Pre-Distillation of Magic States via Composite Schemes

TL;DR

This work addresses the high overhead of magic-state distillation by introducing a platform-aware pre-distillation framework that synthesizes robust $\mathcal{T}$ (and $\mathcal{H}$) gates directly via composite pulses. By canceling leading-order global errors in X–Y, X–Z, and integrated-photonics settings, the authors demonstrate substantial reductions in MSD depth—up to three levels—translating to exponential savings in physical qubits while maintaining fault-tolerance. They formalize a T-magic error metric, establish analytic and numerical constructions for 3- and 5-segment sequences, and show linear scaling of gate-channel errors with leading $T$-state imperfections, reinforcing the practical value of pre-distillation. Across superconducting, trapped-ion, neutral-atom, and photonic platforms, the approach directly lowers the resource cost of universal quantum computation and offers a scalable path toward more resource-efficient architectures. Formally, the results exploit the thresholds and quadratic/cubic error scalings of MSD (e.g., $\epsilon'(b5)=5\u03b5^2+O(b5^3)$ for T-states) to quantify how improved gate fidelities reduce distillation rounds and the associated qubit overhead.

Abstract

Magic state distillation (MSD) is a cornerstone of fault-tolerant quantum computing, enabling non-Clifford gates via state injection into stabilizer circuits. However, the substantial overhead of current MSD protocols remains a major obstacle to scalable implementations. We propose a general framework for pre-distillation, based on composite pulse sequences that suppress systematic errors in the generation of magic states. Unlike typical composite designs that target simple gates such as $X$, $Z$, or Hadamard, our schemes directly implement the non-Clifford $\mathcal{T}$ gate with enhanced robustness. We develop composite sequences tailored to the dominant control imperfections in superconducting, trapped-ion, neutral-atom, and integrated photonic platforms. To quantify improvement in the implementation, we introduce an operationally motivated fidelity measure specifically tailored to the $\mathcal{T}$ gate: the T-magic error, which captures the gate's effectiveness in preparing high-fidelity magic states. We further show that the error in the channel arising from the injection of faulty magic states scales linearly with the leading-order error of the states. Across all platforms, our approach yields high-fidelity $\mathcal{T}$ gates with reduced noise, lowering the number of distillation levels by up to three. This translates to exponential savings in qubit overhead and offers a practical path toward more resource-efficient universal quantum computation.

Figures (16)

• Figure 1: Five-qubit $T$-state distillation and its improvement through pre-distillation. (a) Conceptual schematic of a single round of the $T$-magic-state distillation protocol. One round consumes five noisy copies of $\rho_T(\epsilon) \equiv (1-\epsilon)\ket{T_0}\bra{T_0} + \epsilon \ket{T_1}\bra{T_1}$ (denoted by $\rho_T$ with heavily wavy red symbols to emphasize noise). With probability $p(\epsilon) = \tfrac{1}{6} + O(\epsilon)$, the protocol outputs the logical state $\rho^{(L)}_T(1-\epsilon')$, whose error is quadratically suppressed, $\epsilon'(\epsilon) = 5\epsilon^2 + O(\epsilon^3)$. We represent this improved state using green symbols with reduced waviness, reflecting its lower noise. (b) Distillation curve $\epsilon'(\epsilon)$ (blue): the output error rate as a function of the input error rate, showing quadratic suppression after one round. For reference, the red line shows the identity function $f(\epsilon) = \epsilon$, highlighting the improvement gained over the uncorrected case. (c) Required number of recursive distillation rounds to achieve an error below the $10^{-15}$ fault-tolerance threshold, illustrating the exponential scaling of qubit overhead with the initial error $\epsilon$. (d) Comparison of $\mathcal{T}$-gate implementations under systematic width errors in directional couplers. The plot shows the number of distillation levels needed to reach $10^{-15}$ error. Results are shown for a two-segment design (black) and three composite four-pulse schemes: (A) blue, (B) red, (C) brown.
• Figure 2: Improvements enabled by composite pulse sequences for implementing the $\mathcal{T}$ gate in $X$-$Y$ Systems with Global Rabi Frequency Errors. The black curves correspond to the single-pulse implementation, while the blue, red, and brown curves correspond to the three-, five-, and seven-pulse composite gates. (a) Comparison of Frobenius distance fidelity $\mathcal{F}_{\text{frob}}$ (solid lines) and trace distance fidelity $\mathcal{F}_{\text{trace}}$ (dashed lines) for single-, three-, five-, and seven-pulse implementations under global systematic errors. (b) Reduction in the number of MSD iterations required to achieve a target gate error of $10^{-15}$, where composite sequences reduce the required levels by one, two, or even three in some error regimes.
• Figure 3: Improvements enabled by composite pulse sequences for implementing the $\mathcal{H}$ gate in $X$-$Y$ Systems with global Rabi frequency errors. The black curves correspond to the single-pulse implementation, while the blue, red, and brown curves correspond to the three-, five-, and seven-pulse composite gates. (a) Comparison of Frobenius distance fidelity $\mathcal{F}_{\text{frob}}$ (solid lines) and trace distance fidelity $\mathcal{F}_{\text{trace}}$ (dashed lines) for single-, three-, five-, and seven-pulse implementations under global systematic errors. (b) Reduction in the number of MSD iterations required to achieve a target gate error of $10^{-15}$. In certain error regimes, composite sequences lower the required distillation levels by one or two.
• Figure 4: Performance of $\mathcal{T}$ gate implementations using composite pulse sequences under correlated systematic $Z$ errors in $X$-$Z$ Hamiltonians. The black curve shows the two-segment baseline, while the red, blue, and brown curves correspond to three different three-pulse designs labeled (a), (b), and (c). Panel (a) presents the Frobenius fidelity $\mathcal{F}_{\text{frob}}$ across these schemes. Panel (b) shows the corresponding reduction in the number of magic state distillation levels required to achieve a gate T-magic error below $10^{-15}$, based on Eq. (\ref{['eq:natural error of the gate']}).
• Figure 5: Performance of $\mathcal{T}$ gate implementations using composite pulse sequences under global systematic width errors in directional couplers. The two-segment design (black) is compared with three four-pulse composite schemes: (a) blue, (b) red, and (c) brown. While scheme (a) shows an initial advantage in fidelity, scheme (c) offers superior performance over the entire error range in terms of the number of concatenation levels.