arXiv:2610.19299
Gap-Priced $|T\rangle$-States: Dynamic $|T\rangle$ Post-Selection and Budget-Aware Factory Provisioning
Maxwell Poster, Sayam Sethi, Steven Hofmeyr, Yuhao Liu, Jonathan Baker, Costin Iancu
Abstract
Magic $|T\rangle$-state preparation dominates the cost of fault-tolerant quantum computation on surface codes, yet cultivation and distillation are typically provisioned statically, leaving much of the program's error budget unused. We present Gap-Priced $|T\rangle$ States (GP$|T\rangle$-S), a lattice surgery scheduler and factory allocator that co-optimizes both methods under a total $T$-error budget. GP$|T\rangle$-S uses the complementary gap of each cultivated state as a fidelity estimate and dynamically adjusts the cultivation gap cutoff, converting budget slack into speed by giving the scheduler control over the cultivation post-selection. GP$|T\rangle$-S also characterizes distillation factories by size, $|T\rangle$-state production rate, and budget slack, and introduces cultivating factories that embed cultivation within routing channels, achieving fidelities otherwise requiring two-level distillation at a single-level footprint. GP$|T\rangle$-S organizes these characterizations into a factory catalog, and performs distillation factory allocation over this catalog in tandem with the gap cutoff using dynamic-programming to achieve one of i) minimum spacetime volume, ii) minimum execution time, or iii) minimum space overhead, all within a user-provided $T$-error budget. Across sixteen benchmarks under high and low $T$-error budgets of $10^{-3}$ and $10^{-5}$ respectively, GP$|T\rangle$-S consistently outperforms cultivation-only and single-factory baselines without exceeding the budget, achieving a 5.5$\times$ geometric mean speedup in logical cycle count and a 3.4$\times$ geometric mean reduction in space-time volume.
AI Summary
The paper addresses joint provisioning and scheduling of magic-state cultivation and distillation for surface-code computation under a total $T$-error budget. Its scheduler, GP$|T\rangle$-S, uses decoder complementary gaps to estimate cultivated-state error rates and adjusts post-selection thresholds during execution, spending unused error budget to increase supply. It also introduces cultivation inside distillation-factory routing channels and selects factory configurations from a catalog indexed by footprint, output rate, and infidelity. Cultivation statistics come from existing end-to-end stabilizer simulations; distance-dependent factory fidelities are obtained through density-matrix simulations at physical error rate $p=10^{-3}$. An exact dynamic program computes a Pareto front over qubit count, production rate, and infidelity-weighted rate. A faster allocator relaxes placement constraints and subsequently places factories greedily. Characterized cultivation-fed 15-to-1 factories reach modeled output infidelities down to approximately $2\times10^{-14}$. Across 16 compiled benchmarks with $1.4\times10^4$–$8.2\times10^5$ $T$ gates and budgets $10^{-3}$ and $10^{-5}$, simulations report geometric-mean improvements of $5.5\times$ in logical cycles and $3.4\times$ in spacetime volume. Comparisons use fixed-cutoff cultivation where feasible, otherwise one smallest feasible factory; reported expenditure stays within budget. The budget covers the $T$-state error contribution, not total computation failure. Cutoff selection uses aggregate error rates in expectation; the paper notes possible overspending for short programs or near completion and proposes feasibility guards. High-gap calibration has finite-sample irregularities, factory logical errors limit gains from better inputs, and the relaxed allocator can overprovision space relative to the exact search.
Central claim may not hold: The claimed total T-error-budget compliance is not established beyond estimated-error accounting, because the paper acknowledges possible overspending under suffix-average acceptance and does not specify or verify its proposed safeguard. The allocator's exactness concerns aggregate catalog statistics rather than actual routed execution objectives, so the abstract's unqualified minimum-time and minimum-spacetime claims are stronger than the demonstrated optimization. The paper inconsistently specifies the factory measurement distance as d_m = d_x and d_m = d_z, leaving a production-rate assumption used in provisioning unresolved. The table's cultivating-factory and bare-cultivator costs contradict the figure's blanket claim that cultivating factories dominate bare cultivation in qubit-cycles per output state.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 17 pages, 13 figures