  12. [12]F. Armata, G. Calajo, T. Jaako, M. S. Kim, and P. Rabl, Harvesting multiqubit entanglement from ultrastrong interactions in circuit quantum electrodynamics, Phys. Rev. Lett. 119, 183602 (2017).DOI
  13. [13]C. Hotter, A. M. Miranowicz, and K. Gietka, Quantum metrology in the ultrastrong coupling regime of light-matter interactions: Leveraging virtual excitations without extracting them, Phys. Rev. Lett. 135, 100802 (2025).DOI
  14. [14]T. Niemczyk, F. Deppe, H. Huebl, E. P. Menzel, F. Hocke, M. J. Schwarz, J. J. Garcia-Ripoll, D. Zueco, T. Hümmer, E. Solano, A. Marx, and R. Gross, Circuit quantum electrodynamics in the ultrastrong-coupling regime, Nature Physics 6, 772 (2010). 7, 12964.
  15. [15]F. Yoshihara, T. Fuse, Z. Ao, S. Ashhab, K. Kakuyanagi, S. Saito, T. Aoki, K. Koshino, and K. Semba, Inversion of qubit energy levels in qubit-oscillator circuits in the deep-strong-coupling regime, Phys. Rev. Lett. 120, 183601 (2018).DOI
  16. [16]F. Yoshihara, T. Fuse, S. Ashhab, K. Kakuyanagi, S. Saito, and K. Semba, Characteristic spectra of circuit quantum electrodynamics systems from the ultrastrong-to the deep-strong-coupling regime, 95, 053824.DOI
  17. [17]A. Tomonaga, H. Mukai, F. Yoshihara, and J. S. Tsai, Quasiparticle tunneling and 1/f charge noise in ultrastrongly coupled superconducting qubit and resonator, Physical Review B 104, 224509 (2021).DOI
  18. [18]F. Yoshihara, S. Ashhab, T. Fuse, M. Bamba, and K. Semba, Hamiltonian of a flux qubit-LC oscillator circuit in the deep-strong-coupling regime, Scientific Reports 12, 6764 (2022).DOI
  19. [19]D. Lamberto, A. Mercurio, O. Di Stefano, V. Savona, and S. Savasta, Renormalization and low-energy effective models in cavity and circuit quantum electrodynamics, 8, 430.
  20. [20]Breakdown of gauge invariance in ultrastrong-coupling cavity QED, phys. rev. a.
  21. [21]O. Di Stefano, A. Settineri, V. Macrì, L. Garziano, R. Stassi, S. Savasta, and F. Nori, Resolution of gauge ambiguities in ultrastrong-coupling cavity quantum electrodynamics, 15, 803.
  22. [22]P. Löwdin, A note on the quantum-mechanical perturbation theory, 19, 1396.
  23. [23]F. Yan, S. Gustavsson, A. Kamal, J. Birenbaum, A. P. Sears, D. Hover, T. J. Gudmundsen, D. Rosenberg, G. Samach, S. Weber, J. L. Yoder, T. P. Orlando, J. Clarke, A. J. Kerman, and W. D. Oliver, The flux qubit revisited to enhance coherence and reproducibility, 7, 12964.DOI
  24. [24]K. Inomata, T. Yamamoto, P.-M. Billangeon, Y. Nakamura, and J. S. Tsai, Large dispersive shift of cavity resonance induced by a superconducting flux qubit in the straddling regime, Phys. Rev. B 86, 140508(R).
  25. [25]A. Tomonaga, H. Mukai, K. Mizuno, and J. S. Tsai, Efficient spectrum analysis for multi-junction superconducting circuit, Physica Scripta 100, 095115 (2025).
  26. [26]H. Feshbach, Unified theory of nuclear reactions, 5, 357 ().
  27. [27]H. Feshbach, A unified theory of nuclear reactions. II, 19, 287 ().
  28. [28]P. Löwdin, Studies in perturbation theory. IV. solution of eigenvalue problem by projection operator formalism, 3, 969 ().
  29. [29]S. Bravyi, D. P. DiVincenzo, and D. Loss, Schrieffer–Wolff transformation for quantum many-body systems, 326, 2793.
  30. [30]S. Ashhab and F. Nori, Qubit-oscillator systems in the ultrastrong-coupling regime and their potential for preparing nonclassical states, Physical Review A 81, 042311 (2010).
  31. [31]D. Z. Rossatto, C. J. Villas-Bôas, M. Sanz, and E. Solano, Spectral classification of coupling regimes in the quantum rabi model, Phys. Rev. A 96, 013849 (2017).DOI
  32. [32]R. Harris, T. Lanting, A. J. Berkley, J. Johansson, M. W. Johnson, P. Bunyk, E. Ladizinsky, N. Ladizinsky, T. Oh, and S. Han, Compound josephson-junction coupler for flux qubits with minimal crosstalk, 80, 052506.
  33. [33]T. Miyanaga, A. Tomonaga, H. Ito, H. Mukai, and J. Tsai, Ultrastrong tunable coupler between superconducting $LC$ resonators, 16, 064041.
  34. [34]A. Blais, R.-S. Huang, A. Wallraff, S. M. Girvin, and R. J. Schoelkopf, Cavity quantum electrodynamics for superconducting electrical circuits: An architecture for quantum computation, 69, 062320.

Paper details

vlnc:W_w4838b37
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

Not recorded for this paper

View statement
Looking for related papers…
Contents