Introduction
Quantum error-correcting codes protect quantum information against noise and the loss of physical systems [24 , 34 ] . An erasure is the loss of a physical system at a known location. In a distributed quantum memory, recovery may require access to surviving systems at different locations. Local erasure recovery limits the number of systems accessed by confining recovery to a prescribed repair set [16 ] . Under ( r , δ ) (r,\delta) ( r , δ ) -locality, every coordinate belongs to a repair set of size between δ \delta δ and r + δ − 1 r+\delta-1 r + δ − 1 , within which any set of up to δ − 1 \delta-1 δ − 1 erasures can be corrected [13 ] . In contrast, the global minimum distance d d d guarantees correction of any set of up to d − 1 d-1 d − 1 erasures when recovery can access all surviving systems. Thus, δ \delta δ specifies a local erasure tolerance under an access restriction, whereas d d d describes protection without this restriction. For codes with a nontrivial logical subsystem, ( r , δ ) (r,\delta) ( r , δ ) -locality implies d ≥ δ d \geq\delta d ≥ δ (see Corollary III.10).
Subsystem stabilizer codes (SSCs) decompose the encoded space into a logical subsystem and a gauge subsystem [24 , 34 ] . Recovery preserves the logical subsystem and its correlations with any reference system, while a completely positive trace-preserving (CPTP) map acts on the gauge subsystem. Pauli errors differing by a gauge operator are equivalent for logical recovery. We construct a syndrome decoder that distinguishes these equivalence classes using stabilizers supported in the repair set. Stabilizer and subsystem codes identify errors modulo S S S and G G G , respectively; the additional freedom is described by the quotient space G / S G/S G / S .
In this paper, we study subsystem stabilizer ( r , δ ) (r,\delta) ( r , δ ) -locally recoverable codes (( r , δ ) (r,\delta) ( r , δ ) -SSLRCs) and determine how local recovery requirements constrain the logical dimension, gauge dimension, and global minimum distance. More precisely, we address the following three questions:
Algebraic characterization: what conditions on the stabilizer and gauge label spaces characterize erasure recovery within a prescribed repair set for arbitrary joint states of the logical and gauge subsystems?
Parameter bounds: what bounds follow from ( r , δ ) (r,\delta) ( r , δ ) -locality without a purity assumption, and what further bounds hold for pure codes?
Optimal pure constructions: can explicit pure ( r , δ ) (r,\delta) ( r , δ ) -SSLRCs simultaneously attain the resulting bounds?
Yaozong Zhang is with the School of Mathematical Sciences and LPMC, Nankai University, Tianjin 300071, China (e-mail: zyzsdutms@163.com).
Sihem Mesnager is with the Department of Mathematics, University of Paris VIII, 93526 Saint-Denis, France, with the Laboratory Analysis, Geometry and Applications, LAGA, CNRS, UMR7539, 93430 Villetaneuse, France, and also with the Telecom Paris, Polytechnic Institute of Paris, 91120 Palaiseau, France (e-mail: smesnager@univ-paris8.fr).
Xiaoqiang Wang is with the Hubei Key Laboratory of Applied Mathematics, Faculty of Mathematics and Statistics, Hubei University, Wuhan 430062, China (e-mail: waxiqq@163.com).
Jian Gao is with the School of Mathematics and Statistics, Shandong University of Technology, Zibo 255000, China (e-mail: dezhougaojian@163.com).
Corresponding author: Jian Gao.
Classical locally recoverable codes (cLRCs). In classical coding theory, Gopalan et al. [17 ] introduced the notion of symbol locality. A symbol has locality at most r r r if it can be recovered from at most r r r other symbols. They established the Singleton-type bound d ≤ n − k − ⌈ k / r ⌉ + 2 d \le n-k-\lceil k/r\rceil+2 d ≤ n − k − ⌈ k / r ⌉ + 2 for an [ n , k , d ] q [n,k,d]_q [ n , k , d ] q linear code whose information symbols have locality at most r r r . Prakash et al. [35 ] introduced ( r , δ ) (r,\delta) ( r , δ ) -locality to allow local correction of multiple erasures and obtained the bound d ≤ n − k + 1 − ( ⌈ k / r ⌉ − 1 ) ( δ − 1 ) d \le n-k+1-(\lceil k/r\rceil-1)(\delta-1) d ≤ n − k + 1 − (⌈ k / r ⌉ − 1 ) ( δ − 1 ) , which reduces to the previous bound when δ = 2 \delta=2 δ = 2 . Cadambe and Mazumdar [7 ] derived alphabet-dependent bounds, while Grezet et al. [19 ] obtained such bounds for linear codes with ( r , δ ) (r,\delta) ( r , δ ) -locality by using residual codes. By accounting for the alphabet size, these bounds can improve on Singleton-type bounds over small alphabets.
Quantum locally recoverable codes (qLRCs). Golowich and Guruswami [16 ] studied quantum local recovery from single erasures. Luo et al. [31 ] derived parameter bounds and constructed pure qLRCs attaining them through the CSS construction. Galindo et al. [13 ] characterized ( r , δ ) (r,\delta) ( r , δ ) -locality for stabilizer codes in terms of puncturing and shortening. For stabilizer codes obtained from Euclidean or Hermitian dual-containing classical codes, their Singleton-type bound involves the classical minimum distance, which can be replaced by the quantum minimum distance under purity [13 Theorem 31 and Remark 34] . Galindo et al. [12 ] subsequently constructed impure qLRCs exceeding the corresponding bounds of pure qLRCs. For Hermitian constructions, Li et al. [27 ] established further bounds and constructions for pure qLRCs, while Li et al. [26 ] derived bounds for pure and impure qLRCs and used t t t -designs to determine their locality. Matsumoto [33 ] characterized stabilizer measurements sufficient for erasure correction and determined the minimum number of measurements within the corresponding decoding model. Entanglement-assisted qLRCs were studied in [14 , 28 ] . In these models, local recovery may also access the receiver’s halves of preshared maximally entangled pairs; these halves are assumed to be unaffected by erasures. The ( r , δ ) (r,\delta) ( r , δ ) -SSLRCs considered in this paper use no entanglement assistance; recovery preserves the logical subsystem and allows the gauge subsystem to change.
Quantum bounds for qLRCs. Kumar and Bandi [25 ] obtained explicit Griesmer, Hamming, and Plotkin bounds for pure qLRCs from the Hermitian construction. Ball, Lavrauw, and Popatia [4 ] established a Griesmer-type bound for additive codes without requiring linearity over the extension field. Since the associated classical codes need not be linear over F q 2 \mathbb{F}_{q^2} F q 2 , we use this additive bound. Stylianou and Boche [37 ] used blockwise weight enumerators to derive bounds for qLRCs with disjoint repair sets under a blockwise purity condition. Their formulation does not require a stabilizer structure.
SSCs and gauge freedom. Operator quantum error correction [24 ] and the subsystem stabilizer formalism [34 ] allow recovery to preserve logical information while changing the gauge subsystem. Aly and Klappenecker [1 ] developed additive constructions and propagation rules for trading logical and gauge dimensions with guarantees on minimum distance and purity. Further constructions arise from Hermitian hulls [29 , 30 ] , while Bacon-Shor codes [3 ] provide impure examples with weight-two gauge generators. Quantum code concatenation [23 ] , including concatenation of subsystem codes [9 Lemma 20] , provides bounds on the minimum distance in terms of the constituent-code distances. The geometric locality in [9 ] concerns interaction lengths, whereas the locality studied here concerns erasure recovery within prescribed coordinate sets. Bény et al. [6 ] studied approximate recovery under locality and symmetry constraints. Cao, Cotardo, and Lackey [8 ] studied supported symplectic subspaces and local invariants of stabilizer and subsystem codes through quantum anticodes.
Global and local constraints. Klappenecker and Sarvepalli [22 ] proved a Singleton bound for F q \mathbb{F}_q F q -linear subsystem codes without a purity assumption. This bound constrains the global code parameters without incorporating a prescribed family of repair sets. For ( r , δ ) (r,\delta) ( r , δ ) -SSLRCs, a further question is which errors can be distinguished modulo the gauge space using only stabilizers supported in a repair set. We give an algebraic characterization of this requirement and use it to derive parameter bounds.
Recovery Model and Methods
We require local recovery to return the state to the original code space and preserve the logical subsystem together with its correlations with an arbitrary reference system. We impose this requirement on every joint state of the logical, gauge, and reference systems. The gauge state may change and need not be known.
Let G G G and S S S denote the gauge and stabilizer label spaces, respectively. For a coordinate set T T T , let W T W_T W T be the space of Pauli labels supported in T T T , and let L T = L ∩ W T L_T=L\cap W_T L T = L ∩ W T . Theorem III.8 shows that a nonempty erased set E ⊆ Γ E\subseteq\Gamma E ⊆ Γ is recoverable within Γ \Gamma Γ if and only if W E ∩ S ⊥ t = G E W_E\cap S^{\perp_t}=G_E W E ∩ S ⊥ t = G E , where ⊥ t \perp_t ⊥ t denotes trace-symplectic orthogonality. To prove sufficiency, we construct an explicit local recovery channel based on stabilizer syndrome measurements. Without assuming purity, we derive bounds by studying supported stabilizer and gauge subspaces and counting gauge-equivalence classes of errors. For pure codes with a nontrivial logical subsystem, the stabilizer eigenspace defines a stabilizer code with the same minimum distance [2 Lemma 10] . We show that this reduction preserves the prescribed repair cover and obtain further bounds by shortening an associated additive classical code.
Main Contributions
In this paper, the parameters of an ( r , δ ) (r,\delta) ( r , δ ) -SSC Q \mathcal{Q} Q are denoted by ( ( n , K , R , d ) ) q ((n,K,R,d))_q (( n , K , R , d ) ) q , where K K K and R R R are the logical and gauge dimensions, and d d d is the global minimum distance. When K = q k K=q^k K = q k and R = q g R=q^g R = q g for nonnegative integers k k k and g g g , we say that Q Q Q is an ( r , δ ) (r,\delta) ( r , δ ) -SSC with parameters [ [ n , k , g , d ] ] q [[n,k,g,d]]_q [[ n , k , g , d ] ] q . The notation dim p \dim_p dim p denotes dimension over F p \mathbb{F}_p F p . Our main results for ( r , δ ) (r,\delta) ( r , δ ) -SSLRCs are summarized as follows. Item 1) answers (Q1), items 2) and 3) answer (Q2), and item 4) answers (Q3).
Algebraic characterization. For an ( r , δ ) (r,\delta) ( r , δ ) -SSC, we establish necessary and sufficient algebraic conditions for local erasure recovery within a prescribed repair set without a purity assumption (see Theorem III.8). We construct a recovery channel for arbitrary joint logical-gauge states, including correlations with a reference system. We establish algebraic criteria for the CSS and Hermitian hull constructions of Q Q Q (see Propositions III.12 and III.15). If R = 1 R=1 R = 1 , the criterion reduces to the stabilizer local recovery condition in [13 ] .
Bounds without a purity assumption. We derive a redundancy bound for ( r , δ ) (r,\delta) ( r , δ ) -SSLRCs with disjoint repair sets (Proposition IV.1). When two disjoint erased sets of an ( r , δ ) (r,\delta) ( r , δ ) -SSLRC are globally correctable, their supported gauge spaces and the stabilizers on the remaining coordinates bound K K K and R R R (see Theorem IV.3). The mixed counting bound in Theorem IV.7 constrains K R KR K R for fewer than δ \delta δ known erasures in each of several pairwise disjoint repair sets and at most t t t errors at unknown positions outside their union, where 2 t < d 2t<d 2 t < d .
Bounds for pure codes. We establish a rank-accumulation lemma for additive ( r , δ ) (r,\delta) ( r , δ ) -cLRCs based on the arguments in [7 Lemma 1] and [19 proof of Corollary 2] (see Lemma V.3 ). Using this result to shorten the comparison codes, we derive Singleton-type and alphabet-dependent bounds for pure ( r , δ ) (r,\delta) ( r , δ ) -SSLRCs (see Theorems V.4 and V.7 ). Moreover, by applying the corresponding classical bounds to these shortened codes, we obtain explicit Griesmer, Hamming, and Plotkin bounds (see Theorems V.10 and V.12 ).
Optimal constructions of pure codes. We determine the parameters of tensor products of SSCs and trade logical and gauge dimensions while preserving the stabilizer label space, minimum distance, and repair cover (see Propositions VI.1 and VI.2). We apply these results to construct nine classes of pure ( r , δ ) (r,\delta) ( r , δ ) -SSLRCs from MDS stabilizer codes, Reed-Solomon codes, and MDS Hermitian hulls (see Table I ). These pure codes simultaneously attain the Singleton-type bound (42) , the alphabet-dependent bound (44) , the F q 2 \mathbb{F}_{q^2} F q 2 -linear Griesmer bound (50) , and the redundancy bound (33) .
Table I. Optimal pure F q \mathbb{F}_q F q -linear ( r , δ ) (r,\delta) ( r , δ ) -SSLRCS attaining the bounds (41), (43), (49) and (32).
Organization of the Paper
The rest of this paper is organized as follows. Section II reviews the notation and preliminary results for ( r , δ ) (r,\delta) ( r , δ ) -SSLRCs. Section III establishes the algebraic characterization of local recovery and presents the CSS and Hermitian hull constructions for ( r , δ ) (r,\delta) ( r , δ ) -SSLRCs. Section IV derives bounds without assuming purity, while Section V establishes further bounds for pure ( r , δ ) (r,\delta) ( r , δ ) -SSLRCs. Section VI constructs nine classes of pure ( r , δ ) (r,\delta) ( r , δ ) -SSLRCs that simultaneously attain the stated bounds from Sections IV and V. Section VII concludes the paper and discusses further questions.
Preliminaries
In this section, we recall basic definitions and results on classical ( r , δ ) (r,\delta) ( r , δ ) -cLRCs and SSCs, and state the global erasure correction criterion needed for ( r , δ ) (r,\delta) ( r , δ ) -SSLRCs.
Classical Codes in Coding Theory
Let q = p m q=p^m q = p m , where p p p is prime and m ≥ 1 m\ge1 m ≥ 1 is an integer, and let F q \mathbb{F}_q F q denote the finite field of order q q q . Let F q ∗ \mathbb{F}_q^* F q ∗ denote the multiplicative group of F q \mathbb{F}_q F q , and F q n \mathbb{F}_q^n F q n the n n n -dimensional vector space over F q \mathbb{F}_q F q . A linear code C C C with parameters [ n , k , d ] q [n,k,d]_q [ n , k , d ] q is a k k k -dimensional linear subspace of F q n \mathbb{F}_q^n F q n with minimum Hamming distance d = d ( C ) = min { wt ( c ) : c ∈ C ∖ { 0 } } d=d(C)=\min\{\operatorname{wt}(c):c\in C\setminus\{0\}\} d = d ( C ) = min { wt ( c ) : c ∈ C ∖ { 0 }} , where wt ( c ) \operatorname{wt}(c) wt ( c ) is the number of nonzero coordinates of c c c . The support of c = ( c 1 , … , c n ) c=(c_1,\ldots,c_n) c = ( c 1 , … , c n ) is the set supp ( c ) = { i : c i ≠ 0 , 1 ≤ i ≤ n } \operatorname{supp}(c)=\{i:c_i\ne0,1\le i\le n\} supp ( c ) = { i : c i = 0 , 1 ≤ i ≤ n } . Thus wt ( c ) = ∣ supp ( c ) ∣ \operatorname{wt}(c)=|\operatorname{supp}(c)| wt ( c ) = ∣ supp ( c ) ∣ . A generator matrix for an [ n , k , d ] q [n,k,d]_q [ n , k , d ] q code C C C is a k × n k\times n k × n matrix M M M whose rows form a basis for C C C . Let u = ( u 1 , … , u n ) u=(u_1,\ldots,u_n) u = ( u 1 , … , u n ) , v = ( v 1 , … , v n ) ∈ F q n v=(v_1,\ldots,v_n)\in\mathbb{F}_q^n v = ( v 1 , … , v n ) ∈ F q n . The Euclidean inner product of u u u and v v v is defined by ⟨ u , v ⟩ = ∑ i = 1 n u i v i \langle u,v\rangle=\sum_{i=1}^n u_i v_i ⟨ u , v ⟩ = ∑ i = 1 n u i v i , and the Euclidean dual code of C C C is defined by
C ⊥ = { v ∈ F q n : ⟨ u , v ⟩ = 0 for all u ∈ C } . C^\perp=\{v\in\mathbb{F}_q^n:\langle u,v\rangle=0\text{ for all }u\in C\}. C ⊥ = { v ∈ F q n : ⟨ u , v ⟩ = 0 for all u ∈ C } .
For u , v ∈ F q 2 n u,v\in\mathbb{F}_{q^2}^n u , v ∈ F q 2 n , define the Hermitian inner product by ⟨ u , v ⟩ H = ∑ i = 1 n u i v i q \langle u,v\rangle_H=\sum_{i=1}^n u_i v_i^q ⟨ u , v ⟩ H = ∑ i = 1 n u i v i q . For an F q 2 \mathbb{F}_{q^2} F q 2 -linear code C ≤ F q 2 n C\le\mathbb{F}_{q^2}^n C ≤ F q 2 n , define
C ⊥ H = { v ∈ F q 2 n : ⟨ u , v ⟩ H = 0 for all u ∈ C } . C^{\perp_H}=\{v\in\mathbb{F}_{q^2}^n:\langle u,v\rangle_H=0\text{ for all }u\in C\}. C ⊥ H = { v ∈ F q 2 n : ⟨ u , v ⟩ H = 0 for all u ∈ C } .
Moreover, the Hermitian hull of C C C is defined as
Hull H ( C ) = C ∩ C ⊥ H . \operatorname{Hull}_H(C)=C\cap C^{\perp_H}. Hull H ( C ) = C ∩ C ⊥ H .
The parameters of an [ n , k , d ] q [n,k,d]_q [ n , k , d ] q linear code C C C satisfy the classical Singleton bound d ≤ n − k + 1 d\le n-k+1 d ≤ n − k + 1 [32 ] . We call C C C a maximum distance separable (MDS) code if d = n − k + 1 d=n-k+1 d = n − k + 1 . Generalized Reed-Solomon (GRS) codes are MDS.
Definition II.1. [20 ] Let n n n be any integer with 1 ≤ n ≤ q 1\le n\le q 1 ≤ n ≤ q . Let a = ( a 1 , … , a n ) \boldsymbol{a}=(a_1,\ldots,a_n) a = ( a 1 , … , a n ) be an n n n -tuple of distinct elements of F q \mathbb{F}_q F q , and let v = ( v 1 , … , v n ) \boldsymbol{v}=(v_1,\ldots,v_n) v = ( v 1 , … , v n ) be an n n n -tuple of nonzero, but not necessarily distinct, elements of F q \mathbb{F}_q F q . Let 1 ≤ k ≤ n 1\le k\le n 1 ≤ k ≤ n . Then the code
GRS k ( a , v ) = { ( v 1 f ( a 1 ) , … , v n f ( a n ) ) : f ∈ F q [ x ] , deg f < k } \operatorname{GRS}_k(\boldsymbol{a},\boldsymbol{v})=\{(v_1f(a_1),\ldots,v_nf(a_n)):f\in\mathbb{F}_q[x],\ \deg f<k\} GRS k ( a , v ) = {( v 1 f ( a 1 ) , … , v n f ( a n )) : f ∈ F q [ x ] , deg f < k }
is called a GRS code. For v = 1 v = 1 v = 1 , it is called a Reed-Solomon (RS) code.
We recall the following standard results on GRS codes.
Lemma II.2. [20 ] With the above notation, the following statements hold.
GRS k ( a , v ) \operatorname{GRS}_{k}(\mathbf{a},\mathbf{v}) GRS k ( a , v ) is an [ n , k , n − k + 1 ] q [n,k,n-k+1]_{q} [ n , k , n − k + 1 ] q MDS code,
If k < n k<n k < n , its Euclidean dual code GRS k ( a , v ) ⊥ \operatorname{GRS}_{k}(\mathbf{a},\mathbf{v})^{\perp} GRS k ( a , v ) ⊥ is an [ n , n − k , k + 1 ] q [n,n-k,k+1]_{q} [ n , n − k , k + 1 ] q MDS code,
For a GRS code over F q 2 \mathbb{F}_{q^{2}} F q 2 with 1 ≤ k < n ≤ q 1\leq k<n\leq q 1 ≤ k < n ≤ q , its Hermitian dual has parameters [ n , n − k , k + 1 ] q 2 [n,n-k,k+1]_{q^{2}} [ n , n − k , k + 1 ] q 2 ,
A generator matrix of GRS k ( a , v ) \operatorname{GRS}_{k}(\mathbf{a},\mathbf{v}) GRS k ( a , v ) is
( v 1 v 2 ⋯ v n v 1 a 1 v 2 a 2 ⋯ v n a n v 1 a 1 2 v 2 a 2 2 ⋯ v n a n 2 ⋮ ⋮ ⋱ ⋮ v 1 a 1 k − 1 v 2 a 2 k − 1 ⋯ v n a n k − 1 ) . \begin{pmatrix}
v_{1} & v_{2} & \cdots& v_{n}\\
v_{1}a_{1} & v_{2}a_{2} & \cdots& v_{n}a_{n}\\
v_{1}a_{1}^{2} & v_{2}a_{2}^{2} & \cdots& v_{n}a_{n}^{2}\\
\vdots& \vdots& \ddots& \vdots\\
v_{1}a_{1}^{k-1} & v_{2}a_{2}^{k-1} & \cdots& v_{n}a_{n}^{k-1}
\end{pmatrix}. v 1 v 1 a 1 v 1 a 1 2 ⋮ v 1 a 1 k − 1 v 2 v 2 a 2 v 2 a 2 2 ⋮ v 2 a 2 k − 1 ⋯ ⋯ ⋯ ⋱ ⋯ v n v n a n v n a n 2 ⋮ v n a n k − 1 .
We next recall basic definitions and results on additive codes.
Definition II.3. [5 ] , [10 ] Let Q = p e Q=p^{e} Q = p e , where p p p is prime and e ≥ 1 e\geq1 e ≥ 1 . An additive code of length n n n over F Q \mathbb{F}_{Q} F Q is an additive subgroup C C C of F Q n \mathbb{F}_{Q}^{n} F Q n , or equivalently, an F p \mathbb{F}_{p} F p -linear subspace of F Q n \mathbb{F}_{Q}^{n} F Q n . The normalized dimension of C C C is
κ = log Q ∣ C ∣ = dim p C e . \kappa=\log_{Q}|C|=\frac{\dim_{p}C}{e}. κ = log Q ∣ C ∣ = e dim p C . For C ≠ { 0 } C\neq\{0\} C = { 0 } , its minimum Hamming distance is
d = d ( C ) = min c ∈ C ∖ { 0 } wt ( c ) . d=d(C)=\min_{c\in C\setminus\{0\}}\operatorname{wt}(c). d = d ( C ) = c ∈ C ∖ { 0 } min wt ( c ) . We denote the parameters of C C C by ( n , M , d ) Q (n,M,d)_{Q} ( n , M , d ) Q , where M = ∣ C ∣ M=|C| M = ∣ C ∣ . If C C C is F Q \mathbb{F}_{Q} F Q -linear, then κ = dim Q C \kappa=\dim_{Q}C κ = dim Q C and its parameters are written as [ n , κ , d ] Q [n,\kappa,d]_{Q} [ n , κ , d ] Q .
Let C ⊆ F Q n C\subseteq\mathbb{F}_{Q}^{n} C ⊆ F Q n be an additive code. Let n ≥ 1 n\geq1 n ≥ 1 be an integer, and let [ n ] ≜ { 1 , 2 , … , n } [n]\triangleq\{1,2,\ldots,n\} [ n ] ≜ { 1 , 2 , … , n } . For T ⊆ [ n ] T\subseteq[n] T ⊆ [ n ] , let π T : F Q n → F Q ∣ T ∣ \pi_{T}:\mathbb{F}_{Q}^{n}\to\mathbb{F}_{Q}^{|T|} π T : F Q n → F Q ∣ T ∣ denote the coordinate projection onto T T T . The punctured code obtained by deleting the coordinates in T c = [ n ] ∖ T T^{c}=[n]\setminus T T c = [ n ] ∖ T is C ∣ T = π T ( C ) C|_{T}=\pi_{T}(C) C ∣ T = π T ( C ) . Let
C T = { c ∈ C : supp ( c ) ⊆ T } C_{T}=\{c\in C:\operatorname{supp}(c)\subseteq T\} C T = { c ∈ C : supp ( c ) ⊆ T }
be the subcode of C C C supported in T T T . The shortened code obtained by deleting the coordinates in T c T^{c} T c from the codewords of C T C_{T} C T is denoted by sh T ( C ) = π T ( C T ) \operatorname{sh}_{T}(C)=\pi_{T}(C_{T}) sh T ( C ) = π T ( C T ) (see [4 Definition 20] ).
Let F \mathbb{F} F be a finite field. For a linear map φ : U → V \varphi:U\to V φ : U → V between finite-dimensional vector spaces over F \mathbb{F} F , the rank-nullity formula states that
dim F U = dim F ker φ + rank F φ . (1) \dim_{\mathbb{F}}U=\dim_{\mathbb{F}}\ker\varphi+\operatorname{rank}_{\mathbb{F}}\varphi. \tag*{(1)} dim F U = dim F ker φ + rank F φ . (1)
In particular, the restriction π T c ∣ C \left.\pi_{T^{c}}\right|_{C} π T c ∣ C has kernel C T C_{T} C T and image π T c ( C ) \pi_{T^{c}}(C) π T c ( C ) . Since π T \pi_{T} π T maps C T C_{T} C T isomorphically onto sh T ( C ) \operatorname{sh}_{T}(C) sh T ( C ) , it follows from (1) that
dim p sh T ( C ) = dim p C T = dim p C − dim p π T c ( C ) . \dim_{p}\operatorname{sh}_{T}(C)=\dim_{p}C_{T}=\dim_{p}C-\dim_{p}\pi_{T^{c}}(C). dim p sh T ( C ) = dim p C T = dim p C − dim p π T c ( C ) .
Classical and Quantum ( r , δ ) (r,\delta) ( r , δ ) -LRCs
We begin by defining classical codes with ( r , δ ) (r,\delta) ( r , δ ) -locality.
Definition II.4. [13 ] , [35 ] An [ n , k , d ] q [n,k,d]_{q} [ n , k , d ] q linear code C ⊆ F q n C\subseteq\mathbb{F}_{q}^{n} C ⊆ F q n is an ( r , δ ) (r,\delta) ( r , δ ) -LRC (or an LRC with locality ( r , δ ) (r,\delta) ( r , δ ) ) if for each index i ∈ [ n ] i\in[n] i ∈ [ n ] , there exists a set of indices Γ i ⊆ [ n ] \Gamma_{i}\subseteq[n] Γ i ⊆ [ n ] that contains i i i and satisfies the following two conditions:
δ ≤ ∣ Γ i ∣ ≤ r + δ − 1 \delta\leq|\Gamma_{i}|\leq r+\delta-1 δ ≤ ∣ Γ i ∣ ≤ r + δ − 1 ,
the minimum distance of the code C ∣ Γ i C|_{\Gamma_{i}} C ∣ Γ i obtained by deleting code symbols c i c_{i} c i , i ∈ [ n ] ∖ Γ i i\in[n]\setminus\Gamma_{i} i ∈ [ n ] ∖ Γ i , is at least δ \delta δ .
Remark II.5. The sets Γ i \Gamma_{i} Γ i form a repair cover. If C ∣ Γ i C|_{\Gamma_{i}} C ∣ Γ i is nonzero, then δ ≤ d ( C ∣ Γ i ) ≤ ∣ Γ i ∣ \delta\leq d(C|_{\Gamma_{i}})\leq|\Gamma_{i}| δ ≤ d ( C ∣ Γ i ) ≤ ∣ Γ i ∣ .
For ( r , δ ) (r,\delta) ( r , δ ) -cLRCs, we recall the following Singleton-like bound from [35 ] .
Lemma II.6. Let C C C be an ( r , δ ) (r,\delta) ( r , δ ) -cLRC with parameters [ n , k , d ] q [n,k,d]_{q} [ n , k , d ] q . Then
d ≤ n − k + 1 − ( ⌈ k r ⌉ − 1 ) ( δ − 1 ) . d\leq n-k+1-\left(\left\lceil\frac{k}{r}\right\rceil-1\right)(\delta-1). d ≤ n − k + 1 − ( ⌈ r k ⌉ − 1 ) ( δ − 1 ) .
For standard definitions and notation concerning quantum stabilizer codes, we refer the reader to [13 ] , [21 ] . Following [13 ] , we recall the definitions and basic results for ( r , δ ) (r,\delta) ( r , δ ) -qLRCs.
Definition II.7. [13 Definition 9] Given two sets of indices ∅ ≠ I ⊊ J ⊆ [ n ] \varnothing\ne I \subsetneq J \subseteq[n] ∅ = I ⊊ J ⊆ [ n ] , a quantum code Q ⊆ C q n \mathcal{Q} \subseteq\mathbb{C}^{q^n} Q ⊆ C q n is said to be an ( I , J ) (I,J) ( I , J ) -locally recoverable code if there exists a trace-preserving quantum operation R Q , I J \mathcal{R}^{J}_{\mathcal{Q},I} R Q , I J , which acts only on the qudits corresponding to J J J and leaves the remaining qudits unchanged, such that
R Q , I J ∘ Γ I ( ∣ φ ⟩ ⟨ φ ∣ ) = ∣ φ ⟩ ⟨ φ ∣ \mathcal{R}^{J}_{\mathcal{Q},I} \circ\Gamma^{I}(|\varphi\rangle\langle\varphi|)=|\varphi\rangle\langle\varphi| R Q , I J ∘ Γ I ( ∣ φ ⟩ ⟨ φ ∣ ) = ∣ φ ⟩ ⟨ φ ∣
for all ∣ φ ⟩ ∈ Q |\varphi\rangle\in\mathcal{Q} ∣ φ ⟩ ∈ Q , where Γ I \Gamma^{I} Γ I is defined in [13 p. 9] .
Using this definition, we introduce ( r , δ ) (r,\delta) ( r , δ ) -qLRCs as follows.
Definition II.8. [13 Definition 10] A quantum code Q \mathcal{Q} Q is said to be an ( r , δ ) (r,\delta) ( r , δ ) -qLRC if for each index i ∈ [ n ] i \in[n] i ∈ [ n ] , there exists a set J ⊆ [ n ] J \subseteq[n] J ⊆ [ n ] containing i i i with ∣ J ∣ ≤ r + δ − 1 |J| \le r+\delta-1 ∣ J ∣ ≤ r + δ − 1 such that, for every subset I ⊆ J I \subseteq J I ⊆ J of cardinality δ − 1 \delta-1 δ − 1 , Q \mathcal{Q} Q is ( I , J ) (I,J) ( I , J ) -locally recoverable.
For ( r , δ ) (r,\delta) ( r , δ ) -qLRCs, we recall the Singleton-type bound in [13 Theorem 31] .
Lemma II.9. [13 Theorem 31] An ( r , δ ) (r,\delta) ( r , δ ) -qLRC with parameters [ [ n , k , ≥ d ( C ) ] ] q [[n,k,\ge d(C)]]_q [[ n , k , ≥ d ( C )] ] q satisfies
k + 2 d ( C ) + 2 ( ⌈ n + k 2 r ⌉ − 1 ) ( δ − 1 ) ≤ n + 2 , (2) k+2d(C)+2\left(\left\lceil\frac{n+k}{2r}\right\rceil-1\right)(\delta-1)\le n+2, \tag*{(2)} k + 2 d ( C ) + 2 ( ⌈ 2 r n + k ⌉ − 1 ) ( δ − 1 ) ≤ n + 2 , (2)
where C C C is a Euclidean or Hermitian dual-containing ( r , δ ) (r,\delta) ( r , δ ) -cLRC with parameters [ n , n + k 2 , d ( C ) ] [n,\frac{n+k}{2},d(C)] [ n , 2 n + k , d ( C )] .
Pauli Label Space and Additive Symplectic Geometry
Let q = p m q=p^m q = p m , where p p p is prime and m ≥ 1 m\ge1 m ≥ 1 is an integer. For a quantum register X X X , let H X \mathcal{H}_X H X denote its finite-dimensional complex Hilbert space. The space of linear operators on H X \mathcal{H}_X H X is
B ( H X ) = { M : H X → H X ∣ M is C -linear } . \mathcal{B}(\mathcal{H}_X)=\{M:\mathcal{H}_X\to\mathcal{H}_X\mid M\text{ is }\mathbb{C}\text{-linear}\}. B ( H X ) = { M : H X → H X ∣ M is C -linear } .
Let I X I_X I X denote the identity operator on H X \mathcal{H}_X H X . For M ∈ B ( H X ) M\in\mathcal{B}(\mathcal{H}_X) M ∈ B ( H X ) , let M † M^\dagger M † denote its adjoint and Tr ( M ) \operatorname{Tr}(M) Tr ( M ) its trace. A state of X X X is represented by a density operator ρ ∈ B ( H X ) \rho\in\mathcal{B}(\mathcal{H}_X) ρ ∈ B ( H X ) satisfying ρ ⪰ 0 \rho\succeq0 ρ ⪰ 0 and Tr ( ρ ) = 1 \operatorname{Tr}(\rho)=1 Tr ( ρ ) = 1 .
Each physical coordinate i ∈ [ n ] i\in[n] i ∈ [ n ] corresponds to a qudit with Hilbert space H i = C q \mathcal{H}_i=\mathbb{C}^q H i = C q . For T ⊆ [ n ] T\subseteq[n] T ⊆ [ n ] , write H T = ⨂ i ∈ T H i \mathcal{H}_T=\bigotimes_{i\in T}\mathcal{H}_i H T = ⨂ i ∈ T H i , with H ∅ = C \mathcal{H}_{\varnothing}=\mathbb{C} H ∅ = C . The joint Hilbert space of the n n n qudits is H [ n ] = ⨂ i = 1 n H i = ( C q ) ⊗ n \mathcal{H}_{[n]}=\bigotimes_{i=1}^{n}\mathcal{H}_i=(\mathbb{C}^q)^{\otimes n} H [ n ] = ⨂ i = 1 n H i = ( C q ) ⊗ n . For a state ρ \rho ρ on H [ n ] \mathcal{H}_{[n]} H [ n ] , its reduced state on T c = [ n ] ∖ T T^c=[n]\setminus T T c = [ n ] ∖ T is ρ T c = Tr T ( ρ ) \rho_{T^c}=\operatorname{Tr}_T(\rho) ρ T c = Tr T ( ρ ) , where Tr T \operatorname{Tr}_T Tr T denotes the partial trace over the qudits in T T T .
Let ζ p = e 2 π i / p \zeta_p=e^{2\pi i/p} ζ p = e 2 π i / p , and let Tr F q / F p \operatorname{Tr}_{\mathbb{F}_q/\mathbb{F}_p} Tr F q / F p denote the field trace from F q \mathbb{F}_q F q to F p \mathbb{F}_p F p . For a , b ∈ F q a,b\in\mathbb{F}_q a , b ∈ F q , define the single-qudit Pauli operators X ( a ) , Z ( b ) : C q → C q X(a),Z(b):\mathbb{C}^q\to\mathbb{C}^q X ( a ) , Z ( b ) : C q → C q by
X ( a ) ∣ u ⟩ = ∣ u + a ⟩ , Z ( b ) ∣ u ⟩ = ζ p Tr F q / F p ( b u ) ∣ u ⟩ , X(a)|u\rangle=|u+a\rangle,\qquad Z(b)|u\rangle=\zeta_p^{\operatorname{Tr}_{\mathbb{F}_q/\mathbb{F}_p}(bu)}|u\rangle, X ( a ) ∣ u ⟩ = ∣ u + a ⟩ , Z ( b ) ∣ u ⟩ = ζ p Tr F q / F p ( b u ) ∣ u ⟩ ,
where { ∣ u ⟩ : u ∈ F q } \{|u\rangle:u\in\mathbb{F}_q\} { ∣ u ⟩ : u ∈ F q } is the computational basis of C q \mathbb{C}^q C q .
Definition II.10. [21 ] Let W = F q n × F q n W=\mathbb{F}_q^n\times\mathbb{F}_q^n W = F q n × F q n . For w = ( x , z ) ∈ W w=(x,z)\in W w = ( x , z ) ∈ W , with x = ( x 1 , … , x n ) x=(x_1,\ldots,x_n) x = ( x 1 , … , x n ) and z = ( z 1 , … , z n ) z=(z_1,\ldots,z_n) z = ( z 1 , … , z n ) , define the n n n -qudit Pauli operator
P w = ⨂ i = 1 n X ( x i ) Z ( z i ) : H [ n ] ⟶ H [ n ] . P_w=\bigotimes_{i=1}^{n}X(x_i)Z(z_i):\mathcal{H}_{[n]}\longrightarrow\mathcal{H}_{[n]}. P w = i = 1 ⨂ n X ( x i ) Z ( z i ) : H [ n ] ⟶ H [ n ] .
For every n n n -qudit Pauli operator U U U , there is a unique w ∈ W w\in W w ∈ W such that U = ξ P w U=\xi P_w U = ξ P w for some ξ ∈ C \xi\in\mathbb{C} ξ ∈ C with ∣ ξ ∣ = 1 |\xi|=1 ∣ ξ ∣ = 1 . We call w w w the Pauli label of U U U and W W W the Pauli label space .
For w = ( x , z ) w=(x,z) w = ( x , z ) and w ′ = ( x ′ , z ′ ) w'=(x',z') w ′ = ( x ′ , z ′ ) in W W W , define the symplectic and trace-symplectic forms, respectively, by
[ w , w ′ ] s = x ⋅ z ′ − z ⋅ x ′ , [ w , w ′ ] t = Tr F q / F p ( [ w , w ′ ] s ) . [w,w']_s=x\cdot z'-z\cdot x',\qquad[w,w']_t=\operatorname{Tr}_{\mathbb{F}_q/\mathbb{F}_p}([w,w']_s). [ w , w ′ ] s = x ⋅ z ′ − z ⋅ x ′ , [ w , w ′ ] t = Tr F q / F p ([ w , w ′ ] s ) .
where x ⋅ z ′ = ∑ i = 1 n x i z i x\cdot z'=\sum_{i=1}^{n}x_i z_i x ⋅ z ′ = ∑ i = 1 n x i z i . The support and symplectic weight of w w w are
supp ( w ) = { i ∈ [ n ] : ( x i , z i ) ≠ ( 0 , 0 ) } , swt ( w ) = ∣ supp ( w ) ∣ . \operatorname{supp}(w)=\{i\in[n]:(x_i,z_i)\ne(0,0)\},\qquad\operatorname{swt}(w)=|\operatorname{supp}(w)|. supp ( w ) = { i ∈ [ n ] : ( x i , z i ) = ( 0 , 0 )} , swt ( w ) = ∣ supp ( w ) ∣.
Lemma II.11. [21 ] For w = ( x , z ) w=(x,z) w = ( x , z ) and w ′ = ( x ′ , z ′ ) w'=(x',z') w ′ = ( x ′ , z ′ ) in W W W , the following statements hold.
(i) The multiplication and commutation relations are
P w P w ′ = ζ p Tr F q / F p ( z ⋅ x ′ ) P w + w ′ , P w P w ′ = ζ p − [ w , w ′ ] t P w ′ P w . (3) P_wP_{w'}=\zeta_p^{\operatorname{Tr}_{\mathbb{F}_q/\mathbb{F}_p}(z\cdot x')}P_{w+w'},\qquad P_wP_{w'}=\zeta_p^{-[w,w']_t}P_{w'}P_w. \tag*{(3)} P w P w ′ = ζ p Tr F q / F p ( z ⋅ x ′ ) P w + w ′ , P w P w ′ = ζ p − [ w , w ′ ] t P w ′ P w . (3)
(ii) The set E n = { P w : w ∈ W } \mathcal{E}_n=\{P_w:w\in W\} E n = { P w : w ∈ W } is a nice error basis on H [ n ] \mathcal{H}_{[n]} H [ n ] .
(iii) The operators in E n \mathcal{E}_n E n form an orthogonal basis of B ( H [ n ] ) \mathcal{B}(\mathcal{H}_{[n]}) B ( H [ n ] ) and satisfy
Tr ( P w † P w ′ ) = q n δ w , w ′ . (4) \operatorname{Tr}(P_w^\dagger P_{w'})=q^n\delta_{w,w'}. \tag*{(4)} Tr ( P w † P w ′ ) = q n δ w , w ′ . (4)
By (3), two Pauli operators commute if and only if their labels are trace-symplectically orthogonal. We now consider orthogonality for additive subspaces of W W W .
Definition II.12. An additive subspace of W W W is an F p \mathbb{F}_p F p -linear subspace of W W W . For an additive subspace L ≤ W L \leq W L ≤ W , define its trace-symplectic and symplectic orthogonal complements, respectively, by
L ⊥ t = { w ∈ W : [ w , v ] t = 0 for all v ∈ L } , L^{\perp_t} = \{w \in W : [w,v]_t = 0 \text{ for all } v \in L\}, L ⊥ t = { w ∈ W : [ w , v ] t = 0 for all v ∈ L } , L ⊥ s = { w ∈ W : [ w , v ] s = 0 for all v ∈ L } . L^{\perp_s} = \{w \in W : [w,v]_s = 0 \text{ for all } v \in L\}. L ⊥ s = { w ∈ W : [ w , v ] s = 0 for all v ∈ L } .
We call L L L isotropic if L ⊆ L ⊥ t L \subseteq L^{\perp_t} L ⊆ L ⊥ t . The radical of L L L with respect to the trace-symplectic form is rad L = L ∩ L ⊥ t \operatorname{rad} L = L \cap L^{\perp_t} rad L = L ∩ L ⊥ t .
The trace-symplectic form on W W W is alternating and nondegenerate over F p \mathbb{F}_p F p (see [2 proof of Lemma 6] ). Since dim p W = 2 m n \dim_p W = 2mn dim p W = 2 mn , every additive subspace L ≤ W L \leq W L ≤ W satisfies
dim p L + dim p L ⊥ t = 2 m n . (5) \dim_p L + \dim_p L^{\perp_t} = 2mn. \tag*{(5)} dim p L + dim p L ⊥ t = 2 mn . (5)
If L L L is F q \mathbb{F}_q F q -linear, then L ⊥ t = L ⊥ s L^{\perp_t} = L^{\perp_s} L ⊥ t = L ⊥ s [15 proof of Theorem 2] .
For T ⊆ [ n ] T \subseteq[n] T ⊆ [ n ] , let π T : W → F q T × F q T \pi_T: W \to\mathbb{F}_q^T \times\mathbb{F}_q^T π T : W → F q T × F q T denote the coordinate projection given by
π T ( x , z ) = ( ( x i ) i ∈ T , ( z i ) i ∈ T ) . \pi_T(x,z) = ((x_i)_{i\in T},(z_i)_{i\in T}). π T ( x , z ) = (( x i ) i ∈ T , ( z i ) i ∈ T ) .
Let W T = { w ∈ W : supp ( w ) ⊆ T } W_T = \{w \in W : \operatorname{supp}(w) \subseteq T\} W T = { w ∈ W : supp ( w ) ⊆ T } and L T = L ∩ W T L_T = L \cap W_T L T = L ∩ W T . The puncturing and shortening of L L L to T T T are, respectively,
L ∣ T = π T ( L ) , sh T ( L ) = π T ( L T ) . L|_T = \pi_T(L), \qquad\operatorname{sh}_T(L) = \pi_T(L_T). L ∣ T = π T ( L ) , sh T ( L ) = π T ( L T ) .
Thus π T \pi_T π T maps L T L_T L T isomorphically onto sh T ( L ) \operatorname{sh}_T(L) sh T ( L ) ; its inverse extends each label by zero outside T T T .
Lemma II.13. Let L ≤ W L \leq W L ≤ W be an additive subspace and let T ⊆ [ n ] T \subseteq[n] T ⊆ [ n ] . By projection–shortening duality [11 Sec. II-H] for the trace-symplectic form, we have
sh T ( L ⊥ t ) = ( π T ( L ) ) ⊥ t , \operatorname{sh}_T(L^{\perp_t}) = \bigl(\pi_T(L)\bigr)^{\perp_t}, sh T ( L ⊥ t ) = ( π T ( L ) ) ⊥ t , where the orthogonal complement on the right is taken in F q T × F q T \mathbb{F}_q^T \times\mathbb{F}_q^T F q T × F q T . Consequently,
dim p ( W T ∩ L ⊥ t ) = 2 m ∣ T ∣ − dim p π T ( L ) . \dim_p(W_T \cap L^{\perp_t}) = 2m|T| - \dim_p \pi_T(L). dim p ( W T ∩ L ⊥ t ) = 2 m ∣ T ∣ − dim p π T ( L ) .
We next identify the Pauli label space W W W with F q 2 n \mathbb{F}_{q^2}^n F q 2 n to relate trace-symplectic orthogonality to Hermitian duality. Fix α ∈ F q 2 ∖ F q \alpha\in\mathbb{F}_{q^2} \setminus\mathbb{F}_q α ∈ F q 2 ∖ F q . Since { 1 , α } \{1,\alpha\} { 1 , α } is an F q \mathbb{F}_q F q -basis of F q 2 \mathbb{F}_{q^2} F q 2 , the map
ψ : W ⟶ F q 2 n , ψ ( x , z ) = x + α z , (6) \psi: W \longrightarrow\mathbb{F}_{q^2}^n,\qquad\psi(x,z) = x + \alpha z, \tag*{(6)} ψ : W ⟶ F q 2 n , ψ ( x , z ) = x + α z , (6)
is an F q \mathbb{F}_q F q -linear isomorphism. For each coordinate i i i , we have x i + α z i = 0 x_i + \alpha z_i = 0 x i + α z i = 0 if and only if ( x i , z i ) = ( 0 , 0 ) (x_i,z_i) = (0,0) ( x i , z i ) = ( 0 , 0 ) . Hence, for every w ∈ W w \in W w ∈ W ,
supp ( ψ ( w ) ) = supp ( w ) , wt ( ψ ( w ) ) = swt ( w ) . \operatorname{supp}(\psi(w)) = \operatorname{supp}(w), \qquad\operatorname{wt}(\psi(w)) = \operatorname{swt}(w). supp ( ψ ( w )) = supp ( w ) , wt ( ψ ( w )) = swt ( w ) .
We also use ψ \psi ψ for the corresponding map on coordinates in T T T , so that π T ( ψ ( w ) ) = ψ ( π T ( w ) ) \pi_T(\psi(w)) = \psi(\pi_T(w)) π T ( ψ ( w )) = ψ ( π T ( w )) .
