PaperlayerView exact revision on arXiv

Source-linked AI summary

STAR-Magic Mutation: Even More Efficient Analog Rotation Gates for Early Fault-Tolerant Quantum Computer

Riki Toshio, Shota Kanasugi, Jun Fujisaki, Hirotaka Oshima, Shintaro Sato, Keisuke Fujii

arXiv:2603.22891v1quant-ph

TL;DR

Early fault-tolerant quantum computers need lower-overhead methods for magic-state preparation and arbitrary-angle rotation gates. The paper combines transversal multi-rotation with magic state cultivation in STAR-magic mutation and integrates it into STAR ver. 3. The resulting protocol improves rotation-gate error scaling and supports many-body simulations beyond exact classical simulation at p_ph=10^-3.

  • Problem

    Magic state distillation has substantial spacetime overhead, while arbitrary-angle rotations require costly gate decompositions and remain important in Trotter-based circuits.

  • Method

    STAR-magic mutation combines transversal multi-rotation with magic state cultivation, and STAR ver. 3 compiles circuits over Clifford+T+ϕ using SMM and MSC.

  • Results

    STAR-magic mutation achieves O(θ_L^{2(1-Θ(1/d))}p_ph) rotation-gate error using one surface code patch, while STAR ver. 3 supports many-body simulation beyond exact classical simulation at p_ph=10^-3.

  • Takeaways & Limitations

    The architecture improves analog-rotation performance and enables practical early-FTQC simulations under a physical error rate previously considered too optimistic for comparable STAR tasks.

  • Takeaways & Limitations

    The error-mitigation strategies require efficient tomography to estimate resource-state error parameters and residual errors.

Abstract

from arXiv · show

We introduce STAR-magic mutation, an efficient protocol for implementing logical rotation gates on early fault-tolerant quantum computers. This protocol judiciously combines two of the latest state preparation protocols: transversal multi-rotation protocol and magic state cultivation. It achieves a logical rotation gate with a favorable error scaling of $\mathcal{O}(θ_L^{2(1-Θ(1/d))}p_{\text{ph}})$, while requiring only the ancillary space of a single surface code patch. Here, $θ_L$ is the logical rotation angle, $p_{\text{ph}}$ is the physical error rate, and $d$ is the code distance. This scaling marks a significant improvement over the previous state-of-the-art, $\mathcal{O}(θ_L p_{\text{ph}})$, making our protocol particularly powerful for implementing a sequence of small-angle rotation gates, like Trotter-based circuits. Notably, for $θ_L \lesssim 10^{-5}$, our protocol achieves a two-order-of-magnitude reduction in both the execution time and the error rate of analog rotation gates compared to the standard $T$-gate synthesis using cultivated magic states. Building upon this protocol, we also propose a novel quantum computing architecture designed for early fault-tolerant quantum computers, dubbed ``STAR ver.~3". It employs a refined circuit compilation strategy based on Clifford+$T$+$φ$ gate set, rather than the conventional Clifford+$T$ or Clifford+$φ$ gate sets. We establish a theoretical bound on the feasible circuit size on this architecture and illustrate its capabilities by analyzing the spacetime costs for simulating the dynamics of quantum many-body systems. Specifically, we demonstrate that our architecture can simulate biologically-relevant molecules or lattice models at scales beyond the reach of exact classical simulation, with only a few hundred thousand physical qubits, even assuming a realistic error rate of $p_{\text{ph}}=10^{-3}$.

I. INTRODUCTION

Early fault-tolerant quantum computers face substantial overheads for magic-state preparation and arbitrary-angle rotation gates. STAR-magic mutation addresses these bottlenecks with improved error scaling and supports STAR ver. 3, which targets practical many-body simulation at realistic physical error rates.

