--- title: Generalized Labeled Codes url: https://www.emergentmind.com/topics/generalized-labeled-codes type: topic --- # Generalized Labeled Codes Searching arXiv for recent and foundational uses of “generalized labeled codes” and adjacent terminology. “Generalized labeled codes” does not denote a single universally standardized code family across coding theory. In the literature, the phrase and closely adjacent constructions refer to several distinct frameworks in which coordinates, symbols, or local constraints carry additional structure beyond the uniform coordinate model of a classical linear code. These include codes labeled by local singularity data on algebraic surfaces [2508.16369], group-labeled nonlinear codes arising from generalized Hadamard matrices and cocycles [1906.06220], graph-indexed vertex labelings on Johnson and Hamming graphs [1912.10388], group-algebra-labeled stabilizer codes generalizing Haah’s cubic code [1902.04543], and block-structured generalized algebraic-geometric evaluation codes built from higher-degree points and inner codes [1212.5802]. A common theme is that labels are not merely decorative metadata: they determine admissible local symbols, symmetry actions, recovery neighborhoods, or the algebraic form of parity constraints. ## 1. Terminological scope and principal meanings In the most explicit recent usage, a generalized labeled code is a coding-theoretic object attached to a normal surface with isolated singularities, where the ambient coordinate group is a direct sum of dual local homology groups labeled by singularity type [2508.16369]. This replaces the classical binary nodal code, which lives in \((\mathbf Z/2)^s\), by an embedding \[ \mathcal K \subset \bigoplus_{x\in \Sigma} H_1(x)^\vee, \] with each summand determined by the local singularity at \(x\) [2508.16369]. A different, older line of work studies group-labeled nonlinear codes derived from generalized Hadamard matrices over the additive group of a finite field. There, labels are entries of a generalized Hadamard matrix, codewords are row-labelings up to addition of constant vectors, and the code structure is governed simultaneously by additive field structure, cocycles, and extension-group multiplication [1906.06220]. The associated family is called generalized Hadamard full propelinear codes, or GHFP-codes [1906.06220]. Another meaning appears in graph-based coding. Johnson graph codes are linear subspaces of labelings of the vertices of a Johnson graph such that the labeling is uniquely determined by its restriction to a graph-defined generalized neighborhood \(B_r(A)\) [1912.10388]. The coordinates are thus indexed by \(v\)-subsets of an \(n\)-set rather than by consecutive positions, and information sets are determined by shell structure in the graph [1912.10388]. A further neighboring usage occurs in fracton stabilizer codes. The generalized Haah construction specifies a code by a finite group \(G\), an integer \(q\), two \(q\)-tuples \(A=(A_1,\dots,A_q)\) and \(B=(B_1,\dots,B_q)\) of subsets of \(G\), and commuting \(q\times q\) matrices over \(F_2\); these data label the supports of stabilizer generators on a \(2q\)-qubit-per-site lattice \(L(G,2q)\) [1902.04543]. In that setting, the code is “labeled” by subset and matrix data rather than by a uniform translational template [1902.04543]. A plausible implication is that “generalized labeled codes” is best treated as an umbrella description for code families in which the ambient coordinates are indexed by algebraic, combinatorial, or geometric labels carrying operational significance, rather than as the name of a single canonical construction. ## 2. Geometric generalized labeled codes from normal and ADE surfaces The most direct formal definition is given for normal surfaces. Let \(\mathcal V\) be a finite set of labeled finitely generated abelian groups, let \(\Sigma\) be a finite set, let \(h:\Sigma\to\mathcal V\) assign to each \(x\in\Sigma\) a local group \(H_1(x)=V_{h(x)}\), and