Superconducting Quantum Simulators

Superconducting quantum simulators use artificial atoms, microwave resonators, and controllable couplers to reproduce selected quantum dynamics. The same transmon array can approximate spins in one experiment and interacting bosons in another: the distinction depends on which levels are populated, which processes connect them, and what is measured. This page derives those model reductions and compares their physical limitations.

Required background. Quantum simulation defines the target process, while analog simulation introduces effective-model comparison. Helpful background. Circuit QED develops artificial atoms and resonators; superconducting qubits explains their control and readout.

Circuit excitations and effective models

Fabricated capacitances, Josephson elements, and resonator networks set the bare interactions. Microwave and flux drives select rotating frames, activate couplings, and prepare states. The implementation can therefore differ from the static circuit drawing: off-resonant modes, spectator-dependent shifts, and simultaneous drives can generate additional terms.

An encoding isometry $V$ satisfies $V^\dagger V=I$, with $P=VV^\dagger$. The compressed Hamiltonian $V^\dagger H_{\mathrm{dev}}V$ describes exact encoded dynamics only when the subspace is invariant. With leakage coupling $(I-P)H_{\mathrm{dev}}P\ne0$, virtual excursions generate corrections and real excursions leave the encoding. The derivation below retains enough of the oscillator model to identify both effects.

For a device channel $\Phi$, the encoded accepted branch is

$$\Phi_{\mathrm{acc}}(\rho) =V^\dagger\Phi(V\rho V^\dagger)V.$$

It is completely positive and trace nonincreasing. Its trace is the acceptance probability; normalizing the accepted state is generally nonlinear in the input. Keeping a distinct failure flag gives a trace-preserving channel. Moreover, a compressed device Lindbladian need not generate the accepted evolution at later times: leaked population can return and carry memory. Open-system simulation explains the appropriate system–environment description.

The platform comparison follows the effective degrees of freedom:

Circuit realization	Effective variables	Natural model families	Main approximation or measurement issue
Qubit lattice	two-level sites, hard-core excitations	$XY$, Ising, Heisenberg-type, quantum walks	leakage, residual couplings, rotating-frame calibration
Nonlinear resonator array	bosonic occupations	Bose–Hubbard, Kerr lattices, driven fluids of light	Fock-space cutoff, photon loss, drive calibration
Qubit–resonator array	polaritons	Jaynes–Cummings–Hubbard and related light–matter lattices	branch identification, occupation-dependent interaction
Parametric coupler network	spins or bosons with complex links	synthetic gauge fields, Floquet and topological models	phase conventions, micromotion, unwanted sidebands
Gate-programmable processor	encoded qubits or qudits	broad digital model classes	compilation, gate error, depth, sampling

No row is universally superior. The useful realization is the one whose native degrees of freedom, preparation, observables, and dominant errors fit the scientific question.

From quantized modes to spins and bosons

Normal-mode quantization, expansion of Josephson nonlinearities, rotating-wave approximations, and elimination of off-resonant modes are separate steps. Their small parameters and retained levels need not remain valid under strong driving. Blais and colleagues develop the circuit-QED derivations; Houck, Türeci, and Koch, and Carusotto and colleagues discuss their use for quantum simulation.

A useful hierarchy of models

For a weakly anharmonic network, a common intermediate description is

$$\frac{H_{\mathrm{mode}}}{\hbar} ={} \sum_i \left[ \omega_i n_i + \frac{\alpha_i}{2}n_i(n_i-1) \right] + \sum_{i<j} \left( J_{ij}a_i^\dagger a_j + J_{ij}^*a_j^\dagger a_i \right) + \frac{H_{\mathrm{parasitic}}+H_{\mathrm{drive}}(t)}{\hbar},$$

where $n_i=a_i^\dagger a_i$, $\alpha_i$ is the self-Kerr or anharmonicity, $J_{ij}$ is an exchange amplitude, and $H_{\mathrm{parasitic}}$ collects terms not included in the intended target. This equation may describe transmon excitations, nonlinear resonators, or dressed modes, but the parameter meanings depend on the basis used to derive it.

Here $\omega_i,\alpha_i,J_{ij}$ are angular frequencies, whereas the two explicitly divided Hamiltonians have energy units. For a transmon, $\alpha=\omega_{12}-\omega_{01}<0$: the Duffing approximation is a low-excitation expansion, not a Hamiltonian valid to arbitrarily large occupation. Extending its negative quartic term indefinitely would make it unbounded below. Convergence within this approximate oscillator model and validity of the model for the physical circuit are separate checks.

