Walking Floquet code circuits for zero-overhead leakage reduction

AI summary

Overview

Research area: Quantum error correction (QEC) and fault-tolerant quantum computing, specifically leakage mitigation in dynamically generated (Floquet) codes and circuit-level code design.

Technical level: Advanced. The paper assumes familiarity with stabilizer formalism, instantaneous stabilizer groups, syndrome extraction circuits, and circuit-level noise models, although the motivation and headline results are stated in fairly plain language.

Scope (one sentence): The paper introduces and benchmarks two zero-overhead circuit implementations of the honeycomb Floquet code that periodically flush leakage while preserving distance to Pauli errors, and compares them against the walking surface code under four leakage noise models.

What This Paper Is About

Leakage is what happens when a qubit silently leaves the computational subspace (for example, an atom lost from a neutral-atom trap or a transmon driven into a higher energy level), and it is dangerous for QEC because it produces errors that are correlated across space and time, lowering both the threshold and the effective distance. The authors ask whether the honeycomb Floquet code (hFC), whose check operators are only weight-2, can be equipped with leakage reduction units that cost nothing in extra qubits, gates, or circuit depth, and whether such circuits beat the widely studied walking surface code at realistic error rates.

Key Contributions

1. The swirling hFC circuit. A zero-overhead leakage-reducing implementation of the P6 honeycomb Floquet code that uses only late-SWAP leakage reduction units, together with a periodic physical "swirling" rearrangement of qubits between check subrounds, intended for reconfigurable architectures such as neutral atoms.

2. The sliding hFC circuit. A second implementation for fixed-qubit architectures with square-lattice nearest-neighbor connectivity, which combines early-SWAP and late-SWAP gadgets in a single circuit for the first time, causing the code patch to translate diagonally by one unit every two subrounds.

3. Distance preservation. Unlike previous dynamic circuits proposed for the hFC, both new circuits natively preserve the timelike and spacelike code distances under Pauli noise; the authors provide fault-tolerance proofs for the late- and early-SWAP syndrome extraction circuits in Appendix C using the methods of Ref. [36].

4. Analytical and numerical leakage benchmarking. The paper derives upper bounds on the minimum malignant fault weight for both hFC variants and the walking surface code under each leakage model, and simulates all three at distances $d=6$ and $d=12$ under four noise models.

Main Findings

• Qubit lifetime is halved: Qubits in the walking surface code experience a lifetime of 8 gates between initialization and measurement, whereas the swirling hFC gives a lifetime of 4 gates and the sliding hFC averages 4 gates (non-uniform across the patch). This is an intrinsic advantage because a leaked qubit can only propagate errors over the gates it subsequently touches.

• Asymptotic predictions (upper bounds on $w_\text{min}$): For depolarizing leakage, both the hFC and the surface code are bounded at $\lceil d/4 \rceil$. For tailored leakage, the hFC is bounded at $\lceil d/2 \rceil$ versus $\lceil d/4 \rceil$ for the surface code. For skip-gate leakage, both are bounded at $\lceil d/2 \rceil$. The authors stress these are upper bounds from adversarially planted faults, not proven minimum weights.

• Depolarizing leakage produces a crossover the asymptotics do not predict: Although the asymptotic bounds for the two code families coincide, the swirling hFC shows a steeper fitted scaling exponent $\alpha$ than the surface code at practical error rates, and below a crossover error rate the swirling hFC achieves lower logical error rates. The sliding hFC also scales more steeply than the surface code, but by a smaller margin.

• Tailored leakage shows a delayed crossover: Asymptotics predict a large hFC advantage, but in the simulated regime the surface code's scaling is only slightly weaker than the hFC's because the surface code's logical error rate is dominated by higher-weight configurations; the crossover therefore occurs at a lower physical error rate than under depolarizing leakage.

• Skip-gate leakage favors the surface code: All codes stay close to the expected $\lceil d/2 \rceil$ scaling, and the surface code wins because it has fewer weight-$\lceil d/2 \rceil$ mechanisms that cause logical failure; the authors do not expect a crossover in this case.

• 10% leakage transport reverses that conclusion: Adding leakage transport with $p_\text{transport} = 10%$ lowers the scaling exponents of both codes, but degrades the surface code more than the hFC, producing a crossover below which the hFC wins again.

• Entropic ("waterfall") effects grow with distance: Observed scaling at $d=12$ generally deviates more from asymptotic predictions than at $d=6$, most clearly under depolarizing leakage, so asymptotic scaling alone is insufficient to rank codes.

• Qubit overhead differs between variants: With $2d^2$ data qubits, the swirling hFC equips all $3d^2$ edges with ancillas for $5d^2$ physical qubits total, while the sliding hFC reuses ancillas and needs only $d^2$ of them, for $3d^2$ total, compared with $2d^2$ for the walking surface code.

• Not reported in the available text: Specific logical error rate values, threshold error rates, and the numerical range of simulated physical error rates are not given in the excerpt; these are presumably in figures and Appendix F, which are truncated. The paper states only that error rates were simulated at $d=6$ and $d=12$ over $3d$ QEC rounds.