let \[ k:\bigoplus_{x\in\Sigma} H_1(x)\twoheadrightarrow H_1 \] be a surjection. Dualizing yields \[ k^\vee:H_1^\vee\hookrightarrow \bigoplus_{x\in\Sigma} H_1(x)^\vee, \qquad \mathcal K=\operatorname{Im}(k^\vee), \] and \(\mathcal K\) is the generalized labeled code [2508.16369]. This formalism is motivated by the topology of a normal compact complex surface \(X\) with isolated singularities. If \(X^*=X\setminus \operatorname{Sing}(X)\), there is a natural local-to-global map \[ gl:\bigoplus_{P_i\in \operatorname{Sing}(X)} H_1(P_i)\longrightarrow H_1(X^*,\mathbf Z), \] and under the stated global hypotheses this map is surjective [2508.16369]. The strict generalized labeled code is the dual image \[ \mathcal K=H_1(Y^*,\mathbf Z)^\vee \subset \bigoplus_{x\in\Sigma} H_1(x)^\vee \] for a projective normal surface \(Y\) [2508.16369]. The extended code is \[ \mathcal K' = H_1(Y^*\setminus H,\mathbf Z)^\vee, \] where \(H\) is a smooth hyperplane section [2508.16369]. For ADE singularities, the local groups are the discriminant groups of the corresponding root lattices. The paper lists \[ A_n:\mathbf Z/(n+1),\quad D_{2m+1}:\mathbf Z/4,\quad D_{2m}:(\mathbf Z/2)^2,\quad E_6:\mathbf Z/3,\quad E_7:\mathbf Z/2,\quad E_8:0 \] as the local homology groups \(H_1(x)\) [2508.16369]. The geometric labeling is finer than the abstract isomorphism class of the group: distinguished generators are determined by the exceptional curves in the minimal resolution, and this affects refined weight counts [2508.16369]. The extended code also measures lattice saturation. If \(S=\widetilde Y\) is the minimal resolution and \(\mathbb E=\langle E_1,\dots,E_\nu,H\rangle\subset H^2(S,\mathbf Z)\), then \[ \mathbb E^{sat}/\mathbb E \cong (\mathcal K')^\vee \] [2508.16369]. This ties the generalized labeled code directly to the non-primitivity of the lattice generated by exceptional divisors and polarization. ## 3. Weight structures, shortening, and ancestors in the geometric theory The geometric theory requires a refined notion of weight because the local coordinate groups are heterogeneous. For \[ v\in \mathcal K \subset \bigoplus_{x\in\Sigma} H_1(x)^\vee, \] the paper defines the refined weight \[ w(v;j,\xi)=|\{x\mid h(x)=j,\ v_x=\xi\}|, \] the label weight \[ w(v;j)=|\{x\mid h(x)=j,\ v_x\neq 0\}|, \] and the Hamming weight \[ w(v)=\sum_{j,\xi\neq 0} w(v;j,\xi)=\sum_j w(v;j) \] [2508.16369]. This separates support size from local character type. Several congruence restrictions are then proved. For order \(2\) vectors, a weighted sum involving contributions from \(A_{2m+1}\), \(D_n\), and \(E_7\) is divisible by \(4\), and by \(8\) if \(K_S\) is divisible by \(2\) [2508.16369]. For order \(3\), \[ \sum_s s\, t(A_{3s-1}) - t(E_6) \] is divisible by \(3\) [2508.16369]. For almost simple vectors of order \(N>4\), \[ \sum_s s\, t(A_{Ns-1}) \] is divisible by \(N\) [2508.16369]. These are direct analogues of classical divisibility restrictions for binary nodal codes, but now expressed in label-sensitive form. A central innovation is the extension of code shortening. An elementary shortening replaces one local group \(H_1(z)\) by a quotient \(H_1'(z)\) together with a refined labeling of the replacement local pieces, and simultaneously quotients the global group by the image of the discarded part [2508.16369]. Dually, \[ \mathcal K'=\mathcal K\cap (V')^\vee \] for the shortened ambient group [2508.16369]. In geometric terms, shortening is the algebraic counterpart of partial smoothing of singularities. The model example is \(A_n\), where deleting one exceptional curve in the Dynkin chain corresponds to deforming \[ A_n \rightsquigarrow A_m + A_{n-m-1}, \] and induces \[ \mathbf Z/(n+1)\twoheadrightarrow \mathbf Z/\gcd(m+1,n-m) \] at the local-group level [2508.16369]. The paper’s realization theorem states that, for unobstructed ADE surfaces, geometric driven shortenings are exactly realized by partial smoothings, and conversely every partial smoothing yields a geometric