Qubit Lattices

When only two levels per nonlinear mode are intentionally populated, a superconducting array supplies effective spins. Define

$$Z_i = |0_i\rangle\langle 0_i| - |1_i\rangle\langle 1_i|, \qquad \sigma_i^+ = |1_i\rangle\langle 0_i|.$$

The superscript $+$ here means excitation creation. With the displayed $Z$ convention, $\sigma^+=(X-iY)/2$ and $\sigma^-=(X+iY)/2$. Within that subspace,

$$n_i \mapsto \frac{I-Z_i}{2}, \qquad a_i^\dagger \mapsto \sigma_i^+.$$

For real exchange amplitudes and $\delta_i=\omega_i-\omega_{\mathrm{ref}}$, an exchange-coupled network reduces, up to a scalar, to

$$\frac{H_{XY}}{\hbar} = -\frac12\sum_i\delta_i Z_i + \sum_{i<j}J_{ij} \left( \sigma_i^+\sigma_j^- + \sigma_i^-\sigma_j^+ \right),$$

or equivalently

$$\frac{H_{XY}}{\hbar} = -\frac12\sum_i\delta_i Z_i + \frac12\sum_{i<j}J_{ij} \left(X_iX_j+Y_iY_j\right)$$

for real $J_{ij}$. This is both an $XY$ spin model and a hard-core-boson model. It is not the unconstrained Bose–Hubbard model: the local occupation has been restricted to zero or one.

What supplies the interactions

Exchange can arise from direct capacitance, a bus resonator, a tunable coupler, or a parametrically activated process. Longitudinal and cross-Kerr terms can supply effective $ZZ$ interactions. Microwave drives add local fields, while echo sequences or digital blocks can combine native terms into Heisenberg-type and Ising dynamics.

A useful spin-model description may take the form

$$\frac{H_{\mathrm{spin}}}{\hbar} = \sum_i \left( h_i^xX_i+h_i^yY_i+h_i^zZ_i \right) + \sum_{i<j} \sum_{\mu,\nu=x,y,z} J_{ij}^{\mu\nu} \sigma_i^\mu\sigma_j^\nu.$$

The tensor $J_{ij}^{\mu\nu}$ is an effective description in a specified frame. Its entries may change under drive dressing and refocusing. Reporting only a nominal coupler setting is insufficient when unwanted $ZZ$ terms or spectator-dependent shifts are comparable to the effect under study.

Leakage is model error

The two-level reduction requires more than a well-resolved transition. During an interacting protocol, population can enter $|2\rangle$ through strong drives, near-resonant collisions such as $|11\rangle\leftrightarrow|20\rangle$, or coupler-mediated processes. Relevant small parameters often include

$$\frac{|J_{ij}|}{|\alpha_i|}, \qquad \frac{|\Omega_i|}{|\alpha_i|}, \qquad \frac{1}{|\Delta_{\mathrm{leak}}|\tau},$$

where all rates are angular frequencies and $\tau$ characterizes a smooth pulse envelope. Crucially, the actual $|11\rangle\leftrightarrow|20\rangle$ energy denominator is $\delta_i-\delta_j+\alpha_i$, not just $\alpha_i$. For the other doublon it is $\delta_j-\delta_i+\alpha_j$. Pulse bandwidth and occupation-dependent matrix elements must be compared with those detunings. Small ratios help, but coherent errors may accumulate and resonant conditions can invalidate a naive perturbative estimate.

Leakage should be measured when possible. A postselection rule that deletes leaked shots changes the effective experiment and must be reported with the retained fraction and induced bias.

Resonator and Nonlinear-Mode Arrays

Superconducting resonators provide bosonic modes with lithographically designed connectivity. Josephson elements can add an on-site Kerr nonlinearity or mediate nonlinear coupling. In a rotating frame, a common target is the driven Bose–Hubbard model

$$\frac{H_{\mathrm{BH}}}{\hbar} ={} \sum_i \left[ -\Delta_i n_i + \frac{U_i}{2}n_i(n_i-1) + F_i a_i^\dagger + F_i^*a_i \right] + \sum_{i<j} \left( J_{ij}a_i^\dagger a_j + J_{ij}^*a_j^\dagger a_i \right).$$

Here $\Delta_i=\omega_d-\omega_i$, $U_i$, and $F_i$ are angular-frequency parameters; this sign of detuning is opposite to $\delta_i$ used above. The complex pump $F_i$ sets both amplitude and phase. Loss gives an open-system generator,

$$\dot\rho = -\frac{i}{\hbar}[H_{\mathrm{BH}},\rho] + \sum_i\kappa_i\mathcal D[a_i]\rho + \mathcal L_{\mathrm{other}}\rho,$$

with

$$\mathcal D[L]\rho = L\rho L^\dagger - \frac12 \left\lbrace L^\dagger L, \rho \right\rbrace.$$

The coefficient $\kappa_i$ is the photon-number decay rate: an undriven linear mode has $\langle n_i\rangle\propto e^{-\kappa_i t}$ and $\langle a_i\rangle\propto e^{-\kappa_i t/2}$. This fixes the factor of two in the driven benchmark below.

