Introduction
Let X \mathcal{X} X be a space of matrices over a finite field. The rank metric on X \mathcal{X} X is
d r k ( A , B ) = rk ( A − B ) , d_{\mathrm{rk}}(A,B)=\operatorname{rk}(A-B), d rk ( A , B ) = rk ( A − B ) ,
and a nonempty subset C ⊆ X \mathcal{C} \subseteq\mathcal{X} C ⊆ X is a d d d -code if rk ( A − B ) ≥ d \operatorname{rk}(A-B) \ge d rk ( A − B ) ≥ d whenever A , B ∈ C A,B \in\mathcal{C} A , B ∈ C are distinct. The basic extremal questions are to determine the largest possible cardinality, to construct codes attaining the relevant bound, and to classify the extremal codes under the natural rank-metric isometries.
For unrestricted rectangular matrices, the theory began with two independent constructions. Delsarte used the association scheme of bilinear forms, while Gabidulin used extension fields and linearized polynomials [3 , 6 ] . Both constructions attain the rank-metric Singleton bound and give the classical maximum rank-distance, or MRD, codes. These two viewpoints—bilinear forms and linearized-polynomial evaluation—remain fundamental in rank-metric coding theory.
A parallel extremal theory arises in matrix spaces defined by an adjoint symmetry. We use the term classical form spaces for the spaces of alternating and symmetric bilinear forms and of Hermitian sesquilinear forms; after choosing bases, these become the spaces of alternating, symmetric, and Hermitian matrices. Delsarte and Goethals obtained sharp bounds and optimal constructions in ends_mid=1 the alternating-form space [4 ] . Schmidt subsequently established corresponding results for symmetric forms in odd and even characteristic and for Hermitian forms [18 , 19 , 20 , 21 ] . For some parameters of symmetric-form spaces and Hermitian-form spaces, the sharp upper bound for arbitrary codes is not known, whereas the corresponding additive problem has a sharp bound. Accordingly, a maximum additive d d d -code means an additive subgroup that attains the sharp upper bound in its ambient form space.
Let H n ( F q 2 / F q ) \mathcal{H}_n(\mathbb{F}_{q^2}/\mathbb{F}_q) H n ( F q 2 / F q ) be the space of n × n n \times n n × n Hermitian matrices over F q 2 \mathbb{F}_{q^2} F q 2 . Schmidt proved that every additive Hermitian d d d -code C C C satisfies
∣ C ∣ ≤ q n ( n − d + 1 ) . (1) |C| \leq q^{n(n-d+1)}. \tag*{(1)} ∣ C ∣ ≤ q n ( n − d + 1 ) . (1)
He also constructed codes attaining (1) for all admissible pairs ( n , d ) (n,d) ( n , d ) , except when n n n and d d d are both even and
4 ≤ d ≤ n − 2 ; 4 \leq d \leq n-2; 4 ≤ d ≤ n − 2 ;
the endpoint cases d = 2 d=2 d = 2 and d = n d=n d = n were already known [20 ] . See also [24 ] for another construction of codes attaining the bound in (1) when d = 2 d=2 d = 2 . Thus the even–even range was the last unresolved existence case for maximum additive d d d -codes among the classical form spaces described above.
The first purpose of this paper is to close this gap. More generally, we give a uniform construction over every quadratic extension F q s / F q s / 2 \mathbb{F}_{q^s}/\mathbb{F}_{q^{s/2}} F q s / F q s /2 , where s ≥ 2 s \geq2 s ≥ 2 is even. Set
L = F q n , K = F q , L = \mathbb{F}_{q^n}, \qquad K = \mathbb{F}_q, L = F q n , K = F q ,
let σ \sigma σ generate Gal ( L / K ) \operatorname{Gal}(L/K) Gal ( L / K ) , and consider the skew polynomial ring
R = L [ X ; σ ] , X α = σ ( α ) X ( α ∈ L ) . R = L[X;\sigma], \qquad X\alpha= \sigma(\alpha)X \quad(\alpha\in L). R = L [ X ; σ ] , X α = σ ( α ) X ( α ∈ L ) .
Choose a monic irreducible self-reciprocal polynomial F ( Y ) ∈ K [ Y ] F(Y) \in K[Y] F ( Y ) ∈ K [ Y ] of degree s s s . Since F ( X n ) F(X^n) F ( X n ) is central in R R R , the quotient
R F : = R / R F ( X n ) R_F := R/RF(X^n) R F := R / R F ( X n )
is a central simple algebra of degree n n n over its center E F ≅ F q s \mathbb{E}_F \cong\mathbb{F}_{q^s} E F ≅ F q s . Self-reciprocity of F F F yields an involution of the second kind
Θ : R F ⟶ R F , \Theta: R_F \longrightarrow R_F, Θ : R F ⟶ R F ,
whose restriction ϱ \varrho ϱ to E F \mathbb{E}_F E F has fixed field E F ϱ ≅ F q s / 2 \mathbb{E}_F^\varrho\cong\mathbb{F}_{q^{s/2}} E F ϱ ≅ F q s /2 . Under a suitable splitting isomorphism R F ≅ M n ( E F ) R_F \cong M_n(\mathbb{E}_F) R F ≅ M n ( E F ) , the fixed-point space Fix ( Θ ) \operatorname{Fix}(\Theta) Fix ( Θ ) is identified with
H n ( E F / E F ϱ ) . \mathcal{H}_n(\mathbb{E}_F/\mathbb{E}_F^\varrho). H n ( E F / E F ϱ ) .
Thus the Hermitian matrix space is realized intrinsically as the symmetric part of a central simple algebra with unitary involution.
For 0 ≤ ℓ ≤ n s / 2 − 1 0 \leq\ell\leq ns/2 - 1 0 ≤ ℓ ≤ n s /2 − 1 , define
S ℓ ( F ) : = { α 0 + ∑ i = 1 ℓ ( σ − i ( α i ) X − i + α i X i ) + R F ( X n ) : α 0 , … , α ℓ ∈ F q n } . (2) S_\ell(F) := \left\{ \alpha_0 + \sum_{i=1}^{\ell} \left(\sigma^{-i}(\alpha_i)X^{-i} + \alpha_iX^i\right) + RF(X^n) : \alpha_0,\ldots,\alpha_\ell\in\mathbb{F}_{q^n} \right\}. \tag*{(2)} S ℓ ( F ) := { α 0 + i = 1 ∑ ℓ ( σ − i ( α i ) X − i + α i X i ) + R F ( X n ) : α 0 , … , α ℓ ∈ F q n } . (2)
The paired coefficients are exactly the condition that the displayed elements be fixed by Θ \Theta Θ . For 1 ≤ d ≤ n 1 \leq d \leq n 1 ≤ d ≤ n , set
ℓ = ℓ d : = s ( n − d + 1 ) − 2 2 . \ell= \ell_d := \frac{s(n-d+1)-2}{2}. ℓ = ℓ d := 2 s ( n − d + 1 ) − 2 .
We prove that S ℓ d ( F ) \mathcal{S}_{\ell d}(F) S ℓ d ( F ) has minimum rank distance d d d and cardinality
q s n ( n − d + 1 ) / 2 = ∣ F q s / 2 ∣ n ( n − d + 1 ) . q^{sn(n-d+1)/2}=\left|\mathbb{F}_{q^{s/2}}\right|^{n(n-d+1)}. q s n ( n − d + 1 ) /2 = F q s /2 n ( n − d + 1 ) .
It therefore attains the sharp Hermitian bound. Consequently, for any even s s s our construction yields maximum additive Hermitian d d d -codes in H n ( F q s / F q s / 2 ) \mathcal{H}_n(\mathbb{F}_{q^s}/\mathbb{F}_{q^{s/2}}) H n ( F q s / F q s /2 ) for every ( n , d ) (n,d) ( n , d ) , including the previously open even–even range.
We next study equivalence. Let E / F \mathbb{E}/\mathbb{F} E / F be a quadratic extension of finite fields, let A ∗ : = ( A ϱ ) T A^*:=\left(A^\varrho\right)^{\mathsf{T}} A ∗ := ( A ϱ ) T , and let C , D ⊆ H n ( E / F ) \mathcal{C},\mathcal{D}\subseteq\mathcal{H}_n(\mathbb{E}/\mathbb{F}) C , D ⊆ H n ( E / F ) be additive codes. The full semilinear equivalence relation used in this paper is the following: C \mathcal{C} C and D \mathcal{D} D are Hermitian equivalent if
D = { μ N A φ N ∗ : A ∈ C } (3) \mathcal{D}=\{\mu NA^\varphi N^*:A\in\mathcal{C}\} \tag*{(3)} D = { μ N A φ N ∗ : A ∈ C } (3)
for some N ∈ GL n ( E ) N\in\operatorname{GL}_n(\mathbb{E}) N ∈ GL n ( E ) , μ ∈ F × \mu\in\mathbb{F}^{\times} μ ∈ F × , and φ ∈ Aut ( E ) \varphi\in\operatorname{Aut}(\mathbb{E}) φ ∈ Aut ( E ) , where A φ A^\varphi A φ is obtained by applying φ \varphi φ entrywise. These are precisely the semilinear rank-metric isometries fixing the zero matrix, up to the standard translation term [25 ] .
Our analysis of equivalence is based on two structural invariants that we introduce. The first is the scalar field
F ( C ) : = { λ ∈ F : λ C ⊆ C } . \mathcal{F}(\mathcal{C}):=\{\lambda\in\mathbb{F}:\lambda\mathcal{C}\subseteq\mathcal{C}\}. F ( C ) := { λ ∈ F : λ C ⊆ C } .
If Φ ( A ) = μ N A φ N ∗ \Phi(A)=\mu NA^\varphi N^* Φ ( A ) = μ N A φ N ∗ gives an equivalence from C \mathcal{C} C to D \mathcal{D} D , then
λ Φ ( A ) = Φ ( φ − 1 ( λ ) A ) , \lambda\Phi(A)=\Phi\left(\varphi^{-1}(\lambda)A\right), λ Φ ( A ) = Φ ( φ − 1 ( λ ) A ) ,
and hence
F ( D ) = φ ( F ( C ) ) . \mathcal{F}(\mathcal{D})=\varphi(\mathcal{F}(\mathcal{C})). F ( D ) = φ ( F ( C )) .
Thus the isomorphism type, and in particular the cardinality, of the scalar field is an invariant of Hermitian equivalence. Schmidt’s trace constructions are linear over the entire fixed field, whereas, for s > 2 s>2 s > 2 , our codes have a proper scalar field; when ℓ d < n s / 4 \ell d<ns/4 ℓ d < n s /4 , it is exactly F q \mathbb{F}_q F q . This separates the new codes from Schmidt’s constructions whenever the latter exist with the same parameters.
The second invariant is obtained from Lyapunov-type operators. For b ‾ ∈ R F \overline{b}\in R_F b ∈ R F , put
Δ b ‾ ( z ‾ ) : = b ‾ z ‾ + z ‾ Θ ( b ‾ ) , z ‾ ∈ Fix ( Θ ) , \Delta_{\overline{b}}(\overline{z}):=\overline{b}\,\overline{z}+\overline{z}\,\Theta(\overline{b}),\qquad\overline{z}\in\operatorname{Fix}(\Theta), Δ b ( z ) := b z + z Θ ( b ) , z ∈ Fix ( Θ ) ,
and define
L ( C ) : = { b ‾ ∈ R F : Δ b ‾ ( C ) ⊆ C } . \mathcal{L}(\mathcal{C}):=\{\overline{b}\in R_F:\Delta_{\overline{b}}(\mathcal{C})\subseteq\mathcal{C}\}. L ( C ) := { b ∈ R F : Δ b ( C ) ⊆ C } .
After passage to matrices, Δ b ‾ \Delta_{\overline{b}} Δ b becomes the Hermitian Lyapunov operator A ↦ B A + A B ∗ A\mapsto BA+AB^* A ↦ B A + A B ∗ . The corresponding Lyapunov equations are classical objects in matrix analysis and control theory, and have been studied from several viewpoints; see, for instance [1 , 10 , 13 , 14 ] . Its transformation under (3) is immediate. Since finite-field automorphisms commute with ϱ \varrho ϱ , if B ′ = N B φ N − 1 B'=NB^\varphi N^{-1} B ′ = N B φ N − 1 , then
Δ B ′ ( μ N A φ N ∗ ) = μ N Δ B ( A ) φ N ∗ . \Delta_{B'}(\mu NA^\varphi N^*)=\mu N\Delta_B(A)^\varphi N^*. Δ B ′ ( μ N A φ N ∗ ) = μ N Δ B ( A ) φ N ∗ .
Consequently,
L ( D ) = N L ( C ) φ N − 1 . \mathcal{L}(\mathcal{D})=N\mathcal{L}(\mathcal{C})^\varphi N^{-1}. L ( D ) = N L ( C ) φ N − 1 .
For n ≥ 3 n\geq3 n ≥ 3 and 1 ≤ ℓ < n s / 4 1\leq\ell<ns/4 1 ≤ ℓ < n s /4 , we determine this invariant explicitly:
L ( S ℓ ( F ) ) = L + B F , B F : = { λ ∈ E F : λ + Θ ( λ ) ∈ F q } . \mathcal{L}(\mathcal{S}_{\ell}(F))=\mathbb{L}+\mathbb{B}_F,\qquad\mathbb{B}_F:=\{\lambda\in\mathbb{E}_F:\lambda+\Theta(\lambda)\in\mathbb{F}_q\}. L ( S ℓ ( F )) = L + B F , B F := { λ ∈ E F : λ + Θ ( λ ) ∈ F q } .
The distinguished copy of L = F q n \mathbb{L}=\mathbb{F}_{q^n} L = F q n can be recovered from this additive subgroup. Hence an equivalence between two members of the family must normalize that field, reducing an arbitrary Hermitian isometry to a strongly constrained comparison in a common semilinear model.
Based on this determination of the Lyapunov invariant, Section 6 proves strong necessary conditions for equivalence between S ℓ ( F ) \mathcal{S}_{\ell}(F) S ℓ ( F ) and S ℓ ( G ) \mathcal{S}_{\ell}(G) S ℓ ( G ) . A full Hermitian equivalence may first apply an automorphism of E = F q s \mathbb{E}=\mathbb{F}_{q^s} E = F q s ; the conditions below are therefore understood up to this field automorphism. If τ \tau τ denotes the induced automorphism of K = F q \mathbb{K}=\mathbb{F}_q K = F q , write F τ F^{\tau} F τ for the polynomial obtained by applying τ \tau τ to the coefficients of F F F . Assume now that n > 2 n>2 n > 2 . The conclusions are as follows.
1 ≤ ℓ < min { n , n s / 4 } , 1\leq\ell<\min\{n,ns/4\}, 1 ≤ ℓ < min { n , n s /4 } ,
the equivalence between S ℓ ( F ) \mathcal{S}_{\ell}(F) S ℓ ( F ) and S ℓ ( G ) \mathcal{S}_{\ell}(G) S ℓ ( G ) leads to
F τ ( Y ) = λ s G ( λ − 1 Y ) F^{\tau}(Y)=\lambda^{s}G(\lambda^{-1}Y) F τ ( Y ) = λ s G ( λ − 1 Y )
for some λ ∈ F q × \lambda\in\mathbb{F}_q^{\times} λ ∈ F q × .
n < ℓ < n s 4 , n<\ell<\frac{ns}{4}, n < ℓ < 4 n s ,
choose roots u F , u G ∈ F q s u_F,u_G\in\mathbb{F}_{q^s} u F , u G ∈ F q s and put w H = u H + u H − 1 w_H=u_H+u_H^{-1} w H = u H + u H − 1 . Then the equivalence between S ℓ ( F ) \mathcal{S}_{\ell}(F) S ℓ ( F ) and S ℓ ( G ) \mathcal{S}_{\ell}(G) S ℓ ( G ) leads to
w G = a φ ( w F ) + b c φ ( w F ) + d w_G=\frac{a\varphi(w_F)+b}{c\varphi(w_F)+d} w G = c φ ( w F ) + d a φ ( w F ) + b
for some a , b , c , d ∈ F q a,b,c,d\in\mathbb{F}_q a , b , c , d ∈ F q with a d − b c ≠ 0 ad-bc\neq0 a d − b c = 0 , where φ \varphi φ is the field automorphism occurring in the equivalence.
Thus, up to field automorphisms, possible equivalences are confined to explicit scalar or PGL 2 ( F q ) \operatorname{PGL}_2(\mathbb{F}_q) PGL 2 ( F q ) -orbits of the associated polynomials and elements.
These restrictions give quantitative lower bounds under full Hermitian equivalence. Let N q ( s ) N_q(s) N q ( s ) be the number of monic irreducible self-reciprocal polynomials of degree s s s over F q \mathbb{F}_q F q , write q = p h q=p^h q = p h , and retain ℓ d = ( s ( n − d + 1 ) − 2 ) / 2 \ell_d=(s(n-d+1)-2)/2 ℓ d = ( s ( n − d + 1 ) − 2 ) /2 . Define
L 2 ( q ) : = { q / 2 , q even , ( q − 1 ) / 4 , q ≡ 1 ( m o d 4 ) , ( q + 1 ) / 4 , q ≡ 3 ( m o d 4 ) . L_2(q):=
\begin{cases}
q/2, & q\text{ even},\\
(q-1)/4, & q\equiv1\pmod{4},\\
(q+1)/4, & q\equiv3\pmod{4}.
\end{cases} L 2 ( q ) := ⎩ ⎨ ⎧ q /2 , ( q − 1 ) /4 , ( q + 1 ) /4 , q even , q ≡ 1 ( mod 4 ) , q ≡ 3 ( mod 4 ) .
The family contains at least the following numbers of pairwise Hermitian inequivalent maximum additive codes:
⌈ L 2 ( q ) / ( 2 h ) ⌉ \lceil L_2(q)/(2h)\rceil ⌈ L 2 ( q ) / ( 2 h )⌉ if s = 2 s=2 s = 2 and n / 2 < d < n n/2<d<n n /2 < d < n ;
⌈ N q ( s ) / ( s h gcd ( s , q − 1 ) ) ⌉ \lceil N_q(s)/(sh\gcd(s,q-1))\rceil ⌈ N q ( s ) / ( s h g cd( s , q − 1 ))⌉ if s ≥ 4 s\geq4 s ≥ 4 is even and ℓ d < n \ell_d<n ℓ d < n ;
N q ( s ) / ( s h q ( q 2 − 1 ) ) N_q(s)/(shq(q^2-1)) N q ( s ) / ( s h q ( q 2 − 1 )) if s > 4 s>4 s > 4 is even and n < ℓ d < n s / 4 n<\ell_d<ns/4 n < ℓ d < n s /4 .
Since N q ( s ) = Θ ( q s / 2 ) N_q(s)=\Theta(q^{s/2}) N q ( s ) = Θ ( q s /2 ) as q → ∞ q\to\infty q → ∞ , these lower bounds are respectively on the scales q / h q/h q / h , q s / 2 / h q^{s/2}/h q s /2 / h , and q s / 2 − 3 / h q^{s/2-3}/h q s /2 − 3 / h for fixed admissible s s s . In particular, the construction that closes the last existence gap is far from unique.
Organization of the paper
1.1. Organization of the paper. Section 2 reviews Hermitian rank-metric codes, skew polynomial rings, and the arithmetic of irreducible self-reciprocal polynomials. Section 3 constructs the central simple algebra R F R_F R F , its involution of the second kind, and the identification of its fixed points with a Hermitian matrix space. Section 4 defines the codes S ℓ ( F ) \mathcal{S}_{\ell}(F) S ℓ ( F ) , proves their optimality, recalls Schmidt’s trace constructions, and separates the two families by the scalar-field invariant. § 5 introduces the Lyapunov invariant and computes it for the new codes. Finally, § 6 places the constructions in a common semilinear model, proves the necessary equivalence conditions in the short and large exponent ranges, and derives the lower bounds for full Hermitian equivalence.
Preliminaries
For a field K \mathbb{K} K , let M n ( K ) M_n(\mathbb{K}) M n ( K ) denote the ring of n × n n \times n n × n matrices over K \mathbb{K} K .
Hermitian d-codes
2.1. Hermitian d d d -codes. Let E / F \mathbb{E}/\mathbb{F} E / F be a quadratic extension of finite fields, and let ρ : E ⟶ E \rho: \mathbb{E} \longrightarrow\mathbb{E} ρ : E ⟶ E be the nontrivial element of Gal ( E / F ) \operatorname{Gal}(\mathbb{E}/\mathbb{F}) Gal ( E / F ) . Let V = V ( n , E ) V = V(n,\mathbb{E}) V = V ( n , E ) be an n n n -dimensional vector space over E \mathbb{E} E . A map
H : V × V ⟶ E H : V \times V \longrightarrow\mathbb{E} H : V × V ⟶ E
is called a Hermitian form if it is E \mathbb{E} E -linear in the first variable and satisfies
H ( v , u ) = ρ ( H ( u , v ) ) , H(v,u) = \rho(H(u,v)), H ( v , u ) = ρ ( H ( u , v )) ,
for all u , v ∈ V u,v \in V u , v ∈ V . Fix a basis B = ( ξ 1 , … , ξ n ) \mathcal{B} = (\xi_1,\ldots,\xi_n) B = ( ξ 1 , … , ξ n ) of V V V over E \mathbb{E} E . The form H H H is represented by the matrix
M B ( H ) : = ( H ( ξ i , ξ j ) ) 1 ≤ i , j ≤ n ∈ M n ( E ) . M_{\mathcal{B}}(H) := \left(H(\xi_i,\xi_j)\right)_{1 \leq i,j \leq n} \in M_n(\mathbb{E}). M B ( H ) := ( H ( ξ i , ξ j ) ) 1 ≤ i , j ≤ n ∈ M n ( E ) .
We have M B ( H ) ∗ = M B ( H ) M_{\mathcal{B}}(H)^* = M_{\mathcal{B}}(H) M B ( H ) ∗ = M B ( H ) , where A ∗ : = ( A ρ ) ⊤ A^* := (A^\rho)^\top A ∗ := ( A ρ ) ⊤ denotes the conjugate transpose, with conjugation induced by ρ \rho ρ on E \mathbb{E} E . Thus M B ( H ) M_{\mathcal{B}}(H) M B ( H ) is a Hermitian matrix . Conversely, every Hermitian matrix in M n ( E ) M_n(\mathbb{E}) M n ( E ) determines a unique Hermitian form on V V V . We denote by H ( V ) \mathcal{H}(V) H ( V ) the set of all Hermitian forms on V V V , and by H n ( E / F ) \mathcal{H}_n(\mathbb{E}/\mathbb{F}) H n ( E / F ) the set of all n × n n \times n n × n Hermitian matrices over E \mathbb{E} E . Both sets carry a natural structure of F \mathbb{F} F -vector spaces. Hence, after fixing B \mathcal{B} B , one obtains a natural identification between Hermitian forms on V V V and Hermitian matrices of order n n n , i.e., the map
Ψ B : H ( V ) ⟶ H n ( E / F ) , H ⟼ M B ( H ) , (4) \Psi_{\mathcal{B}} : \mathcal{H}(V) \longrightarrow\mathcal{H}_n(\mathbb{E}/\mathbb{F}), \qquad H \longmapsto M_{\mathcal{B}}(H), \tag*{(4)} Ψ B : H ( V ) ⟶ H n ( E / F ) , H ⟼ M B ( H ) , (4)
is a F \mathbb{F} F -linear isomorphism. In particular,
∣ H ( V ) ∣ = ∣ H n ( E / F ) ∣ = ∣ F ∣ n 2 , \lvert\mathcal{H}(V)\rvert= \lvert\mathcal{H}_n(\mathbb{E}/\mathbb{F})\rvert= \lvert\mathbb{F}\rvert^{n^2}, ∣ H ( V )∣ = ∣ H n ( E / F )∣ = ∣ F ∣ n 2 ,
since a Hermitian matrix is determined by its n n n diagonal entries in F \mathbb{F} F and its n ( n − 1 ) / 2 n(n-1)/2 n ( n − 1 ) /2 entries above the diagonal in E \mathbb{E} E . We define
rk ( H ) : = rk ( M B ( H ) ) , \operatorname{rk}(H) := \operatorname{rk}(M_{\mathcal{B}}(H)), rk ( H ) := rk ( M B ( H )) ,
where the matrix rank is computed over E \mathbb{E} E .
We endow the space H ( V ) \mathcal{H}(V) H ( V ) with the rank metric, defined by
d rk ( H 1 , H 2 ) : = rk ( H 1 − H 2 ) = rk ( M B ( H 1 ) − M B ( H 2 ) ) , d_{\operatorname{rk}}(H_1,H_2) := \operatorname{rk}(H_1-H_2) = \operatorname{rk}(M_{\mathcal{B}}(H_1)-M_{\mathcal{B}}(H_2)), d rk ( H 1 , H 2 ) := rk ( H 1 − H 2 ) = rk ( M B ( H 1 ) − M B ( H 2 )) ,
for all H 1 , H 2 ∈ H ( V ) H_1,H_2 \in\mathcal{H}(V) H 1 , H 2 ∈ H ( V ) . Via (4) , we use the same notation for the induced rank metric on H n ( E / F ) \mathcal{H}_n(\mathbb{E}/\mathbb{F}) H n ( E / F ) . Thus a Hermitian rank-metric code is a nonempty subset C ⊆ H ( V ) \mathcal{C} \subseteq\mathcal{H}(V) C ⊆ H ( V ) , or, after fixing B \mathcal{B} B , a nonempty subset C ⊆ H n ( E / F ) \mathcal{C} \subseteq\mathcal{H}_n(\mathbb{E}/\mathbb{F}) C ⊆ H n ( E / F ) . Its minimum distance is
d rk ( C ) : = min { rk ( H 1 − H 2 ) : H 1 , H 2 ∈ C , H 1 ≠ H 2 } . d_{\operatorname{rk}}(\mathcal{C}) := \min\{\operatorname{rk}(H_1-H_2) : H_1,H_2 \in\mathcal{C},\ H_1 \ne H_2\}. d rk ( C ) := min { rk ( H 1 − H 2 ) : H 1 , H 2 ∈ C , H 1 = H 2 } .
For an integer d d d , we call C \mathcal{C} C a Hermitian d d d -code if
d rk ( C ) ≥ d , d_{\operatorname{rk}}(\mathcal{C}) \geq d, d rk ( C ) ≥ d ,
equivalently if every nonzero difference of two distinct codewords has rank at least d d d .
An additive Hermitian code is a Hermitian rank-metric code that is an additive subgroup of H n ( E / F ) \mathcal{H}_n(\mathbb{E}/\mathbb{F}) H n ( E / F ) . The corresponding extremal problem is to maximize ∣ C ∣ |\mathcal{C}| ∣ C ∣ among additive Hermitian d d d -codes in H n ( E / F ) \mathcal{H}_n(\mathbb{E}/\mathbb{F}) H n ( E / F ) . Schmidt proved that every such code satisfies
∣ C ∣ ≤ ∣ F ∣ n ( n − d + 1 ) (5) |\mathcal{C}| \leq|\mathbb{F}|^{n(n-d+1)} \tag*{(5)} ∣ C ∣ ≤ ∣ F ∣ n ( n − d + 1 ) (5)
see [20 Theorem 1] . Codes attaining (5) are called maximum additive Hermitian d d d -codes . Schmidt also constructed such codes in [20 ] for every admissible pair 1 ≤ d ≤ n 1 \leq d \leq n 1 ≤ d ≤ n , except when n n n and d d d are both even and 4 ≤ d ≤ n − 2 4 \leq d \leq n - 2 4 ≤ d ≤ n − 2 . We recall these constructions in Section 4.2.
Central simple algebras from skew polynomial rings
We begin by recalling that a central simple algebra over a field F \mathbb{F} F is a finite-dimensional associative F \mathbb{F} F -algebra whose center is exactly F \mathbb{F} F and which has no nontrivial two-sided ideals. Since throughout this paper we work over finite fields, the Artin–Wedderburn theorem together with Wedderburn’s little theorem implies that every central simple F \mathbb{F} F -algebra is F \mathbb{F} F -isomorphic to a full matrix algebra over F \mathbb{F} F . We shall realize these matrix algebras in a way that is well suited to our constructions, namely through skew polynomial rings.
Let L / K \mathbb{L}/\mathbb{K} L / K be a finite field extension of degree n n n . Let σ \sigma σ be a generator of the Galois group Gal ( L / K ) \operatorname{Gal}(\mathbb{L}/\mathbb{K}) Gal ( L / K ) , and consider the skew polynomial ring
R = L [ X ; σ ] . R = \mathbb{L}[X;\sigma]. R = L [ X ; σ ] .
The elements of R R R are polynomials in the indeterminate X X X with coefficients in L \mathbb{L} L written on the left of each monomial X i X^i X i . Multiplication in R R R is defined by the rule X α = σ ( α ) X X\alpha= \sigma(\alpha)X X α = σ ( α ) X , for all α ∈ L \alpha\in\mathbb{L} α ∈ L , and extended distributively. The ring R R R is noncommutative unless n = 1 n = 1 n = 1 , and its center is given by Z ( R ) = K [ X n ] Z(R) = \mathbb{K}[X^n] Z ( R ) = K [ X n ] . There is a well-defined notion of degree for nonzero elements of R R R , denoted deg ( f ) \deg(f) deg ( f ) , which behaves analogously to the usual polynomial degree. Moreover, R R R is a left Euclidean domain . That is, for any nonzero elements f , g ∈ R f,g \in R f , g ∈ R , there exist unique q , r ∈ R q,r \in R q , r ∈ R such that
f = q g + r , deg ( r ) < deg ( g ) or r = 0. f = qg + r,\qquad\deg(r) < \deg(g)\ \text{or}\ r = 0. f = q g + r , deg ( r ) < deg ( g ) or r = 0.
Whenever r = 0 r = 0 r = 0 , we say that g g g right-divides f f f , and we write g ∣ r f g \mid_{\mathrm{r}} f g ∣ r f ; equivalently, f f f is called a left multiple of g g g . As a consequence, R R R is a left principal ideal domain . This allows one to define the greatest common right divisor and the least common left multiple of two nonzero elements f 1 , f 2 ∈ R f_1,f_2 \in R f 1 , f 2 ∈ R , denoted respectively by gcrd ( f 1 , f 2 ) \operatorname{gcrd}(f_1,f_2) gcrd ( f 1 , f 2 ) and lclm ( f 1 , f 2 ) \operatorname{lclm}(f_1,f_2) lclm ( f 1 , f 2 ) . If a = gcrd ( f 1 , f 2 ) a = \operatorname{gcrd}(f_1,f_2) a = gcrd ( f 1 , f 2 ) and b = lclm ( f 1 , f 2 ) b = \operatorname{lclm}(f_1,f_2) b = lclm ( f 1 , f 2 ) , then R f 1 + R f 2 = R a Rf_1 + Rf_2 = Ra R f 1 + R f 2 = R a and R f 1 ∩ R f 2 = R b Rf_1 \cap Rf_2 = Rb R f 1 ∩ R f 2 = R b , where R f i Rf_i R f i denotes the left ideal generated by f i f_i f i . The reader is referred to [9 , 11 , 16 ] for further details on the algebraic structure of skew polynomial rings. We shall obtain finite-dimensional central simple algebras by central reduction. Since Z ( R ) = K [ X n ] Z(R) = \mathbb{K}[X^n] Z ( R ) = K [ X n ] , each irreducible polynomial F ( Y ) ∈ K [ Y ] F(Y) \in\mathbb{K}[Y] F ( Y ) ∈ K [ Y ] defines a maximal ideal of Z ( R ) Z(R) Z ( R ) via Y ↦ X n Y \mapsto X^n Y ↦ X n . The corresponding quotient of R R R is then finite-dimensional over the residue field
K [ Y ] K [ Y ] F ( Y ) . \frac{\mathbb{K}[Y]}{\mathbb{K}[Y]F(Y)}. K [ Y ] F ( Y ) K [ Y ] .
Precisely, let F ( Y ) ≠ Y F(Y) \ne Y F ( Y ) = Y be an irreducible polynomial of K [ Y ] \mathbb{K}[Y] K [ Y ] with degree s s s . Then F ( X n ) ∈ Z ( R ) = K [ X n ] F(X^n) \in Z(R) = \mathbb{K}[X^n] F ( X n ) ∈ Z ( R ) = K [ X n ] and R F ( X n ) RF(X^n) R F ( X n ) is a two-sided ideal of R R R . We consider the quotient ring
R F : = R R F ( X n ) = { a 0 + a 1 X + ⋯ + a n s − 1 X n s − 1 + R F ( X n ) : a 0 , … , a n s − 1 ∈ L } . R_F := \frac{R}{RF(X^n)} = \left\{a_0 + a_1X + \cdots+ a_{ns-1}X^{ns-1} + RF(X^n): a_0,\ldots,a_{ns-1} \in\mathbb{L}\right\}. R F := R F ( X n ) R = { a 0 + a 1 X + ⋯ + a n s − 1 X n s − 1 + R F ( X n ) : a 0 , … , a n s − 1 ∈ L } .
In the following, elements of the quotient ring R F R_F R F will be denoted by a ‾ \overline{a} a and we often identify a ‾ \overline{a} a with its canonical representative
a ‾ = a + R F ( X n ) , \overline{a}=a+RF(X^n), a = a + R F ( X n ) ,
where a ∈ R a\in R a ∈ R is the unique skew polynomial in the corresponding equivalence class having degree strictly smaller than n s ns n s . With this identification, the degree of a ‾ \overline{a} a is defined by deg ( a ‾ ) : = deg ( a ) \deg(\overline{a}):=\deg(a) deg ( a ) := deg ( a ) . The center of R F R_F R F is denoted by E F \mathbb{E}_F E F , and
E F ≅ K [ Y ] / ( F ( Y ) ) . \mathbb{E}_F\cong\mathbb{K}[Y]/(F(Y)). E F ≅ K [ Y ] / ( F ( Y )) .
Since F F F is irreducible of degree s s s , the center E F \mathbb{E}_F E F is a field with [ E F : K ] = s [\mathbb{E}_F:\mathbb{K}]=s [ E F : K ] = s . Moreover, R F ( X n ) RF(X^n) R F ( X n ) is a maximal two-sided ideal of R R R , and R F R_F R F is a central simple algebra of dimension n 2 n^2 n 2 over E F \mathbb{E}_F E F , hence of dimension n 2 s n^2s n 2 s over K \mathbb{K} K ; see, for example, [7 , 11 , 22 ] . Consequently, the Artin–Wedderburn theorem gives an E F \mathbb{E}_F E F -algebra isomorphism
R R F ( X n ) ≅ M n ( E F ) . (6) \frac{R}{RF(X^n)}\cong M_n(\mathbb{E}_F). \tag*{(6)} R F ( X n ) R ≅ M n ( E F ) . (6)
For an E F \mathbb{E}_F E F -algebra isomorphism
M F : R F ⟶ ∼ M n ( E F ) , \mathcal{M}_F:R_F\overset{\sim}{\longrightarrow}M_n(\mathbb{E}_F), M F : R F ⟶ ∼ M n ( E F ) ,
we can identify any element a ‾ ∈ R F \overline{a}\in R_F a ∈ R F with its image M F ( a ‾ ) \mathcal{M}_F(\overline{a}) M F ( a ) in M n ( F q s ) M_n(\mathbb{F}_{q^s}) M n ( F q s ) , via the isomorphism M F \mathcal{M}_F M F . Note that if
M F ′ : R F ⟶ ∼ M n ( E F ) \mathcal{M}'_F:R_F\overset{\sim}{\longrightarrow}M_n(\mathbb{E}_F) M F ′ : R F ⟶ ∼ M n ( E F )
is another E F \mathbb{E}_F E F -algebra isomorphism then, by Skolem–Noether’s Theorem, there exists N ∈ GL n ( E F ) N\in\operatorname{GL}_n(\mathbb{E}_F) N ∈ GL n ( E F ) such that M F ′ ( a ‾ ) = N M F ( a ‾ ) N − 1 \mathcal{M}'_F(\overline{a})=N\mathcal{M}_F(\overline{a})N^{-1} M F ′ ( a ) = N M F ( a ) N − 1 , for all a ‾ ∈ R F \overline{a}\in R_F a ∈ R F . Therefore,
rk ( M F ′ ( a ‾ ) ) = rk ( M F ( a ‾ ) ) . \operatorname{rk}(\mathcal{M}'_F(\overline{a}))=\operatorname{rk}(\mathcal{M}_F(\overline{a})). rk ( M F ′ ( a )) = rk ( M F ( a )) .
Throughout, we often implicitly identify an element a ‾ ∈ R F \overline{a}\in R_F a ∈ R F with its corresponding matrix in M n ( E F ) M_n(\mathbb{E}_F) M n ( E F ) . Accordingly, we refer to rk ( a ‾ ) \operatorname{rk}(\overline{a}) rk ( a ) to denote the rank of M F ( a ‾ ) \mathcal{M}_F(\overline{a}) M F ( a ) over E F \mathbb{E}_F E F . Indeed, as observed above, this rank is independent of the choice of E F \mathbb{E}_F E F -algebra isomorphism M F \mathcal{M}_F M F . The following result determines the rank of a polynomial in the matrix ring defined by F ( Y ) F(Y) F ( Y ) .
Theorem 2.1 (see [22 Proposition 7] ). For a nonzero a ‾ = a + R F ( X n ) ∈ R F \overline{a}=a+RF(X^n)\in R_F a = a + R F ( X n ) ∈ R F , we have
rk ( a ‾ ) = 1 s ( deg ( F ( X n ) ) − deg ( gcrd ( a , F ( X n ) ) ) ) . \operatorname{rk}(\overline{a})=\frac{1}{s}\left(\deg(F(X^n))-\deg\bigl(\operatorname{gcrd}(a,F(X^n))\bigr)\right). rk ( a ) = s 1 ( deg ( F ( X n )) − deg ( gcrd ( a , F ( X n )) ) ) .
Central simple algebras with unitary involution
In this section, we endow the central simple algebra R F R_F R F with an involution of the second kind. This yields an intrinsic description of the Hermitian elements of R F R_F R F . Through the corresponding matrix representation, we get an explicit representation of the associated Hermitian matrices.
