[Jump to code hierarchy]

Mitten code[1]

Description

Member of a family of LP codes of stabilizer generator weight nine and encoding rate at least \(20\%\), built from one-by-two base matrices over the group algebra of a non-Abelian group. The shape of the base matrices fixes the rate, while taking the lift group non-Abelian removes a distance ceiling suffered by Abelian lifts of the same shape, allowing distances of about \(20\) with only a few hundred data qubits. Their small block size, together with a logical basis permuted by the group action, admits low-overhead and highly parallel logical operations.

The code is \(\mathrm{LP}(A,B)\), i.e., the \(G\)-lift of base matrices \(A=[a_0~a_1]\) and \(B=[b_0~b_1]\) over the group algebra \(\mathbb{F}_2[G]\) of a non-Abelian group \(G\), in the canonical form \begin{align} A = [\,g_1+g_2+g_3~~e+g_4+g_5\,]~,\qquad B = [\,h_1+h_2+h_3~~e+h_4+h_5\,]~, \tag*{(1)}\end{align} where \(e\) is the identity and \(g_i,h_i\) are group elements, and where the left regular representation of \(a_1\) and the right regular representation of \(b_1\) are required to be full rank. The resulting code has five blocks of \(|G|\) data qubits and two blocks each of \(X\)- and \(Z\)-type checks. The name comes from the block structure of the check matrices, in which each check has four similar columns (the fingers) together with a distinguished one (the thumb).

Instances have been obtained by an automated search [1], including \([[150,30,10]]\) over \(C_5 \times S_3\), \([[200,40,12]]\) over \(C_4 \times D_{10}\) (the dihedral group of order ten), \([[300,60,14]]\) over \(C_{10} \times S_3\), \([[500,100,16]]\) over \(C_5 \rtimes C_{20}\), and \([[540,108,18]]\) over \(C_9 \rtimes C_{12}\), whose distances are exact. Three larger reported instances have parameters \([[630,126,\leq 20]]\), \([[780,156,\leq 22]]\), and \([[975,195,\leq 24]]\).

Protection

Were the lift group Abelian, the minimum-weight codeword of the first base matrix would produce a logical operator of that weight, capping the distance at the weight of the base matrices and hence at six for this shape; a non-Abelian lift group removes the cap [1].

Rate

The one-by-two shape of the base matrices guarantees an encoding rate of at least \(20\%\), and every reported instance realizes exactly that rate [1].

Gates

Mitten codes admit a canonical logical basis whose \(X\)- and \(Z\)-type representatives each form a single orbit of the group action on an identity-element representative [1]. Each representative is supported on only two of the five data blocks, and the supports of a conjugate pair intersect in exactly one data qubit. This structure yields a modular logical toolkit based on graph surgery: all Clifford operations follow either from bridging two reusable seed surgery gadgets of tens of qubits each, or from a single fixed extractor that can measure any logical Pauli product. Many logical measurements can be executed in parallel by parallel surgery, and magic states can be injected into all logical qubits at once, supplying the non-Clifford resource for universal computation [1].

Decoding

Telescoping decoder, which sends progressively harder shots through successive BP, Relay-BP, and integer-programming stages [1].

Fault Tolerance

Searches with sQetch found syndrome-extraction schedules estimated to preserve circuit-level distance for the reported instances other than the \([[150,30,10]]\) code, whose best schedule has estimated circuit-level distance eight [1]. Distance preservation for both the memory and surgery schedules is supported by randomized circuit-distance estimates rather than exhaustive certificates.

Threshold

Circuit-level numerics under depolarizing noise, with no idling noise, are consistent with an effective finite-size threshold of about \(0.7\%\) for the family as a memory [1]. Under the same noise model and without extrapolation, the \([[300,60,14]]\) code attains a block logical error rate of about \(10^{-11}\) per round at \(0.1\%\) physical error rate. At \(0.4\%\), the \([[975,195,\leq 24]]\) code reaches about \(4\times 10^{-8}\), nearly two orders of magnitude below a \([[112320,195,24]]\) stack of rotated surface codes decoded by MWPM while using two orders of magnitude fewer physical qubits [1]. Decoding \(15\) billion logical surgery operations on the \([[540,108,18]]\) code at \(0.1\%\) yielded two logical failures.

Primary Hierarchy

Parents
Mitten codes are qubit CSS codes obtained from a product of chain complexes over a group algebra [1].
Mitten codes are LP codes of one-by-two base matrices over a non-Abelian group algebra, taken in a canonical form in which the left regular representation of \(a_1\) and the right regular representation of \(b_1\) are full rank [1].
Mitten code

References

[1]
A. Bhardwaj, M. Ma, N. Meister, R. King, D. Bluvstein, J. Preskill, M. Cain, Q. Xu, and H.-Y. Huang, “High-rate qLDPC processors”, (2026) arXiv:2607.28795
Page edit log

Your contribution is welcome!

on github.com (edit & pull request)

— see instructions

Zoo Code ID: mitten

Cite as:
“Mitten code”, The Error Correction Zoo (V. V. Albert & P. Faist, eds.), 2026. https://errorcorrectionzoo.org/c/mitten, arXiv:2606.11484
BibTeX:
@incollection{eczoo_mitten,
title={Mitten code},
booktitle={The Error Correction Zoo},
year={2026},
editor={Albert, Victor V. and Faist, Philippe},
eprint={2606.11484},
doi={10.48550/arXiv.2606.11484},
url={https://errorcorrectionzoo.org/c/mitten}
}
Share via:
Twitter | Mastodon |  | E-mail
Permanent link:
https://errorcorrectionzoo.org/c/mitten

Cite as:

“Mitten code”, The Error Correction Zoo (V. V. Albert & P. Faist, eds.), 2026. https://errorcorrectionzoo.org/c/mitten, arXiv:2606.11484

Github: https://github.com/errorcorrectionzoo/eczoo_data/edit/main/codes/quantum/qubits/stabilizer/qldpc/balanced_product/lp/mitten.yml.