Quantum Error Correction with the Gottesman-Kitaev-Preskill Code

I. INTRODUCTION

• GKP encoding protects against small position and momentum shifts, and the first experiments used trapped ions and microwave cavities.The longer-term goal is a fault-tolerant quantum computer using bosonic degrees of freedom.Source: §I · p. 1

A. Basic definitions

• Bosonic codes aim to encode k logical qubits in n modes and exploit each mode’s large Hilbert space; GKP achieves the simplest nontrivial case, k = n = 1.Successful bosonic codes are described as hardware efficient.Source: §I · p. 1, A. Basic definitions · p. 2

I. INTRODUCTION

• A proposed scaling route is to reduce single-mode GKP errors, then concatenate encoded qubits with a code such as the surface code.Its comparative value depends on achieving lower logical error than a comparable number of unencoded qubits.Source: §I · pp. 1–2
• A resource-efficient fault-tolerance milestone is encoded-state preparation, entangling gates, and measurement with fidelities comparable to or better than the best physical qubits.Practical interactions can let a single ancilla error propagate to a logical mode error, limiting encoded-operation fidelity.Source: §I · p. 2
• cQED is the Perspective’s focus because superconducting circuits are viewed as flexible and scalable, with high-quality modes and strong nonlinearities supporting coherent oscillator control.Optical GKP-state generation had not yet been demonstrated, amid photon loss, weak nonlinearities, and demanding detector requirements.Source: §I · pp. 1–2

A. Basic definitions

• GKP logical Pauli operators are oscillator displacements chosen to anticommute, while the codespace is the simultaneous +1 eigenspace of two commuting stabilizers.The stabilizers commute with the logical operators and generate the code’s stabilizer group.Source: A. Basic definitions · p. 2
• The codewords can be expressed as quadrature eigenstate combs, and common choices of displacement lattice include square, rectangular, and hexagonal codes.In the square code, the generalized quadratures are the usual position and momentum quadratures.Source: A. Basic definitions · pp. 2–3

B. Approximate GKP codewords

• Approximate GKP codewords are normalized states that approach the stabilizer and logical-state conditions in a meaningful limit.Ideal codewords are non-normalizable, and in general no physical process prepares a state entirely in the ideal codespace.Source: B. Approximate GKP codewords · p. 3
• A Gaussian envelope over the ideal infinite sums makes the codewords normalizable, with the ideal limit recovered as Δ→0.The envelope replaces each coherent state in the sum with a weighted coherent state.Source: B. Approximate GKP codewords · p. 3
• For Δ=0.3, the figure shows Wigner functions of approximate logical-zero states for square and hexagonal GKP codes.The corresponding lattice panels distinguish the logical states using filled and empty coherent-state sites.Source: Figure 1 · p. 3
• Another approximate construction applies weighted displacements to the ideal codewords, with weights concentrated around zero as Δ→0.The Gaussian-envelope construction is recovered for a specified choice of displacement weights.Source: B. Approximate GKP codewords · p. 3
• Modular squeezing parameters quantify proximity to stabilizer eigenstates: they are nonnegative and vanish exactly for such eigenstates.For the approximate codewords, both parameters approach Δ as Δ→0; the square and hexagonal examples have 10.1 dB and 9.48 dB squeezing, respectively, at approximately 4.6 average photons.Source: B. Approximate GKP codewords · p. 4

C. Error correcting properties

• The ideal GKP code satisfies quantum error-correction criteria for displacements with |u|, |v| < √π/2.Here u and v parameterize displacement along the two generalized quadratures.Source: C. Error correcting properties · p. 4
• Approximate codewords can support high-accuracy computation when their displacement weights are sufficiently localized and logical operations do not amplify approximation errors excessively.The latter condition is the paper’s fault-tolerance requirement for these errors.Source: C. Error correcting properties · p. 4
• Realistic oscillator noise can include displacements beyond the correctable range, so the error-correction criteria are generally only approximate.The discussion considers loss and dephasing followed by a numerically optimized recovery that bounds achievable fidelity rather than providing a practical procedure.Source: C. Error correcting properties · p. 4
• The GKP code has strong potential to correct loss but performs poorly against dephasing compared with some other bosonic codes.Small phase-space rotations create large displacements at large amplitudes; residual coupling and miscalibrated gates therefore make precise control important.Source: C. Error correcting properties · p. 4

