Source-linked AI summary

Logical Operator Decomposition for Distance Analysis of Bivariate Bicycle Codes

Mohammad Rowshan, Simon Devitt

arXiv:2609.39395v2quant-phcs.ITmath.IT

TL;DR

BB-code minimum distance is usually determined numerically rather than read from the defining polynomials. The paper uses projection-defined logical components and support-based verification, proving the distances of six standard codes and enumerating their minimum-weight logical operators. Its component labels distinguish logical classes, but do not by themselves give a closed-form distance.

  • Problem

    BB-code minimum distance is usually established numerically or by search rather than read directly from the defining polynomials.

    Source: §I · p. 1
  • Method

    The paper characterizes logical classes through projections and uses translation symmetry and support-exclusion tests to verify lower bounds.

    Source: A. Basis, dimension, and symmetry · p. 5, §V · p. 9
  • Results

    Distances 4, 6, 10, 10, 12, and 18 were proved for six standard codes, and the minimum-weight logical operators were enumerated.

    Source: §VI · p. 12, B. Scope and implementation · p. 14
  • Takeaways & Limitations

    The same defining-polynomial pair can yield different minimum-weight structures on different tori, as shown by the [[72, 12, 6]] and [[108, 8, 10]] examples.

    Source: B. Scope and implementation · p. 14

    Component labels do not yield a closed-form distance, and the census shows minima in both components for five of the six examples.

    Source: §VII · p. 14

Abstract

from arXiv · show

Bivariate bicycle (BB) quantum codes are a prominent finite-length family of quantum low-density parity-check codes, but their minimum distance is usually established numerically rather than read from the defining polynomials. We study the $Z$-logical quotient $K/S$ over $\mathbb F_2[x,y]/(x^\ell-1,y^m-1)$ and show that it fits into a short exact sequence with an annihilator quotient as kernel and a colon quotient as cokernel. The sequence gives an explicit logical basis, a dimension formula, and a componentwise distance identity $d_Z=\min(d_{\mathcal A},d_{\mathcal C})$. Using the Frobenius structure of the finite group algebra, we prove $r_{\mathcal A}=r_{\mathcal C}=k/2$ for every BB code, including repeated-root cases. The algebraic component of a logical class is distinct from the support shape of its lightest representatives: an annihilator class can have a lighter two-block representative, and a colon class can have a one-sided minimum. For lower bounds we show that every proper subset of a minimum-weight logical operator has nonzero syndrome, and that this property persists inside the colon component but not inside the annihilator component. A translation-anchored cluster search built on it proves the distances $4,6,10,10,12,18$ of the six standard BB codes of lengths $18$ to $288$ and enumerates every minimum-weight logical operator. The resulting census shows that $[[108,8,10]]$ is the only one of the six whose distance is attained in a single component, with $d_{\mathcal C}=10$ and $d_{\mathcal A}=12$.

I. INTRODUCTION

Bivariate bicycle codes are finite-length quantum LDPC codes, but their minimum distance is usually established by numerical or search-based methods rather than read from the defining polynomials. The paper decomposes logical classes into annihilator and colon components, then uses that structure to analyze distance and verify six standard examples.

I. INTRODUCTION

  • BB-code distance is usually established numerically or by search, rather than read directly from the defining polynomials.Existing approaches probe a code after construction, so finite-length design tends to build first and measure distance afterward.Source: §I · p. 1

A. Related work

  • Prior work describes code dimensions, logical bases, and numerical distance methods, while this paper uses the annihilator–colon decomposition to label logical classes and split the distance problem.The decomposition is a relation-module construction specialized to the BB logical quotient, and its splitting interacts with cluster verification.Source: A. Related work · pp. 1–2