driven shortening [2508.16369]. This leads to the notion of an ancestor: a maximally singular model from which others arise by shortening or smoothing [2508.16369]. The paper identifies explicit ancestors among cubic and K3 surfaces, including the Cayley cubic, a cubic with \(3A_2\) singularities, a cubic with \(E_6\), a cubic with \(A_5+A_1\), the Kummer surface, and the nine-cusp K3 surface [2508.16369]. This suggests a deformation-theoretic genealogy organized by generalized labeled codes. ## 4. Group-labeled nonlinear codes from generalized Hadamard matrices In the generalized Hadamard framework, labels are elements of the additive group of \(\mathbb F_q\), viewed multiplicatively when needed for cocycles and extensions [1906.06220]. A normalized generalized Hadamard matrix \(GH(q,v/q)\) over \(\mathbb F_q\) defines the row set \(F_H\) and the generalized Hadamard code \[ C_H=\bigcup_{\alpha\in\mathbb F_q}(F_H+\alpha\mathbf 1), \] with \[ |C_H|=qv,\qquad \text{length } v,\qquad \text{minimum distance } v-\frac vq \] [1906.06220]. These codes are typically nonlinear [1906.06220]. The algebraic backbone is an orthogonal cocycle \[ \psi:G\times G\to U \] satisfying the cocycle identity and the orthogonality condition that the cocyclic matrix \(M_\psi\) is a generalized Hadamard matrix [1906.06220]. Such a cocycle determines a central extension \[ E_\psi=\{(u,g):u\in U,\ g\in G\}, \qquad (u,g)(v,h)=(uv\psi(g,h),gh), \] and the paper proves the equivalence between orthogonal cocycles, cocyclic generalized Hadamard matrices, and central relative \((v,w,v,v/w)\)-difference sets [1906.06220]. The propelinear structure is then introduced. A generalized Hadamard code that is full propelinear is called a generalized Hadamard full propelinear code, or GHFP-code [1906.06220]. In this setting, symbol actions are translations by the label vector, while nonconstant codewords induce fixed-point-free coordinate permutations [1906.06220]. The group of coordinate permutations \(\Pi=\{T_x:x\in C\}\) satisfies \[ C_{\mathbf 1}\le K(C),\qquad \Pi\cong C/C_{\mathbf 1} \] [1906.06220]. The paper proves that GHFP-codes are equivalent to cocyclic generalized Hadamard matrices and to central relative difference sets [1906.06220]. This makes the labeled nature explicit: codewords correspond to extension-group elements, coordinate permutations come from the regular action of \(G\), and labels combine through cocycle multiplication. A plausible implication is that GHFP-codes provide one of the clearest algebraic models of “generalized labeled codes” in the nonlinear finite-alphabet setting. ## 5. Graph-indexed labelings and group-algebra-labeled stabilizer codes Johnson graph codes shift the notion of a code coordinate from positions in \([N]\) to vertices of a graph. For the Johnson graph \(J(n,v)\), coordinates are indexed by all \(v\)-subsets of an \(n\)-set, so the length is \[ N=\binom{n}{v} \] [1912.10388]. For a \(k\)-subset \(A\), the generalized neighborhood \(B_r(A)\) is a union of vertex neighborhoods, and a Johnson graph code \(\mathrm{JGC}(n,v,k,r)\) is defined so that every \(B_r(A)\) is an information set [1912.10388]. The dimension is therefore \[ K=|B_r(A)| \] [1912.10388]. The neighborhood size is given explicitly by \[ |B_r(A)|= \begin{cases} \sum_{i=0}^{r}\binom{k}{v-i}\binom{n-k}{i}, & k\ge v,\\[1ex] \sum_{i=0}^{r}\binom{k}{i}\binom{n-k}{n-v-i}, & k\le v. \end{cases} \] [1912.10388]. The main construction starts from a length-\(n\) MDS code \(C_0\) and the determinantal coordinate map \[ T(M)=\bigl(\det(M_L):L\in J(n,v)\bigr) \] for \(v\times n\) matrices \(M\) having at least \(t\) rows in \(C_0\) [1912.10388]. The resulting code is a Johnson graph code with \[ r=\min(v,\dim C_0)-t \] [1912.10388]. The dual is again a Johnson graph code: \[ \mathrm{JGC}(n,v,k,r)^\perp=\mathrm{JGC}(n,v,n-k,R-r-1) \] [1912.10388]. In distributed storage, these codes serve as outer constraints linking layered exact-repair constructions and enabling recovery from \(k