  • Motivation: Magic state distillation can require more than 30d^3 qubit-cycles for p_L<10^-10, motivating lower-overhead alternatives for early fault-tolerant hardware.Magic state cultivation prepares moderately clean states using one surface code patch and can reduce subsequent distillation overhead.
  • Motivation: Arbitrary-angle rotations remain costly because conventional fault-tolerant implementations decompose them into long Clifford-based gate sequences, despite their prevalence in Trotter and variational circuits.The STAR approach instead directly implements analog rotations using resource states.
  • STAR-magic mutation: RUS feedback events produce the prior O(θ_L p_ph) error scaling because rare order-unity feedback angles incur physical-error-rate errors.These events dominate the average rotation-gate error.
  • STAR-magic mutation: STAR-magic mutation combines transversal multi-rotation and magic state cultivation to achieve O(θ_L^{2(1-Θ(1/d))}p_ph) error scaling using one surface code patch.The protocol uses analog execution below a tunable threshold and switches strategies when the RUS angle exceeds that threshold.
  • Performance: Compared with cultivation-only implementations, SMM reduces analog-rotation execution time by two orders of magnitude and achieves error rates several orders of magnitude smaller.Against the earlier TMR-only STAR approach, it reduces errors by more than one order of magnitude while increasing execution time by only a few tens of percent for θ_L≲10^-4.
  • STAR ver. 3: STAR ver. 3 compiles circuits into Clifford+T+ϕ sequences and supports analog rotations with SMM alongside digital rotations with MSC.Its feasible circuit size is bounded by error-mitigation cost and the L1-norm of the target Hamiltonian.
  • STAR ver. 3: STAR ver. 3 can simulate many-body systems beyond exact classical simulation at p_ph=10^-3, including a 72-spin-orbital [4Fe-4S] cluster using 1.9×10^5 physical qubits for T=10 a.u. within a week.Earlier STAR studies required the more optimistic p_ph=10^-4 for comparable tasks.

B. Non-fault-tolerant implementation of analog rotation gates

STAR ver. 2 implements arbitrary multi-Pauli rotations by teleporting resource states prepared with the transversal multi-rotation protocol. TMR post-selects stabilizer outcomes to retain the target state, but physical noise leaves finite residual infidelity.

  • Resource-state implementation: Resource states |mθL⟩L encode arbitrary logical Z rotations and enable gate teleportation onto any target state.The same resource-state approach extends to multi-Pauli rotations through a multi-Pauli measurement circuit.
  • Transversal multi-rotation: TMR prepares resource states by partitioning logical-Z support into disjoint subsets and applying multiple transversal rotations.For k=2, the construction uses multi-Z rotations composed of nearest-neighbor CNOT and single-qubit Z-rotation gates.
  • Transversal multi-rotation: Stabilizer post-selection extracts the target logical resource state by rejecting outputs with nontrivial error syndromes.In the ideal limit, the post-selection succeeds with probability pideal; under physical noise, erroneous states can overlap the logical subspace.
  • Error scaling: The leading TMR error coefficient scales as O(θ2/k_L p_ph), producing finite resource-state infidelity.A hardware-specific dissipative-noise modification can improve the infidelity scaling to O(θ2(1−1/k)_L p_ph).
  • Optimization and overhead: Optimized post-selection checks only critical stabilizers, while subsequent quantum error correction handles other detected errors.For p_ph = 10^-3 and d = 11, the optimized protocol is reported to complete in about 1.3 clocks, versus around 10 clocks for MSC.
  • Optimization and overhead: TMR uses a single ancilla QEC block and supports parallel rotation gates through local ancilla space for lattice surgery.This contrasts with MSD-based architectures, where parallel T gates require substantially more physical qubits for magic-state supply.

3. Effective quantum channel and probabilistic inverse rotation

The noisy gate channel contains coherent over-rotation errors that dominate the RUS process and yield O(θ_L p_ph) worst-case scaling. Probabilistic coherent-error cancellation removes the leading coherent contribution, but higher-order terms matter for very small target angles.