B. Contributions

  • The short exact sequence identifies the logical quotient’s kernel with an annihilator-derived subspace and its quotient with the colon quotient.The projection π labels classes by their right-block image; a zero image means the class admits a left-only representative, not that its lightest representative must be left-only.Source: B. Contributions · p. 2
  • A Frobenius-pairing argument establishes a component-dimension identity without requiring a semisimplicity assumption.In the semisimple case, the paper also recovers the equality character by character from the common zeros of a and b.Source: B. Contributions · p. 2
  • The distance splits into annihilator and colon minima, but component membership does not determine representative shape or remove the hard minimum-weight search.In [[18, 4, 4]], an annihilator class has a lighter two-block representative; colon classes can also have one-sided representatives.Source: B. Contributions · p. 2
  • The search proves distances 4, 6, 10, 10, 12, and 18 for six standard examples and enumerates every minimum-weight logical operator.For [[108, 8, 10]], the colon component has distance 10 while the annihilator component has distance 12.Source: B. Contributions · p. 2

II. PRELIMINARIES

BB codes use a binary bivariate group algebra to encode two qubit blocks and their checks. Their Z-logical operators are the kernel of the X-check map modulo stabilizers, and the minimum Z-distance is the lightest nonzero class in that quotient.

A. Notation and the ambient ring

  • The ambient ring R represents binary polynomials with x- and y-exponents reduced modulo ℓ and m, and has dimension ℓm over F2.Its monomials form a basis, and multiplication is cyclic convolution in both indices.Source: A. Notation and the ambient ring · p. 2

B. Bivariate bicycle codes and the logical quotient

C. Algebraic and matrix dictionary

The algebraic and matrix dictionary translates ideals and module quotients into linear-algebra objects. A short exact sequence then decomposes the Z-logical quotient into annihilator and colon components without requiring a direct-sum splitting.

C. Algebraic and matrix dictionary

  • The principal ideal (a) is im A, ann(a) is ker A, and the colon ideal (a : b) consists of right-block words v for which bv lies in (a).In matrix form, the colon ideal is B^-1(im A), also expressible as ker(CAB) for a parity-check matrix CA of im A.Source: C. Algebraic and matrix dictionary · p. 4

III. DECOMPOSITION OF THE LOGICAL QUOTIENT

  • The short exact sequence identifies the kernel of π as the annihilator quotient ann(a)/b ann(a) and its quotient as the colon quotient (a : b)/(a).The projection sends [(u, v)] to v+(a), and the sequence applies without excluding any BB code.Source: §III · p. 4
  • The projection labels a logical class as annihilator or colon according to whether its image is zero, but does not determine the support shape of its minimum-weight representative.The algebraic label and representative shape are kept distinct.Source: §III · pp. 4–5
  • The exact sequence is interpreted for the BB logical quotient to separate distance calculations into its two class families.The underlying relation-module construction is standard; the paper’s contribution is its BB interpretation and use in separating the distance calculation.Source: §III · pp. 4–5

A. Basis, dimension, and symmetry

The exact sequence supplies a basis and dimension formula for K/S, while Frobenius duality proves equal dimensions for its annihilator and colon components. The construction is translation-equivariant and uses linear algebra independently of distance search.

A. Basis, dimension, and symmetry

  • The exact-sequence basis combines lifts from ann(a)/b ann(a) with lifts from (a : b)/(a), using a solution auj=bvj for each colon lift.The resulting vectors form a basis of K/S.Source: A. Basis, dimension, and symmetry · p. 5
  • The basis construction requires only Gaussian elimination on matrices of size at most N × N, where N = ℓm.Annihilator vectors come from ker A modulo B(ker A); colon vectors come from ker(CAB) modulo im A, with corresponding solutions of Auj=Bvj.Source: A. Basis, dimension, and symmetry · p. 5
  • For every BB code, the annihilator and colon components each have dimension k/2, including in repeated-root cases.The equality follows from Frobenius structure of the finite group algebra.Source: A. Basis, dimension, and symmetry · p. 5
  • The exact-sequence maps commute with translations, so the annihilator and colon families are unions of translation orbits.The companion sequence obtained by exchanging a and b uses a different coordinate projection on the same quotient K/S.Source: A. Basis, dimension, and symmetry · pp. 5–6

B. Componentwise coset-leader formulation

The distance identity minimizes separately over annihilator and colon classes, using exact coset-leader formulations. These identities expose the algebraic structure but leave minimum-weight searches, which remain hard in general.