Consider a monic irreducible polynomial
F ( Y ) = F 0 + F 1 Y + ⋯ + F s − 1 Y s − 1 + Y s ∈ K [ Y ] . F(Y)=F_0+F_1Y+\cdots+F_{s-1}Y^{s-1}+Y^s\in\mathbb{K}[Y]. F ( Y ) = F 0 + F 1 Y + ⋯ + F s − 1 Y s − 1 + Y s ∈ K [ Y ] .
of degree s s s with nonzero constant coefficient F 0 ≠ 0 F_0\ne0 F 0 = 0 . Since the constant term of F ( Y ) F(Y) F ( Y ) is nonzero, the skew polynomials F ( X n ) F(X^n) F ( X n ) and X X X are right coprime in R R R , that is gcrd ( F ( X n ) , X ) = 1 \operatorname{gcrd}(F(X^n),X)=1 gcrd ( F ( X n ) , X ) = 1 . Consequently, by the Bezout identity there exist u , v ∈ R u,v\in R u , v ∈ R such that
1 = u X + v F ( X n ) . (7) 1=uX+vF(X^n). \tag*{(7)} 1 = u X + v F ( X n ) . (7)
Recall that, in view of the algebra isomorphism in (6) , the left and right inverses of any invertible element h + R F ( X n ) h+RF(X^n) h + R F ( X n ) of R F R_F R F coincide, because h + R F ( X n ) h+RF(X^n) h + R F ( X n ) corresponds to an invertible matrix under this isomorphism. Therefore, from (7) we derive that the coset X + R F ( X n ) X+RF(X^n) X + R F ( X n ) is invertible in the ring R F R_F R F , with inverse given by u + R F ( X n ) u+RF(X^n) u + R F ( X n ) . For convenience, we define the notation
X − 1 + R F ( X n ) : = u + R F ( X n ) . X^{-1}+RF(X^n):=u+RF(X^n). X − 1 + R F ( X n ) := u + R F ( X n ) .
In particular, every power X i + R F ( X n ) X^i+RF(X^n) X i + R F ( X n ) with i ≥ 1 i\ge1 i ≥ 1 is invertible in R F R_F R F . In what follows, we denote the inverse of X i + R F ( X n ) X^i+RF(X^n) X i + R F ( X n ) by
X − i + R F ( X n ) . X^{-i}+RF(X^n). X − i + R F ( X n ) .
Let F ^ ( Y ) \widehat{F}(Y) F ( Y ) denote the monic reciprocal polynomial of F ( Y ) F(Y) F ( Y ) , defined by
F ^ ( Y ) = F 0 − 1 Y s F ( 1 Y ) = F 0 − 1 ( 1 + F s − 1 Y + ⋯ + F 1 Y s − 1 + F 0 Y s ) . \widehat{F}(Y)=F_0^{-1}Y^sF\left(\frac{1}{Y}\right)=F_0^{-1}(1+F_{s-1}Y+\cdots+F_1Y^{s-1}+F_0Y^s). F ( Y ) = F 0 − 1 Y s F ( Y 1 ) = F 0 − 1 ( 1 + F s − 1 Y + ⋯ + F 1 Y s − 1 + F 0 Y s ) .
It is well known that F ( Y ) F(Y) F ( Y ) is irreducible over F q \mathbb{F}_q F q if and only if its reciprocal polynomial F ^ ( Y ) \widehat{F}(Y) F ( Y ) is irreducible.
Theorem 3.1 (see [8 ] ). Define the map
Θ : ∑ i = 0 n s − 1 a i X i + R F ( X n ) ⟼ ∑ i = 0 n s − 1 σ − i ( a i ) X − i + R F ^ ( X n ) . (8) \Theta:\sum_{i=0}^{ns-1}a_iX^i+RF(X^n)\longmapsto\sum_{i=0}^{ns-1}\sigma^{-i}(a_i)X^{-i}+R\widehat{F}(X^n). \tag*{(8)} Θ : i = 0 ∑ n s − 1 a i X i + R F ( X n ) ⟼ i = 0 ∑ n s − 1 σ − i ( a i ) X − i + R F ( X n ) . (8) Then Θ \Theta Θ is a ring anti-isomorphism from R F R_F R F onto R F ^ R_{\widehat{F}} R F . Moreover, its inverse is given by
∑ i = 0 n s − 1 a i X i + R F ^ ( X n ) ⟼ ∑ i = 0 n s − 1 σ − i ( a i ) X − i + R F ( X n ) . (9) \sum_{i=0}^{ns-1}a_iX^i+R\widehat{F}(X^n)\longmapsto\sum_{i=0}^{ns-1}\sigma^{-i}(a_i)X^{-i}+RF(X^n). \tag*{(9)} i = 0 ∑ n s − 1 a i X i + R F ( X n ) ⟼ i = 0 ∑ n s − 1 σ − i ( a i ) X − i + R F ( X n ) . (9)
From now on we restrict to the case in which F F F is self-reciprocal , meaning that it coincides with its reciprocal polynomial:
Since F F F is irreducible, this implies that either deg ( F ) = 1 \deg(F)=1 deg ( F ) = 1 or deg ( F ) \deg(F) deg ( F ) is even. Note that for every even integer s ≥ 2 s\ge2 s ≥ 2 there exist monic irreducible self-reciprocal polynomials over F q \mathbb{F}_q F q of degree s s s . More precisely, write
with e ≥ 0 e\ge0 e ≥ 0 and r r r odd. Carlitz in [2 Theorems 2–3] proved that, if N q ( s ) N_q(s) N q ( s ) denotes the number of monic irreducible self-reciprocal polynomials of degree s s s over F q \mathbb{F}_q F q , then
N q ( s ) = { 2 − e − 1 ( q 2 e − 1 ) , if q is odd and r = 1 , 2 − e − 1 π ( r , q 2 e ) , otherwise . (10) N_q(s)=
\begin{cases}
2^{-e-1}(q^{2^e}-1), & \text{if }q\text{ is odd and }r=1,\\
2^{-e-1}\pi(r,q^{2^e}), & \text{otherwise}.
\end{cases} \tag*{(10)} N q ( s ) = { 2 − e − 1 ( q 2 e − 1 ) , 2 − e − 1 π ( r , q 2 e ) , if q is odd and r = 1 , otherwise . (10)
where π ( r , Q ) \pi(r,Q) π ( r , Q ) denotes the number of monic irreducible polynomials of degree r r r over F Q \mathbb{F}_Q F Q . In particular, N q ( s ) > 0 N_q(s)>0 N q ( s ) > 0 for every even s ≥ 2 s\ge2 s ≥ 2 . Henceforth, we assume that s ≥ 2 s\ge2 s ≥ 2 .
In this case, the anti-isomorphism constructed in Theorem 3.1 specializes to an anti-automorphism
Θ : R F ⟶ R F , Θ ( ∑ i = 0 n s − 1 a i X i + R F ( X n ) ) = ∑ i = 0 n s − 1 σ − i ( a i ) X − i + R F ( X n ) . \Theta:R_F\longrightarrow R_F,\qquad
\Theta\left(\sum_{i=0}^{ns-1}a_iX^i+RF(X^n)\right)=\sum_{i=0}^{ns-1}\sigma^{-i}(a_i)X^{-i}+RF(X^n). Θ : R F ⟶ R F , Θ ( i = 0 ∑ n s − 1 a i X i + R F ( X n ) ) = i = 0 ∑ n s − 1 σ − i ( a i ) X − i + R F ( X n ) .
In this situation the map Θ \Theta Θ becomes an anti-automorphism of the ring R F R_F R F and since Θ − 1 = Θ \Theta^{-1}=\Theta Θ − 1 = Θ , this is an involution of the ring R F R_F R F .
Let E F = Z ( R F ) \mathbb{E}_F=Z(R_F) E F = Z ( R F ) , and denote by
ϱ : = Θ ∣ E F ∈ Aut F q ( E F ) \varrho:=\left.\Theta\right|_{\mathbb{E}_F}\in\operatorname{Aut}_{\mathbb{F}_q}(\mathbb{E}_F) ϱ := Θ ∣ E F ∈ Aut F q ( E F )
the automorphism induced by Θ \Theta Θ on the center.
Proposition 3.2. The automorphism ϱ \varrho ϱ is an involution in Aut ( E F ) \operatorname{Aut}(\mathbb{E}_F) Aut ( E F ) . Moreover, its fixed field is
E F ϱ = F q ( X − n + X n ) = { z ( X − n + X n ) + R F ( X n ) : z ( Y ) ∈ K [ Y ] } , \mathbb{E}_F^\varrho=\mathbb{F}_q(X^{-n}+X^n)=\{z(X^{-n}+X^n)+RF(X^n):z(Y)\in\mathbb{K}[Y]\}, E F ϱ = F q ( X − n + X n ) = { z ( X − n + X n ) + R F ( X n ) : z ( Y ) ∈ K [ Y ]} , and [ E F : E F ϱ ] = 2 [\mathbb{E}_F:\mathbb{E}_F^\varrho]=2 [ E F : E F ϱ ] = 2 .
Proof. Since E F \mathbb{E}_F E F is commutative and Θ 2 = id \Theta^2=\operatorname{id} Θ 2 = id , the map ϱ \varrho ϱ is an F q \mathbb{F}_q F q -automorphism of order dividing 2. By definition of Θ \Theta Θ ,
Θ ( X n + R F ( X n ) ) = X − n + R F ( X n ) . \Theta(X^n+RF(X^n))=X^{-n}+RF(X^n). Θ ( X n + R F ( X n )) = X − n + R F ( X n ) .
It follows immediately that ϱ \varrho ϱ fixes X − n + X n X^{-n}+X^n X − n + X n , hence
F q ( X n + X − n ) ⊆ E F ϱ . \mathbb{F}_q(X^n+X^{-n})\subseteq\mathbb{E}_F^\varrho. F q ( X n + X − n ) ⊆ E F ϱ .
Let
u : = X n + R F ( X n ) ∈ E F . u:=X^n+RF(X^n)\in\mathbb{E}_F. u := X n + R F ( X n ) ∈ E F .
The element u u u is not fixed by ϱ \varrho ϱ . Indeed, if u = u − 1 u=u^{-1} u = u − 1 , then u 2 = 1 u^2=1 u 2 = 1 ; since E F \mathbb{E}_F E F is a field, this would force u = ± 1 u=\pm1 u = ± 1 (and u = 1 u=1 u = 1 in characteristic 2), contradicting the fact that the minimal polynomial F F F of u u u has degree s ≥ 2 s\geq2 s ≥ 2 . Thus ϱ \varrho ϱ is a nontrivial involution and
[ E F : E F ϱ ] = 2. [\mathbb{E}_F:\mathbb{E}_F^\varrho]=2. [ E F : E F ϱ ] = 2.
Moreover, u u u satisfies
Y 2 − ( u + u − 1 ) Y + 1 = 0 Y^2-(u+u^{-1})Y+1=0 Y 2 − ( u + u − 1 ) Y + 1 = 0
over F q ( u + u − 1 ) \mathbb{F}_q(u+u^{-1}) F q ( u + u − 1 ) . Since u ∉ F q ( u + u − 1 ) u\notin\mathbb{F}_q(u+u^{-1}) u ∈ / F q ( u + u − 1 ) , this extension also has degree 2. Together with F q ( u + u − 1 ) ⊆ E F ϱ \mathbb{F}_q(u+u^{-1})\subseteq\mathbb{E}_F^\varrho F q ( u + u − 1 ) ⊆ E F ϱ , this gives
E F ϱ = F q ( u + u − 1 ) = F q ( X n + X − n ) , \mathbb{E}_F^\varrho=\mathbb{F}_q(u+u^{-1})=\mathbb{F}_q(X^n+X^{-n}), E F ϱ = F q ( u + u − 1 ) = F q ( X n + X − n ) ,
as claimed.
The involution Θ \Theta Θ is ϱ \varrho ϱ -semilinear over the center: for every λ ∈ E F \lambda\in\mathbb{E}_F λ ∈ E F and every a ˉ ∈ R F \bar a\in R_F a ˉ ∈ R F ,
Θ ( λ a ˉ ) = Θ ( a ˉ ) ϱ ( λ ) = ϱ ( λ ) Θ ( a ˉ ) . \Theta(\lambda\bar a)=\Theta(\bar a)\varrho(\lambda)=\varrho(\lambda)\Theta(\bar a). Θ ( λ a ˉ ) = Θ ( a ˉ ) ϱ ( λ ) = ϱ ( λ ) Θ ( a ˉ ) .
Thus, once an E F \mathbb{E}_F E F -algebra isomorphism M F : R F ⟶ ∼ M n ( E F ) \mathcal{M}_F:R_F\overset{\sim}{\longrightarrow}M_n(\mathbb{E}_F) M F : R F ⟶ ∼ M n ( E F ) is fixed, the anti-automorphism Θ \Theta Θ induces an anti-automorphism of M n ( E F ) M_n(\mathbb{E}_F) M n ( E F ) which is semilinear with respect to ϱ \varrho ϱ .
Theorem 3.3. There exists N ∈ GL n ( E F ) N\in\operatorname{GL}_n(\mathbb{E}_F) N ∈ GL n ( E F ) such that
M F ( a ˉ ) T = N M F ( Θ ( a ˉ ) ) ϱ N − 1 , (11) \mathcal{M}_F(\bar a)^{\mathsf{T}}=N\mathcal{M}_F(\Theta(\bar a))^\varrho N^{-1}, \tag*{(11)} M F ( a ˉ ) T = N M F ( Θ ( a ˉ ) ) ϱ N − 1 , (11)
for all a ˉ ∈ R F \bar a\in R_F a ˉ ∈ R F .
Proof. By Theorem 3.1 , specialized to the case F = F ^ F=\widehat{F} F = F , the map Θ \Theta Θ is an anti-automorphism of R F R_F R F . Therefore
Φ : = M F Θ M F − 1 : M n ( E F ) ⟶ M n ( E F ) \Phi:=\mathcal{M}_F\Theta\mathcal{M}_F^{-1}:M_n(\mathbb{E}_F)\longrightarrow M_n(\mathbb{E}_F) Φ := M F Θ M F − 1 : M n ( E F ) ⟶ M n ( E F )
is an anti-automorphism of M n ( E F ) M_n(\mathbb{E}_F) M n ( E F ) . Since Θ \Theta Θ restricts to ϱ \varrho ϱ on the center E F \mathbb{E}_F E F , the map Φ \Phi Φ is ϱ \varrho ϱ -semilinear. By the Skolem–Noether theorem, there exist τ ∈ Aut ( E F ) \tau\in\operatorname{Aut}(\mathbb{E}_F) τ ∈ Aut ( E F ) and N ∈ GL n ( E F ) N\in\operatorname{GL}_n(\mathbb{E}_F) N ∈ GL n ( E F ) such that
M F ( a ˉ ) T = N M F ( Θ ( a ˉ ) ) τ N − 1 , (12) \mathcal{M}_F(\bar{a})^{\mathsf{T}}=N\mathcal{M}_F(\Theta(\bar{a}))^\tau N^{-1}, \tag*{(12)} M F ( a ˉ ) T = N M F ( Θ ( a ˉ ) ) τ N − 1 , (12)
for all a ˉ ∈ R F \bar{a}\in R_F a ˉ ∈ R F .
It remains to prove that τ = ϱ \tau=\varrho τ = ϱ . Let λ ∈ E F \lambda\in\mathbb{E}_F λ ∈ E F . Since M F ( λ ) = λ I n \mathcal{M}_F(\lambda)=\lambda I_n M F ( λ ) = λ I n and Θ ( λ ) = ϱ ( λ ) \Theta(\lambda)=\varrho(\lambda) Θ ( λ ) = ϱ ( λ ) , from (12) we obtain
λ I n = N ( τ ( ϱ ( λ ) ) I n ) N − 1 = τ ( ϱ ( λ ) ) I n . \lambda I_n=N\bigl(\tau(\varrho(\lambda))I_n\bigr)N^{-1}=\tau(\varrho(\lambda))I_n. λ I n = N ( τ ( ϱ ( λ )) I n ) N − 1 = τ ( ϱ ( λ )) I n .
Thus λ = τ ( ϱ ( λ ) ) \lambda=\tau(\varrho(\lambda)) λ = τ ( ϱ ( λ )) for every λ ∈ E F \lambda\in\mathbb{E}_F λ ∈ E F , namely τ ∘ ϱ = id E F \tau\circ\varrho=\operatorname{id}_{\mathbb{E}_F} τ ∘ ϱ = id E F . Since ϱ 2 = id \varrho^2=\operatorname{id} ϱ 2 = id , we conclude that τ = ϱ \tau=\varrho τ = ϱ . This proves (11) . □
The preceding theorem shows that the transpose on M n ( E F ) M_n(\mathbb{E}_F) M n ( E F ) corresponds, up to inner conjugation, to the involution a ˉ ↦ Θ ( a ˉ ) \bar{a}\mapsto\Theta(\bar{a}) a ˉ ↦ Θ ( a ˉ ) together with the Galois action ϱ \varrho ϱ on the coefficients. In particular, the pair
( R F , Θ ) (R_F,\Theta) ( R F , Θ )
is a central simple E F \mathbb{E}_F E F -algebra endowed with a unitary involution , i.e. an involution of the second kind; see, for example, [12 p. 20] .
We next show that, after replacing M F \mathcal{M}_F M F by a suitable conjugate representation, one may normalize the matrix N N N away.
Theorem 3.4. Let N ∈ GL n ( E F ) N\in\operatorname{GL}_n(\mathbb{E}_F) N ∈ GL n ( E F ) be as in Theorem 3.3. Then there exists λ ∈ E F × \lambda\in\mathbb{E}_F^\times λ ∈ E F × satisfying
N E F / E ( λ ) = 1 \operatorname{N}_{\mathbb{E}_F/E}(\lambda)=1 N E F / E ( λ ) = 1 and
N T = λ N ϱ . (13) N^{\mathsf{T}}=\lambda N^\varrho. \tag*{(13)} N T = λ N ϱ . (13) As a consequence, there exists an E F \mathbb{E}_F E F -algebra isomorphism
M F ′ : R F → ∼ M n ( E F ) \mathcal{M}'_F:R_F\xrightarrow{\sim}M_n(\mathbb{E}_F) M F ′ : R F ∼ M n ( E F ) such that
M F ′ ( a ˉ ) T = M F ′ ( Θ ( a ˉ ) ) ϱ , (14) \mathcal{M}'_F(\bar{a})^{\mathsf{T}}=\mathcal{M}'_F(\Theta(\bar{a}))^\varrho, \tag*{(14)} M F ′ ( a ˉ ) T = M F ′ ( Θ ( a ˉ ) ) ϱ , (14) for all $\bar{a}\in R_F.
Proof. Set A = M F ( a ˉ ) A=\mathcal{M}_F(\bar{a}) A = M F ( a ˉ ) and B = M F ( Θ ( a ˉ ) ) B=\mathcal{M}_F(\Theta(\bar{a})) B = M F ( Θ ( a ˉ )) . Then (11) reads A T = N B ϱ N − 1 A^{\mathsf{T}}=NB^\varrho N^{-1} A T = N B ϱ N − 1 . Replacing a ˉ \bar{a} a ˉ by Θ ( a ˉ ) \Theta(\bar{a}) Θ ( a ˉ ) and using Θ 2 = id \Theta^2=\operatorname{id} Θ 2 = id , we get
B T = N A ϱ N − 1 . B^{\mathsf{T}}=NA^\varrho N^{-1}. B T = N A ϱ N − 1 .
Taking transpose in the latter identity and then applying ϱ \varrho ϱ , we obtain
B ϱ = ( ( N − 1 ) T ) ϱ A T ( N T ) ϱ . B^\varrho=((N^{-1})^{\mathsf{T}})^\varrho A^{\mathsf{T}}(N^{\mathsf{T}})^\varrho. B ϱ = (( N − 1 ) T ) ϱ A T ( N T ) ϱ .
Therefore, by substituting A T = N B ϱ N − 1 A^{\mathsf{T}}=NB^{\varrho}N^{-1} A T = N B ϱ N − 1 , we get
B ϱ = ( ( N − 1 ) T ) ϱ N B ϱ N − 1 ( N T ) ϱ , B^{\varrho}=\left((N^{-1})^{\mathsf{T}}\right)^{\varrho}NB^{\varrho}N^{-1}(N^{\mathsf{T}})^{\varrho}, B ϱ = ( ( N − 1 ) T ) ϱ N B ϱ N − 1 ( N T ) ϱ ,
and so
B = ( N − 1 ) T N ϱ B ( N − 1 ) ϱ N T . B=(N^{-1})^{\mathsf{T}}N^{\varrho}B(N^{-1})^{\varrho}N^{\mathsf{T}}. B = ( N − 1 ) T N ϱ B ( N − 1 ) ϱ N T .
This implies that
( N − 1 ) ϱ N T B = B ( N − 1 ) ϱ N T , (N^{-1})^{\varrho}N^{\mathsf{T}}B=B(N^{-1})^{\varrho}N^{\mathsf{T}}, ( N − 1 ) ϱ N T B = B ( N − 1 ) ϱ N T ,
for all B ∈ M n ( E F ) B\in M_n(E_F) B ∈ M n ( E F ) . Since the center of M n ( E F ) M_n(E_F) M n ( E F ) is E F I n E_F I_n E F I n , there exists λ ∈ E F × \lambda\in E_F^{\times} λ ∈ E F × such that (13) holds. Transposing that identity gives
N = λ ( N ϱ ) T , N=\lambda(N^{\varrho})^{\mathsf{T}}, N = λ ( N ϱ ) T ,
whereas applying ϱ \varrho ϱ gives
( N ϱ ) T = ϱ ( λ ) N . (N^{\varrho})^{\mathsf{T}}=\varrho(\lambda)N. ( N ϱ ) T = ϱ ( λ ) N .
Consequently, N = λ ϱ ( λ ) N N=\lambda\varrho(\lambda)N N = λ ϱ ( λ ) N , and hence
N E F / E F ϱ ( λ ) = λ ϱ ( λ ) = 1. N_{E_F/E_F^{\varrho}}(\lambda)=\lambda\varrho(\lambda)=1. N E F / E F ϱ ( λ ) = λ ϱ ( λ ) = 1.
By Hilbert’s Theorem 90, there exists μ ∈ E F × \mu\in E_F^{\times} μ ∈ E F × such that
λ = μ − 1 ϱ ( μ ) . \lambda=\mu^{-1}\varrho(\mu). λ = μ − 1 ϱ ( μ ) .
Set H : = μ N H:=\mu N H := μ N . Then, using N T = λ N ϱ N^{\mathsf{T}}=\lambda N^{\varrho} N T = λ N ϱ , we obtain
H T = μ N T = μ λ N ϱ = ϱ ( μ ) N ϱ = H ϱ . H^{\mathsf{T}}=\mu N^{\mathsf{T}}=\mu\lambda N^{\varrho}=\varrho(\mu)N^{\varrho}=H^{\varrho}. H T = μ N T = μ λ N ϱ = ϱ ( μ ) N ϱ = H ϱ .
Thus H H H is a nondegenerate ϱ \varrho ϱ -Hermitian matrix. Over the finite field extension E F / E F ϱ E_F/E_F^{\varrho} E F / E F ϱ , every nondegenerate ϱ \varrho ϱ -Hermitian form is equivalent to the standard Hermitian form, see, e.g., [23 Theorem 7.1] . Therefore, there exists P ∈ GL n ( E F ) P\in\operatorname{GL}_n(E_F) P ∈ GL n ( E F ) such that
( P − 1 ) T H ( P − 1 ) ϱ = I n . (15) (P^{-1})^{\mathsf{T}}H(P^{-1})^{\varrho}=I_n. \tag*{(15)} ( P − 1 ) T H ( P − 1 ) ϱ = I n . (15)
Now define
M F ′ ( a ‾ ) : = P M F ( a ‾ ) P − 1 , a ‾ ∈ R F . \mathcal{M}'_F(\overline{a}):=P\mathcal{M}_F(\overline{a})P^{-1},\qquad\overline{a}\in R_F. M F ′ ( a ) := P M F ( a ) P − 1 , a ∈ R F .
Then M F ′ \mathcal{M}'_F M F ′ is again an E F E_F E F -algebra isomorphism. Using (11) , we compute
M F ′ ( a ‾ ) T = ( P − 1 ) T A T P T = ( P − 1 ) T N B ϱ N − 1 P T . \mathcal{M}'_F(\overline{a})^{\mathsf{T}}=(P^{-1})^{\mathsf{T}}A^{\mathsf{T}}P^{\mathsf{T}}=(P^{-1})^{\mathsf{T}}NB^{\varrho}N^{-1}P^{\mathsf{T}}. M F ′ ( a ) T = ( P − 1 ) T A T P T = ( P − 1 ) T N B ϱ N − 1 P T .
Since H = μ N H=\mu N H = μ N , identity (15) implies
( P − 1 ) T N B ϱ N − 1 P T = P ϱ B ϱ ( P − 1 ) ϱ = ( P B P − 1 ) ϱ = M F ′ ( Θ ( a ‾ ) ) ϱ . (P^{-1})^{\mathsf{T}}NB^{\varrho}N^{-1}P^{\mathsf{T}}=P^{\varrho}B^{\varrho}(P^{-1})^{\varrho}=(PBP^{-1})^{\varrho}=\mathcal{M}'_F(\Theta(\overline{a}))^{\varrho}. ( P − 1 ) T N B ϱ N − 1 P T = P ϱ B ϱ ( P − 1 ) ϱ = ( P B P − 1 ) ϱ = M F ′ ( Θ ( a ) ) ϱ .
This proves (14) .
Equation (14) provides the intrinsic expression of the compatibility between transpose and the involution Θ \Theta Θ . After a suitable normalization, the anti-automorphism Θ \Theta Θ becomes precisely the adjoint involution defined by the standard ϱ \varrho ϱ -Hermitian form on E F n E_F^n E F n . It follows that, whenever ϱ ≠ id \varrho\ne\operatorname{id} ϱ = id , the algebra with involution ( R F , Θ ) (R_F,\Theta) ( R F , Θ ) is isomorphic to
( M n ( E F ) , A ↦ A ∗ ) , A ∗ = ( A ϱ ) T . \left(M_n(E_F),\,A\mapsto A^*\right),\qquad A^*=(A^{\varrho})^{\mathsf{T}}. ( M n ( E F ) , A ↦ A ∗ ) , A ∗ = ( A ϱ ) T .
In the rest of the paper, in view of Theorem 3.4 , we fix an E F E_F E F -algebra isomorphism
M F : R F → ∼ M n ( E F ) \mathcal{M}_F:R_F\xrightarrow{\sim}M_n(E_F) M F : R F ∼ M n ( E F )
such that
M F ( Θ ( a ‾ ) ) = M F ( a ‾ ) ∗ (16) \mathcal{M}_F(\Theta(\overline{a}))=\mathcal{M}_F(\overline{a})^* \tag*{(16)} M F ( Θ ( a )) = M F ( a ) ∗ (16)
for all a ‾ ∈ R F \overline{a}\in R_F a ∈ R F . The relation [16 ] immediately identifies Hermitian matrices with the fixed points of Θ \Theta Θ .
Proposition 3.5. For a ‾ ∈ R F \overline{a}\in R_F a ∈ R F , the matrix M F ( a ‾ ) \mathcal{M}_F(\overline{a}) M F ( a ) is Hermitian if and only if
Θ ( a ‾ ) = a ‾ . \Theta(\overline{a})=\overline{a}. Θ ( a ) = a . Equivalently,
M F ( Fix ( Θ ) ) = { H ∈ M n ( E F ) : H ∗ = H } = H n ( E F / E F 0 ) , \mathcal{M}_F(\operatorname{Fix}(\Theta))=\{H\in M_n(\mathbb{E}_F):H^*=H\}=\mathcal{H}_n(\mathbb{E}_F/\mathbb{E}_F^0), M F ( Fix ( Θ )) = { H ∈ M n ( E F ) : H ∗ = H } = H n ( E F / E F 0 ) , where
Fix ( Θ ) : = { a ‾ ∈ R F : Θ ( a ‾ ) = a ‾ } . \operatorname{Fix}(\Theta):=\{\overline{a}\in R_F:\Theta(\overline{a})=\overline{a}\}. Fix ( Θ ) := { a ∈ R F : Θ ( a ) = a } .
We now determine Fix ( Θ ) \operatorname{Fix}(\Theta) Fix ( Θ ) explicitly inside R F R_F R F . Recall that s s s is the degree of the monic irreducible self-reciprocal polynomial F F F . Thus s s s is even. Set
m : = n s 2 , m:=\frac{ns}{2}, m := 2 n s ,
and write R F R_F R F using symmetric powers of X X X . Since X + R F ( X n ) X+RF(X^n) X + R F ( X n ) is invertible in R F R_F R F , right multiplication by X − m X^{-m} X − m gives an L \mathbb{L} L -linear automorphism of R F R_F R F . Therefore
B : = { X − m , X − m + 1 , … , X − 1 , 1 , X , … , X m − 1 } \mathcal{B}:=\{X^{-m},X^{-m+1},\ldots,X^{-1},1,X,\ldots,X^{m-1}\} B := { X − m , X − m + 1 , … , X − 1 , 1 , X , … , X m − 1 }
defines an L \mathbb{L} L -basis of R F R_F R F . In particular, every element a ‾ ∈ R F \overline{a}\in R_F a ∈ R F can be written uniquely in the form
a ‾ = ∑ i = − m m − 1 α i X i + R F ( X n ) , (17) \overline{a}=\sum_{i=-m}^{m-1}\alpha_iX^i+RF(X^n), \tag*{(17)} a = i = − m ∑ m − 1 α i X i + R F ( X n ) , (17)
with α i ∈ L \alpha_i\in\mathbb{L} α i ∈ L . Recall that
Θ ( ∑ i = − m m − 1 α i X i + R F ( X n ) ) = ∑ i = − m m − 1 σ − i ( α i ) X − i + R F ( X n ) . \Theta\left(\sum_{i=-m}^{m-1}\alpha_iX^i+RF(X^n)\right)=\sum_{i=-m}^{m-1}\sigma^{-i}(\alpha_i)X^{-i}+RF(X^n). Θ ( i = − m ∑ m − 1 α i X i + R F ( X n ) ) = i = − m ∑ m − 1 σ − i ( α i ) X − i + R F ( X n ) .
Hence, after reindexing,
Θ ( a ‾ ) = α 0 + ∑ i = 1 m − 1 σ − i ( α i ) X − i + ∑ i = 1 m σ i ( α − i ) X i + R F ( X n ) . \Theta(\overline{a})=\alpha_0+\sum_{i=1}^{m-1}\sigma^{-i}(\alpha_i)X^{-i}+\sum_{i=1}^{m}\sigma^i(\alpha_{-i})X^i+RF(X^n). Θ ( a ) = α 0 + i = 1 ∑ m − 1 σ − i ( α i ) X − i + i = 1 ∑ m σ i ( α − i ) X i + R F ( X n ) .
This yields an explicit description of the fixed subspace of Θ \Theta Θ .
Theorem 3.6. We have
Fix ( Θ ) = { ∑ i = 1 m − 1 σ − i ( α i ) X − i + α 0 + ∑ i = 1 m − 1 α i X i + R F ( X n ) : α 0 , α 1 , … , α m − 1 ∈ L } . \operatorname{Fix}(\Theta)=\left\{\sum_{i=1}^{m-1}\sigma^{-i}(\alpha_i)X^{-i}+\alpha_0+\sum_{i=1}^{m-1}\alpha_iX^i+RF(X^n):\alpha_0,\alpha_1,\ldots,\alpha_{m-1}\in\mathbb{L}\right\}. Fix ( Θ ) = { i = 1 ∑ m − 1 σ − i ( α i ) X − i + α 0 + i = 1 ∑ m − 1 α i X i + R F ( X n ) : α 0 , α 1 , … , α m − 1 ∈ L } . Proof. Set
D : = { ∑ i = 1 m − 1 σ − i ( α i ) X − i + α 0 + ∑ i = 1 m − 1 α i X i + R F ( X n ) : α 0 , α 1 , … , α m − 1 ∈ L } . D:=\left\{\sum_{i=1}^{m-1}\sigma^{-i}(\alpha_i)X^{-i}+\alpha_0+\sum_{i=1}^{m-1}\alpha_iX^i+RF(X^n):\alpha_0,\alpha_1,\ldots,\alpha_{m-1}\in\mathbb{L}\right\}. D := { i = 1 ∑ m − 1 σ − i ( α i ) X − i + α 0 + i = 1 ∑ m − 1 α i X i + R F ( X n ) : α 0 , α 1 , … , α m − 1 ∈ L } .
We first show that D ‾ ⊆ Fix ( Θ ) \overline{D} \subseteq\operatorname{Fix}(\Theta) D ⊆ Fix ( Θ ) . Let a ‾ = a + R F ( X n ) ∈ D \overline{a}=a+RF(X^n)\in D a = a + R F ( X n ) ∈ D , where
a = ∑ i = 1 m − 1 σ − i ( α i ) X − i + α 0 + ∑ i = 1 m − 1 α i X i a=\sum_{i=1}^{m-1}\sigma^{-i}(\alpha_i)X^{-i}+\alpha_0+\sum_{i=1}^{m-1}\alpha_iX^i a = i = 1 ∑ m − 1 σ − i ( α i ) X − i + α 0 + i = 1 ∑ m − 1 α i X i
for some α 0 , α 1 , … , α m − 1 ∈ L \alpha_0,\alpha_1,\ldots,\alpha_{m-1}\in L α 0 , α 1 , … , α m − 1 ∈ L . Then
Θ ( a ‾ ) = ∑ i = 1 m − 1 Θ ( α i X i ) + Θ ( α 0 ) + ∑ i = 1 m − 1 Θ ( σ − i ( α i ) X − i ) = ∑ i = 1 m − 1 σ − i ( α i ) X − i + α 0 + ∑ i = 1 m − 1 α i X i = a ‾ , \Theta(\overline{a})=\sum_{i=1}^{m-1}\Theta(\alpha_iX^i)+\Theta(\alpha_0)+\sum_{i=1}^{m-1}\Theta\left(\sigma^{-i}(\alpha_i)X^{-i}\right)=\sum_{i=1}^{m-1}\sigma^{-i}(\alpha_i)X^{-i}+\alpha_0+\sum_{i=1}^{m-1}\alpha_iX^i=\overline{a}, Θ ( a ) = i = 1 ∑ m − 1 Θ ( α i X i ) + Θ ( α 0 ) + i = 1 ∑ m − 1 Θ ( σ − i ( α i ) X − i ) = i = 1 ∑ m − 1 σ − i ( α i ) X − i + α 0 + i = 1 ∑ m − 1 α i X i = a ,
and therefore a ‾ ∈ Fix ( Θ ) \overline{a}\in\operatorname{Fix}(\Theta) a ∈ Fix ( Θ ) . A dimension count gives the reverse inclusion. Indeed, by Proposition 3.5 ,
∣ Fix ( Θ ) ∣ = ∣ H n ( E F / E F ϱ ) ∣ = ∣ E F ∣ n 2 2 . |\operatorname{Fix}(\Theta)|=|\mathcal{H}_n(\mathbb{E}_F/\mathbb{E}_F^\varrho)|=|\mathbb{E}_F|^{\frac{n^2}{2}}. ∣ Fix ( Θ ) ∣ = ∣ H n ( E F / E F ϱ ) ∣ = ∣ E F ∣ 2 n 2 .
On the other hand, D D D has size
∣ L ∣ m = ∣ K ∣ n 2 s 2 = ∣ E F ∣ n 2 2 . |L|^m=|K|^{\frac{n^2s}{2}}=|\mathbb{E}_F|^{\frac{n^2}{2}}. ∣ L ∣ m = ∣ K ∣ 2 n 2 s = ∣ E F ∣ 2 n 2 .
Since D ⊆ Fix ( Θ ) D\subseteq\operatorname{Fix}(\Theta) D ⊆ Fix ( Θ ) and the two sets have the same cardinality, the assertion follows. □
New constructions
In this section we provide a new family of maximum additive Hermitian d d d -codes using the central simple algebra with involution ( R F , Θ ) (R_F,\Theta) ( R F , Θ ) studied above. The construction below yields maximum additive Hermitian d d d -codes for all admissible choices of the parameters. We then introduce a scalar-field invariant for additive Hermitian codes, which enables us to distinguish the present family from the classical constructions of Schmidt and shows that these codes are genuinely new.
New constructions
4.1. New constructions. Throughout this section, let L = F q n L=\mathbb{F}_{q^n} L = F q n , K = F q K=\mathbb{F}_q K = F q , let s s s be a positive even integer, and let F ∈ F q [ Y ] F\in\mathbb{F}_q[Y] F ∈ F q [ Y ] be a monic irreducible self-reciprocal polynomial of degree s s s . By R F R_F R F we denote the associated central simple algebra endowed with the involution of the second kind Θ \Theta Θ introduced in § 3 . Via the E F \mathbb{E}_F E F -algebra isomorphism M F : R F ⟶ M n ( E F ) \mathcal{M}_F:R_F\longrightarrow M_n(\mathbb{E}_F) M F : R F ⟶ M n ( E F ) , the Θ \Theta Θ -fixed subspace of R F R_F R F is identified with the space H n ( E F / E F ϱ ) \mathcal{H}_n(\mathbb{E}_F/\mathbb{E}_F^\varrho) H n ( E F / E F ϱ ) of ϱ \varrho ϱ -Hermitian matrices in M n ( E F ) M_n(\mathbb{E}_F) M n ( E F ) . Our aim is to construct explicit additive families attaining the general bound for Hermitian d d d -codes.
For 0 ≤ ℓ ≤ n s / 2 − 1 0\leq\ell\leq ns/2-1 0 ≤ ℓ ≤ n s /2 − 1 , define
S ℓ ( F ) : = { α 0 + ∑ i = 1 ℓ ( σ − i ( α i ) X − i + α i X i ) + R F ( X n ) : α i ∈ F q n } ⊆ R F . (18) S_\ell(F):=\left\{\alpha_0+\sum_{i=1}^{\ell}\left(\sigma^{-i}(\alpha_i)X^{-i}+\alpha_iX^i\right)+RF(X^n):\alpha_i\in\mathbb{F}_{q^n}\right\}\subseteq R_F. \tag*{(18)} S ℓ ( F ) := { α 0 + i = 1 ∑ ℓ ( σ − i ( α i ) X − i + α i X i ) + R F ( X n ) : α i ∈ F q n } ⊆ R F . (18)
Theorem 4.1. For all integers d d d and n n n with 1 ≤ d ≤ n 1\leq d\leq n 1 ≤ d ≤ n , the set S ℓ ( F ) S_\ell(F) S ℓ ( F ) , with
ℓ = s ( n − d + 1 ) − 2 2 , \ell=\frac{s(n-d+1)-2}{2}, ℓ = 2 s ( n − d + 1 ) − 2 , is a maximum additive Hermitian d d d -code in H n ( E F / E F ϱ ) \mathcal{H}_n(\mathbb{E}_F/\mathbb{E}_F^\varrho) H n ( E F / E F ϱ ) .