Lemma II.14. [21 ] Let D ≤ F q 2 n D \leq\mathbb{F}_{q^2}^n D ≤ F q 2 n be an F q 2 \mathbb{F}_{q^2} F q 2 -linear code. Under the map ψ \psi ψ in (6), its Hermitian dual satisfies
D ⊥ H = ψ ( ( ψ − 1 ( D ) ) ⊥ t ) . D^{\perp_H} = \psi\bigl((\psi^{-1}(D))^{\perp_t}\bigr). D ⊥ H = ψ ( ( ψ − 1 ( D ) ) ⊥ t ) .
Subsystem Stabilizer Codes
A subsystem stabilizer code encodes information in a logical subsystem A A A . Its stabilizer group determines the code space, on which gauge operators act only on the gauge subsystem B B B [34 ] .
Let G G G be an additive subspace of the Pauli label space W W W , and let G \mathcal{G} G be the group of all Pauli operators whose labels lie in G G G . We call G \mathcal{G} G the gauge group and G G G its gauge label space . Set
S = G ∩ G ⊥ t . (7) S = G \cap G^{\perp_t}. \tag*{(7)} S = G ∩ G ⊥ t . (7)
By (3), S S S is the label space of the center Z ( G ) Z(\mathcal{G}) Z ( G ) . Choose a subgroup S ≤ Z ( G ) \mathcal{S} \leq Z(\mathcal{G}) S ≤ Z ( G ) with label space S S S whose only scalar operator is the identity. We call S \mathcal{S} S the stabilizer group and denote its common + 1 +1 + 1 eigenspace by
Q = { ∣ ψ ⟩ ∈ H n : U ∣ ψ ⟩ = ∣ ψ ⟩ for every U ∈ S } . \mathcal{Q} = \{|\psi\rangle\in\mathcal{H}_n : U|\psi\rangle= |\psi\rangle\text{ for every } U \in\mathcal{S}\}. Q = { ∣ ψ ⟩ ∈ H n : U ∣ ψ ⟩ = ∣ ψ ⟩ for every U ∈ S } .
Definition II.15. [2 , 34 ] A subsystem stabilizer code (SSC) is the code space Q \mathcal{Q} Q together with the decomposition Q ≃ H A ⊗ H B \mathcal{Q} \simeq\mathcal{H}_A \otimes\mathcal{H}_B Q ≃ H A ⊗ H B induced by G \mathcal{G} G . In this decomposition, each U ∈ G U \in\mathcal{G} U ∈ G acts on Q \mathcal{Q} Q as I A ⊗ U B I_A \otimes U_B I A ⊗ U B for some unitary U B U_B U B on H B \mathcal{H}_B H B , and the operators U B U_B U B act irreducibly on H B \mathcal{H}_B H B . We call A A A the logical subsystem and B B B the gauge subsystem .
Let Q ≃ H A ⊗ H B \mathcal{Q}\simeq\mathcal{H}_{A}\otimes\mathcal{H}_{B} Q ≃ H A ⊗ H B be an SSC with gauge label space G G G , stabilizer label space S S S , and stabilizer group S S S . Write h = dim p G h=\dim_{p}G h = dim p G , s = dim p S s=\dim_{p}S s = dim p S , K = dim H A K=\dim\mathcal{H}_{A} K = dim H A , and R = dim H B R=\dim\mathcal{H}_{B} R = dim H B . Let P P P be the orthogonal projector onto Q \mathcal{Q} Q , and let V : H A ⊗ H B → H [ n ] V:\mathcal{H}_{A}\otimes\mathcal{H}_{B}\to\mathcal{H}_{[n]} V : H A ⊗ H B → H [ n ] be an encoding isometry for this decomposition. Then V † V = I A B V^{\dagger}V=I_{AB} V † V = I A B and V V † = P VV^{\dagger}=P V V † = P . The corresponding encoding channel V : B ( H A ⊗ H B ) → B ( H [ n ] ) \mathcal{V}:\mathcal{B}(\mathcal{H}_{A}\otimes\mathcal{H}_{B})\to\mathcal{B}(\mathcal{H}_{[n]}) V : B ( H A ⊗ H B ) → B ( H [ n ] ) is defined by V ( M ) = V M V † \mathcal{V}(M)=VMV^{\dagger} V ( M ) = V M V † , where M ∈ B ( H A ⊗ H B ) M\in\mathcal{B}(\mathcal{H}_{A}\otimes\mathcal{H}_{B}) M ∈ B ( H A ⊗ H B ) .
Lemma II.16. [2 ] , [21 ] , [34 ] With the above notation, the following statements hold.
(i) The logical and gauge dimensions satisfy
K = p m n − ( h + s ) / 2 , R = p ( h − s ) / 2 . (8) K=p^{mn-(h+s)/2},\qquad R=p^{(h-s)/2}. \tag*{(8)} K = p mn − ( h + s ) /2 , R = p ( h − s ) /2 . (8) (ii) The projector P P P satisfies
P = p − s ∑ U ∈ S U , Tr ( P ) = p m n − s = K R . (9) P=p^{-s}\sum_{U\in S}U,\qquad\operatorname{Tr}(P)=p^{mn-s}=KR. \tag*{(9)} P = p − s U ∈ S ∑ U , Tr ( P ) = p mn − s = K R . (9) (iii) For each w ∈ W w\in W w ∈ W ,
V † P w V = 0 if w ∉ S ⊥ t , V † P w V = I A ⊗ U w , B if w ∈ G , V † P w V ∉ I A ⊗ B ( H B ) if w ∈ S ⊥ t ∖ G , (10) \begin{aligned}
V^{\dagger}P_{w}V&=0 &&\text{if }w\notin S^{\perp_{t}},\\
V^{\dagger}P_{w}V&=I_{A}\otimes U_{w,B} &&\text{if }w\in G,\\
V^{\dagger}P_{w}V&\notin I_{A}\otimes\mathcal{B}(\mathcal{H}_{B}) &&\text{if }w\in S^{\perp_{t}}\setminus G, \tag*{(10)}
\end{aligned} V † P w V V † P w V V † P w V = 0 = I A ⊗ U w , B ∈ / I A ⊗ B ( H B ) if w ∈ / S ⊥ t , if w ∈ G , if w ∈ S ⊥ t ∖ G , (10) where U w , B U_{w,B} U w , B is unitary on H B \mathcal{H}_{B} H B .
Remark II.17. Since S = rad G S=\operatorname{rad}G S = rad G , the induced trace-symplectic form on G / S G/S G / S is nondegenerate, and by (8), we have dim p ( G / S ) = h − s = 2 log p R \dim_{p}(G/S)=h-s=2\log_{p}R dim p ( G / S ) = h − s = 2 log p R . Moreover, Lemma II.16(iii) states the Pauli error classification of [34 ] using the additive notation of [2 Theorem 1] .
The encoding channel is V ( M ) = V M V † \mathcal{V}(M)=VMV^{\dagger} V ( M ) = V M V † for M ∈ B ( H A ⊗ H B ) M\in\mathcal{B}(\mathcal{H}_{A}\otimes\mathcal{H}_{B}) M ∈ B ( H A ⊗ H B ) . The encoded subsystems A A A and B B B need not correspond to disjoint sets of physical coordinates. By (10), the labels in S ⊥ t ∖ G S^{\perp_{t}}\setminus G S ⊥ t ∖ G are precisely those of Pauli operators that preserve Q \mathcal{Q} Q and act nontrivially on A A A . Thus, for K > 1 K>1 K > 1 , the global minimum distance is [2 Theorem 1]
d = min { swt ( w ) : w ∈ S ⊥ t ∖ G } . (11) d=\min\{\operatorname{swt}(w):w\in S^{\perp_{t}}\setminus G\}. \tag*{(11)} d = min { swt ( w ) : w ∈ S ⊥ t ∖ G } . (11)
We denote the parameters of Q \mathcal{Q} Q by ( ( n , K , R , d ) ) q ((n,K,R,d))_{q} (( n , K , R , d ) ) q . When K = q k K=q^{k} K = q k and R = q g R=q^{g} R = q g for nonnegative integers k k k and g g g , we write [ [ n , k , g , d ] ] q [[n,k,g,d]]_{q} [[ n , k , g , d ] ] q . In particular, if G G G is F q \mathbb{F}_{q} F q -linear, then it follows from (8) that
k = n − dim q G + dim q S 2 , g = dim q G − dim q S 2 . (12) k=n-\frac{\dim_{q}G+\dim_{q}S}{2},\qquad g=\frac{\dim_{q}G-\dim_{q}S}{2}. \tag*{(12)} k = n − 2 dim q G + dim q S , g = 2 dim q G − dim q S . (12)
If R = 1 R=1 R = 1 , then G = S G=S G = S and Q \mathcal{Q} Q is an ordinary stabilizer code with parameters ( ( n , K , d ) ) q ((n,K,d))_{q} (( n , K , d ) ) q , or [ [ n , k , d ] ] q [[n,k,d]]_{q} [[ n , k , d ] ] q when K = q k K=q^{k} K = q k .
Definition II.18. [2 ] An SSC with gauge label space G G G and minimum distance d d d is pure if swt ( w ) ≥ d \operatorname{swt}(w)\ge d swt ( w ) ≥ d for every nonzero w ∈ G w\in G w ∈ G . Otherwise, it is impure.
For a pure SSC with K > 1 K>1 K > 1 , every nonzero label in G G G has symplectic weight at least d d d , and by (11), the minimum weight in S ⊥ t S^{\perp_{t}} S ⊥ t is d d d . Hence
d = min { swt ( w ) : w ∈ S ⊥ t ∖ { 0 } } . d=\min\{\operatorname{swt}(w):w\in S^{\perp_{t}}\setminus\{0\}\}. d = min { swt ( w ) : w ∈ S ⊥ t ∖ { 0 }} .
Global Erasure Recovery
A quantum channel is a completely positive trace-preserving (CPTP) linear map between operator spaces [38 Definition 2.13] . We write id X \operatorname{id}_{X} id X for the identity channel on B ( H X ) \mathcal{B}(\mathcal{H}_{X}) B ( H X ) . For a known erased set E ⊆ [ n ] E\subseteq[n] E ⊆ [ n ] , the erasure channel is the partial trace Tr E \operatorname{Tr}_{E} Tr E . In global recovery, the decoder has access to all surviving qudits in E c E^{c} E c and returns a state on the full set of coordinates [ n ] [n] [ n ] .
We use the definition of subsystem correctability in [24 Definition 3.1] to formulate recovery from the erasure of a known set of coordinates.
Definition II.19. Let V \mathcal{V} V be the encoding channel of an SSC Q \mathcal{Q} Q with logical subsystem A A A and gauge subsystem B B B . A known erased set E ⊆ [ n ] E\subseteq[n] E ⊆ [ n ] is globally correctable if there exist CPTP maps
R E : B ( H E c ) ⟶ B ( H [ n ] ) , Φ E : B ( H B ) ⟶ B ( H B ) \mathcal{R}_{E}:\mathcal{B}(\mathcal{H}_{E^{c}})\longrightarrow\mathcal{B}(\mathcal{H}_{[n]}),\qquad
\Phi_{E}:\mathcal{B}(\mathcal{H}_{B})\longrightarrow\mathcal{B}(\mathcal{H}_{B}) R E : B ( H E c ) ⟶ B ( H [ n ] ) , Φ E : B ( H B ) ⟶ B ( H B ) such that
R E ∘ Tr E ∘ V = V ∘ ( id A ⊗ Φ E ) . (13) \mathcal{R}_{E}\circ\operatorname{Tr}_{E}\circ\mathcal{V}
=\mathcal{V}\circ(\operatorname{id}_{A}\otimes\Phi_{E}). \tag*{(13)} R E ∘ Tr E ∘ V = V ∘ ( id A ⊗ Φ E ) . (13) The recovery channel R E \mathcal{R}_{E} R E has access to all surviving qudits in E c E^{c} E c . For each fixed E E E , the same maps R E \mathcal{R}_{E} R E and Φ E \Phi_{E} Φ E satisfy (13) for every joint input state of A A A and B B B . The channel identity also holds after tensoring both sides with id F \operatorname{id}_{F} id F for any reference register F F F . Thus, recovery returns the output to Q \mathcal{Q} Q and preserves the joint state of A A A and F F F , while allowing the gauge state to change.
By the subsystem Knill-Laflamme condition [24 Theorem 3.5] and (10) , we obtain the following criterion for a known erasure.
Lemma II.20. Let Q Q Q be an SSC with gauge label space G G G and stabilizer label space S S S . A known erased set E ⊆ [ n ] E \subseteq[n] E ⊆ [ n ] is globally correctable if and only if
W E ∩ S ⊥ t = G E . (14) W_E \cap S^{\perp t} = G_E. \tag*{(14)} W E ∩ S ⊥ t = G E . (14) By Lemma II.13, this condition is equivalent to
dim F p π E ( S ) + dim F p G E = 2 m ∣ E ∣ . (15) \dim_{\mathbb{F}_p}\pi_E(S) + \dim_{\mathbb{F}_p}G_E = 2m|E|. \tag*{(15)} dim F p π E ( S ) + dim F p G E = 2 m ∣ E ∣. (15)
Remark II.21. If K > 1 K > 1 K > 1 and ∣ E ∣ < d |E| < d ∣ E ∣ < d , then each label in W E W_E W E has symplectic weight less than d d d . By (11) , no such label belongs to S ⊥ t ∖ G S^{\perp t} \setminus G S ⊥ t ∖ G . Hence E E E is globally correctable by Lemma II.20.
Local Recovery and Algebraic Characterization
In this section, we characterize local erasure recovery for ( r , δ ) (r,\delta) ( r , δ ) -SSLRCs without a purity assumption. We also specialize the criterion to CSS and Hermitian hull constructions.
Local Recovery and ( r , δ ) (r,\delta) ( r , δ ) -Locality
Let ∅ ≠ E ⊆ Γ ⊆ [ n ] \varnothing\ne E \subseteq\Gamma\subseteq[n] ∅ = E ⊆ Γ ⊆ [ n ] , and set H = Γ ∖ E H = \Gamma\setminus E H = Γ ∖ E and O = Γ c O = \Gamma^{c} O = Γ c . Recovery within Γ \Gamma Γ has the same correction requirement as global recovery in Definition II.19, but the decoder may access only the surviving qudits in H H H .
Definition III.1. Let V \mathcal{V} V be the encoding channel of an SSC Q Q Q with logical subsystem A A A and gauge subsystem B B B . The erasure of E E E is recoverable within Γ \Gamma Γ if there exist CPTP maps
R Γ , E : B ( H H ) ⟶ B ( H Γ ) , Φ Γ , E : B ( H B ) ⟶ B ( H B ) \mathcal{R}_{\Gamma,E}:\mathcal{B}(\mathcal{H}_H)\longrightarrow\mathcal{B}(\mathcal{H}_{\Gamma}),\quad\Phi_{\Gamma,E}:\mathcal{B}(\mathcal{H}_B)\longrightarrow\mathcal{B}(\mathcal{H}_B) R Γ , E : B ( H H ) ⟶ B ( H Γ ) , Φ Γ , E : B ( H B ) ⟶ B ( H B ) such that
( R Γ , E ⊗ id O ) ∘ Tr E ∘ V = V ∘ ( id A ⊗ Φ Γ , E ) . (16) (\mathcal{R}_{\Gamma,E}\otimes\operatorname{id}_O)\circ\operatorname{Tr}_E\circ\mathcal{V}
=\mathcal{V}\circ(\operatorname{id}_A\otimes\Phi_{\Gamma,E}). \tag*{(16)} ( R Γ , E ⊗ id O ) ∘ Tr E ∘ V = V ∘ ( id A ⊗ Φ Γ , E ) . (16) The maps may depend on Γ \Gamma Γ and E E E , but the same pair must apply to every joint input state of A A A and B B B .
Remark III.2. For every joint input state of the logical subsystem A A A , the gauge subsystem B B B , and a reference register F F F , equation (16) ensures that recovery within Γ \Gamma Γ preserves the reduced state on A F AF A F . When R = 1 R = 1 R = 1 , the gauge subsystem is one-dimensional, so recovery within Γ \Gamma Γ must restore the entire encoded state, as in the stabilizer recovery model of [13 ] .
We now define ( r , δ ) (r,\delta) ( r , δ ) -SSLRCs in terms of local erasure recovery.
Definition III.3. Let r ≥ 1 r \ge1 r ≥ 1 and δ ≥ 2 \delta\ge2 δ ≥ 2 be integers. An SSC Q Q Q of length n n n is a subsystem stabilizer ( r , δ ) (r,\delta) ( r , δ ) -locally recoverable code (( r , δ ) (r,\delta) ( r , δ ) -SSLRC) if, for each i ∈ [ n ] i \in[n] i ∈ [ n ] , there exists a repair set Γ i ⊆ [ n ] \Gamma_i \subseteq[n] Γ i ⊆ [ n ] containing i i i such that
δ ≤ ∣ Γ i ∣ ≤ r + δ − 1 (17) \delta\le|\Gamma_i| \le r+\delta-1 \tag*{(17)} δ ≤ ∣ Γ i ∣ ≤ r + δ − 1 (17) and every nonempty erased set E ⊆ Γ i E \subseteq\Gamma_i E ⊆ Γ i with ∣ E ∣ < δ |E| < \delta ∣ E ∣ < δ is recoverable within Γ i \Gamma_i Γ i in the sense of Definition III.1. The family { Γ i : i ∈ [ n ] } \{\Gamma_i:i\in[n]\} { Γ i : i ∈ [ n ]} is called a repair cover . Repair sets for different coordinates may overlap or coincide.
The erased set E E E need not contain i i i . For e = ∣ E ∣ < δ e = |E| < \delta e = ∣ E ∣ < δ , the number of surviving qudits in Γ i \Gamma_i Γ i is
∣ Γ i ∖ E ∣ = ∣ Γ i ∣ − e ≤ r + δ − 1 − e . |\Gamma_i \setminus E| = |\Gamma_i| - e \le r+\delta-1-e. ∣ Γ i ∖ E ∣ = ∣ Γ i ∣ − e ≤ r + δ − 1 − e . In particular, when e = δ − 1 e = \delta- 1 e = δ − 1 , at most r r r qudits in Γ i \Gamma_i Γ i are available for recovery.
An Algebraic Characterization for Local Recovery
In this subsection, we characterize local erasure recovery in SSCs without assuming purity (Theorem III.8). We first establish two lemmas on the correction operators and the local stabilizer measurement needed for Proposition III.6.
For the next two lemmas, let Q Q Q be an SSC with gauge label space G G G , stabilizer label space S S S , and stabilizer group S S S . Let ∅ ≠ E ⊆ Γ ⊆ [ n ] \varnothing\ne E \subseteq\Gamma\subseteq[n] ∅ = E ⊆ Γ ⊆ [ n ] , and set H = Γ ∖ E H = \Gamma\setminus E H = Γ ∖ E and O = Γ c O = \Gamma^c O = Γ c . Let V : H A ⊗ H B → H [ n ] \mathcal{V}:\mathcal{H}_A\otimes\mathcal{H}_B\to\mathcal{H}_{[n]} V : H A ⊗ H B → H [ n ] be an encoding isometry for Q Q Q , with associated encoding channel V \mathcal{V} V . Let ν = dim F p S Γ \nu= \dim_{\mathbb{F}_p}S_\Gamma ν = dim F p S Γ , and choose an F p \mathbb{F}_p F p -basis s 1 , … , s ν s_1,\ldots,s_\nu s 1 , … , s ν of S Γ S_\Gamma S Γ . For each j j j , let U j ∈ S U_j\in S U j ∈ S be the stabilizer operator with label s j s_j s j . Define the syndrome map by
χ : W E ⟶ F p ν , χ ( a ) = ( [ a , s 1 ] t , … , [ a , s ν ] t ) . \chi:W_E\longrightarrow\mathbb{F}_p^\nu,\quad\chi(a)=([a,s_1]_t,\ldots,[a,s_\nu]_t). χ : W E ⟶ F p ν , χ ( a ) = ([ a , s 1 ] t , … , [ a , s ν ] t ) .
The decoder selects a correction operator according to the measured syndrome. The following lemma provides a sufficient condition under which this operator restores the logical subsystem for all Pauli errors with the same syndrome.
Lemma III.4. With the above notation, suppose that W E ∩ S Γ ⊥ t = G E W_E \cap S_{\Gamma}^{\perp t} = G_E W E ∩ S Γ ⊥ t = G E . Then χ \chi χ is F p \mathbb{F}_p F p -linear and ker χ = G E \ker\chi= G_E ker χ = G E . For each y ∈ χ ( W E ) y \in\chi(W_E) y ∈ χ ( W E ) , choose a representative label a y ∈ W E a_y \in W_E a y ∈ W E with χ ( a y ) = y \chi(a_y) = y χ ( a y ) = y . For every a ∈ W E a \in W_E a ∈ W E , the residual operator D a : = P a χ ( a ) † P a D_a := P_{a\chi(a)}^{\dagger} P_a D a := P a χ ( a ) † P a satisfies
D a V = V ( I A ⊗ U a , B ) D_a V = V(I_A \otimes U_{a,B}) D a V = V ( I A ⊗ U a , B ) for some unitary U a , B U_{a,B} U a , B on H B \mathcal{H}_B H B .
Proof. Let s j = ( x j , z j ) s_j = (x_j,z_j) s j = ( x j , z j ) . Then
U j = ξ j ⨂ i = 1 n X ( x j , i ) Z ( z j , i ) U_j = \xi_j \bigotimes_{i=1}^{n} X(x_{j,i})Z(z_{j,i}) U j = ξ j i = 1 ⨂ n X ( x j , i ) Z ( z j , i )
for some scalar phase ξ j \xi_j ξ j . Since supp ( s j ) ⊆ Γ \operatorname{supp}(s_j) \subseteq\Gamma supp ( s j ) ⊆ Γ , we have ( x j , i , z j , i ) = ( 0 , 0 ) (x_{j,i},z_{j,i}) = (0,0) ( x j , i , z j , i ) = ( 0 , 0 ) for every i ∈ O = Γ c i \in O = \Gamma^c i ∈ O = Γ c . Thus each factor on O O O is X ( 0 ) Z ( 0 ) = I i X(0)Z(0) = I_i X ( 0 ) Z ( 0 ) = I i , and U j = U ^ j ⊗ I O U_j = \widehat{U}_j \otimes I_O U j = U j ⊗ I O , where
U ^ j = ξ j ⨂ i ∈ Γ X ( x j , i ) Z ( z j , i ) , I O = ⨂ i ∈ O I i . \widehat{U}_j = \xi_j \bigotimes_{i\in\Gamma} X(x_{j,i})Z(z_{j,i}), \qquad I_O = \bigotimes_{i\in O} I_i. U j = ξ j i ∈ Γ ⨂ X ( x j , i ) Z ( z j , i ) , I O = i ∈ O ⨂ I i .
For every ∣ φ ⟩ ∈ H A ⊗ H B |\varphi\rangle\in\mathcal{H}_A \otimes\mathcal{H}_B ∣ φ ⟩ ∈ H A ⊗ H B , we have V ∣ φ ⟩ ∈ Q V|\varphi\rangle\in\mathcal{Q} V ∣ φ ⟩ ∈ Q . Since U j ∈ S U_j \in S U j ∈ S , we have U j V ∣ φ ⟩ = V ∣ φ ⟩ U_jV|\varphi\rangle= V|\varphi\rangle U j V ∣ φ ⟩ = V ∣ φ ⟩ for every ∣ φ ⟩ |\varphi\rangle ∣ φ ⟩ , and hence U j V = V U_jV = V U j V = V . For every a ∈ W E a \in W_E a ∈ W E , it follows from (3) that
U j P a V = ξ j P s j P a V = ζ p − [ s j , a ] t P a ξ j P s j V = ζ p − [ s j , a ] t P a U j V = ζ p [ a , s j ] t P a V , (18) \begin{aligned}
U_jP_aV &= \xi_jP_{s_j}P_aV \\
&= \zeta_p^{-[s_j,a]_t}P_a\xi_jP_{s_j}V \\
&= \zeta_p^{-[s_j,a]_t}P_aU_jV \\
&= \zeta_p^{[a,s_j]_t}P_aV, \tag*{(18)}
\end{aligned} U j P a V = ξ j P s j P a V = ζ p − [ s j , a ] t P a ξ j P s j V = ζ p − [ s j , a ] t P a U j V = ζ p [ a , s j ] t P a V , (18)
where the last equality follows from U j V = V U_jV = V U j V = V and − [ s j , a ] t = [ a , s j ] t -[s_j,a]_t = [a,s_j]_t − [ s j , a ] t = [ a , s j ] t . Let ∣ ψ ⟩ = V ∣ φ ⟩ |\psi\rangle= V|\varphi\rangle ∣ ψ ⟩ = V ∣ φ ⟩ be any unit vector in Q \mathcal{Q} Q . After the error P a P_a P a , the state becomes P a ∣ ψ ⟩ P_a|\psi\rangle P a ∣ ψ ⟩ . By (18) ,
U j ( P a ∣ ψ ⟩ ) = ζ p [ a , s j ] t ( P a ∣ ψ ⟩ ) . U_j(P_a|\psi\rangle) = \zeta_p^{[a,s_j]_t}(P_a|\psi\rangle). U j ( P a ∣ ψ ⟩) = ζ p [ a , s j ] t ( P a ∣ ψ ⟩) .
Thus P a ∣ ψ ⟩ P_a|\psi\rangle P a ∣ ψ ⟩ is an eigenvector of U j U_j U j with eigenvalue ζ p [ a , s j ] t \zeta_p^{[a,s_j]_t} ζ p [ a , s j ] t . For fixed j j j , the measurement outcome depends only on the error label a a a and is the same for every encoded state.
The map χ \chi χ is F p \mathbb{F}_p F p -linear. Since s 1 , … , s ν s_1,\ldots,s_\nu s 1 , … , s ν span S Γ S_\Gamma S Γ , the hypothesis implies ker χ = W E ∩ S Γ ⊥ t = G E \ker\chi= W_E \cap S_{\Gamma}^{\perp t} = G_E ker χ = W E ∩ S Γ ⊥ t = G E . Consequently, for a , a ′ ∈ W E a,a' \in W_E a , a ′ ∈ W E ,
χ ( a ) = χ ( a ′ ) ⟺ a − a ′ ∈ G E . \chi(a) = \chi(a') \Longleftrightarrow a-a' \in G_E. χ ( a ) = χ ( a ′ ) ⟺ a − a ′ ∈ G E .
Thus the local stabilizer measurements distinguish errors with different syndromes, whereas errors with the same syndrome differ only by a gauge operator.
For each measured syndrome y y y , we next specify a correction operator and verify that it restores the logical subsystem for every Pauli error with that syndrome. For each y ∈ χ ( W E ) y \in\chi(W_E) y ∈ χ ( W E ) , choose a representative label a y ∈ W E a_y \in W_E a y ∈ W E with χ ( a y ) = y \chi(a_y) = y χ ( a y ) = y . For the measured syndrome y y y , the decoder applies P a y † P_{a_y}^{\dagger} P a y † , where P a y † = P a y − 1 P_{a_y}^{\dagger} = P_{a_y}^{-1} P a y † = P a y − 1 . For y ∈ F p ν ∖ χ ( W E ) y \in\mathbb{F}_p^\nu\setminus\chi(W_E) y ∈ F p ν ∖ χ ( W E ) , set a y = 0 a_y = 0 a y = 0 ; the corresponding correction operator is then P 0 † = I P_0^{\dagger} = I P 0 † = I . Now suppose that the actual error is P a P_a P a , where a ∈ W E a \in W_E a ∈ W E , and let y = χ ( a ) y = \chi(a) y = χ ( a ) . After the decoder applies the correction operator, the residual operator is
D a : = P a y † P a . D_a := P_{a_y}^{\dagger}P_a. D a := P a y † P a .
If a = a y a = a_y a = a y , then D a = I D_a = I D a = I . If a ≠ a y a \ne a_y a = a y , write a = ( x , z ) a = (x,z) a = ( x , z ) and a y = ( x y , z y ) a_y = (x_y,z_y) a y = ( x y , z y ) . By (3),
P a y P a − a y = ζ p Tr F q / F p ( z y ⋅ ( x − x y ) ) P a y + ( a − a y ) = ζ p Tr F q / F p ( z y ⋅ ( x − x y ) ) P a . \begin{aligned}
P_{a_y}P_{a-a_y} &= \zeta_p^{\operatorname{Tr}_{\mathbb{F}_q/\mathbb{F}_p}(z_y\cdot(x-x_y))}P_{a_y+(a-a_y)} \\
&= \zeta_p^{\operatorname{Tr}_{\mathbb{F}_q/\mathbb{F}_p}(z_y\cdot(x-x_y))}P_a.
\end{aligned} P a y P a − a y = ζ p Tr F q / F p ( z y ⋅ ( x − x y )) P a y + ( a − a y ) = ζ p Tr F q / F p ( z y ⋅ ( x − x y )) P a .
Multiplying both sides on the left by P a y † P_{a_y}^{\dagger} P a y † and using P a y † P a y = I P_{a_y}^{\dagger}P_{a_y} = I P a y † P a y = I , we obtain
D a = P a y † P a = ζ p − Tr F q / F p ( z y ⋅ ( x − x y ) ) P a − a y . D_a = P_{a_y}^{\dagger}P_a = \zeta_p^{-\operatorname{Tr}_{\mathbb{F}_q/\mathbb{F}_p}(z_y\cdot(x-x_y))}P_{a-a_y}. D a = P a y † P a = ζ p − Tr F q / F p ( z y ⋅ ( x − x y )) P a − a y .
Thus D a D_a D a has label a − a y a-a_y a − a y . By definition, y = χ ( a ) y = \chi(a) y = χ ( a ) , and the choice of a y a_y a y ensures that χ ( a y ) = y \chi(a_y) = y χ ( a y ) = y . By the linearity of χ \chi χ ,
χ ( a − a y ) = χ ( a ) − χ ( a y ) = y − y = 0. \chi(a-a_y) = \chi(a)-\chi(a_y) = y-y = 0. χ ( a − a y ) = χ ( a ) − χ ( a y ) = y − y = 0.
Hence a − a y ∈ ker χ = G E a-a_y \in\ker\chi= G_E a − a y ∈ ker χ = G E . Since a − a y a-a_y a − a y is the label of D a D_a D a , it follows that D a D_a D a is a gauge operator supported on E E E . Let U ∈ S U \in S U ∈ S have label s s s . Since s ∈ S ⊆ G Γ ⊥ t s \in S \subseteq G_{\Gamma}^{\perp t} s ∈ S ⊆ G Γ ⊥ t and a − a y ∈ G a-a_y \in G a − a y ∈ G , we have [ s , a − a y ] t = 0 [s,a-a_y]_t = 0 [ s , a − a y ] t = 0 . Hence U D a = D a U UD_a = D_aU U D a = D a U by (3). For every ∣ ψ ⟩ ∈ Q |\psi\rangle\in\mathcal{Q} ∣ ψ ⟩ ∈ Q , we have
U D a ∣ ψ ⟩ = D a U ∣ ψ ⟩ = D a ∣ ψ ⟩ , (19) UD_a|\psi\rangle= D_aU|\psi\rangle= D_a|\psi\rangle, \tag*{(19)} U D a ∣ ψ ⟩ = D a U ∣ ψ ⟩ = D a ∣ ψ ⟩ , (19)
which implies that D a ∣ ψ ⟩ ∈ Q D_a|\psi\rangle\in Q D a ∣ ψ ⟩ ∈ Q . For every ∣ ϕ ⟩ ∈ H A ⊗ H B |\phi\rangle\in\mathcal{H}_A \otimes\mathcal{H}_B ∣ ϕ ⟩ ∈ H A ⊗ H B , we have V ∣ ϕ ⟩ = ∣ ψ ⟩ ∈ Q V|\phi\rangle= |\psi\rangle\in Q V ∣ ϕ ⟩ = ∣ ψ ⟩ ∈ Q . Then D a V ∣ ϕ ⟩ ∈ Q D_aV|\phi\rangle\in Q D a V ∣ ϕ ⟩ ∈ Q . Hence P D a V = D a V PD_aV = D_aV P D a V = D a V , where P = V V † P = VV^\dagger P = V V † is the orthogonal projector onto Q Q Q . Since a − a y ∈ G E ⊆ G a-a_y \in G_E \subseteq G a − a y ∈ G E ⊆ G , it follows from (10) that
V † D a V = I A ⊗ U a , B V^\dagger D_aV = I_A \otimes U_{a,B} V † D a V = I A ⊗ U a , B
for some unitary U a , B U_{a,B} U a , B on H B \mathcal{H}_B H B . Therefore,
D a V = P D a V = V ( V † D a V ) = V ( I A ⊗ U a , B ) . (20) D_aV = PD_aV = V(V^\dagger D_aV) = V(I_A \otimes U_{a,B}). \tag*{(20)} D a V = P D a V = V ( V † D a V ) = V ( I A ⊗ U a , B ) . (20)
Hence the correction operator P a y † P^\dagger_{a_y} P a y † returns every encoded state affected by an error of syndrome y y y to Q Q Q and leaves the logical subsystem unchanged.
Lemma III.4 specifies a correction operator for each syndrome under the stated hypothesis. We next construct a projective measurement on Γ \Gamma Γ that determines the syndrome. This construction does not require the condition W E ∩ S Γ ⊥ = G E W_E \cap S_\Gamma^{\perp} = G_E W E ∩ S Γ ⊥ = G E .
Lemma III.5. For the SSC and coordinate sets fixed above, there is a projective measurement { Π y : y ∈ F p ν } \{\Pi_y : y \in\mathbb{F}_p^\nu\} { Π y : y ∈ F p ν } on H Γ \mathcal{H}_\Gamma H Γ such that
( Π y ⊗ I O ) P a V = δ y , χ ( a ) P a V , (\Pi_y \otimes I_O)P_aV = \delta_{y,\chi(a)}P_aV, ( Π y ⊗ I O ) P a V = δ y , χ ( a ) P a V , where a ∈ W E a \in W_E a ∈ W E and y ∈ F p ν y \in\mathbb{F}_p^\nu y ∈ F p ν . For every encoded state affected by P a P_a P a , this measurement returns the syndrome χ ( a ) \chi(a) χ ( a ) with probability 1.
Proof. Let U j = U ^ j ⊗ I O U_j = \widehat{U}_j \otimes I_O U j = U j ⊗ I O , where U ^ j \widehat{U}_j U j acts on H Γ \mathcal{H}_\Gamma H Γ . For any j , k ∈ { 1 , … , ν } j,k \in\{1,\ldots,\nu\} j , k ∈ { 1 , … , ν } , the stabilizer operators U j U_j U j and U k U_k U k commute. Since U j = U ^ j ⊗ I O U_j = \widehat{U}_j \otimes I_O U j = U j ⊗ I O and U k = U ^ k ⊗ I O U_k = \widehat{U}_k \otimes I_O U k = U k ⊗ I O , we have
( U ^ j U ^ k ) ⊗ I O = U j U k = U k U j = ( U ^ k U ^ j ) ⊗ I O . (\widehat{U}_j\widehat{U}_k) \otimes I_O = U_jU_k = U_kU_j = (\widehat{U}_k\widehat{U}_j) \otimes I_O. ( U j U k ) ⊗ I O = U j U k = U k U j = ( U k U j ) ⊗ I O .
Hence U ^ j U ^ k = U ^ k U ^ j \widehat{U}_j\widehat{U}_k = \widehat{U}_k\widehat{U}_j U j U k = U k U j . By (3), the label of U j p U_j^p U j p is p s j = 0 ps_j = 0 p s j = 0 . Thus U j p U_j^p U j p is a scalar operator in S S S . Since the identity is the only scalar operator in S S S , we obtain
U ^ j p ⊗ I O = ( U ^ j ⊗ I O ) p = U j p = I Γ ⊗ I O . \widehat{U}_j^p \otimes I_O = (\widehat{U}_j \otimes I_O)^p = U_j^p = I_\Gamma\otimes I_O. U j p ⊗ I O = ( U j ⊗ I O ) p = U j p = I Γ ⊗ I O .
Therefore U ^ j p = I Γ \widehat{U}_j^p = I_\Gamma U j p = I Γ , and every eigenvalue of U ^ j \widehat{U}_j U j has the form ζ p c \zeta_p^c ζ p c for some c ∈ F p c \in\mathbb{F}_p c ∈ F p .
For each j ∈ { 1 , … , ν } j \in\{1,\ldots,\nu\} j ∈ { 1 , … , ν } and y j ∈ F p y_j \in\mathbb{F}_p y j ∈ F p , we construct the orthogonal projector onto the eigenspace with eigenvalue ζ p y j \zeta_p^{y_j} ζ p y j . Define
Q j , y j = 1 p ∑ b = 0 p − 1 ( ζ p − y j U ^ j ) b = 1 p ∑ b = 0 p − 1 ζ p − b y j U ^ j b . Q_{j,y_j} = \frac{1}{p}\sum_{b=0}^{p-1}(\zeta_p^{-y_j}\widehat{U}_j)^b = \frac{1}{p}\sum_{b=0}^{p-1}\zeta_p^{-by_j}\widehat{U}_j^b. Q j , y j = p 1 b = 0 ∑ p − 1 ( ζ p − y j U j ) b = p 1 b = 0 ∑ p − 1 ζ p − b y j U j b .
Let ∣ v ⟩ |v\rangle ∣ v ⟩ be an eigenvector satisfying U ^ j ∣ v ⟩ = ζ p c ∣ v ⟩ \widehat{U}_j|v\rangle= \zeta_p^c|v\rangle U j ∣ v ⟩ = ζ p c ∣ v ⟩ , where c ∈ F p c \in\mathbb{F}_p c ∈ F p . Then
Q j , y j ∣ v ⟩ = 1 p ∑ b = 0 p − 1 ζ p − b y j U ^ j b ∣ v ⟩ = ( 1 p ∑ b = 0 p − 1 ζ p b ( c − y j ) ) ∣ v ⟩ . \begin{aligned}
Q_{j,y_j}|v\rangle= \frac{1}{p}\sum_{b=0}^{p-1}\zeta_p^{-by_j}\widehat{U}_j^b|v\rangle\\
&= \left(\frac{1}{p}\sum_{b=0}^{p-1}\zeta_p^{b(c-y_j)}\right)|v\rangle.
\end{aligned} Q j , y j ∣ v ⟩ = p 1 b = 0 ∑ p − 1 ζ p − b y j U j b ∣ v ⟩ = ( p 1 b = 0 ∑ p − 1 ζ p b ( c − y j ) ) ∣ v ⟩ .
If c = y j c=y_j c = y j , the coefficient in parentheses equals 1. If c ≠ y j c\ne y_j c = y j , then ζ p c − y j ≠ 1 \zeta_p^{c-y_j}\ne1 ζ p c − y j = 1 , and
1 p ∑ b = 0 p − 1 ζ p b ( c − y j ) = 1 − ζ p p ( c − y j ) p ( 1 − ζ p c − y j ) = 0. \frac{1}{p}\sum_{b=0}^{p-1}\zeta_p^{b(c-y_j)} = \frac{1-\zeta_p^{p(c-y_j)}}{p(1-\zeta_p^{c-y_j})}=0. p 1 b = 0 ∑ p − 1 ζ p b ( c − y j ) = p ( 1 − ζ p c − y j ) 1 − ζ p p ( c − y j ) = 0.
Consequently,
Q j , y j ∣ v ⟩ = { ∣ v ⟩ , c = y j , 0 , c ≠ y j . Q_{j,y_j}|v\rangle=
\begin{cases}
|v\rangle, & c=y_j,\\
0, & c\ne y_j.
\end{cases} Q j , y j ∣ v ⟩ = { ∣ v ⟩ , 0 , c = y j , c = y j .
Since U ^ j \widehat{U}_j U j is unitary, the spectral theorem [38 Theorem 1.3] implies that its eigenspaces form an orthogonal decomposition of H Γ \mathcal{H}_\Gamma H Γ . Hence Q j , y j Q_{j,y_j} Q j , y j is the required orthogonal projector. For each fixed j j j , we therefore have
Q j , c † = Q j , c , Q j , c Q j , c ′ = δ c , c ′ Q j , c , ∑ c ∈ F p Q j , c = I Γ . Q_{j,c}^{\dagger}=Q_{j,c},\quad Q_{j,c}Q_{j,c'}=\delta_{c,c'}Q_{j,c},\quad\sum_{c\in\mathbb{F}_p}Q_{j,c}=I_\Gamma. Q j , c † = Q j , c , Q j , c Q j , c ′ = δ c , c ′ Q j , c , c ∈ F p ∑ Q j , c = I Γ .
Each Q j , c Q_{j,c} Q j , c is a polynomial in U ^ j \widehat{U}_j U j . Since U ^ 1 , … , U ^ ν \widehat{U}_1,\ldots,\widehat{U}_\nu U 1 , … , U ν commute, we have Q j , c Q k , c ′ = Q k , c ′ Q j , c Q_{j,c}Q_{k,c'}=Q_{k,c'}Q_{j,c} Q j , c Q k , c ′ = Q k , c ′ Q j , c for all j , k ∈ { 1 , … , ν } j,k\in\{1,\ldots,\nu\} j , k ∈ { 1 , … , ν } and c , c ′ ∈ F p c,c'\in\mathbb{F}_p c , c ′ ∈ F p . For y = ( y 1 , … , y ν ) ∈ F p ν \mathbf{y}=(y_1,\ldots,y_\nu)\in\mathbb{F}_p^\nu y = ( y 1 , … , y ν ) ∈ F p ν , define
Π y = ∏ j = 1 ν Q j , y j = ∏ j = 1 ν ( 1 p ∑ b = 0 p − 1 ζ p − b y j U ^ j b ) . \Pi_{\mathbf{y}}=\prod_{j=1}^{\nu}Q_{j,y_j}=\prod_{j=1}^{\nu}\left(\frac{1}{p}\sum_{b=0}^{p-1}\zeta_p^{-by_j}\widehat{U}_j^b\right). Π y = j = 1 ∏ ν Q j , y j = j = 1 ∏ ν ( p 1 b = 0 ∑ p − 1 ζ p − b y j U j b ) .
Since the factors commute and are self-adjoint, we have Π y † = Π y \Pi_{\mathbf{y}}^\dagger=\Pi_{\mathbf{y}} Π y † = Π y . Moreover,
Π y Π y ′ = ∏ j = 1 ν ( Q j , y j Q j , y j ′ ) = ( ∏ j = 1 ν δ y j , y j ′ ) ∏ j = 1 ν Q j , y j = δ y , y ′ Π y . \Pi_{\mathbf{y}}\Pi_{\mathbf{y}'}=\prod_{j=1}^{\nu}\left(Q_{j,y_j}Q_{j,y_j'}\right)
=\left(\prod_{j=1}^{\nu}\delta_{y_j,y_j'}\right)\prod_{j=1}^{\nu}Q_{j,y_j}
=\delta_{\mathbf{y},\mathbf{y}'}\Pi_{\mathbf{y}}. Π y Π y ′ = j = 1 ∏ ν ( Q j , y j Q j , y j ′ ) = ( j = 1 ∏ ν δ y j , y j ′ ) j = 1 ∏ ν Q j , y j = δ y , y ′ Π y .
and
∑ y ∈ F p ν Π y = ∑ y 1 ∈ F p ⋯ ∑ y ν ∈ F p Q 1 , y 1 ⋯ Q ν , y ν = ∏ j = 1 ν ( ∑ y j ∈ F p Q j , y j ) = I Γ . \sum_{\mathbf{y}\in\mathbb{F}_p^\nu}\Pi_{\mathbf{y}}
=\sum_{y_1\in\mathbb{F}_p}\cdots\sum_{y_\nu\in\mathbb{F}_p}Q_{1,y_1}\cdots Q_{\nu,y_\nu}
=\prod_{j=1}^{\nu}\left(\sum_{y_j\in\mathbb{F}_p}Q_{j,y_j}\right)
=I_\Gamma. y ∈ F p ν ∑ Π y = y 1 ∈ F p ∑ ⋯ y ν ∈ F p ∑ Q 1 , y 1 ⋯ Q ν , y ν = j = 1 ∏ ν y j ∈ F p ∑ Q j , y j = I Γ .
Thus { Π y : y ∈ F p ν } \{\Pi_{\mathbf{y}}:\mathbf{y}\in\mathbb{F}_p^\nu\} { Π y : y ∈ F p ν } defines a projective measurement on Γ \Gamma Γ . Since the factors are commuting orthogonal projectors, we also have
im Π y = ⋂ j = 1 ν im Q j , y j = { ∣ v ⟩ ∈ H Γ : U ^ j ∣ v ⟩ = ζ p y j ∣ v ⟩ for all j = 1 , … , ν } . \operatorname{im}\Pi_{\mathbf{y}}=\bigcap_{j=1}^{\nu}\operatorname{im}Q_{j,y_j}
=\left\{|v\rangle\in\mathcal{H}_\Gamma:\widehat{U}_j|v\rangle=\zeta_p^{y_j}|v\rangle\text{ for all }j=1,\ldots,\nu\right\}. im Π y = j = 1 ⋂ ν im Q j , y j = { ∣ v ⟩ ∈ H Γ : U j ∣ v ⟩ = ζ p y j ∣ v ⟩ for all j = 1 , … , ν } .
We label the joint measurement outcome ( ζ p y 1 , … , ζ p y ν ) (\zeta_p^{y_1},\ldots,\zeta_p^{y_\nu}) ( ζ p y 1 , … , ζ p y ν ) by y \mathbf{y} y . For ν = 0 \nu=0 ν = 0 , we use the empty-product convention Π 0 = I Γ \Pi_{\mathbf{0}}=I_\Gamma Π 0 = I Γ , where 0 \mathbf{0} 0 is the unique element of F p 0 \mathbb{F}_p^0 F p 0 .
It remains to verify that this measurement returns χ ( a ) \chi(a) χ ( a ) for every a ∈ W E a\in W_E a ∈ W E . Since U j V = V U_jV=V U j V = V , it follows from (3) that
( U ^ j ⊗ I O ) P a V = ζ p [ a , s j ] t P a V . (\widehat{U}_j\otimes I_O)P_aV=\zeta_p^{[a,s_j]_t}P_aV. ( U j ⊗ I O ) P a V = ζ p [ a , s j ] t P a V .
By induction on b b b with 0 ≤ b < p 0\le b<p 0 ≤ b < p , we obtain
( U ^ j b ⊗ I O ) P a V = ζ p b [ a , s j ] t P a V . (\widehat{U}_j^b\otimes I_O)P_aV=\zeta_p^{b[a,s_j]_t}P_aV. ( U j b ⊗ I O ) P a V = ζ p b [ a , s j ] t P a V .
