Constant Space Overhead Magic State Distillation

• The paper presents protocols that achieve constant space overhead in magic state distillation, ensuring resource efficiency independent of target error rates.
• It employs asymptotically good codes and transversal non-Clifford gates to robustly suppress errors, enabling scalable fault-tolerant quantum computation.
• Innovative circuit designs and catalytic methods allow practical implementation across various hardware platforms while significantly reducing physical qubit requirements.

Constant space overhead magic state distillation denotes strategies and codes for generating high-fidelity non-Clifford resource states (such as $|T\rangle$ or $|CCZ\rangle$) with a number of physical qubits per output state that does not grow with the target error rate $\epsilon$ (i.e., space overhead scales as $\mathcal{O}(1)$ in $\epsilon$). These methods are essential for scalable fault-tolerant quantum computation, eliminating the super-polynomial or polylogarithmic spatial overhead of traditional multi-round or concatenated magic state distillers. Recent advances have established both explicit code constructions and protocol frameworks—on qubits and qudits—that realize strictly constant space overhead, including families of LDPC codes with transversal third-level gates, algebraic geometry codes, and catalytic conversions.

1. Definitions and Fundamental Metrics

The space overhead of magic state distillation protocols is formally defined as the number of physical qubits (or raw input magic states) consumed per output distilled state at the target error $\epsilon$:

$$C(\epsilon) = \min \left\{ \frac{\zeta}{\xi(\zeta)} : \text{protocol yields }\xi(\zeta)\text{ output states with error} \leq \epsilon \text{ from } \zeta\text{ inputs} \right\}$$

Traditional protocols exhibit $C(\epsilon) = \mathcal{O}(\log^\gamma(1/\epsilon))$ for various $\gamma>0$. In constant space-overhead protocols, $\gamma=0$: output error can be arbitrarily suppressed by increasing code distance or batch size, yet the $\zeta/\xi$ ratio is bounded (Wills et al., 2024, Cervia et al., 26 Dec 2025).

Additional key metrics:

• Time overhead: number of circuit rounds (depth) per output.
• Success probability: probability of accepting a batch; may trade off with space overhead in catalytic protocols (Fang et al., 2024).
• Error suppression: typically $p_{\text{out}} = \mathcal{O}(p^{d})$, for code distance $d$ and input error $p$.

2. Asymptotically Good Codes and Transversal Non-Clifford Gates

Protocols achieve constant space overhead by using quantum codes of finite (or scalable) block length that combine:

• Non-vanishing rate $(K/N \to \text{const} > 0)$: a fixed fraction of logical qubits per code block.
• Macroscopic distance $(d \sim \Theta(N))$: suppresses logical errors exponentially in code size.
• Transversal gates at $\mathcal{C}_3$: permits block-wise application of non-Clifford gates, notably CCZ or $T$, and magic state preparation at scale.

Prominent code constructions include:

• Algebraic geometry codes: e.g., over $\mathbb{F}_{2^{10}}$ yielding CSS quantum codes supporting transversal third-level gates on 1024-dimensional qudits, then mapped to 10-qubit blocks per qudit (Wills et al., 2024).
• Triorthogonal codes above the TVZ bound: explicit families over $\mathbb{F}_{64}$ constructed via lifting procedures using Artin–Schreier towers, yielding, e.g., $[[42,14,6]]_{64}$ with transversal CCZ, rate $1/3$ and distance $6$ (Cervia et al., 26 Dec 2025).
• Tricycle LDPC codes: generalizing bicycle codes to three homological dimensions, constructed via three-fold balanced products over Abelian groups; empirical cases include $[[72,9,(12,6)]]$, $[[192,27,(16,8)]]$, etc. All support transversal CCZ gates and constant-depth syndrome extraction (Menon et al., 14 Aug 2025).

3. Protocol Design and Circuit Architecture

Distinct protocol architectures are realized depending on code and platform:

• Tricycle code protocols use block-wise encoding, single-shot preparation, one round of constant-depth Z-syndrome extraction (using meta-checks), and depth-18 transversal CCZ followed by optional X-syndrome cleanup. No postselection is required, and error correction employs BP+OSD decoding over limited history (Menon et al., 14 Aug 2025).
• Qudit AG-code protocols encode $K$ logical CCZ states into $N$ physical via a triorthogonal CSS code, apply a transversal CCZ, and measure X & Z stabilizers; postselection yields output error $\mathcal{O}(p^d)$ with space overhead $O(1)$ (Cervia et al., 26 Dec 2025).
• Algebraic-geometry qudit codes on qubits: Each $2^{10}$-dim qudit encoded as 10 qubits, evoking high-overhead CCZ-state conversion recipes but maintaining constant net overhead per output (Wills et al., 2024).
• Unfolded Reed-Muller-based protocols for biased noise: Leverage X-stabilizer unfolding in 2D to prepare magic states with $O(1)$ overhead, exploiting noise bias and nearest-neighbor connectivity (Ruiz et al., 16 Jul 2025).