This is generally a driven nonequilibrium problem, not an equilibrium Bose gas weakly coupled to a thermal bath. Coherent drive does not define a thermodynamic chemical potential by itself, and the steady state need not be a Gibbs state.

Bosonic truncation has to converge

Numerical analysis and some effective derivations truncate each mode at $n_i\leq n_{\max}$. The experiment does not inherit that cutoff unless a strong blockade enforces it. A calculation should demonstrate convergence of the reported observable as $n_{\max}$ increases, while the experiment should bound high-occupation weight through number-selective measurement, calibrated moments, or another appropriate diagnostic.

The hard-core limit $|U/J|\gg1$ is an approximation over a specified energy and occupation range, away from drive-assisted doublon resonances. For $U<0$, avoiding doublons can describe prepared dynamics in an excited manifold; it does not identify the equilibrium ground state with that of a repulsive Bose–Hubbard model. It does not turn a nonlinear oscillator into an exact qubit. Conversely, finite $U$ can be the desired physics when doublons, bound pairs, compressibility, or interaction-dependent transport are being studied.

Qubit–Resonator and Polariton Arrays

A circuit-QED cell containing a resonator and artificial atom is described at leading order by a Jaynes–Cummings Hamiltonian. Coupling cells into a lattice gives

$$\frac{H_{\mathrm{JCH}}}{\hbar} ={} \sum_i \left[ \omega_{r,i}a_i^\dagger a_i - \frac{\omega_{q,i}}{2}Z_i + g_i \left( a_i^\dagger\sigma_i^- + a_i\sigma_i^+ \right) \right] - \sum_{\langle i,j\rangle} t_{ij} \left( a_i^\dagger a_j + a_j^\dagger a_i \right).$$

Here $g_i$ and $t_{ij}$ are taken real, and the qubit term has a minus sign because $Z|0\rangle=|0\rangle$ while $|1\rangle$ is the excited state. The conserved excitation number of an isolated cell is $N=a^\dagger a+(I-Z)/2$. In the basis $(|0,n\rangle,|1,n-1\rangle)$,

$$\frac{H_n}{\hbar}= \begin{pmatrix} n\omega_r-\omega_q/2 & g\sqrt n\\ g\sqrt n & (n-1)\omega_r+\omega_q/2 \end{pmatrix}.$$

Thus the branches have energies $E_{n,\pm}/\hbar=(n-1/2)\omega_r \pm\tfrac12\sqrt{(\omega_q-\omega_r)^2+4g^2n}$ for $n\ge1$. The local eigenstates are polaritons: superpositions of photonic and artificial-atom excitations. Their splitting and composition depend on detuning and excitation number. Photon blockade then produces an occupation-dependent effective interaction.

The Jaynes–Cummings–Hubbard model should not be called a Bose–Hubbard model without stating the reduction. A Bose–Hubbard description can emerge in a selected polariton branch and parameter regime, but its hopping and interaction inherit branch-dependent matrix elements. Near branch crossings, under strong drive, or outside the rotating-wave regime, a one-band reduction can fail.

Useful diagnostics include local spectroscopy of the polariton ladder, normal-mode spectroscopy of the array, response versus drive power, and comparison of observed resonances with the full few-excitation device model. Seeing an avoided crossing establishes hybridization; it does not by itself establish a many-body phase.

Tunable Couplers and Parametric Interactions

Static real hopping is restrictive. Flux-tunable couplers and periodic modulation can activate interactions in selected rotating frames. Suppose modes $i$ and $j$ have a frequency difference $\omega_i-\omega_j$. Modulating a coupling element near that difference can produce an effective exchange

$$\frac{H_{ij}^{\mathrm{eff}}}{\hbar} = J_{ij}^{\mathrm{eff}} e^{i\phi_{ij}} a_i^\dagger a_j + J_{ij}^{\mathrm{eff}} e^{-i\phi_{ij}} a_j^\dagger a_i.$$

Take $J_{ij}^{\mathrm{eff}}\ge0$, with any sign absorbed into $\phi_{ij}$; the term shown transfers an excitation from $j$ to $i$. To express the same Hamiltonian in new operators $a_i'=e^{i\chi_i}a_i$, the coefficient must transform as

$$\phi_{ij} \mapsto \phi_{ij}+\chi_i-\chi_j.$$

An individual link phase is therefore gauge dependent. The phase accumulated around a closed directed loop $C$,

$$\Phi_C = \sum_{(i,j)\in C}\phi_{ij} \pmod{2\pi},$$

is gauge invariant and can represent a synthetic magnetic flux. A nonzero phase written on a single link of an open chain has no gauge-invariant flux; it can be removed by redefining local phases. If coherent local drives or phase-sensitive receivers are present, their phases must transform too; relative drive-to-hopping phases can be measured even on an open chain. Roushan and colleagues use closed loops to study synthetic magnetic flux.

