[Paper Review] Magic state cultivation: growing T states as cheap as CNOT gates

Craig Gidney, Noah Shutty+1 more|arXiv (Cornell University)|Sep 26, 2024

Economic and Technological Innovation4 citations

TL;DR

This paper introduces 'magic state cultivation,' a fault-tolerant method to prepare high-fidelity T states using surface code patches with spacetime costs comparable to a single CNOT gate. By iteratively injecting, growing, and extracting logical T states via stabilizer measurements and error suppression, the technique achieves logical error rates as low as $4 \cdot 10^{-11}$ at $5 \cdot 10^{-4}$ noise, reducing resource costs by an order of magnitude over prior distillation methods.

ABSTRACT

We refine ideas from Knill 1996, Jones 2016, Chamberland 2020, Gidney 2023+2024, Bombin 2024, and Hirano 2024 to efficiently prepare good $|T angle$ states. We call our construction "magic state cultivation" because it gradually grows the size and reliability of one state. Cultivation fits inside a surface code patch and uses roughly the same number of physical gates as a lattice surgery CNOT gate of equivalent reliability. We estimate the infidelity of cultivation (from injection to idling at distance 15) using a mix of state vector simulation, stabilizer simulation, error enumeration, and Monte Carlo sampling. Compared to prior work, cultivation uses an order of magnitude fewer qubit-rounds to reach logical error rates as low as $2 \cdot 10^{-9}$ when subjected to $10^{-3}$ uniform depolarizing circuit noise. Halving the circuit noise to $5 \cdot 10^{-4}$ improves the achievable logical error rate to $4 \cdot 10^{-11}$. Cultivation's efficiency and strong response to improvements in physical noise suggest that further magic state distillation may never be needed in practice.

Motivation & Objective

• To reduce the resource cost of preparing logical T states in surface codes, which are essential for fault-tolerant quantum computation.
• To address the high overhead of magic state distillation, which currently dominates the cost of non-Clifford gates in fault-tolerant architectures.
• To develop a scalable, physically realizable method that avoids large logical operations and instead uses physical gate sequences within a single code patch.
• To demonstrate that cultivation can match or surpass the efficiency of lattice surgery CNOT gates while being highly responsive to improvements in physical noise.
• To explore whether magic state distillation may become obsolete in practice due to the efficiency and scalability of cultivation.

Proposed method

• The method uses a three-stage process: injection of a noisy T state into a surface code, cultivation via repeated stabilizer measurements and error suppression, and escape via a postselection protocol to extract a high-fidelity logical T state.
• Cultivation leverages the surface code's ability to detect and correct errors during state evolution, using a sequence of logical CNOT and measurement operations to increase the logical distance of the state.
• The construction employs a hybrid simulation approach combining state vector simulations, stabilizer simulations, error enumeration, and Monte Carlo sampling to estimate logical error rates.
• The escape stage uses a GHZ-state-assisted measurement protocol to verify and extract the logical T state, with postselection to ensure fidelity.
• The method is designed to fit within a single surface code patch and uses roughly the same number of physical gates as a lattice surgery CNOT gate of equivalent reliability.
• The framework assumes a digitized depolarizing noise model and uses a fault-tolerant decoder to track error propagation across stages.

Experimental results

Research questions

• RQ1 Can T state preparation be achieved with spacetime costs comparable to a CNOT gate in the surface code?
• RQ2 To what extent can error suppression during state evolution reduce the need for full distillation protocols?
• RQ3 How does the logical error rate scale with physical noise strength and code distance in a cultivation-based approach?
• RQ4 Can the cultivation method outperform traditional distillation in terms of qubit·rounds and resource efficiency?
• RQ5 Is there a practical threshold below which magic state distillation becomes obsolete due to cultivation's efficiency and noise responsiveness?

Key findings

• Cultivation achieves a logical error rate of $2 \cdot 10^{-9}$ at $10^{-3}$ noise and distance 15, using an order of magnitude fewer qubit·rounds than prior distillation methods.
• Halving the physical noise to $5 \cdot 10^{-4}$ reduces the logical error rate to $4 \cdot 10^{-11}$, demonstrating strong responsiveness to noise improvements.
• At $10^{-4}$ noise, the method achieves a logical error rate of $6 \cdot 10^{-15}$ at distance 5, indicating high scalability.
• The spacetime cost of cultivation is estimated to be roughly equivalent to that of a lattice surgery CNOT gate of similar reliability.
• The escape stage, though complex, is not a bottleneck when decoded properly—initial simulations were misleading due to decoder limitations.
• The method suggests that further magic state distillation may not be needed in practice, as cultivation’s efficiency and noise sensitivity could outpace algorithmic demands.