  • Effective channel: Gate teleportation with a noisy resource state produces an effective channel combining ideal rotation with a stochastic error channel.The channel is analyzed when teleportation succeeds on the first trial and then extended through the RUS process.
  • Effective channel: Coherent over-rotation dominates the noisy channel, giving a worst-case RUS error rate of O(θ_L p_ph).The scaling arises because inverse and forward RUS rotations occur with equal probability while rare large-angle feedback events contribute errors of O(p_ph).
  • Probabilistic inverse rotation: PCEC applies a stochastic inverse-rotation channel after the noisy rotation to cancel its leading coherent error term.The composed channel is analyzed through an effective error channel for the corrected rotation.
  • Probabilistic inverse rotation: When θ_L can be smaller than p_ph, the residual O~(θ_L p_ph^2) term is not necessarily negligible and motivates modifying PCEC.The modification is introduced to cancel higher-order errors before the RUS process reaches angles of order unity.

4. Resilience to control errors

TMR is relatively vulnerable to coherent over-rotation errors, although randomizing rotation directions suppresses leading effects. Residual control errors can still drift over time and require calibration during computation.

  • Control-error model: Physical control imperfections include pulse-calibration errors, quantization-axis misalignment, and qubit crosstalk.These effects are not captured by assuming only stochastic Pauli error channels.
  • Coherent over-rotation: TMR is particularly weak against coherent over-rotation errors, which shift the logical rotation angle.The logical-angle shift is proportional to the relative error in the input physical rotation angles.
  • Mitigation: Randomizing each transversal rotation direction suppresses the leading coherent-error amplitude at the logical level.The residual error can also be calibrated by measuring the resulting rotation angle with quantum phase estimation.
  • Mitigation: Time-varying device error parameters prevent perfect removal of control errors, motivating QPE-based calibration within auxiliary computational space.The paper frames this as supporting transient simulations of classically intractable circuits within limited qubit budgets.

III. STAR-MAGIC MUTATION: HIGH-FIDELITY LOGICAL ROTATION GATE WITH A SMALL ANGLE

STAR-magic mutation combines TMR with magic state cultivation to suppress higher-order errors in logical analog rotations. It uses TMR below an angle threshold and T-gate synthesis above it, while accounting for cultivation’s preparation stages and post-selection statistics.

  • Proposal: STAR-magic mutation combines TMR and magic state cultivation so their complementary strengths address logical-rotation errors and overhead.The proposal modifies PCEC to cancel higher-order coherent terms before deriving its error scaling.
  • Magic state cultivation: Magic state cultivation comprises injection, iterative check-grow-stabilize cultivation, and escape into a larger QEC code.The escape stage grafts a large surface code onto the cultivated color-code state before transforming it into a fully matchable code.
  • Magic state cultivation: Cultivation simulations report p_m = 2 × 10^-9 at p_ph = 10^-3 with d = 15, but a 99% discard rate and roughly t_m = 10 clocks per state.The reported spacetime estimate is acknowledged as somewhat underestimated because limited-space packing of multiple cultivation processes is ignored.
  • Error model: TMR output is a mixture of the target resource state and non-target states whose error amplitudes depend on their order in p_ph.The coefficient q_j includes sampling and post-selection contributions, with q_j scaling as O(θ2n_j p_ph^j).
  • Higher-order error cancellation: The resulting effective channel is formulated by composing the noisy rotation channel with the correction channel and evaluating its residual error terms.The construction assumes the relevant physical-error expansion and uses the post-selection coefficients to characterize the effective logical error.
  • Higher-order error cancellation: The protocol applies a correction channel using ideal rotations R_−Δ_j to cancel higher-order coherent errors in the noisy resource-state channel.These correction rotations are implemented through T-gate synthesis using high-fidelity states from MSC or MSD.

C. STAR-magic mutation

STAR-magic mutation combines analog TMR-based RUS with a digital T-gate-synthesis stage, switching when the RUS angle reaches a tunable threshold. This hybrid protocol improves error scaling while using only one ancillary surface-code patch.