D. Logical operations on GKP codes

• Apart from state preparation, all logical Clifford operations can be performed using Gaussian operations and homodyne measurements.The interactions are at most quadratic in creation and annihilation operators.Source: D. Logical operations on GKP codes · pp. 4–5
• The gate-fidelity figure compares an approximate square GKP code with trivial encoding in Fock states |0⟩ and |1⟩ under simultaneous loss and dephasing.The horizontal variable is photon number, followed by optimal error correction.Source: Figure 2 · p. 5
• The caption states that GKP outperforms the trivial encoding whenever its fidelity falls below the gray shaded region.Source: Figure 2 · p. 5

1. Pauli quadrature measurements

• A destructive Pauli measurement measures the corresponding quadrature, rounds to the nearest multiple of √π, and assigns the sign by even or odd parity.Rounding makes the procedure robust to small displacement errors.Source: §1 · p. 5
• At η = 75%, the error probability is about 5.6% (∆ = 0.3) and 4.1% (∆ = 0.2), versus 0.61% and 0.15% at η = 90%.The figure plots the probability of mistaking logical zero for one under homodyne measurement.Source: Figure 3 · p. 5, §1 · p. 6
• Rapid quadrature measurement conflicts with keeping a cQED oscillator long-lived, requiring κ to be increased before readout or information to be mapped to a low-Q mode.The passage gives Q ∼ 1/κ.Source: §1 · p. 5
• Measurement inefficiency shrinks the GKP state and worsens codeword discrimination, so known η requires rescaling the bins to multiples of √ηπ.The figure compares homodyne readout and noiseless phase-estimation schemes across measurement efficiencies and rounds.Source: §1 · p. 5, Figure 3 · p. 5
• Amplifying quadrature information before releasing the state may improve efficiency, but whether this approach can meet scalable fault-tolerance demands remains open.Source: §1 · p. 6

2. Pauli phase estimation

• A two-level ancilla enables nondestructive logical Pauli measurement by estimating the phase of a unitary displacement, avoiding direct quadrature measurement.The circuit uses a controlled-displacement gate and measures the ancilla in the X basis.Source: §2 · p. 6
• The improved scheme applies a small controlled displacement orthogonal to the logical displacement to better approximate measurement of the approximate logical Pauli.Source: §2 · p. 6
• For small ∆, optimizing λ to ≃√π∆^2/2 gives p_err ≃ 0.4∆^6.The modified scheme adds a small controlled displacement orthogonal to the logical displacement.Source: §2 · p. 6
• Majority voting over repeated ideal phase-estimation rounds is evaluated for both schemes with a noiseless two-level ancilla.The cited passage reports numerical outcomes for the rounds but is incomplete for some values.Source: §2 · pp. 6–7
• Modified phase estimation can yield very small measurement errors with a noiseless ancilla, but an ancilla bit flip during controlled displacement can cause a large random mode displacement.Source: §2 · p. 7

• The Hadamard, phase, and CNOT gates, together with logical basis measurements and encoded |0L⟩ and |AL⟩ preparation, form a universal set.Source: §3 · p. 7
• These gates are generally approximate on physical GKP codewords and may reduce encoded-information quality by making codewords harder to distinguish.Source: §3 · p. 7
• A Clifford frame replaces single-qubit Clifford gates with Pauli-basis measurements and generalized control gates, while preserving the number of qubits, two-qubit gates, and measurements.Source: §3 · p. 7
• For GKP qubits, updating the Clifford frame can be done by changing local-oscillator or classical-pump phases, with two-qubit Clifford interactions realizable through three- or four-wave mixing.Source: Figure 5 · p. 7, §3 · p. 7, Figure 6 · p. 8

4. Error spread through gates

• A GKP CNOT can spread a control-mode displacement error to the target, but small displacements remain small and can be corrected in a subsequent error-correction round.Source: §4 · p. 8
• Realistic implementations can add spurious higher-order terms that may amplify and spread errors; the goal is sufficiently low error for reduced overhead at the next protection level, not arbitrary suppression.Source: §4 · p. 8