Floquet qualifications

A periodically driven circuit has exact evolution generated by a time-dependent Hamiltonian. A static $H_{\mathrm{eff}}$ is a derived description. Its credibility depends on drive frequency, amplitude, off-resonant transitions, micromotion, and heating or leakage. At stroboscopic times $t=mT$,

$$U(mT) = \left[U(T)\right]^m = e^{-imT H_F/\hbar},$$

but $H_F$ is not unique because quasienergies are defined modulo $2\pi\hbar/T$. Measurements within a period can depend on the micromotion operator even when stroboscopic dynamics agree.

Parametric protocols should report drive frequencies and phases, calibrated effective coupling amplitudes, relevant unwanted sidebands, the observation times relative to the drive period, and the approximation used to infer $H_{\mathrm{eff}}$. The Floquet–Magnus expansion provides one systematic route when its scale hierarchy is appropriate.

Digital and Digital–Analog Routes

A gate-programmable superconducting processor can encode a target into qubits and implement a product formula, qubitization-based algorithm, variational state preparation, or another digital protocol. Platform-specific issues then enter through the native gate set, connectivity, leakage, reset, measurement, and coherent calibration errors. The algorithmic approximation remains separate from hardware noise.

Digital–analog simulation interleaves native many-body evolution with local rotations, refocusing pulses, or entangling gates. A schematic cycle is

$$U_{\mathrm{cycle}} = R_m e^{-iH_m\tau_m/\hbar} \cdots R_2 e^{-iH_2\tau_2/\hbar} R_1 e^{-iH_1\tau_1/\hbar}.$$

This can exploit a high-fidelity exchange interaction while using digital controls to change bases or interaction signs. It can also compound analog model error, pulse error, and product-formula error. Calling the protocol hybrid does not remove any of these terms.

Time reversal illustrates the point. Reversing a nominal coupling sign can enable a Loschmidt echo or an out-of-time-ordered correlator, but a credible reversal test must establish which terms changed sign. Static frequency disorder, loss, residual $ZZ$ interactions, and control transients may not reverse. Echo decay is therefore not automatically a measure of scrambling.

Open-System and Reservoir-Engineered Routes

Loss is unavoidable in microwave circuits, but it can also be designed. Auxiliary resonators, lossy modes, measurement and feedback, and parametric processes can produce effective pumping, cooling, stabilization, or correlated jump operators. For example, adiabatic elimination of a rapidly damped auxiliary mode may yield

$$\dot\rho_s = -\frac{i}{\hbar}[H_{\mathrm{eff}},\rho_s] + \Gamma\mathcal D[L_{\mathrm{eff}}]\rho_s.$$

The elimination requires a scale hierarchy and generates both dissipative and coherent corrections. The auxiliary occupation, separation of rates, frequency dependence of the engineered bath, and additional noise channels should be checked rather than hidden in a fitted $\Gamma$.

Reservoir engineering is especially useful for photonic matter because photon loss continuously removes the excitations whose many-body state is sought. A frequency-selective replenishing process can stabilize an incompressible state against holes. The resulting state is maintained by nonequilibrium fluxes; it is not an equilibrium ground state merely because its correlations resemble those of one. Ma and colleagues demonstrate frequency-selective stabilization of a finite photonic Mott state; preparation by a driven reservoir and equilibrium thermalization are distinct physical statements.

The page on Open-System Simulation develops the general distinction between target dissipation, implementation noise, channel dilation, trajectories, and non-Markovian memory.

State Preparation

Superconducting simulators support several preparation routes, each with a different claim.

Product-state initialization

Measurement, active reset, or passive relaxation prepares a reference state; calibrated rotations then create local occupations or spin orientations. This is natural for quenches and transport. The initial state should be characterized for reset errors, thermal population, leakage, and correlations induced by simultaneous control.

Adiabatic and quasiadiabatic ramps

One prepares an accessible eigenstate and changes parameters toward a target Hamiltonian. A finite ramp competes with the minimum spectral gap, decoherence, calibration drift, and finite-size level structure. Agreement with an intended ground state requires more than a visually smooth ramp. Energy estimates, symmetry checks, reverse ramps, and ramp-time convergence can help.

Spectral loading and photon injection

Frequency-selective pulses can populate chosen normal modes or many-body resonances. The prepared state depends on pulse bandwidth, matrix elements, loss during loading, and spectral crowding. A peak in transmission is not by itself a calibrated Fock-state preparation.

Dissipative stabilization

Engineered loss and replenishment can autonomously attract the device toward a steady manifold. The correct output is then a steady state of an implemented Liouvillian. Preparation quality should include convergence from multiple initial states, stationarity over the measurement window, defect density, and sensitivity to reservoir parameters.

Variational and feedback-assisted preparation

Parameterized circuits or pulse schedules can be optimized from measured costs. The result depends on ansatz expressivity, optimizer behavior, estimator bias, shot noise, and drift during the loop. A low measured cost is evidence only after the relation between that cost and the desired state has been established.