  • Two-stage protocol: SMM uses TMR below θth and switches to fault-tolerant T-gate synthesis when θRUS ≥ θth.Digital-stage T-gates use magic states prepared by MSC or MSD.
  • Two-stage protocol: The probability of entering the digital stage is pswitch ≃ O(θth/θL), so threshold selection limits synthesis overhead.The digital stage is therefore rarely required when the threshold sufficiently exceeds the target angle.
  • Resource requirements: SMM can use MSC locally within one ancillary surface-code patch and parallelize multiple rotation gates through locality-aware preparation.The same locality supports reduced time overhead in Trotter simulation via parallel resource-state preparation.
  • Error scaling: With fixed threshold-to-angle ratio, the effective error scales as O(θL^{2(1−1/k)}p_ph), improving over the prior O(θLp_ph) scaling.The theorem assumes noiseless magic states and neglects T-synthesis approximation error.

D. Numerical results

Numerical results evaluate RUS overhead and effective error under perfect and finite-error magic-state supplies. They show that threshold choice controls both the RUS factor and the balance between analog-stage and digital-stage errors.

  • Perfect magic states: With fixed ratio r, the numerical RUS factor follows the small-angle asymptotic behavior expected from Theorem 1.Figure 7 compares α_RUS, P_L, and the predicted small-angle scaling.
  • Perfect magic states: For p_m = 0, STAR-magic mutation achieves α_RUS ranges of 0.2–0.4 at θth = 0.05 and 0.05–0.11 at θth = 0.01.STAR ver. 2 gives 1.5 < α_RUS < 1.8 for k = 5, 7, 9.
  • Finite magic-state error: For p_m = 2×10^-9, α_RUS increases when the target angle becomes sufficiently small under fixed-ratio thresholding.The increase occurs because accumulated digital-stage error surpasses analog-stage error.
  • Threshold optimization: Optimizing θth balances accumulated analog- and digital-stage errors against execution time and total logical error.Figure 9 varies θth = 2^nθL for n = 0, 1, …, 15 to display this tradeoff.

IV. NOVEL EARLY-FTQC ARCHITECTURE: STAR VER. 3

STAR ver. 3 refines the STAR architecture by combining local analog rotations with fault-tolerant treatment of larger-angle non-Clifford operations. Its design addresses the limitations of TMR fidelity for large-angle rotations while retaining STAR’s locality advantages.

  • Motivation: STAR’s TMR protocol implements arbitrary logical rotations locally, but its fidelity deteriorates for large-angle rotations, especially T-gates.This creates a motivation for combining STAR-style analog rotations with cultivated magic states.
  • Motivation: MSC prepares high-fidelity magic states locally without large magic-state factories, preserving STAR’s removal of factories and parallelizable rotations.The paper identifies this compatibility as the motivation for refining the STAR architecture.

A. Definition

STAR ver. 3 uses a Clifford+T+ϕ compilation strategy that assigns small-angle rotations to SMM and T-gates or large-angle rotations to fault-tolerant synthesis. The architecture also supports local preparation and error mitigation.

  • Definition: The Clifford+T+ϕ gate set decomposes circuits into Clifford operations, T-gates, and analog rotations, with their ratio chosen to minimize total time overhead.The optimization includes sampling cost for error mitigation.
  • Definition: T-gates use MSC or MSD, while large-angle analog rotations use fault-tolerant Clifford+T decompositions.Small-angle analog rotations are implemented with STAR-magic mutation instead.
  • Definition: Residual T-gate and analog-rotation errors are handled with error-mitigation techniques such as probabilistic error cancellation.The architecture’s execution flow is summarized in Fig. 10.
  • Capabilities: Clifford+T+ϕ enables circuits combining T-gates with small-angle rotations, including typical block-encoding and Trotter-based circuits.These circuit classes are described as more complex than those supported by previous STAR architectures.
  • Capabilities: STAR ver. 3 retains locality of magic- and resource-state preparation, supporting parallel non-Clifford operations without conventional factory-mediated routing.The architecture’s locality is presented as a major advantage over MSD-based architectures.