Circuit depths are fixed ($O(1)$) or grow polylogarithmically with target error; all protocols attain parallelizable measurement rounds and compressed resource usage.

4. Error Analysis, Success Probability, and Thresholds

Logical error rates are primarily determined by code distance and single-shot decoding fidelity:

• Tricycle codes with windowed BP+OSD decoding achieve robust error suppression: logical error rates per code block can reach $10^{-6}$–$10^{-10}$ at physical error rates $p=10^{-3}$, with thresholds above $0.4\%$ (Menon et al., 14 Aug 2025).
• AG-code qudit protocols yield $\epsilon = \exp(-\Omega(n))$ logical error for $p_{\text{phys}} < p_{\text{th}}$, where $n$ is code length; $\epsilon$ may be driven to negligibly small values while maintaining constant space overhead (Wills et al., 2024, Cervia et al., 26 Dec 2025).
• For biased-noise qubits, unfolded Reed-Muller protocols yield logical error $\sim 35p_Z^3$ in the infinite bias limit, with only $53$ qubits and $5.5$ rounds at $p_Z=10^{-3}$, $\eta \gtrsim 5\times10^6$ (Ruiz et al., 16 Jul 2025).

Catalytic protocols trade space overhead for success probability: the optimal overhead can be reduced to unity with corresponding success factor, but with constant-size catalyst and quantum block reuse (Fang et al., 2024).

5. Resource Scaling, Platform Implementation, and Experimental Feasibility

Resource summary, finite examples, and hardware considerations:

Code	Physical Qubits/Qudits	Output Magic States	Space Overhead	Circuit Depth
Tricycle (N=192,K=27)	192	27	~7	$O(1)$
AG-qudit (Q₀)	42 ($q=64$ qudits)	14	3	$O(1)$
Unfolded, bias$=5\times10^6$	53	1	53	$\approx5.5$

• Tricycle codes: physical implementation is demonstrated for reconfigurable neutral atom arrays, minimal cycle time ($\sim$ tens of $\mu$s), CNOT-layered syndrome extraction ($D_{\text{CNOT}}=12$ for $w=4$), constant-depth CCZ realized in neutral atom hardware (Menon et al., 14 Aug 2025).
• AG-code qudit protocols: conversion to standard qubit CCZ states via a constant-depth Clifford circuit with at most 9 measurements and 12 single-qudit, 9 two-qudit Cliffords (Cervia et al., 26 Dec 2025).
• Unfolded Reed-Muller protocol: only requires nearest-neighbor CNOTs, no long-range or three-body interactions, and up to 175 qubits in moderate bias regimes; verified via Stim + BP/OSD Monte Carlo to $p_L\sim10^{-7}$ (Ruiz et al., 16 Jul 2025).

6. Catalytic Distillation and Overhead Trade-offs

Recent theoretical results demonstrate one-shot magic state distillation with constant overhead using quantum catalysts. Fang & Liu show that any multi-shot code-based protocol with $n$ input and $m$ output magic states can be lifted to a one-shot catalytic protocol with overhead $k=\lceil n/m \rceil$, success probability $p'=p\cdot m/\lceil n/k \rceil$ (Fang et al., 2024). Setting $k=1$ achieves the optimal constant overhead of unity, at the cost of lower (but still constant) success probability. The catalyst is reused and does not alter the amortized space cost. This framework decouples the batch size from the target fidelity, in contrast to code-based protocols that require extremely large batch sizes as $\epsilon\to 0$.

7. Comparative Analysis and Outlook

Constant space overhead distillation protocols resolve longstanding barriers to resource scaling in fault-tolerant quantum computation. These methods provide:

• Strictly bounded physical qubit resources per output, independent of target error rate.
• Circuit depths and time overheads that are fixed or grow logarithmically.
• Thresholds and error suppression competitive with or superior to traditional concatenated protocols.
• Hardware adaptability: protocols exploit match to Hamiltonian, gate, and decoherence bias, and are validated on platforms from superconducting qubits to neutral atoms.

Current research focuses on explicit code optimization (e.g., minimizing code length for fixed dimension), robust syndrome decoding under non-ideal noise, circuit compiler integration, and experimental demonstration at scale. Modular integration with code-switching and lattice surgery is foreseen, enabling architectures for large-scale universal quantum computation with drastic reduction in required quantum memory and factory resources (Menon et al., 14 Aug 2025, Cervia et al., 26 Dec 2025, Wills et al., 2024).