Measurement and Observable Reconstruction

Superconducting circuits do not directly return a wavefunction. They return voltages, classified outcomes, switching events, or spectra from which target observables are inferred.

Site-resolved bitstrings

Dispersive readout can estimate occupation and spin observables. Repeating a protocol in basis $B$ produces samples $x$ with distribution

$$p_{\mathrm{obs}}(x|B) = \sum_y A_B(x|y) p_{\mathrm{true}}(y|B).$$

This classical assignment model assumes the measurement effects are diagonal in the intended measurement basis, so probabilities alone determine the response. Basis-state calibration does not characterize arbitrary coherent measurement errors. Within this model, $A_B$ is a readout-response matrix. Independent single-site correction assumes

$$A_B(x|y) \approx \prod_i A_{B,i}(x_i|y_i),$$

an approximation that can fail through readout crosstalk, state-dependent resonator shifts, shared amplification, and classifier correlations. Inverting an ill-conditioned response matrix can amplify statistical noise and may produce nonphysical estimates unless regularized.

Resonator output fields

Homodyne or heterodyne detection gives field quadratures and correlation functions. Input–output theory relates an output mode to an intracavity mode, schematically

$$a_{\mathrm{out}}(t) = a_{\mathrm{in}}(t) - \sqrt{\kappa_{\mathrm{ext}}}\,a(t),$$

subject to a convention for phases and coupling signs. Receiver gain, bandwidth, added noise, filtering, and temporal-mode definitions belong to the observable calibration. A transmission coefficient is a response function of a driven open device, not automatically an equilibrium spectral function.

Spectroscopy

Weak-probe spectroscopy can identify normal modes, avoided crossings, interaction shifts, and many-body transitions. Strong probes dress the spectrum and may saturate transitions. Line positions and weights should be compared with an open-system response calculation when linewidths, pumping, and state populations matter.

Correlators, tomography, and randomized measurements

Local rotations before readout give Pauli correlators. Full tomography scales exponentially; symmetry-restricted reconstruction, randomized measurements, and shadow methods can estimate selected properties more economically. Every method trades measurement settings and shots against assumptions, estimator variance, and reconstruction bias.

Out-of-time-ordered correlators and Loschmidt echoes require additional controls, usually approximate time reversal. Their interpretation should be benchmarked against ordinary decoherence and reversal error. A decaying echo alone does not distinguish chaos, leakage, loss, and calibration mismatch.

Testing the effective circuit model

Spectroscopy measures energy differences, Ramsey experiments constrain local detunings, and swap oscillations constrain exchange. These data do not automatically identify a complete Hamiltonian: omitted operators may be absorbed into fitted parameters. Parameters calibrated separately can also change under simultaneous drives through AC Stark shifts, crosstalk, and coupler excitation. Predicting additional initial states and observables without refitting is a useful test of the inferred model.

Distinguish errors in the model from errors in its measured output. Leakage, photon loss, and preparation defects change states; assignment correction changes the estimator and its variance. A sum of error terms is meaningful only when they bound the same output metric. Vary the corresponding physical parameters and numerical cutoffs, then propagate uncertainty to the plotted observable. Device characterization and verification of quantum simulation develop these tests.

Finite-size and time limits remain separate from calibration. A small array can establish finite-instance dynamics. Long switching times, edge response, or a sharp crossover require additional scaling or symmetry information to support a bulk phase interpretation. Fitzpatrick and colleagues provide an experimental example of dissipative-lattice behavior; the definition of the transition determines which finite-device trends matter.

Worked Mapping: Transmon Chain to Hard-Core Bosons

Consider $L$ capacitively coupled transmon-like modes in a rotating frame,

$$\frac{H}{\hbar} = \sum_{i=1}^L \left[ \delta_i n_i + \frac{\alpha_i}{2}n_i(n_i-1) \right] + \sum_{i=1}^{L-1} J_i \left( a_i^\dagger a_{i+1} + a_{i+1}^\dagger a_i \right).$$

Assume the protocol starts in the zero- and one-excitation subspace per site, the bandwidth and interaction do not resonantly reach $|2\rangle$, and $|J_i/\alpha_j|\ll1$ on relevant links. Projection gives

$$\frac{H_{\mathrm{hc}}}{\hbar} = \sum_i\delta_i n_i^{\mathrm{hc}} + \sum_iJ_i \left( b_i^\dagger b_{i+1} + b_{i+1}^\dagger b_i \right),$$

with

$$\left(b_i^\dagger\right)^2=0, \qquad n_i^{\mathrm{hc}}=b_i^\dagger b_i.$$

The spin map is

$$b_i^\dagger\leftrightarrow\sigma_i^+, \qquad n_i^{\mathrm{hc}} = \frac{I-Z_i}{2}.$$