B. Error mitigation and theoretical bound on the feasible circuit size

STAR ver. 3 combines probabilistic error cancellation with a Clifford+T+ϕ compilation strategy and bounds feasible circuit sizes by total mitigation cost. Under the stated assumptions, it expands the supported circuit regime and reduces rotation-gate error and execution time relative to prior architectures.

  • Theoretical bound: The resulting sampling overhead restricts simulatable circuit sizes, creating a fundamental limitation of STAR ver. 3.The restriction follows from the requirement that total mitigation cost remain bounded rather than grow exponentially.
  • Architecture: STAR ver. 3 compiles circuits into Clifford+T+ϕ and assigns Clifford, T, and analog-rotation gates to specialized execution procedures.The flowchart describes standard lattice surgery or transversal Clifford gates, cultivated or distilled magic states for T-gates, and dedicated handling for rotation gates.
  • Error mitigation: Probabilistic error cancellation estimates noise-free observables by sampling correcting Pauli-Z operations, but amplifies estimator variance by γ_T^2.The method virtually realizes inverse noise channels through measurement outcomes rather than direct unitary sampling.
  • Theoretical bound: For d = 21, k = 7, and θ_th = 0.01, STAR-magic mutation gives α_max ≃ 0.1 and permits STAR ver. 3 simulations with λ = 100 up to T ≲ 100 at p_ph = 10^-3.This regime is stated to avoid an exponential cost of error mitigation.
  • Architecture comparison: STAR ver. 3 covers a vastly broader class of circuits than previous STAR architectures and outperforms MSC-based FTQC when small-angle rotations predominate.The reported comparison also gives over two orders of magnitude reductions in total error rate and rotation-gate execution time.

V. PRACTICAL APPLICATIONS

The paper evaluates STAR ver. 3 for quantum many-body dynamics, including molecular and spin-system simulations relevant to nonequilibrium phenomena and spectroscopy. These applications are suitable because their algorithms require many small-angle rotation gates, which STAR ver. 3 is designed to execute efficiently.

  • Applications: The application study estimates spacetime costs for simulating quantum many-body dynamics in molecules and spin systems.The motivation includes nonequilibrium phenomena such as chemical reactions and ultrafast dynamics, alongside NMR and neutron-scattering data interpretation.
  • Applications: Trotterization and randomized many-body algorithms require many small-angle rotation gates, making them suitable targets for STAR ver. 3.The architecture avoids costly T-gate synthesis and magic state distillation for these gates.
  • Method: The resource analysis uses TE-PAI, whose circuit depth and sampling overhead depend on evolution time T and Hamiltonian L1-norm λ rather than detailed Hamiltonian structure.The section presents TE-PAI as the randomized algorithm used for the many-body dynamics estimates.

A. TE-PAI: Exact quantum dynamics simulation by sampling random circuits

TE-PAI simulates quantum dynamics by randomly sampling circuits whose averaged estimator reproduces the target time-evolution superoperator. Its computational cost is controlled by the interpolation angle and a tunable parameter balancing gate count against sampling overhead.

  • Estimator construction: TE-PAI replaces rotation superoperators with unbiased estimators built from randomly selected superoperators A, B_k, and C_k.The selected operations include identity, a signed small-angle rotation, and a π/2 rotation.
  • Estimator construction: The estimator satisfies E[Û] = U, enabling expectation values of time-evolved observables to be obtained from sampled quantum circuits.The sampled superoperators are selected with probabilities proportional to the absolute values of their coefficients.
  • Cost analysis: In the N → ∞ limit, the expected number of non-trivial gates scales linearly with evolution time T and Hamiltonian norm λ.This scaling is identified as an efficiency feature of TE-PAI, subject to practical angle-scaling requirements.
  • Cost analysis: The interpolation angle must scale as O(1/(λT)) to avoid exponential sample complexity, while TE-PAI trades circuit depth against measurement overhead through Δ and Q.Q is tunable, and the reported total gate count is (2(λT)^2/Q + Q)e^Q/ϵ^2.
  • Compatibility with STAR: Fixing one small rotation angle makes TE-PAI circuits well suited to STAR and reduces the cost of calibrating physical-level rotation angles.The circuits combine Pauli-string operations with a unique fixed small-angle rotation.