B. Componentwise coset-leader formulation

  • The Z-distance is the smaller of the minimum weights in the annihilator and colon components.The projection partitions the nonzero logical classes into these two families.Source: B. Componentwise coset-leader… · p. 6, Figure 4 · p. 6
  • The colon minimum selects v outside (a) and adds wt(v) to λa(bv), the lightest left component solving au=bv.The colon minimization can range over v ∈ R\(a), since λa(bv) is infinite unless v ∈ (a : b).Source: B. Componentwise coset-leader… · pp. 6–7
  • The annihilator minimum searches vectors of the form (t+br, ar), with t ∈ ann(a)\b ann(a) and r ∈ R.The matrix form is a minimum-weight search over a coset of im[B ; A] inside ker A ⊕ F2^N.Source: B. Componentwise coset-leader… · p. 6
  • The exact component identities are not closed-form distance formulas because their residual coset-leader minimizations are NP-hard in general.A one-sided annihilator representative gives only an upper bound: in [[18, 4, 4]], wann = 6 while dA = 4.Source: B. Componentwise coset-leader… · p. 7
  • Algebraic component and representative shape are independent: an annihilator class can have a lighter two-block representative, and a colon class can have a one-sided minimum.The annihilator one-sided quantity wann is not necessarily the component minimum dA.Source: B. Componentwise coset-leader… · p. 7

C. Spectral interpretation in the semisimple case

In the semisimple case, common zeros of the defining polynomials determine the nonzero fibers of the logical quotient. Each common zero contributes equally to the annihilator and colon components, recovering the balance rA = rC = k/2.

C. Spectral interpretation in the semisimple case

  • Each character contributes a two-dimensional fiber to K/S exactly when both defining polynomials vanish there.Otherwise the stabilizer fills the one-dimensional syzygy solution space and the quotient fiber is zero.Source: C. Spectral interpretation in the… · p. 7
  • Each common zero contributes one dimension to each component, giving rA = rC = k/2.This character-by-character count recovers the common-zero dimension count.Source: C. Spectral interpretation in the… · p. 7
  • For [[18, 4, 4]], two common-zero characters form one Frobenius orbit and account for k = 4 binary dimensions.More generally, Frobenius-closed character orbits recover the binary dimensions from the splitting-field count.Source: C. Spectral interpretation in the… · p. 7
  • The spectral description is an interpretation and dimension check, not an assumption required for the distance analysis.Source: C. Spectral interpretation in the… · p. 7

IV. ALGEBRAIC COMPONENT VERSUS REPRESENTATIVE SHAPE

The algebraic component of a logical class and the support shape of its minimum-weight representatives are independent labels. This distinction informs how the paper searches for upper-bound witnesses and interprets them.

A. Two independent labels

  • For fixed τ, one-sided, τ-lopsided, and τ-balanced shapes partition nonzero logical classes, and any shape may occur in either component.The examples use τ = 3; one-sided minima occur in both components.Source: A. Two independent labels · p. 7
  • Component membership is fixed by the projection π, whereas adding a stabilizer can change a representative’s support shape.An annihilator class can have a lighter two-block representative, while a colon class can have a one-sided minimum.Source: §IV · p. 7, Figure 6 · p. 8

B. Search order and what each step establishes

  • Search left-only annihilator witnesses, then right-only witnesses and sparse two-block relations, labeling each by π before pursuing lower bounds.This ordering is a heuristic for upper bounds; exact component minima remain dA and dC.Source: B. Search order and what each step… · p. 8

C. Worked examples

Worked examples show that component labels do not determine the support shape or weight of a lightest representative. The examples also distinguish upper-bound witnesses from componentwise distance proofs.