A single excitation performs a continuous-time quantum walk. Several excitations experience the hard-core constraint and produce nontrivial correlations. However, a one-dimensional open nearest-neighbor $XY$ chain with local $Z$ fields maps to free fermions by Jordan–Wigner: this basic case remains an important classically tractable benchmark. Additional density interactions or more general geometries change that conclusion. Yan and colleagues study multiparticle quantum walks in a superconducting processor. The simplest site-density observable is

$$\langle n_i(t)\rangle = \frac{1-\langle Z_i(t)\rangle}{2}.$$

Connected density correlations are

$$C_{ij}(t) = \langle n_i(t)n_j(t)\rangle - \langle n_i(t)\rangle \langle n_j(t)\rangle.$$

The mapping has several falsifiable checks:

• total excitation number should be conserved by the ideal Hamiltonian;
• one-particle dynamics should match the calibrated hopping matrix;
• measured $|2\rangle$ population should remain below the stated tolerance;
• interaction-sensitive observables should change predictably between one- and two-particle sectors;
• loss and dephasing should explain observed number decay and coherence loss over the chosen time window.

For equal anharmonicities and resonant sites, virtual doublons already shift the $|11\rangle$ energy by $\delta E_{11}/\hbar\simeq-4J^2/\alpha$ even when their population remains small (Exercise 8). Thus the projected hard-core Hamiltonian misses an interaction that accumulates phase on a time scale $|\alpha|/J^2$.

Failure of number conservation does not identify its cause. It may indicate relaxation, thermal excitation, leakage, drive error, or an omitted number-nonconserving interaction. Additional measurements are needed to separate them.

Worked Mapping: Driven Photonic Lattice

Consider a coherently driven array with uniform loss. In a common rotating frame,

$$\frac{H}{\hbar} ={} -\Delta\sum_i n_i + \frac{U}{2}\sum_i n_i(n_i-1) - J\sum_{\langle i,j\rangle} \left(a_i^\dagger a_j+a_j^\dagger a_i\right) + \sum_i\left(F_i a_i^\dagger+F_i^*a_i\right),$$

and

$$\dot\rho = -\frac{i}{\hbar}[H,\rho] + \kappa\sum_i\mathcal D[a_i]\rho.$$

The dimensionless ratios $U/\kappa$, $J/\kappa$, $|F|/\kappa$, and $\Delta/\kappa$, together with geometry and boundary conditions, organize the finite-device response. For $U=J=0$, each site is a linear driven oscillator with steady amplitude

$$\langle a_i\rangle_{\mathrm{ss}} = \frac{F_i}{\Delta+i\kappa/2}$$

under the conventions above. This limit calibrates drive, detuning, loss, and receiver response before nonlinear many-body claims are attempted.

At finite $U$ and $J$, observables may include steady occupations, $g^{(2)}$ correlations, hysteresis under finite-rate sweeps, switching-time distributions, and spatial response. Long switching times and bimodal records can signal metastability in a finite system. Calling the observation a phase transition requires care about finite-size scaling, observation time, and the order of limits.

Interpreting spectroscopy and dynamics

An avoided crossing establishes hybridization; its gap alone does not establish a many-body state. Spectral frequencies also do not determine populations or coherences. Compare transition strengths and prepared-state observables with the same model used for the level energies.

Time-reversal diagnostics need a similar distinction. A Loschmidt echo can decay because of loss, leakage, unreversed disorder, or control error. The OTOC experiments of Braumüller and colleagues illustrate the additional controls involved; echo decay by itself is not a unique signature of chaos.

For numerical modeling, increase the retained occupation range and check each observable. A stable mean photon number does not imply a stable intensity correlation, which weights the distribution tail more strongly. For experimental readout inversion, include calibration uncertainty and possible bias from regularization or postselection. These checks connect the calculated effective model to the measured finite-device response.

Exercises

1. Hard-core reduction

Project the weakly anharmonic mode Hamiltonian

$$\frac{H}{\hbar} = \sum_i\delta_i n_i + \sum_{i<j}J_{ij} \left(a_i^\dagger a_j+a_j^\dagger a_i\right)$$

onto $|0\rangle,|1\rangle$ at every site. Derive the $XY$ spin Hamiltonian using the conventions on this page. Which constant term can be discarded?

Solution

Within the truncated subspace,

$$n_i = \frac{I-Z_i}{2}, \qquad a_i^\dagger=\sigma_i^+, \qquad a_i=\sigma_i^-.$$

Therefore

$$\frac{H_{\mathrm{eff}}}{\hbar} ={} \frac12\sum_i\delta_i I - \frac12\sum_i\delta_i Z_i + \sum_{i<j}J_{ij} \left( \sigma_i^+\sigma_j^- + \sigma_i^-\sigma_j^+ \right).$$

The first term is a scalar energy shift and can be discarded. For real $J_{ij}$,

$$\sigma_i^+\sigma_j^- + \sigma_i^-\sigma_j^+ = \frac12(X_iX_j+Y_iY_j).$$

The reduction does not prove that leakage is absent; that is a separate approximation to test.