B. Resource estimation on STAR ver. 3: Simulation of quantum many-body dynamics

The resource analysis estimates STAR ver. 3 costs for TE-PAI simulations using code-distance, layout, and runtime assumptions. It finds feasibility for large molecular systems at hundreds of thousands of physical qubits, but substantial runtime constraints remain.

  • Resource estimation: TE-PAI costs are determined by evolution time T and the target Hamiltonian’s L1-norm λ.The analysis uses these quantities to determine computational resources for many-body dynamics simulations.
  • Runtime model: STAR ver. 3 estimates single-shot TE-PAI runtime as (2(λT)^2/Q + Q)C_smm clock cycles under the fast block layout.The analysis assumes C_smm ≃ 3 clocks for θ_th = 2^6·θ_L and a sufficient supply of magic and resource states.
  • Runtime model: The runtime-to-real-time conversion assumes one code cycle lasts 1 µs on current superconducting qubit chips.The resulting runtime is expressed as (2(λT)^2/Q + Q)C_smm d microseconds.
  • Molecular simulations: 1.0–2.6 × 10^5 physical qubits suffice for real-time simulations of large molecular systems such as [4Fe-4S] enzyme active centers.This is below the several-million-qubit scale typically required in prior FTQC studies.
  • Limitations: Keeping total runtime within one week requires λT below roughly 1.5 × 10^3, corresponding to T ≲ 10 a.u. for [4Fe-4S].The authors state that these resource analyses are preliminary and based on a simplified, not fully optimized approach.
  • STAR-magic mutation: STAR-magic mutation supports arbitrary analog rotations with error O(θ_L^(2(1−Θ(1/d)))p_ph) using one ancillary surface-code patch.The protocol adaptively switches between TMR and MSC-based T-gate synthesis during gate teleportation.
  • Architecture and scope: STAR ver. 3 combines Clifford, T, and φ gates, and TE-PAI analysis targets many-body Hamiltonians at classically intractable sizes with p_ph = 10^-3.The conclusion identifies this as an early-FTQC architecture capable of simulating diverse many-body dynamics under a realistic physical error rate.
  • Open issues: The study leaves optimal compilation, improved algorithms, tomography, and control-error mitigation as important unresolved issues.These issues are presented as future directions for reducing runtime and overcoming the feasible-circuit-size bound.

Appendix A: The spacetime costs of TE-PAI simulations of the 2D Hubbard model

The appendix evaluates TE-PAI resource costs for the 2D Hubbard model under fixed model, accuracy, and physical-error assumptions. It finds that systems with up to 100 sites fit within hundreds of thousands of physical qubits, while runtime becomes prohibitive for larger NT.

  • Model definition: The 2D Hubbard model uses hopping strength t, onsite interaction U, and adjacent-site pairs on a square lattice.The parameters t and U specify hopping and onsite Coulomb interaction, while Jordan–Wigner strings preserve fermionic commutation relations.
  • Model definition: Jordan–Wigner transformation represents the Hubbard Hamiltonian as a linear combination of Pauli-string operators.This representation supports the Pauli-based TE-PAI resource analysis.
  • Simulation settings: The appendix varies N from 4 to 100 sites with t = 1, U = 4, T ∈ [1, 20], p_ph = 10^-3, Q = 1, and ε = 0.05.The logical-qubit count is set to N_L = 2N.
  • Resource results: N ≤ 100 Hubbard sites can be simulated with less than 6 × 10^5 physical qubits.The result is shown for the stated TE-PAI settings and STAR ver. 3 resource model.
  • Runtime results: Total runtime exceeds one week when NT ≳ 300.The appendix identifies this runtime overhead as a motivation for approaches beyond TE-PAI.
  • Cost reduction: Parallel Trotter compilation and fermionic swap gates are proposed as routes to reduce Hubbard-model simulation cost.The appendix points to these approaches as exploiting the high parallelism of periodic models.
Loading 2603.22891v1…