A robust phase of continuous transversal gates in quantum stabilizer codes

A quantum error correcting code protects encoded logical information against errors. Transversal gates are a naturally fault-tolerant way to manipulate logical qubits but cannot be universal themselves. Protocols such as magic state distillation are needed to achieve universality via measurements and postselection. A phase is a region of parameter space with smoothly varying large-scale statistical properties except at its boundaries. Here, we find a phase of continuously tunable logical unitaries for the surface code implemented by transversal operations and decoding that is robust against dephasing errors. The logical unitaries in this phase have an infidelity that is exponentially suppressed in the code distance compared to their rotation angles. We exploit this to design a simple fault-tolerant protocol for continuous-angle logical rotations. This lowers the overhead for applications requiring many small-angle rotations such as quantum simulation.

💡 Research Summary

The paper introduces a novel fault‑tolerant protocol for implementing continuous‑angle logical Z rotations on the surface code using only transversal coherent rotations, syndrome measurements, and decoder‑assisted corrections. While transversal gates are naturally fault‑tolerant, the Eastin‑Knill theorem forbids a universal set of such gates, so conventional approaches rely on magic‑state distillation and gate teleportation, which require mid‑circuit measurements, feed‑forward, and substantial overhead. The authors identify a “robust phase” in the space of physical dephasing rate p and rotation angle θ where the mean relative logical dephasing ⟨q_s/|ϕ_s|⟩ decays exponentially with the code distance d. In this phase, a transversal rotation U_θ = exp(iθZ)⊗n followed by syndrome extraction and a decoder‑determined Pauli correction C_s yields a logical unitary exp(iϕ_s Z) together with a logical dephasing channel of rate q_s. Crucially, the ratio q_s/|ϕ_s| becomes vanishingly small as d grows, meaning that logical rotations can be performed with arbitrarily low infidelity despite the presence of physical dephasing.

The authors model the combined coherent‑dephasing noise using a tensor‑network method and the PyMatching decoder, confirming the existence of the robust phase via numerical phase diagrams. They then propose to repeat the basic rotation step t times, choosing the physical rotation angles θ_j adaptively so that the sum of logical angles Φ_t = ∑ϕ(θ_j) approaches a target Φ_T. This adaptive selection is cast as a stochastic control problem: the state is the current accumulated logical angle Φ, the control input is the physical rotation angle θ, and the cost is the expected number of rounds needed to reach Φ_T. By solving the Bellman optimality equation through value iteration, they obtain an optimal policy θ*(Φ) that minimizes the expected number of rounds. The protocol also tracks the accumulated logical dephasing Q_t, which evolves according to Q_{t+1} = (1−Q_t) q(θ) + (1−q(θ)) Q_t. If Q_t becomes too large, the protocol can be reset and restarted, a strategy that further reduces the final logical error rate when the goal is to prepare a magic state for injection.

Simulation results demonstrate several key performance features. The expected number of rounds E