Recursive 3D self-correcting code [1]
Description
Family of geometrically local 3D CSS codes, each encoding one logical qubit, whose stabilizer Hamiltonians are self-correcting [1]. Each iteration replaces every square of the Tanner square complex by two parallel layers joined by a thickness-two strip, doubling the syndrome weight of one error type [1; Sec. 3.2]. Alternating \(X\)-type and \(Z\)-type iterations makes this doubling occur at all length scales and for both error types.
The Tanner square complex of a code is its Tanner graph with squares \((x,q,z,q')\) attached for pairs of qubits shared by checks \(x\) and \(z\) [2]. The recursion starts from the \([[5,1,2]]\) rotated surface code [1; Sec. 3.3]. Before the replacement, an iteration scales the embedding of this complex in \(\mathbb{R}^3\) by a large constant and randomly perturbs its vertices to lower the density of squares. The replacing complexes, \(U_X\) in the \(X\)-iteration and \(U_Z\) in the \(Z\)-iteration, are glued along the boundaries of neighboring squares. The complex \(U_X\) resembles a piece of a thickness-one 3D surface code slab whose front and back boundaries condense \(X\)-type membranes, and \(U_Z\) is its \(Z\)-type counterpart [1; Remark 3.7]. Stabilizer generators have weight at most seven, and each qubit lies in at most seven generators [1; Lemma 3.14 and Thm. 3.16].
Protection
Distance grows as a power of \(n\) [1; Sec. 1.1]. Each iteration multiplies the number of qubits by a constant factor, so \(n=\exp(\Theta(k))\) after \(2k\) iterations (using asymptotic notation) [1; Sec. 3.3].Decoding
Renormalization-group-like decoder whose locally computable coarse-graining maps apply a correction at one level and pass the residual syndrome to the previous level [1; Sec. 4]. For \(X\)-type errors, coarse-graining at least halves the syndrome weight across an \(X\)-iteration and does not increase it across a \(Z\)-iteration [1; Thms. 4.6 and 4.7].Cousins
- 3D surface code— The local complexes that replace each square in the recursive 3D self-correcting code construction resemble pieces of thickness-one 3D surface code slabs with membrane-condensing front and back boundaries [1; Remark 3.7].
- 3D lattice stabilizer code— The recursive 3D self-correcting code is geometrically local in three dimensions but not translationally invariant, so it evades the no-go results for self-correction in translationally invariant 3D stabilizer codes [1].
- \([[5,1,2]]\) rotated surface code— The recursive 3D self-correcting code family is obtained by recursively transforming the \([[5,1,2]]\) rotated surface code, which is its level-zero code [1; Sec. 3.3].
Primary Hierarchy
References
- [1]
S. Balasubramanian, M. Davydova, and T.-C. Lin, “A passive self-correcting quantum memory in three dimensions”, (2026) arXiv:2605.10943
- [2]
X. Li, T.-C. Lin, and M.-H. Hsieh, “Transform Arbitrary Good Quantum LDPC Codes into Good Geometrically Local Codes in Any Dimension”, (2024) arXiv:2408.01769
Page edit log
- Victor V. Albert (2026-11-08) — most recent
- Victor V. Albert (2026-10-27)
Cite as:
“Recursive 3D self-correcting code”, The Error Correction Zoo (V. V. Albert & P. Faist, eds.), 2026. https://errorcorrectionzoo.org/c/recursive_3d_self_correct, arXiv:2606.11484