By the definition of Q j , y j Q_{j,y_j} Q j , y j , we have
( Q j , y j ⊗ I O ) P a V = 1 p ∑ b = 0 p − 1 ζ p − b y j ( U ^ j b ⊗ I O ) P a V = ( 1 p ∑ b = 0 p − 1 ζ p b ( [ a , s j ] t − y j ) ) P a V = δ y j , [ a , s j ] t P a V , (Q_{j,y_j}\otimes I_O)P_aV
=\frac{1}{p}\sum_{b=0}^{p-1}\zeta_p^{-by_j}(\widehat{U}_j^b\otimes I_O)P_aV
=\left(\frac{1}{p}\sum_{b=0}^{p-1}\zeta_p^{b([a,s_j]_t-y_j)}\right)P_aV
=\delta_{y_j,[a,s_j]_t}P_aV, ( Q j , y j ⊗ I O ) P a V = p 1 b = 0 ∑ p − 1 ζ p − b y j ( U j b ⊗ I O ) P a V = ( p 1 b = 0 ∑ p − 1 ζ p b ([ a , s j ] t − y j ) ) P a V = δ y j , [ a , s j ] t P a V ,
where the last equality follows from the geometric sum computed above. By the definitions of Π y \Pi_{\mathbf{y}} Π y and χ \chi χ , we obtain
( Π y ⊗ I O ) P a V = ( ∏ j = 1 ν δ y j , [ a , s j ] t ) P a V = δ y , χ ( a ) P a V . (21) (\Pi_{\mathbf{y}}\otimes I_O)P_aV
=\left(\prod_{j=1}^{\nu}\delta_{y_j,[a,s_j]_t}\right)P_aV
=\delta_{\mathbf{y},\chi(a)}P_aV. \tag*{(21)} ( Π y ⊗ I O ) P a V = ( j = 1 ∏ ν δ y j , [ a , s j ] t ) P a V = δ y , χ ( a ) P a V . (21)
For any unit vector ∣ φ ⟩ ∈ H A ⊗ H B |\varphi\rangle\in\mathcal{H}_A\otimes\mathcal{H}_B ∣ φ ⟩ ∈ H A ⊗ H B , the probability of outcome y \mathbf{y} y on the state P a V ∣ φ ⟩ P_aV|\varphi\rangle P a V ∣ φ ⟩ is
Pr ( y ) = ∥ ( Π y ⊗ I O ) P a V ∣ φ ⟩ ∥ 2 = δ y , χ ( a ) ∥ P a V ∣ φ ⟩ ∥ 2 = δ y , χ ( a ) , \Pr(\mathbf{y})=\left\|(\Pi_{\mathbf{y}}\otimes I_O)P_aV|\varphi\rangle\right\|^2
=\delta_{\mathbf{y},\chi(a)}\left\|P_aV|\varphi\rangle\right\|^2
=\delta_{\mathbf{y},\chi(a)}, Pr ( y ) = ∥ ( Π y ⊗ I O ) P a V ∣ φ ⟩ ∥ 2 = δ y , χ ( a ) ∥ P a V ∣ φ ⟩ ∥ 2 = δ y , χ ( a ) ,
since V V V is an isometry and P a P_a P a is unitary. By linearity in the density operator, the same probability formula holds for mixed encoded states. Thus, for any encoded state after the error P a P_a P a , the projective measurement { Π y : y ∈ F p ν } \{\Pi_y : y \in\mathbb{F}_p^\nu\} { Π y : y ∈ F p ν } on Γ \Gamma Γ returns the syndrome y = χ ( a ) y = \chi(a) y = χ ( a ) with probability 1.
By Lemmas III.4 and III.5, the measurement outcome χ ( a ) \chi(a) χ ( a ) determines the correction operator P a χ ( a ) † P_{a_{\chi(a)}}^\dagger P a χ ( a ) † whenever W E ∩ S Γ ⊥ t = G E W_E \cap S_\Gamma^{\perp_t} = G_E W E ∩ S Γ ⊥ t = G E . We now use this measurement and the corresponding correction operators to construct a recovery channel for the erasure of E E E .
Proposition III.6. Let Q Q Q be an SSC with gauge label space G G G and stabilizer label space S S S . Let ∅ ≠ E ⊆ Γ ⊆ [ n ] \varnothing\ne E \subseteq\Gamma\subseteq[n] ∅ = E ⊆ Γ ⊆ [ n ] . If W E ∩ S Γ ⊥ t = G E W_E \cap S_\Gamma^{\perp_t} = G_E W E ∩ S Γ ⊥ t = G E , then the erasure of E E E is recoverable within Γ \Gamma Γ .
Proof. Let H = Γ ∖ E H = \Gamma\setminus E H = Γ ∖ E and O = Γ c O = \Gamma^c O = Γ c . Let V : H A ⊗ H B → H [ n ] V : \mathcal{H}_A \otimes\mathcal{H}_B \to\mathcal{H}_{[n]} V : H A ⊗ H B → H [ n ] be an encoding isometry for Q Q Q , with associated encoding channel V \mathcal{V} V . We construct a CPTP recovery map R Γ , E : B ( H H ) → B ( H Γ ) \mathcal{R}_{\Gamma,E} : \mathcal{B}(\mathcal{H}_H) \to\mathcal{B}(\mathcal{H}_\Gamma) R Γ , E : B ( H H ) → B ( H Γ ) and a CPTP gauge map Φ Γ , E : B ( H B ) → B ( H B ) \Phi_{\Gamma,E} : \mathcal{B}(\mathcal{H}_B) \to\mathcal{B}(\mathcal{H}_B) Φ Γ , E : B ( H B ) → B ( H B ) such that (16) holds.
Choose the basis of S Γ S_\Gamma S Γ and the syndrome map χ \chi χ as above. By Lemma III.4, choose the representative labels a y a_y a y for y ∈ χ ( W E ) y \in\chi(W_E) y ∈ χ ( W E ) and the unitaries U a , B U_{a,B} U a , B satisfying (19). Set a y = 0 a_y = 0 a y = 0 for y ∈ F p ν ∖ χ ( W E ) y \in\mathbb{F}_p^\nu\setminus\chi(W_E) y ∈ F p ν ∖ χ ( W E ) , and let { Π y : y ∈ F p ν } \{\Pi_y : y \in\mathbb{F}_p^\nu\} { Π y : y ∈ F p ν } be the measurement from Lemma III.5.
To perform the measurement on Γ \Gamma Γ , the decoder first prepares replacement qudits at E E E in the maximally mixed state. Let d E = q ∣ E ∣ d_E = q^{|E|} d E = q ∣ E ∣ and define the replacement map
J E : B ( H H ) ⟶ B ( H Γ ) , J E ( τ ) = I E d E ⊗ τ . \mathcal{J}_E : \mathcal{B}(\mathcal{H}_H) \longrightarrow\mathcal{B}(\mathcal{H}_\Gamma), \qquad\mathcal{J}_E(\tau) = \frac{I_E}{d_E} \otimes\tau. J E : B ( H H ) ⟶ B ( H Γ ) , J E ( τ ) = d E I E ⊗ τ .
For every operator ρ ∈ B ( H [ n ] ) \rho\in\mathcal{B}(\mathcal{H}_{[n]}) ρ ∈ B ( H [ n ] ) , we have
( J E ⊗ id O ) ( Tr E ρ ) = I E d E ⊗ Tr E ρ = 1 d E 2 ∑ a ∈ W E P a ρ P a † . (22) (\mathcal{J}_E \otimes\operatorname{id}_O)(\operatorname{Tr}_E \rho) = \frac{I_E}{d_E} \otimes\operatorname{Tr}_E \rho= \frac{1}{d_E^2} \sum_{a \in W_E} P_a \rho P_a^\dagger. \tag*{(22)} ( J E ⊗ id O ) ( Tr E ρ ) = d E I E ⊗ Tr E ρ = d E 2 1 a ∈ W E ∑ P a ρ P a † . (22)
To prove the second equality, expand ρ \rho ρ in the Pauli basis on E E E as
ρ = ∑ b ∈ W E P b , E ⊗ M b , \rho= \sum_{b \in W_E} P_{b,E} \otimes M_b, ρ = b ∈ W E ∑ P b , E ⊗ M b ,
where P b = P b , E ⊗ I E c P_b = P_{b,E} \otimes I_{E^c} P b = P b , E ⊗ I E c and M b ∈ B ( H E c ) M_b \in\mathcal{B}(\mathcal{H}_{E^c}) M b ∈ B ( H E c ) . By (4) ,
Tr E ρ = ∑ b ∈ W E Tr ( P b , E ) M b = d E M 0 . \operatorname{Tr}_E \rho= \sum_{b \in W_E} \operatorname{Tr}(P_{b,E})M_b = d_E M_0. Tr E ρ = b ∈ W E ∑ Tr ( P b , E ) M b = d E M 0 .
By (3) , conjugation by P a P_a P a multiplies the term indexed by b b b by ζ p − [ a , b ] t \zeta_p^{-[a,b]_t} ζ p − [ a , b ] t . For b ≠ 0 b \ne0 b = 0 , the nondegeneracy of the trace-symplectic form on W E W_E W E implies that a ↦ [ a , b ] t a \mapsto[a,b]_t a ↦ [ a , b ] t is a nonzero F p \mathbb{F}_p F p -linear functional. Each value in F p \mathbb{F}_p F p therefore occurs d E 2 / p d_E^2/p d E 2 / p times. By the geometric sum formula, we obtain
1 d E 2 ∑ a ∈ W E ζ p − [ a , b ] t = δ b , 0 , \frac{1}{d_E^2}\sum_{a \in W_E} \zeta_p^{-[a,b]_t} = \delta_{b,0}, d E 2 1 a ∈ W E ∑ ζ p − [ a , b ] t = δ b , 0 ,
where all summands equal 1 when b = 0 b = 0 b = 0 . Consequently,
1 d E 2 ∑ a ∈ W E P a ρ P a † = I E ⊗ M 0 = I E d E ⊗ Tr E ρ , \frac{1}{d_E^2}\sum_{a \in W_E} P_a \rho P_a^\dagger= I_E \otimes M_0 = \frac{I_E}{d_E} \otimes\operatorname{Tr}_E \rho, d E 2 1 a ∈ W E ∑ P a ρ P a † = I E ⊗ M 0 = d E I E ⊗ Tr E ρ ,
which proves (22) .
Since a y ∈ W E a_y \in W_E a y ∈ W E and E ⊆ Γ E \subseteq\Gamma E ⊆ Γ , write P a y = C y ⊗ I O P_{a_y} = C_y \otimes I_O P a y = C y ⊗ I O , where C y C_y C y is unitary on H Γ \mathcal{H}_\Gamma H Γ . After the replacement, the decoder performs the projective measurement { Π y : y ∈ F p ν } \{\Pi_y : y \in\mathbb{F}_p^\nu\} { Π y : y ∈ F p ν } and applies C y † C_y^\dagger C y † upon obtaining outcome y y y . The resulting recovery map is
R Γ , E ( τ ) = ∑ y ∈ F p ν C y † Π y J E ( τ ) Π y C y . (23) \mathcal{R}_{\Gamma,E}(\tau) = \sum_{y \in\mathbb{F}_p^\nu} C_y^\dagger\Pi_y \mathcal{J}_E(\tau) \Pi_y C_y. \tag*{(23)} R Γ , E ( τ ) = y ∈ F p ν ∑ C y † Π y J E ( τ ) Π y C y . (23)
The map J E \mathcal{J}_E J E is CPTP because it appends the fixed state I E / d E I_E/d_E I E / d E . The subsequent measurement and correction define a channel with Kraus operators K y = C y † Π y K_y = C_y^\dagger\Pi_y K y = C y † Π y , since
∑ y ∈ F p ν K y † K y = ∑ y ∈ F p ν Π y C y C y † Π y = ∑ y ∈ F p ν Π y = I Γ . \sum_{y \in\mathbb{F}_p^\nu} K_y^\dagger K_y
= \sum_{y \in\mathbb{F}_p^\nu} \Pi_y C_y C_y^\dagger\Pi_y
= \sum_{y \in\mathbb{F}_p^\nu} \Pi_y
= I_\Gamma. y ∈ F p ν ∑ K y † K y = y ∈ F p ν ∑ Π y C y C y † Π y = y ∈ F p ν ∑ Π y = I Γ .
Hence R Γ , E \mathcal{R}_{\Gamma,E} R Γ , E is CPTP, with input on H H H and output on Γ \Gamma Γ . It remains to verify (16). Let M ∈ B ( H A ⊗ H B ) M \in\mathcal{B}(\mathcal{H}_{A} \otimes\mathcal{H}_{B}) M ∈ B ( H A ⊗ H B ) be arbitrary. By (21) and linearity, we may apply the measurement and correction to each Pauli term separately. For the term indexed by a a a , it follows from (20) that only the outcome y = χ ( a ) y=\chi(a) y = χ ( a ) contributes. Thus
( R Γ , E ⊗ id O ) ( Tr E V ( M ) ) = 1 d E 2 ∑ a ∈ W E D a V M V † D a † = 1 d E 2 ∑ a ∈ W E V ( I A ⊗ U a , B ) M ( I A ⊗ U a , B † ) V † . \begin{aligned}
(\mathcal{R}_{\Gamma,E} \otimes\operatorname{id}_{O})(\operatorname{Tr}_{E}\mathcal{V}(M))
= \frac{1}{d_{E}^{2}}\sum_{a\in W_{E}}D_{a}\mathcal{V}M\mathcal{V}^{\dagger}D_{a}^{\dagger} \\
&= \frac{1}{d_{E}^{2}}\sum_{a\in W_{E}}\mathcal{V}(I_{A}\otimes U_{a,B})M(I_{A}\otimes U_{a,B}^{\dagger})\mathcal{V}^{\dagger}.
\end{aligned} ( R Γ , E ⊗ id O ) ( Tr E V ( M )) = d E 2 1 a ∈ W E ∑ D a V M V † D a † = d E 2 1 a ∈ W E ∑ V ( I A ⊗ U a , B ) M ( I A ⊗ U a , B † ) V † .
where the last equality follows from (19). Define
Φ Γ , E ( σ ) = 1 d E 2 ∑ a ∈ W E U a , B σ U a , B † , σ ∈ B ( H B ) . \Phi_{\Gamma,E}(\sigma)=\frac{1}{d_{E}^{2}}\sum_{a\in W_{E}}U_{a,B}\sigma U_{a,B}^{\dagger},\quad\sigma\in\mathcal{B}(\mathcal{H}_{B}). Φ Γ , E ( σ ) = d E 2 1 a ∈ W E ∑ U a , B σ U a , B † , σ ∈ B ( H B ) .
Since ∣ W E ∣ = d E 2 |W_{E}|=d_{E}^{2} ∣ W E ∣ = d E 2 , the map Φ Γ , E \Phi_{\Gamma,E} Φ Γ , E is a uniform average of unitary channels and is therefore CPTP. With this definition, we can write the preceding identity as
( R Γ , E ⊗ id O ) ( Tr E V ( M ) ) = V ( ( id A ⊗ Φ Γ , E ) ( M ) ) . (\mathcal{R}_{\Gamma,E}\otimes\operatorname{id}_{O})(\operatorname{Tr}_{E}\mathcal{V}(M))=\mathcal{V}((\operatorname{id}_{A}\otimes\Phi_{\Gamma,E})(M)). ( R Γ , E ⊗ id O ) ( Tr E V ( M )) = V (( id A ⊗ Φ Γ , E ) ( M )) .
Since this equality holds for every M ∈ B ( H A ⊗ H B ) M\in\mathcal{B}(\mathcal{H}_{A}\otimes\mathcal{H}_{B}) M ∈ B ( H A ⊗ H B ) , we obtain
( R Γ , E ⊗ id O ) ∘ Tr E ∘ V = V ∘ ( id A ⊗ Φ Γ , E ) . (\mathcal{R}_{\Gamma,E}\otimes\operatorname{id}_{O})\circ\operatorname{Tr}_{E}\circ\mathcal{V}
=\mathcal{V}\circ(\operatorname{id}_{A}\otimes\Phi_{\Gamma,E}). ( R Γ , E ⊗ id O ) ∘ Tr E ∘ V = V ∘ ( id A ⊗ Φ Γ , E ) .
By Definition III.1, the erasure of E E E is recoverable within Γ \Gamma Γ .
We next prove a necessary condition for local erasure recovery in terms of the stabilizer label space.
Lemma III.7. Let Q Q Q be an SSC with stabilizer label space S S S , and let ∅ ≠ E ⊆ Γ ⊆ [ n ] \varnothing\ne E\subseteq\Gamma\subseteq[n] ∅ = E ⊆ Γ ⊆ [ n ] . If the erasure of E E E is recoverable within Γ \Gamma Γ in the sense of Definition III.1, then π E ( S Γ ) = π E ( S ) \pi_{E}(S_{\Gamma})=\pi_{E}(S) π E ( S Γ ) = π E ( S ) .
Proof. Set H = Γ ∖ E H=\Gamma\setminus E H = Γ ∖ E and O = Γ c O=\Gamma^{c} O = Γ c . Since S Γ ⊆ S S_{\Gamma}\subseteq S S Γ ⊆ S , we have π E ( S Γ ) ⊆ π E ( S ) \pi_{E}(S_{\Gamma})\subseteq\pi_{E}(S) π E ( S Γ ) ⊆ π E ( S ) . It remains to prove the reverse inclusion. We first show that π O ( S ) = π O ( S E c ) \pi_{O}(S)=\pi_{O}(S_{E^{c}}) π O ( S ) = π O ( S E c ) . Let S S S be the stabilizer group of Q Q Q , and let P P P be the orthogonal projector onto Q Q Q . We consider the maximally mixed state on Q Q Q , whose Pauli expansion contains only stabilizer terms. By (9), Tr P = K R \operatorname{Tr}P=KR Tr P = K R , so ω = P K R \omega=\frac{P}{KR} ω = K R P is a normalized state. Since V ( M ) = V M V † \mathcal{V}(M)=\mathcal{V}M\mathcal{V}^{\dagger} V ( M ) = V M V † and V V † = P \mathcal{V}\mathcal{V}^{\dagger}=P V V † = P , we also have
V ( I A K ⊗ I B R ) = V ( I A ⊗ I B ) V † K R = P K R = ω . \mathcal{V}\left(\frac{I_{A}}{K}\otimes\frac{I_{B}}{R}\right)
=\frac{\mathcal{V}(I_{A}\otimes I_{B})\mathcal{V}^{\dagger}}{KR}
=\frac{P}{KR}=\omega. V ( K I A ⊗ R I B ) = K R V ( I A ⊗ I B ) V † = K R P = ω . For each s ∈ S s\in S s ∈ S , let U s ∈ S U_{s}\in S U s ∈ S be its stabilizer representative. By (9), K R ∣ S ∣ = q n KR|S|=q^{n} K R ∣ S ∣ = q n , and hence
ω = 1 K R ∣ S ∣ ∑ s ∈ S U s = q − n ∑ s ∈ S U s . \omega=\frac{1}{KR|S|}\sum_{s\in S}U_{s}=q^{-n}\sum_{s\in S}U_{s}. ω = K R ∣ S ∣ 1 s ∈ S ∑ U s = q − n s ∈ S ∑ U s . By Definition III.1, there exist CPTP maps
R Γ , E : B ( H H ) ⟶ B ( H Γ ) , Φ Γ , E : B ( H B ) ⟶ B ( H B ) \mathcal{R}_{\Gamma,E}:\mathcal{B}(\mathcal{H}_{H})\longrightarrow\mathcal{B}(\mathcal{H}_{\Gamma}),\quad
\Phi_{\Gamma,E}:\mathcal{B}(\mathcal{H}_{B})\longrightarrow\mathcal{B}(\mathcal{H}_{B}) R Γ , E : B ( H H ) ⟶ B ( H Γ ) , Φ Γ , E : B ( H B ) ⟶ B ( H B ) satisfying (16). Applying this identity to I A / K ⊗ I B / R I_{A}/K\otimes I_{B}/R I A / K ⊗ I B / R , we obtain
η : = ( R Γ , E ⊗ id O ) ( Tr E ω ) = V ( I A K ⊗ Φ Γ , E ( I B R ) ) . \eta:=(\mathcal{R}_{\Gamma,E}\otimes\operatorname{id}_{O})(\operatorname{Tr}_{E}\omega)
=\mathcal{V}\left(\frac{I_{A}}{K}\otimes\Phi_{\Gamma,E}\left(\frac{I_{B}}{R}\right)\right). η := ( R Γ , E ⊗ id O ) ( Tr E ω ) = V ( K I A ⊗ Φ Γ , E ( R I B ) ) . Since the channels are trace-preserving and ω \omega ω is normalized, Tr ( η ) = 1 \operatorname{Tr}(\eta)=1 Tr ( η ) = 1 . The last expression represents an encoded state, and hence P η P = η P\eta P=\eta P η P = η . For every s ∈ S s\in S s ∈ S , the operator U s † U_{s}^{\dagger} U s † acts as the identity on Q Q Q , and therefore U s † P = P U_{s}^{\dagger}P=P U s † P = P . It follows that
Tr ( U s † η ) = Tr ( U s † P η P ) = Tr ( P η P ) = Tr ( η ) = 1. (24) \operatorname{Tr}(U_{s}^{\dagger}\eta)=\operatorname{Tr}(U_{s}^{\dagger}P\eta P)
=\operatorname{Tr}(P\eta P)
=\operatorname{Tr}(\eta)=1. \tag*{(24)} Tr ( U s † η ) = Tr ( U s † P η P ) = Tr ( P η P ) = Tr ( η ) = 1. (24) We next compute the Pauli expansion of η \eta η . Let
U s = ξ s P π Γ ( s ) ⊗ P π O ( s ) , ∣ ξ s ∣ = 1 , U_{s}=\xi_{s}P_{\pi_{\Gamma}(s)}\otimes P_{\pi_{O}(s)},\quad|\xi_{s}|=1, U s = ξ s P π Γ ( s ) ⊗ P π O ( s ) , ∣ ξ s ∣ = 1 ,
where each projected Pauli operator acts on the indicated coordinate set. After grouping the tensor factors according to E E E , H H H , and O O O , we have
Tr E U s = ξ s Tr ( P π E ( s ) ) P π H ( s ) ⊗ P π O ( s ) . \operatorname{Tr}_{E} U_s = \xi_s \operatorname{Tr}\left(P_{\pi_E(s)}\right) P_{\pi_H(s)} \otimes P_{\pi_O(s)}. Tr E U s = ξ s Tr ( P π E ( s ) ) P π H ( s ) ⊗ P π O ( s ) .
By (4) ,
Tr ( P π E ( s ) ) = { q ∣ E ∣ , π E ( s ) = 0 , 0 , π E ( s ) ≠ 0. \operatorname{Tr}\left(P_{\pi_E(s)}\right) =
\begin{cases}
q^{|E|}, & \pi_E(s)=0,\\
0, & \pi_E(s)\ne0.
\end{cases} Tr ( P π E ( s ) ) = { q ∣ E ∣ , 0 , π E ( s ) = 0 , π E ( s ) = 0.
Thus a term survives the partial trace over E E E precisely when its label belongs to S E c S_{E^{c}} S E c . Consequently,
Tr E ω = q − n ∑ s ∈ S Tr E U s = q − ( n − ∣ E ∣ ) ∑ u ∈ S E c ξ u P π H ( u ) ⊗ P π O ( u ) . \operatorname{Tr}_{E}\omega
= q^{-n}\sum_{s\in S}\operatorname{Tr}_{E}U_s
= q^{-(n-|E|)}\sum_{u\in S_{E^{c}}}\xi_u P_{\pi_H(u)}\otimes P_{\pi_O(u)}. Tr E ω = q − n s ∈ S ∑ Tr E U s = q − ( n − ∣ E ∣ ) u ∈ S E c ∑ ξ u P π H ( u ) ⊗ P π O ( u ) .
The recovery channel acts on H H H and leaves O O O unchanged. By linearity,
η = q − ( n − ∣ E ∣ ) ∑ u ∈ S E c ξ u R Γ , E ( P π H ( u ) ) ⊗ P π O ( u ) . \eta=q^{-(n-|E|)}\sum_{u\in S_{E^{c}}}\xi_u\mathcal{R}_{\Gamma,E}\left(P_{\pi_H(u)}\right)\otimes P_{\pi_O(u)}. η = q − ( n − ∣ E ∣ ) u ∈ S E c ∑ ξ u R Γ , E ( P π H ( u ) ) ⊗ P π O ( u ) .
In this expansion, every Pauli factor on O O O has its label in π O ( S E c ) \pi_O(S_{E^{c}}) π O ( S E c ) . Suppose that there exists s ∈ S s\in S s ∈ S such that π O ( s ) ∉ π O ( S E c ) \pi_O(s)\notin\pi_O(S_{E^{c}}) π O ( s ) ∈ / π O ( S E c ) . Then π O ( s ) ≠ π O ( u ) \pi_O(s)\ne\pi_O(u) π O ( s ) = π O ( u ) for every u ∈ S E c u\in S_{E^{c}} u ∈ S E c . By the Pauli orthogonality relation in (4) ,
Tr ( P π O ( s ) † P π O ( u ) ) = 0. \operatorname{Tr}\left(P_{\pi_O(s)}^{\dagger}P_{\pi_O(u)}\right)=0. Tr ( P π O ( s ) † P π O ( u ) ) = 0.
Using the tensor-product formula for the trace, we obtain
Tr ( U s † η ) = q − ( n − ∣ E ∣ ) ∑ u ∈ S E c ξ s ‾ ξ u × Tr ( P π Γ ( s ) † R Γ , E ( P π H ( u ) ) ) × Tr ( P π O ( s ) † P π O ( u ) ) = 0. \operatorname{Tr}\left(U_s^{\dagger}\eta\right)
=q^{-(n-|E|)}\sum_{u\in S_{E^{c}}}\overline{\xi_s}\xi_u
\times\operatorname{Tr}\left(P_{\pi_{\Gamma}(s)}^{\dagger}\mathcal{R}_{\Gamma,E}\left(P_{\pi_H(u)}\right)\right)
\times\operatorname{Tr}\left(P_{\pi_O(s)}^{\dagger}P_{\pi_O(u)}\right)
=0. Tr ( U s † η ) = q − ( n − ∣ E ∣ ) u ∈ S E c ∑ ξ s ξ u × Tr ( P π Γ ( s ) † R Γ , E ( P π H ( u ) ) ) × Tr ( P π O ( s ) † P π O ( u ) ) = 0.
This contradicts (24) . Hence π O ( S ) ⊆ π O ( S E c ) \pi_O(S)\subseteq\pi_O(S_{E^{c}}) π O ( S ) ⊆ π O ( S E c ) . The reverse inclusion follows from S E c ⊆ S S_{E^{c}}\subseteq S S E c ⊆ S . Therefore,
π O ( S ) = π O ( S E c ) . (25) \pi_O(S)=\pi_O(S_{E^{c}}). \tag*{(25)} π O ( S ) = π O ( S E c ) . (25)
Now let s ∈ S s\in S s ∈ S . By (25) , we may choose u ∈ S E c u\in S_{E^{c}} u ∈ S E c such that π O ( u ) = π O ( s ) \pi_O(u)=\pi_O(s) π O ( u ) = π O ( s ) . Since s , u ∈ S s,u\in S s , u ∈ S , their difference belongs to S S S . Moreover,
π O ( s − u ) = π O ( s ) − π O ( u ) = 0. \pi_O(s-u)=\pi_O(s)-\pi_O(u)=0. π O ( s − u ) = π O ( s ) − π O ( u ) = 0.
Thus s − u s-u s − u is supported in Γ \Gamma Γ , and hence s − u ∈ S Γ s-u\in S_{\Gamma} s − u ∈ S Γ . Since u ∈ S E c u\in S_{E^{c}} u ∈ S E c , we also have π E ( u ) = 0 \pi_E(u)=0 π E ( u ) = 0 . It follows that
π E ( s ) = π E ( s ) − π E ( u ) = π E ( s − u ) ∈ π E ( S Γ ) . \pi_E(s)=\pi_E(s)-\pi_E(u)=\pi_E(s-u)\in\pi_E(S_{\Gamma}). π E ( s ) = π E ( s ) − π E ( u ) = π E ( s − u ) ∈ π E ( S Γ ) .
Since s ∈ S s\in S s ∈ S was arbitrary, this proves π E ( S ) ⊆ π E ( S Γ ) \pi_E(S)\subseteq\pi_E(S_{\Gamma}) π E ( S ) ⊆ π E ( S Γ ) . Combining the two inclusions, we obtain π E ( S Γ ) = π E ( S ) \pi_E(S_{\Gamma})=\pi_E(S) π E ( S Γ ) = π E ( S ) .
We can now state the algebraic characterization of local erasure recovery for an SSC.
Theorem III.8. Let Q Q Q be an SSC with additive gauge label space G ≤ W G\leq W G ≤ W and stabilizer label space S = G ∩ G ⊥ t S=G\cap G^{\perp_t} S = G ∩ G ⊥ t . Let ∅ ≠ E ⊆ Γ ⊆ [ n ] \varnothing\ne E\subseteq\Gamma\subseteq[n] ∅ = E ⊆ Γ ⊆ [ n ] . The following conditions are equivalent.
The erasure of E E E is recoverable within Γ \Gamma Γ in the sense of Definition III.1.
W E ∩ S Γ ⊥ t = G E W_E\cap S_{\Gamma}^{\perp_t}=G_E W E ∩ S Γ ⊥ t = G E .
dim p π E ( S Γ ) + dim p G E = 2 m ∣ E ∣ \dim_p\pi_E(S_{\Gamma})+\dim_p G_E=2m|E| dim p π E ( S Γ ) + dim p G E = 2 m ∣ E ∣ .
The erasure of E E E is globally correctable and π E ( S Γ ) = π E ( S ) \pi_E(S_{\Gamma})=\pi_E(S) π E ( S Γ ) = π E ( S ) .
Proof. We first show that conditions 2) and 3) are equivalent. By Lemma II.13, we have
dim p G E = dim p ( W E ∩ S Γ ⊥ t ) = 2 m ∣ E ∣ − dim p π E ( S Γ ) , \dim_p G_E=\dim_p(W_E\cap S_{\Gamma}^{\perp_t})=2m|E|-\dim_p\pi_E(S_{\Gamma}), dim p G E = dim p ( W E ∩ S Γ ⊥ t ) = 2 m ∣ E ∣ − dim p π E ( S Γ ) , which proves that 2) ⇒ \Rightarrow ⇒ 3). By the definition of S S S , we have S = G ∩ G ⊥ t S=G\cap G^{\perp_t} S = G ∩ G ⊥ t and S ⊆ G ⊥ t S\subseteq G^{\perp_t} S ⊆ G ⊥ t . Then G = ( G ⊥ t ) ⊥ t ⊆ S ⊥ t G=(G^{\perp_t})^{\perp_t}\subseteq S^{\perp_t} G = ( G ⊥ t ) ⊥ t ⊆ S ⊥ t . By the definition of G E G_E G E , G E = G ∩ W E ⊆ S ⊥ t ∩ W E G_E=G\cap W_E\subseteq S^{\perp_t}\cap W_E G E = G ∩ W E ⊆ S ⊥ t ∩ W E . By the definition of S Γ S_{\Gamma} S Γ , S Γ = S ∩ W Γ ⊆ S S_{\Gamma}=S\cap W_{\Gamma}\subseteq S S Γ = S ∩ W Γ ⊆ S , which implies that S ⊥ t ⊆ S Γ ⊥ t S^{\perp_t}\subseteq S_{\Gamma}^{\perp_t} S ⊥ t ⊆ S Γ ⊥ t . Then W E ∩ S ⊥ t ⊆ W E ∩ S Γ ⊥ t W_E\cap S^{\perp_t}\subseteq W_E\cap S_{\Gamma}^{\perp_t} W E ∩ S ⊥ t ⊆ W E ∩ S Γ ⊥ t . Therefore, G E ⊆ W E ∩ S Γ ⊥ t G_E\subseteq W_E\cap S_{\Gamma}^{\perp_t} G E ⊆ W E ∩ S Γ ⊥ t . By condition 3) and Lemma II.13, we have
dim p ( W E ∩ S Γ ⊥ t ) = 2 m ∣ E ∣ − dim p π E ( S Γ ) = 2 m ∣ E ∣ − ( 2 m ∣ E ∣ − dim p G E ) = dim p G E , \begin{aligned}
\dim_p(W_E\cap S_{\Gamma}^{\perp_t})
&=2m|E|-\dim_p\pi_E(S_{\Gamma})\\
&=2m|E|-(2m|E|-\dim_p G_E)\\
&=\dim_p G_E,
\end{aligned} dim p ( W E ∩ S Γ ⊥ t ) = 2 m ∣ E ∣ − dim p π E ( S Γ ) = 2 m ∣ E ∣ − ( 2 m ∣ E ∣ − dim p G E ) = dim p G E ,
which proves that 3) ⇒ \Rightarrow ⇒ 2).
We next show that conditions 2) and 4) are equivalent. On the one hand, G E ⊂ W E ∩ S ⊥ t ⊂ W E ∩ S Γ ⊥ t G_E \subset W_E \cap S^{\perp_t} \subset W_E \cap S_\Gamma^{\perp_t} G E ⊂ W E ∩ S ⊥ t ⊂ W E ∩ S Γ ⊥ t . By condition 2), we obtain G E = W E ∩ S Γ ⊥ t G_E = W_E \cap S_\Gamma^{\perp_t} G E = W E ∩ S Γ ⊥ t . By Lemma II.20, the erasure of E E E is globally correctable. On the other hand,
dim p G E = dim p ( W E ∩ S Γ ⊥ t ) = 2 m ∣ E ∣ − dim p π E ( S Γ ) , \dim_p G_E = \dim_p(W_E \cap S_\Gamma^{\perp_t}) = 2m|E|-\dim_p\pi_E(S_\Gamma), dim p G E = dim p ( W E ∩ S Γ ⊥ t ) = 2 m ∣ E ∣ − dim p π E ( S Γ ) ,
and
dim p G E = dim p ( W E ∩ S Γ ⊥ t ) = 2 m ∣ E ∣ − dim p π E ( S ) . \dim_p G_E = \dim_p(W_E \cap S_\Gamma^{\perp_t}) = 2m|E|-\dim_p\pi_E(S). dim p G E = dim p ( W E ∩ S Γ ⊥ t ) = 2 m ∣ E ∣ − dim p π E ( S ) .
Then dim p π E ( S Γ ) = dim p π E ( S ) \dim_p\pi_E(S_\Gamma)=\dim_p\pi_E(S) dim p π E ( S Γ ) = dim p π E ( S ) . Hence π E ( S Γ ) = π E ( S ) \pi_E(S_\Gamma)=\pi_E(S) π E ( S Γ ) = π E ( S ) , since π E ( S Γ ) ⊆ π E ( S ) \pi_E(S_\Gamma)\subseteq\pi_E(S) π E ( S Γ ) ⊆ π E ( S ) . This proves 2) ⇒ \Rightarrow ⇒ 4). Conversely, 4) ⇒ \Rightarrow ⇒ 2) follows directly.
It remains to prove 1) ⇒ \Rightarrow ⇒ 4). Assume that condition 1) holds, and set H = Γ ∖ E H=\Gamma\setminus E H = Γ ∖ E and O = Γ c O=\Gamma^c O = Γ c . By Definition III.1, there exist CPTP maps
R Γ , E : B ( H H ) ⟶ B ( H Γ ) , Φ Γ , E : B ( H B ) ⟶ B ( H B ) \mathcal{R}_{\Gamma,E}:\mathcal{B}(\mathcal{H}_H)\longrightarrow\mathcal{B}(\mathcal{H}_\Gamma),\quad\Phi_{\Gamma,E}:\mathcal{B}(\mathcal{H}_B)\longrightarrow\mathcal{B}(\mathcal{H}_B) R Γ , E : B ( H H ) ⟶ B ( H Γ ) , Φ Γ , E : B ( H B ) ⟶ B ( H B )
satisfying (16). Define R E : = R Γ , E ⊗ id O \mathcal{R}_E:=\mathcal{R}_{\Gamma,E}\otimes\operatorname{id}_O R E := R Γ , E ⊗ id O and Φ E : = Φ Γ , E \Phi_E:=\Phi_{\Gamma,E} Φ E := Φ Γ , E . Since E c = H ⊔ O E^c=H\sqcup O E c = H ⊔ O and [ n ] = Γ ⊔ O [n]=\Gamma\sqcup O [ n ] = Γ ⊔ O , the map R E : B ( H E c ) ⟶ B ( H [ n ] ) \mathcal{R}_E:\mathcal{B}(\mathcal{H}_{E^c})\longrightarrow\mathcal{B}(\mathcal{H}_{[n]}) R E : B ( H E c ) ⟶ B ( H [ n ] ) is CPTP. Moreover, by (16), we have
R E ∘ Tr E ∘ V = ( R Γ , E ⊗ id O ) ∘ Tr E ∘ V = V ∘ ( id A ⊗ Φ Γ , E ) = V ∘ ( id A ⊗ Φ E ) . \begin{aligned}
\mathcal{R}_E\circ\operatorname{Tr}_E\circ\mathcal{V}=(\mathcal{R}_{\Gamma,E}\otimes\operatorname{id}_O)\circ\operatorname{Tr}_E\circ\mathcal{V} \\
&=\mathcal{V}\circ(\operatorname{id}_A\otimes\Phi_{\Gamma,E}) \\
&=\mathcal{V}\circ(\operatorname{id}_A\otimes\Phi_E).
\end{aligned} R E ∘ Tr E ∘ V = ( R Γ , E ⊗ id O ) ∘ Tr E ∘ V = V ∘ ( id A ⊗ Φ Γ , E ) = V ∘ ( id A ⊗ Φ E ) .
Thus R E \mathcal{R}_E R E and Φ E \Phi_E Φ E satisfy the recovery identity in Definition II.19, and the erasure of E E E is globally correctable. Finally, Lemma III.7 implies that π E ( S Γ ) = π E ( S ) \pi_E(S_\Gamma)=\pi_E(S) π E ( S Γ ) = π E ( S ) . This proves 1) ⇒ \Rightarrow ⇒ 4).
Combining the preceding implications with Proposition III.6, we obtain
1 ) ⇒ 4 ) ⇒ 3 ) ⇒ 2 ) ⇒ 1 ) 1)\Rightarrow4)\Rightarrow3)\Rightarrow2)\Rightarrow1) 1 ) ⇒ 4 ) ⇒ 3 ) ⇒ 2 ) ⇒ 1 )
and complete the proof.
Corollary III.10. An ( r , δ ) (r,\delta) ( r , δ ) -SSLRC with K > 1 K>1 K > 1 has d ≥ δ d\ge\delta d ≥ δ .
Proof. Suppose to the contrary that d < δ d<\delta d < δ . By (11), choose w ∈ S ⊥ t ∖ G w\in S^{\perp_t}\setminus G w ∈ S ⊥ t ∖ G with swt ( w ) = d \operatorname{swt}(w)=d swt ( w ) = d . Let i ∈ supp ( w ) i\in\operatorname{supp}(w) i ∈ supp ( w ) , and let Γ i \Gamma_i Γ i be a repair set for i i i . Set E = supp ( w ) ∩ Γ i E=\operatorname{supp}(w)\cap\Gamma_i E = supp ( w ) ∩ Γ i , and let v v v be the zero extension of π Γ i ( w ) \pi_{\Gamma_i}(w) π Γ i ( w ) . Then π Γ i ( v ) = π Γ i ( w ) \pi_{\Gamma_i}(v)=\pi_{\Gamma_i}(w) π Γ i ( v ) = π Γ i ( w ) , π Γ i c ( v ) = 0 \pi_{\Gamma_i^c}(v)=0 π Γ i c ( v ) = 0 . Hence, supp ( v ) = supp ( w ) ∩ Γ i = E \operatorname{supp}(v)=\operatorname{supp}(w)\cap\Gamma_i=E supp ( v ) = supp ( w ) ∩ Γ i = E , v ∈ W E v\in W_E v ∈ W E . Since i ∈ E i\in E i ∈ E and E ⊆ supp ( w ) E\subseteq\operatorname{supp}(w) E ⊆ supp ( w ) , we have 0 < ∣ E ∣ ≤ ∣ supp ( w ) ∣ = d < δ 0<|E|\le|\operatorname{supp}(w)|=d<\delta 0 < ∣ E ∣ ≤ ∣ supp ( w ) ∣ = d < δ . Thus, by Definition III.1, the erasure of E E E is recoverable within Γ i \Gamma_i Γ i .
For every z ∈ S Γ i z\in S_{\Gamma_i} z ∈ S Γ i , the support of z z z is contained in Γ i \Gamma_i Γ i . Since π Γ i ( v ) = π Γ i ( w ) \pi_{\Gamma_i}(v)=\pi_{\Gamma_i}(w) π Γ i ( v ) = π Γ i ( w ) and w ∈ S ⊥ t w\in S^{\perp_t} w ∈ S ⊥ t , we obtain [ v , z ] t = [ w , z ] t = 0 [v,z]_t=[w,z]_t=0 [ v , z ] t = [ w , z ] t = 0 . Therefore, v ∈ W E ∩ S Γ i ⊥ t v\in W_E\cap S_{\Gamma_i}^{\perp_t} v ∈ W E ∩ S Γ i ⊥ t . By condition 2) of Theorem III.8, W E ∩ S Γ i ⊥ t = G E W_E\cap S_{\Gamma_i}^{\perp_t}=G_E W E ∩ S Γ i ⊥ t = G E , and hence v ∈ G E v\in G_E v ∈ G E . Since G E ⊆ G ⊆ S ⊥ t G_E\subseteq G\subseteq S^{\perp_t} G E ⊆ G ⊆ S ⊥ t , we have w − v ∈ S ⊥ t w-v\in S^{\perp_t} w − v ∈ S ⊥ t . Moreover, w − v ∉ G w-v\notin G w − v ∈ / G . Indeed, if w − v ∈ G w-v\in G w − v ∈ G , then w = ( w − v ) + v ∈ G w=(w-v)+v\in G w = ( w − v ) + v ∈ G , contrary to the choice of w w w . By the construction of w − v w-v w − v , we have π Γ i ( w − v ) = 0 \pi_{\Gamma_i}(w-v)=0 π Γ i ( w − v ) = 0 , π Γ i c ( w − v ) = π Γ i c ( w ) \pi_{\Gamma_i^c}(w-v)=\pi_{\Gamma_i^c}(w) π Γ i c ( w − v ) = π Γ i c ( w ) . Consequently, supp ( w − v ) = supp ( w ) ∖ Γ i = supp ( w ) ∖ E \operatorname{supp}(w-v)=\operatorname{supp}(w)\setminus\Gamma_i=\operatorname{supp}(w)\setminus E supp ( w − v ) = supp ( w ) ∖ Γ i = supp ( w ) ∖ E . Since E ≠ ∅ E\ne\varnothing E = ∅ , it follows that w − v ∈ S ⊥ t ∖ G w-v\in S^{\perp_t}\setminus G w − v ∈ S ⊥ t ∖ G , swt ( w − v ) = d − ∣ E ∣ < d \operatorname{swt}(w-v)=d-|E|<d swt ( w − v ) = d − ∣ E ∣ < d . This contradicts (11). Therefore, d ≥ δ d\ge\delta d ≥ δ .
CSS and Hermitian Hull Constructions
In this subsection, we apply Theorem III.8 to the CSS and Hermitian hull constructions of SSCs [2 Corollaries 4 and 5] . Let C X , C Z ≤ F q n C_X,C_Z \leq\mathbb{F}_q^n C X , C Z ≤ F q n be F q \mathbb{F}_q F q -linear codes with full-row-rank generator matrices M X M_X M X and M Z M_Z M Z . Define
G = C X × C Z , H X = C X ∩ C Z ⊥ , H Z = C Z ∩ C X ⊥ , G=C_X\times C_Z,\quad H_X=C_X\cap C_Z^\perp,\quad H_Z=C_Z\cap C_X^\perp, G = C X × C Z , H X = C X ∩ C Z ⊥ , H Z = C Z ∩ C X ⊥ ,
and let c X = dim q C X c_X=\dim_q C_X c X = dim q C X and c Z = dim q C Z c_Z=\dim_q C_Z c Z = dim q C Z .
We first determine the parameters of the CSS SSC.
Lemma III.11. Let b = rank q ( M X M Z T ) b=\operatorname{rank}_q(M_XM_Z^{T}) b = rank q ( M X M Z T ) . The gauge label space G G G defines a CSS SSC Q Q Q with stabilizer label space S = H X × H Z S=H_X\times H_Z S = H X × H Z , logical dimension K = q n − c X − c Z + b K=q^{n-c_X-c_Z+b} K = q n − c X − c Z + b , and gauge dimension R = q b R=q^b R = q b . If K > 1 K>1 K > 1 , its minimum distance is d = min { d X , d Z } d=\min\{d_X,d_Z\} d = min { d X , d Z } , where
d X = min a ∈ H Z ⊥ ∖ C X wt ( a ) , d Z = min b ∈ H X ⊥ ∖ C Z wt ( b ) . (26) d_X=\min_{a\in H_Z^\perp\setminus C_X}\operatorname{wt}(a),\quad d_Z=\min_{b\in H_X^\perp\setminus C_Z}\operatorname{wt}(b). \tag*{(26)} d X = a ∈ H Z ⊥ ∖ C X min wt ( a ) , d Z = b ∈ H X ⊥ ∖ C Z min wt ( b ) . (26)
Proof. By [2 Corollary 4] , G = C X × C Z G=C_X\times C_Z G = C X × C Z defines a CSS SSC Q Q Q with stabilizer label space
S = ( C X ∩ C Z ⊥ ) × ( C Z ∩ C X ⊥ ) = H X × H Z . S=(C_X\cap C_Z^\perp)\times(C_Z\cap C_X^\perp)=H_X\times H_Z. S = ( C X ∩ C Z ⊥ ) × ( C Z ∩ C X ⊥ ) = H X × H Z .
The gauge label space G G G has the full-row-rank generator matrix
M G = ( M X 0 0 M Z ) . M_G=\begin{pmatrix}M_X&0\\0&M_Z\end{pmatrix}. M G = ( M X 0 0 M Z ) .
whose two row blocks consist of X X X -type and Z Z Z -type labels, respectively. The parameter formulas for this construction imply that
k = n − c X + c Z + dim q S 2 , g = c X + c Z − dim q S 2 , k=n-\frac{c_X+c_Z+\dim_q S}{2},\quad g=\frac{c_X+c_Z-\dim_q S}{2}, k = n − 2 c X + c Z + dim q S , g = 2 c X + c Z − dim q S ,
where K = q k K=q^k K = q k and R = q g R=q^g R = q g are the logical and gauge dimensions, respectively.
