R&D Innovate

← All articles

Quantum

Synthetic loop phases cancel non-adiabatic errors in a superconducting qutrit

Driving a quantum system quickly without corrupting its state remains one of the central dilemmas of quantum control. Standard adiabatic protocols offer high fidelity by steering states gently along an instantaneous eigenpath, but their operational speed is strictly limited by the energy gap to adjacent levels. Accelerating the drive inevitably triggers non-adiabatic transitions, kicking population out of the intended subspace and into unwanted excited states. Counterdiabatic driving provides an analytical escape from this trade-off by introducing auxiliary control fields tailored to cancel non-adiabatic transitions as they emerge. Yet in multi-level architectures, synthesising these auxiliary fields typically demands complex, multi-axis control schemes that scale poorly across physical hardware.

In cyclic three-level systems, or qutrits, a geometric mechanism offers a more compact alternative. By closing a triangular interaction loop between three quantum levels, quantum interference among the couplings can be configured to cancel transitions selectively. Because the sign of a coupling field determines how probability amplitudes interfere, reversing a coupling sign can flip an interaction from destructive cancellation to constructive leakage. Turning this principle from a mathematical curiosity into an experimental tool, however, requires precise, simultaneous control over both single-photon and multi-photon transitions in a solid-state circuit.

Engineering cyclic driving in a transmon

Superconducting transmons are naturally anharmonic ladders whose lowest three energy levels—labeled $|0\rangle$, $|1\rangle$, and $|2\rangle$—form a qutrit. In standard operation, microwave pulses address the electric-dipole allowed transitions $|0\rangle \leftrightarrow |1\rangle$ and $|1\rangle \leftrightarrow |2\rangle$. The direct transition between the ground and second excited state, $|0\rangle \leftrightarrow |2\rangle$, is electric-dipole forbidden to first order by parity and selection rules. To establish a closed cyclic interaction loop, an artificial connection between $|0\rangle$ and $|2\rangle$ must be synthesised without disrupting the individual controls on the other two legs.

This closed loop was realised by combining two resonant single-photon microwave drives with a phase-controlled effective two-photon drive. The two resonant single-photon drives address the $|0\rangle \leftrightarrow |1\rangle$ and $|1\rangle \leftrightarrow |2\rangle$ transitions, respectively. Meanwhile, the effective two-photon drive bridges the $|0\rangle \leftrightarrow |2\rangle$ gap, completing the triangular loop.

Because each leg of the closed loop is driven independently, the relative phase accumulated along the perimeter—known as the effective loop phase—can be adjusted at will. The experiment defined two primary configurations separated by a phase shift of $\pi$: a left-handed loop phase $\Phi_L = +\pi/2$ and a right-handed loop phase $\Phi_R = -\pi/2$. This phase control effectively sets an analogue of handedness, governing how auxiliary driving terms interfere with the non-adiabatic terms that arise from pushing the system beyond the adiabatic speed limit.

Resolving sign-selective dark pathways

In a typical three-level transfer protocol, such as stimulated Raman adiabatic passage, population moves from an initial state to a target state along an instantaneous dark state—a superposition that theoretically avoids populating the lossy intermediate bright manifold. When the drive rate increases, non-adiabatic coupling between the dark state and the bright manifold opens an error channel, destroying the transfer fidelity.

The synthetic loop phase alters this dynamic fundamentally. At $\Phi_L = +\pi/2$, the auxiliary contribution generated by the effective two-photon pathway interferes destructively with the intrinsic non-adiabatic term. The cancellation suppresses transitions into the bright manifold, preserving state transfer along a quiet dark pathway even at drive speeds where adiabatic approximations fail entirely. Conversely, when the loop phase is inverted to $\Phi_R = -\pi/2$, the sign of the auxiliary contribution flips, interfering constructively with the non-adiabatic coupling. Rather than suppressing error, the drive reinforces leakage, pulling the population strongly into the bright manifold.

Time-resolved state trajectories directly capture this divergence in dynamics. Under the $\Phi_L$ condition, population smoothly transitions toward the designated target state, with the intermediate level remaining largely unpopulated throughout the pulse sequence. Under the $\Phi_R$ condition, the time-domain traces reveal substantial excursion into the bright manifold, scattering population away from the intended target.

By measuring the late-time population averages, the experiment demonstrated a target-state population contrast of 0.929 between the two opposite loop phases. This substantial difference confirms that the sign of the drive phase alone dictates whether non-adiabatic leakage is extinguished or magnified. Further characterisation through continuous loop-phase sweeps mapped out the phase selectivity of the mechanism, while systematic pulse-area and common-mode amplitude scans bounded the operating regime and quantified the system’s sensitivity to pulse-shaping errors.

Practical boundaries and operational limits

Demonstrating sign-selective counterdiabatic interference in an isolated transmon qutrit proves the underlying physical concept, but significant engineering hurdles separate this bench demonstration from operational quantum processors.

First, the protocol relies on an effective two-photon drive to bridge the $|0\rangle \leftrightarrow |2\rangle$ transition. Two-photon transitions require higher microwave drive powers and careful calibration to suppress unwanted Stark shifts, which can detune adjacent energy levels and degrade coherence. Second, while a population contrast of 0.929 demonstrates clear phase selectivity between destructive and constructive pathways, it falls short of the fault-tolerant thresholds required for gate operations in error-corrected architectures. Residual leakage, drive amplitude miscalibrations, and intrinsic transmon dephasing all limit late-time state fidelity.

Third, the experiment was conducted on a single programmable qutrit. Scaling cyclic drive schemes to multi-qudit processors introduces severe crosstalk risks. Delivering three concurrent microwave tones—including a high-power two-photon drive—to multiple closely spaced qubits or qutrits will inevitably lead to stray driving on neighboring circuit elements unless isolation and frequency allocation are managed with extreme stringency. The control overhead required to tune three drive amplitudes and relative loop phases simultaneously on every node represents a major hurdle for large-scale integration.

Sources

The R&D takeaway

For hardware planners and quantum control teams, this work demonstrates that multi-photon loop engineering can substitute for complex multi-axis control when mitigating non-adiabatic errors in multilevel nodes. However, integrating cyclic driving into processor roadmaps will require balancing the speed gains against the thermal and calibration overhead of high-power two-photon microwave lines. R&D strategies should treat synthetic loop phases as specialised acceleration tools for high-value bottlenecks, rather than universal replacements for standard two-level gate operations.

The R&D Innovate desk