Methodology in Plain English

The hFC measures only two-qubit check operators, cycling through $XX$ on red edges, $YY$ on green edges, and $ZZ$ on blue edges according to a three-step periodic schedule. The authors exploit two circuit identities: a SWAP can be compiled into gates already present in the syndrome extraction circuit, and a SWAP placed at certain positions in that circuit becomes trivial. This lets them insert a SWAP between a data qubit and a freshly initialized ancilla — the essence of a leakage reduction unit — at zero cost in qubits, gates, or depth. Placing the SWAP at the end of the circuit gives the "late-SWAP" gadget; placing it just after the first gate gives the "early-SWAP" gadget. The swirling circuit uses late-SWAPs on a bipartite coloring of data qubits, resetting black qubits on even subrounds and gray qubits on odd subrounds, with a conceptual physical rearrangement restoring the layout. The sliding circuit uses both gadget types, assigned according to which qubit adjacencies exist on a square lattice, so that the patch distorts and then restores itself up to a diagonal translation. For the analysis, the authors count gate lifetimes analytically, plant adversarial leakage faults along minimum-length nontrivial loops and add faults until decoding fails (giving upper bounds on $w_\text{min}$), and then run circuit-level simulations of both hFC variants and the walking surface code at $d=6$ and $d=12$, fitting logical error rates to a power law $\sim p^\alpha$ to extract effective scaling exponents.

Why This Matters

Leakage is becoming a larger share of the error budget as Pauli error fidelities improve, so QEC protocols increasingly need co-designed leakage mitigation rather than Pauli-only design. This work shows that a weight-2-check Floquet code can flush leakage faster than the walking surface code at no circuit overhead, and that finite-error-rate entropic effects can make the hFC the better choice in regimes where asymptotic analysis says it should not be.

• Neutral-atom quantum computers, where atom loss is a native leakage mechanism and where the swirling circuit's local rearrangement can be folded into the atom movement already required for two-qubit gates.
• Superconducting transmon processors, where heating out of the computational subspace is a known leakage source and where fixed-connectivity variants of the sliding circuit are relevant.
• Hardware with fixed heavy-hex or square connectivity, for which the paper sketches circuit variants (Appendix D) that forgo physical motion at the cost of one extra teleportation gate.
• Benchmarking and co-design of QEC stacks, since the results argue for comparing codes at realistic rather than asymptotic error rates when selecting an architecture.

Industry relevance: The comparison targets a practical decision — hFC versus surface code — for near-term fault-tolerant hardware, and quantifies the trade in physical qubit overhead ($3d^2$ or $5d^2$ versus $2d^2$) against faster leakage removal and steeper sub-threshold scaling.

Future Directions

• Planar variants with boundaries. The analysis is done on the toric honeycomb lattice; benchmarking a planar code with boundaries is needed for practical implementation, and the authors expect equal or better scaling because boundary stabilizers are generally lower weight.
• Better low-error-rate simulation and ansätze. More accurate simulators for low error rates and better logical-error-rate ansätze for leakage noise are needed, since the splitting method cannot be applied directly.
• Exploiting leakage detection information. Analyzing gains from erasure checks or three-state readout for leakage detection, which may require new decoding strategies to use the extra measurement information optimally.
• Searching for better dynamical schedules. Since weight-2 checks appear to limit leakage spread, the authors ask whether other low-weight-check dynamical or subsystem code schedules could beat the walking surface code in even more regimes, including newer constructions from Refs. [41, 3].
• Pareto-frontier optimization. Jointly optimizing circuit structure, leakage effects, connectivity constraints, and decoding to map the frontier of favorable architectures under different hardware constraints.

Target Audience

Researchers and graduate students in quantum error correction and fault-tolerant quantum computing, especially those working on Floquet codes, dynamical codes, and leakage-aware circuit design; hardware teams on neutral-atom and superconducting platforms evaluating code choices; and decoder developers who need to understand how leakage-induced correlated faults appear in syndrome data. Readers without a QEC background will find the motivation accessible but the technical content demanding.

Authors’ abstract

Leakage, occurring when a qubit undetectably exits the computational subspace, poses a significant challenge for quantum error correction by inducing correlated errors in space and time. These correlations reduce both the code threshold and the effective code distance. To address this challenge, we introduce two zero-overhead walking circuits for the honeycomb Floquet code (hFC), termed the swirling and sliding circuits, which periodically remove leakage while dynamically protecting logical qubits via a schedule of anticommuting two-body measurements. Unlike previous dynamic circuits for the hFC, ours preserves the distance to Pauli errors. We further study their performance under leakage and find that, for two leakage noise models, they may correct as many leakage errors as Pauli errors, improving on the more widely known walking surface code. Through numerical simulation, we observe that finite-error-rate performance under leakage is strongly influenced by entropic effects, producing a pronounced waterfall regime in which the logical error rate decreases much more rapidly with physical error rate than the expected distance-limited scaling. In this regime, even when asymptotic predictions suggest otherwise, the hFC can outperform the same-distance walking surface code in error rates and sub-threshold logical error scaling.