Proof. By Theorem 3.6 , every element of S ℓ ( F ) S_\ell(F) S ℓ ( F ) is fixed by Θ \Theta Θ , and therefore S ℓ ( F ) ⊆ Fix ( Θ ) S_\ell(F)\subseteq\operatorname{Fix}(\Theta) S ℓ ( F ) ⊆ Fix ( Θ ) . Hence, via the algebra isomorphism M F : R F ⟶ ∼ M n ( E F ) \mathcal{M}_F:R_F\overset{\sim}{\longrightarrow}M_n(\mathbb{E}_F) M F : R F ⟶ ∼ M n ( E F ) , such that (16) holds, the set S ℓ ( F ) S_\ell(F) S ℓ ( F ) is identified with a family of Hermitian matrices, namely with a subset of H n ( E F / E F q ) \mathcal{H}_n(\mathbb{E}_F/\mathbb{E}_F^q) H n ( E F / E F q ) . Clearly, S ℓ ( F ) \mathcal{S}_{\ell}(F) S ℓ ( F ) is an F q \mathbb{F}_q F q -subspace of R F R_F R F , with
∣ S ℓ ( F ) ∣ = ( q n ) s ( n − d + 1 ) 2 = q s n ( n − d + 1 ) 2 . |\mathcal{S}_{\ell}(F)|=(q^n)^{\frac{s(n-d+1)}{2}}=q^{\frac{sn(n-d+1)}{2}}. ∣ S ℓ ( F ) ∣ = ( q n ) 2 s ( n − d + 1 ) = q 2 s n ( n − d + 1 ) .
It remains to prove that S ℓ ( F ) \mathcal{S}_{\ell}(F) S ℓ ( F ) is a d d d -code in H n ( E F / E F q ) \mathcal{H}_n(\mathbb{E}_F/\mathbb{E}_F^q) H n ( E F / E F q ) , that is, every nonzero element has rank at least d d d . Let
a ˉ = α 0 + ∑ i = 1 s ( n − d + 1 ) − 2 2 ( σ − i ( α i ) X − i + α i X i ) + R F ( X n ) ∈ S ℓ ( F ) ∖ { 0 } . \bar{a}=\alpha_0+\sum_{i=1}^{\frac{s(n-d+1)-2}{2}}\left(\sigma^{-i}(\alpha_i)X^{-i}+\alpha_iX^i\right)+RF(X^n)\in\mathcal{S}_{\ell}(F)\setminus\{0\}. a ˉ = α 0 + i = 1 ∑ 2 s ( n − d + 1 ) − 2 ( σ − i ( α i ) X − i + α i X i ) + R F ( X n ) ∈ S ℓ ( F ) ∖ { 0 } .
We claim that rk ( a ˉ ) ≥ d \operatorname{rk}(\bar{a})\ge d rk ( a ˉ ) ≥ d . First note that X ℓ + R F ( X n ) X^\ell+RF(X^n) X ℓ + R F ( X n ) is invertible in R F R_F R F . Hence right multiplication by X ℓ + R F ( X n ) X^\ell+RF(X^n) X ℓ + R F ( X n ) preserves rank, and therefore
rk ( a ˉ ) = rk ( a ˉ X ℓ ) . \operatorname{rk}(\bar{a})=\operatorname{rk}(\bar{a}X^\ell). rk ( a ˉ ) = rk ( a ˉ X ℓ ) .
Moreover, no negative powers of X X X occur in a ˉ X ℓ \bar{a}X^\ell a ˉ X ℓ . Thus a ˉ X ℓ \bar{a}X^\ell a ˉ X ℓ admits a representative b ∈ R b\in R b ∈ R such that
deg ( b ) ≤ ℓ + ℓ = 2 ℓ = s ( n − d + 1 ) − 2 = s ( n − d ) + s − 2. \deg(b)\le\ell+\ell=2\ell=s(n-d+1)-2=s(n-d)+s-2. deg ( b ) ≤ ℓ + ℓ = 2 ℓ = s ( n − d + 1 ) − 2 = s ( n − d ) + s − 2.
Now every common right divisor of F ( X n ) F(X^n) F ( X n ) and b b b is also a right divisor of F ( X n ) F(X^n) F ( X n ) . Since F ( Y ) F(Y) F ( Y ) has degree s s s , it follows from [22 Theorem 2] that every right divisor of F ( X n ) F(X^n) F ( X n ) has degree divisible by s s s . Therefore
deg ( gcrd ( F ( X n ) , b ) ) ≤ s ( n − d ) + s − 2 , \deg(\operatorname{gcrd}(F(X^n),b))\le s(n-d)+s-2, deg ( gcrd ( F ( X n ) , b )) ≤ s ( n − d ) + s − 2 ,
and, because deg ( gcrd ( F ( X n ) , b ) ) \deg(\operatorname{gcrd}(F(X^n),b)) deg ( gcrd ( F ( X n ) , b )) is a multiple of s s s , we in fact obtain
deg ( gcrd ( F ( X n ) , b ) ) ≤ s ( n − d ) . \deg(\operatorname{gcrd}(F(X^n),b))\le s(n-d). deg ( gcrd ( F ( X n ) , b )) ≤ s ( n − d ) .
Applying Theorem 2.1 , we conclude that
rk ( a ˉ ) = rk ( a ˉ X ℓ ) = n − 1 s deg ( gcrd ( F ( X n ) , b ) ) ≥ n − 1 s s ( n − d ) = d . \operatorname{rk}(\bar{a})=\operatorname{rk}(\bar{a}X^\ell)=n-\frac{1}{s}\deg(\operatorname{gcrd}(F(X^n),b))\ge n-\frac{1}{s}s(n-d)=d. rk ( a ˉ ) = rk ( a ˉ X ℓ ) = n − s 1 deg ( gcrd ( F ( X n ) , b )) ≥ n − s 1 s ( n − d ) = d .
This proves that S ℓ ( F ) \mathcal{S}_{\ell}(F) S ℓ ( F ) is an additive Hermitian d d d -code in H n ( E F / E F q ) \mathcal{H}_n(\mathbb{E}_F/\mathbb{E}_F^q) H n ( E F / E F q ) . Finally, by (5) , every additive Hermitian d d d -code in H n ( E F / E F q ) \mathcal{H}_n(\mathbb{E}_F/\mathbb{E}_F^q) H n ( E F / E F q ) has cardinality at most
q s n ( n − d + 1 ) 2 . q^{\frac{sn(n-d+1)}{2}}. q 2 s n ( n − d + 1 ) .
Since S ℓ ( F ) \mathcal{S}_{\ell}(F) S ℓ ( F ) attains this bound, it is maximum. □
By Proposition 3.2 , the extension E F / E F q \mathbb{E}_F/\mathbb{E}_F^q E F / E F q is quadratic; equivalently, E F q ≅ F q s / 2 \mathbb{E}_F^q\cong\mathbb{F}_{q^{s/2}} E F q ≅ F q s /2 . Thus, for every even value of s s s and every self-reciprocal irreducible polynomial F F F as above, the construction takes place over the Hermitian space associated with the quadratic extension
E F / E F q ≅ F q s / F q s / 2 . \mathbb{E}_F/\mathbb{E}_F^q\cong\mathbb{F}_{q^s}/\mathbb{F}_{q^{s/2}}. E F / E F q ≅ F q s / F q s /2 .
Consequently, Theorem 4.1 gives, for every length n n n and every minimum distance d d d , an explicit maximum additive Hermitian d d d -code of size
∣ E F ∣ n ( n − d + 1 ) 2 , |\mathbb{E}_F|^{\frac{n(n-d+1)}{2}}, ∣ E F ∣ 2 n ( n − d + 1 ) ,
in H n ( E F / E F q ) \mathcal{H}_n(\mathbb{E}_F/\mathbb{E}_F^q) H n ( E F / E F q ) , attaining the Schmidt bound (5) . In particular, no parity restriction on the pair ( n , d ) (n,d) ( n , d ) remains. This fills the exceptional even–even range 4 ≤ d ≤ n − 2 4\le d\le n-2 4 ≤ d ≤ n − 2 , which was left open by Schmidt’s constructions. At the same time, this construction places all admissible values of n n n and d d d inside a single uniform family.
In the next section we recall Schmidt’s original constructions, in the cases where they exist, in a skew-polynomial form compatible with the present notation. This will allow us to compare the classical trace constructions with the new family above.
Schmidt's constructions in the skew-polynomial framework
We now recall Schmidt’s trace constructions for maximum additive Hermitian rank-metric codes [20 ] , recast in a skew-polynomial setting. We work over the quadratic extension
E : = F q 2 , F : = F q , \mathbb{E} := \mathbb{F}_{q^2}, \qquad\mathbb{F} := \mathbb{F}_q, E := F q 2 , F := F q ,
and we regard
L : = F q 2 n \mathbb{L} := \mathbb{F}_{q^{2n}} L := F q 2 n
as an n n n -dimensional vector space over E \mathbb{E} E . Let
S : = L [ X ; σ ] , σ ( α ) = α q 2 , S := \mathbb{L}[X;\sigma], \qquad\sigma(\alpha) = \alpha^{q^2}, S := L [ X ; σ ] , σ ( α ) = α q 2 ,
and set
S 0 : = S S ( X n − 1 ) . S_0 := \frac{S}{S(X^n-1)}. S 0 := S ( X n − 1 ) S .
Since σ n = id L \sigma^n = \operatorname{id}_{\mathbb{L}} σ n = id L , the polynomial X n − 1 X^n-1 X n − 1 is central in S S S . Thus, S 0 S_0 S 0 is an E \mathbb{E} E -algebra isomorphic to M n ( E ) M_n(\mathbb{E}) M n ( E ) . Also, we can identify S 0 S_0 S 0 with the algebra of q 2 q^2 q 2 -linearized polynomials over L \mathbb{L} L modulo Z q 2 n − Z Z^{q^{2n}}-Z Z q 2 n − Z via
f = ∑ i = 0 n − 1 c i X i ↦ P f : = ∑ i = 0 n − 1 c i Z q 2 i . f = \sum_{i=0}^{n-1} c_i X^i \mapsto P_f := \sum_{i=0}^{n-1} c_i Z^{q^{2i}}. f = i = 0 ∑ n − 1 c i X i ↦ P f := i = 0 ∑ n − 1 c i Z q 2 i .
To f ∈ S 0 f \in S_0 f ∈ S 0 we associate the E \mathbb{E} E -sesquilinear form
H f ( u , v ) : = Tr L / E ( P f ( u ) v q ) , u , v ∈ L . (19) H_f(u,v) := \operatorname{Tr}_{\mathbb{L}/\mathbb{E}}\left(P_f(u)v^{q}\right), \qquad u,v \in\mathbb{L}. \tag*{(19)} H f ( u , v ) := Tr L / E ( P f ( u ) v q ) , u , v ∈ L . (19)
The Hermitian condition can now be written directly on the coefficients of f f f . Namely, if f = ∑ i = 0 n − 1 c i X i f = \sum_{i=0}^{n-1} c_i X^i f = ∑ i = 0 n − 1 c i X i , then H f H_f H f is Hermitian over E / F \mathbb{E}/\mathbb{F} E / F if and only if
c n − i + 1 = c i q 2 n − 2 i + 1 , (20) c_{n-i+1} = c_i^{q^{2n-2i+1}}, \tag*{(20)} c n − i + 1 = c i q 2 n − 2 i + 1 , (20)
for every i ∈ { 0 , … , n − 1 } i \in\{0,\ldots,n-1\} i ∈ { 0 , … , n − 1 } , where the indices are read modulo n n n ; see also [15 ] . Equivalently, the Hermitian forms are the fixed points of the semilinear involution
Θ c y c : S 0 ⟶ S 0 , Θ c y c ( ∑ i = 0 n − 1 c i X i ) = ∑ i = 0 n − 1 c i q 2 n − 2 i + 1 X n − i + 1 . \Theta_{\mathrm{cyc}}: S_0 \longrightarrow S_0, \qquad
\Theta_{\mathrm{cyc}}\left(\sum_{i=0}^{n-1} c_i X^i\right)
= \sum_{i=0}^{n-1} c_i^{q^{2n-2i+1}}X^{n-i+1}. Θ cyc : S 0 ⟶ S 0 , Θ cyc ( i = 0 ∑ n − 1 c i X i ) = i = 0 ∑ n − 1 c i q 2 n − 2 i + 1 X n − i + 1 .
Thus
H f is Hermitian over E / F ⟺ Θ c y c ( f ) = f . (21) H_f \text{ is Hermitian over } \mathbb{E}/\mathbb{F}
\Longleftrightarrow
\Theta_{\mathrm{cyc}}(f)=f. \tag*{(21)} H f is Hermitian over E / F ⟺ Θ cyc ( f ) = f . (21)
We now recall the two constructions of maximum additive Hermitian rank-metric codes given by Schmidt [20 ] .
Theorem 4.2 (see [20 Theorem 4] ). Assume that 1 ≤ d ≤ n − 1 1 \le d \le n-1 1 ≤ d ≤ n − 1 and that n n n and d d d have opposite parity. Set t : = n − d + 1 2 t := \frac{n-d+1}{2} t := 2 n − d + 1 . Define
K d 1 : = { ∑ j = 1 t ( α j q X j + α j q 2 n − 2 j + 2 X n − j + 1 ) + S ( X n − 1 ) : α 1 , … , α t ∈ L } ⊆ S 0 . (22) \mathcal{K}_d^1 := \left\{
\sum_{j=1}^{t}\left(\alpha_j^q X^j+\alpha_j^{q^{2n-2j+2}}X^{n-j+1}\right)+S(X^n-1)
:\alpha_1,\ldots,\alpha_t\in\mathbb{L}
\right\}\subseteq S_0. \tag*{(22)} K d 1 := { j = 1 ∑ t ( α j q X j + α j q 2 n − 2 j + 2 X n − j + 1 ) + S ( X n − 1 ) : α 1 , … , α t ∈ L } ⊆ S 0 . (22) Then, under the identification above, K d 1 \mathcal{K}_d^1 K d 1 is a maximum additive Hermitian d d d -code in H n ( E / F ) \mathcal{H}_n(\mathbb{E}/\mathbb{F}) H n ( E / F ) .
Theorem 4.3 (see [20 Theorem 5] ). Assume that n n n and d d d are odd and that 1 ≤ d ≤ n 1 \le d \le n 1 ≤ d ≤ n . Set t : = n − d 2 t := \frac{n-d}{2} t := 2 n − d . Define
K d 2 : = { α 0 X n + 1 2 + ∑ j = 1 t ( α j q X n − 2 j + 1 2 + α j q n + 2 j + 1 X n + 2 j + 1 2 ) + S ( X n − 1 ) : α 0 ∈ F q n , α 1 , … , α t ∈ L } ⊆ S 0 . (23) \mathcal{K}_{d}^{2} := \left\{ \alpha_{0}X^{\frac{n+1}{2}}+\sum_{j=1}^{t}\left(\alpha_{j}^{q}X^{\frac{n-2j+1}{2}}+\alpha_{j}^{q^{n+2j+1}}X^{\frac{n+2j+1}{2}}\right)+S(X^{n}-1):\alpha_{0}\in\mathbb{F}_{q^{n}},\ \alpha_{1},\ldots,\alpha_{t}\in L\right\}\subseteq S_{0}. \tag*{(23)} K d 2 := { α 0 X 2 n + 1 + j = 1 ∑ t ( α j q X 2 n − 2 j + 1 + α j q n + 2 j + 1 X 2 n + 2 j + 1 ) + S ( X n − 1 ) : α 0 ∈ F q n , α 1 , … , α t ∈ L } ⊆ S 0 . (23) Then, under the identification above, K d 2 \mathcal{K}_{d}^{2} K d 2 is a maximum additive Hermitian d d d -code in H n ( E / F ) \mathcal{H}_{n}(\mathbb{E}/\mathbb{F}) H n ( E / F ) .
A scalar-field invariant
We introduce a simple invariant which separates the codes constructed in Theorem 4.1 from the trace constructions of Schmidt, whenever the latter are available over the same Hermitian space. The invariant measures the largest field of scalars, inside the fixed field of the Hermitian involution, which preserves the code.
Let E / F \mathbb{E}/\mathbb{F} E / F be a quadratic extension of finite fields, and let ρ \rho ρ be the nontrivial element of Gal ( E / F ) \operatorname{Gal}(\mathbb{E}/\mathbb{F}) Gal ( E / F ) . We use the standard notion of equivalence for Hermitian rank-metric codes. For N ∈ GL n ( E ) N \in\operatorname{GL}_{n}(\mathbb{E}) N ∈ GL n ( E ) , μ ∈ F × \mu\in\mathbb{F}^{\times} μ ∈ F × , φ ∈ Aut ( E ) \varphi\in\operatorname{Aut}(\mathbb{E}) φ ∈ Aut ( E ) , and B ∈ H n ( E / F ) B \in\mathcal{H}_{n}(\mathbb{E}/\mathbb{F}) B ∈ H n ( E / F ) , the map
Φ μ , N , φ , B : H n ( E / F ) ⟶ H n ( E / F ) \Phi_{\mu,N,\varphi,B}:\mathcal{H}_{n}(\mathbb{E}/\mathbb{F})\longrightarrow\mathcal{H}_{n}(\mathbb{E}/\mathbb{F}) Φ μ , N , φ , B : H n ( E / F ) ⟶ H n ( E / F )
defined by
Φ μ , N , φ , B ( A ) = μ N A φ N ∗ + B \Phi_{\mu,N,\varphi,B}(A)=\mu NA^{\varphi}N^{*}+B Φ μ , N , φ , B ( A ) = μ N A φ N ∗ + B
preserves the rank distance. Here A φ A^{\varphi} A φ denotes the matrix obtained by applying φ \varphi φ entrywise.
Conversely, every semilinear isometry of the Hermitian space preserving the rank-metric is of this form; see [25 ] . Accordingly, two codes C , D ⊆ H n ( E / F ) C,D\subseteq\mathcal{H}_{n}(\mathbb{E}/\mathbb{F}) C , D ⊆ H n ( E / F ) are said to be Hermitian equivalent if
D = Φ μ , N , φ , B ( C ) D=\Phi_{\mu,N,\varphi,B}(C) D = Φ μ , N , φ , B ( C )
for some N ∈ GL n ( E ) N\in\operatorname{GL}_{n}(\mathbb{E}) N ∈ GL n ( E ) , μ ∈ F × \mu\in\mathbb{F}^{\times} μ ∈ F × , φ ∈ Aut ( E ) \varphi\in\operatorname{Aut}(\mathbb{E}) φ ∈ Aut ( E ) , and B ∈ H n ( E / F ) B\in\mathcal{H}_{n}(\mathbb{E}/\mathbb{F}) B ∈ H n ( E / F ) . For additive codes the translation term may be omitted; equivalently,
D = Φ μ , N , φ ( C ) = { μ N A φ N ∗ : A ∈ C } . (24) D=\Phi_{\mu,N,\varphi}(C)=\{\mu NA^{\varphi}N^{*}:A\in C\}. \tag*{(24)} D = Φ μ , N , φ ( C ) = { μ N A φ N ∗ : A ∈ C } . (24)
When φ = id E \varphi=\operatorname{id}_{\mathbb{E}} φ = id E , we say that C C C and D D D are linearly Hermitian equivalent .
The following elementary invariant records the largest subfield of the fixed field over which an additive Hermitian code remains linear.
Definition 4.4. Let C ⊆ H n ( E / F ) C\subseteq\mathcal{H}_{n}(\mathbb{E}/\mathbb{F}) C ⊆ H n ( E / F ) be a nonzero additive code. We define
F ( C ) : = { λ ∈ F : λ C ⊆ C } . \mathcal{F}(C):=\{\lambda\in\mathbb{F}:\lambda C\subseteq C\}. F ( C ) := { λ ∈ F : λ C ⊆ C } .
Lemma 4.5. For every nonzero additive code C ⊆ H n ( E / F ) C\subseteq\mathcal{H}_{n}(\mathbb{E}/\mathbb{F}) C ⊆ H n ( E / F ) , the set F ( C ) \mathcal{F}(C) F ( C ) is a subfield of F \mathbb{F} F .
Proof. Clearly 0 , 1 ∈ F ( C ) 0,1\in\mathcal{F}(C) 0 , 1 ∈ F ( C ) . If λ , η ∈ F ( C ) \lambda,\eta\in\mathcal{F}(C) λ , η ∈ F ( C ) , then, since C C C is additive,
( λ + η ) C ⊆ C , λ η C ⊆ C . (\lambda+\eta)C\subseteq C,\qquad\lambda\eta C\subseteq C. ( λ + η ) C ⊆ C , λ η C ⊆ C .
Moreover, − λ ∈ F ( C ) -\lambda\in\mathcal{F}(C) − λ ∈ F ( C ) whenever λ ∈ F ( C ) \lambda\in\mathcal{F}(C) λ ∈ F ( C ) . If λ ≠ 0 \lambda\ne0 λ = 0 , then the inclusion λ C ⊆ C \lambda C\subseteq C λ C ⊆ C is an equality because C C C is finite, and hence λ − 1 C = C \lambda^{-1}C=C λ − 1 C = C . Thus F ( C ) \mathcal{F}(C) F ( C ) is a subfield of F \mathbb{F} F . □ \square □
Proposition 4.6. Let C , D ⊆ H n ( E / F ) \mathcal{C},\mathcal{D}\subseteq\mathcal{H}_n(\mathbb{E}/\mathbb{F}) C , D ⊆ H n ( E / F ) be nonzero additive codes. Suppose that there exist N ∈ GL n ( E ) N\in\operatorname{GL}_n(\mathbb{E}) N ∈ GL n ( E ) , μ ∈ F × \mu\in\mathbb{F}^{\times} μ ∈ F × , and φ ∈ Aut ( E ) \varphi\in\operatorname{Aut}(\mathbb{E}) φ ∈ Aut ( E ) , such that
D = Φ μ , N , φ ( C ) . \mathcal{D}=\Phi_{\mu,N,\varphi}(\mathcal{C}). D = Φ μ , N , φ ( C ) . Then
F ( D ) = φ ( F ( C ) ) . \mathcal{F}(\mathcal{D})=\varphi(\mathcal{F}(\mathcal{C})). F ( D ) = φ ( F ( C )) . In particular, the cardinality of F ( C ) \mathcal{F}(\mathcal{C}) F ( C ) is invariant under Hermitian equivalence.
Proof. Let λ ∈ F \lambda\in\mathbb{F} λ ∈ F . For every A ∈ C A\in\mathcal{C} A ∈ C , we have
λ Φ μ , N , φ ( A ) = μ N ( ( φ − 1 ( λ ) A ) φ ) N ∗ = Φ μ , N , φ ( φ − 1 ( λ ) A ) . \lambda\Phi_{\mu,N,\varphi}(A)=\mu N((\varphi^{-1}(\lambda)A)^\varphi)N^*=\Phi_{\mu,N,\varphi}(\varphi^{-1}(\lambda)A). λ Φ μ , N , φ ( A ) = μ N (( φ − 1 ( λ ) A ) φ ) N ∗ = Φ μ , N , φ ( φ − 1 ( λ ) A ) . Thus
λ D ⊆ D ⟺ φ − 1 ( λ ) C ⊆ C . \lambda\mathcal{D}\subseteq\mathcal{D}\quad\Longleftrightarrow\quad\varphi^{-1}(\lambda)\mathcal{C}\subseteq\mathcal{C}. λ D ⊆ D ⟺ φ − 1 ( λ ) C ⊆ C . Equivalently,
λ ∈ F ( D ) ⟺ φ − 1 ( λ ) ∈ F ( C ) . \lambda\in\mathcal{F}(\mathcal{D})\quad\Longleftrightarrow\quad\varphi^{-1}(\lambda)\in\mathcal{F}(\mathcal{C}). λ ∈ F ( D ) ⟺ φ − 1 ( λ ) ∈ F ( C ) . Therefore F ( D ) = φ ( F ( C ) ) \mathcal{F}(\mathcal{D})=\varphi(\mathcal{F}(\mathcal{C})) F ( D ) = φ ( F ( C )) , proving the assertion. □
We first compute the invariant for Schmidt’s constructions. Here, E = F q 2 \mathbb{E}=\mathbb{F}_{q^2} E = F q 2 , F = F q \mathbb{F}=\mathbb{F}_q F = F q , L = F q 2 n L=\mathbb{F}_{q^{2n}} L = F q 2 n , S = L [ X ; σ ] S=L[X;\sigma] S = L [ X ; σ ] , with σ ( α ) = α q 2 \sigma(\alpha)=\alpha^{q^2} σ ( α ) = α q 2 , and S 0 = S / S ( X n − 1 ) S_0=S/S(X^n-1) S 0 = S / S ( X n − 1 ) . Via the identification described above, the codes
K d 1 ⊆ S 0 ≅ M n ( F q 2 ) and K d 2 ⊆ S 0 ≅ M n ( F q 2 ) \mathcal{K}^{1}_{d}\subseteq S_0\cong M_n(\mathbb{F}_{q^2})\qquad\text{and}\qquad\mathcal{K}^{2}_{d}\subseteq S_0\cong M_n(\mathbb{F}_{q^2}) K d 1 ⊆ S 0 ≅ M n ( F q 2 ) and K d 2 ⊆ S 0 ≅ M n ( F q 2 )
defined in (22) and (23) are regarded as additive codes in H n ( F q 2 / F q ) \mathcal{H}_n(\mathbb{F}_{q^2}/\mathbb{F}_q) H n ( F q 2 / F q ) .
Proposition 4.7. Let C ⊆ H n ( F q 2 / F q ) \mathcal{C}\subseteq\mathcal{H}_n(\mathbb{F}_{q^2}/\mathbb{F}_q) C ⊆ H n ( F q 2 / F q ) be either the code corresponding to K d 1 \mathcal{K}^{1}_{d} K d 1 in (22) , when 1 ≤ d ≤ n − 1 1\leq d\leq n-1 1 ≤ d ≤ n − 1 and n , d n,d n , d have opposite parity, or the code corresponding to K d 2 \mathcal{K}^{2}_{d} K d 2 in (23) , when n , d n,d n , d are odd. Then
F ( C ) = F q . \mathcal{F}(\mathcal{C})=\mathbb{F}_q. F ( C ) = F q . Proof. By definition, F ( C ) ⊆ F q \mathcal{F}(\mathcal{C})\subseteq\mathbb{F}_q F ( C ) ⊆ F q . Thus, it is enough to prove that F q \mathbb{F}_q F q acts on the corresponding code in S 0 S_0 S 0 . Assume first that C \mathcal{C} C corresponds to K d 1 \mathcal{K}^{1}_{d} K d 1 . For λ ∈ F q \lambda\in\mathbb{F}_q λ ∈ F q , we have
λ ( a j q X j + a j q 2 n − 2 j + 2 X n − j + 1 ) = ( λ a j ) q X j + ( λ a j ) q 2 n − 2 j + 2 X n − j + 1 , \lambda\left(a_j^qX^j+a_j^{q^{2n-2j+2}}X^{n-j+1}\right)=(\lambda a_j)^qX^j+(\lambda a_j)^{q^{2n-2j+2}}X^{n-j+1}, λ ( a j q X j + a j q 2 n − 2 j + 2 X n − j + 1 ) = ( λ a j ) q X j + ( λ a j ) q 2 n − 2 j + 2 X n − j + 1 , since λ q = λ \lambda^q=\lambda λ q = λ . This shows that λ K d 1 ⊆ K d 1 \lambda\mathcal{K}^{1}_{d}\subseteq\mathcal{K}^{1}_{d} λ K d 1 ⊆ K d 1 . The same calculation gives λ K d 2 ⊆ K d 2 \lambda\mathcal{K}^{2}_{d}\subseteq\mathcal{K}^{2}_{d} λ K d 2 ⊆ K d 2 . In either case F q ⊆ F ( C ) \mathbb{F}_q\subseteq\mathcal{F}(\mathcal{C}) F q ⊆ F ( C ) , and the reverse inclusion was already observed. The assertion follows. □
We now consider the family constructed in Theorem 4.1. We use the notation of § 4.1 . Thus L = F q n L=\mathbb{F}_{q^n} L = F q n , K = F q K=\mathbb{F}_q K = F q , and F ∈ F q [ Y ] F\in\mathbb{F}_q[Y] F ∈ F q [ Y ] is a monic, irreducible, self-reciprocal polynomial of even degree s s s . We set R F : = R / R F ( X n ) R_F:=R/RF(X^n) R F := R / R F ( X n ) . The center of R F R_F R F is
E = E F : = Z ( R F ) ≅ F q s . \mathbb{E}=\mathbb{E}_F:=Z(R_F)\cong\mathbb{F}_{q^s}. E = E F := Z ( R F ) ≅ F q s .
The involution Θ \Theta Θ induces on E F \mathbb{E}_F E F an involution ϱ \varrho ϱ , whose fixed field is
F : = E F ϱ ≅ F q s / 2 . \mathbb{F}:=\mathbb{E}_F^\varrho\cong\mathbb{F}_{q^{s/2}}. F := E F ϱ ≅ F q s /2 .
Thus the ambient Hermitian space associated with F F F is H n ( E F / E F ϱ ) \mathcal{H}_n(\mathbb{E}_F/\mathbb{E}_F^\varrho) H n ( E F / E F ϱ ) .
Proposition 4.8. Assume that 2 ≤ d ≤ n 2 \le d \le n 2 ≤ d ≤ n , and put ℓ : = s ( n − d + 1 ) − 2 2 \ell:= \frac{s(n-d+1)-2}{2} ℓ := 2 s ( n − d + 1 ) − 2 . Let
C : = S ℓ ( F ) = { α 0 + ∑ i = 1 ℓ ( σ − i ( α i ) X − i + α i X i ) + R F ( X n ) : α 0 , … , α ℓ ∈ L } ⊆ H n ( E / F ) . C := S_{\ell}(F) = \left\{ \alpha_0 + \sum_{i=1}^{\ell} \left(\sigma^{-i}(\alpha_i)X^{-i}+\alpha_iX^i\right)+RF(X^n):\alpha_0,\ldots,\alpha_{\ell}\in\mathbb{L}\right\}\subseteq\mathcal{H}_n(E/F). C := S ℓ ( F ) = { α 0 + i = 1 ∑ ℓ ( σ − i ( α i ) X − i + α i X i ) + R F ( X n ) : α 0 , … , α ℓ ∈ L } ⊆ H n ( E / F ) . Then
F ( C ) ⊊ F \mathcal{F}(C)\subsetneq\mathbb{F} F ( C ) ⊊ F whenever s > 2 s>2 s > 2 . Moreover, if ℓ < n s 4 \ell<\frac{ns}{4} ℓ < 4 n s , then
F ( C ) = F q . \mathcal{F}(C)=\mathbb{F}_q. F ( C ) = F q .
Proof. We have F ( C ) = { λ ∈ F : λ C ⊆ C } \mathcal{F}(C)=\{\lambda\in\mathbb{F}:\lambda C\subseteq C\} F ( C ) = { λ ∈ F : λ C ⊆ C } . Every element λ ∈ F \lambda\in\mathbb{F} λ ∈ F has a unique expression
λ = ∑ h = 0 s / 2 − 1 λ h ( X − n + X n ) h + R F ( X n ) , (25) \lambda=\sum_{h=0}^{s/2-1}\lambda_h(X^{-n}+X^n)^h+RF(X^n), \tag*{(25)} λ = h = 0 ∑ s /2 − 1 λ h ( X − n + X n ) h + R F ( X n ) , (25)
with λ h ∈ F q \lambda_h\in\mathbb{F}_q λ h ∈ F q . Let m : = n s 2 m:=\frac{ns}{2} m := 2 n s . By Theorem 3.6, every element of Fix ( Θ ) \operatorname{Fix}(\Theta) Fix ( Θ ) has a unique symmetric expansion with exponents in the range [ − m + 1 , m − 1 ] [-m+1,m-1] [ − m + 1 , m − 1 ] . In this expansion, the elements of C = S ℓ ( F ) C=S_{\ell}(F) C = S ℓ ( F ) are precisely those fixed elements whose symmetric support is contained in the interval [ − ℓ , ℓ ] [-\ell,\ell] [ − ℓ , ℓ ] . We first prove that F ( C ) ⊊ F \mathcal{F}(C)\subsetneq\mathbb{F} F ( C ) ⊊ F when s > 2 s>2 s > 2 . Set
η : = X − n + X n + R F ( X n ) ∈ F . \eta:=X^{-n}+X^n+RF(X^n)\in\mathbb{F}. η := X − n + X n + R F ( X n ) ∈ F .
It is enough to show that η C ⊈ C \eta C\nsubseteq C η C ⊈ C .
If ℓ < n \ell<n ℓ < n , then 1 + R F ( X n ) ∈ C 1+RF(X^n)\in C 1 + R F ( X n ) ∈ C , whereas
η ( 1 + R F ( X n ) ) = X − n + X n + R F ( X n ) \eta(1+RF(X^n))=X^{-n}+X^n+RF(X^n) η ( 1 + R F ( X n )) = X − n + X n + R F ( X n )
has symmetric support { − n , n } \{-n,n\} { − n , n } , which is not contained in [ − ℓ , ℓ ] [-\ell,\ell] [ − ℓ , ℓ ] . Hence η ∉ F ( C ) \eta\notin\mathcal{F}(C) η ∈ / F ( C ) . Assume now that ℓ ≥ n \ell\ge n ℓ ≥ n . We claim that there exists an integer i i i such that
0 < i ≤ ℓ , ℓ < i + n < m . 0<i\le\ell,\qquad\ell<i+n<m. 0 < i ≤ ℓ , ℓ < i + n < m .
Indeed, if ℓ < m − n \ell<m-n ℓ < m − n , we may take i = ℓ i=\ell i = ℓ . If ℓ ≥ m − n \ell\ge m-n ℓ ≥ m − n , we take
i : = m − n − 1. i:=m-n-1. i := m − n − 1.
This integer is positive since s > 2 s>2 s > 2 and n ≥ 2 n\ge2 n ≥ 2 . Moreover, i ≤ ℓ i\le\ell i ≤ ℓ , and
On the other hand, since 2 ≤ d ≤ n 2\le d\le n 2 ≤ d ≤ n , we have
ℓ = s ( n − d + 1 ) − 2 2 ≤ s ( n − 1 ) − 2 2 = m − s 2 − 1 < m − 1. \ell=\frac{s(n-d+1)-2}{2}\le\frac{s(n-1)-2}{2}=m-\frac{s}{2}-1<m-1. ℓ = 2 s ( n − d + 1 ) − 2 ≤ 2 s ( n − 1 ) − 2 = m − 2 s − 1 < m − 1.
Thus ℓ < i + n < m \ell<i+n<m ℓ < i + n < m , as required. Now
u i ‾ : = X i + X − i + R F ( X n ) \overline{u_i}:=X^i+X^{-i}+RF(X^n) u i := X i + X − i + R F ( X n )
belongs to C C C . Multiplying by X n + X − n + R F ( X n ) X^n+X^{-n}+RF(X^n) X n + X − n + R F ( X n ) , we obtain
X i + n + X i − n + X n − i + X − i − n + R F ( X n ) . X^{i+n}+X^{i-n}+X^{n-i}+X^{-i-n}+RF(X^n). X i + n + X i − n + X n − i + X − i − n + R F ( X n ) .
The exponent i + n i+n i + n satisfies ℓ < i + n < m \ell<i+n<m ℓ < i + n < m . Therefore, ( X − n + X n ) u i ‾ ∉ C (X^{-n}+X^n)\overline{u_i}\notin C ( X − n + X n ) u i ∈ / C and consequently
F ( C ) ⊊ F \mathcal{F}(C)\subsetneq\mathbb{F} F ( C ) ⊊ F
for s > 2 s>2 s > 2 . We now assume that ℓ < n s 4 = m 2 \ell< \frac{ns}{4} = \frac{m}{2} ℓ < 4 n s = 2 m and prove that F ( C ) = F q \mathcal{F}(C) = \mathbb{F}_q F ( C ) = F q . The inclusion F q ⊆ F ( C ) \mathbb{F}_q \subseteq\mathcal{F}(C) F q ⊆ F ( C ) is immediate, since C C C is an F q \mathbb{F}_q F q -subspace. Conversely, let
λ ∈ F ( C ) ⊆ F , \lambda\in\mathcal{F}(C) \subseteq\mathbb{F}, λ ∈ F ( C ) ⊆ F ,
and write λ \lambda λ as in (25) . Let t t t be the largest index such that λ t ≠ 0 \lambda_t \ne0 λ t = 0 . We prove that t = 0 t = 0 t = 0 .
Suppose, to the contrary, that t > 0 t > 0 t > 0 . Since 1 + R F ( X n ) ∈ C 1 + RF(X^n) \in C 1 + R F ( X n ) ∈ C and λ C ⊆ C \lambda C \subseteq C λ C ⊆ C , we have λ ∈ C \lambda\in C λ ∈ C . In the expansion
λ = ∑ h = 0 s / 2 − 1 λ h ( X n + X − n ) h + R F ( X n ) , \lambda= \sum_{h=0}^{s/2-1} \lambda_h(X^n+X^{-n})^h + RF(X^n), λ = h = 0 ∑ s /2 − 1 λ h ( X n + X − n ) h + R F ( X n ) ,
the highest positive exponent is t n tn t n , and its coefficient is λ t ≠ 0 \lambda_t \ne0 λ t = 0 . Thus, since λ ∈ C \lambda\in C λ ∈ C , we get t n ≤ ℓ tn \le\ell t n ≤ ℓ . Now consider
u ‾ ℓ : = X ℓ + X − ℓ + R F ( X n ) ∈ C . \overline{u}_{\ell}:=X^\ell+X^{-\ell}+RF(X^n)\in C. u ℓ := X ℓ + X − ℓ + R F ( X n ) ∈ C .
Since λ ∈ F ( C ) \lambda\in\mathcal{F}(C) λ ∈ F ( C ) , we must have λ u ‾ ℓ ∈ C \lambda\overline{u}_{\ell}\in C λ u ℓ ∈ C . However, the product λ u ‾ ℓ \lambda\overline{u}_{\ell} λ u ℓ has a nonzero component at the exponent ℓ + t n \ell+tn ℓ + t n . Since t > 0 t>0 t > 0 , we have ℓ + t n > ℓ \ell+tn>\ell ℓ + t n > ℓ . Moreover, using t n ≤ ℓ tn\le\ell t n ≤ ℓ and ℓ < m / 2 \ell<m/2 ℓ < m /2 , we get
ℓ + t n ≤ 2 ℓ < m . \ell+tn\le2\ell<m. ℓ + t n ≤ 2 ℓ < m .