Let B = M X M Z T B=M_XM_Z^T B = M X M Z T and rank q B = b \operatorname{rank}_q B=b rank q B = b . Since M X M_X M X and M Z M_Z M Z have full row rank, the codewords in C X C_X C X and C Z C_Z C Z can be written uniquely as u M X uM_X u M X and v M Z vM_Z v M Z , respectively, where u ∈ F q c X u\in\mathbb{F}_q^{c_X} u ∈ F q c X and v ∈ F q c Z v\in\mathbb{F}_q^{c_Z} v ∈ F q c Z . By the definitions of H X H_X H X and H Z H_Z H Z , we have
u M X ∈ H X ⟺ u B = 0 , v M Z ∈ H Z ⟺ v B T = 0. uM_X\in H_X\Longleftrightarrow uB=0,\quad vM_Z\in H_Z\Longleftrightarrow vB^T=0. u M X ∈ H X ⟺ u B = 0 , v M Z ∈ H Z ⟺ v B T = 0.
By (1) ,
dim q H X = c X − b , dim q H Z = c Z − b . \dim_q H_X=c_X-b,\quad\dim_q H_Z=c_Z-b. dim q H X = c X − b , dim q H Z = c Z − b .
Since S = H X × H Z S=H_X\times H_Z S = H X × H Z , it follows that dim q S = dim q H X + dim q H Z = c X + c Z − 2 b \dim_q S=\dim_q H_X+\dim_q H_Z=c_X+c_Z-2b dim q S = dim q H X + dim q H Z = c X + c Z − 2 b . Hence
k = n − c X − c Z + b , g = b , k=n-c_X-c_Z+b,\quad g=b, k = n − c X − c Z + b , g = b ,
which imply that K = q n − c X − c Z + b K=q^{n-c_X-c_Z+b} K = q n − c X − c Z + b and R = q b R=q^b R = q b .
To determine the minimum distance of Q Q Q , suppose that K > 1 K>1 K > 1 , or equivalently, k > 0 k>0 k > 0 . By the above dimension formulas, we have
dim q H Z ⊥ − dim q C X = n − ( c Z − b ) − c X = k > 0 , \dim_q H_Z^\perp-\dim_q C_X=n-(c_Z-b)-c_X=k>0, dim q H Z ⊥ − dim q C X = n − ( c Z − b ) − c X = k > 0 ,
dim q H X ⊥ − dim q C Z = n − ( c X − b ) − c Z = k > 0. \dim_q H_X^\perp-\dim_q C_Z=n-(c_X-b)-c_Z=k>0. dim q H X ⊥ − dim q C Z = n − ( c X − b ) − c Z = k > 0.
Thus both H Z ⊥ ∖ C X H_Z^\perp\setminus C_X H Z ⊥ ∖ C X and H X ⊥ ∖ C Z H_X^\perp\setminus C_Z H X ⊥ ∖ C Z are nonempty. Since S = H X × H Z S=H_X\times H_Z S = H X × H Z is F q \mathbb{F}_q F q -linear, S ⊥ t = S ⊥ s = H Z ⊥ × H X ⊥ S^{\perp_t}=S^{\perp_s}=H_Z^\perp\times H_X^\perp S ⊥ t = S ⊥ s = H Z ⊥ × H X ⊥ . By (11) ,
d = min { swt ( a ∣ b ) : ( a ∣ b ) ∈ ( H Z ⊥ × H X ⊥ ) ∖ ( C X × C Z ) } . d=\min\left\{\operatorname{swt}(a\mid b):(a\mid b)\in(H_Z^\perp\times H_X^\perp)\setminus(C_X\times C_Z)\right\}. d = min { swt ( a ∣ b ) : ( a ∣ b ) ∈ ( H Z ⊥ × H X ⊥ ) ∖ ( C X × C Z ) } .
For every label in this set, either a ∈ H Z ⊥ ∖ C X a\in H_Z^\perp\setminus C_X a ∈ H Z ⊥ ∖ C X or b ∈ H X ⊥ ∖ C Z b\in H_X^\perp\setminus C_Z b ∈ H X ⊥ ∖ C Z . Moreover,
swt ( a ∣ b ) = ∣ supp ( a ) ∪ supp ( b ) ∣ ≥ max { wt ( a ) , wt ( b ) } . \operatorname{swt}(a\mid b)=\lvert\operatorname{supp}(a)\cup\operatorname{supp}(b)\rvert\geq\max\{\operatorname{wt}(a),\operatorname{wt}(b)\}. swt ( a ∣ b ) = ∣ supp ( a ) ∪ supp ( b )∣ ≥ max { wt ( a ) , wt ( b )} .
It follows that d ≥ min { d X , d Z } d\geq\min\{d_X,d_Z\} d ≥ min { d X , d Z } . Conversely, choose a ∈ H Z ⊥ ∖ C X a\in H_Z^\perp\setminus C_X a ∈ H Z ⊥ ∖ C X and b ∈ H X ⊥ ∖ C Z b\in H_X^\perp\setminus C_Z b ∈ H X ⊥ ∖ C Z such that wt ( a ) = d X , wt ( b ) = d Z \operatorname{wt}(a)=d_X,\operatorname{wt}(b)=d_Z wt ( a ) = d X , wt ( b ) = d Z . Then ( a ∣ 0 ) , ( 0 ∣ b ) ∈ S ⊥ t ∖ G (a\mid0),(0\mid b)\in S^{\perp_t}\setminus G ( a ∣ 0 ) , ( 0 ∣ b ) ∈ S ⊥ t ∖ G . Hence,
d ≤ swt ( a ∣ 0 ) = d X , d ≤ swt ( 0 ∣ b ) = d Z , d\leq\operatorname{swt}(a\mid0)=d_X,\quad d\leq\operatorname{swt}(0\mid b)=d_Z, d ≤ swt ( a ∣ 0 ) = d X , d ≤ swt ( 0 ∣ b ) = d Z ,
which, together with the preceding lower bound, imply that d = min { d X , d Z } d=\min\{d_X,d_Z\} d = min { d X , d Z } .
We now characterize the ( r , δ ) (r,\delta) ( r , δ ) -locality of the CSS SSC in Lemma III.11.
Proposition III.12. Let Q Q Q be the CSS SSC in Lemma III.11. Then Q Q Q is an ( r , δ ) (r,\delta) ( r , δ ) -SSLRC if and only if it admits a repair cover satisfying (17) such that, for every repair set Γ i \Gamma_i Γ i and every nonempty E ⊆ Γ i E\subseteq\Gamma_i E ⊆ Γ i with ∣ E ∣ < δ \lvert E\rvert<\delta ∣ E ∣ < δ ,
dim q π E ( ( H X ) Γ i ) + dim q ( C Z ) E = ∣ E ∣ , (27) \dim_q\pi_E((H_X)_{\Gamma_i})+\dim_q(C_Z)_E=\lvert E\rvert, \tag*{(27)} dim q π E (( H X ) Γ i ) + dim q ( C Z ) E = ∣ E ∣ , (27) dim q π E ( ( H Z ) Γ i ) + dim q ( C X ) E = ∣ E ∣ . (28) \dim_q\pi_E((H_Z)_{\Gamma_i})+\dim_q(C_X)_E=\lvert E\rvert. \tag*{(28)} dim q π E (( H Z ) Γ i ) + dim q ( C X ) E = ∣ E ∣ . (28)
Proof. By Definition III.3, Q Q Q is an ( r , δ ) (r,\delta) ( r , δ ) -SSLRC if and only if it admits a repair cover satisfying (17) such that every nonempty E ⊆ Γ i E \subseteq\Gamma_i E ⊆ Γ i with ∣ E ∣ < δ |E| < \delta ∣ E ∣ < δ is recoverable within Γ i \Gamma_i Γ i . We first express the recovery criterion for a single erased set in terms of the dimensions of the component codes.
Fix ∅ ≠ E ⊆ Γ ⊆ [ n ] \varnothing\ne E \subseteq\Gamma\subseteq[n] ∅ = E ⊆ Γ ⊆ [ n ] . By condition 3) of Theorem III.8, the erasure of E E E is recoverable within Γ \Gamma Γ if and only if
dim F p π E ( S Γ ) + dim F p G E = 2 m ∣ E ∣ . \dim_{\mathbb{F}_p} \pi_E(S_\Gamma) + \dim_{\mathbb{F}_p} G_E = 2m|E|. dim F p π E ( S Γ ) + dim F p G E = 2 m ∣ E ∣.
Since the spaces π E ( S Γ ) \pi_E(S_\Gamma) π E ( S Γ ) and G E G_E G E are F q \mathbb{F}_q F q -linear, the recovery condition is equivalent to
dim F q π E ( S Γ ) + dim F q G E = 2 ∣ E ∣ . \dim_{\mathbb{F}_q} \pi_E(S_\Gamma) + \dim_{\mathbb{F}_q} G_E = 2|E|. dim F q π E ( S Γ ) + dim F q G E = 2∣ E ∣.
We now express these dimensions in terms of C X C_X C X , C Z C_Z C Z , H X H_X H X , and H Z H_Z H Z . By Lemma III.11, G = C X × C Z G = C_X \times C_Z G = C X × C Z and S = H X × H Z S = H_X \times H_Z S = H X × H Z . Hence, S Γ = ( H X ) Γ × ( H Z ) Γ S_\Gamma= (H_X)_\Gamma\times(H_Z)_\Gamma S Γ = ( H X ) Γ × ( H Z ) Γ and G E = ( C X ) E × ( C Z ) E G_E = (C_X)_E \times(C_Z)_E G E = ( C X ) E × ( C Z ) E . Since the projection π E \pi_E π E acts on each component, we have
π E ( S Γ ) = π E ( ( H X ) Γ ) × π E ( ( H Z ) Γ ) . \pi_E(S_\Gamma) = \pi_E((H_X)_\Gamma) \times\pi_E((H_Z)_\Gamma). π E ( S Γ ) = π E (( H X ) Γ ) × π E (( H Z ) Γ ) .
Therefore,
dim F q π E ( S Γ ) = dim F q π E ( ( H X ) Γ ) + dim F q π E ( ( H Z ) Γ ) , dim F q G E = dim F q ( C X ) E + dim F q ( C Z ) E . \begin{aligned}
\dim_{\mathbb{F}_q} \pi_E(S_\Gamma) &= \dim_{\mathbb{F}_q} \pi_E((H_X)_\Gamma) + \dim_{\mathbb{F}_q} \pi_E((H_Z)_\Gamma),\\
\dim_{\mathbb{F}_q} G_E &= \dim_{\mathbb{F}_q}(C_X)_E + \dim_{\mathbb{F}_q}(C_Z)_E.
\end{aligned} dim F q π E ( S Γ ) dim F q G E = dim F q π E (( H X ) Γ ) + dim F q π E (( H Z ) Γ ) , = dim F q ( C X ) E + dim F q ( C Z ) E .
It follows that the erasure of E E E is recoverable within Γ \Gamma Γ if and only if
( dim F q π E ( ( H X ) Γ ) + dim F q ( C Z ) E ) + ( dim F q π E ( ( H Z ) Γ ) + dim F q ( C X ) E ) = 2 ∣ E ∣ . \begin{aligned}
\bigl(\dim_{\mathbb{F}_q} \pi_E((H_X)_\Gamma) + \dim_{\mathbb{F}_q}(C_Z)_E\bigr)\\
+ \bigl(\dim_{\mathbb{F}_q} \pi_E((H_Z)_\Gamma) + \dim_{\mathbb{F}_q}(C_X)_E\bigr) = 2|E|.
\end{aligned} ( dim F q π E (( H X ) Γ ) + dim F q ( C Z ) E ) + ( dim F q π E (( H Z ) Γ ) + dim F q ( C X ) E ) = 2∣ E ∣.
We next prove that dim F q π E ( ( H X ) Γ ) + dim F q ( C Z ) E ≤ ∣ E ∣ \dim_{\mathbb{F}_q} \pi_E((H_X)_\Gamma) + \dim_{\mathbb{F}_q}(C_Z)_E \le|E| dim F q π E (( H X ) Γ ) + dim F q ( C Z ) E ≤ ∣ E ∣ . To this end, we show that π E ( ( H X ) Γ ) \pi_E((H_X)_\Gamma) π E (( H X ) Γ ) and sh E ( C Z ) \operatorname{sh}_E(C_Z) sh E ( C Z ) are orthogonal in F q E \mathbb{F}_q^E F q E . Let x ∈ ( H X ) Γ x \in(H_X)_\Gamma x ∈ ( H X ) Γ and z ∈ ( C Z ) E z \in(C_Z)_E z ∈ ( C Z ) E . Since z z z is supported on E E E , we have z j = 0 z_j = 0 z j = 0 for every j ∉ E j \notin E j ∈ / E . Hence
π E ( x ) ⋅ π E ( z ) = ∑ j ∈ E x j z j = ∑ j = 1 n x j z j = x ⋅ z = 0. \pi_E(x) \cdot\pi_E(z) = \sum_{j \in E} x_j z_j = \sum_{j=1}^{n} x_j z_j = x \cdot z = 0. π E ( x ) ⋅ π E ( z ) = j ∈ E ∑ x j z j = j = 1 ∑ n x j z j = x ⋅ z = 0.
The last equality follows from x ∈ H X ⊆ C Z ⊥ x \in H_X \subseteq C_Z^\perp x ∈ H X ⊆ C Z ⊥ and z ∈ C Z z \in C_Z z ∈ C Z . Since x x x and z z z are arbitrary, we obtain π E ( ( C Z ) E ) ⊆ ( π E ( ( H X ) Γ ) ) ⊥ \pi_E((C_Z)_E) \subseteq\bigl(\pi_E((H_X)_\Gamma)\bigr)^\perp π E (( C Z ) E ) ⊆ ( π E (( H X ) Γ ) ) ⊥ , where the Euclidean orthogonal complement is taken in F q E \mathbb{F}_q^E F q E . By the definition of shortening, sh E ( C Z ) = π E ( ( C Z ) E ) \operatorname{sh}_E(C_Z) = \pi_E((C_Z)_E) sh E ( C Z ) = π E (( C Z ) E ) . Moreover, since π E \pi_E π E is injective on ( C Z ) E (C_Z)_E ( C Z ) E , we have dim F q ( C Z ) E = dim F q sh E ( C Z ) \dim_{\mathbb{F}_q}(C_Z)_E = \dim_{\mathbb{F}_q}\operatorname{sh}_E(C_Z) dim F q ( C Z ) E = dim F q sh E ( C Z ) . The Euclidean inner product on F q E \mathbb{F}_q^E F q E is nondegenerate. Therefore,
dim F q ( C Z ) E = dim F q sh E ( C Z ) ≤ dim F q ( π E ( ( H X ) Γ ) ) ⊥ = ∣ E ∣ − dim F q π E ( ( H X ) Γ ) . \begin{aligned}
\dim_{\mathbb{F}_q}(C_Z)_E &= \dim_{\mathbb{F}_q}\operatorname{sh}_E(C_Z)\\
&\le\dim_{\mathbb{F}_q}\bigl(\pi_E((H_X)_\Gamma)\bigr)^\perp\\
&= |E| - \dim_{\mathbb{F}_q}\pi_E((H_X)_\Gamma).
\end{aligned} dim F q ( C Z ) E = dim F q sh E ( C Z ) ≤ dim F q ( π E (( H X ) Γ ) ) ⊥ = ∣ E ∣ − dim F q π E (( H X ) Γ ) .
Thus
dim F q π E ( ( H X ) Γ ) + dim F q ( C Z ) E ≤ ∣ E ∣ . \dim_{\mathbb{F}_q} \pi_E((H_X)_\Gamma) + \dim_{\mathbb{F}_q}(C_Z)_E \le|E|. dim F q π E (( H X ) Γ ) + dim F q ( C Z ) E ≤ ∣ E ∣.
The same argument with X X X and Z Z Z interchanged shows that
dim F q π E ( ( H Z ) Γ ) + dim F q ( C X ) E ≤ ∣ E ∣ . \dim_{\mathbb{F}_q} \pi_E((H_Z)_\Gamma) + \dim_{\mathbb{F}_q}(C_X)_E \le|E|. dim F q π E (( H Z ) Γ ) + dim F q ( C X ) E ≤ ∣ E ∣.
Each left-hand side is at most ∣ E ∣ |E| ∣ E ∣ . Their sum equals 2 ∣ E ∣ 2|E| 2∣ E ∣ if and only if both inequalities are equalities. By condition 3) of Theorem III.8 and the preceding CSS dimension formulas, the erasure of E E E is therefore recoverable within Γ \Gamma Γ if and only if
dim F q π E ( ( H X ) Γ ) + dim F q ( C Z ) E = ∣ E ∣ , dim F q π E ( ( H Z ) Γ ) + dim F q ( C X ) E = ∣ E ∣ . \begin{aligned}
\dim_{\mathbb{F}_q} \pi_E((H_X)_\Gamma) + \dim_{\mathbb{F}_q}(C_Z)_E &= |E|,\\
\dim_{\mathbb{F}_q} \pi_E((H_Z)_\Gamma) + \dim_{\mathbb{F}_q}(C_X)_E &= |E|.
\end{aligned} dim F q π E (( H X ) Γ ) + dim F q ( C Z ) E dim F q π E (( H Z ) Γ ) + dim F q ( C X ) E = ∣ E ∣ , = ∣ E ∣.
Suppose that Q Q Q is an ( r , δ ) (r,\delta) ( r , δ ) -SSLRC. By Definition III.3, there is a repair cover satisfying (17) such that every nonempty E ⊆ Γ i E \subseteq\Gamma_i E ⊆ Γ i with ∣ E ∣ < δ |E| < \delta ∣ E ∣ < δ is recoverable within Γ i \Gamma_i Γ i . The preceding equivalence with Γ = Γ i \Gamma= \Gamma_i Γ = Γ i implies that (27) and (28) hold for every such E E E . Conversely, suppose that a repair cover satisfies (17) . Assume that (27) and (28) hold for every repair set Γ i \Gamma_i Γ i and every nonempty E ⊆ Γ i E \subseteq\Gamma_i E ⊆ Γ i with ∣ E ∣ < δ |E| < \delta ∣ E ∣ < δ . By the same equivalence, each such E E E is recoverable within Γ i \Gamma_i Γ i . Hence Q Q Q is an ( r , δ ) (r,\delta) ( r , δ ) -SSLRC by Definition III.3. □ \square □
Remark III.13. If C X ⊆ C Z ⊥ C_X \subseteq C_Z^\perp C X ⊆ C Z ⊥ , then H X = C X H_X = C_X H X = C X , H Z = C Z H_Z = C_Z H Z = C Z , and b = 0 b = 0 b = 0 . Thus G = S = C X × C Z G = S = C_X \times C_Z G = S = C X × C Z , and Proposition III.12 reduces to the CSS stabilizer construction [2 Corollary 4] .
We next consider the Hermitian hull construction of SSCs, using the coordinate map ψ \psi ψ in (6) and Lemma II.14. Let C ≤ F q 2 [ n ] C \le\mathbb{F}_{q^2}^{[n]} C ≤ F q 2 [ n ] be an F q 2 \mathbb{F}_{q^2} F q 2 -linear code with Hermitian hull H = Hull H ( C ) = C ∩ C ⊥ H H = \operatorname{Hull}_H(C) = C \cap C^{\perp_H} H = Hull H ( C ) = C ∩ C ⊥ H . Write c = dim F q 2 C c = \dim_{\mathbb{F}_{q^2}} C c = dim F q 2 C and h H = dim F q 2 H h_H = \dim_{\mathbb{F}_{q^2}} H h H = dim F q 2 H .
Lemma III.14. The gauge label space G = ψ − 1 ( C ) G=\psi^{-1}(C) G = ψ − 1 ( C ) defines an SSC Q Q Q with stabilizer label space S = ψ − 1 ( H ) S=\psi^{-1}(H) S = ψ − 1 ( H ) , logical dimension K = q n − c − h H K=q^{n-c-h_H} K = q n − c − h H , and gauge dimension R = q c − h H R=q^{c-h_H} R = q c − h H . If K > 1 K>1 K > 1 , or equivalently c + h H < n c+h_H<n c + h H < n , its minimum distance is
d = min { wt ( v ) : v ∈ H ⊥ H ∖ C } . d=\min\{\operatorname{wt}(v):v\in H^{\perp_H}\setminus C\}. d = min { wt ( v ) : v ∈ H ⊥ H ∖ C } . Proof. By the Hermitian subsystem construction of [2 Corollary 5] , the gauge label space G = ψ − 1 ( C ) G=\psi^{-1}(C) G = ψ − 1 ( C ) defines an SSC Q Q Q . Since C C C is F q 2 \mathbb{F}_{q^2} F q 2 -linear, Lemma II.14 implies G ⊥ t = ψ − 1 ( C ⊥ H ) G^{\perp_t}=\psi^{-1}(C^{\perp_H}) G ⊥ t = ψ − 1 ( C ⊥ H ) . Hence the stabilizer label space is
S = G ∩ G ⊥ t = ψ − 1 ( C ) ∩ ψ − 1 ( C ⊥ H ) = ψ − 1 ( C ∩ C ⊥ H ) = ψ − 1 ( H ) . \begin{aligned}
S&=G\cap G^{\perp_t}\\
&=\psi^{-1}(C)\cap\psi^{-1}(C^{\perp_H})\\
&=\psi^{-1}(C\cap C^{\perp_H})\\
&=\psi^{-1}(H).
\end{aligned} S = G ∩ G ⊥ t = ψ − 1 ( C ) ∩ ψ − 1 ( C ⊥ H ) = ψ − 1 ( C ∩ C ⊥ H ) = ψ − 1 ( H ) . We next determine the logical and gauge dimensions. The map ψ \psi ψ is an F q \mathbb{F}_q F q -linear isomorphism, and F q 2 \mathbb{F}_{q^2} F q 2 has dimension two over F q \mathbb{F}_q F q . Therefore, dim q G = 2 c \dim_qG=2c dim q G = 2 c and dim q S = 2 h H \dim_qS=2h_H dim q S = 2 h H . By (12) ,
k = n − dim q G + dim q S 2 = n − 2 c + 2 h H 2 = n − c − h H , k=n-\frac{\dim_qG+\dim_qS}{2}=n-\frac{2c+2h_H}{2}=n-c-h_H, k = n − 2 dim q G + dim q S = n − 2 2 c + 2 h H = n − c − h H , and
g = dim q G − dim q S 2 = 2 c − 2 h H 2 = c − h H . g=\frac{\dim_qG-\dim_qS}{2}=\frac{2c-2h_H}{2}=c-h_H. g = 2 dim q G − dim q S = 2 2 c − 2 h H = c − h H . Thus K = q n − c − h H K=q^{n-c-h_H} K = q n − c − h H , R = q c − h H R=q^{c-h_H} R = q c − h H .
To determine the minimum distance of Q Q Q , assume that K > 1 K>1 K > 1 . Then k = n − c − h H > 0 k=n-c-h_H>0 k = n − c − h H > 0 . Since H ⊆ C ⊥ H H\subseteq C^{\perp_H} H ⊆ C ⊥ H , Hermitian duality implies C ⊆ H ⊥ H C\subseteq H^{\perp_H} C ⊆ H ⊥ H . Moreover,
dim q 2 H ⊥ H − dim q 2 C = n − h H − c = k > 0. \dim_{q^2}H^{\perp_H}-\dim_{q^2}C=n-h_H-c=k>0. dim q 2 H ⊥ H − dim q 2 C = n − h H − c = k > 0. Thus H ⊥ H ∖ C H^{\perp_H}\setminus C H ⊥ H ∖ C is nonempty. The hull H H H is F q 2 \mathbb{F}_{q^2} F q 2 -linear. By Lemma II.14, S ⊥ t = ψ − 1 ( H ⊥ H ) S^{\perp_t}=\psi^{-1}(H^{\perp_H}) S ⊥ t = ψ − 1 ( H ⊥ H ) . Consequently, ψ \psi ψ restricts to a bijection
S ⊥ t ∖ G ⟶ H ⊥ H ∖ C . S^{\perp_t}\setminus G\longrightarrow H^{\perp_H}\setminus C. S ⊥ t ∖ G ⟶ H ⊥ H ∖ C . Since the coordinate map ψ \psi ψ preserves support, we have swt ( w ) = wt ( ψ ( w ) ) \operatorname{swt}(w)=\operatorname{wt}(\psi(w)) swt ( w ) = wt ( ψ ( w )) for w ∈ W w\in W w ∈ W . By (11) ,
d = min w ∈ S ⊥ t ∖ G swt ( w ) = min v ∈ H ⊥ H ∖ C wt ( v ) . d=\min_{w\in S^{\perp_t}\setminus G}\operatorname{swt}(w)=\min_{v\in H^{\perp_H}\setminus C}\operatorname{wt}(v). d = w ∈ S ⊥ t ∖ G min swt ( w ) = v ∈ H ⊥ H ∖ C min wt ( v ) .
We now characterize the ( r , δ ) (r,\delta) ( r , δ ) -locality of the SSC in Lemma III.14.
Proposition III.15. Let Q Q Q be the SSC in Lemma III.14. Then Q Q Q is an ( r , δ ) (r,\delta) ( r , δ ) -SSLRC if and only if it admits a repair cover satisfying (17) such that, for every repair set Γ i \Gamma_i Γ i and every nonempty E ⊆ Γ i E\subseteq\Gamma_i E ⊆ Γ i with ∣ E ∣ < δ |E|<\delta ∣ E ∣ < δ ,
dim q 2 π E ( H Γ i ) + dim q 2 C E = ∣ E ∣ . (29) \dim_{q^2}\pi_E(H_{\Gamma_i})+\dim_{q^2}C_E=|E|. \tag*{(29)} dim q 2 π E ( H Γ i ) + dim q 2 C E = ∣ E ∣. (29) Proof. We proceed as in the proof of Proposition III.12, using the corresponding dimension formulas. By Lemma III.14, ψ ( G ) = C \psi(G)=C ψ ( G ) = C and ψ ( S ) = H \psi(S)=H ψ ( S ) = H . The coordinate map ψ \psi ψ is an F p \mathbb{F}_p F p -linear isomorphism. It preserves support and commutes with coordinate projections. Since C C C and H H H are F q 2 \mathbb{F}_{q^2} F q 2 -linear and q = p m q=p^m q = p m , we have
dim p G E = 2 m dim q 2 C E , \dim_pG_E=2m\dim_{q^2}C_E, dim p G E = 2 m dim q 2 C E , dim p π E ( S Γ ) = 2 m dim q 2 π E ( H Γ ) \dim_p\pi_E(S_\Gamma)=2m\dim_{q^2}\pi_E(H_\Gamma) dim p π E ( S Γ ) = 2 m dim q 2 π E ( H Γ ) for every ∅ ≠ E ⊆ Γ ⊆ [ n ] \varnothing\ne E\subseteq\Gamma\subseteq[n] ∅ = E ⊆ Γ ⊆ [ n ] . The assertion follows from condition 3) of Theorem III.8 and Definition III.3.
Remark III.16. If C ⊆ C ⊥ H C\subseteq C^{\perp_H} C ⊆ C ⊥ H , then H = C H=C H = C and g = 0 g=0 g = 0 . Thus Proposition III.15 reduces to the Hermitian stabilizer construction [2 Corollary 5] .
Bounds for ( r , δ ) (r,\delta) ( r , δ ) -SSLRCs Without Assuming Purity
In this section, we derive three bounds for ( r , δ ) (r,\delta) ( r , δ ) -SSLRCs without assuming purity: a redundancy bound for disjoint repair sets, a supported-subspace bound from supported stabilizer and gauge subspaces, and a mixed counting bound for known erasures within selected repair sets and errors at unknown positions outside their union. Throughout this section, let Q Q Q be an ( r , δ ) (r,\delta) ( r , δ ) -SSLRC with parameters ( ( n , K , R , d ) ) q ((n,K,R,d))_q (( n , K , R , d ) ) q , where q = p m q=p^m q = p m and K > 1 K>1 K > 1 . The gauge label space G G G is F p \mathbb{F}_p F p -linear. Let S = G ∩ G ⊥ t S=G\cap G^{\perp_t} S = G ∩ G ⊥ t be the stabilizer label space, and set s = dim p S s=\dim_pS s = dim p S and h = dim p G h=\dim_pG h = dim p G . By (8), Λ = log q ( K R ) = n − s / m \Lambda=\log_q(KR)=n-s/m Λ = log q ( K R ) = n − s / m . When K = q k K=q^k K = q k and R = q g R=q^g R = q g for nonnegative integers k , g k,g k , g , the parameters of Q Q Q are [ [ n , k , g , d ] ] q [[n,k,g,d]]_q [[ n , k , g , d ] ] q and Λ = k + g \Lambda=k+g Λ = k + g . Every repair set satisfies δ ≤ ∣ Γ ∣ ≤ r + δ − 1 \delta\le|\Gamma|\le r+\delta-1 δ ≤ ∣Γ∣ ≤ r + δ − 1 , and d ≥ δ d\ge\delta d ≥ δ by Corollary III.10.
A Redundancy Bound for Disjoint Repair Sets
We apply Theorem III.8 to pairwise disjoint repair sets to obtain the following bound on Λ \Lambda Λ .
Proposition IV.1. Let M ≥ 1 M \ge1 M ≥ 1 be an integer, and let Γ 1 , … , Γ M \Gamma_1,\ldots,\Gamma_M Γ 1 , … , Γ M be pairwise disjoint repair sets of Q Q Q . For each j ∈ [ M ] j \in[M] j ∈ [ M ] , let E j E_j E j be a nonempty subset of Γ j \Gamma_j Γ j with ∣ E j ∣ < δ |E_j| < \delta ∣ E j ∣ < δ . Then
Λ ≤ n − ∑ j = 1 M ( 2 ∣ E j ∣ − dim p G E j m ) . (30) \Lambda\le n-\sum_{j=1}^{M}\left(2|E_j|-\frac{\dim_{p}G_{E_j}}{m}\right). \tag*{(30)} Λ ≤ n − j = 1 ∑ M ( 2∣ E j ∣ − m dim p G E j ) . (30) Moreover, for every j ∈ [ M ] j \in[M] j ∈ [ M ] ,
r + δ − 1 ≥ ∣ Γ j ∣ ≥ 2 ∣ E j ∣ − dim p G E j m . (31) r+\delta-1 \ge|\Gamma_j| \ge2|E_j|-\frac{\dim_{p}G_{E_j}}{m}. \tag*{(31)} r + δ − 1 ≥ ∣ Γ j ∣ ≥ 2∣ E j ∣ − m dim p G E j . (31)
Proof. Since the repair sets are pairwise disjoint, the stabilizer label subspaces S Γ j S_{\Gamma_j} S Γ j supported in Γ j \Gamma_j Γ j form a direct sum in S S S . Hence
⨁ j = 1 M S Γ j ⊆ S , ∑ j = 1 M dim p S Γ j ≤ s . \bigoplus_{j=1}^{M}S_{\Gamma_j}\subseteq S,\qquad\sum_{j=1}^{M}\dim_{p}S_{\Gamma_j}\le s. j = 1 ⨁ M S Γ j ⊆ S , j = 1 ∑ M dim p S Γ j ≤ s .
Since ∣ E j ∣ < δ |E_j|<\delta ∣ E j ∣ < δ , condition 3) of Theorem III.8 implies
2 m ∣ E j ∣ − dim p G E j = dim p π E j ( S Γ j ) ≤ dim p S Γ j . (32) 2m|E_j|-\dim_{p}G_{E_j}=\dim_{p}\pi_{E_j}(S_{\Gamma_j})\le\dim_{p}S_{\Gamma_j}. \tag*{(32)} 2 m ∣ E j ∣ − dim p G E j = dim p π E j ( S Γ j ) ≤ dim p S Γ j . (32)
Then
∑ j = 1 M ( 2 m ∣ E j ∣ − dim p G E j ) ≤ ∑ j = 1 M dim p S Γ j ≤ s = m ( n − Λ ) , \sum_{j=1}^{M}\left(2m|E_j|-\dim_{p}G_{E_j}\right)\le\sum_{j=1}^{M}\dim_{p}S_{\Gamma_j}
\le s=m(n-\Lambda), j = 1 ∑ M ( 2 m ∣ E j ∣ − dim p G E j ) ≤ j = 1 ∑ M dim p S Γ j ≤ s = m ( n − Λ ) ,
which proves (29).
The space W Γ j W_{\Gamma_j} W Γ j of Pauli labels supported in Γ j \Gamma_j Γ j is nondegenerate under the trace-symplectic form and has F p \mathbb{F}_p F p -dimension 2 m ∣ Γ j ∣ 2m|\Gamma_j| 2 m ∣ Γ j ∣ . Since S Γ j S_{\Gamma_j} S Γ j is isotropic, it follows from (5) that
dim p S Γ j ≤ dim p ( W Γ j ∩ S Γ j ⊥ ) = 2 m ∣ Γ j ∣ − dim p S Γ j , \dim_{p}S_{\Gamma_j}\le\dim_{p}(W_{\Gamma_j}\cap S_{\Gamma_j}^{\perp})=2m|\Gamma_j|-\dim_{p}S_{\Gamma_j}, dim p S Γ j ≤ dim p ( W Γ j ∩ S Γ j ⊥ ) = 2 m ∣ Γ j ∣ − dim p S Γ j ,
which implies that dim p S Γ j ≤ m ∣ Γ j ∣ \dim_{p}S_{\Gamma_j}\le m|\Gamma_j| dim p S Γ j ≤ m ∣ Γ j ∣ . Combining this inequality with (31), we obtain
2 m ∣ E j ∣ − dim p G E j ≤ dim p S Γ j ≤ m ∣ Γ j ∣ . 2m|E_j|-\dim_{p}G_{E_j}\le\dim_{p}S_{\Gamma_j}\le m|\Gamma_j|. 2 m ∣ E j ∣ − dim p G E j ≤ dim p S Γ j ≤ m ∣ Γ j ∣.
Dividing by m m m and applying ∣ Γ j ∣ ≤ r + δ − 1 |\Gamma_j|\le r+\delta-1 ∣ Γ j ∣ ≤ r + δ − 1 , we obtain (30), which completes the proof.
Bounds from Supported Stabilizer and Gauge Subspaces
For two disjoint globally correctable erased sets, we derive bounds on K K K and R R R from their supported gauge spaces and the stabilizers on the remaining coordinates. For E ⊆ [ n ] E\subseteq[n] E ⊆ [ n ] , let S E S_E S E denote the subspace of stabilizer labels supported in E E E . Since S E ⊆ G E S_E\subseteq G_E S E ⊆ G E , define
β ( E ) = dim p G E − dim p S E = dim p ( G E / S E ) . (34) \beta(E)=\dim_{p}G_E-\dim_{p}S_E=\dim_{p}(G_E/S_E). \tag*{(34)} β ( E ) = dim p G E − dim p S E = dim p ( G E / S E ) . (34)
Theorem IV.3. Let E 1 , E 2 ⊆ [ n ] E_1,E_2\subseteq[n] E 1 , E 2 ⊆ [ n ] be disjoint sets such that the erasure of each E i E_i E i is correctable with access to all surviving coordinates in E i c E_i^c E i c . Let T = [ n ] ∖ ( E 1 ∪ E 2 ) T=[n]\setminus(E_1\cup E_2) T = [ n ] ∖ ( E 1 ∪ E 2 ) . Then
log q K + log q R ≤ ∣ T ∣ − dim p S T m + β ( E 1 ) + β ( E 2 ) 2 m , (35) \log_q K+\log_q R\le|T|-\frac{\dim_{p}S_T}{m}+\frac{\beta(E_1)+\beta(E_2)}{2m}, \tag*{(35)} log q K + log q R ≤ ∣ T ∣ − m dim p S T + 2 m β ( E 1 ) + β ( E 2 ) , (35) where S T S_T S T denotes the subspace of stabilizer labels supported in T T T . Moreover, β ( E 1 ) + β ( E 2 ) ≤ 2 log p R \beta(E_1)+\beta(E_2)\le2\log_p R β ( E 1 ) + β ( E 2 ) ≤ 2 log p R and
log q K ≤ ∣ T ∣ − dim p S T m . (36) \log_q K\le|T|-\frac{\dim_{p}S_T}{m}. \tag*{(36)} log q K ≤ ∣ T ∣ − m dim p S T . (36)
Proof. Let s i = dim p S E i s_i=\dim_{p}S_{E_i} s i = dim p S E i and b i = dim p G E i b_i=\dim_{p}G_{E_i} b i = dim p G E i for i = 1 , 2 i=1,2 i = 1 , 2 , and let s T = dim p S T s_T=\dim_{p}S_T s T = dim p S T . The map π E 1 ∣ S \pi_{E_1}|_S π E 1 ∣ S is the coordinate projection onto E 1 E_1 E 1 restricted to S S S . Since E 1 c = E 2 ∪ T E_1^c=E_2\cup T E 1 c = E 2 ∪ T , its kernel consists of the stabilizer labels supported in E 2 ∪ T E_2\cup T E 2 ∪ T . Thus ker ( π E 1 ∣ S ) = S E 2 ∪ T \ker(\pi_{E_1}|_S)=S_{E_2\cup T} ker ( π E 1 ∣ S ) = S E 2 ∪ T . Since S E 2 , S T ≤ S E 2 ∪ T S_{E_2},S_T\le S_{E_2\cup T} S E 2 , S T ≤ S E 2 ∪ T and S E 2 ∩ S T = S E 2 ∩ T = { 0 } S_{E_2}\cap S_T=S_{E_2\cap T}=\{0\} S E 2 ∩ S T = S E 2 ∩ T = { 0 } , we have S E 2 ⊕ S T ⊆ S E 2 ∪ T S_{E_2}\oplus S_T\subseteq S_{E_2\cup T} S E 2 ⊕ S T ⊆ S E 2 ∪ T . By (15), dim p π E 1 ( S ) = 2 m ∣ E 1 ∣ − b 1 \dim_p\pi_{E_1}(S)=2m|E_1|-b_1 dim p π E 1 ( S ) = 2 m ∣ E 1 ∣ − b 1 , and by rank-nullity (1),
s = dim p π E 1 ( S ) + dim p S E 2 ∪ T ≥ 2 m ∣ E 1 ∣ − b 1 + s 2 + s T . s=\dim_p\pi_{E_1}(S)+\dim_p S_{E_2\cup T}\ge2m|E_1|-b_1+s_2+s_T. s = dim p π E 1 ( S ) + dim p S E 2 ∪ T ≥ 2 m ∣ E 1 ∣ − b 1 + s 2 + s T .
Similarly, for E 2 E_2 E 2 , we obtain s ≥ 2 m ∣ E 2 ∣ − b 2 + s 1 + s T s\ge2m|E_2|-b_2+s_1+s_T s ≥ 2 m ∣ E 2 ∣ − b 2 + s 1 + s T . Then
2 s ≥ 2 m ( ∣ E 1 ∣ + ∣ E 2 ∣ ) − ( b 1 − s 1 ) − ( b 2 − s 2 ) + 2 s T = 2 m ( n − ∣ T ∣ ) − β ( E 1 ) − β ( E 2 ) + 2 s T , \begin{aligned}
2s\ge2m(|E_1|+|E_2|)-(b_1-s_1)-(b_2-s_2)+2s_T \\
&=2m(n-|T|)-\beta(E_1)-\beta(E_2)+2s_T,
\end{aligned} 2 s ≥ 2 m ( ∣ E 1 ∣ + ∣ E 2 ∣ ) − ( b 1 − s 1 ) − ( b 2 − s 2 ) + 2 s T = 2 m ( n − ∣ T ∣ ) − β ( E 1 ) − β ( E 2 ) + 2 s T ,
which implies that
n − s m ≤ ∣ T ∣ − dim p S T m + β ( E 1 ) + β ( E 2 ) 2 m . n-\frac{s}{m}\le|T|-\frac{\dim_p S_T}{m}+\frac{\beta(E_1)+\beta(E_2)}{2m}. n − m s ≤ ∣ T ∣ − m dim p S T + 2 m β ( E 1 ) + β ( E 2 ) .
By (8), we have log q ( K R ) = log p m ( p m n − s ) = n − s m \log_q(KR)=\log_{p^m}(p^{mn-s})=n-\frac{s}{m} log q ( K R ) = log p m ( p mn − s ) = n − m s . Hence (34) holds.
Let ϑ : G → G / S \vartheta:G\to G/S ϑ : G → G / S be the quotient map, ϑ ( g ) = g + S \vartheta(g)=g+S ϑ ( g ) = g + S . For i = 1 , 2 i=1,2 i = 1 , 2 , let M i = ( G E i + S ) / S ≤ G / S M_i=(G_{E_i}+S)/S\le G/S M i = ( G E i + S ) / S ≤ G / S be the image ϑ ( G E i ) \vartheta(G_{E_i}) ϑ ( G E i ) of G E i G_{E_i} G E i in G / S G/S G / S . The restriction ϑ ∣ G E i \vartheta|_{G_{E_i}} ϑ ∣ G E i has kernel
ker ( ϑ ∣ G E i ) = G E i ∩ S = ( G ∩ W E i ) ∩ S = S ∩ W E i = S E i . \ker(\vartheta|_{G_{E_i}})=G_{E_i}\cap S=(G\cap W_{E_i})\cap S=S\cap W_{E_i}=S_{E_i}. ker ( ϑ ∣ G E i ) = G E i ∩ S = ( G ∩ W E i ) ∩ S = S ∩ W E i = S E i .
By rank-nullity (1), dim p M i = b i − s i = β ( E i ) \dim_p M_i=b_i-s_i=\beta(E_i) dim p M i = b i − s i = β ( E i ) . Since E 1 ∩ E 2 = ∅ E_1\cap E_2=\varnothing E 1 ∩ E 2 = ∅ , we have [ G E 1 , G E 2 ] t = 0 [G_{E_1},G_{E_2}]_t=0 [ G E 1 , G E 2 ] t = 0 , and hence [ M 1 , M 2 ] t = 0 [M_1,M_2]_t=0 [ M 1 , M 2 ] t = 0 .
Let M 1 ⊥ t M_1^{\perp_t} M 1 ⊥ t denote the orthogonal complement of M 1 M_1 M 1 in G / S G/S G / S with respect to the induced trace-symplectic form. Since G / S G/S G / S is nondegenerate, it follows from the orthogonal-complement dimension formula in (5) that
dim p M 2 ≤ dim p M 1 ⊥ t = dim p ( G / S ) − dim p M 1 . \dim_p M_2\le\dim_p M_1^{\perp_t}=\dim_p(G/S)-\dim_p M_1. dim p M 2 ≤ dim p M 1 ⊥ t = dim p ( G / S ) − dim p M 1 .
Since S ⊆ G S\subseteq G S ⊆ G , we have
dim p ( G / S ) = dim p G − dim p S = h − s . \dim_p(G/S)=\dim_p G-\dim_p S=h-s. dim p ( G / S ) = dim p G − dim p S = h − s .
By (8), log p R = h − s 2 \log_p R=\frac{h-s}{2} log p R = 2 h − s . Thus β ( E 1 ) + β ( E 2 ) ≤ 2 log p R \beta(E_1)+\beta(E_2)\le2\log_p R β ( E 1 ) + β ( E 2 ) ≤ 2 log p R . Substituting this estimate into (34), we obtain
log q K + log q R ≤ ∣ T ∣ − s T m + 2 log p R 2 m = ∣ T ∣ − s T m + m log p m R m , \log_q K+\log_q R\le|T|-\frac{s_T}{m}+\frac{2\log_p R}{2m}=|T|-\frac{s_T}{m}+\frac{m\log_{p^m}R}{m}, log q K + log q R ≤ ∣ T ∣ − m s T + 2 m 2 log p R = ∣ T ∣ − m s T + m m log p m R ,
which proves (35).
Remark IV.4. The condition K > 1 K>1 K > 1 ensures that 2 ( d − 1 ) ≤ n 2(d-1)\le n 2 ( d − 1 ) ≤ n . Indeed, if 2 ( d − 1 ) > n 2(d-1)>n 2 ( d − 1 ) > n , then d − 1 ≥ ⌈ n / 2 ⌉ d-1\ge\lceil n/2\rceil d − 1 ≥ ⌈ n /2 ⌉ . A partition of [ n ] [n] [ n ] into sets of sizes ⌊ n / 2 ⌋ \lfloor n/2\rfloor ⌊ n /2 ⌋ and ⌈ n / 2 ⌉ \lceil n/2\rceil ⌈ n /2 ⌉ would then consist of two globally correctable sets. Since T = ∅ T=\varnothing T = ∅ and S ∅ = { 0 } S_{\varnothing}=\{0\} S ∅ = { 0 } , equation (35) implies log q K ≤ 0 \log_q K\le0 log q K ≤ 0 , a contradiction. Therefore, we may choose disjoint sets E 1 , E 2 ⊆ [ n ] E_1,E_2\subseteq[n] E 1 , E 2 ⊆ [ n ] with ∣ E 1 ∣ = ∣ E 2 ∣ = d − 1 |E_1|=|E_2|=d-1 ∣ E 1 ∣ = ∣ E 2 ∣ = d − 1 . Both are globally correctable by Lemma II.20 and (11). Let T = [ n ] ∖ ( E 1 ∪ E 2 ) T=[n]\setminus(E_1\cup E_2) T = [ n ] ∖ ( E 1 ∪ E 2 ) . Since ∣ T ∣ = n − 2 ( d − 1 ) |T|=n-2(d-1) ∣ T ∣ = n − 2 ( d − 1 ) , it follows from Theorem IV.3 that
Λ + 2 ( d − 1 ) ≤ n − dim p S T m + β ( E 1 ) + β ( E 2 ) 2 m . \Lambda+2(d-1)\le n-\frac{\dim_p S_T}{m}+\frac{\beta(E_1)+\beta(E_2)}{2m}. Λ + 2 ( d − 1 ) ≤ n − m dim p S T + 2 m β ( E 1 ) + β ( E 2 ) .
If the code is pure, then G E 1 = G E 2 = { 0 } G_{E_1}=G_{E_2}=\{0\} G E 1 = G E 2 = { 0 } by Definition II.18, and hence β ( E 1 ) = β ( E 2 ) = 0 \beta(E_1)=\beta(E_2)=0 β ( E 1 ) = β ( E 2 ) = 0 . Since dim p S T ≥ 0 \dim_p S_T\ge0 dim p S T ≥ 0 , we obtain Λ + 2 ( d − 1 ) ≤ n \Lambda+2(d-1)\le n Λ + 2 ( d − 1 ) ≤ n , which is the pure subsystem Singleton bound K R ≤ q n − 2 d + 2 KR\le q^{n-2d+2} K R ≤ q n − 2 d + 2 of [2 Theorem 11] .
As in Proposition IV.1, we use pairwise disjoint repair sets in T T T to obtain a lower bound on dim p S T \dim_p S_T dim p S T . Substituting this bound into Theorem IV.3, we obtain the following corollary.
