Description
Qubit CSS code whose stabilizer generators are an active subset of the rows of a larger pair of block-circulant parent matrices, so that the CSS orthogonality condition is imposed only on the retained rows. The construction is designed so that low-weight combinations of the discarded latent rows fail to commute with the active checks, and therefore do not automatically become logical operators as they would were orthogonality enforced on every row. Evading this mechanism circumvents a barrier that otherwise caps the distance of a row-deleted low-density construction at the stabilizer generator weight, allowing sparse, regular, high-girth checks to be combined with encoding rate at or above one half.
In the strict monomial construction, two block-circulant parent matrices \(\hat{H}_X\) and \(\hat{H}_Z\), with \(L/2\) block rows and \(L\) block columns, are assembled by a \(G\)-lift in which every nonzero proto-matrix entry is replaced by a permutation matrix drawn from a group \(G\). Writing \(\mathcal{F}=\{F_i\}\) and \(\mathcal{G}=\{G_i\}\) for the two lists of lifts and \(\Gamma\) for a chosen active set of index pairs, the lifts are required to commute, \([F_i,G_j]=0\), for every \((i,j)\in\Gamma\), while commutation is deliberately broken for at least one pair outside \(\Gamma\). The lift group is generally a product of two factors with distinct roles. The top factor is the one whose non-commutativity creates the broken commutators that active orthogonality relies on, while the remaining bottom factor is Abelian and supplies code automorphisms. In the sectors whose top factor is Abelian or trivial no pair need fail to commute at all, and the parent matrices are then fully orthogonal [2]. The stabilizer generator matrices \(H_X\) and \(H_Z\) are then the first \(J\) block rows of the parents. Because the block-circulant structure makes the \((i,j)\) block of \(\hat{H}_X\hat{H}_Z^T\) depend only on the offset \(j-i\), collecting the commutators of all lift pairs whose indices sum to that offset, a suitable choice of \(\Gamma\) makes these blocks vanish for all \(0 \leq i,j < J\), so the retained rows define a CSS code [1].
Polynomial variants replace a single permutation in each lift entry by a sum of group elements, while loose active orthogonality requires only the aggregate commutator sum at each active offset to vanish, rather than requiring every contributing pair to commute individually [2]. These variants remain actively orthogonal, but the pairwise condition above and the sharp monomial bounds below no longer apply.
Protection
Retaining \(J\) block rows in each basis yields encoding rate at least \(1-2J/L\), so that \(J \leq L/4\) guarantees rate at least one half [2]. Sparse, regular, high-girth checks are targeted alongside rate, girth eight being attained by the flagship Kasai code and girth at least six by most of the compact GALA instances.
Distance is constrained by the latent rows. Any latent-row combination that commutes with the opposite-type active checks but lies outside the same-type stabilizer row space represents a logical operator, and the least weight of such a combination upper bounds the minimum distance [1]. Controlled non-orthogonality can therefore let the distance exceed the stabilizer generator weight by preventing low-weight latent rows from becoming logical operators.
The sharpest limitations are proved for strict monomial GALA and Kasai lifts, for which every stabilizer row has weight \(L\) [2]. Taking \(J>L/4\) forces full orthogonality of the parents. Provided at least one latent row lies outside the retained row space of its type, that row is a weight-\(L\) logical operator and \(d\leq L\). If every latent row is already in the retained row space, the retained code equals its fully orthogonal parent and this augmentation argument supplies no distance bound. At the other extreme, \(J\leq 2\) gives \(d\leq g/2\), where \(g\) is the Tanner-graph girth. Thus a monomial lift can break the weight barrier only in the window \(L\geq 12\) and \(2<J\leq L/4\), apart from the possibility of a block length exponential in the target distance.
Cousins
- Pair-partition CPM code— Both families impose CSS orthogonality algebraically rather than through a product construction, but by different mechanisms. Pair-partition CPM codes match column types in pairs so that every overlap between an \(X\)-type and a \(Z\)-type generator occurs an even number of times [3], whereas actively orthogonal CSS codes impose the condition only on the active rows retained from a larger pair of parent matrices [1].
- Two-branch coset CSS code— Both families impose CSS orthogonality algebraically rather than through a product construction, but by different mechanisms. Two-branch coset CSS codes reduce the commutation condition to equalities and disjointness conditions on multiplicative cosets of a finite field [4], whereas actively orthogonal CSS codes impose the condition only on the active rows retained from a larger pair of parent matrices [1].
- Lifted-product (LP) code— LP codes enforce the CSS orthogonality condition as an algebraic identity on all rows, whereas actively orthogonal CSS codes impose it only on the retained rows [1,2].
Member of code lists
Primary Hierarchy
References
- [1]
- K. Kasai, “Breaking the Orthogonality Barrier in Quantum LDPC Codes”, (2026) arXiv:2601.08824
- [2]
- W. Yang, C. Duckering, and A. Dua, “Designer Codes from GALA: Compact, Self-Dual, and Rate-1/2 QEC on Reconfigurable Atom Arrays”, (2026) arXiv:2608.07431
- [3]
- K. Okada and K. Kasai, “Pair-Partition Constructions for CPM-Based Quantum LDPC Codes”, (2026) arXiv:2607.14091
- [4]
- K. Okada and K. Kasai, “A Two-Branch Finite-Field Construction for Regular CSS LDPC Bases”, (2026) arXiv:2605.23894
Page edit log
- Victor V. Albert (2026-08-25) — most recent
- Victor V. Albert (2026-08-22)
- Victor V. Albert (2026-08-17)
- Victor V. Albert (2026-08-15)
Cite as:
“Actively orthogonal CSS code”, The Error Correction Zoo (V. V. Albert & P. Faist, eds.), 2026. https://errorcorrectionzoo.org/c/actively_orthogonal_css, arXiv:2606.11484