Magic state cultivation with lattice-surgery escape on superconducting hardware
Magic state cultivation (MSC) is a low-overhead protocol for fault-tolerantly preparing logical magic states. In this work, we implement two MSC variants to prepare a magic state in a distance-5 code on a publicly available IBM Nighthawk processor. The first is a lattice-surgery escape, recently proposed by Hirano et al. and experimentally studied here for the first time. The second is a grafting escape, previously demonstrated by Rosenfeld et al., and extended here by incorporating a switch into a matchable code. Both protocols begin with fault-tolerant cultivation of a T state on a color code, followed by an escape into a larger code. We decode the results using minimum-weight perfect matching or Tesseract on circuit-level error models calibrated to the hardware noise. Compared with the previous experiment, our cultivated T state achieves a logical error probability of 2e-5 at the T-state injection angle, nearly 10x lower with about 3x the retainment. Moreover, our execution of the grafting escape has a 2x lower logical error rate per stabilizer round at a higher retainment. When comparing the two protocols under identical execution conditions and complementary gap postselection based on decoder confidence, lattice-surgery escape into the surface code preserves a T-state fidelity above 0.99 ten rounds after escape at about 5% retainment. By contrast, matchable-switched grafting drops below 0.95 fidelity at comparable retainment. Furthermore, the lattice-surgery escape reaches a logical error rate per stabilizer round of about 0.14% with around 11% average shot retainment. These results establish lattice-surgery escape as a practical MSC protocol that benefits from a direct transition into a surface code, and demonstrate the lowest MSC error and the highest surface-code magic-state fidelity reported to date, advancing the experimental realization of fault-tolerant non-Clifford gadgets.
Authors
- Yuval Baum (ORCID: https://orcid.org/0000-0003-4631-8551)
- Haoran Liao (ORCID: https://orcid.org/0000-0002-6399-006X)
- Michael Schutzman
- William Munizzi (ORCID: https://orcid.org/0000-0002-2675-0251)
- Pranav Mundada (ORCID: https://orcid.org/0000-0003-3316-0724)
- Gavin Hartnett
Publication Details
- Journal
- arXiv (Cornell University)
- Published
- 2026-11-17
- DOI
- https://doi.org/10.48550/arxiv.2611.22035
- Primary Topic
- Quantum Computing Algorithms and Architecture
- Type
- preprint