2. A gauge-invariant loop phase

Three modes form a directed triangle with hopping terms $J_{ij}e^{i\phi_{ij}}a_i^\dagger a_j$. Re-express the same Hamiltonian using $a_i'=e^{i\chi_i}a_i$ and show that this leaves $\Phi=\phi_{12}+\phi_{23}+\phi_{31}$ invariant modulo $2\pi$. Explain why the phase on one link of an open chain is not by itself observable.

Solution

Each link transforms as

$$\phi_{ij}' = \phi_{ij}+\chi_i-\chi_j.$$

Thus

$$\Phi' ={} \phi_{12}+\chi_1-\chi_2 + \phi_{23}+\chi_2-\chi_3 + \phi_{31}+\chi_3-\chi_1 = \Phi.$$

All local phases cancel around the loop. On a tree, including an open chain, one can choose the $\chi_i$ recursively to make every hopping real. There is no closed loop and hence no gauge-invariant flux. Coherent drives transform at the same time: $F_i a_i^\dagger$ becomes $F_i e^{i\chi_i}a_i^{\prime\dagger}$. Keeping laboratory drive phases fixed while changing a hopping phase is a physical change, not merely a change of operator convention.

3. Linear driven-cavity benchmark

For one mode with

$$\frac{H}{\hbar} = -\Delta a^\dagger a + F a^\dagger+F^*a, \qquad \dot\rho = -\frac{i}{\hbar}[H,\rho] + \kappa\mathcal D[a]\rho,$$

derive the steady coherent amplitude. Why is this a useful calibration before turning on a nonlinear many-body protocol?

Solution

The first moment obeys

$$\frac{d}{dt}\langle a\rangle = \left(i\Delta-\frac\kappa2\right) \langle a\rangle - iF.$$

Setting the derivative to zero gives

$$\langle a\rangle_{\mathrm{ss}} = \frac{F}{\Delta+i\kappa/2}.$$

Its amplitude and phase jointly test the drive calibration, detuning, linewidth, receiver transfer function, and sign conventions. These quantities otherwise enter a nonlinear array fit simultaneously and can become poorly identifiable.

4. Number loss versus leakage

In a hard-core-boson experiment, the measured computational-subspace number $N_{01}=\sum_i |1_i\rangle\langle1_i|$ decreases. Give at least three physical mechanisms that can cause this observation and one additional measurement that helps distinguish each mechanism.

Solution

Energy relaxation $|1\rangle\to|0\rangle$ decreases both total excitation and $N_{01}$; an interleaved lifetime experiment under comparable bias and drive conditions constrains it. An exchange process $|11\rangle\to|20\rangle$ decreases $N_{01}$ while conserving the total physical excitation number. An unintended driven $|1\rangle\to|2\rangle$ transition can even increase total excitation. Three-level readout or shelving tests these leakage processes. Readout drift can change classified populations without changing the state; interleaved prepared-state references test assignment stability. A number-nonconserving parametric sideband or unintended drive can also change the number; frequency scans and phase-dependent controls help identify that mechanism.

5. Bosonic cutoff convergence

A simulation of a driven Kerr resonator is run with local cutoffs $n_{\max}=4,6,8$. The estimated occupation is $2.10$, $2.34$, and $2.35$, while the estimated $g^{(2)}(0)$ is $0.72$, $0.91$, and $0.98$. What can be claimed about convergence? What further check would you request?

Solution

The mean occupation appears nearly converged between cutoffs 6 and 8, but the second-order correlation has not demonstrated comparable stability. Higher moments weight the tail of the number distribution more strongly, so they can remain cutoff sensitive after the mean has stabilized. One should increase $n_{\max}$ further and inspect the probability near the cutoff, especially $p(n_{\max})$ and the high-occupation tail. Convergence must be established for every observable used in the claim, not inferred from one low moment.

6. Readout-response inversion

Suppose measured bitstring probabilities satisfy $p_{\mathrm{obs}}=A p_{\mathrm{true}}$. Explain why a formally invertible but ill-conditioned $A$ can make corrected correlators less reliable. Name two ways to diagnose or manage the problem.

Solution

The corrected estimate is

$$\widehat p_{\mathrm{true}} = A^{-1}\widehat p_{\mathrm{obs}}.$$

Small singular values of $A$ make $A^{-1}$ amplify shot noise and calibration error. The result can have large covariance or even negative components. One can inspect the singular-value spectrum or condition number and propagate the full sampling and calibration covariance through the inversion. Constrained maximum-likelihood reconstruction, regularization, more informative readout, or reporting observables that avoid unstable directions can manage the issue. Any regularization bias should be included in the uncertainty analysis.