C. Worked examples

  • The [[72, 12, 6]] code has weight-6 one-sided logicals in both the annihilator and colon components, obtained from ring arithmetic.For the annihilator witness, the cluster search with radius 5 finds no logical, establishing d = 6.Source: C. Worked examples · p. 8
  • For [[18, 4, 4]], the weight-4 logical (1, b) lies in the annihilator component, although its displayed minimum representatives are 3-lopsided.The annihilator quotient’s minimum one-sided weight is 6, so “annihilator” does not mean “one-sided.”Source: C. Worked examples · pp. 8–9
  • For [[108, 8, 10]], the minimum colon logicals have weight 10, while the annihilator component has minimum weight 12.The radius-9 cluster search finds no logical; the radius-11 census returns 54 weight-10 translates in nine colon classes, all with split (6, 4).Source: C. Worked examples · p. 9

D. Explicit witnesses

The paper combines explicit logical witnesses with exact lower-bound tests based on syndrome connectivity and translation-anchored search. Component-specific connectivity gives an exact search for the colon distance, but does not extend to the annihilator component.

D. Explicit witnesses

  • Table III lists verified low-weight witnesses for the standard codes, each in K\S, with weight splits recorded.The Gross-code pair extends the one-sided pattern of [[72, 12, 6]] with an extra sparse factor (1 + x6).Source: D. Explicit witnesses · p. 9

V. EXACT LOWER BOUNDS

  • The support-exclusion criterion proves dZ ≥ d0 exactly when K(E) = S(E) for every support E smaller than d0.Translation symmetry reduces the test to one support per orbit; supports with independent check columns pass after one rank computation.Source: §V · p. 9

A. Syndrome connectivity and anchored cluster search

  • Every proper nonempty subset of a minimum-weight logical has nonzero syndrome, enabling the anchored cluster search to certify lower bounds.The search grows candidate supports from a translated anchor and, if it terminates without output, certifies dZ ≥ W + 1.Source: A. Syndrome connectivity and anchored… · p. 10
  • Minimum colon logicals also have this connectivity property, so the corresponding search computes dC exactly; the annihilator analogue can fail.An annihilator logical may split into two lighter colon logicals, so dA requires a census at dZ or a one-sided witness with Proposition 4.Source: A. Syndrome connectivity and anchored… · pp. 10–11

B. Complexity of the structural and support steps

The structural vector spaces are computable by Gaussian elimination in O(N^3) arithmetic operations, but the support search remains exponential in the target distance. Translation anchoring and rank pruning reduce the search without making distance computation polynomial-time.

B. Complexity of the structural and support steps

  • O(N^3) arithmetic operations suffice to obtain the structural vector spaces from O(N) × O(N) matrices.This bound covers the structural computation, not extracting a low-weight quotient element or finding a minimum-weight lift.Source: B. Complexity of the structural and… · p. 11
  • Anchoring one occupied coordinate reduces the number of weight-i supports, with leading savings of order n/i before further deduplication.Rank pruning rejects supports whose restricted check columns are independent.Source: B. Complexity of the structural and… · p. 11
  • The support-exclusion procedure remains exponential in the target distance; symmetry and early rejection reduce work but do not yield polynomial-time distance computation.For each anchored support, it computes dim K(E) and compares it with dim S(E) when dim K(E) is nonzero.Source: B. Complexity of the structural and… · p. 11

C. Results of the lower-bound computations

The anchored cluster search and explicit logical witnesses establish exact distances for all six standard BB codes, and the complete minimum-weight census separates component labels from representative shapes. The results also show that the support-exclusion test is limited to the shortest examples and that simple support-based bounds can be weak or false.