A. State preparation using two-level ancilla

• For fixed ϵ, preparation cannot approach zero phase arbitrarily closely, and experimental imperfections such as unwanted nonlinearities and dephasing constrain increasing photon number.Source: A. State preparation using two-level… · p. 9
• Alternating phase estimation of a stabilizer displacement and a small orthogonal displacement prepares approximate GKP states with Δ ∼ √ϵ, sharpening peaks and trimming the envelope.Source: A. State preparation using two-level… · p. 9
• To prepare a logical state, repeated Sharpen-and-Trim cycles for both stabilizers project into the logical subspace, followed by a logical Z measurement to select a basis state.Source: A. State preparation using two-level… · p. 9, Figure 8 · p. 9

B. Fault-tolerance in state preparation

• An ancilla bit flip during the Sharpen controlled displacement can create a potentially large random GKP displacement, making the simple circuits non-fault-tolerant to relaxation.Source: B. Fault-tolerance in state preparation · p. 9
• Ancilla phase flips commute with phase-estimation gates and cause measurement errors; repeated final logical-Z measurements can suppress the resulting logical errors.Source: B. Fault-tolerance in state preparation · p. 9

1. Biased noise ancilla

• A biased-noise ancilla with strongly suppressed bit flips is proposed as a path to fault-tolerant state preparation, provided controlled-displacement gates preserve that suppression.Source: §1 · p. 10
• For a Kerr-cat ancilla, the controlled displacement becomes faster and more accurate as α increases, while the ancilla’s bit-flip suppression also strengthens.Source: §1 · pp. 10–11
• Optimized measurement and stabilizer protocols require an Rx rotation, which can be implemented by turning off the two-photon pump and allowing Kerr evolution; propagated errors are the measured logical operator or a stabilizer.Source: §1 · p. 11
• A SNAIL-based Kerr-cat could use the same nonlinear element for controlled displacement and GKP two-qubit gates, but the effect of flux bias on 3D-cavity lifetime requires further exploration.Source: §1 · p. 11

C. Error correction with GKP ancillae

• Knill correction avoids active recovery by tracking a logical Pauli in a Pauli frame, while its displacement-error propagation is essentially equivalent to Steane correction.A beam-splitter version of Knill correction was reported to lower logical-error probability.Source: C. Error correction with GKP ancillae · p. 11
• Single-shot GKP correction needs two additional GKP states per round and very high-efficiency quadrature measurement; preparing those ancillas may itself require repeated stabilizer measurements.Phase estimation with a two-state ancilla yields only one bit per measurement, whereas GKP ancillae can provide single-shot syndrome information with accuracy set by ancilla quality and measurement efficiency.Source: C. Error correction with GKP ancillae · p. 11

IV. THE BIG PICTURE: SCALABILITY AND FAULT-TOLERANCE

• In concatenated GKP–surface codes, repeated inner-code stabilizer measurements estimate and correct each GKP mode’s noise, while outer-code parity checks correct residual errors.The inner measurements may use GKP-encoded or discrete-qubit ancillae; surface-code ancillae may also be either type.Source: §IV · p. 12
• If logical-operation errors fall sufficiently below the surface-code threshold, modest code distance may reach useful logical error rates with less overhead than a conventional-qubit surface code.The proposed comparison concerns hardware and software-control resources.Source: §IV · p. 12

A. All-GKP surface code

• Modified surface-code checks use CZ gates for Z checks and a mix of CX and CX† gates for X checks to minimize displacement-error amplification.These gates can be implemented using nonlinear couplers such as a transmon or SNAIL.Source: A. All-GKP surface code · p. 13
• MWPM decoding can incorporate conditional Pauli-error probabilities from analog measurement outcomes as dynamic matching-graph edge weights.This uses continuous-variable measurement information in the surface-code decoder.Source: A. All-GKP surface code · p. 13
• Numerical studies commonly model state and circuit errors as independent Gaussian displacements, but that model is unphysical and underestimates logical error relative to coherent approximate GKP states.The Gaussian displacement channel also does not generally represent oscillator loss, dephasing, or heating at arbitrary rates.Source: A. All-GKP surface code · p. 13
• S = 18.6 dB under equally noisy displacement errors, versus S = 11.2 dB with state-preparation noise only; better decoding and protocol optimization reduce the latter threshold to S = 9.9 dB.The analog-syndrome decoding strategy can also dramatically reduce overhead for a target logical error rate.Source: A. All-GKP surface code · p. 13
• Rectangular GKP states can produce biased logical noise from isotropic environmental displacements, but their advantage with realistic circuit-level noise remains undetermined.For r = 3, the reported rectangular-code threshold corresponds to ∼1.7 dB squeezing, compared with 2.3 dB for the square-lattice code under the stated simplistic data-noise model.Source: A. All-GKP surface code · p. 14