7. Designing a held-out test

You calibrate an $L$-site exchange matrix from single-excitation quantum walks. Design a held-out test that probes whether the same model predicts the two-excitation sector. State one discrepancy that would indicate a missing interaction rather than merely hopping miscalibration.

Solution

Prepare two excitations at several separations, evolve at times not used in the hopping fit, and measure the full two-particle distribution together with connected correlations $C_{ij}(t)$. Predict these records using the fixed single-particle hopping matrix and the stated hard-core constraint, without refitting $J_{ij}$. A separation-dependent phase shift, bound-pair feature, or correlation pattern that cannot be reproduced while single-particle dynamics remain accurate points toward a missing two-body term such as residual $ZZ$, cross-Kerr interaction, or finite-anharmonicity correction. Uniform timing or hopping error would normally also spoil the held-out one-particle prediction.

8. Virtual doublons and an accumulated phase

Two identical resonant modes have anharmonicity $\alpha\ne0$ and real exchange rate $J$. In the two-excitation sector, subtract the bare $|11\rangle$ energy and derive the matrix on $\{|11\rangle,|20\rangle,|02\rangle\}$. Find the eigenfrequency connected continuously to $|11\rangle$ as $J\to0$ and the maximum doublon probability after starting in $|11\rangle$. Why does small leakage not imply a negligible phase error at arbitrarily long times?

Solution

The creation-operator factors give

$$\frac{H_2}{\hbar}= \begin{pmatrix} 0&\sqrt2J&\sqrt2J\\ \sqrt2J&\alpha&0\\ \sqrt2J&0&\alpha \end{pmatrix}.$$

Only $|D_+\rangle=(|20\rangle+|02\rangle)/\sqrt2$ couples to $|11\rangle$, with matrix element $2J$; the antisymmetric doublon is dark. The connected eigenfrequency is

$$\zeta=\frac{\alpha-\operatorname{sgn}(\alpha) \sqrt{\alpha^2+16J^2}}2 =-\frac{4J^2}{\alpha}+O(J^4/\alpha^3).$$

It is positive for the usual transmon $\alpha<0$. The exact two-state oscillation gives

$$P_D(t)=\frac{16J^2}{\alpha^2+16J^2} \sin^2\!\left(\frac{\sqrt{\alpha^2+16J^2}}2t\right).$$

Leakage remains of order $(J/\alpha)^2$, while the phase difference relative to the naively projected Hamiltonian grows as $\zeta t$. A simulation can therefore have small leaked population and an order-one interaction-induced phase error. Preparing the dressed eigenstate instead removes these population oscillations but retains the shifted eigenfrequency.

References

• D. G. Angelakis, M. F. Santos, and S. Bose, “Photon-blockade-induced Mott transitions and $XY$ spin models in coupled cavity arrays,” Physical Review A 76, 031805(R) (2007), doi:10.1103/PhysRevA.76.031805.

• A. Blais, A. L. Grimsmo, S. M. Girvin, and A. Wallraff, “Circuit quantum electrodynamics,” Reviews of Modern Physics 93, 025005 (2021), doi:10.1103/RevModPhys.93.025005.

• J. Braumüller et al., “Probing quantum information propagation with out-of-time-ordered correlators,” Nature Physics 18, 172–178 (2022), doi:10.1038/s41567-021-01430-w.

• I. Carusotto, A. A. Houck, A. J. Kollár, P. Roushan, D. I. Schuster, and J. Simon, “Photonic materials in circuit quantum electrodynamics,” Nature Physics 16, 268–279 (2020), doi:10.1038/s41567-020-0815-y.

• M. Fitzpatrick, N. M. Sundaresan, A. C. Y. Li, J. Koch, and A. A. Houck, “Observation of a dissipative phase transition in a one-dimensional circuit QED lattice,” Physical Review X 7, 011016 (2017), doi:10.1103/PhysRevX.7.011016.

• A. A. Houck, H. E. Türeci, and J. Koch, “On-chip quantum simulation with superconducting circuits,” Nature Physics 8, 292–299 (2012), doi:10.1038/nphys2251.

• R. Ma et al., “A dissipatively stabilized Mott insulator of photons,” Nature 566, 51–57 (2019), doi:10.1038/s41586-019-0897-9.

• P. Roushan et al., “Chiral ground-state currents of interacting photons in a synthetic magnetic field,” Nature Physics 13, 146–151 (2017), doi:10.1038/nphys3930.

• P. Roushan et al., “Spectroscopic signatures of localization with interacting photons in superconducting qubits,” Science 358, 1175–1179 (2017), doi:10.1126/science.aao1401.

• Y. Salathé et al., “Digital quantum simulation of spin models with circuit quantum electrodynamics,” Physical Review X 5, 021027 (2015), doi:10.1103/PhysRevX.5.021027.

• Z. Yan et al., “Strongly correlated quantum walks with a 12-qubit superconducting processor,” Science 364, 753–756 (2019), doi:10.1126/science.aaw1611.