Corollary IV.5. Let E 1 , E 2 ⊆ [ n ] E_1,E_2\subseteq[n] E 1 , E 2 ⊆ [ n ] be disjoint sets of size d − 1 d-1 d − 1 , and let T = [ n ] ∖ ( E 1 ∪ E 2 ) T=[n]\setminus(E_1\cup E_2) T = [ n ] ∖ ( E 1 ∪ E 2 ) . Let M ∈ Z ≥ 0 M\in\mathbb{Z}_{\ge0} M ∈ Z ≥ 0 , and let Γ 1 , … , Γ M ⊆ T \Gamma_1,\ldots,\Gamma_M\subseteq T Γ 1 , … , Γ M ⊆ T be pairwise disjoint repair sets of Q Q Q . For each j ∈ [ M ] j\in[M] j ∈ [ M ] , let F j ⊆ Γ j F_j\subseteq\Gamma_j F j ⊆ Γ j with ∣ F j ∣ = δ − 1 |F_j|=\delta-1 ∣ F j ∣ = δ − 1 . Then
Λ + 2 ( d − 1 ) + ∑ j = 1 M ( 2 ( δ − 1 ) − dim p G F j m ) ≤ n + β ( E 1 ) + β ( E 2 ) 2 m . (37) \Lambda+2(d-1)+\sum_{j=1}^{M}\left(2(\delta-1)-\frac{\dim_p G_{F_j}}{m}\right)\le n+\frac{\beta(E_1)+\beta(E_2)}{2m}. \tag*{(37)} Λ + 2 ( d − 1 ) + j = 1 ∑ M ( 2 ( δ − 1 ) − m dim p G F j ) ≤ n + 2 m β ( E 1 ) + β ( E 2 ) . (37) where G F j G_{F_j} G F j denotes the subspace of gauge labels supported in F j F_j F j .
Proof. Since Γ j ⊆ T \Gamma_j\subseteq T Γ j ⊆ T , the subspace of S T S_T S T consisting of labels supported in Γ j \Gamma_j Γ j satisfies
( S T ) Γ j = S T ∩ W Γ j = S ∩ W T ∩ W Γ j = S ∩ W T ∩ Γ j = S ∩ W Γ j = S Γ j . \begin{aligned}
(S_T)_{\Gamma_j}=S_T\cap W_{\Gamma_j} \\
&=S\cap W_T\cap W_{\Gamma_j} \\
&=S\cap W_{T\cap\Gamma_j} \\
&=S\cap W_{\Gamma_j} \\
&=S_{\Gamma_j}.
\end{aligned} ( S T ) Γ j = S T ∩ W Γ j = S ∩ W T ∩ W Γ j = S ∩ W T ∩ Γ j = S ∩ W Γ j = S Γ j .
For F j ⊆ Γ j F_j \subseteq\Gamma_j F j ⊆ Γ j with 0 < ∣ F j ∣ = δ − 1 < δ 0<|F_j|=\delta-1<\delta 0 < ∣ F j ∣ = δ − 1 < δ , condition 3) of Theorem III.8 implies that dim p π F j ( S Γ j ) = 2 m ∣ F j ∣ − dim p G F j \dim_p \pi_{F_j}(\mathcal{S}_{\Gamma_j})=2m|F_j|-\dim_p G_{F_j} dim p π F j ( S Γ j ) = 2 m ∣ F j ∣ − dim p G F j . Since Γ i ∩ Γ j = ∅ \Gamma_i\cap\Gamma_j=\emptyset Γ i ∩ Γ j = ∅ for i ≠ j i\ne j i = j and Γ j ⊆ T \Gamma_j\subseteq T Γ j ⊆ T for every j ∈ [ M ] j\in[M] j ∈ [ M ] , we have
⨁ j = 1 M S Γ j ⊆ S T . \bigoplus_{j=1}^{M}\mathcal{S}_{\Gamma_j}\subseteq\mathcal{S}_T. j = 1 ⨁ M S Γ j ⊆ S T .
Hence
dim p S T ≥ ∑ j = 1 M dim p S Γ j ≥ ∑ j = 1 M dim p π F j ( S Γ j ) = ∑ j = 1 M ( 2 m ( δ − 1 ) − dim p G F j ) , \begin{aligned}
\dim_p\mathcal{S}_T\ge\sum_{j=1}^{M}\dim_p\mathcal{S}_{\Gamma_j} \\
&\ge\sum_{j=1}^{M}\dim_p\pi_{F_j}(\mathcal{S}_{\Gamma_j}) \\
&=\sum_{j=1}^{M}\left(2m(\delta-1)-\dim_p G_{F_j}\right),
\end{aligned} dim p S T ≥ j = 1 ∑ M dim p S Γ j ≥ j = 1 ∑ M dim p π F j ( S Γ j ) = j = 1 ∑ M ( 2 m ( δ − 1 ) − dim p G F j ) ,
which implies that
dim p S T m ≥ ∑ j = 1 M ( 2 ( δ − 1 ) − dim p G F j m ) . \frac{\dim_p\mathcal{S}_T}{m}\ge\sum_{j=1}^{M}\left(2(\delta-1)-\frac{\dim_p G_{F_j}}{m}\right). m dim p S T ≥ j = 1 ∑ M ( 2 ( δ − 1 ) − m dim p G F j ) .
Since ∣ T ∣ = n − 2 ( d − 1 ) |T|=n-2(d-1) ∣ T ∣ = n − 2 ( d − 1 ) , it follows from (34) that
log q K + log q R ≤ n − 2 ( d − 1 ) − dim p S T m + β ( E 1 ) + β ( E 2 ) 2 m ≤ n − 2 ( d − 1 ) − ∑ j = 1 M ( 2 ( δ − 1 ) − dim p G F j m ) + β ( E 1 ) + β ( E 2 ) 2 m , \begin{aligned}
\log_q K+\log_q R
\le n-2(d-1)-\frac{\dim_p\mathcal{S}_T}{m}+\frac{\beta(E_1)+\beta(E_2)}{2m} \\
&\le n-2(d-1)-\sum_{j=1}^{M}\left(2(\delta-1)-\frac{\dim_p G_{F_j}}{m}\right)+\frac{\beta(E_1)+\beta(E_2)}{2m},
\end{aligned} log q K + log q R ≤ n − 2 ( d − 1 ) − m dim p S T + 2 m β ( E 1 ) + β ( E 2 ) ≤ n − 2 ( d − 1 ) − j = 1 ∑ M ( 2 ( δ − 1 ) − m dim p G F j ) + 2 m β ( E 1 ) + β ( E 2 ) ,
which proves (36).
Remark IV.6. For F q \mathbb{F}_q F q -linear gauge spaces, the global subsystem Singleton bound k + g ≤ n − 2 ( d − 1 ) k+g\le n-2(d-1) k + g ≤ n − 2 ( d − 1 ) also applies [22 Theorem 3.6] . The bound in Corollary IV.5 depends explicitly on the chosen pairwise disjoint repair sets and supported gauge spaces and requires only F p \mathbb{F}_p F p -linearity.
A Mixed Counting Bound via Gauge Equivalence Classes
Local recovery corrects fewer than δ \delta δ known erasures within each of several pairwise disjoint repair sets. We consider these erasures together with errors at unknown positions outside the union of the repair sets, where the local recovery maps act trivially.
Let b ∈ Z ≥ 0 b\in\mathbb{Z}_{\ge0} b ∈ Z ≥ 0 , and let Γ 1 , … , Γ b \Gamma_1,\ldots,\Gamma_b Γ 1 , … , Γ b be repair sets of Q \mathcal{Q} Q satisfying Γ i ∩ Γ j = ∅ \Gamma_i\cap\Gamma_j=\emptyset Γ i ∩ Γ j = ∅ for 1 ≤ i < j ≤ b 1\le i<j\le b 1 ≤ i < j ≤ b . For each j ∈ [ b ] j\in[b] j ∈ [ b ] , let E j ⊆ Γ j E_j\subseteq\Gamma_j E j ⊆ Γ j be a known erased set with e j = ∣ E j ∣ e_j=|E_j| e j = ∣ E j ∣ and 0 ≤ e j ≤ δ − 1 0\le e_j\le\delta-1 0 ≤ e j ≤ δ − 1 . Let
T = ⋃ j = 1 b Γ j , U = T c , N = ∣ U ∣ . T=\bigcup_{j=1}^{b}\Gamma_j,\quad U=T^c,\quad N=|U|. T = j = 1 ⋃ b Γ j , U = T c , N = ∣ U ∣.
Since the repair sets are pairwise disjoint and ∣ Γ j ∣ ≤ r + δ − 1 |\Gamma_j|\le r+\delta-1 ∣ Γ j ∣ ≤ r + δ − 1 , we have
N = n − ∑ j = 1 b ∣ Γ j ∣ ≥ n − b ( r + δ − 1 ) . N=n-\sum_{j=1}^{b}|\Gamma_j|\ge n-b(r+\delta-1). N = n − j = 1 ∑ b ∣ Γ j ∣ ≥ n − b ( r + δ − 1 ) .
Let t ∈ Z ≥ 0 t\in\mathbb{Z}_{\ge0} t ∈ Z ≥ 0 satisfy 2 t < d 2t<d 2 t < d . Define the set of Pauli labels for errors affecting at most t t t qudits at unknown locations in U U U by
B U ( t ) = { u ∈ W U : swt ( u ) ≤ t } . \mathcal{B}_U(t)=\{u\in W_U:\operatorname{swt}(u)\le t\}. B U ( t ) = { u ∈ W U : swt ( u ) ≤ t } .
For each j ∈ [ b ] j\in[b] j ∈ [ b ] , we replace the erased qudits in E j E_j E j with the maximally mixed state I E j / q e j I_{E_j}/q^{e_j} I E j / q e j . By (21), erasure followed by this replacement is equivalent to applying a uniformly random Pauli operator supported in E j E_j E j . We therefore consider the error-label family
E = { ∑ j = 1 b a j + u : a j ∈ W E j , u ∈ B U ( t ) } , \mathcal{E}=\left\{\sum_{j=1}^{b}a_j+u:a_j\in W_{E_j},\ u\in\mathcal{B}_U(t)\right\}, E = { j = 1 ∑ b a j + u : a j ∈ W E j , u ∈ B U ( t ) } ,
where a j a_j a j ranges over all Pauli labels supported in E j E_j E j .
Two Pauli labels are gauge-equivalent if their difference belongs to G G G . For D ∈ { E 1 , … , E b , U } D \in\{E_1,\ldots,E_b,U\} D ∈ { E 1 , … , E b , U } , let G D = G ∩ W D G_D=G\cap W_D G D = G ∩ W D . Then, for w , w ′ ∈ W D w,w' \in W_D w , w ′ ∈ W D ,
w − w ′ ∈ G ⟺ w − w ′ ∈ G D ⟺ w + G D = w ′ + G D . w-w' \in G \Longleftrightarrow w-w' \in G_D \Longleftrightarrow w+G_D=w'+G_D. w − w ′ ∈ G ⟺ w − w ′ ∈ G D ⟺ w + G D = w ′ + G D .
For each j ∈ [ b ] j \in[b] j ∈ [ b ] , let
ν j = 2 m e j − dim p G E j = dim p ( W E j / G E j ) , ν = ∑ j = 1 b ν j . \nu_j=2me_j-\dim_p G_{E_j}=\dim_p(W_{E_j}/G_{E_j}),\qquad\nu=\sum_{j=1}^{b}\nu_j. ν j = 2 m e j − dim p G E j = dim p ( W E j / G E j ) , ν = j = 1 ∑ b ν j .
Define the number of distinct gauge-equivalence classes represented in B U ( t ) \mathcal{B}_U(t) B U ( t ) by
M U ( t ) = ∣ { u + G U : u ∈ B U ( t ) } ∣ . M_U(t)=\left|\{u+G_U:u\in\mathcal{B}_U(t)\}\right|. M U ( t ) = ∣ { u + G U : u ∈ B U ( t )} ∣ .
The following bound compares the number of gauge-equivalence classes represented in E \mathcal{E} E with the number of available stabilizer syndromes.
Theorem IV.7. Under the assumptions and notation above,
p ν M U ( t ) ≤ q n − Λ . (38) p^\nu M_U(t)\le q^{n-\Lambda}. \tag*{(38)} p ν M U ( t ) ≤ q n − Λ . (38) Equivalently,
Λ + ∑ j = 1 b ( 2 e j − dim p G E j m ) + log q M U ( t ) ≤ n . (39) \Lambda+\sum_{j=1}^{b}\left(2e_j-\frac{\dim_p G_{E_j}}{m}\right)+\log_q M_U(t)\le n. \tag*{(39)} Λ + j = 1 ∑ b ( 2 e j − m dim p G E j ) + log q M U ( t ) ≤ n . (39)
Proof. Let s 1 , … , s s s_1,\ldots,s_s s 1 , … , s s be an F p \mathbb{F}_p F p -basis of S S S , with s = m ( n − Λ ) s=m(n-\Lambda) s = m ( n − Λ ) . For each Pauli label v ∈ W v\in W v ∈ W , define the syndrome map σ : W → F p s \sigma:W\to\mathbb{F}_p^s σ : W → F p s by
σ ( v ) = ( [ v , s i ] t ) i = 1 s . \sigma(v)=([v,s_i]_t)_{i=1}^{s}. σ ( v ) = ([ v , s i ] t ) i = 1 s .
It follows from the definition of S ⊥ t S^{\perp_t} S ⊥ t and (5) and (1) that
ker σ = S ⊥ t , rank p σ = 2 m n − dim p S ⊥ t = s . \ker\sigma=S^{\perp_t},\qquad\operatorname{rank}_p\sigma=2mn-\dim_p S^{\perp_t}=s. ker σ = S ⊥ t , rank p σ = 2 mn − dim p S ⊥ t = s .
Define
Θ : ( ∏ j = 1 b W E j / G E j ) × { u + G U : u ∈ B U ( t ) } ⟶ σ ( E ) \Theta:\left(\prod_{j=1}^{b}W_{E_j}/G_{E_j}\right)\times\{u+G_U:u\in\mathcal{B}_U(t)\}\longrightarrow\sigma(\mathcal{E}) Θ : ( j = 1 ∏ b W E j / G E j ) × { u + G U : u ∈ B U ( t )} ⟶ σ ( E )
by
Θ ( a 1 + G E 1 , … , a b + G E b , u + G U ) = σ ( ∑ j = 1 b a j + u ) , \Theta(a_1+G_{E_1},\ldots,a_b+G_{E_b},u+G_U)
=\sigma\left(\sum_{j=1}^{b}a_j+u\right), Θ ( a 1 + G E 1 , … , a b + G E b , u + G U ) = σ ( j = 1 ∑ b a j + u ) ,
where the representatives satisfy a j ∈ W E j a_j\in W_{E_j} a j ∈ W E j and u ∈ B U ( t ) u\in\mathcal{B}_U(t) u ∈ B U ( t ) . Suppose that a j + G E j = a j ′ + G E j a_j+G_{E_j}=a'_j+G_{E_j} a j + G E j = a j ′ + G E j for every j j j and u + G U = u ′ + G U u+G_U=u'+G_U u + G U = u ′ + G U . Then a j − a j ′ ∈ G E j a_j-a'_j\in G_{E_j} a j − a j ′ ∈ G E j and u − u ′ ∈ G U u-u'\in G_U u − u ′ ∈ G U , so
∑ j = 1 b ( a j − a j ′ ) + ( u − u ′ ) ∈ G ⊆ ker σ . \sum_{j=1}^{b}(a_j-a'_j)+(u-u')\in G\subseteq\ker\sigma. j = 1 ∑ b ( a j − a j ′ ) + ( u − u ′ ) ∈ G ⊆ ker σ .
By linearity of σ \sigma σ ,
σ ( ∑ j = 1 b a j + u ) − σ ( ∑ j = 1 b a j ′ + u ′ ) = σ ( ∑ j = 1 b ( a j − a j ′ ) + ( u − u ′ ) ) = 0. \sigma\left(\sum_{j=1}^{b}a_j+u\right)-\sigma\left(\sum_{j=1}^{b}a'_j+u'\right)
=\sigma\left(\sum_{j=1}^{b}(a_j-a'_j)+(u-u')\right)=0. σ ( j = 1 ∑ b a j + u ) − σ ( j = 1 ∑ b a j ′ + u ′ ) = σ ( j = 1 ∑ b ( a j − a j ′ ) + ( u − u ′ ) ) = 0.
Thus Θ \Theta Θ does not depend on the choice of representatives, and Θ \Theta Θ is well defined.
To prove injectivity, let x = ∑ j a j + u x=\sum_j a_j+u x = ∑ j a j + u and x ′ = ∑ j a j ′ + u ′ x'=\sum_j a'_j+u' x ′ = ∑ j a j ′ + u ′ be elements of E \mathcal{E} E with the same syndrome, and set v = x − x ′ v=x-x' v = x − x ′ . By the linearity of σ \sigma σ , we have σ ( v ) = σ ( x − x ′ ) = σ ( x ) − σ ( x ′ ) = 0 \sigma(v)=\sigma(x-x')=\sigma(x)-\sigma(x')=0 σ ( v ) = σ ( x − x ′ ) = σ ( x ) − σ ( x ′ ) = 0 , and hence v ∈ ker σ = S ⊥ t v\in\ker\sigma=S^{\perp_t} v ∈ ker σ = S ⊥ t . Fix j ∈ [ b ] j\in[b] j ∈ [ b ] . For every z ∈ S Γ j z\in S_{\Gamma_j} z ∈ S Γ j , we have
supp ( z ) ⊆ Γ j , supp ( u − u ′ ) ⊆ U , supp ( a i − a i ′ ) ⊆ E i ⊆ Γ i . \operatorname{supp}(z)\subseteq\Gamma_j,\qquad\operatorname{supp}(u-u')\subseteq U,\qquad\operatorname{supp}(a_i-a'_i)\subseteq E_i\subseteq\Gamma_i. supp ( z ) ⊆ Γ j , supp ( u − u ′ ) ⊆ U , supp ( a i − a i ′ ) ⊆ E i ⊆ Γ i .
Since Γ j ∩ U = ∅ \Gamma_j \cap U = \varnothing Γ j ∩ U = ∅ and Γ j ∩ Γ i = ∅ \Gamma_j \cap\Gamma_i = \varnothing Γ j ∩ Γ i = ∅ for i ≠ j i \ne j i = j , [ u − u ′ , z ] t = 0 [u-u',z]_t=0 [ u − u ′ , z ] t = 0 and [ a i − a i ′ , z ] t = 0 [a_i-a_i',z]_t=0 [ a i − a i ′ , z ] t = 0 . Since v ∈ S ⊥ t v \in S^{\perp t} v ∈ S ⊥ t and z ∈ S Γ j ⊆ S z \in S_{\Gamma_j} \subseteq S z ∈ S Γ j ⊆ S , we obtain
0 = [ v , z ] t = ∑ i = 1 b [ a i − a i ′ , z ] t + [ u − u ′ , z ] t = [ a j − a j ′ , z ] t . \begin{aligned}
0=[v,z]_t=\sum_{i=1}^{b}[a_i-a_i',z]_t+[u-u',z]_t \\
&=[a_j-a_j',z]_t.
\end{aligned} 0 = [ v , z ] t = i = 1 ∑ b [ a i − a i ′ , z ] t + [ u − u ′ , z ] t = [ a j − a j ′ , z ] t .
Since this holds for every z ∈ S Γ j z \in S_{\Gamma_j} z ∈ S Γ j and a j − a j ′ ∈ W E j a_j-a_j' \in W_{E_j} a j − a j ′ ∈ W E j , it follows from condition 2) of Theorem III.8 that
a j − a j ′ ∈ W E j ∩ S Γ j ⊥ t = G E j , a_j-a_j' \in W_{E_j}\cap S_{\Gamma_j}^{\perp t}=G_{E_j}, a j − a j ′ ∈ W E j ∩ S Γ j ⊥ t = G E j ,
provided that E j ≠ ∅ E_j \ne\varnothing E j = ∅ . For E j = ∅ E_j=\varnothing E j = ∅ , we have a j − a j ′ = 0 ∈ G E j = { 0 } a_j-a_j'=0 \in G_{E_j}=\{0\} a j − a j ′ = 0 ∈ G E j = { 0 } . Since v ∈ S ⊥ t v \in S^{\perp t} v ∈ S ⊥ t and a j − a j ′ ∈ G E j ⊆ G ⊆ S ⊥ t a_j-a_j' \in G_{E_j} \subseteq G \subseteq S^{\perp t} a j − a j ′ ∈ G E j ⊆ G ⊆ S ⊥ t for every j j j , we have
u − u ′ = v − ∑ j = 1 b ( a j − a j ′ ) ∈ S ⊥ t . u-u'=v-\sum_{j=1}^{b}(a_j-a_j')\in S^{\perp t}. u − u ′ = v − j = 1 ∑ b ( a j − a j ′ ) ∈ S ⊥ t .
Moreover, since u , u ′ ∈ B U ( t ) ⊆ W U u,u' \in B_U(t) \subseteq W_U u , u ′ ∈ B U ( t ) ⊆ W U , we have u − u ′ ∈ W U u-u' \in W_U u − u ′ ∈ W U and
swt ( u − u ′ ) ≤ swt ( u ) + swt ( u ′ ) ≤ 2 t < d . \operatorname{swt}(u-u')\le\operatorname{swt}(u)+\operatorname{swt}(u')\le2t<d. swt ( u − u ′ ) ≤ swt ( u ) + swt ( u ′ ) ≤ 2 t < d .
By (11) , u − u ′ ∈ G U u-u' \in G_U u − u ′ ∈ G U . Consequently, for j ∈ [ b ] j \in[b] j ∈ [ b ] ,
a j + G E j = a j ′ + G E j , u + G U = u ′ + G U . a_j+G_{E_j}=a_j'+G_{E_j},\qquad u+G_U=u'+G_U. a j + G E j = a j ′ + G E j , u + G U = u ′ + G U .
Thus equal images under Θ \Theta Θ imply equality of all component cosets, and Θ \Theta Θ is injective.
For any y ∈ σ ( E ) y \in\sigma(\mathcal{E}) y ∈ σ ( E ) , by the definition of E \mathcal{E} E , we can choose labels a j ∈ W E j a_j \in W_{E_j} a j ∈ W E j and u ∈ B U ( t ) u \in B_U(t) u ∈ B U ( t ) such that
y = σ ( ∑ j = 1 b a j + u ) = Θ ( a 1 + G E 1 , … , a b + G E b , u + G U ) . \begin{aligned}
y=\sigma\left(\sum_{j=1}^{b}a_j+u\right) \\
&=\Theta(a_1+G_{E_1},\ldots,a_b+G_{E_b},u+G_U).
\end{aligned} y = σ ( j = 1 ∑ b a j + u ) = Θ ( a 1 + G E 1 , … , a b + G E b , u + G U ) .
Thus Θ \Theta Θ is surjective and therefore bijective. By the definitions of ν j \nu_j ν j and M U ( t ) M_U(t) M U ( t ) , we obtain
∣ σ ( E ) ∣ = ( ∏ j = 1 b ∣ W E j / G E j ∣ ) M U ( t ) = ( ∏ j = 1 b p 2 m e j − dim F p G E j ) M U ( t ) = p ν M U ( t ) ≤ ∣ F p s ∣ = p s , \begin{aligned}
|\sigma(\mathcal{E})|=\left(\prod_{j=1}^{b}|W_{E_j}/G_{E_j}|\right)M_U(t) \\
&=\left(\prod_{j=1}^{b}p^{2m e_j-\dim_{\mathbb{F}_p}G_{E_j}}\right)M_U(t) \\
&=p^\nu M_U(t)\le|F_p^s|=p^s,
\end{aligned} ∣ σ ( E ) ∣ = ( j = 1 ∏ b ∣ W E j / G E j ∣ ) M U ( t ) = ( j = 1 ∏ b p 2 m e j − d i m F p G E j ) M U ( t ) = p ν M U ( t ) ≤ ∣ F p s ∣ = p s ,
where the inequality follows from σ ( E ) ⊆ F p s \sigma(\mathcal{E})\subseteq F_p^s σ ( E ) ⊆ F p s . Since p s = q n − Λ p^s=q^{n-\Lambda} p s = q n − Λ , this proves (38) . We take base-q q q logarithms and use log q p = 1 / m \log_q p=1/m log q p = 1/ m to obtain (39) . ▫
Remark IV.8. The Pauli error family { P x : x ∈ E } \{P_x:x\in\mathcal{E}\} { P x : x ∈ E } is jointly correctable on the logical subsystem. Let P = V V † P=VV^\dagger P = V V † be the projector onto Q \mathcal{Q} Q . For x , x ′ ∈ E x,x'\in\mathcal{E} x , x ′ ∈ E , distinct syndromes imply x ′ − x ∉ S ⊥ t x'-x\notin S^{\perp t} x ′ − x ∈ / S ⊥ t and hence
P P x † P x ′ P = 0. PP_{x}^{\dagger}P_{x'}P=0. P P x † P x ′ P = 0.
For equal syndromes, it follows from the proof above that x ′ − x ∈ G x'-x\in G x ′ − x ∈ G . Equation (10) then shows that V † P x † P x ′ V V^\dagger P_x^\dagger P_{x'}V V † P x † P x ′ V acts only on the gauge subsystem. In both cases,
V † P x † P x ′ V = I A ⊗ M x , x ′ , M x , x ′ ∈ B ( H B ) , V^\dagger P_x^\dagger P_{x'}V=I_A\otimes M_{x,x'},\qquad M_{x,x'}\in\mathcal{B}(\mathcal{H}_B), V † P x † P x ′ V = I A ⊗ M x , x ′ , M x , x ′ ∈ B ( H B ) ,
where M x , x ′ = 0 M_{x,x'}=0 M x , x ′ = 0 for distinct syndromes; for equal syndromes, the scalar Pauli phase is absorbed into M x , x ′ M_{x,x'} M x , x ′ . Thus the error family satisfies the subsystem Knill-Laflamme condition [24 Theorem 3.5] . Recovery of this combined error family may use global syndrome measurements. The prescribed local recovery operations correct the erasures E j E_j E j within their respective repair sets.
Remark IV.9. For t = 0 t=0 t = 0 , we have M U ( 0 ) = 1 M_U(0)=1 M U ( 0 ) = 1 . Thus (39) reduces to the disjoint redundancy bound (30) , with empty erased sets contributing zero. For b = 0 b=0 b = 0 , U = [ n ] U=[n] U = [ n ] and ν = 0 \nu=0 ν = 0 , so (38) is a Hamming-type bound that counts gauge cosets represented by errors of weight at most t t t .
For a pure code, every nonzero gauge label has weight at least d d d by Definition II.18. Since e j < δ ≤ d e_j<\delta\le d e j < δ ≤ d by Corollary III.10, we have G E j = 0 G_{E_j}=0 G E j = 0 . For distinct u , u ′ ∈ B U ( t ) u,u'\in B_U(t) u , u ′ ∈ B U ( t ) , 0 < swt ( u − u ′ ) ≤ 2 t < d 0<\operatorname{swt}(u-u')\le2t<d 0 < swt ( u − u ′ ) ≤ 2 t < d . Purity therefore implies u − u ′ ∉ G U u-u'\notin G_U u − u ′ ∈ / G U , so the labels represent distinct gauge cosets. Let V Q ( N , t ) V_Q(N,t) V Q ( N , t ) denote the volume of a Hamming ball of radius t t t in N N N coordinates over an alphabet of size Q Q Q . Then
M U ( t ) = V q 2 ( N , t ) , V Q ( N , t ) : = ∑ i = 0 min ( t , N ) ( N i ) ( Q − 1 ) i . M_U(t)=V_{q^2}(N,t),\quad V_Q(N,t):=\sum_{i=0}^{\min(t,N)}\binom{N}{i}(Q-1)^i. M U ( t ) = V q 2 ( N , t ) , V Q ( N , t ) := i = 0 ∑ m i n ( t , N ) ( i N ) ( Q − 1 ) i .
For b = 0 b=0 b = 0 and t = ⌊ ( d − 1 ) / 2 ⌋ t=\lfloor(d-1)/2\rfloor t = ⌊( d − 1 ) /2 ⌋ , it follows from (38) that
V q 2 ( n , t ) ≤ q n K R , V_{q^2}(n,t)\leq\frac{q^n}{KR}, V q 2 ( n , t ) ≤ K R q n ,
which is the pure subsystem Hamming bound of [2 Lemma 13] .
Bounds for Pure ( r , δ ) (r,\delta) ( r , δ ) -SSLRCs
In this section, we establish Singleton-type and alphabet-dependent bounds for pure ( r , δ ) (r,\delta) ( r , δ ) -SSLRCs, together with explicit Griesmer, Hamming, and Plotkin bounds. These bounds apply to overlapping repair sets. Throughout this section, let Q Q Q be a pure ( r , δ ) (r,\delta) ( r , δ ) -SSLRC with parameters ( ( n , K , R , d ) ) q ((n,K,R,d))_q (( n , K , R , d ) ) q , where q = p m q=p^m q = p m and K > 1 K>1 K > 1 . Let G G G be its F p \mathbb{F}_p F p -linear gauge label space and S = G ∩ G ⊥ t S=G\cap G^{\perp_t} S = G ∩ G ⊥ t its stabilizer label space. Set s = dim p S s=\dim_p S s = dim p S and Λ = log q ( K R ) \Lambda=\log_q(KR) Λ = log q ( K R ) . For parameters [ [ n , k , g , d ] ] q [[n,k,g,d]]_q [[ n , k , g , d ] ] q , we have Λ = k + g \Lambda=k+g Λ = k + g .
Reduction to Stabilizer Codes and Classical Comparison Codes
By [2 Lemma 10] , a pure ( ( n , K , R , d ) ) q ((n,K,R,d))_q (( n , K , R , d ) ) q SSC can be regarded as a pure ( ( n , K R , d ) ) q ((n,KR,d))_q (( n , K R , d ) ) q stabilizer code.
Proposition V.1. Let Q Q Q be a pure ( r , δ ) (r,\delta) ( r , δ ) -SSLRC with parameters ( ( n , K , R , d ) ) q ((n,K,R,d))_q (( n , K , R , d ) ) q , where K > 1 K>1 K > 1 . The common + 1 +1 + 1 eigenspace of its stabilizer group S S S is an ( r , δ ) (r,\delta) ( r , δ ) -locally recoverable stabilizer code with parameters ( ( n , K R , d ) ) q ((n,KR,d))_q (( n , K R , d ) ) q and the same repair cover as Q Q Q . If Q Q Q has parameters [ [ n , k , g , d ] ] q [[n,k,g,d]]_q [[ n , k , g , d ] ] q , then this stabilizer code has parameters [ [ n , k + g , d ] ] q [[n,k+g,d]]_q [[ n , k + g , d ] ] q .
Proof. Let Q 0 Q_0 Q 0 denote the common + 1 +1 + 1 eigenspace of S S S , regarded as a stabilizer code. The physical code space is unchanged, and the entire space H A ⊗ H B \mathcal{H}_A\otimes\mathcal{H}_B H A ⊗ H B is treated as its logical space. By [2 Lemma 10] , Q 0 Q_0 Q 0 has dimension K R KR K R and minimum distance d d d . In the subsystem formalism, its gauge label space is G 0 = S G_0=S G 0 = S , and its stabilizer label space remains S S S .
Let Γ \Gamma Γ be a repair set in the prescribed repair cover of Q Q Q , and let ∅ ≠ E ⊆ Γ \varnothing\ne E\subseteq\Gamma ∅ = E ⊆ Γ with ∣ E ∣ < δ |E|<\delta ∣ E ∣ < δ . By Corollary III.10, we have ∣ E ∣ < δ ≤ d |E|<\delta\leq d ∣ E ∣ < δ ≤ d . Every label in G E G_E G E has symplectic weight at most ∣ E ∣ < d |E|<d ∣ E ∣ < d , whereas purity requires every nonzero gauge label to have weight at least d d d . Thus G E = { 0 } G_E=\{0\} G E = { 0 } by Definition II.18. Since S ⊆ G S\subseteq G S ⊆ G , we also have S E = { 0 } S_E=\{0\} S E = { 0 } . Since E E E is recoverable within Γ \Gamma Γ for Q Q Q , condition 2) of Theorem III.8 implies that
W E ∩ S Γ ⊥ t = G E = { 0 } = S E = ( G 0 ) E . W_E\cap S_{\Gamma}^{\perp_t}=G_E=\{0\}=S_E=(G_0)_E. W E ∩ S Γ ⊥ t = G E = { 0 } = S E = ( G 0 ) E . By the sufficient direction of the same criterion, E E E is recoverable within Γ \Gamma Γ for Q 0 Q_0 Q 0 . Since Γ \Gamma Γ and E E E were arbitrary and the repair-set sizes are unchanged, Q 0 Q_0 Q 0 has the same ( r , δ ) (r,\delta) ( r , δ ) repair cover as Q Q Q .
To bound K R KR K R , we use the additive classical code C = ψ ( S ⊥ t ) C=\psi(S^{\perp_t}) C = ψ ( S ⊥ t ) , where ψ \psi ψ is the coordinatewise map defined in (6) . For an additive code C ≤ F q 2 n C\leq\mathbb{F}_{q^2}^n C ≤ F q 2 n , we define its normalized dimension as log q 2 ∣ C ∣ \log_{q^2}|C| log q 2 ∣ C ∣ , which need not be an integer. We call C C C an additive ( r , δ ) (r,\delta) ( r , δ ) -c c c LRC if it admits a repair cover satisfying (17) and d ( C ∣ Γ ) ≥ δ d(C|_\Gamma)\geq\delta d ( C ∣ Γ ) ≥ δ for every repair set Γ \Gamma Γ .
Proposition V.2. Let Q Q Q be a pure ( r , δ ) (r,\delta) ( r , δ ) -SSLRC with parameters ( ( n , K , R , d ) ) q ((n,K,R,d))_q (( n , K , R , d ) ) q , where K > 1 K>1 K > 1 . Then C = ψ ( S ⊥ t ) C=\psi(S^{\perp_t}) C = ψ ( S ⊥ t ) is an additive ( r , δ ) (r,\delta) ( r , δ ) -c c c LRC with the same repair cover as Q Q Q , minimum distance d d d , and normalized dimension
κ = log q 2 ∣ C ∣ = n − s 2 m = n + Λ 2 . (40) \kappa=\log_{q^2}|C|=n-\frac{s}{2m}=\frac{n+\Lambda}{2}. \tag*{(40)} κ = log q 2 ∣ C ∣ = n − 2 m s = 2 n + Λ . (40) If Q Q Q has parameters [ [ n , k , g , d ] ] q [[n,k,g,d]]_q [[ n , k , g , d ] ] q , then κ = ( n + k + g ) / 2 \kappa=(n+k+g)/2 κ = ( n + k + g ) /2 . If C C C is also F q 2 \mathbb{F}_{q^2} F q 2 -linear, then κ \kappa κ is an integer and C C C has parameters [ n , κ , d ] q 2 [n,\kappa,d]_{q^2} [ n , κ , d ] q 2 .
Proof. We first determine the normalized dimension of C C C . Since S ⊥ t S^{\perp_t} S ⊥ t is F p \mathbb{F}_p F p -linear and ψ \psi ψ is an F p \mathbb{F}_p F p -linear isomorphism, C C C is additive. By (5), we have
∣ C ∣ = ∣ S ⊥ t ∣ = p 2 m n − s , |C|=|S^{\perp_t}|=p^{2mn-s}, ∣ C ∣ = ∣ S ⊥ t ∣ = p 2 mn − s , κ = log q 2 ∣ C ∣ = 2 m n − s 2 m = n − s 2 m . \kappa=\log_{q^2}|C|=\frac{2mn-s}{2m}=n-\frac{s}{2m}. κ = log q 2 ∣ C ∣ = 2 m 2 mn − s = n − 2 m s . By (8), Λ = n − s / m \Lambda=n-s/m Λ = n − s / m , which proves (39). If Q Q Q has parameters [ [ n , k , g , d ] ] q [[n,k,g,d]]_q [[ n , k , g , d ] ] q , then Λ = k + g \Lambda=k+g Λ = k + g , so κ = ( n + k + g ) / 2 \kappa=(n+k+g)/2 κ = ( n + k + g ) /2 . If C C C is F q 2 \mathbb{F}_{q^2} F q 2 -linear, then κ = dim q 2 C \kappa=\dim_{q^2}C κ = dim q 2 C is an integer.
We next prove that d ( C ) ≥ d d(C)\geq d d ( C ) ≥ d . Let G G G be the gauge label space of Q Q Q . By Definition II.18, every nonzero label in G G G has symplectic weight at least d d d . By (11), every label in S ⊥ t ∖ G S^{\perp_t}\setminus G S ⊥ t ∖ G also has symplectic weight at least d d d . Since K > 1 K>1 K > 1 , the set S ⊥ t ∖ G S^{\perp_t}\setminus G S ⊥ t ∖ G is nonempty, and we may choose w 0 ∈ S ⊥ t ∖ G w_0\in S^{\perp_t}\setminus G w 0 ∈ S ⊥ t ∖ G with swt ( w 0 ) = d \operatorname{swt}(w_0)=d swt ( w 0 ) = d . Since ψ \psi ψ preserves weight, we obtain
d ( C ) = min 0 ≠ w ∈ S ⊥ t swt ( w ) = d . d(C)=\min_{0\ne w\in S^{\perp_t}}\operatorname{swt}(w)=d. d ( C ) = 0 = w ∈ S ⊥ t min swt ( w ) = d .
Finally, we prove that C C C has the same repair cover as Q Q Q . Let Γ \Gamma Γ be a prescribed repair set, and let c ∈ C ∣ Γ c \in C|_{\Gamma} c ∈ C ∣ Γ satisfy wt ( c ) < δ \operatorname{wt}(c) < \delta wt ( c ) < δ . Suppose that c ≠ 0 c \ne0 c = 0 , and choose v ∈ S ⊥ t v \in S^{\perp_t} v ∈ S ⊥ t such that ψ ( v ) ∣ Γ = c \psi(v)|_{\Gamma} = c ψ ( v ) ∣ Γ = c . Let w w w be the zero extension of π Γ ( v ) \pi_{\Gamma}(v) π Γ ( v ) , and set E = supp ( w ) E = \operatorname{supp}(w) E = supp ( w ) . Since ψ \psi ψ acts coordinatewise and preserves support, we have
∅ ≠ E ⊆ Γ , ∣ E ∣ = swt ( w ) = wt ( c ) < δ . \varnothing\ne E \subseteq\Gamma,\quad|E| = \operatorname{swt}(w) = \operatorname{wt}(c) < \delta. ∅ = E ⊆ Γ , ∣ E ∣ = swt ( w ) = wt ( c ) < δ .
For every s ∈ S Γ s \in S_{\Gamma} s ∈ S Γ , we have
[ w , s ] t = [ v , s ] t = 0 , [w,s]_t = [v,s]_t = 0, [ w , s ] t = [ v , s ] t = 0 ,
since w w w and v v v agree on Γ \Gamma Γ and s s s is supported in Γ \Gamma Γ . Thus w ∈ W E ∩ S Γ ⊥ t w \in W_E \cap S^{\perp_t}_{\Gamma} w ∈ W E ∩ S Γ ⊥ t . Since E E E is recoverable within Γ \Gamma Γ for Q Q Q , it follows from condition 2) of Theorem III.8 that
w ∈ W E ∩ S Γ ⊥ t = G E ⊆ G . w \in W_E \cap S^{\perp_t}_{\Gamma} = G_E \subseteq G. w ∈ W E ∩ S Γ ⊥ t = G E ⊆ G .
By Corollary III.10, we have swt ( w ) < δ ≤ d \operatorname{swt}(w) < \delta\le d swt ( w ) < δ ≤ d . By Definition II.18, we obtain w = 0 w = 0 w = 0 , contrary to c ≠ 0 c \ne0 c = 0 . Hence d ( C ∣ Γ ) ≥ δ d(C|_{\Gamma}) \ge\delta d ( C ∣ Γ ) ≥ δ for every prescribed repair set. Since the repair sets still cover [ n ] [n] [ n ] and satisfy the same size bounds, C C C is an additive ( r , δ ) (r,\delta) ( r , δ ) -cLRC with the same repair cover as Q Q Q .
Singleton-Type and Alphabet-Dependent Bounds
For an additive code C ≤ F q 2 n C \le\mathbb{F}_{q^2}^{n} C ≤ F q 2 n and T ⊆ [ n ] T \subseteq[n] T ⊆ [ n ] , let ρ C ( T ) \rho_C(T) ρ C ( T ) denote the normalized dimension of the punctured code C ∣ T C|_T C ∣ T , that is,
ρ C ( T ) = log q 2 ∣ π T ( C ) ∣ = dim p π T ( C ) 2 m . \rho_C(T) = \log_{q^2} |\pi_T(C)| = \frac{\dim_p \pi_T(C)}{2m}. ρ C ( T ) = log q 2 ∣ π T ( C ) ∣ = 2 m dim p π T ( C ) .
We obtain the following lemma by adapting the arguments in [7 Lemma 1] and [19 proof of Corollary 2] to additive codes.
Lemma V.3. Let C = ψ ( S ⊥ t ) C = \psi(S^{\perp_t}) C = ψ ( S ⊥ t ) be the comparison code in Proposition V.2 with normalized dimension κ \kappa κ , and let ℓ = r + δ − 1 \ell= r + \delta- 1 ℓ = r + δ − 1 . For every integer a ≥ 0 a \ge0 a ≥ 0 with a r < κ ar < \kappa a r < κ , there exists a coordinate set T ⊆ [ n ] T \subseteq[n] T ⊆ [ n ] such that
∣ T ∣ ≤ a ℓ , ρ C ( T ) ≤ a r , ∣ T ∣ − ρ C ( T ) ≥ a ( δ − 1 ) . |T| \le a\ell,\quad\rho_C(T) \le ar,\quad|T| - \rho_C(T) \ge a(\delta- 1). ∣ T ∣ ≤ a ℓ , ρ C ( T ) ≤ a r , ∣ T ∣ − ρ C ( T ) ≥ a ( δ − 1 ) . If a ℓ ≤ n a\ell\le n a ℓ ≤ n , the set T T T can be enlarged to a set T ′ T' T ′ satisfying ∣ T ′ ∣ = a ℓ |T'| = a\ell ∣ T ′ ∣ = a ℓ and ρ C ( T ′ ) ≤ a r \rho_C(T') \le ar ρ C ( T ′ ) ≤ a r .
Proof. For a = 0 a = 0 a = 0 , take T = T ′ = ∅ T = T' = \varnothing T = T ′ = ∅ . Assume a ≥ 1 a \ge1 a ≥ 1 . Expand each coordinate of C C C in a fixed F p \mathbb{F}_p F p -basis of F q 2 \mathbb{F}_{q^2} F q 2 , and let M M M be a generator matrix of the resulting F p \mathbb{F}_p F p -linear code. Let
M = [ M 1 ∣ M 2 ∣ ⋯ ∣ M n ] , M = [M_1 \mid M_2 \mid\cdots\mid M_n], M = [ M 1 ∣ M 2 ∣ ⋯ ∣ M n ] ,
where M i M_i M i consists of the 2 m 2m 2 m scalar columns corresponding to coordinate i ∈ [ n ] i \in[n] i ∈ [ n ] . For T ⊆ [ n ] T \subseteq[n] T ⊆ [ n ] , let M T M_T M T be the submatrix formed by the blocks indexed by T T T , and let C ( T ) \mathcal{C}(T) C ( T ) be its column space over F p \mathbb{F}_p F p . The row space of M T M_T M T is the coordinate expansion of π T ( C ) \pi_T(C) π T ( C ) . Since row rank equals column rank, we have
dim p C ( T ) = rank p M T = dim p π T ( C ) = 2 m ρ C ( T ) . \dim_p \mathcal{C}(T) = \operatorname{rank}_p M_T = \dim_p \pi_T(C) = 2m\rho_C(T). dim p C ( T ) = rank p M T = dim p π T ( C ) = 2 m ρ C ( T ) .
Let Γ \Gamma Γ be a repair set in the prescribed repair cover, and let D ⊆ Γ D \subseteq\Gamma D ⊆ Γ with ∣ D ∣ ≤ δ − 1 |D| \le\delta- 1 ∣ D ∣ ≤ δ − 1 . The puncturing map C ∣ Γ → C ∣ Γ ∖ D C|_{\Gamma} \to C|_{\Gamma\setminus D} C ∣ Γ → C ∣ Γ ∖ D is surjective. Any codeword in its kernel is supported in D D D and has weight at most δ − 1 \delta- 1 δ − 1 . Since d ( C ∣ Γ ) ≥ δ d(C|_{\Gamma}) \ge\delta d ( C ∣ Γ ) ≥ δ by Proposition V.2 , the puncturing map has trivial kernel. By rank-nullity (1), the two punctured codes have the same F p \mathbb{F}_p F p -dimension. Consequently,
C ( Γ ∖ D ) ⊆ C ( Γ ) , dim p C ( Γ ∖ D ) = dim p C ( Γ ) . \mathcal{C}(\Gamma\setminus D) \subseteq\mathcal{C}(\Gamma),\quad\dim_p \mathcal{C}(\Gamma\setminus D) = \dim_p \mathcal{C}(\Gamma). C ( Γ ∖ D ) ⊆ C ( Γ ) , dim p C ( Γ ∖ D ) = dim p C ( Γ ) .