Thus ℓ + t n \ell+tn ℓ + t n lies in the range [ − m + 1 , m − 1 ] [-m+1,m-1] [ − m + 1 , m − 1 ] but outside the interval [ − ℓ , ℓ ] [-\ell,\ell] [ − ℓ , ℓ ] . This contradicts λ u ‾ ℓ ∈ C \lambda\overline{u}_{\ell}\in C λ u ℓ ∈ C . Hence t = 0 t=0 t = 0 , and therefore λ ∈ F q \lambda\in\mathbb{F}_q λ ∈ F q . This proves the claim. □
We can now separate the new family from Schmidt’s trace constructions.
Theorem 4.9. Assume that s > 2 s>2 s > 2 and 2 ≤ d ≤ n 2\le d\le n 2 ≤ d ≤ n . Let
C = S ℓ ( F ) ⊆ H n ( E / F ) , ℓ = s ( n − d + 1 ) − 2 2 , C=S_{\ell}(\mathbb{F})\subseteq\mathcal{H}_n(\mathbb{E}/\mathbb{F}),\qquad\ell=\frac{s(n-d+1)-2}{2}, C = S ℓ ( F ) ⊆ H n ( E / F ) , ℓ = 2 s ( n − d + 1 ) − 2 , be the maximum additive Hermitian d d d -code constructed in Theorem 4.1, where E = E F ≅ F q s \mathbb{E}=\mathbb{E}_{\mathbb{F}}\cong\mathbb{F}_{q^s} E = E F ≅ F q s , F = E F ≅ F q s / 2 \mathbb{F}=\mathbb{E}_{\mathbb{F}}\cong\mathbb{F}_{q^{s/2}} F = E F ≅ F q s /2 . Let
C ′ ⊆ H n ( E / F ) C'\subseteq\mathcal{H}_n(\mathbb{E}/\mathbb{F}) C ′ ⊆ H n ( E / F ) be a Schmidt trace construction from Theorem 4.2 or Theorem 4.3 , whichever applies to the parameters n n n and d d d , realized in the Hermitian space H n ( E / F ) \mathcal{H}_n(\mathbb{E}/\mathbb{F}) H n ( E / F ) . Then C C C and C ′ C' C ′ are not Hermitian equivalent.
Proof. Apply Proposition 4.7 with the base field there replaced by F \mathbb{F} F (equivalently, with q q q there replaced by q s / 2 q^{s/2} q s /2 ). The code C ′ C' C ′ then satisfies
F ( C ′ ) = F . \mathcal{F}(C')=\mathbb{F}. F ( C ′ ) = F .
On the other hand, by Proposition 4.8 , the code C C C satisfies
F ( C ) ⊊ F . \mathcal{F}(C)\subsetneq\mathbb{F}. F ( C ) ⊊ F .
Since the cardinality of F ( ⋅ ) \mathcal{F}(\cdot) F ( ⋅ ) is invariant under Hermitian equivalence by Theorem 4.6, the two codes cannot be Hermitian equivalent. □
independently of the choice of F F F . Hence this invariant coincides with the corresponding invariant of the classical trace construction and cannot, by itself, distinguish the quadratic members of our family from the previously known construction.
This limitation is only for the invariant, not for the construction. The case s = 2 s=2 s = 2 is treated by a different argument in Subsection 6.7, where we use the equivalence criterion to prove that, in the range n / 2 < d < n n/2<d<n n /2 < d < n , the codes
S n − d ( F ) , F ∈ I q , 2 s r S_{n-d}(F), \qquad F \in\mathcal{I}_{q,2}^{\mathrm{sr}} S n − d ( F ) , F ∈ I q , 2 sr
give rise, under full Hermitian equivalence, to at least
⌊ L 2 ( q ) 2 h ⌋ \left\lfloor\frac{L_2(q)}{2h} \right\rfloor ⌊ 2 h L 2 ( q ) ⌋
pairwise inequivalent codes, where q = p h q=p^h q = p h ; see Corollary 6.3. Thus, already in the quadratic case, the construction produces many new equivalence classes, although the scalar-field invariant alone does not separate them from the classical construction.
Moreover, the case d = 1 d=1 d = 1 is excluded from the separation statement because then
ℓ = s n − 2 2 , \ell=\frac{sn-2}{2}, ℓ = 2 s n − 2 ,
and the construction gives the whole Hermitian space.
Lyapunov maps and structural invariants
In this section, we introduce Lyapunov maps as structural tools for the central simple algebra R F R_F R F endowed with the involution of the second kind Θ \Theta Θ . For any additive Hermitian code in this setting, these maps give rise to an additive subgroup consisting of all matrices whose associated Lyapunov operators stabilize the code. We prove that this subgroup is invariant under Hermitian semilinear equivalence. We then compute it explicitly for the new family of codes constructed in the previous section. The resulting formula provides a key structural invariant for the equivalence analysis carried out in the next section.
The Lyapunov invariant
Let E / F E/F E / F be a quadratic extension of finite fields, and let ϱ \varrho ϱ be the nontrivial element of Gal ( E / F ) \operatorname{Gal}(E/F) Gal ( E / F ) . Consider the space of Hermitian matrices H n ( E / F ) \mathcal{H}_n(E/F) H n ( E / F ) .
For B ∈ M n ( E ) B\in M_n(E) B ∈ M n ( E ) , define the F F F -linear map
Δ B : M n ( E ) ⟶ M n ( E ) , Δ B ( A ) : = B A + A B ∗ , \Delta_B:M_n(E)\longrightarrow M_n(E), \qquad\Delta_B(A):=BA+AB^*, Δ B : M n ( E ) ⟶ M n ( E ) , Δ B ( A ) := B A + A B ∗ ,
where B ∗ : = ( B ϱ ) ⊤ B^*:=(B^\varrho)^\top B ∗ := ( B ϱ ) ⊤ , consistently with the notation of § 2 .
If A ∈ H n ( E / F ) A\in\mathcal{H}_n(E/F) A ∈ H n ( E / F ) , then
Δ B ( A ) ∗ = ( B A + A B ∗ ) ∗ = A ∗ B ∗ + B A ∗ = A B ∗ + B A = Δ B ( A ) . \Delta_B(A)^*=(BA+AB^*)^*=A^*B^*+BA^*=AB^*+BA=\Delta_B(A). Δ B ( A ) ∗ = ( B A + A B ∗ ) ∗ = A ∗ B ∗ + B A ∗ = A B ∗ + B A = Δ B ( A ) .
Thus Δ B \Delta_B Δ B maps H n ( E / F ) \mathcal{H}_n(E/F) H n ( E / F ) into itself. This is the Hermitian analogue of the classical Lyapunov map A ⟼ B A + A B ∗ A\longmapsto BA+AB^* A ⟼ B A + A B ∗ , in the context of algebras with involution of the second kind.
Definition 5.1. Let C ⊆ H n ( E / F ) C\subseteq\mathcal{H}_n(E/F) C ⊆ H n ( E / F ) be an additive code. We define the Lyapunov invariant of C C C by
L ( C ) : = { B ∈ M n ( E ) : Δ B ( C ) ⊆ C } . \mathcal{L}(C):=\{B\in M_n(E):\Delta_B(C)\subseteq C\}. L ( C ) := { B ∈ M n ( E ) : Δ B ( C ) ⊆ C } .
It is straightforward to show that the set L ( C ) \mathcal{L}(\mathcal{C}) L ( C ) is an additive subgroup of M n ( E ) M_n(\mathbb{E}) M n ( E ) .
Proposition 5.2. Let C , D ⊆ H n ( E / F ) \mathcal{C},\mathcal{D} \subseteq\mathcal{H}_n(\mathbb{E}/\mathbb{F}) C , D ⊆ H n ( E / F ) be additive Hermitian codes. Suppose that
D = { μ N A φ N ∗ : A ∈ C } \mathcal{D} = \{\mu N A^{\varphi}N^{*}: A \in\mathcal{C}\} D = { μ N A φ N ∗ : A ∈ C } for some N ∈ GL n ( E ) N \in\operatorname{GL}_n(\mathbb{E}) N ∈ GL n ( E ) , μ ∈ F × \mu\in\mathbb{F}^{\times} μ ∈ F × , φ ∈ Aut ( E ) \varphi\in\operatorname{Aut}(\mathbb{E}) φ ∈ Aut ( E ) . Then
L ( D ) = { N B φ N − 1 : B ∈ L ( C ) } = N L ( C ) φ N − 1 . \mathcal{L}(\mathcal{D}) = \{N B^{\varphi}N^{-1}: B \in\mathcal{L}(\mathcal{C})\} = N\mathcal{L}(\mathcal{C})^{\varphi}N^{-1}. L ( D ) = { N B φ N − 1 : B ∈ L ( C )} = N L ( C ) φ N − 1 . In particular, ∣ L ( C ) ∣ = ∣ L ( D ) ∣ |\mathcal{L}(\mathcal{C})| = |\mathcal{L}(\mathcal{D})| ∣ L ( C ) ∣ = ∣ L ( D ) ∣ .
Proof. Let B ∈ M n ( E ) B \in M_n(\mathbb{E}) B ∈ M n ( E ) , and set B ′ : = N B φ N − 1 B' := N B^{\varphi}N^{-1} B ′ := N B φ N − 1 . For A ∈ C A \in\mathcal{C} A ∈ C , we compute
Δ B ′ ( μ N A φ N ∗ ) = B ′ ( μ N A φ N ∗ ) + ( μ N A φ N ∗ ) B ′ ∗ = μ N B φ A φ N ∗ + μ N A φ ( B φ ) ∗ N ∗ = μ N ( B A + A B ∗ ) φ N ∗ = μ N Δ B ( A ) φ N ∗ . \begin{aligned}
\Delta_{B'}(\mu N A^{\varphi}N^{*}) &= B'(\mu N A^{\varphi}N^{*}) + (\mu N A^{\varphi}N^{*})B'^{*} \\
&= \mu N B^{\varphi} A^{\varphi}N^{*} + \mu N A^{\varphi}(B^{\varphi})^{*}N^{*} \\
&= \mu N(BA + AB^{*})^{\varphi}N^{*} \\
&= \mu N\Delta_B(A)^{\varphi}N^{*}.
\end{aligned} Δ B ′ ( μ N A φ N ∗ ) = B ′ ( μ N A φ N ∗ ) + ( μ N A φ N ∗ ) B ′ ∗ = μ N B φ A φ N ∗ + μ N A φ ( B φ ) ∗ N ∗ = μ N ( B A + A B ∗ ) φ N ∗ = μ N Δ B ( A ) φ N ∗ .
Hence
Δ B ′ ( D ) ⊆ D ⟺ Δ B ( C ) ⊆ C . \Delta_{B'}(\mathcal{D}) \subseteq\mathcal{D} \quad\Longleftrightarrow\quad\Delta_B(\mathcal{C}) \subseteq\mathcal{C}. Δ B ′ ( D ) ⊆ D ⟺ Δ B ( C ) ⊆ C .
Therefore
B ∈ L ( C ) ⟺ N B φ N − 1 ∈ L ( D ) , B \in\mathcal{L}(\mathcal{C}) \quad\Longleftrightarrow\quad N B^{\varphi}N^{-1} \in\mathcal{L}(\mathcal{D}), B ∈ L ( C ) ⟺ N B φ N − 1 ∈ L ( D ) ,
which proves the asserted equality. □
Computation of the Lyapunov invariant for S ℓ ( F ) S_{\ell}(F) S ℓ ( F )
We keep the notation of § 4.1 .
Thus
L = F q n , K = F q , R = L [ X ; σ ] , ⟨ σ ⟩ = Gal ( L / K ) . \mathbb{L} = \mathbb{F}_{q^n}, \qquad\mathbb{K} = \mathbb{F}_q, \qquad R = \mathbb{L}[X;\sigma], \qquad\langle\sigma\rangle= \operatorname{Gal}(\mathbb{L}/\mathbb{K}). L = F q n , K = F q , R = L [ X ; σ ] , ⟨ σ ⟩ = Gal ( L / K ) .
Let F ∈ F q [ Y ] F \in\mathbb{F}_q[Y] F ∈ F q [ Y ] be monic, irreducible, and self-reciprocal of even degree s s s , and set
R F : = R / R F ( X n ) . R_F := R/RF(X^n). R F := R / R F ( X n ) .
The center of R F R_F R F is
E = E F : = Z ( R F ) ≃ F q s . \mathbb{E} = \mathbb{E}_F := Z(R_F) \simeq\mathbb{F}_{q^s}. E = E F := Z ( R F ) ≃ F q s .
The involution Θ \Theta Θ induces an involution ϱ \varrho ϱ on E \mathbb{E} E , and we write
F : = E ϱ ≃ F q s / 2 . \mathbb{F} := \mathbb{E}^{\varrho} \simeq\mathbb{F}_{q^{s/2}}. F := E ϱ ≃ F q s /2 .
Thus Fix ( Θ ) ⊆ R F \operatorname{Fix}(\Theta) \subseteq R_F Fix ( Θ ) ⊆ R F is identified with the Hermitian space
H n ( E / F ) . \mathcal{H}_n(\mathbb{E}/\mathbb{F}). H n ( E / F ) .
Throughout this subsection, we compute the preimage in R F R_F R F of the Lyapunov invariant under the normalized representation M F \mathcal{M}_F M F . Equivalently, for a code C ⊆ Fix ( Θ ) \mathcal{C} \subseteq\operatorname{Fix}(\Theta) C ⊆ Fix ( Θ ) , we write
L ( C ) : = { b ˉ ∈ R F : b ˉ a ˉ + a ˉ Θ ( b ˉ ) ∈ C for all a ˉ ∈ C } . \mathcal{L}(\mathcal{C}) := \{\bar{b} \in R_F : \bar{b}\bar{a} + \bar{a}\Theta(\bar{b}) \in\mathcal{C}\text{ for all }\bar{a} \in\mathcal{C}\}. L ( C ) := { b ˉ ∈ R F : b ˉ a ˉ + a ˉ Θ ( b ˉ ) ∈ C for all a ˉ ∈ C } .
Under M F \mathcal{M}_F M F , this agrees with the matrix invariant introduced above.
We now compute the Lyapunov invariant of the Hermitian code S ℓ ( F ) S_{\ell}(F) S ℓ ( F ) . The result is stated in Theorem 5.4 . As in § 4.1 , set
m : = n s 2 . m := \frac{ns}{2}. m := 2 n s .
Since F F F is a monic irreducible self-reciprocal polynomial of degree s s s , the integer s s s is even, and hence m m m is an integer. Every element b ˉ ∈ R F \bar{b} \in R_F b ˉ ∈ R F admits a unique expansion
b ˉ = ∑ i = − m m − 1 β i X i + R F ( X n ) , β i ∈ L , \bar{b} = \sum_{i=-m}^{m-1} \beta_i X^i + RF(X^n), \qquad\beta_i \in\mathbb{L}, b ˉ = i = − m ∑ m − 1 β i X i + R F ( X n ) , β i ∈ L ,
with respect to the interval [ − m , m − 1 ] [-m,m-1] [ − m , m − 1 ] .
We shall repeatedly use the following elementary fact without further mention: if I ⊆ Z I \subseteq\mathbb{Z} I ⊆ Z is an interval of length < n s < ns < n s , then the monomials X i + R F ( X n ) X^i+RF(X^n) X i + R F ( X n ) , i ∈ I i \in I i ∈ I , are L \mathbb{L} L -linearly independent in R F R_F R F .
Define
B F : = { λ ∈ E : λ + Θ ( λ ) ∈ K } . \mathcal{B}_F := \{\lambda\in\mathbb{E} : \lambda+\Theta(\lambda) \in\mathbb{K}\}. B F := { λ ∈ E : λ + Θ ( λ ) ∈ K } .
Proposition 5.3. For every 0 ≤ ℓ ≤ n s / 2 − 1 0 \leq\ell\leq ns/2-1 0 ≤ ℓ ≤ n s /2 − 1 , one has
L + B F ⊆ L ( S ℓ ( F ) ) . \mathbb{L}+\mathcal{B}_F \subseteq\mathcal{L}(S_\ell(F)). L + B F ⊆ L ( S ℓ ( F )) .
Proof. Let α ∈ L \alpha\in\mathbb{L} α ∈ L . Since Θ ( α + R F ( X n ) ) = α + R F ( X n ) \Theta(\alpha+RF(X^n))=\alpha+RF(X^n) Θ ( α + R F ( X n )) = α + R F ( X n ) , for every
a ˉ = α 0 + ∑ i = 1 ℓ ( σ − i ( α i ) X − i + α i X i ) + R F ( X n ) ∈ S ℓ ( F ) \bar{a}=\alpha_0+\sum_{i=1}^{\ell}\left(\sigma^{-i}(\alpha_i)X^{-i}+\alpha_iX^i\right)+RF(X^n)\in S_\ell(F) a ˉ = α 0 + i = 1 ∑ ℓ ( σ − i ( α i ) X − i + α i X i ) + R F ( X n ) ∈ S ℓ ( F )
we have
Δ α ( a ˉ ) = α a ˉ + a ˉ α . \Delta_\alpha(\bar{a})=\alpha\bar{a}+\bar{a}\alpha. Δ α ( a ˉ ) = α a ˉ + a ˉ α .
Hence, the coefficient at X i X^i X i is
α α i + α i σ i ( α ) , \alpha\alpha_i+\alpha_i\sigma^i(\alpha), α α i + α i σ i ( α ) ,
while the coefficient at X − i X^{-i} X − i is
α σ − i ( α i ) + σ − i ( α i ) σ − i ( α ) = σ − i ( α α i + α i σ i ( α ) ) . \alpha\sigma^{-i}(\alpha_i)+\sigma^{-i}(\alpha_i)\sigma^{-i}(\alpha)
=\sigma^{-i}\left(\alpha\alpha_i+\alpha_i\sigma^i(\alpha)\right). α σ − i ( α i ) + σ − i ( α i ) σ − i ( α ) = σ − i ( α α i + α i σ i ( α ) ) .
Thus Δ α ( a ˉ ) ∈ S ℓ ( F ) \Delta_\alpha(\bar{a})\in S_\ell(F) Δ α ( a ˉ ) ∈ S ℓ ( F ) , and hence L ⊆ L ( S ℓ ( F ) ) \mathbb{L}\subseteq\mathcal{L}(S_\ell(F)) L ⊆ L ( S ℓ ( F )) .
Now let λ ∈ B F \lambda\in\mathcal{B}_F λ ∈ B F . Since λ \lambda λ is central, for every a ˉ ∈ S ℓ ( F ) \bar{a}\in S_\ell(F) a ˉ ∈ S ℓ ( F ) we have
Δ λ ( a ˉ ) = λ a ˉ + a ˉ Θ ( λ ) = ( λ + Θ ( λ ) ) a ˉ . \Delta_\lambda(\bar{a})=\lambda\bar{a}+\bar{a}\Theta(\lambda)=(\lambda+\Theta(\lambda))\bar{a}. Δ λ ( a ˉ ) = λ a ˉ + a ˉ Θ ( λ ) = ( λ + Θ ( λ )) a ˉ .
By definition, λ + Θ ( λ ) ∈ K = F q \lambda+\Theta(\lambda)\in\mathbb{K}=\mathbb{F}_q λ + Θ ( λ ) ∈ K = F q . Since S ℓ ( F ) S_\ell(F) S ℓ ( F ) is an F q \mathbb{F}_q F q -space, it follows that Δ λ ( a ˉ ) ∈ S ℓ ( F ) \Delta_\lambda(\bar{a})\in S_\ell(F) Δ λ ( a ˉ ) ∈ S ℓ ( F ) . Thus
B F ⊆ L ( S ℓ ( F ) ) . \mathcal{B}_F\subseteq\mathcal{L}(S_\ell(F)). B F ⊆ L ( S ℓ ( F )) .
Since L ( S ℓ ( F ) ) \mathcal{L}(S_\ell(F)) L ( S ℓ ( F )) is additive, the assertion follows. □
We now prove the reverse inclusion in the regime ℓ < m / 2 \ell<m/2 ℓ < m /2 .
Theorem 5.4. Assume that 1 ≤ ℓ < m / 2 1\leq\ell<m/2 1 ≤ ℓ < m /2 and n ≥ 3 n\geq3 n ≥ 3 . Then
L ( S ℓ ( F ) ) = L + B F . \mathcal{L}(S_\ell(F))=\mathbb{L}+\mathcal{B}_F. L ( S ℓ ( F )) = L + B F .
Proof. By Proposition 5.3 , it remains to prove the reverse inclusion.
Let
b ˉ = ∑ i = − m m − 1 β i X i + R F ( X n ) ∈ L ( S ℓ ( F ) ) , \bar{b}=\sum_{i=-m}^{m-1}\beta_iX^i+RF(X^n)\in\mathcal{L}(S_\ell(F)), b ˉ = i = − m ∑ m − 1 β i X i + R F ( X n ) ∈ L ( S ℓ ( F )) ,
with β i ∈ L \beta_i \in\mathbb{L} β i ∈ L , be written with respect to the basis
X − m , X − m + 1 , … , X − 1 , 1 , X , … , X m − 1 X^{-m}, X^{-m+1}, \ldots, X^{-1}, 1, X, \ldots, X^{m-1} X − m , X − m + 1 , … , X − 1 , 1 , X , … , X m − 1
of R F R_F R F . Write also
Θ ( b ˉ ) = ∑ i = − m m − 1 γ i X i + R F ( X n ) , \Theta(\bar{b}) = \sum_{i=-m}^{m-1} \gamma_i X^i + R_F(X^n), Θ ( b ˉ ) = i = − m ∑ m − 1 γ i X i + R F ( X n ) ,
with γ i ∈ L \gamma_i \in\mathbb{L} γ i ∈ L , with respect to the same basis.
We derive the required restrictions on the coefficients β i \beta_i β i in four steps.
Step 1. Derive coefficient relation imposed by the condition 1 + R F ( X n ) ∈ S ℓ ( F ) 1 + R_F(X^n) \in S_\ell(F) 1 + R F ( X n ) ∈ S ℓ ( F ) , and use it to prove that all noncentral coefficients with exponents outside the interval [ − ℓ , ℓ ] [-\ell,\ell] [ − ℓ , ℓ ] vanish.
Step 2. Consider the remaining central coefficients outside [ − ℓ , ℓ ] [-\ell,\ell] [ − ℓ , ℓ ] and show that they belong to K \mathbb{K} K , thereby identifying the corresponding contribution as central.
Step 3. Treat the coefficients with exponents in [ − ℓ , ℓ ] [-\ell,\ell] [ − ℓ , ℓ ] by a descending induction to show that, modulo central terms, all nonconstant coefficients vanish.
Step 4. Determine the possible central summand and prove that it belongs to B F B_F B F , which yields the desired inclusion.
On the noncentral coefficients outside the interval [ − ℓ , ℓ ] [-\ell,\ell] [ − ℓ , ℓ ] .
Since 1 + R F ( X n ) ∈ S ℓ ( F ) 1 + R_F(X^n) \in S_\ell(F) 1 + R F ( X n ) ∈ S ℓ ( F ) , we have
b ˉ + Θ ( b ˉ ) = Δ b ˉ ( 1 + R F ( X n ) ) ∈ S ℓ ( F ) . \bar{b} + \Theta(\bar{b}) = \Delta_{\bar{b}}(1 + R_F(X^n)) \in S_\ell(F). b ˉ + Θ ( b ˉ ) = Δ b ˉ ( 1 + R F ( X n )) ∈ S ℓ ( F ) .
Therefore, for every i ∈ [ − m , m − 1 ] i \in[-m,m-1] i ∈ [ − m , m − 1 ] with ∣ i ∣ > ℓ |i| > \ell ∣ i ∣ > ℓ , the coefficient of X i X^i X i in b ˉ + Θ ( b ˉ ) \bar{b} + \Theta(\bar{b}) b ˉ + Θ ( b ˉ ) is zero, that is,
β i + γ i = 0. (26) \beta_i + \gamma_i = 0. \tag*{(26)} β i + γ i = 0. (26)
In this step, we eliminate all noncentral coefficients outside the interval [ − ℓ , ℓ ] [-\ell,\ell] [ − ℓ , ℓ ] . Let
∣ i ∣ > ℓ , i ≢ 0 ( m o d n ) . |i| > \ell,\qquad i \not\equiv0 \pmod{n}. ∣ i ∣ > ℓ , i ≡ 0 ( mod n ) .
For every α ∈ L \alpha\in\mathbb{L} α ∈ L , the constant element α + R F ( X n ) \alpha+ R_F(X^n) α + R F ( X n ) belongs to S ℓ ( F ) S_\ell(F) S ℓ ( F ) . Hence
Δ b ˉ ( α + R F ( X n ) ) = b ˉ α + α Θ ( b ˉ ) ∈ S ℓ ( F ) . \Delta_{\bar{b}}(\alpha+ R_F(X^n)) = \bar{b}\alpha+ \alpha\Theta(\bar{b}) \in S_\ell(F). Δ b ˉ ( α + R F ( X n )) = b ˉ α + α Θ ( b ˉ ) ∈ S ℓ ( F ) .
The coefficient of X i + R F ( X n ) X^i + R_F(X^n) X i + R F ( X n ) in Δ b ˉ ( α ) \Delta_{\bar{b}}(\alpha) Δ b ˉ ( α ) is
β i σ i ( α ) + α γ i . \beta_i\sigma^i(\alpha) + \alpha\gamma_i. β i σ i ( α ) + α γ i .
Since X i X^i X i lies outside the interval [ − ℓ , ℓ ] [-\ell,\ell] [ − ℓ , ℓ ] , this coefficient is zero for every α ∈ L \alpha\in\mathbb{L} α ∈ L . Using (26) , we obtain
β i ( σ i ( α ) − α ) = 0 , \beta_i(\sigma^i(\alpha)-\alpha)=0, β i ( σ i ( α ) − α ) = 0 ,
for every α ∈ L \alpha\in\mathbb{L} α ∈ L . Since i ≢ 0 ( m o d n ) i \not\equiv0 \pmod n i ≡ 0 ( mod n ) , one has σ i ≠ id L \sigma^i \ne\operatorname{id}_{\mathbb{L}} σ i = id L . Thus β i = 0 \beta_i=0 β i = 0 . Then (26) also gives γ i = 0 \gamma_i=0 γ i = 0 . Consequently, all noncentral coefficients of b ˉ \bar{b} b ˉ outside [ − ℓ , ℓ ] [-\ell,\ell] [ − ℓ , ℓ ] vanish.
On the central coefficients outside the interval [ − ℓ , ℓ ] [-\ell,\ell] [ − ℓ , ℓ ] .
Write s = 2 r s=2r s = 2 r , so that m = n r m=nr m = n r , and put Y = X n Y=X^n Y = X n . Since F F F is monic, self-reciprocal and has constant term 1, the relation F ( Y ) = 0 F(Y)=0 F ( Y ) = 0 , after multiplication by Y − r Y^{-r} Y − r , gives
X m = − X − m + w in R F , (27) X^m=-X^{-m}+w \qquad\text{in } R_F, \tag*{(27)} X m = − X − m + w in R F , (27)
where
w ∈ ∑ − m < ν n < m K X ν n . w \in\sum_{-m<\nu n<m} \mathbb{K}X^{\nu n}. w ∈ − m < ν n < m ∑ K X ν n .
Thus rewriting the boundary term X m X^m X m only produces central powers with coefficients in K \mathbb{K} K .
Let t t t be a positive multiple of n n n such that
For α ∈ L \alpha\in\mathbb{L} α ∈ L , set
Z 1 ( α ) : = σ − 1 ( α ) X − 1 + α X + R F ( X n ) ∈ S ℓ ( F ) . Z_1(\alpha):=\sigma^{-1}(\alpha)X^{-1}+\alpha X+RF(X^n)\in S_\ell(F). Z 1 ( α ) := σ − 1 ( α ) X − 1 + α X + R F ( X n ) ∈ S ℓ ( F ) .
Since
Δ b ˉ ( Z 1 ( α ) ) ∈ S ℓ ( F ) , \Delta_{\bar b}(Z_1(\alpha))\in S_\ell(F), Δ b ˉ ( Z 1 ( α )) ∈ S ℓ ( F ) ,
its coefficient of X t + 1 X^{t+1} X t + 1 is zero for every α ∈ L \alpha\in\mathbb{L} α ∈ L .
We claim that this coefficient forces β t ∈ K \beta_t\in\mathbb{K} β t ∈ K . Indeed, since t t t is a positive multiple of n n n and t < m t<m t < m , we have
As n ≥ 3 n\ge3 n ≥ 3 , it follows that
t + 2 ≤ m − n + 2 ≤ m − 1. t+2\le m-n+2\le m-1. t + 2 ≤ m − n + 2 ≤ m − 1.
Thus both t + 1 t+1 t + 1 and t + 2 t+2 t + 2 belong to the chosen set of representatives [ − m , m − 1 ] [-m,m-1] [ − m , m − 1 ] . Moreover,
ℓ < t + 1 < m , \ell<t+1<m, ℓ < t + 1 < m ,
so X t + 1 X^{t+1} X t + 1 lies outside the interval [ − ℓ , ℓ ] [-\ell,\ell] [ − ℓ , ℓ ] . Using the multiplication rule X i a = σ i ( a ) X i X^i a=\sigma^i(a)X^i X i a = σ i ( a ) X i , the only direct contributions to the coefficient of X t + 1 X^{t+1} X t + 1 in
Δ b ˉ ( Z 1 ( α ) ) = b ˉ Z 1 ( α ) + Z 1 ( α ) Θ ( b ˉ ) \Delta_{\bar b}(Z_1(\alpha))=\bar bZ_1(\alpha)+Z_1(\alpha)\Theta(\bar b) Δ b ˉ ( Z 1 ( α )) = b ˉ Z 1 ( α ) + Z 1 ( α ) Θ ( b ˉ )
come from
β t X t ⋅ α X , β t + 2 X t + 2 ⋅ σ − 1 ( α ) X − 1 , \beta_tX^t\cdot\alpha X,\qquad\beta_{t+2}X^{t+2}\cdot\sigma^{-1}(\alpha)X^{-1}, β t X t ⋅ α X , β t + 2 X t + 2 ⋅ σ − 1 ( α ) X − 1 ,
and from
α X ⋅ γ t X t , σ − 1 ( α ) X − 1 ⋅ γ t + 2 X t + 2 . \alpha X\cdot\gamma_tX^t,\qquad\sigma^{-1}(\alpha)X^{-1}\cdot\gamma_{t+2}X^{t+2}. α X ⋅ γ t X t , σ − 1 ( α ) X − 1 ⋅ γ t + 2 X t + 2 .
Their contributions are respectively
β t σ t ( α ) , β t + 2 σ t + 2 ( σ − 1 ( α ) ) , \beta_t\sigma^t(\alpha),\qquad\beta_{t+2}\sigma^{t+2}(\sigma^{-1}(\alpha)), β t σ t ( α ) , β t + 2 σ t + 2 ( σ − 1 ( α )) ,
and
α σ ( γ t ) , σ − 1 ( α ) σ − 1 ( γ t + 2 ) . \alpha\sigma(\gamma_t),\qquad\sigma^{-1}(\alpha)\sigma^{-1}(\gamma_{t+2}). α σ ( γ t ) , σ − 1 ( α ) σ − 1 ( γ t + 2 ) .
Since t ≡ 0 ( m o d n ) t\equiv0\pmod n t ≡ 0 ( mod n ) , we have σ t = id L \sigma^t=\operatorname{id}_{\mathbb{L}} σ t = id L , and hence σ t + 2 ( σ − 1 ( α ) ) = σ ( α ) \sigma^{t+2}(\sigma^{-1}(\alpha))=\sigma(\alpha) σ t + 2 ( σ − 1 ( α )) = σ ( α ) . Therefore the coefficient of X t + 1 X^{t+1} X t + 1 is
( β t + σ ( γ t ) ) α + β t + 2 σ ( α ) + σ − 1 ( γ t + 2 ) σ − 1 ( α ) . \left(\beta_t+\sigma(\gamma_t)\right)\alpha+\beta_{t+2}\sigma(\alpha)+\sigma^{-1}(\gamma_{t+2})\sigma^{-1}(\alpha). ( β t + σ ( γ t ) ) α + β t + 2 σ ( α ) + σ − 1 ( γ t + 2 ) σ − 1 ( α ) .
Now t + 2 > ℓ t+2>\ell t + 2 > ℓ , and, since t ≡ 0 ( m o d n ) t\equiv0\pmod n t ≡ 0 ( mod n ) and n ≥ 3 n\ge3 n ≥ 3 ,
t + 2 ≠ 0 ( m o d n ) . t+2\ne0\pmod n. t + 2 = 0 ( mod n ) .
Thus t + 2 t+2 t + 2 is the index of a noncentral exponent outside the interval [ − ℓ , ℓ ] [-\ell,\ell] [ − ℓ , ℓ ] . By the previous step,
β t + 2 = 0. \beta_{t+2}=0. β t + 2 = 0.
Moreover, by (26) ,
γ t + 2 = − β t + 2 = 0. \gamma_{t+2}=-\beta_{t+2}=0. γ t + 2 = − β t + 2 = 0.
Hence the coefficient of X t + 1 X^{t+1} X t + 1 reduces to
( β t + σ ( γ t ) ) α . \left(\beta_t+\sigma(\gamma_t)\right)\alpha. ( β t + σ ( γ t ) ) α .
It is zero for every α ∈ L \alpha\in\mathbb{L} α ∈ L . Taking α = 1 \alpha=1 α = 1 , we get
β t + σ ( γ t ) = 0. \beta_t+\sigma(\gamma_t)=0. β t + σ ( γ t ) = 0.
On the other hand, by applying again (26) , we have
γ t = − β t . \gamma_t=-\beta_t. γ t = − β t .
Consequently
0 = β t + σ ( γ t ) = β t − σ ( β t ) . 0=\beta_t+\sigma(\gamma_t)=\beta_t-\sigma(\beta_t). 0 = β t + σ ( γ t ) = β t − σ ( β t ) .
Thus
β t ∈ Fix ( σ ) = K . \beta_t\in\operatorname{Fix}(\sigma)=\mathbb{K}. β t ∈ Fix ( σ ) = K .
This proves that every positive central coefficient of b ˉ \bar{b} b ˉ having support in the set [ − m + 1 , m − 1 ] ∖ [ − ℓ , ℓ ] [-m+1,m-1]\setminus[-\ell,\ell] [ − m + 1 , m − 1 ] ∖ [ − ℓ , ℓ ] belongs to K \mathbb{K} K .
We next handle the boundary coefficient β − m \beta_{-m} β − m . We compare the coefficient of X − m + 1 X^{-m+1} X − m + 1 in
Δ b ˉ ( Z 1 ( α ) ) . \Delta_{\bar{b}}(Z_1(\alpha)). Δ b ˉ ( Z 1 ( α )) .
The exponent − m + 1 -m+1 − m + 1 lies outside the interval, since ℓ < m / 2 \ell<m/2 ℓ < m /2 . The relevant terms are
β − m X − m ⋅ α X , β − m + 2 X − m + 2 ⋅ σ − 1 ( α ) X − 1 , \beta_{-m}X^{-m}\cdot\alpha X,\qquad\beta_{-m+2}X^{-m+2}\cdot\sigma^{-1}(\alpha)X^{-1}, β − m X − m ⋅ α X , β − m + 2 X − m + 2 ⋅ σ − 1 ( α ) X − 1 ,
and
α X ⋅ γ − m X − m , σ − 1 ( α ) X − 1 ⋅ γ − m + 2 X − m + 2 . \alpha X\cdot\gamma_{-m}X^{-m},\qquad\sigma^{-1}(\alpha)X^{-1}\cdot\gamma_{-m+2}X^{-m+2}. α X ⋅ γ − m X − m , σ − 1 ( α ) X − 1 ⋅ γ − m + 2 X − m + 2 .
Their contributions to the coefficient of X − m + 1 X^{-m+1} X − m + 1 are
β − m σ − m ( α ) , β − m + 2 σ − m + 2 ( σ − 1 ( α ) ) , \beta_{-m}\sigma^{-m}(\alpha),\qquad\beta_{-m+2}\sigma^{-m+2}(\sigma^{-1}(\alpha)), β − m σ − m ( α ) , β − m + 2 σ − m + 2 ( σ − 1 ( α )) ,
and
α σ ( γ − m ) , σ − 1 ( α ) σ − 1 ( γ − m + 2 ) . \alpha\sigma(\gamma_{-m}),\qquad\sigma^{-1}(\alpha)\sigma^{-1}(\gamma_{-m+2}). α σ ( γ − m ) , σ − 1 ( α ) σ − 1 ( γ − m + 2 ) .
Here possible boundary rewritings cannot contribute to this term since n ≥ 3 n\geq3 n ≥ 3 . Since m ≡ 0 ( m o d n ) m\equiv0\pmod{n} m ≡ 0 ( mod n ) , this gives
( β − m + σ ( γ − m ) ) α + β − m + 2 σ ( α ) + σ − 1 ( γ − m + 2 ) σ − 1 ( α ) = 0 \left(\beta_{-m}+\sigma(\gamma_{-m})\right)\alpha+\beta_{-m+2}\sigma(\alpha)+\sigma^{-1}(\gamma_{-m+2})\sigma^{-1}(\alpha)=0 ( β − m + σ ( γ − m ) ) α + β − m + 2 σ ( α ) + σ − 1 ( γ − m + 2 ) σ − 1 ( α ) = 0
for every α ∈ L \alpha\in\mathbb{L} α ∈ L . Since n ≥ 3 n\geq3 n ≥ 3 , the three automorphisms
id L , σ , σ − 1 \operatorname{id}_{\mathbb{L}},\qquad\sigma,\qquad\sigma^{-1} id L , σ , σ − 1
are pairwise distinct and they are linearly independent over L \mathbb{L} L as maps L → L \mathbb{L}\to\mathbb{L} L → L . Consequently, the identity above, valid for every α ∈ L \alpha\in\mathbb{L} α ∈ L , implies
β − m + 2 = 0 , γ − m + 2 = 0 , β − m + σ ( γ − m ) = 0. \beta_{-m+2}=0,\qquad\gamma_{-m+2}=0,\qquad\beta_{-m}+\sigma(\gamma_{-m})=0. β − m + 2 = 0 , γ − m + 2 = 0 , β − m + σ ( γ − m ) = 0.