B. Hybrid-GKP surface code

• For distance d, the illustrative Hybrid-GKP layout requires d^2 resonators and 2d^2 − 1 nonlinear couplers, versus 3d^2 − 1 resonators and 5d^2 − 4d couplers for the All-GKP layout.The hybrid layout retains GKP-encoded data while using discrete-qubit ancillae.Source: B. Hybrid-GKP surface code · p. 15
• A discrete ancilla measures GKP stabilizers and surface-code parity checks through controlled-displacement gates, followed by ancilla measurement.Approximate GKP states cause parity-measurement errors even with a noiseless ancilla; the cited modified phase-estimation approach gives perr ∼0.8∆^6 for X checks in the small-∆ limit.Source: B. Hybrid-GKP surface code · p. 15
• With a ∼14 dB GKP state and η = 75% measurement efficiency, the estimated total parity-check infidelity is ∼2% with a discrete ancilla versus ∼4% with a GKP ancilla.The discrete-ancilla estimate combines ∼0.0025% phase-estimation error with ∼2% readout error.Source: B. Hybrid-GKP surface code · p. 15
• Transmon relaxation limits controlled-displacement performance and may cap GKP-codeword lifetime near the transmon T1; a biased-noise ancilla such as a Kerr-cat is proposed as an alternative.The authors identify hardware efficiency as a motivation for further analysis of the hybrid approach.Source: B. Hybrid-GKP surface code · p. 15
• The architecture need not use a single ancilla type: discrete qubits may measure inner-code stabilizers while GKP modes serve as surface-code data and syndrome qubits.The choice can depend on the properties and performance of the operations.Source: B. Hybrid-GKP surface code · p. 15

C. Universality

• Lattice surgery can implement logical Clifford gates in the concatenated GKP–surface code using logical Pauli measurements and controlled-Pauli gates.The required controlled-Pauli gates can couple two GKP codewords or a GKP codeword and a discrete qubit.Source: C. Universality · pp. 15–16
• GKP magic-state preparation, injection, and distillation combined with Clifford operations provide universal computation; one-bit teleportation can transfer an arbitrary ancilla state to a GKP code.The teleportation circuit uses a controlled displacement to implement the CX gate.Source: C. Universality · p. 16, Figure 13 · p. 16

V. SUMMARY AND OUTLOOK

• State preparation must produce sufficiently small-∆ codewords with low logical-error probability, including errors propagated from the preparation ancilla.The stated goal is to outperform physical-qubit alternatives such as transmons and trapped ions; Kerr-cat ancillae are described as promising but need further analysis.Source: §V · p. 16
• High-fidelity GKP gates require clean two-mode interactions that switch on quickly relative to decoherence while minimizing spurious nonlinearities and errors the code corrects poorly.The passage particularly flags dephasing and spurious nonlinearities.Source: §V · p. 16
• Measurements remain a challenge: standard homodyne may be inadequate, and the authors discuss raising effective efficiency probably past 90% or using discrete-ancilla phase estimation.How low measurement error can be made in practice remains open.Source: §V · p. 16
• Scaling requires developing a cQED architecture and more realistic noise models; All-GKP thresholds look promising under studied models, while Hybrid-GKP quantitative analysis remains limited.The authors identify whether hybrid schemes can outperform conventional discrete-qubit approaches as a key question.Source: §V · pp. 16–17
• The outlook also proposes passive stabilization, noise-tailored codes, and color-code alternatives, while calling for performance estimates under realistic circuit-level noise.Software single-qubit Clifford operations on GKP codes could support software Clifford operations at the color-code level.Source: §V · p. 17