Hence C ( Γ ∖ D ) = C ( Γ ) \mathcal{C}(\Gamma\setminus D) = \mathcal{C}(\Gamma) C ( Γ ∖ D ) = C ( Γ ) for every such D D D . Since ∣ Γ ∣ ≥ δ |\Gamma| \ge\delta ∣Γ∣ ≥ δ , we may take D 0 ⊆ Γ D_0 \subseteq\Gamma D 0 ⊆ Γ with ∣ D 0 ∣ = δ − 1 |D_0| = \delta- 1 ∣ D 0 ∣ = δ − 1 . By the equality of the column spaces, we have ρ C ( Γ ) = ρ C ( Γ ∖ D 0 ) \rho_C(\Gamma) = \rho_C(\Gamma\setminus D_0) ρ C ( Γ ) = ρ C ( Γ ∖ D 0 ) . The matrix M Γ ∖ D 0 M_{\Gamma\setminus D_0} M Γ ∖ D 0 contains ∣ Γ ∣ − ( δ − 1 ) |\Gamma| - (\delta- 1) ∣Γ∣ − ( δ − 1 ) coordinate blocks, each consisting of 2 m 2m 2 m scalar columns. Since its rank is at most its number of columns,
ρ C ( Γ ) = rank p M Γ ∖ D 0 2 m ≤ 2 m ( ∣ Γ ∣ − ( δ − 1 ) ) 2 m = ∣ Γ ∣ − ( δ − 1 ) ≤ r , \begin{aligned}
\rho_C(\Gamma) &= \frac{\operatorname{rank}_p M_{\Gamma\setminus D_0}}{2m} \\
&\le\frac{2m(|\Gamma| - (\delta- 1))}{2m} \\
&= |\Gamma| - (\delta- 1) \le r,
\end{aligned} ρ C ( Γ ) = 2 m rank p M Γ ∖ D 0 ≤ 2 m 2 m ( ∣Γ∣ − ( δ − 1 )) = ∣Γ∣ − ( δ − 1 ) ≤ r ,
where the last inequality follows from ∣ Γ ∣ ≤ r + δ − 1 |\Gamma| \le r + \delta- 1 ∣Γ∣ ≤ r + δ − 1 . Combining the column-space equality with this bound on ρ C ( Γ ) \rho_C(\Gamma) ρ C ( Γ ) , we obtain, for every D ⊆ Γ D \subseteq\Gamma D ⊆ Γ with ∣ D ∣ ≤ δ − 1 |D| \le\delta- 1 ∣ D ∣ ≤ δ − 1 ,
C ( Γ ∖ D ) = C ( Γ ) , \mathcal{C}(\Gamma\setminus D) = \mathcal{C}(\Gamma), C ( Γ ∖ D ) = C ( Γ ) ,
ρ C ( Γ ∖ D ) = ρ C ( Γ ) ≤ ∣ Γ ∣ − ( δ − 1 ) ≤ r . (41) \rho_C(\Gamma\setminus D) = \rho_C(\Gamma) \le|\Gamma| - (\delta- 1) \le r. \tag*{(41)} ρ C ( Γ ∖ D ) = ρ C ( Γ ) ≤ ∣Γ∣ − ( δ − 1 ) ≤ r . (41)
Now let T ⊆ [ n ] T \subseteq[n] T ⊆ [ n ] satisfy ρ C ( T ) < κ \rho_C(T) < \kappa ρ C ( T ) < κ . Since ρ C ( [ n ] ) = κ \rho_C([n]) = \kappa ρ C ([ n ]) = κ , we have
dim p C ( T ) = 2 m ρ C ( T ) < 2 m κ = dim p C ( [ n ] ) . \dim_p \mathcal{C}(T) = 2m\rho_C(T) < 2m\kappa= \dim_p \mathcal{C}([n]). dim p C ( T ) = 2 m ρ C ( T ) < 2 mκ = dim p C ([ n ]) .
Hence there is a coordinate i ∈ [ n ] ∖ T i \in[n] \setminus T i ∈ [ n ] ∖ T whose block M i M_i M i contains a column outside C ( T ) \mathcal{C}(T) C ( T ) , so that C ( { i } ) ⊈ C ( T ) \mathcal{C}(\{i\}) \not\subseteq\mathcal{C}(T) C ({ i }) ⊆ C ( T ) . Let Γ \Gamma Γ be a repair set containing i i i in the prescribed repair cover. Then
C ( T ) ⊊ C ( T ) + C ( Γ ) = C ( T ∪ Γ ) . \mathcal{C}(T) \subsetneq\mathcal{C}(T) + \mathcal{C}(\Gamma) = \mathcal{C}(T \cup\Gamma). C ( T ) ⊊ C ( T ) + C ( Γ ) = C ( T ∪ Γ ) .
Let u u u denote the number of new coordinates and b b b the increase in normalized dimension, that is,
u = ∣ Γ ∖ T ∣ = ∣ T ∪ Γ ∣ − ∣ T ∣ , b = ρ C ( T ∪ Γ ) − ρ C ( T ) > 0. u = |\Gamma\setminus T| = |T \cup\Gamma| - |T|,\quad b = \rho_{\mathcal{C}}(T \cup\Gamma) - \rho_{\mathcal{C}}(T) > 0. u = ∣Γ ∖ T ∣ = ∣ T ∪ Γ∣ − ∣ T ∣ , b = ρ C ( T ∪ Γ ) − ρ C ( T ) > 0.
If u ≤ δ − 1 u \le\delta- 1 u ≤ δ − 1 , apply (41) with D = Γ ∖ T D = \Gamma\setminus T D = Γ ∖ T . Since Γ ∖ D = Γ ∩ T \Gamma\setminus D = \Gamma\cap T Γ ∖ D = Γ ∩ T , we obtain C ( Γ ) = C ( Γ ∩ T ) ⊆ C ( T ) \mathcal{C}(\Gamma) = \mathcal{C}(\Gamma\cap T) \subseteq\mathcal{C}(T) C ( Γ ) = C ( Γ ∩ T ) ⊆ C ( T ) . It follows that
C ( T ∪ Γ ) = C ( T ) + C ( Γ ) = C ( T ) . \mathcal{C}(T \cup\Gamma) = \mathcal{C}(T) + \mathcal{C}(\Gamma) = \mathcal{C}(T). C ( T ∪ Γ ) = C ( T ) + C ( Γ ) = C ( T ) .
Thus ρ C ( T ∪ Γ ) = ρ C ( T ) \rho_{\mathcal{C}}(T \cup\Gamma) = \rho_{\mathcal{C}}(T) ρ C ( T ∪ Γ ) = ρ C ( T ) , and hence b = 0 b = 0 b = 0 , contrary to b > 0 b > 0 b > 0 . Hence u ≥ δ u \ge\delta u ≥ δ . Let D ⊆ Γ ∖ T D \subseteq\Gamma\setminus T D ⊆ Γ ∖ T satisfy ∣ D ∣ = δ − 1 |D| = \delta- 1 ∣ D ∣ = δ − 1 . By (41) ,
C ( T ∪ Γ ) = C ( T ) + C ( Γ ∖ D ) = C ( T ) + C ( ( Γ ∖ T ) ∖ D ) , \begin{aligned}
\mathcal{C}(T \cup\Gamma) &= \mathcal{C}(T) + \mathcal{C}(\Gamma\setminus D) \\
&= \mathcal{C}(T) + \mathcal{C}((\Gamma\setminus T) \setminus D),
\end{aligned} C ( T ∪ Γ ) = C ( T ) + C ( Γ ∖ D ) = C ( T ) + C (( Γ ∖ T ) ∖ D ) ,
where the second equality uses C ( Γ ∩ T ) ⊆ C ( T ) \mathcal{C}(\Gamma\cap T) \subseteq\mathcal{C}(T) C ( Γ ∩ T ) ⊆ C ( T ) . The set ( Γ ∖ T ) ∖ D (\Gamma\setminus T) \setminus D ( Γ ∖ T ) ∖ D has u − ( δ − 1 ) u - (\delta- 1) u − ( δ − 1 ) coordinates, so its column space is generated by 2 m ( u − ( δ − 1 ) ) 2m(u - (\delta- 1)) 2 m ( u − ( δ − 1 )) scalar columns. Therefore,
2 m b = dim F p C ( T ∪ Γ ) − dim F p C ( T ) ≤ dim F p C ( ( Γ ∖ T ) ∖ D ) ≤ 2 m ( u − ( δ − 1 ) ) . \begin{aligned}
2mb &= \dim_{\mathbb{F}_p} \mathcal{C}(T \cup\Gamma) - \dim_{\mathbb{F}_p} \mathcal{C}(T) \\
&\le\dim_{\mathbb{F}_p} \mathcal{C}((\Gamma\setminus T) \setminus D) \\
&\le2m(u - (\delta- 1)).
\end{aligned} 2 mb = dim F p C ( T ∪ Γ ) − dim F p C ( T ) ≤ dim F p C (( Γ ∖ T ) ∖ D ) ≤ 2 m ( u − ( δ − 1 )) .
Since u ≤ ∣ Γ ∣ ≤ ℓ = r + δ − 1 u \le|\Gamma| \le\ell= r + \delta- 1 u ≤ ∣Γ∣ ≤ ℓ = r + δ − 1 , we obtain
u ≤ ℓ , 0 < b ≤ u − ( δ − 1 ) ≤ ∣ Γ ∣ − ( δ − 1 ) ≤ r . u \le\ell,\quad0 < b \le u - (\delta- 1) \le|\Gamma| - (\delta- 1) \le r. u ≤ ℓ , 0 < b ≤ u − ( δ − 1 ) ≤ ∣Γ∣ − ( δ − 1 ) ≤ r .
Moreover,
( ∣ T ∪ Γ ∣ − ρ C ( T ∪ Γ ) ) − ( ∣ T ∣ − ρ C ( T ) ) = ( ( ∣ T ∣ + u ) − ( ρ C ( T ) + b ) ) − ( ∣ T ∣ − ρ C ( T ) ) = u − b ≥ δ − 1. \begin{aligned}
\bigl(|T \cup\Gamma| - \rho_{\mathcal{C}}(T \cup\Gamma)\bigr) - \bigl(|T| - \rho_{\mathcal{C}}(T)\bigr)
&= \bigl((|T| + u) - (\rho_{\mathcal{C}}(T) + b)\bigr) - \bigl(|T| - \rho_{\mathcal{C}}(T)\bigr) \\
&= u - b \ge\delta- 1.
\end{aligned} ( ∣ T ∪ Γ∣ − ρ C ( T ∪ Γ ) ) − ( ∣ T ∣ − ρ C ( T ) ) = ( ( ∣ T ∣ + u ) − ( ρ C ( T ) + b ) ) − ( ∣ T ∣ − ρ C ( T ) ) = u − b ≥ δ − 1.
Thus, for every T ⊆ [ n ] T \subseteq[n] T ⊆ [ n ] with ρ C ( T ) < κ \rho_{\mathcal{C}}(T) < \kappa ρ C ( T ) < κ , there exists a repair set Γ \Gamma Γ such that
∣ T ∪ Γ ∣ − ∣ T ∣ ≤ ℓ , |T \cup\Gamma| - |T| \le\ell, ∣ T ∪ Γ∣ − ∣ T ∣ ≤ ℓ ,
0 < ρ C ( T ∪ Γ ) − ρ C ( T ) ≤ r , 0 < \rho_{\mathcal{C}}(T \cup\Gamma) - \rho_{\mathcal{C}}(T) \le r, 0 < ρ C ( T ∪ Γ ) − ρ C ( T ) ≤ r ,
∣ T ∪ Γ ∣ − ρ C ( T ∪ Γ ) ≥ ∣ T ∣ − ρ C ( T ) + δ − 1. |T \cup\Gamma| - \rho_{\mathcal{C}}(T \cup\Gamma) \ge|T| - \rho_{\mathcal{C}}(T) + \delta- 1. ∣ T ∪ Γ∣ − ρ C ( T ∪ Γ ) ≥ ∣ T ∣ − ρ C ( T ) + δ − 1.
Let T 0 = ∅ T_0 = \varnothing T 0 = ∅ . We construct T 1 , … , T a T_1, \ldots, T_a T 1 , … , T a inductively. For 0 ≤ j < a 0 \le j < a 0 ≤ j < a , suppose that T j T_j T j has been constructed. Since ρ C ( T 0 ) = 0 \rho_{\mathcal{C}}(T_0) = 0 ρ C ( T 0 ) = 0 and each of the first j j j steps increases the normalized dimension by at most r r r , we have
ρ C ( T j ) ≤ ρ C ( T 0 ) + j r = j r ≤ a r < κ . \rho_{\mathcal{C}}(T_j) \le\rho_{\mathcal{C}}(T_0) + jr = jr \le ar < \kappa. ρ C ( T j ) ≤ ρ C ( T 0 ) + j r = j r ≤ a r < κ .
There is therefore a repair set Γ j + 1 \Gamma_{j+1} Γ j + 1 satisfying the three inequalities above with T = T j T = T_j T = T j . Set T j + 1 = T j ∪ Γ j + 1 T_{j+1} = T_j \cup\Gamma_{j+1} T j + 1 = T j ∪ Γ j + 1 . Then
ρ C ( T j + 1 ) ≤ ρ C ( T j ) + r ≤ ( j + 1 ) r . \rho_{\mathcal{C}}(T_{j+1}) \le\rho_{\mathcal{C}}(T_j) + r \le(j + 1)r. ρ C ( T j + 1 ) ≤ ρ C ( T j ) + r ≤ ( j + 1 ) r .
Applying the three increment bounds successively for a a a steps, we obtain
∣ T a ∣ ≤ a ℓ , ρ C ( T a ) ≤ a r , ∣ T a ∣ − ρ C ( T a ) ≥ a ( δ − 1 ) . |T_a| \le a\ell,\quad\rho_{\mathcal{C}}(T_a) \le ar,\quad|T_a| - \rho_{\mathcal{C}}(T_a) \ge a(\delta- 1). ∣ T a ∣ ≤ a ℓ , ρ C ( T a ) ≤ a r , ∣ T a ∣ − ρ C ( T a ) ≥ a ( δ − 1 ) .
Hence T = T a T = T_a T = T a satisfies the three required inequalities.
Finally, suppose a ℓ ≤ n a\ell\le n a ℓ ≤ n . Since ∣ T a ∣ ≤ a ℓ |T_a| \le a\ell ∣ T a ∣ ≤ a ℓ , take T ′ ⊇ T a T' \supseteq T_a T ′ ⊇ T a with ∣ T ′ ∣ = a ℓ |T'| = a\ell ∣ T ′ ∣ = a ℓ . Each of the a ℓ − ∣ T a ∣ a\ell- |T_a| a ℓ − ∣ T a ∣ added coordinates contributes 2 m 2m 2 m scalar columns and therefore increases the normalized dimension by at most one. Hence
ρ C ( T ′ ) ≤ ρ C ( T a ) + a ℓ − ∣ T a ∣ = a ℓ − ( ∣ T a ∣ − ρ C ( T a ) ) ≤ a ℓ − a ( δ − 1 ) = a r . \begin{aligned}
\rho_{\mathcal{C}}(T') &\le\rho_{\mathcal{C}}(T_a) + a\ell- |T_a| \\
&= a\ell- \bigl(|T_a| - \rho_{\mathcal{C}}(T_a)\bigr) \\
&\le a\ell- a(\delta- 1) = ar.
\end{aligned} ρ C ( T ′ ) ≤ ρ C ( T a ) + a ℓ − ∣ T a ∣ = a ℓ − ( ∣ T a ∣ − ρ C ( T a ) ) ≤ a ℓ − a ( δ − 1 ) = a r .
We now apply Lemma V.3 to derive a Singleton-type bound for ( r , δ ) (r,\delta) ( r , δ ) -SSLRCs.
Theorem V.4. A pure ( r , δ ) (r,\delta) ( r , δ ) -SSLRC with parameters [ [ n , k , g , d ] ] q [[n,k,g,d]]_q [[ n , k , g , d ] ] q and k > 0 k > 0 k > 0 satisfies
k + g + 2 ( d − 1 ) + 2 ( ⌈ n + k + g 2 r ⌉ − 1 ) ( δ − 1 ) ≤ n . (42) k + g + 2(d - 1) + 2\left(\left\lceil\frac{n + k + g}{2r}\right\rceil- 1\right)(\delta- 1) \le n. \tag*{(42)} k + g + 2 ( d − 1 ) + 2 ( ⌈ 2 r n + k + g ⌉ − 1 ) ( δ − 1 ) ≤ n . (42)
Moreover, a pure ( r , δ ) (r,\delta) ( r , δ ) -SSLRC with parameters ( ( n , K , R , d ) ) q ((n,K,R,d))_q (( n , K , R , d ) ) q and K > 1 K>1 K > 1 satisfies
Λ + 2 ( d − 1 ) + 2 ( ⌈ n + Λ 2 r ⌉ − 1 ) ( δ − 1 ) ≤ n , Λ = log q ( K R ) . (43) \Lambda+2(d-1)+2\left(\left\lceil\frac{n+\Lambda}{2r}\right\rceil-1\right)(\delta-1)\le n,\qquad\Lambda=\log_q(KR). \tag*{(43)} Λ + 2 ( d − 1 ) + 2 ( ⌈ 2 r n + Λ ⌉ − 1 ) ( δ − 1 ) ≤ n , Λ = log q ( K R ) . (43)
Proof. Let C = ψ ( S ⊥ t ) C=\psi(S^{\perp t}) C = ψ ( S ⊥ t ) and κ = ( n + Λ ) / 2 \kappa=(n+\Lambda)/2 κ = ( n + Λ ) /2 , as in Proposition V.2. Set a = ⌈ κ / r ⌉ − 1 a=\lceil\kappa/r\rceil-1 a = ⌈ κ / r ⌉ − 1 . Then a ≥ 0 a\ge0 a ≥ 0 and a r < κ ar<\kappa a r < κ . Choose T T T as in Lemma V.3. The kernel of π T ∣ C \pi_T|_C π T ∣ C is C T c C_{T^c} C T c . The shortened code D = sh T c ( C ) D=\operatorname{sh}_{T^c}(C) D = sh T c ( C ) is obtained from C T c C_{T^c} C T c by deleting its zero coordinates in T T T . By rank-nullity (1) and the definition of shortening, we have
log q 2 ∣ D ∣ = κ − ρ C ( T ) > 0 , d ( D ) ≥ d . \log_{q^2}|D|=\kappa-\rho_C(T)>0,\qquad d(D)\ge d. log q 2 ∣ D ∣ = κ − ρ C ( T ) > 0 , d ( D ) ≥ d .
Since the shortened code is nonzero, the cardinality form of the classical Singleton bound [32 ] implies that
∣ D ∣ ≤ ( q 2 ) n − ∣ T ∣ − d + 1 , κ − ρ C ( T ) ≤ n − ∣ T ∣ − d + 1. |D|\le(q^2)^{n-|T|-d+1},\qquad\kappa-\rho_C(T)\le n-|T|-d+1. ∣ D ∣ ≤ ( q 2 ) n − ∣ T ∣ − d + 1 , κ − ρ C ( T ) ≤ n − ∣ T ∣ − d + 1.
It follows that
d ≤ n − κ + 1 − ( ∣ T ∣ − ρ C ( T ) ) ≤ n − κ + 1 − a ( δ − 1 ) . \begin{aligned}
d\le n-\kappa+1-\bigl(|T|-\rho_C(T)\bigr) \\
&\le n-\kappa+1-a(\delta-1).
\end{aligned} d ≤ n − κ + 1 − ( ∣ T ∣ − ρ C ( T ) ) ≤ n − κ + 1 − a ( δ − 1 ) .
Since 2 κ = n + Λ 2\kappa=n+\Lambda 2 κ = n + Λ , we obtain
Λ + 2 ( d − 1 ) + 2 a ( δ − 1 ) ≤ n , a = ⌈ n + Λ 2 r ⌉ − 1 , \Lambda+2(d-1)+2a(\delta-1)\le n,\qquad a=\left\lceil\frac{n+\Lambda}{2r}\right\rceil-1, Λ + 2 ( d − 1 ) + 2 a ( δ − 1 ) ≤ n , a = ⌈ 2 r n + Λ ⌉ − 1 ,
which proves (42). For integer subsystem parameters, Λ = k + g \Lambda=k+g Λ = k + g . Thus (41) follows from (42).
Remark V.5. For g = 0 g=0 g = 0 , (41) reduces to the Singleton-type bound for the pure stabilizer LRCs considered in [13 Theorem 31 and Remark 34] .
Corollary V.6. A pure ( r , δ ) (r,\delta) ( r , δ ) -SSLRC with parameters ( ( n , K , R , d ) ) q ((n,K,R,d))_q (( n , K , R , d ) ) q and K > 1 K>1 K > 1 satisfies r ≥ δ r\ge\delta r ≥ δ .
Proof. Let κ = ( n + Λ ) / 2 \kappa=(n+\Lambda)/2 κ = ( n + Λ ) /2 , t = δ − 1 ≥ 1 t=\delta-1\ge1 t = δ − 1 ≥ 1 , and c = ⌈ κ / r ⌉ c=\lceil\kappa/r\rceil c = ⌈ κ / r ⌉ . By Corollary III.10 and Theorem V.4, we have
t + 1 ≤ d ≤ n − κ + 1 − ( c − 1 ) t , κ + c t ≤ n . t+1\le d\le n-\kappa+1-(c-1)t,\qquad\kappa+ct\le n. t + 1 ≤ d ≤ n − κ + 1 − ( c − 1 ) t , κ + c t ≤ n . Suppose r ≤ t r\le t r ≤ t . Since c = ⌈ κ / r ⌉ ≥ κ / r c=\lceil\kappa/r\rceil\ge\kappa/r c = ⌈ κ / r ⌉ ≥ κ / r , t / r ≥ 1 t/r\ge1 t / r ≥ 1 , and Λ = log q ( K R ) > 0 \Lambda=\log_q(KR)>0 Λ = log q ( K R ) > 0 , we obtain
n ≥ κ + c t ≥ κ + κ t r ≥ 2 κ = n + Λ > n . n\ge\kappa+ct\ge\kappa+\frac{\kappa t}{r}\ge2\kappa=n+\Lambda>n. n ≥ κ + c t ≥ κ + r κ t ≥ 2 κ = n + Λ > n . This contradiction implies r > δ − 1 r>\delta-1 r > δ − 1 , and hence r ≥ δ r\ge\delta r ≥ δ .
For Q = q 2 Q=q^2 Q = q 2 and integers N ≥ 0 N\ge0 N ≥ 0 and d ≥ 1 d\ge1 d ≥ 1 , let k a d d ( Q ) ( N , d ) k_{\mathrm{add}}^{(Q)}(N,d) k add ( Q ) ( N , d ) denote the maximum of log Q ∣ D ∣ \log_Q|D| log Q ∣ D ∣ over F p \mathbb{F}_p F p -linear codes D ≤ F Q N D\le\mathbb{F}_Q^N D ≤ F Q N in which every nonzero codeword has Hamming weight at least d d d . Let k o p t ( Q ) ( N , d ) k_{\mathrm{opt}}^{(Q)}(N,d) k opt ( Q ) ( N , d ) denote the maximum of dim Q D \dim_Q D dim Q D over F Q \mathbb{F}_Q F Q -linear codes satisfying the same weight condition.
Theorem V.7. Let Q \mathcal{Q} Q be a pure ( r , δ ) (r,\delta) ( r , δ ) -SSLRC with parameters ( ( n , K , R , d ) ) q ((n,K,R,d))_q (( n , K , R , d ) ) q , where q = p m q=p^m q = p m and K > 1 K>1 K > 1 . Let Λ = log q ( K R ) \Lambda=\log_q(KR) Λ = log q ( K R ) and define
A = { a ∈ Z ≥ 0 : 2 a r < n + Λ , a ( r + δ − 1 ) ≤ n } , \mathcal{A}=\{a\in\mathbb{Z}_{\ge0}:2ar<n+\Lambda,\ a(r+\delta-1)\le n\}, A = { a ∈ Z ≥ 0 : 2 a r < n + Λ , a ( r + δ − 1 ) ≤ n } , where a a a counts the repair sets added in Lemma V.3. Then
Λ ≤ min a ∈ A { 2 a r + 2 k a d d ( q 2 ) ( n − a ( r + δ − 1 ) , d ) − n } . (44) \Lambda\le\min_{a\in\mathcal{A}}\{2ar+2k_{\mathrm{add}}^{(q^2)}(n-a(r+\delta-1),d)-n\}. \tag*{(44)} Λ ≤ a ∈ A min { 2 a r + 2 k add ( q 2 ) ( n − a ( r + δ − 1 ) , d ) − n } . (44) If Q \mathcal{Q} Q has parameters [ [ n , k , g , d ] ] q [[n,k,g,d]]_q [[ n , k , g , d ] ] q , then
k + g ≤ min a ∈ Z ≥ 0 , 2 a r < n + k + g a ( r + δ − 1 ) ≤ n { 2 a r + 2 k a d d ( q 2 ) ( n − a ( r + δ − 1 ) , d ) − n } . (45) k+g\le\min_{\substack{a\in\mathbb{Z}_{\ge0},\ 2ar<n+k+g\\a(r+\delta-1)\le n}}\{2ar+2k_{\mathrm{add}}^{(q^2)}(n-a(r+\delta-1),d)-n\}. \tag*{(45)} k + g ≤ a ∈ Z ≥ 0 , 2 a r < n + k + g a ( r + δ − 1 ) ≤ n min { 2 a r + 2 k add ( q 2 ) ( n − a ( r + δ − 1 ) , d ) − n } . (45) Moreover, for C = ψ ( S ⊥ t ) C=\psi(S^{\perp t}) C = ψ ( S ⊥ t ) and every T ⊆ [ n ] T\subseteq[n] T ⊆ [ n ] , we have
Λ ≤ 2 ρ C ( T ) + 2 k a d d ( q 2 ) ( n − ∣ T ∣ , d ) − n . (46) \Lambda\le2\rho_C(T)+2k_{\mathrm{add}}^{(q^2)}(n-|T|,d)-n. \tag*{(46)} Λ ≤ 2 ρ C ( T ) + 2 k add ( q 2 ) ( n − ∣ T ∣ , d ) − n . (46) If C C C is F q 2 \mathbb{F}_{q^2} F q 2 -linear, k a d d ( q 2 ) k_{\mathrm{add}}^{(q^2)} k add ( q 2 ) can be replaced by k o p t ( q 2 ) k_{\mathrm{opt}}^{(q^2)} k opt ( q 2 ) in all three inequalities.
Proof. Let κ = log q 2 ∣ C ∣ = ( n + Λ ) / 2 \kappa=\log_{q^2}|C|=(n+\Lambda)/2 κ = log q 2 ∣ C ∣ = ( n + Λ ) /2 , as in Proposition V.2. For every T ⊆ [ n ] T\subseteq[n] T ⊆ [ n ] , we have the exact sequence
0 ⟶ C T c ⟶ C → π T C ∣ T ⟶ 0. 0\longrightarrow C_{T^c}\longrightarrow C\xrightarrow{\pi_T}C|_T\longrightarrow0. 0 ⟶ C T c ⟶ C π T C ∣ T ⟶ 0.
The shortened code D T = sh T c ( C ) D_T = \operatorname{sh}_{T^c}(C) D T = sh T c ( C ) is obtained from C T c C_{T^c} C T c by deleting the zero coordinates in T T T , which preserves its F p \mathbb{F}_p F p -dimension and Hamming weights. By rank-nullity (1), we have
log q 2 ∣ D T ∣ = κ − ρ C ( T ) , d ( D T ) ≥ d . \log_{q^2} |D_T| = \kappa- \rho_C(T),\ d(D_T) \ge d. log q 2 ∣ D T ∣ = κ − ρ C ( T ) , d ( D T ) ≥ d .
By the definition of k a d d ( q 2 ) k_{\mathrm{add}}^{(q^2)} k add ( q 2 ) , we have
κ − ρ C ( T ) ≤ k a d d ( q 2 ) ( n − ∣ T ∣ , d ) , \kappa- \rho_C(T) \le k_{\mathrm{add}}^{(q^2)}(n-|T|,d), κ − ρ C ( T ) ≤ k add ( q 2 ) ( n − ∣ T ∣ , d ) ,
n + Λ = 2 κ ≤ 2 ρ C ( T ) + 2 k a d d ( q 2 ) ( n − ∣ T ∣ , d ) . n+\Lambda= 2\kappa\le2\rho_C(T) + 2k_{\mathrm{add}}^{(q^2)}(n-|T|,d). n + Λ = 2 κ ≤ 2 ρ C ( T ) + 2 k add ( q 2 ) ( n − ∣ T ∣ , d ) .
Thus (45) holds.
Let ℓ = r + δ − 1 \ell= r+\delta-1 ℓ = r + δ − 1 and a ∈ A a \in\mathcal{A} a ∈ A . Then a r < κ ar < \kappa a r < κ and a ℓ ≤ n a\ell\le n a ℓ ≤ n . By Lemma V.3, we can choose a set T ′ T' T ′ with ∣ T ′ ∣ = a ℓ |T'| = a\ell ∣ T ′ ∣ = a ℓ and ρ C ( T ′ ) ≤ a r \rho_C(T') \le ar ρ C ( T ′ ) ≤ a r . Applying (45) to T ′ T' T ′ , we obtain
Λ ≤ 2 ρ C ( T ′ ) + 2 k a d d ( q 2 ) ( n − a ℓ , d ) − n \Lambda\le2\rho_C(T') + 2k_{\mathrm{add}}^{(q^2)}(n-a\ell,d)-n Λ ≤ 2 ρ C ( T ′ ) + 2 k add ( q 2 ) ( n − a ℓ , d ) − n
≤ 2 a r + 2 k a d d ( q 2 ) ( n − a ( r + δ − 1 ) , d ) − n . \le2ar + 2k_{\mathrm{add}}^{(q^2)}(n-a(r+\delta-1),d)-n. ≤ 2 a r + 2 k add ( q 2 ) ( n − a ( r + δ − 1 ) , d ) − n .
Taking the minimum over a ∈ A a \in\mathcal{A} a ∈ A , we obtain (43). For integer subsystem parameters, Λ = k + g \Lambda= k+g Λ = k + g , which implies (44).
If C C C is F q 2 \mathbb{F}_{q^2} F q 2 -linear, every shortened code D T D_T D T is also F q 2 \mathbb{F}_{q^2} F q 2 -linear. In this case,
log q 2 ∣ D T ∣ = dim q 2 D T ≤ k o p t ( q 2 ) ( n − ∣ T ∣ , d ) . \log_{q^2}|D_T| = \dim_{q^2}D_T \le k_{\mathrm{opt}}^{(q^2)}(n-|T|,d). log q 2 ∣ D T ∣ = dim q 2 D T ≤ k opt ( q 2 ) ( n − ∣ T ∣ , d ) .
The same argument establishes all three inequalities with k o p t ( q 2 ) k_{\mathrm{opt}}^{(q^2)} k opt ( q 2 ) in place of k a d d ( q 2 ) k_{\mathrm{add}}^{(q^2)} k add ( q 2 ) .
Remark V.8. For g = 0 g=0 g = 0 and an F q 2 \mathbb{F}_{q^2} F q 2 -linear comparison code C = ψ ( S ⊥ t ) C=\psi(S^{\perp_t}) C = ψ ( S ⊥ t ) , the linear version of (43) reduces to the classical alphabet-dependent bound in [19 Corollary 2] under the substitution κ = ( n + k ) / 2 \kappa=(n+k)/2 κ = ( n + k ) /2 , with the same admissible range of a a a .
Remark V.9. The alphabet-dependent bound (43) is at least as strong as the Singleton-type bound (42). By the classical Singleton bound [32 ] , we have
k a d d ( q 2 ) ( N , d ) ≤ max { 0 , N − d + 1 } . k_{\mathrm{add}}^{(q^2)}(N,d) \le\max\{0,N-d+1\}. k add ( q 2 ) ( N , d ) ≤ max { 0 , N − d + 1 } .
Let a ∗ = ⌈ κ / r ⌉ − 1 a_*=\lceil\kappa/r\rceil-1 a ∗ = ⌈ κ / r ⌉ − 1 . Then a ∗ ≥ 0 a_* \ge0 a ∗ ≥ 0 and a ∗ r < κ a_*r<\kappa a ∗ r < κ . By Theorem V.4 and Corollary III.10, we have
n ≥ κ + ( a ∗ + 1 ) ( δ − 1 ) > a ∗ r + a ∗ ( δ − 1 ) = a ∗ ( r + δ − 1 ) . \begin{aligned}
n \ge\kappa+(a_*+1)(\delta-1) \\
&> a_*r+a_*(\delta-1) \\
&= a_*(r+\delta-1).
\end{aligned} n ≥ κ + ( a ∗ + 1 ) ( δ − 1 ) > a ∗ r + a ∗ ( δ − 1 ) = a ∗ ( r + δ − 1 ) .
Thus a ∗ ∈ A a_* \in\mathcal{A} a ∗ ∈ A . For a = a ∗ a=a_* a = a ∗ , let T ′ T' T ′ be the coordinate set used in the proof of Theorem V.7. The shortened code D T ′ D_{T'} D T ′ has length N ∗ = n − a ∗ ( r + δ − 1 ) N_* = n-a_*(r+\delta-1) N ∗ = n − a ∗ ( r + δ − 1 ) , and
log q 2 ∣ D T ′ ∣ = κ − ρ C ( T ′ ) ≥ κ − a ∗ r > 0. \log_{q^2}|D_{T'}| = \kappa-\rho_C(T') \ge\kappa-a_*r>0. log q 2 ∣ D T ′ ∣ = κ − ρ C ( T ′ ) ≥ κ − a ∗ r > 0.
Hence D T ′ D_{T'} D T ′ is nonzero. Since d ( D T ′ ) ≥ d d(D_{T'}) \ge d d ( D T ′ ) ≥ d , its length satisfies N ∗ ≥ d N_* \ge d N ∗ ≥ d . By the Singleton estimate, we therefore have k a d d ( q 2 ) ( N ∗ , d ) ≤ N ∗ − d + 1 k_{\mathrm{add}}^{(q^2)}(N_*,d) \le N_*-d+1 k add ( q 2 ) ( N ∗ , d ) ≤ N ∗ − d + 1 . At a = a ∗ a=a_* a = a ∗ , (43) yields
Λ ≤ 2 a ∗ r + 2 k a d d ( q 2 ) ( N ∗ , d ) − n \Lambda\le2a_*r+2k_{\mathrm{add}}^{(q^2)}(N_*,d)-n Λ ≤ 2 a ∗ r + 2 k add ( q 2 ) ( N ∗ , d ) − n
≤ 2 a ∗ r + 2 ( N ∗ − d + 1 ) − n = n − 2 ( d − 1 ) − 2 a ∗ ( δ − 1 ) . \begin{aligned}
\le2a_*r+2(N_*-d+1)-n \\
&= n-2(d-1)-2a_*(\delta-1).
\end{aligned} ≤ 2 a ∗ r + 2 ( N ∗ − d + 1 ) − n = n − 2 ( d − 1 ) − 2 a ∗ ( δ − 1 ) .
This is equivalent to (42). If k a d d ( q 2 ) ( N ∗ , d ) < N ∗ − d + 1 k_{\mathrm{add}}^{(q^2)}(N_*,d)<N_*-d+1 k add ( q 2 ) ( N ∗ , d ) < N ∗ − d + 1 , the bound at a = a ∗ a=a_* a = a ∗ is strictly stronger.
Explicit Griesmer, Hamming and Plotkin Bounds
Let C = ψ ( S ⊥ t ) ≤ F q 2 [ n ] C=\psi(S^{\perp_t}) \le\mathbb{F}_{q^2}^{[n]} C = ψ ( S ⊥ t ) ≤ F q 2 [ n ] be the additive classical code in Proposition V.2. Recall that κ = ( n + Λ ) / 2 \kappa=(n+\Lambda)/2 κ = ( n + Λ ) /2 and ℓ = r + δ − 1 \ell=r+\delta-1 ℓ = r + δ − 1 , and define
A + = { a ∈ Z ≥ 0 : a r < κ , a ℓ < n } , \mathcal{A}_+ = \{a \in\mathbb{Z}_{\ge0}: ar<\kappa,\ a\ell<n\}, A + = { a ∈ Z ≥ 0 : a r < κ , a ℓ < n } ,
N a = n − a ℓ , b a = 2 m ( κ − a r ) , M a = p b a = q n + Λ − 2 a r . N_a=n-a\ell,\quad b_a=2m(\kappa-ar),\quad M_a=p^{b_a}=q^{n+\Lambda-2ar}. N a = n − a ℓ , b a = 2 m ( κ − a r ) , M a = p b a = q n + Λ − 2 a r .
For each a ∈ A + a \in\mathcal{A}_+ a ∈ A + , by Lemma V.3, we can choose a set T ′ T' T ′ with ∣ T ′ ∣ = a ℓ |T'|=a\ell ∣ T ′ ∣ = a ℓ and ρ C ( T ′ ) ≤ a r \rho_C(T')\le ar ρ C ( T ′ ) ≤ a r . We select all codewords of C C C that vanish on T ′ T' T ′ and delete these coordinates from each selected codeword. The resulting shortened code D a = sh ( T ′ ) c ( C ) D_a=\operatorname{sh}_{(T')^c}(C) D a = sh ( T ′ ) c ( C ) has length n − ∣ T ′ ∣ = n − a ℓ = N a n-|T'|=n-a\ell=N_a n − ∣ T ′ ∣ = n − a ℓ = N a . By rank-nullity (1), we have
dim p D a = 2 m ( κ − ρ C ( T ′ ) ) ≥ 2 m ( κ − a r ) = b a . \dim_p D_a=2m(\kappa-\rho_C(T'))\ge2m(\kappa-ar)=b_a. dim p D a = 2 m ( κ − ρ C ( T ′ )) ≥ 2 m ( κ − a r ) = b a .
Since ∣ D a ∣ = p dim p D a |D_a| = p^{\dim_p D_a} ∣ D a ∣ = p d i m p D a and shortening does not decrease the minimum distance, we obtain
dim p D a ≥ b a , ∣ D a ∣ ≥ M a , d ( D a ) ≥ d . (47) \dim_p D_a \ge b_a,\quad|D_a| \ge M_a,\quad d(D_a) \ge d. \tag*{(47)} dim p D a ≥ b a , ∣ D a ∣ ≥ M a , d ( D a ) ≥ d . (47)
Theorem V.10. Let Q Q Q be a pure ( r , δ ) (r,\delta) ( r , δ ) -SSLRC with parameters ( ( n , K , R , d ) ) q ((n,K,R,d))_q (( n , K , R , d ) ) q , where q = p m q = p^m q = p m and K > 1 K > 1 K > 1 . Set L p = ( p 2 m − 1 ) / ( p − 1 ) L_p = (p^{2m} - 1)/(p - 1) L p = ( p 2 m − 1 ) / ( p − 1 ) and ω p = p 2 m − 1 \omega_p = p^{2m-1} ω p = p 2 m − 1 . For every a ∈ A + a \in\mathcal{A}_+ a ∈ A + ,
n ≥ a ( r + δ − 1 ) + ⌈ 1 L p ∑ i = 0 b a − 1 ⌈ ω p d p i ⌉ ⌉ . (48) n \ge a(r+\delta-1) + \left\lceil\frac{1}{L_p}\sum_{i=0}^{b_a-1}\left\lceil\frac{\omega_p d}{p^i}\right\rceil\right\rceil. \tag*{(48)} n ≥ a ( r + δ − 1 ) + ⌈ L p 1 i = 0 ∑ b a − 1 ⌈ p i ω p d ⌉ ⌉ . (48) If b a ≥ 2 m b_a \ge2m b a ≥ 2 m , (47) is equivalent to
n ≥ a ( r + δ − 1 ) + d + ⌈ p − 1 p 2 m − 1 ∑ j = 1 b a − 2 m ⌈ d p j ⌉ ⌉ . (49) n \ge a(r+\delta-1) + d + \left\lceil\frac{p-1}{p^{2m}-1}\sum_{j=1}^{b_a-2m}\left\lceil\frac{d}{p^j}\right\rceil\right\rceil. \tag*{(49)} n ≥ a ( r + δ − 1 ) + d + ⌈ p 2 m − 1 p − 1 j = 1 ∑ b a − 2 m ⌈ p j d ⌉ ⌉ . (49) If C = ψ ( S ⊥ t ) C = \psi(S^{\perp t}) C = ψ ( S ⊥ t ) is F q 2 \mathbb{F}_{q^2} F q 2 -linear, then h a = κ − a r h_a = \kappa-ar h a = κ − a r is a positive integer and
n ≥ a ( r + δ − 1 ) + ∑ i = 0 h a − 1 ⌈ d q 2 i ⌉ . (50) n \ge a(r+\delta-1) + \sum_{i=0}^{h_a-1}\left\lceil\frac{d}{q^{2i}}\right\rceil. \tag*{(50)} n ≥ a ( r + δ − 1 ) + i = 0 ∑ h a − 1 ⌈ q 2 i d ⌉ . (50)
Proof. Fix a ∈ A + a \in\mathcal{A}_+ a ∈ A + . By (46), D a D_a D a contains an F p \mathbb{F}_p F p -linear subcode D a ′ D'_a D a ′ of dimension b a b_a b a . Its minimum distance is at least d d d . Choose an F p \mathbb{F}_p F p -basis of F q 2 \mathbb{F}_{q^2} F q 2 and identify each symbol with its coordinate vector in F p 2 m \mathbb{F}_p^{2m} F p 2 m . We concatenate D a ′ D'_a D a ′ with a p p p -ary simplex code with parameters [ L p , 2 m , ω p ] p [L_p,2m,\omega_p]_p [ L p , 2 m , ω p ] p , as in the proof of [4 Theorem 12] . The simplex encoder is F p \mathbb{F}_p F p -linear and injective, and every nonzero input vector is mapped to a codeword of weight ω p \omega_p ω p . The concatenated code therefore has parameters
[ L p N a , b a , d ′ ] p , d a ′ ≥ ω p d . [L_pN_a,b_a,d']_p,\quad d'_a \ge\omega_p d. [ L p N a , b a , d ′ ] p , d a ′ ≥ ω p d .
By the classical Griesmer bound [32 ] ,
L p N a ≥ ∑ i = 0 b a − 1 ⌈ d a ′ p i ⌉ ≥ ∑ i = 0 b a − 1 ⌈ ω p d p i ⌉ . L_pN_a \ge\sum_{i=0}^{b_a-1}\left\lceil\frac{d'_a}{p^i}\right\rceil\ge\sum_{i=0}^{b_a-1}\left\lceil\frac{\omega_p d}{p^i}\right\rceil. L p N a ≥ i = 0 ∑ b a − 1 ⌈ p i d a ′ ⌉ ≥ i = 0 ∑ b a − 1 ⌈ p i ω p d ⌉ .
We divide by L p L_p L p and take the ceiling, since N a N_a N a is an integer. Using n = a ℓ + N a n = a\ell+ N_a n = a ℓ + N a , we obtain (47).
Suppose b a ≥ 2 m b_a \ge2m b a ≥ 2 m . Since ω p = p 2 m − 1 \omega_p = p^{2m-1} ω p = p 2 m − 1 , the first 2 m 2m 2 m terms and the remaining terms satisfy
∑ i = 0 2 m − 1 ⌈ p 2 m − 1 d p i ⌉ = d ∑ j = 0 2 m − 1 p j = L p d , \sum_{i=0}^{2m-1}\left\lceil\frac{p^{2m-1}d}{p^i}\right\rceil= d\sum_{j=0}^{2m-1}p^j = L_p d, i = 0 ∑ 2 m − 1 ⌈ p i p 2 m − 1 d ⌉ = d j = 0 ∑ 2 m − 1 p j = L p d ,
∑ i = 2 m b a − 1 ⌈ p 2 m − 1 d p i ⌉ = ∑ j = 1 b a − 2 m ⌈ d p j ⌉ . \sum_{i=2m}^{b_a-1}\left\lceil\frac{p^{2m-1}d}{p^i}\right\rceil= \sum_{j=1}^{b_a-2m}\left\lceil\frac{d}{p^j}\right\rceil. i = 2 m ∑ b a − 1 ⌈ p i p 2 m − 1 d ⌉ = j = 1 ∑ b a − 2 m ⌈ p j d ⌉ .
The second sum is zero when b a = 2 m b_a = 2m b a = 2 m . Since d d d is an integer, we have
⌈ 1 L p ∑ i = 0 b a − 1 ⌈ ω p d p i ⌉ ⌉ = d + ⌈ 1 L p ∑ j = 1 b a − 2 m ⌈ d p j ⌉ ⌉ . \left\lceil\frac{1}{L_p}\sum_{i=0}^{b_a-1}\left\lceil\frac{\omega_p d}{p^i}\right\rceil\right\rceil= d + \left\lceil\frac{1}{L_p}\sum_{j=1}^{b_a-2m}\left\lceil\frac{d}{p^j}\right\rceil\right\rceil. ⌈ L p 1 i = 0 ∑ b a − 1 ⌈ p i ω p d ⌉ ⌉ = d + ⌈ L p 1 j = 1 ∑ b a − 2 m ⌈ p j d ⌉ ⌉ .
Thus (47) takes the form (48), since 1 / L p = ( p − 1 ) / ( p 2 m − 1 ) 1/L_p = (p-1)/(p^{2m}-1) 1/ L p = ( p − 1 ) / ( p 2 m − 1 ) .
If C C C is F q 2 \mathbb{F}_{q^2} F q 2 -linear, so is D a D_a D a , and dim q 2 D a ≥ κ − a r = h a \dim_{q^2}D_a \ge\kappa-ar = h_a dim q 2 D a ≥ κ − a r = h a . Choose an h a h_a h a -dimensional F q 2 \mathbb{F}_{q^2} F q 2 -linear subcode of D a D_a D a . Since its minimum distance is at least d d d , the classical Griesmer bound over F q 2 \mathbb{F}_{q^2} F q 2 implies
N a ≥ ∑ i = 0 h a − 1 ⌈ d q 2 i ⌉ . N_a \ge\sum_{i=0}^{h_a-1}\left\lceil\frac{d}{q^{2i}}\right\rceil. N a ≥ i = 0 ∑ h a − 1 ⌈ q 2 i d ⌉ .
Since N a = n − a ( r + δ − 1 ) N_a = n-a(r+\delta-1) N a = n − a ( r + δ − 1 ) , we obtain (49). □
Theorem V.12. Let Q Q Q be a pure ( r , δ ) (r,\delta) ( r , δ ) -SSLRC with parameters ( ( n , K , R , d ) ) q ((n,K,R,d))_q (( n , K , R , d ) ) q , where q = p m q = p^m q = p m and K > 1 K > 1 K > 1 . Let t = ⌊ ( d − 1 ) / 2 ⌋ t = \lfloor(d-1)/2\rfloor t = ⌊( d − 1 ) /2 ⌋ . For every a ∈ A + a \in\mathcal{A}_+ a ∈ A + ,
Λ + 2 a ( δ − 1 ) + log q V q 2 ( N a , t ) ≤ n , \Lambda+ 2a(\delta-1) + \log_q V_{q^2}(N_a,t) \le n, Λ + 2 a ( δ − 1 ) + log q V q 2 ( N a , t ) ≤ n , d ≤ ⌊ ( q 2 − 1 ) N a M a q 2 ( M a − 1 ) ⌋ , (51) d \le\left\lfloor\frac{(q^2-1)N_aM_a}{q^2(M_a-1)}\right\rfloor, \tag*{(51)} d ≤ ⌊ q 2 ( M a − 1 ) ( q 2 − 1 ) N a M a ⌋ , (51)
where V q 2 ( N a , t ) V_{q^{2}}(N_a,t) V q 2 ( N a , t ) is the Hamming-ball volume defined in Remark IV.9.