By (26) , γ − m = − β − m \gamma_{-m}=-\beta_{-m} γ − m = − β − m . Hence
β − m − σ ( β − m ) = 0 , \beta_{-m}-\sigma(\beta_{-m})=0, β − m − σ ( β − m ) = 0 ,
and therefore
β − m ∈ K . \beta_{-m}\in\mathbb{K}. β − m ∈ K .
It remains to handle the negative central coefficients outside the interval [ − ℓ , ℓ ] [-\ell,\ell] [ − ℓ , ℓ ] . Arguing as before, we get
β r = − γ r ∈ K , \beta_r=-\gamma_r\in\mathbb{K}, β r = − γ r ∈ K ,
for every central exponent r r r such that − m < r < − ℓ -m < r < -\ell − m < r < − ℓ .
Thus every central coefficient of b ˉ \bar{b} b ˉ outside the interval [ − ℓ , ℓ ] [-\ell,\ell] [ − ℓ , ℓ ] belongs to K \mathbb{K} K . Consequently the outside central part of b ˉ \bar{b} b ˉ is an element of the center E \mathbb{E} E . We may therefore write
b ˉ = c ˉ + λ 0 , \bar{b}=\bar{c}+\lambda_0, b ˉ = c ˉ + λ 0 ,
where λ 0 ∈ E F \lambda_0\in\mathbb{E}_F λ 0 ∈ E F and
c ˉ = ∑ i = − ℓ ℓ β i X i + R F ( X n ) . \bar{c}=\sum_{i=-\ell}^{\ell}\beta_iX^i+RF(X^n). c ˉ = i = − ℓ ∑ ℓ β i X i + R F ( X n ) .
Since c ˉ \bar{c} c ˉ is supported in [ − ℓ , ℓ ] [-\ell,\ell] [ − ℓ , ℓ ] , and since ℓ < m / 2 < m \ell<m/2<m ℓ < m /2 < m , applying Θ \Theta Θ to c ˉ \bar{c} c ˉ , we get
Θ ( c ˉ ) = ∑ i = − ℓ ℓ σ − i ( β i ) X − i + R F ( X n ) . \Theta(\bar{c})=\sum_{i=-\ell}^{\ell}\sigma^{-i}(\beta_i)X^{-i}+RF(X^n). Θ ( c ˉ ) = i = − ℓ ∑ ℓ σ − i ( β i ) X − i + R F ( X n ) .
Therefore c ˉ + Θ ( c ˉ ) \bar{c}+\Theta(\bar{c}) c ˉ + Θ ( c ˉ ) is fixed by Θ \Theta Θ and supported in [ − ℓ , ℓ ] [-\ell,\ell] [ − ℓ , ℓ ] . This implies
c ˉ + Θ ( c ˉ ) ∈ S ℓ ( F ) . \bar{c}+\Theta(\bar{c})\in S_\ell(F). c ˉ + Θ ( c ˉ ) ∈ S ℓ ( F ) .
Since also
b ˉ + Θ ( b ˉ ) ∈ S ℓ ( F ) , \bar{b}+\Theta(\bar{b})\in S_\ell(F), b ˉ + Θ ( b ˉ ) ∈ S ℓ ( F ) ,
we get
u ˉ : = λ 0 + Θ ( λ 0 ) = ( b ˉ + Θ ( b ˉ ) ) − ( c ˉ + Θ ( c ˉ ) ) ∈ S ℓ ( F ) . \bar{u}:=\lambda_0+\Theta(\lambda_0)=(\bar{b}+\Theta(\bar{b}))-(\bar{c}+\Theta(\bar{c}))\in S_\ell(F). u ˉ := λ 0 + Θ ( λ 0 ) = ( b ˉ + Θ ( b ˉ )) − ( c ˉ + Θ ( c ˉ )) ∈ S ℓ ( F ) .
In particular, u ˉ \bar{u} u ˉ is a central element supported in the interval [ − ℓ , ℓ ] [-\ell,\ell] [ − ℓ , ℓ ] , and its central coefficients are in K \mathbb{K} K .
Step 3: On the coefficients inside the interval [ − ℓ , ℓ ] [-\ell,\ell] [ − ℓ , ℓ ] .
We now eliminate the nonconstant part of c ˉ \bar{c} c ˉ , modulo central terms. For 1 ≤ j ≤ ℓ 1\leq j\leq\ell 1 ≤ j ≤ ℓ and α ∈ L \alpha\in\mathbb{L} α ∈ L , set
Z j ( α ) : = σ − j ( α ) X − j + α X j + R F ( X n ) ∈ S ℓ ( F ) . Z_j(\alpha):=\sigma^{-j}(\alpha)X^{-j}+\alpha X^j+RF(X^n)\in S_\ell(F). Z j ( α ) := σ − j ( α ) X − j + α X j + R F ( X n ) ∈ S ℓ ( F ) .
We proceed by descending induction on h = ℓ , ℓ − 1 , … , 1 h=\ell,\ell-1,\ldots,1 h = ℓ , ℓ − 1 , … , 1 . At the beginning of the step h h h , we write
b ˉ = c ˉ h + λ h , \bar{b}=\bar{c}_h+\lambda_h, b ˉ = c ˉ h + λ h ,
where λ h ∈ E \lambda_h\in\mathbb{E} λ h ∈ E ,
u ˉ h : = λ h + Θ ( λ h ) ∈ S ℓ ( F ) , \bar{u}_h:=\lambda_h+\Theta(\lambda_h)\in S_\ell(F), u ˉ h := λ h + Θ ( λ h ) ∈ S ℓ ( F ) ,
and c ˉ h \bar{c}_h c ˉ h is supported in [ − h , h ] [-h,h] [ − h , h ] together with the still untreated coefficients at ± h \pm h ± h . More precisely, all coefficients with absolute value strictly larger than h h h have already either vanished, if their exponent is not divisible by n n n , or have been absorbed into λ h \lambda_h λ h , if their exponent is divisible by n n n . Put
j : = ℓ + 1 − h . j:=\ell+1-h. j := ℓ + 1 − h .
Then 1 ≤ j ≤ ℓ 1\leq j\leq\ell 1 ≤ j ≤ ℓ . We compare the coefficient of X ℓ + 1 X^{\ell+1} X ℓ + 1 in
Δ b ˉ ( Z j ( α ) ) = Δ c ˉ h ( Z j ( α ) ) + u ˉ h Z j ( α ) . \Delta_{\bar{b}}(Z_j(\alpha))=\Delta_{\bar{c}_h}(Z_j(\alpha))+\bar{u}_hZ_j(\alpha). Δ b ˉ ( Z j ( α )) = Δ c ˉ h ( Z j ( α )) + u ˉ h Z j ( α ) .
Here c ˉ h \bar{c}_h c ˉ h , u ˉ h \bar{u}_h u ˉ h , and Z j ( α ) Z_j(\alpha) Z j ( α ) are all supported in exponents whose sums lie in [ − 2 ℓ , 2 ℓ ] [-2\ell,2\ell] [ − 2 ℓ , 2 ℓ ] . Since ℓ < m / 2 \ell<m/2 ℓ < m /2 , this interval is contained in [ − m + 1 , m − 1 ] [-m+1,m-1] [ − m + 1 , m − 1 ] . Hence no reduction modulo F ( X n ) F(X^n) F ( X n ) is involved in this coefficient comparison.
The only contribution from c ˉ h \bar{c}_h c ˉ h to the coefficient of X ℓ + 1 X^{\ell+1} X ℓ + 1 comes from the pair β h , β − h \beta_h,\beta_{-h} β h , β − h . It is
β h σ h ( α ) + α σ ℓ + 1 ( β − h ) . \beta_h\sigma^h(\alpha)+\alpha\sigma^{\ell+1}(\beta_{-h}). β h σ h ( α ) + α σ ℓ + 1 ( β − h ) .
The central summand contributes through u ˉ h Z j ( α ) \bar{u}_h Z_j(\alpha) u ˉ h Z j ( α ) . Such a contribution can occur only when h ≡ 0 ( m o d n ) h \equiv0 \pmod{n} h ≡ 0 ( mod n ) . Let u h ∈ K u_h \in\mathbb{K} u h ∈ K denote the coefficient of X h X^h X h in u ˉ h \bar{u}_h u ˉ h , with the convention u h = 0 u_h = 0 u h = 0 if h ≢ 0 ( m o d n ) h \not\equiv0 \pmod{n} h ≡ 0 ( mod n ) . Since X ℓ + 1 X^{\ell+1} X ℓ + 1 lies outside the interval, we obtain
β h σ h ( α ) + α σ ℓ + 1 ( β − h ) + u h α = 0 , (28) \beta_h \sigma^h(\alpha) + \alpha\sigma^{\ell+1}(\beta_{-h}) + u_h \alpha= 0, \tag*{(28)} β h σ h ( α ) + α σ ℓ + 1 ( β − h ) + u h α = 0 , (28)
for every α ∈ L \alpha\in\mathbb{L} α ∈ L . Assume first that h ≢ 0 ( m o d n ) h \not\equiv0 \pmod{n} h ≡ 0 ( mod n ) . Then u h = 0 u_h = 0 u h = 0 . Taking α = 1 \alpha= 1 α = 1 in (28) gives
β h + σ ℓ + 1 ( β − h ) = 0. \beta_h + \sigma^{\ell+1}(\beta_{-h}) = 0. β h + σ ℓ + 1 ( β − h ) = 0.
Substituting this into (28) , we get
β h ( σ h ( α ) − α ) = 0 , \beta_h\left(\sigma^h(\alpha)-\alpha\right)=0, β h ( σ h ( α ) − α ) = 0 ,
for every α ∈ L \alpha\in\mathbb{L} α ∈ L . Since σ h ≠ id L \sigma^h \ne\operatorname{id}_{\mathbb{L}} σ h = id L , it follows that
β h = 0 , β − h = 0. \beta_h = 0,\qquad\beta_{-h} = 0. β h = 0 , β − h = 0.
Assume now that h ≡ 0 ( m o d n ) h \equiv0 \pmod{n} h ≡ 0 ( mod n ) . Then σ h = id L \sigma^h = \operatorname{id}_{\mathbb{L}} σ h = id L , and
ℓ + 1 = h + j . \ell+ 1 = h + j. ℓ + 1 = h + j .
Thus (28) gives
β h + σ j ( β − h ) + u h = 0. (29) \beta_h + \sigma^j(\beta_{-h}) + u_h = 0. \tag*{(29)} β h + σ j ( β − h ) + u h = 0. (29)
Since n ≥ 3 n \ge3 n ≥ 3 , the condition h ≡ 0 ( m o d n ) h \equiv0 \pmod{n} h ≡ 0 ( mod n ) implies h ≥ 3 h \ge3 h ≥ 3 , and hence j + 1 ≤ ℓ j+1 \le\ell j + 1 ≤ ℓ . Applying the same comparison to Z j + 1 ( α ) Z_{j+1}(\alpha) Z j + 1 ( α ) and to the coefficient of X ℓ + 2 X^{\ell+2} X ℓ + 2 , we obtain
β h + σ j + 1 ( β − h ) + u h = 0. (30) \beta_h + \sigma^{j+1}(\beta_{-h}) + u_h = 0. \tag*{(30)} β h + σ j + 1 ( β − h ) + u h = 0. (30)
Subtracting (29) from (30) gives
σ j + 1 ( β − h ) = σ j ( β − h ) , \sigma^{j+1}(\beta_{-h}) = \sigma^j(\beta_{-h}), σ j + 1 ( β − h ) = σ j ( β − h ) ,
and therefore
σ ( β − h ) = β − h . \sigma(\beta_{-h}) = \beta_{-h}. σ ( β − h ) = β − h .
Hence
β − h ∈ K . \beta_{-h} \in\mathbb{K}. β − h ∈ K .
Since u h ∈ K u_h \in\mathbb{K} u h ∈ K , equation (29) also gives
β h ∈ K . \beta_h \in\mathbb{K}. β h ∈ K .
Therefore
β h X h + β − h X − h ∈ E . \beta_h X^h + \beta_{-h}X^{-h} \in\mathbb{E}. β h X h + β − h X − h ∈ E .
We absorb this central pair into the central summand λ h \lambda_h λ h , and continue the induction. Iterating this argument for h = ℓ , ℓ − 1 , … , 1 h=\ell,\ell-1,\ldots,1 h = ℓ , ℓ − 1 , … , 1 , we obtain
b ˉ = β + λ \bar{b} = \beta+ \lambda b ˉ = β + λ
with β ∈ L \beta\in\mathbb{L} β ∈ L and λ ∈ E \lambda\in\mathbb{E} λ ∈ E .
Step 4: Determination of the central summand.
It remains to determine which central elements λ \lambda λ may occur. Since β ∈ L ⊆ L ( S ℓ ( F ) ) \beta\in\mathbb{L} \subseteq\mathcal{L}(\mathcal{S}_{\ell}(F)) β ∈ L ⊆ L ( S ℓ ( F )) and b ˉ ∈ L ( S ℓ ( F ) ) \bar{b} \in\mathcal{L}(\mathcal{S}_{\ell}(F)) b ˉ ∈ L ( S ℓ ( F )) , we have
λ = b ˉ − β ∈ L ( S ℓ ( F ) ) . \lambda= \bar{b} - \beta\in\mathcal{L}(\mathcal{S}_{\ell}(F)). λ = b ˉ − β ∈ L ( S ℓ ( F )) .
Because λ \lambda λ is central, for every Z ∈ S ℓ ( F ) Z \in\mathcal{S}_{\ell}(F) Z ∈ S ℓ ( F ) ,
Δ λ ( Z ) = λ Z + Z Θ ( λ ) = ( λ + Θ ( λ ) ) Z . \Delta_{\lambda}(Z) = \lambda Z + Z\Theta(\lambda) = (\lambda+ \Theta(\lambda))Z. Δ λ ( Z ) = λ Z + Z Θ ( λ ) = ( λ + Θ ( λ )) Z .
Set
u ‾ : = λ + Θ ( λ ) ∈ E . \overline{u} := \lambda+ \Theta(\lambda) \in E. u := λ + Θ ( λ ) ∈ E .
Since 1 + R F ( X n ) ∈ S ℓ ( F ) 1 + RF(X^n) \in\mathcal{S}_{\ell}(F) 1 + R F ( X n ) ∈ S ℓ ( F ) , we have
u ‾ = Δ λ ( 1 + R F ( X n ) ) ∈ S ℓ ( F ) . \overline{u} = \Delta_{\lambda}(1 + RF(X^n)) \in\mathcal{S}_{\ell}(F). u = Δ λ ( 1 + R F ( X n )) ∈ S ℓ ( F ) .
We claim that u ‾ ∈ K \overline{u} \in K u ∈ K .
Suppose, to the contrary, that u ‾ ∉ K \overline{u} \notin K u ∈ / K . Since u ‾ ∈ E ∩ S ℓ ( F ) \overline{u} \in E \cap\mathcal{S}_{\ell}(F) u ∈ E ∩ S ℓ ( F ) , its support consists only of central exponents lying in [ − ℓ , ℓ ] [-\ell,\ell] [ − ℓ , ℓ ] . Let t = ξ n > 0 t = \xi n > 0 t = ξ n > 0 be the largest positive central exponent occurring in monomials of u ‾ \overline{u} u , and let u t ∈ K × u_t \in K^{\times} u t ∈ K × be its coefficient. Then t ≤ ℓ t \leq\ell t ≤ ℓ . Put
j : = ℓ + 1 − t . j := \ell+ 1 - t. j := ℓ + 1 − t .
Then 1 ≤ j ≤ ℓ 1 \leq j \leq\ell 1 ≤ j ≤ ℓ , and hence Z j ( α ) ∈ S ℓ ( F ) Z_j(\alpha) \in\mathcal{S}_{\ell}(F) Z j ( α ) ∈ S ℓ ( F ) , for every α ∈ L \alpha\in L α ∈ L . The coefficient of X ℓ + 1 X^{\ell+1} X ℓ + 1 in u ‾ Z j ( α ) \overline{u}Z_j(\alpha) u Z j ( α ) is precisely u t α u_t\alpha u t α . Since
u ‾ Z j ( α ) = Δ λ ( Z j ( α ) ) ∈ S ℓ ( F ) , \overline{u}Z_j(\alpha) = \Delta_{\lambda}(Z_j(\alpha)) \in\mathcal{S}_{\ell}(F), u Z j ( α ) = Δ λ ( Z j ( α )) ∈ S ℓ ( F ) ,
and X ℓ + 1 X^{\ell+1} X ℓ + 1 lies outside the interval [ − ℓ , ℓ ] [-\ell,\ell] [ − ℓ , ℓ ] , this coefficient must be zero for every α ∈ L \alpha\in L α ∈ L . This forces u t = 0 u_t = 0 u t = 0 , a contradiction. Therefore u ‾ \overline{u} u has no positive nonconstant central term. Since u ‾ = Θ ( u ‾ ) \overline{u} = \Theta(\overline{u}) u = Θ ( u ) , it has no negative nonconstant central term either. Hence
u ‾ ∈ K . \overline{u} \in K. u ∈ K .
Thus λ + Θ ( λ ) ∈ K \lambda+ \Theta(\lambda) \in K λ + Θ ( λ ) ∈ K , that is,
λ ∈ B F . \lambda\in B_F. λ ∈ B F .
Consequently
b ‾ = β + λ ∈ L + B F . \overline{b} = \beta+ \lambda\in L + B_F. b = β + λ ∈ L + B F .
This proves
L ( S ℓ ( F ) ) ⊆ L + B F . \mathcal{L}(\mathcal{S}_{\ell}(F)) \subseteq L + B_F. L ( S ℓ ( F )) ⊆ L + B F .
Together with Proposition 5.3 , the theorem follows. □
Equivalence
Main results
The construction in § 4.1 is parametrized by monic irreducible self-reciprocal polynomials. Since many such polynomials exist in each prescribed even degree, a natural question is how many inequivalent members arise from this construction.
Our main goal in this section is to answer this question. We begin with the two main equivalence theorems—one for each parameter range—and their enumerative consequences for full Hermitian equivalence.
We first fix the notation used throughout this section:
L = F q n , K = F q , R = L [ X ; σ ] . L = \mathbb{F}_{q^n}, \qquad K = \mathbb{F}_q, \qquad R = L[X;\sigma]. L = F q n , K = F q , R = L [ X ; σ ] .
where σ \sigma σ generates Gal ( L / K ) \operatorname{Gal}(L/K) Gal ( L / K ) , and we assume that n > 2 n > 2 n > 2 . Let s ≥ 2 s \geq2 s ≥ 2 be even, and put
E : = F q s , F : = F q s / 2 . E := \mathbb{F}_{q^s}, \qquad F := \mathbb{F}_{q^{s/2}}. E := F q s , F := F q s /2 .
The nontrivial automorphism of the quadratic extension E / F E/F E / F is
ϱ : E ⟶ E , x ⟼ x q s / 2 . \varrho: E \longrightarrow E, \qquad x \longmapsto x^{q^{s/2}}. ϱ : E ⟶ E , x ⟼ x q s /2 .
Let F ( Y ) , G ( Y ) ∈ K [ Y ] F(Y),G(Y) \in K[Y] F ( Y ) , G ( Y ) ∈ K [ Y ] be monic irreducible self-reciprocal polynomials of degree s s s .
Choose roots u F , u G ∈ E u_F,u_G \in E u F , u G ∈ E of F F F and G G G , respectively. Since u H θ = u H − 1 u_H^{\theta}=u_H^{-1} u H θ = u H − 1 for H ∈ { F , G } H \in\{F,G\} H ∈ { F , G } , the elements
w H : = u H + u H − 1 w_H := u_H+u_H^{-1} w H := u H + u H − 1
belong to F \mathbb{F} F .
The next two theorems are the main results of this section about the linear Hermitian equivalence between S ℓ ( F ) \mathcal{S}_{\ell}(F) S ℓ ( F ) and S ℓ ( G ) \mathcal{S}_{\ell}(G) S ℓ ( G ) .
Theorem 6.1 (Short-range equivalence). Assume that
n > 2 and 1 ≤ ℓ < min { n , n s / 4 } . n>2 \qquad\text{and} \qquad1 \leq\ell< \min\{n,ns/4\}. n > 2 and 1 ≤ ℓ < min { n , n s /4 } . If S ℓ ( F ) \mathcal{S}_{\ell}(F) S ℓ ( F ) and S ℓ ( G ) \mathcal{S}_{\ell}(G) S ℓ ( G ) are linearly Hermitian equivalent, then there exists λ ∈ F q × \lambda\in\mathbb{F}_q^{\times} λ ∈ F q × such that
F ( Y ) = λ s G ( λ − 1 Y ) . F(Y)=\lambda^sG\left(\lambda^{-1}Y\right). F ( Y ) = λ s G ( λ − 1 Y ) .
Theorem 6.2 (Large-range equivalence). Assume that
n > 2 , s > 4 , n < ℓ < n s 4 . n>2,\qquad s>4,\qquad n<\ell<\frac{ns}{4}. n > 2 , s > 4 , n < ℓ < 4 n s . If S ℓ ( F ) \mathcal{S}_{\ell}(F) S ℓ ( F ) and S ℓ ( G ) \mathcal{S}_{\ell}(G) S ℓ ( G ) are linearly Hermitian equivalent, then there exist a , b , c , d ∈ K a,b,c,d \in K a , b , c , d ∈ K , with a d − b c ≠ 0 ad-bc \ne0 a d − b c = 0 , such that
w G = a w F + b c w F + d . w_G=\frac{aw_F+b}{cw_F+d}. w G = c w F + d a w F + b .
Since w F ∉ K w_F \notin K w F ∈ / K , the denominator c w F + d cw_F+d c w F + d is nonzero for every nonzero pair ( c , d ) ∈ K 2 (c,d) \in K^2 ( c , d ) ∈ K 2 . Hence the fractional-linear expression in Theorem 6.2 is well defined.
For the counting statement, let
I q , s s r : = { H ( Y ) ∈ F q [ Y ] : H is monic, irreducible, self-reciprocal, and deg ( H ) = s } , \mathcal{I}_{q,s}^{\mathrm{sr}}:=\{H(Y)\in\mathbb{F}_q[Y]:H\text{ is monic, irreducible, self-reciprocal, and }\deg(H)=s\}, I q , s sr := { H ( Y ) ∈ F q [ Y ] : H is monic, irreducible, self-reciprocal, and deg ( H ) = s } ,
and write
N q ( s ) : = ∣ I q , s s r ∣ . N_q(s):=\left|\mathcal{I}_{q,s}^{\mathrm{sr}}\right|. N q ( s ) := I q , s sr .
For an integer d d d with 1 ≤ d ≤ n 1 \leq d \leq n 1 ≤ d ≤ n , set
ℓ d : = s ( n − d + 1 ) − 2 2 . (31) \ell_d:=\frac{s(n-d+1)-2}{2}. \tag*{(31)} ℓ d := 2 s ( n − d + 1 ) − 2 . (31)
Thus S ℓ d ( H ) \mathcal{S}_{\ell_d}(H) S ℓ d ( H ) is the maximum additive Hermitian d d d -code constructed in Theorem 4.1. For the quadratic case, i.e., s = 2 s=2 s = 2 , put
L 2 ( q ) : = { q 2 , q even , q − 1 4 , q ≡ 1 ( m o d 4 ) , q + 1 4 , q ≡ 3 ( m o d 4 ) . L_2(q):=
\begin{cases}
\frac{q}{2}, & q\text{ even},\\
\frac{q-1}{4}, & q\equiv1\pmod{4},\\
\frac{q+1}{4}, & q\equiv3\pmod{4}.
\end{cases} L 2 ( q ) := ⎩ ⎨ ⎧ 2 q , 4 q − 1 , 4 q + 1 , q even , q ≡ 1 ( mod 4 ) , q ≡ 3 ( mod 4 ) .
Corollary 6.3 (Hermitian enumerative bounds). Write q = p h q=p^h q = p h and assume that n > 2 n>2 n > 2 . For each even integer s ≥ 2 s\geq2 s ≥ 2 , let
Λ s , d : = { S ℓ d ( H ) : H ∈ I q , s s r } . \Lambda_{s,d}:=\{\mathcal{S}_{\ell_d}(H):H\in\mathcal{I}_{q,s}^{\mathrm{sr}}\}. Λ s , d := { S ℓ d ( H ) : H ∈ I q , s sr } . The following lower bounds hold.
If s = 2 s=2 s = 2 and n / 2 < d < n n/2<d<n n /2 < d < n , then Λ s , d \Lambda_{s,d} Λ s , d contains at least
⌊ L 2 ( q ) 2 h ⌋ \left\lfloor\frac{L_2(q)}{2h} \right\rfloor ⌊ 2 h L 2 ( q ) ⌋
pairwise Hermitian inequivalent maximum additive Hermitian d d d -codes in H n ( F q 2 / F q ) \mathcal{H}_n(\mathbb{F}_{q^2}/\mathbb{F}_q) H n ( F q 2 / F q ) .
If s ≥ 4 s\geq4 s ≥ 4 is even and ℓ d < n \ell_d<n ℓ d < n , then Λ s , d \Lambda_{s,d} Λ s , d contains at least
⌊ N q ( s ) s h gcd ( s , q − 1 ) ⌋ \left\lfloor\frac{N_q(s)}{sh\gcd(s,q-1)} \right\rfloor ⌊ s h g cd( s , q − 1 ) N q ( s ) ⌋
pairwise Hermitian inequivalent maximum additive Hermitian d d d -codes in H n ( F q s / F q s / 2 ) \mathcal{H}_n(\mathbb{F}_{q^s}/\mathbb{F}_{q^{s/2}}) H n ( F q s / F q s /2 ) .
If s > 4 s>4 s > 4 is even and
n < ℓ d < n s 4 , n<\ell_d<\frac{ns}{4}, n < ℓ d < 4 n s ,
then Λ s , d \Lambda_{s,d} Λ s , d contains at least
⌊ N q ( s ) s h q ( q 2 − 1 ) ⌋ \left\lfloor\frac{N_q(s)}{shq(q^2-1)} \right\rfloor ⌊ s h q ( q 2 − 1 ) N q ( s ) ⌋
pairwise Hermitian inequivalent maximum additive Hermitian d d d -codes in H n ( F q s / F q s / 2 ) \mathcal{H}_n(\mathbb{F}_{q^s}/\mathbb{F}_{q^{s/2}}) H n ( F q s / F q s /2 ) .
Since N q ( s ) = Θ ( q s / 2 ) N_q(s)=\Theta(q^{s/2}) N q ( s ) = Θ ( q s /2 ) as q → ∞ q\to\infty q → ∞ , the three parts have orders at least q / h q/h q / h , q s / 2 / h q^{s/2}/h q s /2 / h , and q s / 2 − 3 / h q^{s/2-3}/h q s /2 − 3 / h , respectively, for fixed admissible s s s . The sharper linearly Hermitian bounds used to prove this corollary are stated later in Subsection 6.7, where the relevant parameter-orbit arguments occur.
Proof strategy
For H ∈ { F , G } H\in\{F,G\} H ∈ { F , G } , put
R H : = R / R H ( X n ) . R_H:=R/R_H(X^n). R H := R / R H ( X n ) .
For F ≠ G F\neq G F = G , the codes S ℓ ( F ) ⊆ R F \mathcal{S}_{\ell}(F)\subseteq R_F S ℓ ( F ) ⊆ R F and S ℓ ( G ) ⊆ R G \mathcal{S}_{\ell}(G)\subseteq R_G S ℓ ( G ) ⊆ R G lie in different quotient algebras, so a direct comparison is inconvenient. To place them in a common setting, Theorem 6.4 embeds each quotient R H R_H R H , for H ∈ { F , G } H\in\{F,G\} H ∈ { F , G } , into End E ( V ) \operatorname{End}_{\mathbb{E}}(V) End E ( V ) . More precisely, it provides an irreducible σ \sigma σ -semilinear operator T H T_H T H satisfying T H n = u H id V T_H^n=u_H\operatorname{id}_V T H n = u H id V and an E \mathbb{E} E -algebra isomorphism Ψ H : R H → End E ( V ) \Psi_H:R_H\to\operatorname{End}_{\mathbb{E}}(V) Ψ H : R H → End E ( V ) that sends the class of X X X to T H T_H T H .
We write
C ℓ ( H ) : = Ψ H ( S ℓ ( H ) ) and B : = { λ ∈ E : λ + λ ϱ ∈ K } . \mathcal{C}_{\ell}(H):=\Psi_H(\mathcal{S}_{\ell}(H)) \qquad\text{and}\qquad\mathcal{B}:=\{\lambda\in\mathbb{E}:\lambda+\lambda^{\varrho}\in K\}. C ℓ ( H ) := Ψ H ( S ℓ ( H )) and B := { λ ∈ E : λ + λ ϱ ∈ K } .
The involutions induced by F F F and G G G on the common endomorphism algebra need not coincide. Thus a linear Hermitian equivalence does not immediately have the form of a single Hermitian congruence inside End E ( V ) \operatorname{End}_{\mathbb{E}}(V) End E ( V ) . Instead, transporting the equivalence through separate identifications with the standard Hermitian matrix space produces an E \mathbb{E} E -algebra automorphism
η ∈ Aut E ( End E ( V ) ) \eta\in\operatorname{Aut}_{\mathbb{E}}(\operatorname{End}_{\mathbb{E}}(V)) η ∈ Aut E ( End E ( V ))
and an invertible element C ∈ End E ( V ) × C\in\operatorname{End}_{\mathbb{E}}(V)^{\times} C ∈ End E ( V ) × such that
C ℓ ( G ) = η ( C ℓ ( F ) ) C . \mathcal{C}_{\ell}(G)=\eta(\mathcal{C}_{\ell}(F))C. C ℓ ( G ) = η ( C ℓ ( F )) C .
The Lyapunov invariant then gives
η ( L + B ) = L + B . \eta(\mathcal{L}+\mathcal{B})=\mathcal{L}+\mathcal{B}. η ( L + B ) = L + B .
The proofs of both main equivalence theorems begin with the same chain of reductions:
linear Hermitian equivalence of S ℓ ( F ) and S ℓ ( G ) \text{linear Hermitian equivalence of }\mathcal{S}_{\ell}(F)\text{ and }\mathcal{S}_{\ell}(G) linear Hermitian equivalence of S ℓ ( F ) and S ℓ ( G )
common semilinear realization
linear Hermitian equivalence of C ℓ ( F ) and C ℓ ( G ) \text{linear Hermitian equivalence of }\mathcal{C}_{\ell}(F)\text{ and }\mathcal{C}_{\ell}(G) linear Hermitian equivalence of C ℓ ( F ) and C ℓ ( G )
Lyapunov invariance
Q ( L + B ) Q − 1 = L + B Q(\mathcal{L}+B)Q^{-1}=\mathcal{L}+B Q ( L + B ) Q − 1 = L + B
field normalization
Q L Q − 1 = L . Q\mathcal{L}Q^{-1}=\mathcal{L}. Q L Q − 1 = L .
The semilinear realization is established in Theorem 6.4 and Proposition 6.5 . The Lyapunov computation gives L ( C ℓ ( H ) ) = L + B \mathcal{L}(\mathcal{C}_{\ell}(H))=\mathcal{L}+B L ( C ℓ ( H )) = L + B , while Theorem 6.9 shows that stabilizing L + B \mathcal{L}+B L + B already forces normalization of L \mathcal{L} L .
After this common reduction, the proof divides according to the value of ℓ \ell ℓ :
1 ≤ ℓ < min { n , n s / 4 } n < ℓ < n s / 4 1\leq\ell<\min\{n,ns/4\}
\qquad\qquad
n<\ell<ns/4 1 ≤ ℓ < min { n , n s /4 } n < ℓ < n s /4
L-linear span
⟨ T H − ℓ , … , 1 , … , T H ℓ ⟩ L \left\langle T_H^{-\ell},\ldots,1,\ldots,T_H^{\ell}\right\rangle_{\mathcal{L}} ⟨ T H − ℓ , … , 1 , … , T H ℓ ⟩ L
degree-zero component
⟨ 1 , w H , … , w H ⌊ ℓ / n ⌋ ⟩ L \left\langle1,w_H,\ldots,w_H^{\lfloor\ell/n\rfloor}\right\rangle_{\mathcal{L}} ⟨ 1 , w H , … , w H ⌊ ℓ / n ⌋ ⟩ L
short power-space classification
F ( Y ) = λ s G ( λ − 1 Y ) F(Y)=\lambda^sG(\lambda^{-1}Y) F ( Y ) = λ s G ( λ − 1 Y )
central interval determination
w G = a w F + b c w F + d w_G=\frac{aw_F+b}{cw_F+d} w G = c w F + d a w F + b
In the short range, Lemma 6.11 recovers the full interval of powers and Proposition 6.12 yields the scaling relation in Theorem 6.1 . In the large range, Lemmas 6.15 and 6.14 recover the fractional-linear relation between w F w_F w F and w G w_G w G , proving Theorem 6.2 .
Finally, the linearly Hermitian enumeration results follow by bounding the size of the relevant parameter orbits: the quadratic scaling orbit is governed by c ↦ − c c\mapsto-c c ↦ − c on Y 2 + c Y + 1 Y^2+cY+1 Y 2 + c Y + 1 , the short-range scaling orbit has size at most gcd ( s , q − 1 ) \gcd(s,q-1) g cd( s , q − 1 ) , and a self-reciprocal PGL 2 ( F q ) \operatorname{PGL}_2(\mathbb{F}_q) PGL 2 ( F q ) -orbit has size at most q ( q 2 − 1 ) q(q^2-1) q ( q 2 − 1 ) . Passing from linear to full Hermitian equivalence introduces at most ∣ Aut ( F q s ) ∣ = s h |\operatorname{Aut}(\mathbb{F}_{q^s})|=sh ∣ Aut ( F q s ) ∣ = s h further identifications (and 2 h 2h 2 h when s = 2 s=2 s = 2 ). This gives the three parts of Corollary 6.3.
A common semilinear realization
Although S ℓ ( F ) ⊆ R F \mathcal{S}_{\ell}(F)\subseteq R_F S ℓ ( F ) ⊆ R F and S ℓ ( G ) ⊆ R G \mathcal{S}_{\ell}(G)\subseteq R_G S ℓ ( G ) ⊆ R G lie in different central simple algebras, the following construction places both codes in a single endomorphism algebra. We first record the required semilinear terminology.
Let V V V be a finite-dimensional L \mathcal{L} L -vector space. For α ∈ L \alpha\in\mathcal{L} α ∈ L , we denote by
m α : V ⟶ V , v ⟼ α v , m_{\alpha}:V\longrightarrow V,\qquad v\longmapsto\alpha v, m α : V ⟶ V , v ⟼ α v ,
the L \mathbb{L} L -linear scalar multiplication by α \alpha α . A map T : V ⟶ V T: V \longrightarrow V T : V ⟶ V is called σ \sigma σ -semilinear if it is additive and
T ( α v ) = σ ( α ) T ( v ) T(\alpha v)=\sigma(\alpha)T(v) T ( α v ) = σ ( α ) T ( v )
for every α ∈ L \alpha\in\mathbb{L} α ∈ L , v ∈ V v \in V v ∈ V .
In this paper, semilinear operators are always assumed to be bijective. If T T T is σ \sigma σ -semilinear, then T i T^i T i is σ i \sigma^i σ i -semilinear for every i ≥ 0 i \ge0 i ≥ 0 . Since σ \sigma σ has order n n n , the operator T n T^n T n is L \mathbb{L} L -linear. Moreover, T n T = T T n T^nT=T T^n T n T = T T n and T n m α = m α T n T^n m_\alpha=m_\alpha T^n T n m α = m α T n , for every α ∈ L \alpha\in\mathbb{L} α ∈ L . Hence K [ T n ] \mathbb{K}[T^n] K [ T n ] is a central subalgebra of the K \mathbb{K} K -algebra generated by m α m_\alpha m α and T T T . This is the semilinear counterpart of the fact that X n X^n X n is central in the skew-polynomial ring L [ X ; σ ] \mathbb{L}[X;\sigma] L [ X ; σ ] .
An L \mathbb{L} L -subspace W ≤ V W \le V W ≤ V is said to be T T T -invariant if T ( W ) ⊆ W T(W) \subseteq W T ( W ) ⊆ W . Since T T T is bijective, this is equivalent to T ( W ) = W T(W)=W T ( W ) = W . We say that T T T is irreducible if the only T T T -invariant L \mathbb{L} L -subspaces of V V V are { 0 } \{0\} { 0 } and V V V .
Set
r : = gcd ( n , s ) , n 0 : = n r , s 0 : = s r . r:=\gcd(n,s), \qquad n_0:=\frac{n}{r}, \qquad s_0:=\frac{s}{r}. r := g cd( n , s ) , n 0 := r n , s 0 := r s .
and put
Ω : = F q n s / r . \Omega:=\mathbb{F}_{q^{ns/r}}. Ω := F q n s / r .
We shall work on the fixed additive group
We endow V V V with the following L \mathbb{L} L -vector space structure:
α ⋅ ( x 0 , … , x r − 1 ) = ( α x 0 , σ ( α ) x 1 , … , σ r − 1 ( α ) x r − 1 ) , α ∈ L . \alpha\cdot(x_0,\ldots,x_{r-1})=(\alpha x_0,\sigma(\alpha)x_1,\ldots,\sigma^{r-1}(\alpha)x_{r-1}), \qquad\alpha\in\mathbb{L}. α ⋅ ( x 0 , … , x r − 1 ) = ( α x 0 , σ ( α ) x 1 , … , σ r − 1 ( α ) x r − 1 ) , α ∈ L .
Then
dim L V = r [ Ω : L ] = r s 0 = s . \dim_{\mathbb{L}}V=r[\Omega:\mathbb{L}]=rs_0=s. dim L V = r [ Ω : L ] = r s 0 = s .
Now let H ( Y ) ≠ Y H(Y)\ne Y H ( Y ) = Y be a monic irreducible polynomial in K [ Y ] \mathbb{K}[Y] K [ Y ] of degree s s s , and choose a root u H ∈ E = F q s u_H\in\mathbb{E}=\mathbb{F}_{q^s} u H ∈ E = F q s of H H H . For u ∈ E ⊆ Ω u\in\mathbb{E}\subseteq\Omega u ∈ E ⊆ Ω , it is clear that
u id V : V ⟶ V ( x 0 , … , x r − 1 ) ⟼ ( u x 0 , … , u x r − 1 ) . u\operatorname{id}_V:V\longrightarrow V \qquad(x_0,\ldots,x_{r-1})\longmapsto(ux_0,\ldots,ux_{r-1}). u id V : V ⟶ V ( x 0 , … , x r − 1 ) ⟼ ( u x 0 , … , u x r − 1 ) .