D. When the bounds meet: exact distance

VI. EXAMPLES AND RESULTS

  • 4, 6, 10, 10, 12, and 18 are proved as the distances of the six standard examples, including a confirmed length-288 value previously reported as a numerical upper bound.Each explicit witness meets its cluster-search lower bound; the implementation was also checked against exhaustive enumeration on 200 random BB codes of length at most 32.Source: D. When the bounds meet: exact distance · p. 11, C. Results of the lower-bound… · p. 11, §VI · p. 12
  • The complete census finds that only [[108, 8, 10]] attains its distance in a single component; one-sided and balanced minimum classes coexist in three other examples.A colon class with a one-sided minimum also occurs in [[72, 12, 6]].Source: §VI · p. 12, Table V · p. 12
  • For the Gross code [[144, 12, 12]], 246 minimum-weight classes include one-sided and balanced annihilator classes as well as lopsided and balanced colon classes.The weight-12 one-sided annihilator witness is therefore representative of only a small part of its minimum-weight structure.Source: §VI · p. 12

A. Decomposition profile

  • A convolution-based bound returns 3 against d = 6 for [[72, 12, 6]], while the proposed difference-set inequality is false over R for [[18, 4, 4]].The latter would predict weight at least 5, although (a, 1) is a logical of weight 4.Source: A. Decomposition profile · p. 12

B. Screening on random pairs

Random-pair screens illustrate complementary ways to flag sparse witnesses, and the paper combines those screens with component-labelled bounds and an anchored cluster search. The discussion distinguishes the proved distance results from decoder-dependent logical error rates and identifies extensions that remain undeveloped.

B. Screening on random pairs

  • On the sampled weight-three pairs, the one-sided annihilator and lopsided-colon screens trade off across the two tori, but the reported marginal rates do not establish how many pairs survive all screens.The fixed-seed counts are demonstrations on unconstrained samples, not optimized BB-code searches or evidence that the curated codes are rare.Source: B. Screening on random pairs · pp. 12–13, Table VIII · p. 13

C. Search and design procedure

  • The search procedure moves from sparse-relation and annihilator screens to colon searches, then applies an anchored lower bound to competitive candidates and reports component and shape data.The workflow starts with structural spaces and uses the cluster search at radius dtarget − 1; Fig. 10 summarizes the route from code specification to bounds.Source: C. Search and design procedure · p. 13, Figure 10 · p. 14

VII. DISCUSSION AND LIMITATIONS

  • The decomposition supplies a component invariant alongside weight, but the census shows that component labels alone do not determine the minimum or its representative shape.The minimum occurs in both components for five of six examples, and the colon component remains compatible with cluster verification.Source: §VII · p. 14

A. Results and open problems

  • The method proves all six standard distances and determines dA and dC separately, but worst-case search grows exponentially with target distance.The length-288 run visits 455,504,708 nodes at radius 17; larger-automorphism canonicalisation is suggested for extending the method.Source: A. Results and open problems · p. 14

B. Scope and implementation

  • Minimum Hamming weight is a code property, not a prediction of logical error rate under a particular decoder or noise model.The paper does not perform the decoder and noise simulations needed to establish logical error rates.Source: B. Scope and implementation · p. 14

C. Beyond distance: further applications

  • The exact-sequence decomposition applies to two-block group-algebra codes and other CSS constructions defined by a pair of ring elements, but its further applications are proposed rather than developed.Suggested directions include hardware-oriented logical representatives, decoder design, and large-scale code search.Source: C. Beyond distance: further applications · p. 14

APPENDIX B CODE DATA AND EXPLICIT WITNESSES

The appendix gives defining polynomials and convenient minimum-weight witnesses for the standard codes, with component membership determined separately by the projection map.

APPENDIX B CODE DATA AND EXPLICIT WITNESSES

  • Table IX pairs each standard code’s defining polynomials with a convenient minimum-weight witness verified as a non-stabilizer element of K \ S.Each listed pair satisfies au + bv = 0 over F2, and component membership is determined separately by π[(u, v)] = v + (a).Source: Appendix B · p. 15
  • The displayed minimum witness for [[108, 8, 10]] is balanced and belongs to one of the code’s nine minimum-weight classes, all 3-balanced colon classes.The excerpt specifies the code’s polynomials and begins listing u’s terms.Source: Appendix B · p. 15
  • The [[288, 12, 18]] one-sided witness has v = 0 and weight 18, and exact checks verify Au = 0 and u ∉ b ann(a).Those checks establish that it is a logical rather than a stabilizer.Source: Appendix B · p. 15
Loading 2609.39395v2…