Proof. Fix a ∈ A + a \in A_{+} a ∈ A + , and let Q = q 2 Q=q^{2} Q = q 2 and M = ∣ D a ∣ M=|D_a| M = ∣ D a ∣ . By (46), M ≥ M a > 1 M \ge M_a > 1 M ≥ M a > 1 and d ( D a ) ≥ d d(D_a) \ge d d ( D a ) ≥ d . Since 2 t < d 2t<d 2 t < d , the Hamming balls of radius t t t centered at distinct codewords of D a D_a D a are disjoint. By the classical Hamming bound [32 ] , we have
M a V Q ( N a , t ) ≤ M V Q ( N a , t ) ≤ Q N a . M_a V_Q(N_a,t) \le M V_Q(N_a,t) \le Q^{N_a}. M a V Q ( N a , t ) ≤ M V Q ( N a , t ) ≤ Q N a .
We take base-q q q logarithms and use M a = q n + Λ − 2 a r M_a=q^{n+\Lambda-2ar} M a = q n + Λ − 2 a r and N a = n − a ℓ N_a=n-a\ell N a = n − a ℓ to obtain
n + Λ − 2 a r + log q V Q ( N a , t ) ≤ 2 ( n − a ℓ ) , n+\Lambda-2ar+\log_q V_Q(N_a,t) \le2(n-a\ell), n + Λ − 2 a r + log q V Q ( N a , t ) ≤ 2 ( n − a ℓ ) ,
Λ + 2 a ( ℓ − r ) + log q V Q ( N a , t ) ≤ n . \Lambda+2a(\ell-r)+\log_q V_Q(N_a,t) \le n. Λ + 2 a ( ℓ − r ) + log q V Q ( N a , t ) ≤ n .
Since ℓ − r = δ − 1 \ell-r=\delta-1 ℓ − r = δ − 1 , we obtain the stated Hamming bound.
By the distance form of the classical Plotkin bound [32 ] , we have
d ≤ d ( D a ) ≤ ( Q − 1 ) N a M Q ( M − 1 ) . d \le d(D_a) \le\frac{(Q-1)N_aM}{Q(M-1)}. d ≤ d ( D a ) ≤ Q ( M − 1 ) ( Q − 1 ) N a M .
The inequality M ≥ M a > 1 M \ge M_a>1 M ≥ M a > 1 implies
M M − 1 − M a M a − 1 = M a − M ( M − 1 ) ( M a − 1 ) ≤ 0. \frac{M}{M-1}-\frac{M_a}{M_a-1}
=\frac{M_a-M}{(M-1)(M_a-1)} \le0. M − 1 M − M a − 1 M a = ( M − 1 ) ( M a − 1 ) M a − M ≤ 0.
Consequently,
d ≤ ( Q − 1 ) N a M a Q ( M a − 1 ) . d \le\frac{(Q-1)N_aM_a}{Q(M_a-1)}. d ≤ Q ( M a − 1 ) ( Q − 1 ) N a M a .
Since d d d is an integer, we take the floor of the right-hand side and substitute Q = q 2 Q=q^2 Q = q 2 to obtain (50).
Some Constructions of Optimal F q \mathbb{F}_q F q -Linear Pure ( r , δ ) (r,\delta) ( r , δ ) -SSLRCs
In this section, we construct nine classes of F q \mathbb{F}_q F q -linear pure ( r , δ ) (r,\delta) ( r , δ ) -SSLRCs from MDS stabilizer codes, RS codes, and linear codes with MDS Hermitian hulls. The explicit families have d = δ d=\delta d = δ and simultaneously attain the Singleton-type bound (41), the alphabet-dependent bound (43), the F q 2 \mathbb{F}_{q^2} F q 2 -linear Griesmer bound (49), and the disjoint redundancy bound (32). These optimal constructions are summarized in Table I , where q q q is a prime power, B ≥ 1 B \ge1 B ≥ 1 , and all remaining parameters are integers.
Tensor Products and Dimension Trading
In this subsection, we determine the parameters of tensor products of SSCs and give conditions under which their coordinate blocks form repair sets. We then show how to trade the logical and gauge dimensions of pure ( r , δ ) (r,\delta) ( r , δ ) -SSLRCs while preserving the minimum distance and the repair cover.
Proposition VI.1. Let B ≥ 1 B \ge1 B ≥ 1 be an integer. For each j ∈ [ B ] j \in[B] j ∈ [ B ] , let Q j Q_j Q j be an SSC over F q \mathbb{F}_q F q with parameters [ [ ℓ j , k j , g j , d j ] ] q [[\ell_j,k_j,g_j,d_j]]_q [[ ℓ j , k j , g j , d j ] ] q , where k j > 0 k_j>0 k j > 0 . Set n = ∑ j = 1 B ℓ j n=\sum_{j=1}^{B}\ell_j n = ∑ j = 1 B ℓ j and Q = ⨂ j = 1 B Q j Q=\bigotimes_{j=1}^{B}Q_j Q = ⨂ j = 1 B Q j . Let Γ j \Gamma_j Γ j denote the coordinate block of Q j Q_j Q j in Q Q Q so that [ n ] = ⋃ j = 1 B Γ j [n]=\bigcup_{j=1}^{B}\Gamma_j [ n ] = ⋃ j = 1 B Γ j and ∣ Γ j ∣ = ℓ j |\Gamma_j|=\ell_j ∣ Γ j ∣ = ℓ j . Then the following statements hold.
(i) Q Q Q is an SSC with parameters
[ [ n , ∑ j = 1 B k j , ∑ j = 1 B g j , min j ∈ [ B ] d j ] ] q . \left[\left[n,\sum_{j=1}^{B}k_j,\sum_{j=1}^{B}g_j,\min_{j\in[B]}d_j\right]\right]_q. [ [ n , j = 1 ∑ B k j , j = 1 ∑ B g j , j ∈ [ B ] min d j ] ] q . (ii) If each Q j Q_j Q j is pure, then Q Q Q is pure.
(iii) If r ≥ 1 r \ge1 r ≥ 1 and δ ≥ 2 \delta\ge2 δ ≥ 2 are integers satisfying δ ≤ min j ∈ [ B ] d j \delta\le\min_{j\in[B]}d_j δ ≤ min j ∈ [ B ] d j and δ ≤ ℓ j ≤ r + δ − 1 \delta\le\ell_j \le r+\delta-1 δ ≤ ℓ j ≤ r + δ − 1 for every j ∈ [ B ] j \in[B] j ∈ [ B ] , then Q Q Q is an ( r , δ ) (r,\delta) ( r , δ ) -SSLRC with Γ 1 , … , Γ B \Gamma_1,\ldots,\Gamma_B Γ 1 , … , Γ B as repair sets.
Proof. For each j ∈ [ B ] j \in[B] j ∈ [ B ] , let G j ≤ F q ℓ j × F q ℓ j G_j \le\mathbb{F}_q^{\ell_j}\times\mathbb{F}_q^{\ell_j} G j ≤ F q ℓ j × F q ℓ j be the gauge label space of Q j Q_j Q j , and set S j = G j ∩ G j ⊥ t S_j=G_j\cap G_j^{\perp_t} S j = G j ∩ G j ⊥ t , where G j ⊥ t G_j^{\perp_t} G j ⊥ t and S j ⊥ t S_j^{\perp_t} S j ⊥ t denote the orthogonal complements taken in F q ℓ j × F q ℓ j \mathbb{F}_q^{\ell_j}\times\mathbb{F}_q^{\ell_j} F q ℓ j × F q ℓ j . We extend G j G_j G j , S j S_j S j , and their orthogonal complements by zero outside Γ j \Gamma_j Γ j . The tensor product Q Q Q has gauge label space G = ⨁ j = 1 B G j G=\bigoplus_{j=1}^{B}G_j G = ⨁ j = 1 B G j . Since labels supported on distinct blocks are orthogonal under the trace-symplectic form, we obtain
G ⊥ t = ⨁ j = 1 B G j ⊥ t , G^{\perp_t}=\bigoplus_{j=1}^{B}G_j^{\perp_t}, G ⊥ t = j = 1 ⨁ B G j ⊥ t ,
S = G ∩ G ⊥ t = ⨁ j = 1 B ( G j ∩ G j ⊥ t ) = ⨁ j = 1 B S j , S=G\cap G^{\perp_t}=\bigoplus_{j=1}^{B}(G_j\cap G_j^{\perp_t})=\bigoplus_{j=1}^{B}S_j, S = G ∩ G ⊥ t = j = 1 ⨁ B ( G j ∩ G j ⊥ t ) = j = 1 ⨁ B S j ,
S ⊥ t = ⨁ j = 1 B S j ⊥ t . S^{\perp_t}=\bigoplus_{j=1}^{B}S_j^{\perp_t}. S ⊥ t = j = 1 ⨁ B S j ⊥ t .
By (8) and the additivity of dimensions under direct sums,
K = ∏ j = 1 B q k j = q ∑ j = 1 B k j , R = ∏ j = 1 B q g j = q ∑ j = 1 B g j . K=\prod_{j=1}^{B}q^{k_j}=q^{\sum_{j=1}^{B}k_j},\qquad R=\prod_{j=1}^{B}q^{g_j}=q^{\sum_{j=1}^{B}g_j}. K = j = 1 ∏ B q k j = q ∑ j = 1 B k j , R = j = 1 ∏ B q g j = q ∑ j = 1 B g j .
Let w = ∑ j = 1 B w j ∈ S ⊥ t ∖ G w=\sum_{j=1}^{B}w_j\in S^{\perp_t}\setminus G w = ∑ j = 1 B w j ∈ S ⊥ t ∖ G , where w j ∈ S j ⊥ t w_j\in S_j^{\perp_t} w j ∈ S j ⊥ t . Then w j ∉ G j w_j\notin G_j w j ∈ / G j for some j ∈ [ B ] j\in[B] j ∈ [ B ] . By (11),
swt ( w ) = ∑ i = 1 B swt ( w i ) ≥ swt ( w j ) ≥ d j ≥ min i ∈ [ B ] d i , \operatorname{swt}(w)=\sum_{i=1}^{B}\operatorname{swt}(w_i)\ge\operatorname{swt}(w_j)\ge d_j\ge\min_{i\in[B]}d_i, swt ( w ) = i = 1 ∑ B swt ( w i ) ≥ swt ( w j ) ≥ d j ≥ i ∈ [ B ] min d i ,
which implies that d ( Q ) ≥ min i ∈ [ B ] d i d(Q)\ge\min_{i\in[B]}d_i d ( Q ) ≥ min i ∈ [ B ] d i . Conversely, choose j 0 ∈ [ B ] j_0\in[B] j 0 ∈ [ B ] with d j 0 = min j ∈ [ B ] d j d_{j_0}=\min_{j\in[B]}d_j d j 0 = min j ∈ [ B ] d j . Since k j 0 > 0 k_{j_0}>0 k j 0 > 0 , we may choose w 0 ∈ S j 0 ⊥ t ∖ G j 0 w_0\in S_{j_0}^{\perp_t}\setminus G_{j_0} w 0 ∈ S j 0 ⊥ t ∖ G j 0 of weight d j 0 d_{j_0} d j 0 by (11). Its zero extension belongs to S ⊥ t ∖ G S^{\perp_t}\setminus G S ⊥ t ∖ G and has the same weight. Hence d ( Q ) = min j ∈ [ B ] d j d(Q)=\min_{j\in[B]}d_j d ( Q ) = min j ∈ [ B ] d j . This proves (i).
Suppose that every Q j Q_j Q j is pure. For any nonzero v = ∑ j = 1 B v j ∈ G v=\sum_{j=1}^{B}v_j\in G v = ∑ j = 1 B v j ∈ G , choose j j j with v j ≠ 0 v_j\ne0 v j = 0 . By Definition II.18,
swt ( v ) = ∑ i = 1 B swt ( v i ) ≥ swt ( v j ) ≥ d j ≥ d ( Q ) . \operatorname{swt}(v)=\sum_{i=1}^{B}\operatorname{swt}(v_i)\ge\operatorname{swt}(v_j)\ge d_j\ge d(Q). swt ( v ) = i = 1 ∑ B swt ( v i ) ≥ swt ( v j ) ≥ d j ≥ d ( Q ) .
Thus Q Q Q is pure, proving (ii).
To prove (iii), fix j ∈ [ B ] j\in[B] j ∈ [ B ] and ∅ ≠ E ⊆ Γ j \varnothing\ne E\subseteq\Gamma_j ∅ = E ⊆ Γ j with ∣ E ∣ < δ |E|<\delta ∣ E ∣ < δ . Since ∣ E ∣ < δ ≤ d j |E|<\delta\le d_j ∣ E ∣ < δ ≤ d j , no label in S j ⊥ t ∖ G j S_j^{\perp_t}\setminus G_j S j ⊥ t ∖ G j is supported in E E E by (11). Together with G j ⊆ S j ⊥ t G_j\subseteq S_j^{\perp_t} G j ⊆ S j ⊥ t , this implies W E ∩ S j ⊥ t = ( G j ) E W_E\cap S_j^{\perp_t}=(G_j)_E W E ∩ S j ⊥ t = ( G j ) E . Since S Γ j = S j S_{\Gamma_j}=S_j S Γ j = S j , the orthogonal complement of S Γ j S_{\Gamma_j} S Γ j in W W W satisfies
S Γ j ⊥ t = S j ⊥ t ⊕ W Γ j c . S_{\Gamma_j}^{\perp_t}=S_j^{\perp_t}\oplus W_{\Gamma_j^c}. S Γ j ⊥ t = S j ⊥ t ⊕ W Γ j c .
Since E ⊆ Γ j E\subseteq\Gamma_j E ⊆ Γ j and G = ⨁ i = 1 B G i G=\bigoplus_{i=1}^{B}G_i G = ⨁ i = 1 B G i , we have
W E ∩ S Γ j ⊥ t = W E ∩ S j ⊥ t = ( G j ) E = G E . W_E\cap S_{\Gamma_j}^{\perp_t}=W_E\cap S_j^{\perp_t}=(G_j)_E=G_E. W E ∩ S Γ j ⊥ t = W E ∩ S j ⊥ t = ( G j ) E = G E .
By condition 2) of Theorem III.8, E E E is recoverable within Γ j \Gamma_j Γ j . The blocks partition [ n ] [n] [ n ] and satisfy δ ≤ ∣ Γ j ∣ ≤ r + δ − 1 \delta\leq|\Gamma_j| \leq r+\delta-1 δ ≤ ∣ Γ j ∣ ≤ r + δ − 1 . Therefore, Q \mathcal{Q} Q is an ( r , δ ) (r,\delta) ( r , δ ) -SSLRC with the stated repair sets. This completes the proof.
By [1 Theorem 7] , an F q \mathbb{F}_q F q -linear MDS stabilizer code with parameters [ [ n , k 0 , d ] ] q [[n,k_0,d]]_q [[ n , k 0 , d ] ] q can be converted into a pure F q \mathbb{F}_q F q -linear SSC with parameters [ [ n , k 0 − γ , γ , d ] ] q [[n,k_0-\gamma,\gamma,d]]_q [[ n , k 0 − γ , γ , d ] ] q for 0 ≤ γ < k 0 0 \leq\gamma< k_0 0 ≤ γ < k 0 . We next apply the symplectic construction underlying [1 Theorems 2 and 3] to trade dimensions in pure F q \mathbb{F}_q F q -linear ( r , δ ) (r,\delta) ( r , δ ) -SSLRCs with a prescribed repair cover.
Proposition VI.2. Let Q \mathcal{Q} Q be a pure F q \mathbb{F}_q F q -linear ( r , δ ) (r,\delta) ( r , δ ) -SSLRC with parameters [ [ n , k , g , d ] ] q [[n,k,g,d]]_q [[ n , k , g , d ] ] q , where k > 0 k>0 k > 0 . Let S S S be its stabilizer label space, and fix a repair cover R \mathcal{R} R of Q \mathcal{Q} Q . For every integer 0 ≤ γ ≤ k + g − 1 0 \leq\gamma\leq k+g-1 0 ≤ γ ≤ k + g − 1 , there exists a pure F q \mathbb{F}_q F q -linear ( r , δ ) (r,\delta) ( r , δ ) -SSLRC Q γ \mathcal{Q}_\gamma Q γ with parameters [ [ n , k + g − γ , γ , d ] ] q [[n,k+g-\gamma,\gamma,d]]_q [[ n , k + g − γ , γ , d ] ] q , stabilizer label space S S S , and repair cover R \mathcal{R} R .
Proof. We first construct F q \mathbb{F}_q F q -linear spaces G γ G_\gamma G γ with G γ ∩ G γ ⊥ t = S G_\gamma\cap G_\gamma^{\perp_t}=S G γ ∩ G γ ⊥ t = S . Let G G G be the gauge label space of Q \mathcal{Q} Q . Since S = G ∩ G ⊥ t S=G\cap G^{\perp_t} S = G ∩ G ⊥ t , we have S ⊆ G S\subseteq G S ⊆ G . For any v ∈ G v\in G v ∈ G and s ∈ S ⊆ G ⊥ t s\in S\subseteq G^{\perp_t} s ∈ S ⊆ G ⊥ t , we have [ v , s ] t = − [ s , v ] t = 0 [v,s]_t=-[s,v]_t=0 [ v , s ] t = − [ s , v ] t = 0 . Hence G ⊆ S ⊥ t G\subseteq S^{\perp_t} G ⊆ S ⊥ t . By Definition II.18, every nonzero label in G G G has weight at least d d d . By (11), every label in S ⊥ t ∖ G S^{\perp_t}\setminus G S ⊥ t ∖ G also has weight at least d d d . Therefore,
swt ( v ) ≥ d for every 0 ≠ v ∈ S ⊥ t . (52) \operatorname{swt}(v)\geq d\text{ for every }0\neq v\in S^{\perp_t}. \tag*{(52)} swt ( v ) ≥ d for every 0 = v ∈ S ⊥ t . (52)
Since k > 0 k>0 k > 0 , choose w ∗ ∈ S ⊥ t ∖ G w_*\in S^{\perp_t}\setminus G w ∗ ∈ S ⊥ t ∖ G with swt ( w ∗ ) = d \operatorname{swt}(w_*)=d swt ( w ∗ ) = d by (11). Since S ⊆ G S\subseteq G S ⊆ G , we have w ∗ ∉ S w_*\notin S w ∗ ∈ / S . By Corollary III.10, we have d ≥ δ ≥ 2 d\geq\delta\geq2 d ≥ δ ≥ 2 . Hence S ≠ { 0 } S\neq\{0\} S = { 0 } ; otherwise, S ⊥ t = W S^{\perp_t}=W S ⊥ t = W would contain a label of weight one, contrary to (51). Since S S S is F q \mathbb{F}_q F q -linear, we have S ⊥ t = S ⊥ s S^{\perp_t}=S^{\perp_s} S ⊥ t = S ⊥ s . By (12) and (5), we obtain
dim q S = n − k − g , dim q S ⊥ s − dim q S = 2 ( k + g ) . \dim_q S=n-k-g,\qquad\dim_q S^{\perp_s}-\dim_q S=2(k+g). dim q S = n − k − g , dim q S ⊥ s − dim q S = 2 ( k + g ) .
Moreover, S ⊥ s ∩ ( S ⊥ s ) ⊥ s = S S^{\perp_s}\cap(S^{\perp_s})^{\perp_s}=S S ⊥ s ∩ ( S ⊥ s ) ⊥ s = S . Set x 1 = w ∗ x_1=w_* x 1 = w ∗ . Since ( S ⊥ s ) ⊥ s = S (S^{\perp_s})^{\perp_s}=S ( S ⊥ s ) ⊥ s = S and x 1 ∉ S x_1\notin S x 1 ∈ / S , choose z 1 ∈ S ⊥ s z_1\in S^{\perp_s} z 1 ∈ S ⊥ s with [ x 1 , z 1 ] s = 1 [x_1,z_1]_s=1 [ x 1 , z 1 ] s = 1 . By symplectic basis extension, choose x i , z i ∈ S ⊥ s x_i,z_i\in S^{\perp_s} x i , z i ∈ S ⊥ s for 2 ≤ i ≤ k + g 2\leq i\leq k+g 2 ≤ i ≤ k + g such that
S ⊥ s = S ⊕ span F q { x i , z i : 1 ≤ i ≤ k + g } S^{\perp_s}=S\oplus\operatorname{span}_{\mathbb{F}_q}\{x_i,z_i:1\leq i\leq k+g\} S ⊥ s = S ⊕ span F q { x i , z i : 1 ≤ i ≤ k + g }
and for 1 ≤ i , j ≤ k + g 1\leq i,j\leq k+g 1 ≤ i , j ≤ k + g ,
[ x i , z j ] s = δ i j , [ x i , x j ] s = [ z i , z j ] s = 0. [x_i,z_j]_s=\delta_{ij},\qquad[x_i,x_j]_s=[z_i,z_j]_s=0. [ x i , z j ] s = δ ij , [ x i , x j ] s = [ z i , z j ] s = 0.
For 0 ≤ γ ≤ k + g − 1 0\leq\gamma\leq k+g-1 0 ≤ γ ≤ k + g − 1 , define
G γ = S ⊕ span F q { x i , z i : 2 ≤ i ≤ γ + 1 } , (53) G_\gamma=S\oplus\operatorname{span}_{\mathbb{F}_q}\{x_i,z_i:2\leq i\leq\gamma+1\}, \tag*{(53)} G γ = S ⊕ span F q { x i , z i : 2 ≤ i ≤ γ + 1 } , (53)
where the span of the empty set is { 0 } \{0\} { 0 } . Since S ⊆ G γ ⊆ S ⊥ s S\subseteq G_\gamma\subseteq S^{\perp_s} S ⊆ G γ ⊆ S ⊥ s , we have S ⊆ G γ ∩ G γ ⊥ s S\subseteq G_\gamma\cap G_\gamma^{\perp_s} S ⊆ G γ ∩ G γ ⊥ s . Conversely, let v ∈ G γ ∩ G γ ⊥ s v\in G_\gamma\cap G_\gamma^{\perp_s} v ∈ G γ ∩ G γ ⊥ s and write
v = s + ∑ i = 2 γ + 1 ( a i x i + b i z i ) , v=s+\sum_{i=2}^{\gamma+1}(a_i x_i+b_i z_i), v = s + i = 2 ∑ γ + 1 ( a i x i + b i z i ) ,
where s ∈ S s\in S s ∈ S and a i , b i ∈ F q a_i,b_i\in\mathbb{F}_q a i , b i ∈ F q . For every 2 ≤ j ≤ γ + 1 2\leq j\leq\gamma+1 2 ≤ j ≤ γ + 1 , we have
a j = [ v , z j ] s = 0 , − b j = [ v , x j ] s = 0. a_j=[v,z_j]_s=0,\qquad-b_j=[v,x_j]_s=0. a j = [ v , z j ] s = 0 , − b j = [ v , x j ] s = 0.
These equalities follow from the symplectic basis relations and v ∈ G γ ⊥ s v\in G_\gamma^{\perp_s} v ∈ G γ ⊥ s . Hence v = s ∈ S v=s\in S v = s ∈ S . Since G γ G_\gamma G γ is F q \mathbb{F}_q F q -linear, we conclude that
G γ ∩ G γ ⊥ t = G γ ∩ G γ ⊥ s = S . G_\gamma\cap G_\gamma^{\perp_t}=G_\gamma\cap G_\gamma^{\perp_s}=S. G γ ∩ G γ ⊥ t = G γ ∩ G γ ⊥ s = S .
We next determine the logical and gauge dimensions. By [1 Theorem 1] , we obtain an F q \mathbb{F}_q F q -linear SSC Q γ \mathcal{Q}_\gamma Q γ with gauge label space G γ G_\gamma G γ and stabilizer label space S S S . By (52), we have
dim q G γ = dim q S + 2 γ = n − k − g + 2 γ . \dim_q G_\gamma=\dim_q S+2\gamma=n-k-g+2\gamma. dim q G γ = dim q S + 2 γ = n − k − g + 2 γ .
By (12), we obtain
k γ = n − dim q G γ + dim q S 2 = k + g − γ > 0 , k_\gamma=n-\frac{\dim_q G_\gamma+\dim_q S}{2}=k+g-\gamma>0, k γ = n − 2 dim q G γ + dim q S = k + g − γ > 0 ,
g γ = dim q G γ − dim q S 2 = γ . g_\gamma=\frac{\dim_q G_\gamma-\dim_q S}{2}=\gamma. g γ = 2 dim q G γ − dim q S = γ .
We next prove that d ( Q γ ) = d d(\mathcal{Q}_\gamma)=d d ( Q γ ) = d and that Q γ \mathcal{Q}_\gamma Q γ is pure. By the direct sum decomposition of S ⊥ s S^{\perp_s} S ⊥ s , we obtain w ∗ = x 1 ∉ G γ w_*=x_1\notin G_\gamma w ∗ = x 1 ∈ / G γ . Thus w ∗ ∈ S ⊥ t ∖ G γ w_*\in S^{\perp_t}\setminus G_\gamma w ∗ ∈ S ⊥ t ∖ G γ . By (11) and (51), we obtain d ≤ d ( Q γ ) ≤ swt ( w ∗ ) = d d\leq d(\mathcal{Q}_\gamma)\leq\operatorname{swt}(w_*)=d d ≤ d ( Q γ ) ≤ swt ( w ∗ ) = d . Hence d ( Q γ ) = d d(\mathcal{Q}_\gamma)=d d ( Q γ ) = d . For every 0 ≠ v ∈ G γ ⊆ S ⊥ t 0\neq v\in G_\gamma\subseteq S^{\perp_t} 0 = v ∈ G γ ⊆ S ⊥ t , we also have swt ( v ) ≥ d \operatorname{swt}(v)\geq d swt ( v ) ≥ d by (51). Thus Q γ \mathcal{Q}_\gamma Q γ is pure by Definition II.18.
Finally, we prove that R \mathcal{R} R is a repair cover of Q γ \mathcal{Q}_\gamma Q γ . Fix Γ ∈ R \Gamma\in\mathcal{R} Γ ∈ R and ∅ ≠ E ⊆ Γ \varnothing\neq E\subseteq\Gamma ∅ = E ⊆ Γ with ∣ E ∣ < δ |E|<\delta ∣ E ∣ < δ . For every v ∈ W E v\in W_E v ∈ W E , we have swt ( v ) ≤ ∣ E ∣ < δ ≤ d \operatorname{swt}(v)\leq|E|<\delta\leq d swt ( v ) ≤ ∣ E ∣ < δ ≤ d . By (51), we obtain W E ∩ S ⊥ t = { 0 } W_E\cap S^{\perp_t}=\{0\} W E ∩ S ⊥ t = { 0 } . Since G γ ⊆ S ⊥ t G_\gamma\subseteq S^{\perp_t} G γ ⊆ S ⊥ t , we have G E = ( G γ ) E = { 0 } G_E=(G_\gamma)_E=\{0\} G E = ( G γ ) E = { 0 } . By condition 2) of Theorem III.8, applied to Q \mathcal{Q} Q , we obtain W E ∩ S Γ ⊥ t = G E = { 0 } = ( G γ ) E W_E\cap S_\Gamma^{\perp_t}=G_E=\{0\}=(G_\gamma)_E W E ∩ S Γ ⊥ t = G E = { 0 } = ( G γ ) E . Since Q γ \mathcal{Q}_\gamma Q γ has the same stabilizer label space S S S , we conclude by the same theorem that E E E is recoverable within Γ \Gamma Γ for Q γ \mathcal{Q}_\gamma Q γ . Since Γ \Gamma Γ and E E E were arbitrary, R \mathcal{R} R is also a repair cover of Q γ Q_\gamma Q γ . Therefore, for every 0 ≤ γ ≤ k + g − 1 0 \le\gamma\le k+g-1 0 ≤ γ ≤ k + g − 1 , there exists a pure F q \mathbb{F}_q F q -linear ( r , δ ) (r,\delta) ( r , δ ) -SSLRC with parameters [ [ n , k + g − γ , γ , d ] ] q [[n,k+g-\gamma,\gamma,d]]_q [[ n , k + g − γ , γ , d ] ] q , stabilizer label space S S S , and repair cover R \mathcal{R} R .
Combining Propositions VI.1 and VI.2, we obtain the following corollary.
Corollary VI.3. With the notation of Proposition VI.1, suppose that every Q j Q_j Q j is pure and F q \mathbb{F}_q F q -linear. Let r ≥ 1 r \ge1 r ≥ 1 and δ ≥ 2 \delta\ge2 δ ≥ 2 be integers such that δ ≤ min j ∈ [ B ] d j \delta\le\min_{j\in[B]} d_j δ ≤ min j ∈ [ B ] d j and δ ≤ ℓ j ≤ r + δ − 1 \delta\le\ell_j \le r+\delta-1 δ ≤ ℓ j ≤ r + δ − 1 for every j ∈ [ B ] j \in[B] j ∈ [ B ] . For every integer 0 ≤ γ ≤ ∑ j = 1 B ( k j + g j ) − 1 0 \le\gamma\le\sum_{j=1}^{B}(k_j+g_j)-1 0 ≤ γ ≤ ∑ j = 1 B ( k j + g j ) − 1 , there exists a pure F q \mathbb{F}_q F q -linear ( r , δ ) (r,\delta) ( r , δ ) -SSLRC with parameters
[ [ n , ∑ j = 1 B ( k j + g j ) − γ , γ , min j ∈ [ B ] d j ] ] q \left[\left[n,\sum_{j=1}^{B}(k_j+g_j)-\gamma,\gamma,\min_{j\in[B]}d_j\right]\right]_q [ [ n , j = 1 ∑ B ( k j + g j ) − γ , γ , j ∈ [ B ] min d j ] ] q and repair sets Γ 1 , … , Γ B \Gamma_1,\ldots,\Gamma_B Γ 1 , … , Γ B . Its stabilizer label space is ⨁ j = 1 B S j \bigoplus_{j=1}^{B}S_j ⨁ j = 1 B S j , where S j S_j S j is the stabilizer label space of Q j Q_j Q j .
Optimal Pure ( r , δ ) (r,\delta) ( r , δ ) -SSLRCs from MDS Stabilizer Codes
Applying Corollary VI.3 to MDS stabilizer codes, we obtain the following construction.
Theorem VI.4. Let Q 0 Q_0 Q 0 be an F q \mathbb{F}_q F q -linear MDS stabilizer code with parameters [ [ N , k 0 , d 0 ] ] q [[N,k_0,d_0]]_q [[ N , k 0 , d 0 ] ] q , where k 0 > 0 k_0>0 k 0 > 0 and d 0 ≥ 2 d_0\ge2 d 0 ≥ 2 . Let S 0 S_0 S 0 be its stabilizer label space. For every pair of integers B ≥ 1 B\ge1 B ≥ 1 and 0 ≤ g ≤ B k 0 − 1 0\le g\le Bk_0-1 0 ≤ g ≤ B k 0 − 1 , there exists a pure F q \mathbb{F}_q F q -linear ( r , δ ) (r,\delta) ( r , δ ) -SSLRC Q Q Q with parameters
[ [ B N , B k 0 − g , g , d 0 ] ] q , ( r , δ ) = ( N − d 0 + 1 , d 0 ) . (54) [[BN,Bk_0-g,g,d_0]]_q,\qquad(r,\delta)=(N-d_0+1,d_0). \tag*{(54)} [[ B N , B k 0 − g , g , d 0 ] ] q , ( r , δ ) = ( N − d 0 + 1 , d 0 ) . (54) Its stabilizer label space is S = S 0 ⊕ B S=S_0^{\oplus B} S = S 0 ⊕ B , and its B B B coordinate blocks of length N N N form a pairwise disjoint repair cover. The code attains the Singleton-type bound (42) , the alphabet-dependent bound (44) , and the redundancy bound (33) for these blocks. If ψ ( S 0 ⊥ ) \psi(S_0^{\perp}) ψ ( S 0 ⊥ ) is F q 2 \mathbb{F}_{q^2} F q 2 -linear, then Q Q Q also attains the F q 2 \mathbb{F}_{q^2} F q 2 -linear Griesmer bound (50) .
Proof. We first construct Q Q Q . Since Q 0 Q_0 Q 0 is MDS, it is pure and k 0 = N − 2 d 0 + 2 k_0=N-2d_0+2 k 0 = N − 2 d 0 + 2 ; see the proof of [1 Theorem 7] . Let r = N − d 0 + 1 r=N-d_0+1 r = N − d 0 + 1 and δ = d 0 \delta=d_0 δ = d 0 . Since k 0 > 0 k_0>0 k 0 > 0 and δ ≤ N = r + δ − 1 \delta\le N=r+\delta-1 δ ≤ N = r + δ − 1 , we may apply Corollary VI.3 to B B B copies of Q 0 Q_0 Q 0 . For every 0 ≤ g ≤ B k 0 − 1 0\le g\le Bk_0-1 0 ≤ g ≤ B k 0 − 1 , we obtain a pure F q \mathbb{F}_q F q -linear ( r , δ ) (r,\delta) ( r , δ ) -SSLRC Q Q Q with the parameters in (54) and stabilizer label space S = S 0 ⊕ B S=S_0^{\oplus B} S = S 0 ⊕ B . Its B B B coordinate blocks of length N N N form a pairwise disjoint repair cover.
We next show that equality holds in the Singleton-type bound (42) and the disjoint redundancy bound (33) . Let n = B N n=BN n = B N , k = B k 0 − g k=Bk_0-g k = B k 0 − g , and d = d 0 d=d_0 d = d 0 . Since N + k 0 = 2 ( N − d 0 + 1 ) = 2 r N+k_0=2(N-d_0+1)=2r N + k 0 = 2 ( N − d 0 + 1 ) = 2 r , we have
⌈ n + k + g 2 r ⌉ = ⌈ B ( N + k 0 ) 2 r ⌉ = B . \left\lceil\frac{n+k+g}{2r}\right\rceil
=
\left\lceil\frac{B(N+k_0)}{2r}\right\rceil
=B. ⌈ 2 r n + k + g ⌉ = ⌈ 2 r B ( N + k 0 ) ⌉ = B .
Consequently,
k + g + 2 ( d − 1 ) + 2 ( B − 1 ) ( δ − 1 ) = B k 0 + 2 ( d 0 − 1 ) + 2 ( B − 1 ) ( d 0 − 1 ) = B ( k 0 + 2 d 0 − 2 ) = B N = n . \begin{aligned}
k+g+2(d-1)+2(B-1)(\delta-1)
&=Bk_0+2(d_0-1)+2(B-1)(d_0-1)\\
&=B(k_0+2d_0-2)=BN=n.
\end{aligned} k + g + 2 ( d − 1 ) + 2 ( B − 1 ) ( δ − 1 ) = B k 0 + 2 ( d 0 − 1 ) + 2 ( B − 1 ) ( d 0 − 1 ) = B ( k 0 + 2 d 0 − 2 ) = B N = n .
Thus Q Q Q attains the Singleton-type bound (42) . Since Q Q Q is pure, the redundancy bound (33) applies with M = B M=B M = B . Moreover,
n − k − g = B ( N − k 0 ) = 2 B ( d 0 − 1 ) = 2 B ( δ − 1 ) . n-k-g=B(N-k_0)=2B(d_0-1)=2B(\delta-1). n − k − g = B ( N − k 0 ) = 2 B ( d 0 − 1 ) = 2 B ( δ − 1 ) .
Hence the redundancy bound (33) is also attained.
We next prove that the alphabet-dependent bound (44) is attained. Set κ = ( n + k + g ) / 2 = B r \kappa=(n+k+g)/2=Br κ = ( n + k + g ) /2 = B r and choose a = B − 1 a=B-1 a = B − 1 . This value is admissible because
a r = ( B − 1 ) r < κ , a ( r + δ − 1 ) = ( B − 1 ) N < n . ar=(B-1)r<\kappa,\qquad a(r+\delta-1)=(B-1)N<n. a r = ( B − 1 ) r < κ , a ( r + δ − 1 ) = ( B − 1 ) N < n .
For this choice, the remaining length is n − a ( r + δ − 1 ) = N n-a(r+\delta-1)=N n − a ( r + δ − 1 ) = N . By purity and (11), every nonzero label in S 0 ⊥ S_0^{\perp} S 0 ⊥ has weight at least d 0 d_0 d 0 , and some label in S 0 ⊥ ∖ S 0 S_0^{\perp}\setminus S_0 S 0 ⊥ ∖ S 0 has weight d 0 d_0 d 0 . Since ψ \psi ψ preserves weight, C 0 : = ψ ( S 0 ⊥ ) C_0:=\psi(S_0^{\perp}) C 0 := ψ ( S 0 ⊥ ) has minimum distance d 0 d_0 d 0 . By (12) and (5), we have
∣ C 0 ∣ = q dim q S 0 ⊥ = q N + k 0 = ( q 2 ) r . |C_0|=q^{\dim_q S_0^{\perp}}=q^{N+k_0}=(q^2)^r. ∣ C 0 ∣ = q d i m q S 0 ⊥ = q N + k 0 = ( q 2 ) r .
Combining this equality with the cardinality form of the classical Singleton bound [32 ] , we obtain
r ≤ k a d d ( q 2 ) ( N , d 0 ) ≤ N − d 0 + 1 = r . r\le k_{\mathrm{add}}^{(q^2)}(N,d_0)\le N-d_0+1=r. r ≤ k add ( q 2 ) ( N , d 0 ) ≤ N − d 0 + 1 = r .
Therefore, k a d d ( q 2 ) ( N , d 0 ) = r k_{\mathrm{add}}^{(q^2)}(N,d_0)=r k add ( q 2 ) ( N , d 0 ) = r . Substituting a = B − 1 a=B-1 a = B − 1 into the right-hand side of (44) , we obtain
k + g ≤ min a ∈ A { 2 a r + 2 k a d d ( q 2 ) ( n − a ( r + δ − 1 ) , d ) − n } ≤ 2 ( B − 1 ) r + 2 r − B N = B ( 2 r − N ) = B k 0 = k + g . \begin{aligned}
k+g &\le\min_{a\in A}\{2ar+2k_{\mathrm{add}}^{(q^2)}(n-a(r+\delta-1),d)-n\}\\
&\le2(B-1)r+2r-BN\\
&=B(2r-N)=Bk_0=k+g.
\end{aligned} k + g ≤ a ∈ A min { 2 a r + 2 k add ( q 2 ) ( n − a ( r + δ − 1 ) , d ) − n } ≤ 2 ( B − 1 ) r + 2 r − B N = B ( 2 r − N ) = B k 0 = k + g .
Hence the alphabet-dependent bound (44) is attained.
Finally, suppose that C 0 C_0 C 0 is F q 2 \mathbb{F}_{q^2} F q 2 -linear. Since S = S 0 ⊕ B S=S_0^{\oplus B} S = S 0 ⊕ B , we have
C : = ψ ( S ⊥ t ) = C 0 ⊕ B . C:=\psi(S^{\perp_{\mathrm{t}}})=C_0^{\oplus B}. C := ψ ( S ⊥ t ) = C 0 ⊕ B .
Thus C C C is F q 2 \mathbb{F}_{q^2} F q 2 -linear. By Proposition V.2,
dim q 2 C = n + k + g 2 = κ = B r . \dim_{q^2} C=\frac{n+k+g}{2}=\kappa=Br. dim q 2 C = 2 n + k + g = κ = B r .
Therefore, Q Q Q satisfies the F q 2 \mathbb{F}_{q^2} F q 2 -linear Griesmer bound (50) . For the same admissible choice a = B − 1 a=B-1 a = B − 1 , we have N a = N N_a=N N a = N and h a = κ − a r = r h_a=\kappa-ar=r h a = κ − a r = r . Since ⌈ d 0 / q 2 i ⌉ ≥ 1 \left\lceil d_0/q^{2i}\right\rceil\ge1 ⌈ d 0 / q 2 i ⌉ ≥ 1 for i ≥ 1 i\ge1 i ≥ 1 , we obtain
B N ≥ ( B − 1 ) N + ∑ i = 0 r − 1 ⌈ d 0 q 2 i ⌉ ≥ ( B − 1 ) N + d 0 + r − 1 = B N . \begin{aligned}
BN&\ge(B-1)N+\sum_{i=0}^{r-1}\left\lceil\frac{d_0}{q^{2i}}\right\rceil\\
&\ge(B-1)N+d_0+r-1\\
&=BN.
\end{aligned} B N ≥ ( B − 1 ) N + i = 0 ∑ r − 1 ⌈ q 2 i d 0 ⌉ ≥ ( B − 1 ) N + d 0 + r − 1 = B N .
Thus Q Q Q also attains the F q 2 \mathbb{F}_{q^2} F q 2 -linear Griesmer bound (50) .
For an integer 1 ≤ t < q / 2 1\le t<q/2 1 ≤ t < q /2 , let H t = { ( f ( a ) ) a ∈ F q : f ∈ F q [ x ] , deg f < t } H_t=\{(f(a))_{a\in\mathbb{F}_q}:f\in\mathbb{F}_q[x],\ \deg f<t\} H t = {( f ( a ) ) a ∈ F q : f ∈ F q [ x ] , deg f < t } . The code H t H_t H t is an [ q , t , q − t + 1 ] q [q,t,q-t+1]_q [ q , t , q − t + 1 ] q Reed-Solomon code with generator matrix A t = ( a j ) 0 ≤ j < t , a ∈ F q A_t=(a^j)_{0\le j<t,\ a\in\mathbb{F}_q} A t = ( a j ) 0 ≤ j < t , a ∈ F q by [32 Ch. 10, Sec. 8] . Since H t = GRS t ( a , 1 ) H_t=\operatorname{GRS}_t(a,1) H t = GRS t ( a , 1 ) , we have H t ⊥ = H q − t H_t^\perp=H_{q-t} H t ⊥ = H q − t by the GRS duality theorem [32 Ch. 10, Theorem 4] .
Lemma VI.5. For 1 ≤ t < q / 2 1\le t<q/2 1 ≤ t < q /2 , H t ⊆ H t ⊥ H_t\subseteq H_t^\perp H t ⊆ H t ⊥ , and H t ⊥ H_t^\perp H t ⊥ is an [ q , q − t , t + 1 ] q [q,q-t,t+1]_q [ q , q − t , t + 1 ] q MDS code. The isotropic space S 0 = H t × H t S_0=H_t\times H_t S 0 = H t × H t defines a pure F q \mathbb{F}_q F q -linear MDS stabilizer code with parameters [ [ q , q − 2 t , t + 1 ] ] q [[q,q-2t,t+1]]_q [[ q , q − 2 t , t + 1 ] ] q .
Proof. Since t < q − t t<q-t t < q − t , we have H t ⊆ H q − t = H t ⊥ H_t\subseteq H_{q-t}=H_t^\perp H t ⊆ H q − t = H t ⊥ . By Lemma II.2 (ii), H t ⊥ H_t^\perp H t ⊥ has parameters [ q , q − t , t + 1 ] q [q,q-t,t+1]_q [ q , q − t , t + 1 ] q . Apply Lemma III.11 with C X = C Z = H t C_X=C_Z=H_t C X = C Z = H t and M X = M Z = A t M_X=M_Z=A_t M X = M Z = A t . Since A t A t T = 0 A_tA_t^{\mathsf{T}}=0 A t A t T = 0 , we have b = 0 b=0 b = 0 and
G 0 = S 0 = H t × H t , K = q q − 2 t > 1 , R = 1. G_0=S_0=H_t\times H_t,\quad K=q^{q-2t}>1,\quad R=1. G 0 = S 0 = H t × H t , K = q q − 2 t > 1 , R = 1.
Thus S 0 = G 0 ∩ G 0 ⊥ t ⊆ S 0 ⊥ s S_0=G_0\cap G_0^{\perp_{\mathrm{t}}}\subseteq S_0^{\perp_{\mathrm{s}}} S 0 = G 0 ∩ G 0 ⊥ t ⊆ S 0 ⊥ s , and the resulting code Q 0 Q_0 Q 0 is an F q \mathbb{F}_q F q -linear stabilizer code. By (25) in Lemma III.11,
d ( Q 0 ) = min { wt ( u ) : u ∈ H t ⊥ ∖ H t } = t + 1 , d(Q_0)=\min\{\operatorname{wt}(u):u\in H_t^\perp\setminus H_t\}=t+1, d ( Q 0 ) = min { wt ( u ) : u ∈ H t ⊥ ∖ H t } = t + 1 ,
where the last equality follows from d ( H t ) = q − t + 1 > t + 1 = d ( H t ⊥ ) d(H_t)=q-t+1>t+1=d(H_t^\perp) d ( H t ) = q − t + 1 > t + 1 = d ( H t ⊥ ) . For every nonzero ( x , z ) ∈ G 0 (x,z)\in G_0 ( x , z ) ∈ G 0 , at least one component is a nonzero codeword of H t H_t H t , and hence
swt ( x , z ) = ∣ supp ( x ) ∪ supp ( z ) ∣ ≥ q − t + 1 > t + 1. \operatorname{swt}(x,z)=|\operatorname{supp}(x)\cup\operatorname{supp}(z)|\ge q-t+1>t+1. swt ( x , z ) = ∣ supp ( x ) ∪ supp ( z ) ∣ ≥ q − t + 1 > t + 1.
Therefore, Q 0 Q_0 Q 0 is pure by Definition II.18. Finally, ( q − 2 t ) + 2 ( d ( Q 0 ) − 1 ) = q (q-2t)+2(d(Q_0)-1)=q ( q − 2 t ) + 2 ( d ( Q 0 ) − 1 ) = q , so Q 0 Q_0 Q 0 attains the quantum Singleton bound [2 Theorem 11] .
Applying Corollary VI.3 to copies of the stabilizer code in Lemma VI.5, we obtain the following family of pure ( r , δ ) (r,\delta) ( r , δ ) -SSLRCs.
Theorem VI.6. Let q q q be a prime power, and let t t t , B B B , and g g g be integers such that 1 ≤ t < q / 2 1\le t<q/2 1 ≤ t < q /2 , B ≥ 1 B\ge1 B ≥ 1 and 0 ≤ g ≤ B ( q − 2 t ) − 1 0\le g\le B(q-2t)-1 0 ≤ g ≤ B ( q − 2 t ) − 1 . Then there exists a pure F q \mathbb{F}_q F q -linear ( r , δ ) (r,\delta) ( r , δ ) -SSLRC Q Q Q with parameters
[ [ B q , B ( q − 2 t ) − g , g , t + 1 ] ] q , ( r , δ ) = ( q − t , t + 1 ) . (55) [[Bq,B(q-2t)-g,g,t+1]]_q,\quad(r,\delta)=(q-t,t+1). \tag*{(55)} [[ B q , B ( q − 2 t ) − g , g , t + 1 ] ] q , ( r , δ ) = ( q − t , t + 1 ) . (55) Moreover, Q Q Q attains the Singleton-type bound (42) , the alphabet-dependent bound (44) , and the F q 2 \mathbb{F}_{q^2} F q 2 -linear Griesmer bound (50) . It also attains the redundancy bound (33) for its B B B pairwise disjoint coordinate blocks.