The important point is that the underlying L \mathbb{L} L -vector space V V V is fixed independently of the defining polynomial H H H . Moreover, if T H T_H T H is chosen so that T H n = u H id V T_H^n=u_H\operatorname{id}_V T H n = u H id V , then
E T H : = K [ T H n ] = K [ u H id V ] \mathbb{E}_{T_H}:=\mathbb{K}[T_H^n]=\mathbb{K}[u_H\operatorname{id}_V] E T H := K [ T H n ] = K [ u H id V ]
is a field of endomorphisms of V V V , naturally isomorphic to
K ( u H ) = E . \mathbb{K}(u_H)=\mathbb{E}. K ( u H ) = E .
Thus V V V becomes a vector space over E T H \mathbb{E}_{T_H} E T H , with dim E T H V = n \dim_{\mathbb{E}_{T_H}}V=n dim E T H V = n .
The following result is proved in [17 Theorem 3.6] .
Theorem 6.4. Let H ( Y ) ≠ Y H(Y)\ne Y H ( Y ) = Y be a monic irreducible polynomial in K [ Y ] \mathbb{K}[Y] K [ Y ] of degree s s s , and let u H ∈ E u_H\in\mathbb{E} u H ∈ E be a root of H H H . Then there exists an irreducible σ \sigma σ -semilinear operator
T H : V ⟶ V T_H:V\longrightarrow V T H : V ⟶ V on the s s s -dimensional L \mathbb{L} L -vector space V = Ω r V=\Omega^r V = Ω r such that T H n = u H id V T_H^n=u_H\operatorname{id}_V T H n = u H id V . Moreover, the map
Ψ H : R H ⟶ End E T H ( V ) \Psi_H:R_H\longrightarrow\operatorname{End}_{\mathbb{E}_{T_H}}(V) Ψ H : R H ⟶ End E T H ( V )
defined by
Ψ H ( ∑ i = 0 n s − 1 α i X i + R H ( X n ) ) = ∑ i = 0 n s − 1 m α i T H i , α i ∈ L , \Psi_H\left(\sum_{i=0}^{n_s-1}\alpha_iX^i+R_H(X^n)\right)=\sum_{i=0}^{n_s-1}m_{\alpha_i}T_H^i,\qquad\alpha_i\in\mathbb{L}, Ψ H ( i = 0 ∑ n s − 1 α i X i + R H ( X n ) ) = i = 0 ∑ n s − 1 m α i T H i , α i ∈ L ,
is an E \mathbb{E} E -algebra isomorphism and an L \mathbb{L} L -vector-space isomorphism.
In what follows, we shall write α i T H i \alpha_iT_H^i α i T H i for m α i T H i m_{\alpha_i}T_H^i m α i T H i , by abuse of notation.
For H ∈ { F , G } H\in\{F,G\} H ∈ { F , G } , fix a root u H ∈ E u_H\in\mathbb{E} u H ∈ E of H H H and a map Ψ H \Psi_H Ψ H as in Theorem 6.4 . Since H H H is self-reciprocal of even degree, one has u H ϱ = u H − 1 u_H^{\varrho}=u_H^{-1} u H ϱ = u H − 1 .
Define
C ℓ ( H ) : = Ψ H ( S ℓ ( H ) ) ⊆ End E ( V ) . \mathcal{C}_{\ell}(H):=\Psi_H(\mathcal{S}_{\ell}(H))\subseteq\operatorname{End}_{\mathbb{E}}(V). C ℓ ( H ) := Ψ H ( S ℓ ( H )) ⊆ End E ( V ) .
Then
C ℓ ( H ) = { α 0 + ∑ i = 1 ℓ ( σ − i ( α i ) T H − i + α i T H i ) : α 0 , … , α ℓ ∈ L } . (32) \mathcal{C}_{\ell}(H)=\left\{\alpha_0+\sum_{i=1}^{\ell}\left(\sigma^{-i}(\alpha_i)T_H^{-i}+\alpha_iT_H^i\right):\alpha_0,\ldots,\alpha_{\ell}\in\mathbb{L}\right\}. \tag*{(32)} C ℓ ( H ) = { α 0 + i = 1 ∑ ℓ ( σ − i ( α i ) T H − i + α i T H i ) : α 0 , … , α ℓ ∈ L } . (32)
Let
Θ H : = Ψ H ∘ Θ ∘ Ψ H − 1 . \Theta_H:=\Psi_H\circ\Theta\circ\Psi_H^{-1}. Θ H := Ψ H ∘ Θ ∘ Ψ H − 1 .
Then Θ H \Theta_H Θ H is an anti-automorphism of End E ( V ) \operatorname{End}_{\mathbb{E}}(V) End E ( V ) , and
Θ H ( α ) = α for every α ∈ L , Θ H ( T H ) = T H − 1 . (33) \Theta_H(\alpha)=\alpha\quad\text{for every }\alpha\in\mathbb{L},\qquad\Theta_H(T_H)=T_H^{-1}. \tag*{(33)} Θ H ( α ) = α for every α ∈ L , Θ H ( T H ) = T H − 1 . (33)
Moreover, Θ H \Theta_H Θ H restricts to ϱ \varrho ϱ on the center E id V \mathbb{E}\operatorname{id}_V E id V . Indeed,
Θ H ( u H id V ) = Θ H ( T H n ) = T H − n = u H − 1 id V = u H ϱ id V . \Theta_H(u_H\operatorname{id}_V)=\Theta_H(T_H^n)=T_H^{-n}=u_H^{-1}\operatorname{id}_V=u_H^{\varrho}\operatorname{id}_V. Θ H ( u H id V ) = Θ H ( T H n ) = T H − n = u H − 1 id V = u H ϱ id V .
Thus Θ H \Theta_H Θ H is a unitary involution of End E ( V ) \operatorname{End}_{\mathbb{E}}(V) End E ( V ) , and C ℓ ( H ) ⊆ Fix ( Θ H ) \mathcal{C}_{\ell}(H)\subseteq\operatorname{Fix}(\Theta_H) C ℓ ( H ) ⊆ Fix ( Θ H ) .
We next show that linear Hermitian equivalence is preserved when one passes from the skew-polynomial model to the common semilinear model. This justifies working from now on in End E ( V ) \operatorname{End}_{\mathbb{E}}(V) End E ( V ) .
Let
M H : R H ⟶ ∼ M n ( E ) , H ∈ { F , G } , \mathcal{M}_H:R_H\overset{\sim}{\longrightarrow}M_n(\mathbb{E}),\qquad H\in\{F,G\}, M H : R H ⟶ ∼ M n ( E ) , H ∈ { F , G } ,
be E \mathbb{E} E -algebra isomorphisms identifying the unitary involution on R H R_H R H with the standard Hermitian adjoint A ↦ A ∗ : = ( A ϱ ) T A\mapsto A^*:=(A^{\varrho})^{\mathsf{T}} A ↦ A ∗ := ( A ϱ ) T on M n ( E ) M_n(\mathbb{E}) M n ( E ) . Likewise, let
N H : End E ( V ) ⟶ ∼ M n ( E ) , H ∈ { F , G } , \mathcal{N}_H:\operatorname{End}_{\mathbb{E}}(V)\overset{\sim}{\longrightarrow}M_n(\mathbb{E}),\qquad H\in\{F,G\}, N H : End E ( V ) ⟶ ∼ M n ( E ) , H ∈ { F , G } ,
be E \mathbb{E} E -algebra isomorphisms identifying the unitary involution Θ H \Theta_H Θ H on End E ( V ) \operatorname{End}_{\mathbb{E}}(V) End E ( V ) with the same adjoint A ↦ A ∗ A\mapsto A^* A ↦ A ∗ . For convenience, we say that
two codes S ℓ ( F ) \mathcal{S}_{\ell}(F) S ℓ ( F ) and S ℓ ( G ) \mathcal{S}_{\ell}(G) S ℓ ( G ) are linearly Hermitian equivalent if M F ( S ℓ ( F ) ) \mathcal{M}_F(\mathcal{S}_{\ell}(F)) M F ( S ℓ ( F )) and M G ( S ℓ ( G ) ) \mathcal{M}_G(\mathcal{S}_{\ell}(G)) M G ( S ℓ ( G )) are linearly Hermitian equivalent;
the codes C ℓ ( F ) \mathcal{C}_{\ell}(F) C ℓ ( F ) and C ℓ ( G ) \mathcal{C}_{\ell}(G) C ℓ ( G ) are linearly Hermitian equivalent if N F ( C ℓ ( F ) ) \mathcal{N}_F(\mathcal{C}_{\ell}(F)) N F ( C ℓ ( F )) and N G ( C ℓ ( G ) ) \mathcal{N}_G(\mathcal{C}_{\ell}(G)) N G ( C ℓ ( G )) are linearly Hermitian equivalent.
Proposition 6.5. The codes S ℓ ( F ) \mathcal{S}_{\ell}(F) S ℓ ( F ) and S ℓ ( G ) \mathcal{S}_{\ell}(G) S ℓ ( G ) are linearly Hermitian equivalent if and only if the codes C ℓ ( F ) \mathcal{C}_{\ell}(F) C ℓ ( F ) and C ℓ ( G ) \mathcal{C}_{\ell}(G) C ℓ ( G ) are linearly Hermitian equivalent.
Proof. For H ∈ { F , G } H \in\{F,G\} H ∈ { F , G } , the map Ψ H : R H ⟶ End E ( V ) \Psi_H : R_H \longrightarrow\operatorname{End}_{\mathbb{E}}(V) Ψ H : R H ⟶ End E ( V ) is an E \mathbb{E} E -algebra isomorphism and maps S ℓ ( H ) S_\ell(H) S ℓ ( H ) onto C ℓ ( H ) \mathcal{C}_\ell(H) C ℓ ( H ) . For each H ∈ { F , G } H \in\{F,G\} H ∈ { F , G } , consider the commutative diagram
R H → Ψ H End E ( V ) M H ↓ ↓ N H M n ( E ) → Φ H M n ( E ) \begin{CD}
R_H @>{\Psi_H}>> \operatorname{End}_{\mathbb{E}}(V) \\
@V{\mathcal{M}_H}VV @VV{\mathcal{N}_H}V \\
M_n(\mathbb{E}) @>{\Phi_H}>> M_n(\mathbb{E})
\end{CD} R H M H ↓ ⏐ M n ( E ) Ψ H Φ H End E ( V ) ↓ ⏐ N H M n ( E )
where
Φ H : = N H ∘ Ψ H ∘ M H − 1 . \Phi_H := \mathcal{N}_H \circ\Psi_H \circ\mathcal{M}_H^{-1}. Φ H := N H ∘ Ψ H ∘ M H − 1 .
Then Φ H \Phi_H Φ H is an E \mathbb{E} E -algebra automorphism of M n ( E ) M_n(\mathbb{E}) M n ( E ) . By the Skolem–Noether theorem, there exists P H ∈ GL n ( E ) P_H \in\operatorname{GL}_n(\mathbb{E}) P H ∈ GL n ( E ) such that
Φ H ( A ) = P H A P H − 1 for every A ∈ M n ( E ) . \Phi_H(A) = P_H A P_H^{-1} \qquad\text{for every } A \in M_n(\mathbb{E}). Φ H ( A ) = P H A P H − 1 for every A ∈ M n ( E ) .
Since M H \mathcal{M}_H M H , Ψ H \Psi_H Ψ H , and N H \mathcal{N}_H N H all respect the corresponding unitary involutions, the automorphism Φ H \Phi_H Φ H commutes with A ↦ A ∗ A \mapsto A^* A ↦ A ∗ . Hence
( P H A P H − 1 ) ∗ = P H A ∗ P H − 1 for every A ∈ M n ( E ) . (P_H A P_H^{-1})^* = P_H A^* P_H^{-1} \qquad\text{for every } A \in M_n(\mathbb{E}). ( P H A P H − 1 ) ∗ = P H A ∗ P H − 1 for every A ∈ M n ( E ) .
Equivalently,
A P H ∗ P H = P H ∗ P H A for every A ∈ M n ( E ) . A P_H^* P_H = P_H^* P_H A \qquad\text{for every } A \in M_n(\mathbb{E}). A P H ∗ P H = P H ∗ P H A for every A ∈ M n ( E ) .
Therefore
P H ∗ P H = c H I n P_H^* P_H = c_H I_n P H ∗ P H = c H I n
for some c H ∈ E × c_H \in\mathbb{E}^{\times} c H ∈ E × . Since P H ∗ P H P_H^*P_H P H ∗ P H is Hermitian, we have c H ϱ = c H c_H^\varrho= c_H c H ϱ = c H , and so
c H ∈ F × . c_H \in\mathbb{F}^{\times}. c H ∈ F × .
Thus
P H − 1 = c H − 1 P H ∗ . P_H^{-1} = c_H^{-1}P_H^*. P H − 1 = c H − 1 P H ∗ .
Consequently,
N H ( C ℓ ( H ) ) = c H − 1 P H M H ( S ℓ ( H ) ) P H ∗ . (34) \mathcal{N}_H(\mathcal{C}_\ell(H)) = c_H^{-1}P_H\mathcal{M}_H(\mathcal{S}_\ell(H))P_H^*. \tag*{(34)} N H ( C ℓ ( H )) = c H − 1 P H M H ( S ℓ ( H )) P H ∗ . (34)
Assume first that S ℓ ( F ) \mathcal{S}_\ell(F) S ℓ ( F ) and S ℓ ( G ) \mathcal{S}_\ell(G) S ℓ ( G ) are linearly Hermitian equivalent. Then there exist U ∈ GL n ( E ) U \in\operatorname{GL}_n(\mathbb{E}) U ∈ GL n ( E ) and λ ∈ F × \lambda\in\mathbb{F}^{\times} λ ∈ F × such that
M G ( S ℓ ( G ) ) = λ U M F ( S ℓ ( F ) ) U ∗ . \mathcal{M}_G(\mathcal{S}_\ell(G)) = \lambda U\mathcal{M}_F(\mathcal{S}_\ell(F))U^*. M G ( S ℓ ( G )) = λ U M F ( S ℓ ( F )) U ∗ .
Using (34) for H = F H=F H = F and H = G H=G H = G , we get
N G ( C ℓ ( G ) ) = λ c F c G ( P G U P F − 1 ) N F ( C ℓ ( F ) ) ( P G U P F − 1 ) ∗ . \mathcal{N}_G(\mathcal{C}_\ell(G)) = \lambda\frac{c_F}{c_G}(P_GUP_F^{-1})\mathcal{N}_F(\mathcal{C}_\ell(F))(P_GUP_F^{-1})^*. N G ( C ℓ ( G )) = λ c G c F ( P G U P F − 1 ) N F ( C ℓ ( F )) ( P G U P F − 1 ) ∗ .
Since λ c F c G ∈ F × \lambda\frac{c_F}{c_G} \in\mathbb{F}^{\times} λ c G c F ∈ F × , this proves that C ℓ ( F ) \mathcal{C}_\ell(F) C ℓ ( F ) and C ℓ ( G ) \mathcal{C}_\ell(G) C ℓ ( G ) are linearly Hermitian equivalent. The converse is obtained by applying the same argument to Ψ F − 1 \Psi_F^{-1} Ψ F − 1 and Ψ G − 1 \Psi_G^{-1} Ψ G − 1 .
The Lyapunov invariant and its stabilizer
We now translate the Lyapunov invariant to the common semilinear model.
When the Lyapunov invariant is applied to C ℓ ( H ) \mathcal{C}_{\ell}(H) C ℓ ( H ) , it is computed with respect to the involution Θ H \Theta_{H} Θ H . Thus, temporarily writing the involution as a subscript,
L Θ H ( C ℓ ( H ) ) : = { B ∈ End E ( V ) : B A + A Θ H ( B ) ∈ C ℓ ( H ) for every A ∈ C ℓ ( H ) } . \mathcal{L}_{\Theta_{H}}\left(\mathcal{C}_{\ell}(H)\right):=\left\{B\in\operatorname{End}_{\mathbb{E}}(V):BA+A\Theta_{H}(B)\in\mathcal{C}_{\ell}(H)\text{ for every }A\in\mathcal{C}_{\ell}(H)\right\}. L Θ H ( C ℓ ( H ) ) := { B ∈ End E ( V ) : B A + A Θ H ( B ) ∈ C ℓ ( H ) for every A ∈ C ℓ ( H ) } .
We shall continue to omit the subscript whenever the involution is clear from the code under consideration.
Recall that
B = { λ ∈ E : λ + λ ϱ ∈ K } . \mathcal{B}=\{\lambda\in\mathbb{E}:\lambda+\lambda^{\varrho}\in\mathbb{K}\}. B = { λ ∈ E : λ + λ ϱ ∈ K } .
Throughout, we identify E \mathbb{E} E with the scalar subfield E id V ⊆ End E ( V ) \mathbb{E}\operatorname{id}_{V}\subseteq\operatorname{End}_{\mathbb{E}}(V) E id V ⊆ End E ( V ) , and we identify L \mathbb{L} L with the copy m L : = { m α : α ∈ L } ⊆ End E ( V ) \mathfrak{m}_{\mathbb{L}}:=\{m_{\alpha}:\alpha\in\mathbb{L}\}\subseteq\operatorname{End}_{\mathbb{E}}(V) m L := { m α : α ∈ L } ⊆ End E ( V ) coming from the fixed L \mathbb{L} L -vector space structure on V V V . By abuse of notation, we simply write L \mathbb{L} L for m L \mathfrak{m}_{\mathbb{L}} m L .
Proposition 6.6. Assume n > 2 n>2 n > 2 and 1 ≤ ℓ < n s / 4 1\leq\ell<ns/4 1 ≤ ℓ < n s /4 . For every H ∈ { F , G } H\in\{F,G\} H ∈ { F , G } , one has
L ( C ℓ ( H ) ) = L + B . \mathcal{L}\left(\mathcal{C}_{\ell}(H)\right)=\mathbb{L}+\mathcal{B}. L ( C ℓ ( H ) ) = L + B .
Proof. The assertion follows by applying Ψ H \Psi_{H} Ψ H to Theorem 5.4 . Under Ψ H \Psi_{H} Ψ H , the coefficient field L ⊆ R H \mathbb{L}\subseteq R_{H} L ⊆ R H is identified with the copy { m α : α ∈ L } ⊆ End E ( V ) \{m_{\alpha}:\alpha\in\mathbb{L}\}\subseteq\operatorname{End}_{\mathbb{E}}(V) { m α : α ∈ L } ⊆ End E ( V ) . Moreover, the space
B H = { λ ∈ E F : λ + Θ ( λ ) ∈ K } \mathcal{B}_{H}=\{\lambda\in\mathbb{E}_{F}:\lambda+\Theta(\lambda)\in\mathbb{K}\} B H = { λ ∈ E F : λ + Θ ( λ ) ∈ K }
is identified with B \mathcal{B} B , because Θ H \Theta_{H} Θ H restricts to ϱ \varrho ϱ on the scalar field E id V \mathbb{E}\operatorname{id}_{V} E id V .
We next determine the stabilizer of L + B \mathbb{L}+\mathcal{B} L + B in GL E ( V ) \operatorname{GL}_{\mathbb{E}}(V) GL E ( V ) . We start with a preliminary lemma.
Lemma 6.7. The map
E ⊗ K L ⟶ End E ( V ) , e ⊗ α ⟼ e α , \mathbb{E}\otimes_{\mathbb{K}}\mathbb{L}\longrightarrow\operatorname{End}_{\mathbb{E}}(V),\qquad e\otimes\alpha\longmapsto e\alpha, E ⊗ K L ⟶ End E ( V ) , e ⊗ α ⟼ e α , is injective. Also, E ∩ L = K \mathbb{E}\cap\mathbb{L}=\mathbb{K} E ∩ L = K .
Proof. Fix H ∈ { F , G } H\in\{F,G\} H ∈ { F , G } , let u H ∈ E u_{H}\in\mathbb{E} u H ∈ E be a root of H H H , and write
z H : = X n + R H ( X n ) ∈ Z ( R H ) . z_{H}:=X^{n}+R_{H}(X^{n})\in Z(R_{H}). z H := X n + R H ( X n ) ∈ Z ( R H ) .
Via the map
Ψ H : R H ⟶ End E ( V ) , \Psi_{H}:R_{H}\longrightarrow\operatorname{End}_{\mathbb{E}}(V), Ψ H : R H ⟶ End E ( V ) ,
the center Z ( R H ) ≅ K [ X n ] / ( H ( X n ) ) Z(R_{H})\cong\mathbb{K}[X^{n}]/(H(X^{n})) Z ( R H ) ≅ K [ X n ] / ( H ( X n )) is identified with the scalar field E id V \mathbb{E}\operatorname{id}_{V} E id V , via
z H ⟼ u H id V , z_{H}\longmapsto u_{H}\operatorname{id}_{V}, z H ⟼ u H id V ,
whereas the coefficient field L ⊆ R H \mathbb{L}\subseteq R_{H} L ⊆ R H is mapped to the subfield m L : = { m α : α ∈ L } ⊆ End E ( V ) \mathfrak{m}_{\mathbb{L}}:=\{m_{\alpha}:\alpha\in\mathbb{L}\}\subseteq\operatorname{End}_{\mathbb{E}}(V) m L := { m α : α ∈ L } ⊆ End E ( V ) defined by the fixed L \mathbb{L} L -vector space structure on V V V . Since H H H is irreducible of degree s s s , the elements
1 , u H , … , u H s − 1 1,u_{H},\ldots,u_{H}^{s-1} 1 , u H , … , u H s − 1
form a K \mathbb{K} K -basis of E \mathbb{E} E . Hence every element of E ⊗ K L \mathbb{E}\otimes_{\mathbb{K}}\mathbb{L} E ⊗ K L can be written uniquely as
∑ j = 0 s − 1 u H j ⊗ α j , α j ∈ L . \sum_{j=0}^{s-1}u_{H}^{j}\otimes\alpha_{j},\qquad\alpha_{j}\in\mathbb{L}. j = 0 ∑ s − 1 u H j ⊗ α j , α j ∈ L .
::: Its image under the multiplication map is
∑ j = 0 s − 1 u H j α j ∈ End E ( V ) . \sum_{j=0}^{s-1} u_H^j\alpha_j \in\operatorname{End}_{E}(V). j = 0 ∑ s − 1 u H j α j ∈ End E ( V ) .
Therefore it is enough to prove that every relation
∑ j = 0 s − 1 u H j α j = 0 , α j ∈ L , \sum_{j=0}^{s-1} u_H^j\alpha_j = 0,\qquad\alpha_j \in L, j = 0 ∑ s − 1 u H j α j = 0 , α j ∈ L ,
inside End E ( V ) \operatorname{End}_{E}(V) End E ( V ) , is trivial. Pulling this relation back through Ψ H \Psi_H Ψ H , we obtain
∑ j = 0 s − 1 z H j α j = 0 in R H . \sum_{j=0}^{s-1} z_H^j\alpha_j = 0 \qquad\text{in } R_H. j = 0 ∑ s − 1 z H j α j = 0 in R H .
Equivalently,
∑ j = 0 s − 1 α j X n j ∈ R H ( X n ) . \sum_{j=0}^{s-1} \alpha_j X^{nj} \in R_H(X^n). j = 0 ∑ s − 1 α j X nj ∈ R H ( X n ) .
Since, deg ( ∑ j = 0 s − 1 α j X n j ) < n s \deg\left(\sum_{j=0}^{s-1}\alpha_jX^{nj}\right)<ns deg ( ∑ j = 0 s − 1 α j X nj ) < n s , we get α 0 = α 1 = ⋯ = α s − 1 = 0 \alpha_0=\alpha_1=\cdots=\alpha_{s-1}=0 α 0 = α 1 = ⋯ = α s − 1 = 0 . This proves that the multiplication map
E ⊗ K L ⟶ End E ( V ) , e ⊗ α ⟼ e α , E\otimes_K L \longrightarrow\operatorname{End}_{E}(V),\qquad e\otimes\alpha\longmapsto e\alpha, E ⊗ K L ⟶ End E ( V ) , e ⊗ α ⟼ e α ,
is injective. In particular,
Lemma 6.8. Assume n > 2 n>2 n > 2 . Let W ⊆ End E ( V ) W\subseteq\operatorname{End}_{E}(V) W ⊆ End E ( V ) be a unital subfield such that
K ⊆ W , dim K W = n , W ∩ E = K , K\subseteq W,\qquad\dim_K W=n,\qquad W\cap E=K, K ⊆ W , dim K W = n , W ∩ E = K , and
W ⊆ L + E . W\subseteq L+E. W ⊆ L + E . Then
Proof. Choose a K K K -linear complement E 0 E_0 E 0 of K K K in E E E , so that
E = K ⊕ E 0 . E=K\oplus E_0. E = K ⊕ E 0 . By Theorem 6.7 , we have L ∩ E = K L\cap E=K L ∩ E = K . Hence
L ∩ E 0 = 0. L\cap E_0=0. L ∩ E 0 = 0. Since K ⊆ L K\subseteq L K ⊆ L , we obtain the direct sum decomposition
L + E = L + K + E 0 = L ⊕ E 0 L+E=L+K+E_0=L\oplus E_0 L + E = L + K + E 0 = L ⊕ E 0 as K K K -vector spaces. Let
π : L ⊕ E 0 ⟶ L \pi:L\oplus E_0\longrightarrow L π : L ⊕ E 0 ⟶ L be the projection onto the first component.
We claim that the restriction
π ∣ W : W ⟶ L \pi|_W:W\longrightarrow L π ∣ W : W ⟶ L
is injective. Indeed, if w ∈ W w \in W w ∈ W and π ( w ) = 0 \pi(w)=0 π ( w ) = 0 , then w ∈ W ∩ E 0 w \in W \cap E_0 w ∈ W ∩ E 0 . Since E 0 ⊆ E E_0 \subseteq E E 0 ⊆ E , we get
w ∈ W ∩ E = K . w \in W \cap E = K. w ∈ W ∩ E = K .
But w ∈ E 0 w \in E_0 w ∈ E 0 , and E 0 ∩ K = 0 E_0 \cap K = 0 E 0 ∩ K = 0 . Therefore w = 0 w=0 w = 0 . Thus π ∣ W \pi|_W π ∣ W is injective. Since
dim K W = n = dim K L , \dim_K W = n = \dim_K L, dim K W = n = dim K L ,
it follows that π ∣ W \pi|_W π ∣ W is an isomorphism.
Consequently, there exists a unique K K K -linear map
f : L ⟶ E 0 f : L \longrightarrow E_0 f : L ⟶ E 0
such that
W = { a + f ( a ) : a ∈ L } . W = \{a+f(a):a \in L\}. W = { a + f ( a ) : a ∈ L } .
Since 1 ∈ W 1 \in W 1 ∈ W , we have f ( 1 ) = 0 f(1)=0 f ( 1 ) = 0 .
We now prove that f = 0 f=0 f = 0 . Choose a K K K -basis
e 0 = 1 , e 1 , … , e s − 1 e_0=1,e_1,\ldots,e_{s-1} e 0 = 1 , e 1 , … , e s − 1
of E E E such that e 1 , … , e s − 1 e_1,\ldots,e_{s-1} e 1 , … , e s − 1 is a K K K -basis of E 0 E_0 E 0 . For every a ∈ L a \in L a ∈ L , write
f ( a ) = ∑ i = 1 s − 1 φ i ( a ) e i . f(a)=\sum_{i=1}^{s-1}\varphi_i(a)e_i. f ( a ) = i = 1 ∑ s − 1 φ i ( a ) e i .
where φ i : L → K \varphi_i:L \to K φ i : L → K are K K K -linear maps.
Let a , b ∈ L a,b \in L a , b ∈ L . Since W W W is closed under multiplication, we have
( a + f ( a ) ) ( b + f ( b ) ) ∈ W ⊆ L + E . (a+f(a))(b+f(b)) \in W \subseteq L+E. ( a + f ( a )) ( b + f ( b )) ∈ W ⊆ L + E .
Expanding the product, and using that a b ∈ L ab \in L ab ∈ L and f ( a ) f ( b ) ∈ E f(a)f(b) \in E f ( a ) f ( b ) ∈ E , we obtain
f ( a ) b + f ( b ) a ∈ L + E . f(a)b+f(b)a \in L+E. f ( a ) b + f ( b ) a ∈ L + E .
Equivalently,
∑ i = 1 s − 1 e i ( φ i ( a ) b + φ i ( b ) a ) ∈ L + E . \sum_{i=1}^{s-1}e_i\bigl(\varphi_i(a)b+\varphi_i(b)a\bigr) \in L+E. i = 1 ∑ s − 1 e i ( φ i ( a ) b + φ i ( b ) a ) ∈ L + E .
By Lemma 6.7, applied with respect to the basis e 0 = 1 , e 1 , … , e s − 1 e_0=1,e_1,\ldots,e_{s-1} e 0 = 1 , e 1 , … , e s − 1 , this implies that
φ i ( a ) b + φ i ( b ) a ∈ K \varphi_i(a)b+\varphi_i(b)a \in K φ i ( a ) b + φ i ( b ) a ∈ K
for every i = 1 , … , s − 1 i=1,\ldots,s-1 i = 1 , … , s − 1 and every a , b ∈ L a,b \in L a , b ∈ L .
We show that each φ i \varphi_i φ i is zero. Suppose, by contradiction, that φ i ≠ 0 \varphi_i \ne0 φ i = 0 for some i i i . Since φ i : L → K \varphi_i:L \to K φ i : L → K is a nonzero K K K -linear map and dim K L = n \dim_K L=n dim K L = n , we have
dim K ker ( φ i ) = n − 1. \dim_K\ker(\varphi_i)=n-1. dim K ker ( φ i ) = n − 1.
Because n > 2 n>2 n > 2 , this dimension is greater than 1 1 1 . Hence ker ( φ i ) \ker(\varphi_i) ker ( φ i ) is not contained in the one-dimensional K K K -subspace K ⊆ L K \subseteq L K ⊆ L . Choose
b ∈ ker ( φ i ) ∖ K . b \in\ker(\varphi_i) \setminus K. b ∈ ker ( φ i ) ∖ K .
Since φ i ≠ 0 \varphi_i \ne0 φ i = 0 , choose α ∈ L \alpha\in L α ∈ L such that
φ i ( α ) ≠ 0. \varphi_i(\alpha) \ne0. φ i ( α ) = 0.
Then
φ i ( α ) b + φ i ( b ) α ∈ K \varphi_i(\alpha)b+\varphi_i(b)\alpha\in K φ i ( α ) b + φ i ( b ) α ∈ K
and φ i ( b ) = 0 \varphi_i(b)=0 φ i ( b ) = 0 , so
φ i ( α ) b ∈ K . \varphi_i(\alpha)b\in K. φ i ( α ) b ∈ K .
Since φ i ( α ) ∈ K × \varphi_i(\alpha)\in K^\times φ i ( α ) ∈ K × , this forces b ∈ K b\in K b ∈ K , a contradiction. Therefore φ i = 0 \varphi_i=0 φ i = 0 for every i = 1 , … , s − 1 i=1,\ldots,s-1 i = 1 , … , s − 1 . Hence f = 0 f=0 f = 0 , and so
W = { α : α ∈ L } = L . W=\{\alpha:\alpha\in L\}=L. W = { α : α ∈ L } = L .
This proves the claim. □
Theorem 6.9. Assume n > 2 n>2 n > 2 . Let U ⊆ E U\subseteq E U ⊆ E be a K K K -subspace with K ⊆ U K\subseteq U K ⊆ U . Then
Stab G L E ( V ) ( L + U ) = Stab G L E ( V ) ( L ) , \operatorname{Stab}_{\mathrm{GL}_E(V)}(L+U)=\operatorname{Stab}_{\mathrm{GL}_E(V)}(L), Stab GL E ( V ) ( L + U ) = Stab GL E ( V ) ( L ) , where both stabilizers are taken with respect to the conjugation action of G L E ( V ) \mathrm{GL}_E(V) GL E ( V ) on End E ( V ) \operatorname{End}_E(V) End E ( V ) .
Proof. We need to prove that, for every Q ∈ G L E ( V ) Q\in\mathrm{GL}_E(V) Q ∈ GL E ( V ) ,
Q ( L + U ) Q − 1 = L + U Q(L+U)Q^{-1}=L+U Q ( L + U ) Q − 1 = L + U
if and only if
Q L Q − 1 = L . QLQ^{-1}=L. Q L Q − 1 = L .
Assume first that Q L Q − 1 = L QLQ^{-1}=L Q L Q − 1 = L . Since U ⊆ E id V U\subseteq E\operatorname{id}_V U ⊆ E id V , and since Q Q Q is E E E -linear, one has Q U Q − 1 = U QUQ^{-1}=U Q U Q − 1 = U . Therefore
Q ( L + U ) Q − 1 = Q L Q − 1 + Q U Q − 1 = L + U . Q(L+U)Q^{-1}=QLQ^{-1}+QUQ^{-1}=L+U. Q ( L + U ) Q − 1 = Q L Q − 1 + Q U Q − 1 = L + U .
Thus
Stab G L E ( V ) ( L ) ⊆ Stab G L E ( V ) ( L + U ) . \operatorname{Stab}_{\mathrm{GL}_E(V)}(L)\subseteq\operatorname{Stab}_{\mathrm{GL}_E(V)}(L+U). Stab GL E ( V ) ( L ) ⊆ Stab GL E ( V ) ( L + U ) .
Conversely, suppose that Q ( L + U ) Q − 1 = L + U Q(L+U)Q^{-1}=L+U Q ( L + U ) Q − 1 = L + U . Put W : = Q L Q − 1 W:=QLQ^{-1} W := Q L Q − 1 . Then
W ⊆ L + U ⊆ L + E . W\subseteq L+U\subseteq L+E. W ⊆ L + U ⊆ L + E .
Moreover, W W W is a unital subfield of A = End E ( V ) A=\operatorname{End}_E(V) A = End E ( V ) , contains K K K , and dim K W = n \dim_K W=n dim K W = n .
We prove that W ∩ E = K W\cap E=K W ∩ E = K . Let w ∈ W ∩ E w\in W\cap E w ∈ W ∩ E . Then w = Q α Q − 1 w=Q\alpha Q^{-1} w = Q α Q − 1 for some α ∈ L \alpha\in L α ∈ L . Since w ∈ E id V w\in E\operatorname{id}_V w ∈ E id V , and since Q ∈ G L E ( V ) Q\in\mathrm{GL}_E(V) Q ∈ GL E ( V ) , we have Q − 1 w Q = w Q^{-1}wQ=w Q − 1 w Q = w . Hence
α = Q − 1 w Q = w . \alpha=Q^{-1}wQ=w. α = Q − 1 w Q = w .
Thus w = α ∈ L ∩ E = K w=\alpha\in L\cap E=K w = α ∈ L ∩ E = K by Theorem 6.7 . Therefore W ∩ E ⊆ K W\cap E\subseteq K W ∩ E ⊆ K . The reverse inclusion is clear, since K ⊆ W K\subseteq W K ⊆ W and K ⊆ E K\subseteq E K ⊆ E . Hence
By Theorem 6.8, we obtain W = L W=L W = L . Therefore
Q L Q − 1 = L . QLQ^{-1}=L. Q L Q − 1 = L .
This proves the reverse inclusion, and hence the theorem. □
We can now derive the precise form of a linear Hermitian equivalence in the common endomorphism algebra. Notice that the following argument does not require the involutions Θ F \Theta_F Θ F and Θ G \Theta_G Θ G to coincide.
Lemma 6.10. Assume that n > 2 n>2 n > 2 and 1 ≤ ℓ < n s 4 1\leq\ell<\frac{ns}{4} 1 ≤ ℓ < 4 n s . Suppose that S ℓ ( F ) \mathcal{S}_{\ell}(F) S ℓ ( F ) and S ℓ ( G ) \mathcal{S}_{\ell}(G) S ℓ ( G ) are linearly Hermitian equivalent. Then there exist
Q ∈ GL E ( V ) and C ∈ End E ( V ) × Q\in\operatorname{GL}_{\mathbb{E}}(V) \qquad\text{and}\qquad C\in\operatorname{End}_{\mathbb{E}}(V)^{\times} Q ∈ GL E ( V ) and C ∈ End E ( V ) × such that
C ℓ ( G ) = Q C ℓ ( F ) Q − 1 C , \mathcal{C}_{\ell}(G)=Q\mathcal{C}_{\ell}(F)Q^{-1}C, C ℓ ( G ) = Q C ℓ ( F ) Q − 1 C , Q L Q − 1 = L , Q\mathbb{L}Q^{-1}=\mathbb{L}, Q L Q − 1 = L , and
C ∈ Cen End E ( V ) ( L ) × . C\in\operatorname{Cen}_{\operatorname{End}_{\mathbb{E}}(V)}(\mathbb{L})^{\times}. C ∈ Cen End E ( V ) ( L ) × .
Proof. For H ∈ { F , G } H\in\{F,G\} H ∈ { F , G } , choose an E \mathbb{E} E -algebra isomorphism
N H : End E ( V ) ⟶ ∼ M n ( E ) \mathcal{N}_H:\operatorname{End}_{\mathbb{E}}(V)\overset{\sim}{\longrightarrow}M_n(\mathbb{E}) N H : End E ( V ) ⟶ ∼ M n ( E )
identifying the involution Θ H \Theta_H Θ H with the standard Hermitian adjoint A ↦ A ∗ A\mapsto A^* A ↦ A ∗ . By Proposition 6.5 , the codes C ℓ ( F ) \mathcal{C}_{\ell}(F) C ℓ ( F ) and C ℓ ( G ) \mathcal{C}_{\ell}(G) C ℓ ( G ) are linearly Hermitian equivalent. Hence there exist S ∈ GL n ( E ) S\in\operatorname{GL}_n(\mathbb{E}) S ∈ GL n ( E ) and ν ∈ F × \nu\in\mathbb{F}^{\times} ν ∈ F × such that
N G ( C ℓ ( G ) ) = ν S N F ( C ℓ ( F ) ) S ∗ . \mathcal{N}_G\bigl(\mathcal{C}_{\ell}(G)\bigr)=\nu S\mathcal{N}_F\bigl(\mathcal{C}_{\ell}(F)\bigr)S^*. N G ( C ℓ ( G ) ) = ν S N F ( C ℓ ( F ) ) S ∗ .
Define
η : = N G − 1 ∘ τ S ∘ N F , \eta:=\mathcal{N}_G^{-1}\circ\tau_S\circ\mathcal{N}_F, η := N G − 1 ∘ τ S ∘ N F ,
where τ S ( A ) : = S A S − 1 \tau_S(A):=SAS^{-1} τ S ( A ) := S A S − 1 , for every A ∈ M n ( E ) A\in M_n(\mathbb{E}) A ∈ M n ( E ) , and put
C : = N G − 1 ( ν S S ∗ ) . C:=\mathcal{N}_G^{-1}(\nu SS^*). C := N G − 1 ( ν S S ∗ ) .
Then η \eta η is an E \mathbb{E} E -algebra automorphism of End E ( V ) \operatorname{End}_{\mathbb{E}}(V) End E ( V ) , the element C C C is invertible, and
C ℓ ( G ) = η ( C ℓ ( F ) ) C . \mathcal{C}_{\ell}(G)=\eta\bigl(\mathcal{C}_{\ell}(F)\bigr)C. C ℓ ( G ) = η ( C ℓ ( F ) ) C .
We next transport the Lyapunov invariant. Since N H \mathcal{N}_H N H is an isomorphism of algebras with involution, one has
N H ( L Θ H ( C ℓ ( H ) ) ) = L ( N H ( C ℓ ( H ) ) ) . \mathcal{N}_H\bigl(\mathcal{L}_{\Theta_H}(\mathcal{C}_{\ell}(H))\bigr)=\mathcal{L}\bigl(\mathcal{N}_H(\mathcal{C}_{\ell}(H))\bigr). N H ( L Θ H ( C ℓ ( H )) ) = L ( N H ( C ℓ ( H )) ) .
Therefore, by Proposition 5.2 ,
N G ( L Θ G ( C ℓ ( G ) ) ) = S N F ( L Θ F ( C ℓ ( F ) ) ) S − 1 . \mathcal{N}_G\bigl(\mathcal{L}_{\Theta_G}(\mathcal{C}_{\ell}(G))\bigr)=S\mathcal{N}_F\bigl(\mathcal{L}_{\Theta_F}(\mathcal{C}_{\ell}(F))\bigr)S^{-1}. N G ( L Θ G ( C ℓ ( G )) ) = S N F ( L Θ F ( C ℓ ( F )) ) S − 1 .
Equivalently,
L Θ G ( C ℓ ( G ) ) = η ( L Θ F ( C ℓ ( F ) ) ) . \mathcal{L}_{\Theta_G}(\mathcal{C}_{\ell}(G))=\eta\bigl(\mathcal{L}_{\Theta_F}(\mathcal{C}_{\ell}(F))\bigr). L Θ G ( C ℓ ( G )) = η ( L Θ F ( C ℓ ( F )) ) .
By Proposition 6.6,
L Θ F ( C ℓ ( F ) ) = L + B = L Θ G ( C ℓ ( G ) ) . \mathcal{L}_{\Theta_F}(\mathcal{C}_{\ell}(F))=\mathbb{L}+\mathcal{B}=\mathcal{L}_{\Theta_G}(\mathcal{C}_{\ell}(G)). L Θ F ( C ℓ ( F )) = L + B = L Θ G ( C ℓ ( G )) .
Consequently,
η ( L + B ) = L + B . \eta(\mathbb{L}+\mathcal{B})=\mathbb{L}+\mathcal{B}. η ( L + B ) = L + B .
Since End E ( V ) ≅ M n ( E ) \operatorname{End}_{\mathbb{E}}(V)\cong M_n(\mathbb{E}) End E ( V ) ≅ M n ( E ) , the Skolem–Noether theorem gives an element Q ∈ GL E ( V ) Q\in\operatorname{GL}_{\mathbb{E}}(V) Q ∈ GL E ( V ) such that
Thus
Q ( L + B ) Q − 1 = L + B . Q(\mathbb{L}+\mathcal{B})Q^{-1}=\mathbb{L}+\mathcal{B}. Q ( L + B ) Q − 1 = L + B .
By Theorem 6.9 , it follows that
Q L Q − 1 = L . Q\mathbb{L}Q^{-1}=\mathbb{L}. Q L Q − 1 = L .
Moreover,
C ℓ ( G ) = Q C ℓ ( F ) Q − 1 C . \mathcal{C}_{\ell}(G)=Q\mathcal{C}_{\ell}(F)Q^{-1}C. C ℓ ( G ) = Q C ℓ ( F ) Q − 1 C .
It remains to prove that C C C centralizes L \mathbb{L} L . Since 1 ∈ C ℓ ( F ) 1\in\mathcal{C}_{\ell}(F) 1 ∈ C ℓ ( F ) , the equality above gives
C ∈ C ℓ ( G ) . C\in\mathcal{C}_{\ell}(G). C ∈ C ℓ ( G ) .
Hence
Θ G ( C ) = C . \Theta_G(C)=C. Θ G ( C ) = C .
Furthermore, since L ⊆ C ℓ ( F ) \mathbb{L}\subseteq\mathcal{C}_{\ell}(F) L ⊆ C ℓ ( F ) and η ( L ) = L \eta(\mathbb{L})=\mathbb{L} η ( L ) = L , we have α C ∈ C ℓ ( G ) \alpha C\in\mathcal{C}_{\ell}(G) α C ∈ C ℓ ( G ) , for every α ∈ L \alpha\in\mathbb{L} α ∈ L . Every element of C ℓ ( G ) \mathcal{C}_{\ell}(G) C ℓ ( G ) is fixed by Θ G \Theta_G Θ G . Using also that Θ G \Theta_G Θ G fixes L \mathbb{L} L pointwise, we obtain
α C = Θ G ( α C ) = Θ G ( C ) Θ G ( α ) = C α . \alpha C=\Theta_G(\alpha C)=\Theta_G(C)\Theta_G(\alpha)=C\alpha. α C = Θ G ( α C ) = Θ G ( C ) Θ G ( α ) = C α .
Thus C C C centralizes L \mathbb{L} L , and therefore C ∈ Cen End E ( V ) ( L ) × C\in\operatorname{Cen}_{\operatorname{End}_{\mathbb{E}}(V)}(\mathbb{L})^{\times} C ∈ Cen End E ( V ) ( L ) × .
The range 1 ≤ ℓ < n 1\leq\ell<n 1 ≤ ℓ < n .
Lemma 6.11. Assume 1 ≤ ℓ < n 1\leq\ell<n 1 ≤ ℓ < n . For H ∈ { F , G } H\in\{F,G\} H ∈ { F , G } , set
P ℓ ( H ) : = ⟨ C ℓ ( H ) ⟩ L . \mathfrak{P}_{\ell}(H):=\langle\mathcal{C}_{\ell}(H)\rangle_{\mathbb{L}}. P ℓ ( H ) := ⟨ C ℓ ( H ) ⟩ L . Then
P ℓ ( H ) = ⟨ T H − ℓ , T H − ℓ + 1 , … , T H − 1 , 1 , T H , … , T H ℓ ⟩ L . \mathfrak{P}_{\ell}(H)=\left\langle T_H^{-\ell},T_H^{-\ell+1},\ldots,T_H^{-1},1,T_H,\ldots,T_H^{\ell}\right\rangle_{\mathbb{L}}. P ℓ ( H ) = ⟨ T H − ℓ , T H − ℓ + 1 , … , T H − 1 , 1 , T H , … , T H ℓ ⟩ L . Proof. From (32) , we have
C ℓ ( H ) = { α 0 + ∑ i = 1 ℓ ( σ − i ( α i ) T H − i + α i T H i ) : α 0 , … , α ℓ ∈ L } . \mathcal{C}_{\ell}(H)=\left\{\alpha_0+\sum_{i=1}^{\ell}\left(\sigma^{-i}(\alpha_i)T_H^{-i}+\alpha_iT_H^i\right):\alpha_0,\ldots,\alpha_{\ell}\in\mathbb{L}\right\}. C ℓ ( H ) = { α 0 + i = 1 ∑ ℓ ( σ − i ( α i ) T H − i + α i T H i ) : α 0 , … , α ℓ ∈ L } . Hence
P ℓ ( H ) ⊆ ⟨ T H − ℓ , T H − ℓ + 1 , … , T H − 1 , 1 , T H , … , T H ℓ ⟩ L . \mathfrak{P}_{\ell}(H)\subseteq\left\langle T_H^{-\ell},T_H^{-\ell+1},\ldots,T_H^{-1},1,T_H,\ldots,T_H^{\ell}\right\rangle_{\mathbb{L}}. P ℓ ( H ) ⊆ ⟨ T H − ℓ , T H − ℓ + 1 , … , T H − 1 , 1 , T H , … , T H ℓ ⟩ L . We prove the reverse inclusion. Clearly 1 ∈ C ℓ ( H ) 1\in\mathcal{C}_{\ell}(H) 1 ∈ C ℓ ( H ) , and therefore
L ⊆ P ℓ ( H ) . \mathbb{L}\subseteq\mathfrak{P}_{\ell}(H). L ⊆ P ℓ ( H ) . Let i ∈ { 1 , … , ℓ } i\in\{1,\ldots,\ell\} i ∈ { 1 , … , ℓ } . Since ℓ < n \ell<n ℓ < n , we have i < n i<n i < n , and hence
σ i ≠ id L . \sigma^i\ne\operatorname{id}_{\mathbb{L}}. σ i = id L . Thus there exists η i ∈ L \eta_i\in\mathbb{L} η i ∈ L such that
σ − i ( η i ) ≠ η i . \sigma^{-i}(\eta_i)\ne\eta_i. σ − i ( η i ) = η i . The two elements
T H − i + T H i T_H^{-i}+T_H^i T H − i + T H i and
σ − i ( η i ) T H − i + η i T H i \sigma^{-i}(\eta_i)T_H^{-i}+\eta_iT_H^i σ − i ( η i ) T H − i + η i T H i belong to C ℓ ( H ) ⊆ P ℓ ( H ) \mathcal{C}_{\ell}(H)\subseteq\mathfrak{P}_{\ell}(H) C ℓ ( H ) ⊆ P ℓ ( H ) . Since
det ( 1 1 σ − i ( η i ) η i ) = η i − σ − i ( η i ) ≠ 0 , (35) \det\begin{pmatrix}1&1\\\sigma^{-i}(\eta_i)&\eta_i\end{pmatrix}
=\eta_i-\sigma^{-i}(\eta_i)\ne0, \tag*{(35)} det ( 1 σ − i ( η i ) 1 η i ) = η i − σ − i ( η i ) = 0 , (35)
their L \mathbb{L} L -span contains both T H − i T_H^{-i} T H − i and T H i T_H^i T H i . Therefore
T H − i , T H i ∈ P ℓ ( H ) for every i = 1 , … , ℓ . T_H^{-i},T_H^i \in\mathcal{P}_{\ell}(H) \quad\text{for every } i=1,\ldots,\ell. T H − i , T H i ∈ P ℓ ( H ) for every i = 1 , … , ℓ .
This proves
P ℓ ( H ) = ⟨ T H − ℓ , T H − ℓ + 1 , … , T H − 1 , 1 , T H , … , T H ℓ ⟩ L . \mathcal{P}_{\ell}(H)=\left\langle T_H^{-\ell},T_H^{-\ell+1},\ldots,T_H^{-1},1,T_H,\ldots,T_H^{\ell}\right\rangle_{\mathbb{L}}. P ℓ ( H ) = ⟨ T H − ℓ , T H − ℓ + 1 , … , T H − 1 , 1 , T H , … , T H ℓ ⟩ L .
Note that T H ℓ T_H^\ell T H ℓ is invertible, and right multiplication by T H ℓ T_H^\ell T H ℓ gives
S 2 ℓ ( H ) : = P ℓ ( H ) T H ℓ = ⟨ 1 , T H , … , T H 2 ℓ ⟩ L . \mathcal{S}_{2\ell}(H):=\mathcal{P}_{\ell}(H)T_H^\ell=\left\langle1,T_H,\ldots,T_H^{2\ell}\right\rangle_{\mathbb{L}}. S 2 ℓ ( H ) := P ℓ ( H ) T H ℓ = ⟨ 1 , T H , … , T H 2 ℓ ⟩ L .
In [17 ] , linear equivalence was studied for L \mathbb{L} L -linear power spaces of the form
⟨ 1 , T H , … , T H s k − 1 ⟩ L . \left\langle1,T_H,\ldots,T_H^{sk-1}\right\rangle_{\mathbb{L}}. ⟨ 1 , T H , … , T H s k − 1 ⟩ L .
Although the result there is stated with last exponent s k − 1 sk-1 s k − 1 , its proof uses only that the exponents form an interval and that the largest exponent is strictly smaller than n s / 2 ns/2 n s /2 . The same argument therefore gives the following more flexible statement.
Proposition 6.12. Assume n > 2 n>2 n > 2 , and let j j j be an integer such that
1 ≤ j < n s 2 . 1\leq j<\frac{ns}{2}. 1 ≤ j < 2 n s . For H ∈ { F , G } H\in\{F,G\} H ∈ { F , G } , set
S j ( H ) : = ⟨ 1 , T H , … , T H j ⟩ L . \mathcal{S}_j(H):=\left\langle1,T_H,\ldots,T_H^j\right\rangle_{\mathbb{L}}. S j ( H ) := ⟨ 1 , T H , … , T H j ⟩ L . Suppose that there exist M , N ∈ GL E ( V ) M,N\in\operatorname{GL}_{\mathbb{E}}(V) M , N ∈ GL E ( V ) such that
S m ( G ) = M S m ( F ) N \mathcal{S}_m(G)=M\mathcal{S}_m(F)N S m ( G ) = M S m ( F ) N and
M L M − 1 = L . M\mathbb{L}M^{-1}=\mathbb{L}. M L M − 1 = L . Then there exists λ ∈ F q × \lambda\in\mathbb{F}_q^\times λ ∈ F q × such that
F ( Y ) = λ s G ( λ − 1 Y ) . F(Y)=\lambda^sG(\lambda^{-1}Y). F ( Y ) = λ s G ( λ − 1 Y ) .
We are now ready to prove the short-range main equivalence theorem stated at the beginning of this section.
Proof of Theorem 6.1 . Assume that C ℓ ( F ) \mathcal{C}_{\ell}(F) C ℓ ( F ) and C ℓ ( G ) \mathcal{C}_{\ell}(G) C ℓ ( G ) are linearly Hermitian equivalent. By Lemma 6.10, there exist
Q ∈ GL E ( V ) and C ∈ Cen End E ( V ) ( L ) × Q\in\operatorname{GL}_{\mathbb{E}}(V) \qquad\text{and}\qquad C\in\operatorname{Cen}_{\operatorname{End}_{\mathbb{E}}(V)}(\mathbb{L})^\times Q ∈ GL E ( V ) and C ∈ Cen End E ( V ) ( L ) ×
such that
C ℓ ( G ) = Q C ℓ ( F ) Q − 1 C \mathcal{C}_{\ell}(G)=Q\mathcal{C}_{\ell}(F)Q^{-1}C C ℓ ( G ) = Q C ℓ ( F ) Q − 1 C
and
Q L Q − 1 = L . Q\mathbb{L}Q^{-1}=\mathbb{L}. Q L Q − 1 = L .
We now pass from the Hermitian codes to their L \mathbb{L} L -linear spans. For H ∈ { F , G } H\in\{F,G\} H ∈ { F , G } , set
P ℓ ( H ) : = ⟨ C ℓ ( H ) ⟩ L . \mathcal{P}_{\ell}(H):=\left\langle\mathcal{C}_{\ell}(H)\right\rangle_{\mathbb{L}}. P ℓ ( H ) := ⟨ C ℓ ( H ) ⟩ L .
Since Q Q Q normalizes L \mathbb{L} L , taking the L \mathbb{L} L -linear span in the equivalence above gives
P ℓ ( G ) = Q P ℓ ( F ) Q − 1 C . \mathcal{P}_{\ell}(G)=Q\mathcal{P}_{\ell}(F)Q^{-1}C. P ℓ ( G ) = Q P ℓ ( F ) Q − 1 C .
By Lemma 6.11 , since ℓ < n \ell<n ℓ < n ,
P ℓ ( H ) = ⟨ T H − ℓ , … , T H − 1 , 1 , T H , … , T H ℓ ⟩ L . \mathcal{P}_{\ell}(H)=\left\langle T_H^{-\ell},\ldots,T_H^{-1},1,T_H,\ldots,T_H^{\ell}\right\rangle_{\mathbb{L}}. P ℓ ( H ) = ⟨ T H − ℓ , … , T H − 1 , 1 , T H , … , T H ℓ ⟩ L .
Right multiplication by T H ℓ T_H^{\ell} T H ℓ gives
S 2 ℓ ( H ) = P ℓ ( H ) T H ℓ = ⟨ 1 , T H , … , T H 2 ℓ ⟩ L . \mathcal{S}_{2\ell}(H)=\mathcal{P}_{\ell}(H)T_H^{\ell}=\left\langle1,T_H,\ldots,T_H^{2\ell}\right\rangle_{\mathbb{L}}. S 2 ℓ ( H ) = P ℓ ( H ) T H ℓ = ⟨ 1 , T H , … , T H 2 ℓ ⟩ L .
Consequently,
S 2 ℓ ( G ) = P ℓ ( G ) T G ℓ = Q P ℓ ( F ) Q − 1 C T G ℓ = Q S 2 ℓ ( F ) ( T F − ℓ Q − 1 C T G ℓ ) . \begin{aligned}
\mathcal{S}_{2\ell}(G)&=\mathcal{P}_{\ell}(G)T_G^{\ell}\\
&=Q\mathcal{P}_{\ell}(F)Q^{-1}CT_G^{\ell}\\
&=Q\mathcal{S}_{2\ell}(F)\left(T_F^{-\ell}Q^{-1}CT_G^{\ell}\right).
\end{aligned} S 2 ℓ ( G ) = P ℓ ( G ) T G ℓ = Q P ℓ ( F ) Q − 1 C T G ℓ = Q S 2 ℓ ( F ) ( T F − ℓ Q − 1 C T G ℓ ) .
Thus Proposition 6.12 applies with
M : = Q and N : = T F − ℓ Q − 1 C T G ℓ , M:=Q \qquad\text{and}\qquad N:=T_F^{-\ell}Q^{-1}CT_G^{\ell}, M := Q and N := T F − ℓ Q − 1 C T G ℓ ,
because
M L M − 1 = Q L Q − 1 = L MLM^{-1}=QLQ^{-1}=L M L M − 1 = Q L Q − 1 = L
and 2 ℓ < n s / 2 2\ell<ns/2 2 ℓ < n s /2 .
The range n < ℓ < n s / 4 n < \ell< ns/4 n < ℓ < n s /4
We now consider the range
n < ℓ < n s 4 . n<\ell<\frac{ns}{4}. n < ℓ < 4 n s .
This range is nonempty only when s > 4 s>4 s > 4 . The short-range argument from Theorem 6.1 no longer applies, because the L \mathbb{L} L -span of the Hermitian code need not contain every power of T H T_H T H in the interval [ − ℓ , ℓ ] [-\ell,\ell] [ − ℓ , ℓ ] ; the obstruction occurs precisely at exponents divisible by n n n . Nevertheless, the central elements in this interval still provide a useful invariant. They lead to a classification through the natural PGL 2 ( K ) \operatorname{PGL}_2(\mathbb{K}) PGL 2 ( K ) -action on
u H + u H − 1 . u_H+u_H^{-1}. u H + u H − 1 .
For H ∈ { F , G } H\in\{F,G\} H ∈ { F , G } , recall that u H ∈ E = F q s u_H\in\mathbb{E}=\mathbb{F}_{q^s} u H ∈ E = F q s is a root of H H H . Since H H H is self-reciprocal of even degree, we have
u H ϱ = u H − 1 , u_H^{\varrho}=u_H^{-1}, u H ϱ = u H − 1 ,
where ϱ : x ↦ x q s / 2 \varrho:x\mapsto x^{q^{s/2}} ϱ : x ↦ x q s /2 is the nontrivial automorphism of E / F \mathbb{E}/\mathbb{F} E / F . We put
w H : = u H + u H − 1 . w_H:=u_H+u_H^{-1}. w H := u H + u H − 1 .
Then w H ∈ F = F q s / 2 w_H\in\mathbb{F}=\mathbb{F}_{q^{s/2}} w H ∈ F = F q s /2 .
Lemma 6.13. Let H ∈ K [ Y ] H\in\mathbb{K}[Y] H ∈ K [ Y ] be monic, irreducible, and self-reciprocal of even degree s > 2 s>2 s > 2 , and let u H u_H u H be a root of H H H . Then
[ K ( w H ) : K ] = s 2 . [\mathbb{K}(w_H):\mathbb{K}]=\frac{s}{2}. [ K ( w H ) : K ] = 2 s .
Proof. We know that w H ∈ E ϱ = F q s / 2 w_H \in\mathbb{E}^{\varrho} = \mathbb{F}_{q^{s/2}} w H ∈ E ϱ = F q s /2 , and hence
[ K ( w H ) : K ] ≤ s 2 . [\mathbb{K}(w_H) : \mathbb{K}] \le\frac{s}{2}. [ K ( w H ) : K ] ≤ 2 s .
On the other hand, u H u_H u H satisfies
Y 2 − w H Y + 1 = 0 Y^2-w_HY+1=0 Y 2 − w H Y + 1 = 0
over K ( w H ) \mathbb{K}(w_H) K ( w H ) . Therefore
[ K ( u H ) : K ( w H ) ] ≤ 2. [\mathbb{K}(u_H) : \mathbb{K}(w_H)] \le2. [ K ( u H ) : K ( w H )] ≤ 2.
Since H H H has degree s s s , we have [ K ( u H ) : K ] = s [\mathbb{K}(u_H) : \mathbb{K}] = s [ K ( u H ) : K ] = s . Hence
s = [ K ( u H ) : K ] = [ K ( u H ) : K ( w H ) ] [ K ( w H ) : K ] ≤ 2 [ K ( w H ) : K ] . s = [\mathbb{K}(u_H) : \mathbb{K}] = [\mathbb{K}(u_H) : \mathbb{K}(w_H)][\mathbb{K}(w_H) : \mathbb{K}] \le2[\mathbb{K}(w_H) : \mathbb{K}]. s = [ K ( u H ) : K ] = [ K ( u H ) : K ( w H )] [ K ( w H ) : K ] ≤ 2 [ K ( w H ) : K ] .
It follows that
[ K ( w H ) : K ] ≥ s 2 . [\mathbb{K}(w_H) : \mathbb{K}] \ge\frac{s}{2}. [ K ( w H ) : K ] ≥ 2 s .
Therefore
[ K ( w H ) : K ] = s 2 . [\mathbb{K}(w_H) : \mathbb{K}] = \frac{s}{2}. [ K ( w H ) : K ] = 2 s .
We work in the commutative L \mathbb{L} L -algebra
A : = F ⊗ K L . \mathcal{A} := \mathbb{F} \otimes_{\mathbb{K}} \mathbb{L}. A := F ⊗ K L .
We identify F \mathbb{F} F with the subalgebra F ⊗ 1 \mathbb{F} \otimes1 F ⊗ 1 and L \mathbb{L} L with the subalgebra 1 ⊗ L 1 \otimes\mathbb{L} 1 ⊗ L . Thus A \mathcal{A} A is naturally an L \mathbb{L} L -vector space. For z ∈ F z \in\mathbb{F} z ∈ F , we write simply z z z instead of z ⊗ 1 z \otimes1 z ⊗ 1 . With this convention,
⟨ 1 , w H , … , w H t ⟩ L \langle1,w_H,\ldots,w_H^t\rangle_{\mathbb{L}} ⟨ 1 , w H , … , w H t ⟩ L
denotes the L \mathbb{L} L -subspace of A \mathcal{A} A generated by 1 , w H , … , w H t 1,w_H,\ldots,w_H^t 1 , w H , … , w H t , where t t t is a nonnegative integer.
Lemma 6.14. Let ℓ \ell ℓ be an integer such that
n < ℓ < n s 4 , n < \ell< \frac{ns}{4}, n < ℓ < 4 n s , and set
a 0 : = ⌊ ℓ n ⌋ . a_0 := \left\lfloor\frac{\ell}{n}\right\rfloor. a 0 := ⌊ n ℓ ⌋ . Assume that
⟨ 1 , w G , … , w G a 0 ⟩ L = ⟨ 1 , w F , … , w F a 0 ⟩ L C \langle1,w_G,\ldots,w_G^{a_0}\rangle_{\mathbb{L}}
=
\langle1,w_F,\ldots,w_F^{a_0}\rangle_{\mathbb{L}} C ⟨ 1 , w G , … , w G a 0 ⟩ L = ⟨ 1 , w F , … , w F a 0 ⟩ L C for some invertible element C ∈ A × C \in\mathcal{A}^{\times} C ∈ A × . Then there exist a , b , c , d ∈ K a,b,c,d \in\mathbb{K} a , b , c , d ∈ K , with
a d − b c ≠ 0 , ad-bc \ne0, a d − b c = 0 , such that
w G = a w F + b c w F + d . w_G=\frac{aw_F+b}{cw_F+d}. w G = c w F + d a w F + b .
Proof. By Lemma 6.13, both w F w_F w F and w G w_G w G have degree s / 2 s/2 s /2 over K \mathbb{K} K . Set
t : = s 2 . t := \frac{s}{2}. t := 2 s .
Since
n < ℓ < n s 4 , n < \ell< \frac{ns}{4}, n < ℓ < 4 n s ,
we have
1 ≤ a 0 = ⌊ ℓ n ⌋ < s 4 = t 2 . 1 \le a_0 = \left\lfloor\frac{\ell}{n}\right\rfloor< \frac{s}{4} = \frac{t}{2}. 1 ≤ a 0 = ⌊ n ℓ ⌋ < 4 s = 2 t .
In particular,
Since w F w_{F} w F has degree t t t over K \mathbb{K} K , the elements
1 , w F , … , w F t − 1 1,w_{F},\ldots,w_{F}^{t-1} 1 , w F , … , w F t − 1
form a K \mathbb{K} K -basis of F \mathbb{F} F . After extension of scalars from K \mathbb{K} K to L \mathbb{L} L , they form an L \mathbb{L} L -basis of
F ⊗ K L . \mathbb{F}\otimes_{\mathbb{K}}\mathbb{L}. F ⊗ K L .
Hence evaluation at w F w_{F} w F is injective on the set of polynomials in L [ X ] \mathbb{L}[X] L [ X ] of degree at most 2 a 0 2a_{0} 2 a 0 . The same holds for w G w_{G} w G .
By the assumed equality of the two L \mathbb{L} L -subspaces, for every i = 0 , … , a 0 i=0,\ldots,a_{0} i = 0 , … , a 0 there exists a polynomial
B i ( X ) ∈ L [ X ] , deg B i ≤ a 0 , B_{i}(X)\in\mathbb{L}[X],\qquad\deg B_{i}\leq a_{0}, B i ( X ) ∈ L [ X ] , deg B i ≤ a 0 ,
such that
w G i = B i ( w F ) C . w_{G}^{i}=B_{i}(w_{F})C. w G i = B i ( w F ) C .
For i = 0 i=0 i = 0 , this gives
1 = B 0 ( w F ) C . 1=B_{0}(w_{F})C. 1 = B 0 ( w F ) C .
Thus B 0 ( w F ) B_{0}(w_{F}) B 0 ( w F ) is invertible and
C = B 0 ( w F ) − 1 . C=B_{0}(w_{F})^{-1}. C = B 0 ( w F ) − 1 .
In particular,
w G = B 1 ( w F ) B 0 ( w F ) . w_{G}=\frac{B_{1}(w_{F})}{B_{0}(w_{F})}. w G = B 0 ( w F ) B 1 ( w F ) .
For i = 1 , … , a 0 i=1,\ldots,a_{0} i = 1 , … , a 0 , using
w G i = w G i − 1 w G , w_{G}^{i}=w_{G}^{i-1}w_{G}, w G i = w G i − 1 w G ,
we obtain
B i ( w F ) C = B i − 1 ( w F ) C B 1 ( w F ) C . B_{i}(w_{F})C=B_{i-1}(w_{F})C B_{1}(w_{F})C. B i ( w F ) C = B i − 1 ( w F ) C B 1 ( w F ) C .
Multiplying by C − 1 = B 0 ( w F ) C^{-1}=B_{0}(w_{F}) C − 1 = B 0 ( w F ) , we get
B i ( w F ) B 0 ( w F ) = B i − 1 ( w F ) B 1 ( w F ) . B_{i}(w_{F})B_{0}(w_{F})=B_{i-1}(w_{F})B_{1}(w_{F}). B i ( w F ) B 0 ( w F ) = B i − 1 ( w F ) B 1 ( w F ) .
Both sides are evaluations at w F w_{F} w F of polynomials of degree at most 2 a 0 2a_{0} 2 a 0 . Since
evaluation at w F w_{F} w F is injective in this degree range. Therefore
B i ( X ) B 0 ( X ) = B i − 1 ( X ) B 1 ( X ) , 1 ≤ i ≤ a 0 , B_{i}(X)B_{0}(X)=B_{i-1}(X)B_{1}(X),\qquad1\leq i\leq a_{0}, B i ( X ) B 0 ( X ) = B i − 1 ( X ) B 1 ( X ) , 1 ≤ i ≤ a 0 ,
as polynomial identities in L [ X ] \mathbb{L}[X] L [ X ] . Set
Since 1 1 1 and w G w_{G} w G are linearly independent over L \mathbb{L} L , we have w G ∉ L w_{G}\notin\mathbb{L} w G ∈ / L . Moreover, each polynomial B i B_{i} B i is nonzero: indeed, B i ( w F ) C = w G i ≠ 0 B_{i}(w_{F})C=w_{G}^{i}\neq0 B i ( w F ) C = w G i = 0 , and evaluation at w F w_{F} w F is injective for polynomials of degree at most k k k . Suppose first that k = 1 k=1 k = 1 . Write
B 1 ( X ) = a 1 X + b 1 and B 0 ( X ) = c 1 X + d 1 , B_{1}(X)=a_{1}X+b_{1}\qquad\text{and}\qquad B_{0}(X)=c_{1}X+d_{1}, B 1 ( X ) = a 1 X + b 1 and B 0 ( X ) = c 1 X + d 1 ,
with a 1 , b 1 , c 1 , d 1 ∈ L a_{1},b_{1},c_{1},d_{1}\in\mathbb{L} a 1 , b 1 , c 1 , d 1 ∈ L . Since B 0 ( w F ) B_{0}(w_{F}) B 0 ( w F ) is invertible, we obtain
w G = B 1 ( w F ) B 0 ( w F ) = a 1 w F + b 1 c 1 w F + d 1 . w_{G}=\frac{B_{1}(w_{F})}{B_{0}(w_{F})}
=\frac{a_{1}w_{F}+b_{1}}{c_{1}w_{F}+d_{1}}. w G = B 0 ( w F ) B 1 ( w F ) = c 1 w F + d 1 a 1 w F + b 1 .
Furthermore, B 1 B_1 B 1 and B 0 B_0 B 0 cannot be proportional over L L L , since otherwise w G ∈ L w_G \in L w G ∈ L . Therefore,
a 1 d 1 − b 1 c 1 ≠ 0. a_1d_1-b_1c_1 \ne0. a 1 d 1 − b 1 c 1 = 0.
Assume now that k > 1 k>1 k > 1 . We distinguish two cases.
Case 1: B 1 B_1 B 1 does not divide B 0 B_0 B 0 . Since B 1 ∤ B 0 B_1 \nmid B_0 B 1 ∤ B 0 , there exist an irreducible polynomial P ∈ L [ X ] P \in L[X] P ∈ L [ X ] , an integer r ≥ 1 r \ge1 r ≥ 1 , and a polynomial G 1 ∈ L [ X ] G_1 \in L[X] G 1 ∈ L [ X ] such that
B 1 = G 1 P r , gcd ( G 1 , P ) = 1 , B_1=G_1P^r,\qquad\gcd(G_1,P)=1, B 1 = G 1 P r , g cd( G 1 , P ) = 1 ,
and
P r ∤ B 0 . P^r \nmid B_0. P r ∤ B 0 .
We claim that
P r + i − 1 ∣ B i P^{r+i-1}\mid B_i P r + i − 1 ∣ B i
for every i = 1 , … , k i=1,\ldots,k i = 1 , … , k . For i = 1 i=1 i = 1 , this follows immediately from
B 1 = G 1 P r . B_1=G_1P^r. B 1 = G 1 P r .
Assume that
P r + i − 2 ∣ B i − 1 P^{r+i-2}\mid B_{i-1} P r + i − 2 ∣ B i − 1
for some i ∈ { 2 , … , k } i\in\{2,\ldots,k\} i ∈ { 2 , … , k } . From
B i B 0 = B i − 1 B 1 B_iB_0=B_{i-1}B_1 B i B 0 = B i − 1 B 1
it follows that
P 2 r + i − 2 ∣ B i B 0 . P^{2r+i-2}\mid B_iB_0. P 2 r + i − 2 ∣ B i B 0 .
Since P r ∤ B 0 P^r\nmid B_0 P r ∤ B 0 , at most r − 1 r-1 r − 1 factors equal to P P P divide B 0 B_0 B 0 . Therefore,
P r + i − 1 ∣ B i . P^{r+i-1}\mid B_i. P r + i − 1 ∣ B i .
This proves the claim by induction. In particular,
P r + k − 1 ∣ B k . P^{r+k-1}\mid B_k. P r + k − 1 ∣ B k .
Since B k ≠ 0 B_k\ne0 B k = 0 and deg B k ≤ k \deg B_k\le k deg B k ≤ k , we obtain
k ≥ deg B k ≥ ( r + k − 1 ) deg P . k\ge\deg B_k\ge(r+k-1)\deg P. k ≥ deg B k ≥ ( r + k − 1 ) deg P .
Since r ≥ 1 r\ge1 r ≥ 1 and deg P ≥ 1 \deg P\ge1 deg P ≥ 1 , this forces
r = 1 and deg P = 1. r=1\qquad\text{and}\qquad\deg P=1. r = 1 and deg P = 1.
Moreover, P k ∣ B k P^k\mid B_k P k ∣ B k and deg B k ≤ k \deg B_k\le k deg B k ≤ k , so
B k = λ P k B_k=\lambda P^k B k = λ P k
for some λ ∈ L × \lambda\in L^\times λ ∈ L × . We now have
B 1 = G 1 P , gcd ( G 1 , P ) = 1. B_1=G_1P,\qquad\gcd(G_1,P)=1. B 1 = G 1 P , g cd( G 1 , P ) = 1.
Using the identity
B k B 0 = B k − 1 B 1 , B_kB_0=B_{k-1}B_1, B k B 0 = B k − 1 B 1 ,
we obtain
λ P k B 0 = B k − 1 G 1 P . \lambda P^kB_0=B_{k-1}G_1P. λ P k B 0 = B k − 1 G 1 P .
After cancelling P P P , this becomes
λ P k − 1 B 0 = B k − 1 G 1 . \lambda P^{k-1}B_0=B_{k-1}G_1. λ P k − 1 B 0 = B k − 1 G 1 .
Thus G 1 G_1 G 1 divides P k − 1 B 0 P^{k-1}B_0 P k − 1 B 0 . Since
gcd ( G 1 , P ) = 1 , \gcd(G_1,P)=1, g cd( G 1 , P ) = 1 ,
it follows that G 1 ∣ B 0 G_1 \mid B_0 G 1 ∣ B 0 . Set
E : = B 0 G 1 . E:=\frac{B_0}{G_1}. E := G 1 B 0 .
Then
B k − 1 = λ P k − 1 E . B_{k-1}=\lambda P^{k-1}E. B k − 1 = λ P k − 1 E .
Since deg B k − 1 ≤ k \deg B_{k-1}\leq k deg B k − 1 ≤ k and deg P = 1 \deg P=1 deg P = 1 , we conclude that
deg E ≤ 1. \deg E\leq1. deg E ≤ 1.
Furthermore,
B 0 ( w F ) = G 1 ( w F ) E ( w F ) . B_0(w_F)=G_1(w_F)E(w_F). B 0 ( w F ) = G 1 ( w F ) E ( w F ) .
Therefore,
w G = B 1 ( w F ) B 0 ( w F ) = P ( w F ) E ( w F ) . w_G=\frac{B_1(w_F)}{B_0(w_F)}=\frac{P(w_F)}{E(w_F)}. w G = B 0 ( w F ) B 1 ( w F ) = E ( w F ) P ( w F ) .
Write
P ( X ) = a 1 X + b 1 and E ( X ) = c 1 X + d 1 , P(X)=a_1X+b_1 \qquad\text{and}\qquad E(X)=c_1X+d_1, P ( X ) = a 1 X + b 1 and E ( X ) = c 1 X + d 1 ,
with a 1 , b 1 , c 1 , d 1 ∈ L a_1,b_1,c_1,d_1\in\mathbb{L} a 1 , b 1 , c 1 , d 1 ∈ L . We obtain
w G = a 1 w F + b 1 c 1 w F + d 1 . w_G=\frac{a_1w_F+b_1}{c_1w_F+d_1}. w G = c 1 w F + d 1 a 1 w F + b 1 .
Finally, P P P and E E E cannot be proportional over L \mathbb{L} L , since otherwise their quotient would belong to L \mathbb{L} L , and hence w G ∈ L w_G\in\mathbb{L} w G ∈ L , a contradiction. Consequently,
a 1 d 1 − b 1 c 1 ≠ 0. a_1d_1-b_1c_1\ne0. a 1 d 1 − b 1 c 1 = 0.
Case 2: B 1 B_1 B 1 divides B 0 B_0 B 0 . Write
for some E ∈ L [ X ] E\in\mathbb{L}[X] E ∈ L [ X ] . Since
w G = B 1 ( w F ) B 0 ( w F ) = E ( w F ) − 1 , w_G=\frac{B_1(w_F)}{B_0(w_F)}=E(w_F)^{-1}, w G = B 0 ( w F ) B 1 ( w F ) = E ( w F ) − 1 ,
the polynomial E E E cannot be constant, as otherwise w G ∈ L w_G\in\mathbb{L} w G ∈ L . Thus
deg E ≥ 1. \deg E\geq1. deg E ≥ 1.