Proof. We first construct Q Q Q . By Lemma VI.5, there exists a pure F q \mathbb{F}_q F q -linear stabilizer code Q 0 Q_0 Q 0 with parameters [ [ q , q − 2 t , t + 1 ] ] q [[q,q-2t,t+1]]_q [[ q , q − 2 t , t + 1 ] ] q and stabilizer label space S 0 = H t × H t S_0=H_t\times H_t S 0 = H t × H t . Let r = q − t r=q-t r = q − t and δ = t + 1 \delta=t+1 δ = t + 1 . Since q − 2 t ≥ 1 q-2t\ge1 q − 2 t ≥ 1 , d ( Q 0 ) = δ d(Q_0)=\delta d ( Q 0 ) = δ , and δ ≤ q = r + δ − 1 \delta\le q=r+\delta-1 δ ≤ q = r + δ − 1 , we may apply Corollary VI.3 to B B B copies of Q 0 Q_0 Q 0 . For every 0 ≤ g ≤ B ( q − 2 t ) − 1 0\le g\le B(q-2t)-1 0 ≤ g ≤ B ( q − 2 t ) − 1 , we obtain a pure F q \mathbb{F}_q F q -linear ( r , δ ) (r,\delta) ( r , δ ) -SSLRC Q Q Q with the parameters in (55) and stabilizer label space S = S 0 ⊕ B S=S_0^{\oplus B} S = S 0 ⊕ B . Its B B B coordinate blocks of length q q q form a pairwise disjoint repair cover.
Finally, we verify the F q 2 \mathbb{F}_{q^2} F q 2 -linearity required in Theorem VI.4. Since S 0 = H t × H t S_0=H_t\times H_t S 0 = H t × H t is F q \mathbb{F}_q F q -linear, its trace-symplectic and symplectic orthogonal complements coincide, and hence
S 0 ⊥ t = S 0 ⊥ s = H t ⊥ × H t ⊥ . S_0^{\perp_t}=S_0^{\perp_s}=H_t^\perp\times H_t^\perp. S 0 ⊥ t = S 0 ⊥ s = H t ⊥ × H t ⊥ .
By the coordinatewise definition of ψ \psi ψ in (6), we obtain
ψ ( S 0 ⊥ t ) = H t ⊥ + α H t ⊥ = span F q 2 H t ⊥ , \psi(S_0^{\perp_t})=H_t^\perp+\alpha H_t^\perp=\operatorname{span}_{\mathbb{F}_{q^2}}H_t^\perp, ψ ( S 0 ⊥ t ) = H t ⊥ + α H t ⊥ = span F q 2 H t ⊥ ,
where the last equality follows from the F q \mathbb{F}_q F q -linearity of H t ⊥ H_t^\perp H t ⊥ and F q 2 = F q + α F q \mathbb{F}_{q^2}=\mathbb{F}_q+\alpha\mathbb{F}_q F q 2 = F q + α F q . Thus ψ ( S 0 ⊥ t ) \psi(S_0^{\perp_t}) ψ ( S 0 ⊥ t ) is F q 2 \mathbb{F}_{q^2} F q 2 -linear. By Theorem VI.4, Q Q Q attains the four stated bounds. □ \square □
We use the following Lagrange interpolation identity [32 ] to choose the multipliers of a GRS code.
Lemma VI.7. [32 ] Let N ≥ 1 N\ge1 N ≥ 1 , let a 1 , … , a N a_1,\ldots,a_N a 1 , … , a N be distinct elements of a field, and set ℓ i = ∏ j ≠ i ( a i − a j ) \ell_i=\prod_{j\ne i}(a_i-a_j) ℓ i = ∏ j = i ( a i − a j ) . For every integer 0 ≤ u ≤ N − 1 0\le u\le N-1 0 ≤ u ≤ N − 1 ,
∑ i = 1 N a i u ℓ i = { 0 , 0 ≤ u ≤ N − 2 , 1 , u = N − 1. \sum_{i=1}^{N}\frac{a_i^u}{\ell_i}=
\begin{cases}
0, & 0\le u\le N-2,\\
1, & u=N-1.
\end{cases} i = 1 ∑ N ℓ i a i u = { 0 , 1 , 0 ≤ u ≤ N − 2 , u = N − 1.
Using the parameter families in [1 Lemma 8] , we obtain the following optimal ( r , δ ) (r,\delta) ( r , δ ) -SSLRCs. We use s s s to denote the distance parameter in parts (iii)–(vi) of that lemma.
Corollary VI.8. Let q q q be a prime power, and let B ≥ 1 B\ge1 B ≥ 1 be an integer. Then the following statements hold.
A family of F q \mathbb{F}_q F q -linear pure ( N − d 0 + 1 , d 0 ) (N-d_0+1,d_0) ( N − d 0 + 1 , d 0 ) -SSLRCs with parameters [ [ B N , B ( N − 2 d 0 + 2 ) − g , g , d 0 ] ] q [[BN,B(N-2d_0+2)-g,g,d_0]]_q [[ B N , B ( N − 2 d 0 + 2 ) − g , g , d 0 ] ] q exists for integers 3 ≤ N ≤ q 3\le N\le q 3 ≤ N ≤ q , 2 ≤ d 0 ≤ ⌊ ( N + 1 ) / 2 ⌋ 2\le d_0\le\lfloor(N+1)/2\rfloor 2 ≤ d 0 ≤ ⌊( N + 1 ) /2 ⌋ , and 0 ≤ g ≤ B ( N − 2 d 0 + 2 ) − 1 0\le g\le B(N-2d_0+2)-1 0 ≤ g ≤ B ( N − 2 d 0 + 2 ) − 1 .
A family of F q \mathbb{F}_q F q -linear pure ( ( ν + 1 ) ( q − 1 ) , ν + 2 ) ((\nu+1)(q-1),\nu+2) (( ν + 1 ) ( q − 1 ) , ν + 2 ) -SSLRCs with parameters [ [ B ( ν + 1 ) q , B ( ν + 1 ) ( q − 2 ) − g , g , ν + 2 ] ] q [[B(\nu+1)q,B(\nu+1)(q-2)-g,g,\nu+2]]_q [[ B ( ν + 1 ) q , B ( ν + 1 ) ( q − 2 ) − g , g , ν + 2 ] ] q exists for q ≥ 3 q\ge3 q ≥ 3 and integers 0 ≤ ν ≤ q − 2 0\le\nu\le q-2 0 ≤ ν ≤ q − 2 and 0 ≤ g ≤ B ( ν + 1 ) ( q − 2 ) − 1 0\le g\le B(\nu+1)(q-2)-1 0 ≤ g ≤ B ( ν + 1 ) ( q − 2 ) − 1 .
A family of F q \mathbb{F}_q F q -linear pure ( q − s − 1 , s + 1 ) (q-s-1,s+1) ( q − s − 1 , s + 1 ) -SSLRCs with parameters [ [ B ( q − 1 ) , B ( q − 2 s − 1 ) − g , g , s + 1 ] ] q [[B(q-1),B(q-2s-1)-g,g,s+1]]_q [[ B ( q − 1 ) , B ( q − 2 s − 1 ) − g , g , s + 1 ] ] q exists for integers 1 ≤ s < ( q − 1 ) / 2 1\le s<(q-1)/2 1 ≤ s < ( q − 1 ) /2 and 0 ≤ g ≤ B ( q − 2 s − 1 ) − 1 0\le g\le B(q-2s-1)-1 0 ≤ g ≤ B ( q − 2 s − 1 ) − 1 .
A family of F q \mathbb{F}_q F q -linear pure ( q − s − 1 , s + 2 ) (q-s-1,s+2) ( q − s − 1 , s + 2 ) -SSLRCs with parameters [ [ B q , B ( q − 2 s − 2 ) − g , g , s + 2 ] ] q [[Bq,B(q-2s-2)-g,g,s+2]]_q [[ B q , B ( q − 2 s − 2 ) − g , g , s + 2 ] ] q exists for integers 0 ≤ s ≤ ⌊ ( q − 3 ) / 2 ⌋ 0\le s\le\lfloor(q-3)/2\rfloor 0 ≤ s ≤ ⌊( q − 3 ) /2 ⌋ and 0 ≤ g ≤ B ( q − 2 s − 2 ) − 1 0\le g\le B(q-2s-2)-1 0 ≤ g ≤ B ( q − 2 s − 2 ) − 1 .
A family of F q \mathbb{F}_q F q -linear pure ( q 2 − s − 1 , s + 1 ) (q^2-s-1,s+1) ( q 2 − s − 1 , s + 1 ) -SSLRCs with parameters [ [ B ( q 2 − 1 ) , B ( q 2 − 2 s − 1 ) − g , g , s + 1 ] ] q [[B(q^2-1),B(q^2-2s-1)-g,g,s+1]]_q [[ B ( q 2 − 1 ) , B ( q 2 − 2 s − 1 ) − g , g , s + 1 ] ] q exists for integers 1 ≤ s < q − 1 1\le s<q-1 1 ≤ s < q − 1 and 0 ≤ g ≤ B ( q 2 − 2 s − 1 ) − 1 0\le g\le B(q^2-2s-1)-1 0 ≤ g ≤ B ( q 2 − 2 s − 1 ) − 1 .
A family of F q \mathbb{F}_q F q -linear pure ( q 2 − s − 1 , s + 2 ) (q^2-s-1,s+2) ( q 2 − s − 1 , s + 2 ) -SSLRCs with parameters [ [ B q 2 , B ( q 2 − 2 s − 2 ) − g , g , s + 2 ] ] q [[Bq^2,B(q^2-2s-2)-g,g,s+2]]_q [[ B q 2 , B ( q 2 − 2 s − 2 ) − g , g , s + 2 ] ] q exists for integers 0 ≤ s < q − 1 0\le s<q-1 0 ≤ s < q − 1 and 0 ≤ g ≤ B ( q 2 − 2 s − 2 ) − 1 0\le g\le B(q^2-2s-2)-1 0 ≤ g ≤ B ( q 2 − 2 s − 2 ) − 1 .
Moreover, these families of pure SSLRCs attain the four bounds stated in Theorem VI.4. Their B B B coordinate blocks of length n / B n/B n / B form a pairwise disjoint repair cover, where n n n denotes the code length.
Proof. We first verify that the MDS stabilizer codes can be chosen with F q 2 \mathbb{F}_{q^2} F q 2 -linear comparison codes. For each case, let N N N be the block length and let d 0 = δ d_0=\delta d 0 = δ . Set k 0 = N − 2 d 0 + 2 k_0=N-2d_0+2 k 0 = N − 2 d 0 + 2 . For parts (i), (iii), and (iv), we have N < q N<q N < q and k 0 > 0 k_0>0 k 0 > 0 . Choose distinct a 1 , … , a N ∈ F q a_1,\ldots,a_N\in\mathbb{F}_q a 1 , … , a N ∈ F q , set ℓ i = ∏ j ≠ i ( a i − a j ) \ell_i=\prod_{j\ne i}(a_i-a_j) ℓ i = ∏ j = i ( a i − a j ) , and choose v i ∈ F q 2 × v_i\in\mathbb{F}_{q^2}^{\times} v i ∈ F q 2 × with v i q + 1 = ℓ i − 1 v_i^{q+1}=\ell_i^{-1} v i q + 1 = ℓ i − 1 . Since ℓ i − 1 ∈ F q × \ell_i^{-1}\in\mathbb{F}_q^{\times} ℓ i − 1 ∈ F q × and the norm map onto F q × \mathbb{F}_q^{\times} F q × is surjective, such multipliers exist. Let X = GRS d 0 − 1 ( a , v ) X=\operatorname{GRS}_{d_0-1}(a,v) X = GRS d 0 − 1 ( a , v ) over F q 2 \mathbb{F}_{q^2} F q 2 . For 0 ≤ j , j ′ ≤ d 0 − 2 0\le j,j'\le d_0-2 0 ≤ j , j ′ ≤ d 0 − 2 , we have
∑ i = 1 N ( v i a i j ) ( v i a i j ′ ) q = ∑ i = 1 N a i j + j ′ ℓ i = 0 \sum_{i=1}^{N}(v_i a_i^j)(v_i a_i^{j'})^q=\sum_{i=1}^{N}\frac{a_i^{j+ j'}}{\ell_i}=0 i = 1 ∑ N ( v i a i j ) ( v i a i j ′ ) q = i = 1 ∑ N ℓ i a i j + j ′ = 0
by Lemma VI.7, since a i q = a i a_i^q=a_i a i q = a i and j + j ′ ≤ 2 d 0 − 4 ≤ N − 2 j+j'\le2d_0-4\le N-2 j + j ′ ≤ 2 d 0 − 4 ≤ N − 2 . Hence X ⊆ X ⊥ H X\subseteq X^{\perp_H} X ⊆ X ⊥ H .
For part (ii), let A A A be the length-q 2 q^2 q 2 set obtained by evaluating bivariate polynomials of degree at most ν \nu ν over F q 2 \mathbb{F}_{q^2} F q 2 . By [36 Lemma 8] and the proof of [36 Theorem 5] , there exists a vector λ ∈ F q 2 q 2 \lambda\in\mathbb{F}_{q^2}^{q^2} λ ∈ F q 2 q 2 of weight N = ( ν + 1 ) q N=(\nu+1)q N = ( ν + 1 ) q such that
∑ i = 1 q 2 λ i a i b i q = 0 for all a , b ∈ A . \sum_{i=1}^{q^2}\lambda_i a_i b_i^q=0\quad\text{for all }a,b\in A. i = 1 ∑ q 2 λ i a i b i q = 0 for all a , b ∈ A .
Let I = supp ( λ ) I=\operatorname{supp}(\lambda) I = supp ( λ ) , and choose v i ∈ F q 2 × v_i\in\mathbb{F}_{q^2}^{\times} v i ∈ F q 2 × with v i q + 1 = λ i v_i^{q+1}=\lambda_i v i q + 1 = λ i for i ∈ I i\in I i ∈ I . Then
X = { ( v i a i ) i ∈ I : a ∈ A } X=\{(v_i a_i)_{i\in I}:a\in A\} X = {( v i a i ) i ∈ I : a ∈ A }
is an F q 2 \mathbb{F}_{q^2} F q 2 -linear [ N , ν + 1 , N − ν ] q 2 [N,\nu+1,N-\nu]_{q^2} [ N , ν + 1 , N − ν ] q 2 GRS code. Moreover, for all a , b ∈ A a,b\in A a , b ∈ A ,
∑ i ∈ I ( v i a i ) ( v i b i ) q = ∑ i ∈ I λ i a i b i q = 0. \sum_{i\in I}(v_i a_i)(v_i b_i)^q=\sum_{i\in I}\lambda_i a_i b_i^q=0. i ∈ I ∑ ( v i a i ) ( v i b i ) q = i ∈ I ∑ λ i a i b i q = 0.
Thus X ⊆ X ⊥ H X\subseteq X^{\perp_H} X ⊆ X ⊥ H . For parts (v) and (vi), we may choose a Hermitian self-orthogonal [ N , d 0 − 1 , N − d 0 + 2 ] q 2 [N,d_0-1,N-d_0+2]_{q^2} [ N , d 0 − 1 , N − d 0 + 2 ] q 2 MDS code X X X by the constructions in the proof of [18 Theorem 10] . We next determine the parameters of the stabilizer codes. In every case, X X X is an [ N , d 0 − 1 , N − d 0 + 2 ] q 2 [N,d_0-1,N-d_0+2]_{q^2} [ N , d 0 − 1 , N − d 0 + 2 ] q 2 MDS code and X ⊆ X ⊥ H X\subseteq X^{\perp_H} X ⊆ X ⊥ H . By MDS duality [32 ] , X ⊥ H X^{\perp_H} X ⊥ H has parameters [ N , N − d 0 + 1 , d 0 ] q 2 [N,N-d_0+1,d_0]_{q^2} [ N , N − d 0 + 1 , d 0 ] q 2 . Since k 0 = N − 2 d 0 + 2 > 0 k_0=N-2d_0+2>0 k 0 = N − 2 d 0 + 2 > 0 , we have d ( X ) = N − d 0 + 2 > d 0 d(X)=N-d_0+2>d_0 d ( X ) = N − d 0 + 2 > d 0 . Hence
min { wt ( w ) : w ∈ X ⊥ H ∖ X } = d 0 . \min\{\operatorname{wt}(w):w\in X^{\perp_H}\setminus X\}=d_0. min { wt ( w ) : w ∈ X ⊥ H ∖ X } = d 0 .
By Lemma III.14, S 0 = ψ − 1 ( X ) S_0=\psi^{-1}(X) S 0 = ψ − 1 ( X ) defines a pure F q \mathbb{F}_q F q -linear MDS stabilizer code with parameters [ [ N , k 0 , d 0 ] ] q [[N,k_0,d_0]]_q [[ N , k 0 , d 0 ] ] q and
ψ ( S 0 ⊥ t ) = X ⊥ H . \psi(S_0^{\perp_t})=X^{\perp_H}. ψ ( S 0 ⊥ t ) = X ⊥ H .
Finally, X ⊥ H X^{\perp_H} X ⊥ H is F q 2 \mathbb{F}_{q^2} F q 2 -linear, so the result follows from Theorem VI.4.
Example VI.9. Theorem VI.6 and Corollary VI.8 provide the pure ( r , δ ) (r,\delta) ( r , δ ) -SSLRCs listed in Table 2 . For example, take q = 3 q=3 q = 3 , ν = 1 \nu=1 ν = 1 , and B = 2 B=2 B = 2 in Corollary VI.8(ii). Then N = 6 N=6 N = 6 , k 0 = 2 k_0=2 k 0 = 2 , and d 0 = 3 d_0=3 d 0 = 3 . For every 0 ≤ g ≤ 3 0\leq g\leq3 0 ≤ g ≤ 3 , we obtain
Table 2. Optimal pure ( r , δ ) (r,\delta) ( r , δ ) -SSLRCs from Theorem VI.6 and Corollary VI.8.
[ [ 12 , 4 − g , g , 3 ] ] 3 , ( r , δ ) = ( 4 , 3 ) . [[12,4-g,g,3]]_3,\quad(r,\delta)=(4,3). [[ 12 , 4 − g , g , 3 ] ] 3 , ( r , δ ) = ( 4 , 3 ) .
For g = 1 g=1 g = 1 , the parameters are [ [ 12 , 3 , 1 , 3 ] ] 3 [[12,3,1,3]]_3 [[ 12 , 3 , 1 , 3 ] ] 3 . The two coordinate blocks of length 6 form a pairwise disjoint repair cover. Every code in Table 2 attains the four bounds in Theorem VI.4.
Optimal Pure ( r , δ ) (r,\delta) ( r , δ ) -SSLRCs from MDS Hermitian Hulls
In this subsection, we construct pure F q \mathbb{F}_q F q -linear ( r , δ ) (r,\delta) ( r , δ ) -SSLRCs from Hermitian self-orthogonal MDS codes. We then apply the construction to explicit MDS Hermitian hulls. The resulting codes attain the four bounds in Theorem VI.4.
Theorem VI.10. Let q q q be a prime power, and let H H H be an [ N , h H , N − h H + 1 ] q 2 [N,h_H,N-h_H+1]_{q^2} [ N , h H , N − h H + 1 ] q 2 MDS code. Suppose that H ⊂ H ⊥ H H\subset H^{\perp_H} H ⊂ H ⊥ H and 1 ≤ h H < N / 2 1\leq h_H<N/2 1 ≤ h H < N /2 . For every pair of integers B ≥ 1 B\geq1 B ≥ 1 and 0 ≤ g ≤ B ( N − 2 h H ) − 1 0\leq g\leq B(N-2h_H)-1 0 ≤ g ≤ B ( N − 2 h H ) − 1 , there exists a pure F q \mathbb{F}_q F q -linear ( r , δ ) (r,\delta) ( r , δ ) -SSLRC with parameters
[ [ B N , B ( N − 2 h H ) − g , g , h H + 1 ] ] q , ( r , δ ) = ( N − h H , h H + 1 ) . [[BN,B(N-2h_H)-g,g,h_H+1]]_q,\quad(r,\delta)=(N-h_H,h_H+1). [[ B N , B ( N − 2 h H ) − g , g , h H + 1 ] ] q , ( r , δ ) = ( N − h H , h H + 1 ) . Its stabilizer label space is S = ( ψ − 1 ( H ) ) ⊕ B S=(\psi^{-1}(H))^{\oplus B} S = ( ψ − 1 ( H ) ) ⊕ B . The code attains the Singleton-type bound (42) , the alphabet-dependent bound (44) , and the F q 2 \mathbb{F}_{q^2} F q 2 -linear Griesmer bound (50) . It also attains the redundancy bound (33) for its B B B pairwise disjoint coordinate blocks of length N N N .
Proof. We first construct a pure MDS stabilizer code. By MDS duality [32 ] , H ⊥ H H^{\perp_H} H ⊥ H is an [ N , N − h H , h H + 1 ] q 2 [N,N-h_H,h_H+1]_{q^2} [ N , N − h H , h H + 1 ] q 2 MDS code. Since 2 h H < N 2h_H<N 2 h H < N , we have
d ( H ) = N − h H + 1 > h H + 1 = d ( H ⊥ H ) . d(H)=N-h_H+1>h_H+1=d(H^{\perp_H}). d ( H ) = N − h H + 1 > h H + 1 = d ( H ⊥ H ) .
Let w 0 ∈ H ⊥ H w_0\in H^{\perp_H} w 0 ∈ H ⊥ H have weight h H + 1 h_H+1 h H + 1 . Since d ( H ) > h H + 1 d(H)>h_H+1 d ( H ) > h H + 1 , we have w 0 ∉ H w_0\notin H w 0 ∈ / H . Thus
min w ∈ H ⊥ H ∖ H wt ( w ) = d ( H ⊥ H ) = h H + 1. \min_{w\in H^{\perp_H}\setminus H}\operatorname{wt}(w)=d(H^{\perp_H})=h_H+1. w ∈ H ⊥ H ∖ H min wt ( w ) = d ( H ⊥ H ) = h H + 1.
By Lemma III.14, S 0 = ψ − 1 ( H ) S_0=\psi^{-1}(H) S 0 = ψ − 1 ( H ) defines an F q \mathbb{F}_q F q -linear stabilizer code Q 0 Q_0 Q 0 with parameters
[ [ N , N − 2 h H , h H + 1 ] ] q . [[N,N-2h_H,h_H+1]]_q. [[ N , N − 2 h H , h H + 1 ] ] q .
Since ψ \psi ψ preserves weight, every nonzero label in S 0 S_0 S 0 has symplectic weight at least d ( H ) > h H + 1 d(H)>h_H+1 d ( H ) > h H + 1 . Thus Q 0 Q_0 Q 0 is pure. Moreover, ( N − 2 h H ) + 2 ( ( h H + 1 ) − 1 ) = N (N-2h_H)+2((h_H+1)-1)=N ( N − 2 h H ) + 2 (( h H + 1 ) − 1 ) = N , so Q 0 Q_0 Q 0 is MDS. Finally, we apply Theorem VI.4. By Lemma II.14, we have ψ ( S 0 ⊥ t ) = H ⊥ H \psi(S_0^{\perp_{\mathrm{t}}})=H^{\perp_H} ψ ( S 0 ⊥ t ) = H ⊥ H , which is F q 2 \mathbb{F}_{q^2} F q 2 -linear. Since N − 2 h H > 0 N-2h_H>0 N − 2 h H > 0 , Theorem VI.4 applies with k 0 = N − 2 h H k_0=N-2h_H k 0 = N − 2 h H and d 0 = h H + 1 d_0=h_H+1 d 0 = h H + 1 . For every 0 ≤ g ≤ B ( N − 2 h H ) − 1 0\leq g\leq B(N-2h_H)-1 0 ≤ g ≤ B ( N − 2 h H ) − 1 , we obtain a code with the stated parameters, stabilizer label space, and repair cover. This code attains all four bounds.
For any F q 2 \mathbb{F}_{q^2} F q 2 -linear code C C C with Hermitian hull H = C ∩ C ⊥ H H=C\cap C^{\perp_H} H = C ∩ C ⊥ H , we have H ⊂ C H\subset C H ⊂ C and H ⊂ C ⊥ H ⊆ H ⊥ H H\subset C^{\perp_H}\subseteq H^{\perp_H} H ⊂ C ⊥ H ⊆ H ⊥ H . Thus Theorem VI.10 applies whenever H H H is MDS and 1 ≤ dim q 2 H < N / 2 1\leq\dim_{q^2}H<N/2 1 ≤ dim q 2 H < N /2 . If C C C is an [ N , c , N − c + 1 ] q 2 [N,c,N-c+1]_{q^2} [ N , c , N − c + 1 ] q 2 MDS code and c + h H < N c+h_H<N c + h H < N , then choosing g = B ( c − h H ) g=B(c-h_H) g = B ( c − h H ) recovers the parameters [ [ B N , B ( N − c − h H ) , B ( c − h H ) , h H + 1 ] ] q [[BN,B(N-c-h_H),B(c-h_H),h_H+1]]_q [[ B N , B ( N − c − h H ) , B ( c − h H ) , h H + 1 ] ] q .
We next apply Lemma VI.7 to determine the Hermitian hull of an explicit GRS code.
Theorem VI.11. Let q q q be a prime power, and let N , c , h H N,c,h_H N , c , h H be integers satisfying 1 ≤ h H < c 1\leq h_H<c 1 ≤ h H < c and c + h H < N ≤ q − 1 c+h_H<N\leq q-1 c + h H < N ≤ q − 1 . Let a 1 , … , a N ∈ F q × a_1,\ldots,a_N\in\mathbb{F}_q^{\times} a 1 , … , a N ∈ F q × be distinct, set ℓ i = ∏ j ≠ i ( a i − a j ) \ell_i=\prod_{j\ne i}(a_i-a_j) ℓ i = ∏ j = i ( a i − a j ) , and choose v i ∈ F q 2 × v_i\in\mathbb{F}_{q^2}^{\times} v i ∈ F q 2 × such that
μ = N − c − h H , v i q + 1 = a i μ ℓ i ( 1 ≤ i ≤ N ) . (56) \mu=N-c-h_H,\qquad v_i^{q+1}=\frac{a_i^\mu}{\ell_i}\quad(1\leq i\leq N). \tag*{(56)} μ = N − c − h H , v i q + 1 = ℓ i a i μ ( 1 ≤ i ≤ N ) . (56) For a = ( a 1 , … , a N ) \boldsymbol{a}=(a_1,\ldots,a_N) a = ( a 1 , … , a N ) and v = ( v 1 , … , v N ) \boldsymbol{v}=(v_1,\ldots,v_N) v = ( v 1 , … , v N ) , the code C = GRS c ( a , v ) C=\operatorname{GRS}_c(\boldsymbol{a},\boldsymbol{v}) C = GRS c ( a , v ) has Hermitian hull C ∩ C ⊥ H = GRS h H ( a , v ) C\cap C^{\perp_H}=\operatorname{GRS}_{h_H}(\boldsymbol{a},\boldsymbol{v}) C ∩ C ⊥ H = GRS h H ( a , v ) . Consequently, for every pair of integers B ≥ 1 B\geq1 B ≥ 1 and 0 ≤ g ≤ B ( N − 2 h H ) − 1 0\leq g\leq B(N-2h_H)-1 0 ≤ g ≤ B ( N − 2 h H ) − 1 , there exists a pure F q \mathbb{F}_q F q -linear ( r , δ ) (r,\delta) ( r , δ ) -SSLRC with parameters
[ [ B N , B ( N − 2 h H ) − g , g , h H + 1 ] ] q , ( r , δ ) = ( N − h H , h H + 1 ) . [[BN,B(N-2h_H)-g,g,h_H+1]]_q,\qquad(r,\delta)=(N-h_H,h_H+1). [[ B N , B ( N − 2 h H ) − g , g , h H + 1 ] ] q , ( r , δ ) = ( N − h H , h H + 1 ) . Its stabilizer label space is S = ( ψ − 1 ( GRS h H ( a , v ) ) ) ⊕ B S=(\psi^{-1}(\operatorname{GRS}_{h_H}(\boldsymbol{a},\boldsymbol{v})))^{\oplus B} S = ( ψ − 1 ( GRS h H ( a , v )) ) ⊕ B . The code attains all the bounds stated in Theorem VI.10.
Proof. We first show that the required multipliers exist. Since a 1 , … , a N ∈ F q × a_1,\ldots,a_N\in\mathbb{F}_q^{\times} a 1 , … , a N ∈ F q × are distinct, a i μ / ℓ i ∈ F q × a_i^\mu/\ell_i\in\mathbb{F}_q^{\times} a i μ / ℓ i ∈ F q × for 1 ≤ i ≤ N 1\leq i\leq N 1 ≤ i ≤ N . The norm map v ↦ v q + 1 v\mapsto v^{q+1} v ↦ v q + 1 from F q 2 × \mathbb{F}_{q^2}^{\times} F q 2 × to F q × \mathbb{F}_q^{\times} F q × is surjective. Thus, for each i i i , there exists v i ∈ F q 2 × v_i\in\mathbb{F}_{q^2}^{\times} v i ∈ F q 2 × satisfying (55). We next determine the rank of the Hermitian Gram matrix. Let M = ( v i a i j ) 0 ≤ j < c , 1 ≤ i ≤ N M=(v_i a_i^j)_{0\leq j<c,\,1\leq i\leq N} M = ( v i a i j ) 0 ≤ j < c , 1 ≤ i ≤ N be the generator matrix of C C C , and set Ω = M ( M ( q ) ) T \Omega=M(M^{(q)})^{\mathsf{T}} Ω = M ( M ( q ) ) T . Since a i q = a i a_i^q=a_i a i q = a i , the ( j , j ′ ) (j,j') ( j , j ′ ) -entry of Ω \Omega Ω satisfies
Ω j j ′ = ∑ i = 1 N ( v i a i j ) ( v i a i j ′ ) q = ∑ i = 1 N v i q + 1 a i j + j ′ = ∑ i = 1 N a i μ + j + j ′ ℓ i , \Omega_{jj'}=\sum_{i=1}^{N}(v_i a_i^j)(v_i a_i^{j'})^q=\sum_{i=1}^{N}v_i^{q+1}a_i^{j+j'}=\sum_{i=1}^{N}\frac{a_i^{\mu+j+j'}}{\ell_i}, Ω j j ′ = i = 1 ∑ N ( v i a i j ) ( v i a i j ′ ) q = i = 1 ∑ N v i q + 1 a i j + j ′ = i = 1 ∑ N ℓ i a i μ + j + j ′ ,
with 0 ≤ j , j ′ < c 0\leq j,j'<c 0 ≤ j , j ′ < c . If j < h H j<h_H j < h H or j ′ < h H j'<h_H j ′ < h H , then
0 ≤ μ + j + j ′ ≤ ( N − c − h H ) + ( h H − 1 ) + ( c − 1 ) = N − 2. 0\leq\mu+j+j'\leq(N-c-h_H)+(h_H-1)+(c-1)=N-2. 0 ≤ μ + j + j ′ ≤ ( N − c − h H ) + ( h H − 1 ) + ( c − 1 ) = N − 2.
By Lemma VI.7, we have Ω j j ′ = 0 \Omega_{jj'}=0 Ω j j ′ = 0 whenever j < h H j<h_H j < h H or j ′ < h H j'<h_H j ′ < h H . Let g 0 = c − h H g_0=c-h_H g 0 = c − h H , and let Δ = ( Δ a b ) 0 ≤ a , b < g 0 \Delta=(\Delta_{ab})_{0\leq a,b<g_0} Δ = ( Δ ab ) 0 ≤ a , b < g 0 be the lower-right block of Ω \Omega Ω .
For 0 ≤ a , b < g 0 0\leq a,b<g_0 0 ≤ a , b < g 0 , set j = h H + a j=h_H+a j = h H + a and j ′ = h H + b j'=h_H+b j ′ = h H + b . Since μ = N − c − h H \mu=N-c-h_H μ = N − c − h H and g 0 = c − h H g_0=c-h_H g 0 = c − h H , we have
Δ a b = Ω h H + a , h H + b = ∑ i = 1 N a i μ + 2 h H + a + b ℓ i = ∑ i = 1 N a i N − g 0 + a + b ℓ i . \Delta_{ab}=\Omega_{h_H+a,h_H+b}=\sum_{i=1}^{N}\frac{a_i^{\mu+2h_H+a+b}}{\ell_i}=\sum_{i=1}^{N}\frac{a_i^{N-g_0+a+b}}{\ell_i}. Δ ab = Ω h H + a , h H + b = i = 1 ∑ N ℓ i a i μ + 2 h H + a + b = i = 1 ∑ N ℓ i a i N − g 0 + a + b .
By Lemma VI.7,
Δ a b = { 0 , if a + b ≤ g 0 − 2 , 1 , if a + b = g 0 − 1. \Delta_{ab}=
\begin{cases}
0, & \text{if }a+b\leq g_0-2,\\
1, & \text{if }a+b=g_0-1.
\end{cases} Δ ab = { 0 , 1 , if a + b ≤ g 0 − 2 , if a + b = g 0 − 1.
Thus
Ω = ( 0 0 0 Δ ) . \Omega=\begin{pmatrix}0&0\\0&\Delta\end{pmatrix}. Ω = ( 0 0 0 Δ ) .
Let J g 0 J_{g_0} J g 0 be the g 0 × g 0 g_0\times g_0 g 0 × g 0 matrix with ones on the anti-diagonal and zeros elsewhere. Then Δ J g 0 \Delta J_{g_0} Δ J g 0 is obtained from Δ \Delta Δ by reversing the order of its columns. Since ( Δ J g 0 ) a b = Δ a , g 0 − 1 − b (\Delta J_{g_0})_{ab}=\Delta_{a,g_0-1-b} ( Δ J g 0 ) ab = Δ a , g 0 − 1 − b , we have ( Δ J g 0 ) a b = 0 (\Delta J_{g_0})_{ab}=0 ( Δ J g 0 ) ab = 0 for a < b a<b a < b and ( Δ J g 0 ) a a = 1 (\Delta J_{g_0})_{aa}=1 ( Δ J g 0 ) aa = 1 . Thus Δ J g 0 \Delta J_{g_0} Δ J g 0 is lower triangular with unit diagonal, and
det Δ ⋅ det J g 0 = det ( Δ J g 0 ) = 1. \det\Delta\cdot\det J_{g_0}=\det(\Delta J_{g_0})=1. det Δ ⋅ det J g 0 = det ( Δ J g 0 ) = 1.
Hence det Δ ≠ 0 \det\Delta\ne0 det Δ = 0 and rank q 2 Ω = g 0 \operatorname{rank}_{q^2}\Omega=g_0 rank q 2 Ω = g 0 .
We then determine the Hermitian hull of C C C . Let x = ( x 0 , x 1 , … , x c − 1 ) \boldsymbol{x}=(x_0,x_1,\ldots,x_{c-1}) x = ( x 0 , x 1 , … , x c − 1 ) and y = ( y 0 , y 1 , … , y c − 1 ) \boldsymbol{y}=(y_0,y_1,\ldots,y_{c-1}) y = ( y 0 , y 1 , … , y c − 1 ) be coefficient rows in F q 2 \mathbb{F}_{q^2} F q 2 . For every 1 ≤ i ≤ N 1\leq i\leq N 1 ≤ i ≤ N , we have
( x M ) i = ∑ j = 0 c − 1 x j v i a i j , ( y M ) i = ∑ j ′ = 0 c − 1 y j ′ v i a i j ′ . (57) (\boldsymbol{x}M)_i=\sum_{j=0}^{c-1}x_jv_i a_i^j,\qquad(\boldsymbol{y}M)_i=\sum_{j'=0}^{c-1}y_{j'}v_i a_i^{j'}. \tag*{(57)} ( x M ) i = j = 0 ∑ c − 1 x j v i a i j , ( y M ) i = j ′ = 0 ∑ c − 1 y j ′ v i a i j ′ . (57)
Then
⟨ x M , y M ⟩ H = ∑ i = 1 N ( x M ) i ( ( y M ) i ) q = ∑ i = 1 N ( ∑ j = 0 c − 1 x j v i a i j ) ( ∑ j ′ = 0 c − 1 y j ′ v i a i j ′ ) q = ∑ i = 1 N ∑ j = 0 c − 1 ∑ j ′ = 0 c − 1 x j y j ′ q ( v i a i j ) ( v i a i j ′ ) q = ∑ j = 0 c − 1 ∑ j ′ = 0 c − 1 x j ( ∑ i = 1 N ( v i a i j ) ( v i a i j ′ ) q ) y j ′ q = ∑ j = 0 c − 1 ∑ j ′ = 0 c − 1 x j Ω j j ′ y j ′ q = x Ω ( y ( q ) ) T . \langle xM,yM\rangle_H=\sum_{i=1}^{N}(xM)_i((yM)_i)^q
=\sum_{i=1}^{N}\left(\sum_{j=0}^{c-1}x_jv_ia_i^j\right)\left(\sum_{j'=0}^{c-1}y_{j'}v_ia_i^{j'}\right)^q
=\sum_{i=1}^{N}\sum_{j=0}^{c-1}\sum_{j'=0}^{c-1}x_jy_{j'}^q(v_ia_i^j)(v_ia_i^{j'})^q
=\sum_{j=0}^{c-1}\sum_{j'=0}^{c-1}x_j\left(\sum_{i=1}^{N}(v_ia_i^j)(v_ia_i^{j'})^q\right)y_{j'}^q
=\sum_{j=0}^{c-1}\sum_{j'=0}^{c-1}x_j\Omega_{jj'}y_{j'}^q
=x\Omega(y^{(q)})^{T}. ⟨ x M , y M ⟩ H = i = 1 ∑ N ( x M ) i (( y M ) i ) q = i = 1 ∑ N ( j = 0 ∑ c − 1 x j v i a i j ) j ′ = 0 ∑ c − 1 y j ′ v i a i j ′ q = i = 1 ∑ N j = 0 ∑ c − 1 j ′ = 0 ∑ c − 1 x j y j ′ q ( v i a i j ) ( v i a i j ′ ) q = j = 0 ∑ c − 1 j ′ = 0 ∑ c − 1 x j ( i = 1 ∑ N ( v i a i j ) ( v i a i j ′ ) q ) y j ′ q = j = 0 ∑ c − 1 j ′ = 0 ∑ c − 1 x j Ω j j ′ y j ′ q = x Ω ( y ( q ) ) T .
Since C = { y M : y ∈ F q 2 c } C=\{yM:y\in\mathbb{F}_{q^2}^{c}\} C = { y M : y ∈ F q 2 c } and y ↦ y ( q ) y\mapsto y^{(q)} y ↦ y ( q ) is a bijection on F q 2 c \mathbb{F}_{q^2}^{c} F q 2 c , we have
x M ∈ C ∩ C ⊥ H ⟺ x Ω ( y ( q ) ) T = 0 for every y ∈ F q 2 c xM\in C\cap C^{\perp_H}\Longleftrightarrow x\Omega(y^{(q)})^{T}=0\quad\text{for every }y\in\mathbb{F}_{q^2}^{c} x M ∈ C ∩ C ⊥ H ⟺ x Ω ( y ( q ) ) T = 0 for every y ∈ F q 2 c
⟺ x Ω = 0. \Longleftrightarrow x\Omega=0. ⟺ x Ω = 0.
By the block form of Ω \Omega Ω and the nonsingularity of Δ \Delta Δ ,
x Ω = 0 ⟺ ( x h H , … , x c − 1 ) Δ = 0 x\Omega=0\Longleftrightarrow(x_{h_H},\ldots,x_{c-1})\Delta=0 x Ω = 0 ⟺ ( x h H , … , x c − 1 ) Δ = 0
⟺ x j = 0 for every h H ≤ j < c . \Longleftrightarrow x_j=0\quad\text{for every }h_H\leq j<c. ⟺ x j = 0 for every h H ≤ j < c .
Thus the hull is generated by the first h H h_H h H rows of M M M , and hence C ∩ C ⊥ H = GRS h H ( a , v ) C\cap C^{\perp_H}=\operatorname{GRS}_{h_H}(a,v) C ∩ C ⊥ H = GRS h H ( a , v ) . The codes C C C and its hull are MDS, with parameters [ N , c , N − c + 1 ] q 2 [N,c,N-c+1]_{q^2} [ N , c , N − c + 1 ] q 2 and [ N , h H , N − h H + 1 ] q 2 [N,h_H,N-h_H+1]_{q^2} [ N , h H , N − h H + 1 ] q 2 , respectively. Finally, since H = C ∩ C ⊥ H H=C\cap C^{\perp_H} H = C ∩ C ⊥ H is Hermitian self-orthogonal and 2 h H < c + h H < N 2h_H<c+h_H<N 2 h H < c + h H < N , we may apply Theorem VI.10. For every 0 ≤ g ≤ B ( N − 2 h H ) − 1 0\leq g\leq B(N-2h_H)-1 0 ≤ g ≤ B ( N − 2 h H ) − 1 , we obtain a pure code with the stated quantum parameters, stabilizer label space, and locality. This code attains all four bounds.
Combining the hull identity in [30 Theorem 10] with Theorem VI.10, we obtain the following family.
Proposition VI.12. Let q q q be a prime power, let a a a list all elements of F q 2 \mathbb{F}_{q^2} F q 2 , and let 1 \mathbf{1} 1 be the all-one vector of length q 2 q^2 q 2 . Then Hull H ( GRS q ( a , 1 ) ) = GRS q − 1 ( a , 1 ) \operatorname{Hull}_H(\operatorname{GRS}_q(a,\mathbf{1}))=\operatorname{GRS}_{q-1}(a,\mathbf{1}) Hull H ( GRS q ( a , 1 )) = GRS q − 1 ( a , 1 ) . For every pair of integers B ≥ 1 B\geq1 B ≥ 1 and 0 ≤ g ≤ B ( q 2 − 2 q + 2 ) − 1 0\leq g\leq B(q^2-2q+2)-1 0 ≤ g ≤ B ( q 2 − 2 q + 2 ) − 1 , there exists a pure F q \mathbb{F}_q F q -linear ( r , δ ) (r,\delta) ( r , δ ) -SSLRC with parameters
[ [ B q 2 , B ( q 2 − 2 q + 2 ) − g , g , q ] ] q , ( r , δ ) = ( q 2 − q + 1 , q ) . [[Bq^2,B(q^2-2q+2)-g,g,q]]_q,\qquad(r,\delta)=(q^2-q+1,q). [[ B q 2 , B ( q 2 − 2 q + 2 ) − g , g , q ] ] q , ( r , δ ) = ( q 2 − q + 1 , q ) . Its stabilizer label space is S = ( ψ − 1 ( GRS q − 1 ( a , 1 ) ) ) ⊕ B S=(\psi^{-1}(\operatorname{GRS}_{q-1}(a,\mathbf{1})))^{\oplus B} S = ( ψ − 1 ( GRS q − 1 ( a , 1 )) ) ⊕ B . The code attains all the bounds stated in Theorem VI.10.
Proof. By [30 Theorem 10] , we have
H : = Hull H ( GRS q ( a , 1 ) ) = GRS q − 1 ( a , 1 ) . H:=\operatorname{Hull}_H(\operatorname{GRS}_q(a,\mathbf{1}))=\operatorname{GRS}_{q-1}(a,\mathbf{1}). H := Hull H ( GRS q ( a , 1 )) = GRS q − 1 ( a , 1 ) .
Thus H H H is an [ q 2 , q − 1 , q 2 − q + 2 ] q 2 [q^2,q-1,q^2-q+2]_{q^2} [ q 2 , q − 1 , q 2 − q + 2 ] q 2 MDS code and H ⊆ H ⊥ H H\subseteq H^{\perp_H} H ⊆ H ⊥ H . Since q 2 − 2 ( q − 1 ) = ( q − 1 ) 2 + 1 > 0 q^2-2(q-1)=(q-1)^2+1>0 q 2 − 2 ( q − 1 ) = ( q − 1 ) 2 + 1 > 0 , we may apply Theorem VI.10 with N = q 2 N=q^2 N = q 2 and h H = q − 1 h_H=q-1 h H = q − 1 . Hence the resulting codes have the stated parameters and attain all four bounds.
Example VI.13. For Table III , let ζ \zeta ζ be a primitive element of F q \mathbb{F}_q F q , set a i = ζ i − 1 a_i=\zeta^{i-1} a i = ζ i − 1 for 1 ≤ i ≤ N 1\leq i\leq N 1 ≤ i ≤ N , and choose the multipliers as in (55). For Table IV , let a a a list all elements of F q 2 \mathbb{F}_{q^2} F q 2 and set v = 1 v=\mathbf{1} v = 1 . By Theorem VI.11 and Proposition VI.12, respectively, we obtain the pure F q \mathbb{F}_q F q -linear ( r , δ ) (r,\delta) ( r , δ ) -SSLRCs in the two tables. Each code has d = δ d=\delta d = δ , and its B B B coordinate blocks form pairwise disjoint repair sets. The choices g = B ( c − h H ) g=B(c-h_H) g = B ( c − h H ) and g = B g=B g = B recover the original examples in Table III and Table IV , respectively. The remaining rows correspond to other admissible gauge dimensions. All listed codes attain the Singleton-type bound (41), the alphabet-dependent bound (43), the F q 2 \mathbb{F}_{q^2} F q 2 -linear Griesmer bound (49), and the disjoint redundancy bound (32).
Table III. Optimal Pure ( r , δ ) (r,\delta) ( r , δ ) -SSLRCS from Theorem VI.11.
Table IV. Optimal Pure ( r , δ ) (r,\delta) ( r , δ ) -SSLRCS from Proposition VI.12.
Conclusion and Outlook
In this paper, we characterized exact local erasure recovery for ( r , δ ) (r,\delta) ( r , δ ) -SSLRCS algebraically without assuming purity. Using this criterion, we derived redundancy, supported-subspace, and mixed counting bounds. For pure codes, we used associated additive classical codes to obtain Singleton-type and alphabet-dependent bounds, together with explicit Griesmer, Hamming, and Plotkin inequalities. We also proved that logical and gauge dimensions can be traded while preserving the stabilizer label space, minimum distance, and repair cover. Using MDS stabilizer codes, RS codes, and MDS Hermitian hulls, we constructed pure families with d = δ d=\delta d = δ that simultaneously attain the Singleton-type, alphabet-dependent, F q 2 \mathbb{F}_{q^2} F q 2 -linear Griesmer, and disjoint redundancy bounds.
The constructions in Section VI do not address equality in every bound proved here. One problem for future work is to construct pure families attaining the mixed counting bound (38) in the presence of both local erasures and errors outside the selected repair sets. Another problem is to construct pure ( r , δ ) (r,\delta) ( r , δ ) -SSLRCS with d > δ d>\delta d > δ that attain the bounds in Section V while preserving the prescribed repair cover. The supported-subspace bounds also suggest a separate study of optimal impure families and their gauge-equivalence classes.