Substituting B 0 = B 1 E B_0=B_1E B 0 = B 1 E into
B i B 0 = B i − 1 B 1 B_iB_0=B_{i-1}B_1 B i B 0 = B i − 1 B 1
and cancelling the nonzero polynomial B 1 B_1 B 1 , we obtain
B i E = B i − 1 , 1 ≤ i ≤ k . B_iE=B_{i-1},\qquad1\leq i\leq k. B i E = B i − 1 , 1 ≤ i ≤ k .
It follows inductively that
B i = B k E k − i , 0 ≤ i ≤ k . B_i=B_kE^{k-i},\qquad0\leq i\leq k. B i = B k E k − i , 0 ≤ i ≤ k .
In particular,
B 0 = B k E k . B_0=B_kE^k. B 0 = B k E k .
Therefore,
k ≥ deg B 0 = deg B k + k deg E . k\geq\deg B_0=\deg B_k+k\deg E. k ≥ deg B 0 = deg B k + k deg E .
Since B k ≠ 0 B_k \ne0 B k = 0 and deg E ≥ 1 \deg E \ge1 deg E ≥ 1 , we conclude that
B k ∈ L × and deg E = 1. B_k \in\mathbb{L}^{\times} \qquad\text{and}\qquad\deg E = 1. B k ∈ L × and deg E = 1.
Write
E ( X ) = c 1 X + d 1 , c 1 ≠ 0. E(X) = c_1X + d_1, \qquad c_1 \ne0. E ( X ) = c 1 X + d 1 , c 1 = 0.
Then
w G = 1 c 1 w F + d 1 = 0 w F + 1 c 1 w F + d 1 . w_G = \frac{1}{c_1w_F+d_1} = \frac{0w_F+1}{c_1w_F+d_1}. w G = c 1 w F + d 1 1 = c 1 w F + d 1 0 w F + 1 .
Thus, taking
a 1 = 0 and b 1 = 1 , a_1 = 0 \qquad\text{and}\qquad b_1 = 1, a 1 = 0 and b 1 = 1 ,
we have
a 1 d 1 − b 1 c 1 = − c 1 ≠ 0. a_1d_1-b_1c_1=-c_1 \ne0. a 1 d 1 − b 1 c 1 = − c 1 = 0.
In all cases, there exist a 1 , b 1 , c 1 , d 1 ∈ L a_1,b_1,c_1,d_1 \in\mathbb{L} a 1 , b 1 , c 1 , d 1 ∈ L , with
a 1 d 1 − b 1 c 1 ≠ 0 , a_1d_1-b_1c_1 \ne0, a 1 d 1 − b 1 c 1 = 0 ,
such that
w G = a 1 w F + b 1 c 1 w F + d 1 . w_G = \frac{a_1w_F+b_1}{c_1w_F+d_1}. w G = c 1 w F + d 1 a 1 w F + b 1 .
Equivalently,
c 1 w F w G + d 1 w G − a 1 w F − b 1 = 0 c_1w_Fw_G+d_1w_G-a_1w_F-b_1=0 c 1 w F w G + d 1 w G − a 1 w F − b 1 = 0
inside F ⊗ K L \mathbb{F}\otimes_{\mathbb{K}}\mathbb{L} F ⊗ K L .
It remains to descend the coefficients from L \mathbb{L} L to K \mathbb{K} K . The last identity gives an L \mathbb{L} L -linear dependence among the four elements
1 , w F , w G , w F w G 1,\quad w_F,\quad w_G,\quad w_Fw_G 1 , w F , w G , w F w G
of the K \mathbb{K} K -vector space F \mathbb{F} F . Since
F ⊗ K L \mathbb{F}\otimes_{\mathbb{K}}\mathbb{L} F ⊗ K L
is obtained from F \mathbb{F} F by extension of scalars, these four elements are L \mathbb{L} L -linearly dependent after extension of scalars if and only if they are K \mathbb{K} K -linearly dependent before extension of scalars. Hence there exist a , b , c , d ∈ K a,b,c,d \in\mathbb{K} a , b , c , d ∈ K , not all zero, such that
c w F w G + d w G − a w F − b = 0. cw_Fw_G+dw_G-aw_F-b=0. c w F w G + d w G − a w F − b = 0.
Equivalently,
( c w F + d ) w G = a w F + b . (cw_F+d)w_G=aw_F+b. ( c w F + d ) w G = a w F + b .
We claim that
c w F + d ≠ 0. cw_F+d \ne0. c w F + d = 0.
Indeed, if c w F + d = 0 cw_F+d=0 c w F + d = 0 , then either c = 0 = d c=0=d c = 0 = d , or w F ∈ K w_F \in\mathbb{K} w F ∈ K . The latter is impossible because w F w_F w F has degree t = s / 2 > 1 t=s/2>1 t = s /2 > 1 over K \mathbb{K} K . If c = d = 0 c=d=0 c = d = 0 , then the relation above gives
a w F + b = 0 , aw_F+b=0, a w F + b = 0 ,
again forcing w F ∈ K w_F \in\mathbb{K} w F ∈ K , unless a = b = 0 a=b=0 a = b = 0 , which is excluded. Thus
c w F + d ≠ 0. cw_F+d \ne0. c w F + d = 0.
Since F \mathbb{F} F is a field, c w F + d cw_F+d c w F + d is invertible. Therefore
w G = a w F + b c w F + d . w_G=\frac{aw_F+b}{cw_F+d}. w G = c w F + d a w F + b .
Finally, the matrix
( a b c d ) \begin{pmatrix} a & b \\ c & d \end{pmatrix} ( a c b d )
is nonsingular. If it were singular, then the two linear polynomials
a X + b and c X + d aX+b \qquad\text{and}\qquad cX+d a X + b and c X + d
would be proportional over K \mathrm{K} K , and the quotient
a X + b c X + d \frac{aX+b}{cX+d} c X + d a X + b
would be constant in K \mathrm{K} K . Hence the equality above would force
w G ∈ K , w_G \in\mathrm{K}, w G ∈ K ,
contradicting the fact that w G w_G w G has degree t = s / 2 > 1 t=s/2>1 t = s /2 > 1 over K \mathrm{K} K . Thus
a d − b c ≠ 0. ad-bc \ne0. a d − b c = 0.
This proves the assertion. □
We now isolate the semilinearity component needed to detect it.
For r ∈ Z / n Z r\in\mathbb{Z}/n\mathbb{Z} r ∈ Z / n Z , define
A r : = { A ∈ End E ( V ) : A α = σ r ( α ) A for every α ∈ L } . \mathcal{A}_r:=\{A\in\operatorname{End}_{\mathrm{E}}(V):A\alpha=\sigma^r(\alpha)A\text{ for every }\alpha\in\mathbb{L}\}. A r := { A ∈ End E ( V ) : A α = σ r ( α ) A for every α ∈ L } .
Then
End E ( V ) = ⨁ r ∈ Z / n Z A r , \operatorname{End}_{\mathrm{E}}(V)=\bigoplus_{r\in\mathbb{Z}/n\mathbb{Z}}\mathcal{A}_r, End E ( V ) = r ∈ Z / n Z ⨁ A r ,
see, e.g., [5 Lemma 2.3] . The component A 0 \mathcal{A}_0 A 0 consists precisely of the operators commuting with L \mathbb{L} L .
Lemma 6.15. Let H ∈ { F , G } H\in\{F,G\} H ∈ { F , G } , and put
a : = ⌊ ℓ n ⌋ . a:=\left\lfloor\frac{\ell}{n}\right\rfloor. a := ⌊ n ℓ ⌋ . Then
C ℓ ( H ) ∩ A 0 = ⟨ 1 , w H , w H 2 , … , w H a ⟩ L . \mathcal{C}_{\ell}(H)\cap\mathcal{A}_0=\langle1,w_H,w_H^2,\ldots,w_H^a\rangle_{\mathbb{L}}. C ℓ ( H ) ∩ A 0 = ⟨ 1 , w H , w H 2 , … , w H a ⟩ L . Proof. From the explicit expression
C ℓ ( H ) = { α 0 + ∑ i = 1 ℓ ( σ − i ( α i ) T H − i + α i T H i ) : α 0 , … , α ℓ ∈ L } , \mathcal{C}_{\ell}(H)=\left\{\alpha_0+\sum_{i=1}^{\ell}\left(\sigma^{-i}(\alpha_i)T_H^{-i}+\alpha_iT_H^i\right):\alpha_0,\ldots,\alpha_{\ell}\in\mathbb{L}\right\}, C ℓ ( H ) = { α 0 + i = 1 ∑ ℓ ( σ − i ( α i ) T H − i + α i T H i ) : α 0 , … , α ℓ ∈ L } , the terms lying in A 0 \mathcal{A}_0 A 0 are precisely those with exponent divisible by n n n . Indeed, for every integer j ≥ 0 j\ge0 j ≥ 0 , one has
T H j ∈ A j . T_H^j\in\mathcal{A}_j. T H j ∈ A j . Since End E ( V ) = ⨁ r ∈ Z / n Z A r \operatorname{End}_{\mathrm{E}}(V)=\bigoplus_{r\in\mathbb{Z}/n\mathbb{Z}}\mathcal{A}_r End E ( V ) = ⨁ r ∈ Z / n Z A r , an element of C ℓ ( H ) \mathcal{C}_{\ell}(H) C ℓ ( H ) belongs to A 0 \mathcal{A}_0 A 0 only if all its components in A r \mathcal{A}_r A r , with r ≠ 0 r\ne0 r = 0 , vanish. This implies that
α i = 0 whenever n ∤ i . \alpha_i=0\qquad\text{whenever }n\nmid i. α i = 0 whenever n ∤ i . Thus the only possible nonzero indices are
j = t n , t = 1 , … , a . j=tn,\qquad t=1,\ldots,a. j = t n , t = 1 , … , a .
For these indices,
σ − j = id L , T H j = u H t id V , T H − j = u H − t id V . \sigma^{-j}=\operatorname{id}_{\mathbb{L}},\qquad T_H^j=u_H^t\operatorname{id}_V,\qquad T_H^{-j}=u_H^{-t}\operatorname{id}_V. σ − j = id L , T H j = u H t id V , T H − j = u H − t id V .
Hence
E ℓ ( H ) ∩ A 0 = ⟨ 1 , u H + u H − 1 , u H 2 + u H − 2 , … , u H a + u H − a ⟩ L . \mathcal{E}_{\ell}(H)\cap\mathcal{A}_0=\langle1,u_H+u_H^{-1},u_H^2+u_H^{-2},\ldots,u_H^a+u_H^{-a}\rangle_{\mathbb{L}}. E ℓ ( H ) ∩ A 0 = ⟨ 1 , u H + u H − 1 , u H 2 + u H − 2 , … , u H a + u H − a ⟩ L .
Now put
w H = u H + u H − 1 . w_H=u_H+u_H^{-1}. w H = u H + u H − 1 .
For every j ≥ 1 j\ge1 j ≥ 1 , the element
u H j + u H − j u_H^j+u_H^{-j} u H j + u H − j
is a monic polynomial of degree j j j in w H w_H w H . Indeed,
u H j + 1 + u H − ( j + 1 ) = w H ( u H j + u H − j ) − ( u H j − 1 + u H − ( j − 1 ) ) . u_H^{j+1}+u_H^{-(j+1)}=w_H(u_H^j+u_H^{-j})-(u_H^{j-1}+u_H^{-(j-1)}). u H j + 1 + u H − ( j + 1 ) = w H ( u H j + u H − j ) − ( u H j − 1 + u H − ( j − 1 ) ) .
Therefore
⟨ 1 , u H + u H − 1 , u H 2 + u H − 2 , … , u H a + u H − a ⟩ L = ⟨ 1 , w H , w H 2 , … , w H a ⟩ L . \langle1,u_H+u_H^{-1},u_H^2+u_H^{-2},\ldots,u_H^a+u_H^{-a}\rangle_{\mathbb{L}}=\langle1,w_H,w_H^2,\ldots,w_H^a\rangle_{\mathbb{L}}. ⟨ 1 , u H + u H − 1 , u H 2 + u H − 2 , … , u H a + u H − a ⟩ L = ⟨ 1 , w H , w H 2 , … , w H a ⟩ L .
This proves the claim. □
We also need the following simple consequence of the Lyapunov computation.
Lemma 6.16. Let Q Q Q and C C C be as in Lemma 6.10 . Then the map
A ⟼ Q A Q − 1 C A\longmapsto QAQ^{-1}C A ⟼ Q A Q − 1 C preserves each A r \mathcal{A}_r A r .
Proof. Since
Q L Q − 1 = L , Q\mathbb{L}Q^{-1}=\mathbb{L}, Q L Q − 1 = L ,
there exists h ∈ Z / n Z h\in\mathbb{Z}/n\mathbb{Z} h ∈ Z / n Z such that
Q α Q − 1 = σ h ( α ) for every α ∈ L . Q\alpha Q^{-1}=\sigma^h(\alpha)\qquad\text{for every }\alpha\in\mathbb{L}. Q α Q − 1 = σ h ( α ) for every α ∈ L .
Equivalently,
Q − 1 α = σ − h ( α ) Q − 1 . Q^{-1}\alpha=\sigma^{-h}(\alpha)Q^{-1}. Q − 1 α = σ − h ( α ) Q − 1 .
Let A ∈ A r A\in\mathcal{A}_r A ∈ A r . Then, for every α ∈ L \alpha\in\mathbb{L} α ∈ L ,
( Q A Q − 1 ) α = Q A σ − h ( α ) Q − 1 = Q σ r − h ( α ) A Q − 1 = σ r ( α ) Q A Q − 1 . (QAQ^{-1})\alpha=QA\sigma^{-h}(\alpha)Q^{-1}
=Q\sigma^{r-h}(\alpha)AQ^{-1}
=\sigma^r(\alpha)QAQ^{-1}. ( Q A Q − 1 ) α = Q A σ − h ( α ) Q − 1 = Q σ r − h ( α ) A Q − 1 = σ r ( α ) Q A Q − 1 .
Hence
Q A Q − 1 ∈ A r . QAQ^{-1}\in\mathcal{A}_r. Q A Q − 1 ∈ A r .
Moreover,
C ∈ Cen End E ( V ) ( L ) = A 0 . C\in\operatorname{Cen}_{\operatorname{End}_E(V)}(\mathbb{L})=\mathcal{A}_0. C ∈ Cen End E ( V ) ( L ) = A 0 .
Therefore right multiplication by C C C preserves every A r \mathcal{A}_r A r . Consequently,
Q A Q − 1 C ∈ A r QAQ^{-1}C\in\mathcal{A}_r Q A Q − 1 C ∈ A r
for every A ∈ A r A\in\mathcal{A}_r A ∈ A r . Since the map is bijective, it preserves each component A r \mathcal{A}_r A r . □
We are now ready to prove the large-range main equivalence theorem stated at the beginning of this section.
Proof of Theorem 6.2 . Assume that S ℓ ( F ) \mathcal{S}_{\ell}(F) S ℓ ( F ) and S ℓ ( G ) \mathcal{S}_{\ell}(G) S ℓ ( G ) are linearly Hermitian equivalent. By Lemma 6.10 , there exist
Q ∈ G L E ( V ) and C ∈ Cen End E ( V ) ( L ) × Q \in\mathrm{GL}_{\mathbb{E}}(V) \qquad\text{and} \qquad C \in\operatorname{Cen}_{\operatorname{End}_{\mathbb{E}}(V)}(\mathbb{L})^{\times} Q ∈ GL E ( V ) and C ∈ Cen End E ( V ) ( L ) ×
such that
C ℓ ( G ) = Q C ℓ ( F ) Q − 1 C \mathcal{C}_{\ell}(G)=Q\mathcal{C}_{\ell}(F)Q^{-1}C C ℓ ( G ) = Q C ℓ ( F ) Q − 1 C
and
Q L Q − 1 = L . Q\mathbb{L}Q^{-1}=\mathbb{L}. Q L Q − 1 = L .
By Lemma 6.16 , the map
A ⟼ Q A Q − 1 C A \longmapsto QAQ^{-1}C A ⟼ Q A Q − 1 C
preserves A 0 A_{0} A 0 . Consequently, we have
C ℓ ( G ) ∩ A 0 = Q ( C ℓ ( F ) ∩ A 0 ) Q − 1 C . \mathcal{C}_{\ell}(G)\cap A_{0}=Q\left(\mathcal{C}_{\ell}(F)\cap A_{0}\right)Q^{-1}C. C ℓ ( G ) ∩ A 0 = Q ( C ℓ ( F ) ∩ A 0 ) Q − 1 C .
Put
a 0 : = ⌊ ℓ n ⌋ . a_{0}:=\left\lfloor\frac{\ell}{n}\right\rfloor. a 0 := ⌊ n ℓ ⌋ .
By Lemma 6.15 , we get
⟨ 1 , w G , … , w G a 0 ⟩ L = Q ⟨ 1 , w F , … , w F a 0 ⟩ L Q − 1 C . \langle1,w_{G},\ldots,w_{G}^{a_{0}}\rangle_{\mathbb{L}}=Q\langle1,w_{F},\ldots,w_{F}^{a_{0}}\rangle_{\mathbb{L}}Q^{-1}C. ⟨ 1 , w G , … , w G a 0 ⟩ L = Q ⟨ 1 , w F , … , w F a 0 ⟩ L Q − 1 C .
Since Q Q Q normalizes L \mathbb{L} L and fixes the scalar field E \mathbb{E} E pointwise under conjugation, one has
Q ⟨ 1 , w F , … , w F a 0 ⟩ L Q − 1 = ⟨ 1 , w F , … , w F a 0 ⟩ L . Q\langle1,w_{F},\ldots,w_{F}^{a_{0}}\rangle_{\mathbb{L}}Q^{-1}=\langle1,w_{F},\ldots,w_{F}^{a_{0}}\rangle_{\mathbb{L}}. Q ⟨ 1 , w F , … , w F a 0 ⟩ L Q − 1 = ⟨ 1 , w F , … , w F a 0 ⟩ L .
Therefore
⟨ 1 , w G , … , w G a 0 ⟩ L = ⟨ 1 , w F , … , w F a 0 ⟩ L C . \langle1,w_{G},\ldots,w_{G}^{a_{0}}\rangle_{\mathbb{L}}=\langle1,w_{F},\ldots,w_{F}^{a_{0}}\rangle_{\mathbb{L}}C. ⟨ 1 , w G , … , w G a 0 ⟩ L = ⟨ 1 , w F , … , w F a 0 ⟩ L C .
The element C C C is invertible. Moreover, since 1 1 1 belongs to
⟨ 1 , w F , … , w F a 0 ⟩ L , \langle1,w_{F},\ldots,w_{F}^{a_{0}}\rangle_{\mathbb{L}}, ⟨ 1 , w F , … , w F a 0 ⟩ L ,
the equality above implies
C ∈ ⟨ 1 , w G , … , w G a 0 ⟩ L ⊆ F ⊗ K L . C\in\langle1,w_{G},\ldots,w_{G}^{a_{0}}\rangle_{\mathbb{L}}\subseteq\mathbb{F}\otimes_{\mathbb{K}}\mathbb{L}. C ∈ ⟨ 1 , w G , … , w G a 0 ⟩ L ⊆ F ⊗ K L .
Thus the equality takes place inside the commutative algebra
F ⊗ K L . \mathbb{F}\otimes_{\mathbb{K}}\mathbb{L}. F ⊗ K L .
Now, by Lemma 6.13, both w F w_{F} w F and w G w_{G} w G have degree
t : = s 2 t:=\frac{s}{2} t := 2 s
over K \mathbb{K} K . Since
n < ℓ < n s 4 , n<\ell<\frac{ns}{4}, n < ℓ < 4 n s ,
we have
1 ≤ a 0 = ⌊ ℓ n ⌋ < s 4 = t 2 . 1\leq a_{0}=\left\lfloor\frac{\ell}{n}\right\rfloor<\frac{s}{4}=\frac{t}{2}. 1 ≤ a 0 = ⌊ n ℓ ⌋ < 4 s = 2 t .
Therefore Lemma 6.14 , applied with
u = w F , v = w G , m = a 0 , u=w_{F},\qquad v=w_{G},\qquad m=a_{0}, u = w F , v = w G , m = a 0 ,
gives
w G = a w F + b c w F + d w_{G}=\frac{aw_{F}+b}{cw_{F}+d} w G = c w F + d a w F + b
for some a , b , c , d ∈ K a,b,c,d \in\mathbb{K} a , b , c , d ∈ K , with
a d − b c ≠ 0. ad-bc \ne0. a d − b c = 0.
□ \square □
Remark 6.17. The conclusion of Theorem 6.2 is weaker than the conclusion of Theorem 6.1 . In the short range ℓ < n \ell<n ℓ < n , the interval [ − ℓ , ℓ ] [-\ell,\ell] [ − ℓ , ℓ ] can be recovered after taking the L-span, and one obtains the stronger relation
F ( Y ) = λ s G ( λ − 1 Y ) F(Y)=\lambda^{s}G(\lambda^{-1}Y) F ( Y ) = λ s G ( λ − 1 Y )
for some λ ∈ K × \lambda\in\mathbb{K}^{\times} λ ∈ K × . In the range n < ℓ < n s / 4 n<\ell<ns/4 n < ℓ < n s /4 , the exponents divisible by n n n occur only through the symmetric combinations
u H j + u H − j . u_H^j+u_H^{-j}. u H j + u H − j .
Thus the natural invariant is not the PGL 2 ( K ) \operatorname{PGL}_2(\mathbb{K}) PGL 2 ( K ) -orbit of u H u_H u H , but rather the PGL 2 ( K ) \operatorname{PGL}_2(\mathbb{K}) PGL 2 ( K ) -orbit of
w H = u H + u H − 1 . w_H=u_H+u_H^{-1}. w H = u H + u H − 1 .
Enumerative consequences
The complete bounds for full Hermitian equivalence were collected in Corollary 6.3. We now state and prove the sharper linearly Hermitian bounds in this subsection.
Recall that
ℓ d = s ( n − d + 1 ) − 2 2 . \ell_d=\frac{s(n-d+1)-2}{2}. ℓ d = 2 s ( n − d + 1 ) − 2 .
and
Λ s , d : = { S ℓ d ( F ) : F ∈ I q , s s r } . \Lambda_{s,d}:=\left\{S_{\ell_d}(F):F\in\mathcal{I}_{q,s}^{\mathrm{sr}}\right\}. Λ s , d := { S ℓ d ( F ) : F ∈ I q , s sr } .
For the short range ℓ d < n \ell_d<n ℓ d < n , the equivalence among members in Λ s , d \Lambda_{s,d} Λ s , d is controlled by the scaling relation in Theorem 6.1 , whereas for the large range n < ℓ d < n s / 4 n<\ell_d<ns/4 n < ℓ d < n s /4 the equivalence is controlled by the self-reciprocal PGL 2 ( F q ) \operatorname{PGL}_2(\mathbb{F}_q) PGL 2 ( F q ) -relation in Theorem 6.2 .
We shall use the following elementary observation several times. If F ∈ I q , s s r F\in\mathcal{I}_{q,s}^{\mathrm{sr}} F ∈ I q , s sr , then
Indeed, since F F F is irreducible of degree s > 1 s>1 s > 1 , neither 1 1 1 nor − 1 -1 − 1 is a root of F F F . The roots of a self-reciprocal polynomial are stable under α ↦ α − 1 \alpha\mapsto\alpha^{-1} α ↦ α − 1 , and therefore they split into inverse pairs. Since s s s is even, the product of all roots is 1 1 1 , and hence the constant coefficient of the monic polynomial F F F is equal to 1 1 1 .
We first treat separately the quadratic case s = 2 s=2 s = 2 .
Corollary 6.18. Assume that s = 2 s=2 s = 2 , n > 2 n>2 n > 2 , and
n 2 < d < n . \frac{n}{2}<d<n. 2 n < d < n . Then Λ s , d \Lambda_{s,d} Λ s , d contains at least
L 2 ( q ) : = { q 2 , q even , q − 1 4 , q ≡ 1 ( m o d 4 ) , q + 1 4 , q ≡ 3 ( m o d 4 ) , L_2(q):=
\begin{cases}
\frac{q}{2}, & q\text{ even},\\
\frac{q-1}{4}, & q\equiv1\pmod{4},\\
\frac{q+1}{4}, & q\equiv3\pmod{4},
\end{cases} L 2 ( q ) := ⎩ ⎨ ⎧ 2 q , 4 q − 1 , 4 q + 1 , q even , q ≡ 1 ( mod 4 ) , q ≡ 3 ( mod 4 ) , pairwise linearly Hermitian inequivalent maximum additive Hermitian d d d -codes in H n ( F q 2 / F q ) \mathcal{H}_n(\mathbb{F}_{q^2}/\mathbb{F}_q) H n ( F q 2 / F q ) .
Proof. For s = 2 s=2 s = 2 , one has
The short-range condition
1 ≤ ℓ < n 2 1\leq\ell<\frac{n}{2} 1 ≤ ℓ < 2 n
is therefore equivalent to
n 2 < d < n . \frac{n}{2}<d<n. 2 n < d < n .
Every monic self-reciprocal polynomial of degree 2 has the form
F c ( Y ) = Y 2 + c Y + 1 , c ∈ F q . F_c(Y)=Y^2+cY+1,\qquad c\in\mathbb{F}_q. F c ( Y ) = Y 2 + c Y + 1 , c ∈ F q .
By (10) , we get
∣ I q , 2 s r ∣ = N q ( 2 ) = { ( q − 1 ) / 2 , q odd , q / 2 , q even . \left|\mathcal{I}^{\mathrm{sr}}_{q,2}\right|=N_q(2)=
\begin{cases}
(q-1)/2, & q\text{ odd},\\
q/2, & q\text{ even}.
\end{cases} I q , 2 sr = N q ( 2 ) = { ( q − 1 ) /2 , q /2 , q odd , q even .
Let F c , F c ′ ∈ I q , 2 s r F_c,F_{c'}\in\mathcal{I}^{\mathrm{sr}}_{q,2} F c , F c ′ ∈ I q , 2 sr . If S n − d ( F c ) S_{n-d}(F_c) S n − d ( F c ) and S n − d ( F c ′ ) S_{n-d}(F_{c'}) S n − d ( F c ′ ) are linearly Hermitian equivalent, then, by Theorem 6.1 , there exists λ ∈ F q × \lambda\in\mathbb{F}_q^{\times} λ ∈ F q × such that
F c ( Y ) = λ 2 F c ′ ( λ − 1 Y ) . F_c(Y)=\lambda^2F_{c'}(\lambda^{-1}Y). F c ( Y ) = λ 2 F c ′ ( λ − 1 Y ) .
Since
λ 2 F c ′ ( λ − 1 Y ) = Y 2 + λ c ′ Y + λ 2 , \lambda^2F_{c'}(\lambda^{-1}Y)=Y^2+\lambda c'Y+\lambda^2, λ 2 F c ′ ( λ − 1 Y ) = Y 2 + λ c ′ Y + λ 2 ,
comparison of the constant terms gives
If q q q is even, then λ = 1 \lambda=1 λ = 1 , and hence c = c ′ c=c' c = c ′ . Therefore distinct polynomials in I q , 2 s r \mathcal{I}^{\mathrm{sr}}_{q,2} I q , 2 sr yield pairwise linearly Hermitian inequivalent codes. This gives q / 2 q/2 q /2 such codes.
Assume now that q q q is odd. Then λ = ± 1 \lambda=\pm1 λ = ± 1 , so F c F_c F c can only be identified with F c F_c F c or with F − c F_{-c} F − c . Thus it is enough to count the orbits of the involution
on the set of parameters c ∈ F q c\in\mathbb{F}_q c ∈ F q for which F c ( Y ) F_c(Y) F c ( Y ) is irreducible. This set has cardinality ( q − 1 ) / 2 (q-1)/2 ( q − 1 ) /2 . The only possible fixed point is c = 0 c=0 c = 0 , which belongs to the set precisely when Y 2 + 1 Y^2+1 Y 2 + 1 is irreducible over F q \mathbb{F}_q F q , equivalently when − 1 -1 − 1 is not a square in F q \mathbb{F}_q F q . This happens exactly when q ≡ 3 ( m o d 4 ) q\equiv3\pmod{4} q ≡ 3 ( mod 4 ) . Hence the number of orbits is
( q − 1 ) / 2 2 = q − 1 4 \frac{(q-1)/2}{2}=\frac{q-1}{4} 2 ( q − 1 ) /2 = 4 q − 1
if q ≡ 1 ( m o d 4 ) q\equiv1\pmod{4} q ≡ 1 ( mod 4 ) , and
( q − 1 ) / 2 + 1 2 = q + 1 4 \frac{(q-1)/2+1}{2}=\frac{q+1}{4} 2 ( q − 1 ) /2 + 1 = 4 q + 1
if q ≡ 3 ( m o d 4 ) q\equiv3\pmod{4} q ≡ 3 ( mod 4 ) . This proves the claim. □
We now turn to the case s ≥ 4 s\geq4 s ≥ 4 . We first consider the short range
In terms of d d d , this condition is
n > 2 , s ( n − d + 1 ) − 2 2 < n . (36) n>2,\qquad\frac{s(n-d+1)-2}{2}<n. \tag*{(36)} n > 2 , 2 s ( n − d + 1 ) − 2 < n . (36)
For s ≥ 4 s \ge4 s ≥ 4 and d ≤ n d \le n d ≤ n , the lower bound
1 ≤ s ( n − d + 1 ) − 2 2 1 \le\frac{s(n-d+1)-2}{2} 1 ≤ 2 s ( n − d + 1 ) − 2
is automatic.
Theorem 6.19. Assume that s ≥ 4 s \ge4 s ≥ 4 is even, n > 2 n > 2 n > 2 , and
ℓ = s ( n − d + 1 ) − 2 2 < n . \ell= \frac{s(n-d+1)-2}{2} < n. ℓ = 2 s ( n − d + 1 ) − 2 < n . Then the family Λ s , d \Lambda_{s,d} Λ s , d contains at least
⌈ N q ( s ) gcd ( s , q − 1 ) ⌉ \left\lceil\frac{N_q(s)}{\gcd(s,q-1)} \right\rceil ⌈ g cd( s , q − 1 ) N q ( s ) ⌉ pairwise linearly Hermitian inequivalent maximum additive Hermitian d d d -codes in H n ( F q s / F q s / 2 ) \mathcal{H}_n(\mathbb{F}_{q^s}/\mathbb{F}_{q^{s/2}}) H n ( F q s / F q s /2 ) .
Proof. Let F , G ∈ I q , s s r F,G \in\mathcal{I}_{q,s}^{\mathrm{sr}} F , G ∈ I q , s sr . Suppose that S ℓ ( F ) \mathcal{S}_{\ell}(F) S ℓ ( F ) and S ℓ ( G ) \mathcal{S}_{\ell}(G) S ℓ ( G ) are linearly Hermitian equivalent. By Theorem 6.1, there exists λ ∈ F q × \lambda\in\mathbb{F}_q^\times λ ∈ F q × such that
F ( Y ) = λ s G ( λ − 1 Y ) . F(Y) = \lambda^sG(\lambda^{-1}Y). F ( Y ) = λ s G ( λ − 1 Y ) .
Since F ( 0 ) = G ( 0 ) = 1 F(0)=G(0)=1 F ( 0 ) = G ( 0 ) = 1 , comparison of the constant terms gives
1 = F ( 0 ) = λ s G ( 0 ) = λ s . 1=F(0)=\lambda^sG(0)=\lambda^s. 1 = F ( 0 ) = λ s G ( 0 ) = λ s .
Thus λ ∈ F q × \lambda\in\mathbb{F}_q^\times λ ∈ F q × must satisfy
For each fixed G G G , there are therefore at most
gcd ( s , q − 1 ) \gcd(s,q-1) g cd( s , q − 1 )
possible polynomials F ∈ I q , s s r F \in\mathcal{I}_{q,s}^{\mathrm{sr}} F ∈ I q , s sr which can give rise to a code linearly Hermitian equivalent to S ℓ ( G ) \mathcal{S}_{\ell}(G) S ℓ ( G ) . Since
∣ I q , s s r ∣ = N q ( s ) , \left|\mathcal{I}_{q,s}^{\mathrm{sr}}\right|=N_q(s), I q , s sr = N q ( s ) ,
the asserted lower bound follows.
We now consider the larger range
n < ℓ < n s 4 . n < \ell< \frac{ns}{4}. n < ℓ < 4 n s .
This range is nonempty only when s > 4 s>4 s > 4 . In terms of d d d , the required condition is
n > 2 , n < s ( n − d + 1 ) − 2 2 < n s 4 . (37) n>2,\qquad n<\frac{s(n-d+1)-2}{2}<\frac{ns}{4}. \tag*{(37)} n > 2 , n < 2 s ( n − d + 1 ) − 2 < 4 n s . (37)
Theorem 6.20. Assume that s > 4 s>4 s > 4 is even, n > 2 n>2 n > 2 , and
n < ℓ = s ( n − d + 1 ) − 2 2 < n s 4 . n<\ell=\frac{s(n-d+1)-2}{2}<\frac{ns}{4}. n < ℓ = 2 s ( n − d + 1 ) − 2 < 4 n s . Then Λ s , d \Lambda_{s,d} Λ s , d contains at least
⌈ N q ( s ) q ( q 2 − 1 ) ⌉ \left\lceil\frac{N_q(s)}{q(q^2-1)} \right\rceil ⌈ q ( q 2 − 1 ) N q ( s ) ⌉ pairwise linearly Hermitian inequivalent maximum additive Hermitian d d d -codes in H n ( F q s / F q s / 2 ) \mathcal{H}_n(\mathbb{F}_{q^s}/\mathbb{F}_{q^{s/2}}) H n ( F q s / F q s /2 ) .
Proof. Let F , G ∈ I q , s s r F,G \in\mathcal{I}^{\mathrm{sr}}_{q,s} F , G ∈ I q , s sr . Suppose that S ℓ ( F ) \mathcal{S}_{\ell}(F) S ℓ ( F ) and S ℓ ( G ) \mathcal{S}_{\ell}(G) S ℓ ( G ) are linearly Hermitian equivalent. Since
n < ℓ < n s 4 , n < \ell< \frac{ns}{4}, n < ℓ < 4 n s ,
Theorem 6.2 provides the existence of a , b , c a,b,c a , b , c and d ∈ K d \in\mathbb{K} d ∈ K with a d − b c ≠ 0 ad-bc \ne0 a d − b c = 0 such that
w G = a w F + b c w F + d . w_G=\frac{aw_F+b}{cw_F+d}. w G = c w F + d a w F + b .
Thus every linear Hermitian equivalence class can contain polynomials from at most one self-reciprocal PGL 2 ( F q ) \operatorname{PGL}_2(\mathbb{F}_q) PGL 2 ( F q ) -orbit. Each such orbit has cardinality at most
∣ PGL 2 ( F q ) ∣ = q ( q 2 − 1 ) . \lvert\operatorname{PGL}_2(\mathbb{F}_q)\rvert=q(q^2-1). ∣ PGL 2 ( F q )∣ = q ( q 2 − 1 ) .
Therefore each linear Hermitian equivalence class contains at most q ( q 2 − 1 ) q(q^2-1) q ( q 2 − 1 ) elements of I q , s s r \mathcal{I}^{\mathrm{sr}}_{q,s} I q , s sr . Since
∣ I q , s s r ∣ = N q ( s ) , \lvert\mathcal{I}^{\mathrm{sr}}_{q,s}\rvert=N_q(s), ∣ I q , s sr ∣ = N q ( s ) ,
the asserted lower bound follows. □
Proof of Corollary 6.3. For part (i), the full Hermitian equivalence relation allows field automorphisms of F q 2 \mathbb{F}_{q^2} F q 2 . Since q = p h q=p^h q = p h ,
∣ Aut ( F q 2 ) ∣ = [ F q 2 : F p ] = 2 h . \lvert\operatorname{Aut}(\mathbb{F}_{q^2})\rvert=[\mathbb{F}_{q^2}:\mathbb{F}_p]=2h. ∣ Aut ( F q 2 )∣ = [ F q 2 : F p ] = 2 h .
Consequently, every full Hermitian equivalence class is a union of at most 2 h 2h 2 h linearly Hermitian equivalence classes. Applying Corollary 6.18 gives the first bound.
For parts (ii) and (iii), the relevant field is F q s \mathbb{F}_{q^s} F q s , and
∣ Aut ( F q s ) ∣ = [ F q s : F p ] = s h . \lvert\operatorname{Aut}(\mathbb{F}_{q^s})\rvert=[\mathbb{F}_{q^s}:\mathbb{F}_p]=sh. ∣ Aut ( F q s )∣ = [ F q s : F p ] = s h .
Thus every full Hermitian equivalence class is a union of at most s h sh s h linear classes. Dividing the bounds in Theorem 6.19 and Theorem 6.20 by this factor, and taking ceilings, gives parts (ii) and (iii), respectively. □
Finally, (10) gives
N q ( s ) = Θ ( q s / 2 ) as q → ∞ . N_q(s)=\Theta(q^{s/2})\quad\text{as }q\to\infty. N q ( s ) = Θ ( q s /2 ) as q → ∞.
Hence the linear counts have order q q q in the quadratic case, order q s / 2 q^{s/2} q s /2 in the short range for fixed even s ≥ 4 s\ge4 s ≥ 4 , and order at least q s / 2 − 3 q^{s/2-3} q s /2 − 3 in the large range for fixed even s > 4 s>4 s > 4 . The corresponding full Hermitian counts lose only the field-automorphism factor: they have orders at least q / h q/h q / h , q s / 2 / h q^{s/2}/h q s /2 / h , and q s / 2 − 3 / h q^{s/2-3}/h q s /2 − 3 / h , respectively. Thus our construction yields many inequivalent new maximum additive Hermitian rank-metric codes, in contrast with the single classical trace construction available for the same parameters.
Acknowledgments
Paolo Santonastaso is partially supported by the Italian National Group for Algebraic and Geometric Structures and their Applications (GNSAGA–INdAM). Yue Zhou is partially supported by the National Natural Science Foundation of China (No. 12371337).
The authors are deeply grateful to the KIMSI Institute in Changsha, China, for hosting them and providing office space and facilities during the visit in which most of this work was completed.
Paolo Santonastaso, Department of Mathematics and Applications “R. Caccioppoli”, University of Naples Federico II, Via Cintia, Monte Sant’Angelo, 80126 Naples, Italy
Email address: paolo.santonastaso@unina.it
Yue Zhou, College of Science, National University of Defense Technology, 410073 Changsha, China
Email address: yue.zhou.ovgu@gmail.com