Introduction
Finite semifields are finite not-necessarily associative unital division algebras. Semifields have received a lot of attention in the last around 100 years, starting with the work of Dickson [6 ] who constructed the first finite semifield that is not a field. Dickson’s motivation to study semifields was purely of algebraic nature. Subsequent inquiries into constructions and properties of semifields were often motivated by applications in both pure and applied mathematics, with semifields for instance exhibiting an intimate relation to non-Desarguesian projective planes in finite geometry [17 ] as well as optimal rank-metric codes [33 ] .
The probably most fundamental question on finite semifields is about their ubiquity. Kantor [15 ] asks if the number of non-isotopic semifields grows superpolynomially in the order of the semifield, providing an affirmative answer in the case of even order semifields [14 ] . In the odd order case, this question is still wide open. Indeed, the current best known bound is not even linear in the order of the semifield [9 ] . This suggests that current constructions of odd order semifields are severely lacking. In particular, mimicking the approach of [14 ] that led to the superpolynomial count for even order semifields seems instructive. One key reason why Kantor and Williams [14 ] were able to achieve these counts was that their construction relied on an iterative approach, starting with a known semifield.
In this paper, we attempt a similar approach, realizing an iterative construction of semifields starting from a known family of semifields. We are however unable to achieve the superpolynomial count using our construction. However, our new construction still yields some very interesting results: Firstly, outside of the Kantor-Williams construction and the Hiramine-Matsumoto-Oyama (HMO) construction [10 ] , our construction is the only iterative construction of semifields, opening the door to future refinements to attack Kantor’s question on the number of odd order non-isotopic semifields. We expect that the techniques we develop here can yield crucial tools for a potential resolution of this question. Secondly, the semifields we construct have the peculiar property that they are always 2-dimensional over a nucleus. While that is also probably a reason why the number of non-isotopic members in this family is not large enough to resolve Kantor’s conjecture, the counts we do achieve show that the number of such semifields is still surprisingly large. Semifields two dimensional over a nucleus have a strong intrinsic structure (see the definition of nuclei later), for instance assuming some extra conditions on the other nuclei, semifields two dimensional over a nucleus have been completely classified [2 ] , with only a few families exhibiting these parameters. Our result shows that without the additional condition on the other nuclei, large additional families are possible. Since nuclei lead for instance immediately to large autotopism groups (as well as collineation groups for the related projective planes), our construction thus yields many non-isotopic examples of highly symmetric semifields and semifield planes.
After the preliminaries, Section 3 establishes our baseline construction and its nucleus calculations. The iterated construction is then developed in Section 4. In Section 5, we give explicit examples of our general construction, and Section 6 treats the isotopy question and establishes bounds on the number of non-isotopic semifields obtained from our construction. Finally, Section 7 compares these semifields with previously known constructions and proves that the family contains new isotopy classes.
Preliminaries
Semifields
Let F \mathbb{F} F be a finite field. A finite-dimensional F \mathbb{F} F -algebra is a finite-dimensional F \mathbb{F} F -vector space S \mathbb{S} S endowed with an F \mathbb{F} F -bilinear map
⋆ : S × S ⟶ S , \star: \mathbb{S} \times\mathbb{S} \longrightarrow\mathbb{S}, ⋆ : S × S ⟶ S ,
which is not assumed to be associative. For a ∈ S a \in\mathbb{S} a ∈ S , we denote by
L a : b ⟼ a ⋆ b , R a : b ⟼ b ⋆ a L_a : b \longmapsto a \star b,\qquad R_a : b \longmapsto b \star a L a : b ⟼ a ⋆ b , R a : b ⟼ b ⋆ a
the left and right multiplication maps. The algebra ( S , + , ⋆ ) (\mathbb{S}, +, \star) ( S , + , ⋆ ) is called a presemifield if L a L_a L a and R a R_a R a are bijective for every a ≠ 0 a \ne0 a = 0 . If, in addition, there is an element $1 \in \mathbb{S}s a t i s f y i n g $ 1 ⋆ a = a ⋆ 1 = a satisfying \$1 \star a = a \star1 = a s a t i s f y in g $1 ⋆ a = a ⋆ 1 = a , for all a ∈ S a \in\mathbb{S} a ∈ S , then S \mathbb{S} S is called a semifield . Thus a finite semifield is a unital finite division algebra. Isotopy, introduced by Albert in [1 ] , is the standard equivalence relation for finite semifields. Two presemifields ( S , + , ⋆ ) (\mathbb{S}, +, \star) ( S , + , ⋆ ) and ( S ′ , + , ⋄ ) (\mathbb{S}', +, \diamond) ( S ′ , + , ⋄ ) defined over the same prime field, are said to be isotopic if there exist additive bijections
f , g , h : S ⟶ S ′ f,g,h : \mathbb{S} \longrightarrow\mathbb{S}' f , g , h : S ⟶ S ′
such that
f ( a ⋆ b ) = g ( a ) ⋄ h ( b ) , f(a \star b) = g(a) \diamond h(b), f ( a ⋆ b ) = g ( a ) ⋄ h ( b ) ,
for all a , b ∈ S a,b \in\mathbb{S} a , b ∈ S . If one can take f = g = h f = g = h f = g = h , then the two presemifields are isomorphic. By Kaplansky’s trick, every finite presemifield is isotopic to a semifield. Therefore, throughout the rest of the paper, we may restrict our attention to semifields.
Let S = ( S , + , ⋆ ) \mathbb{S}=(\mathbb{S},+,\star) S = ( S , + , ⋆ ) be a semifield. Its left, middle, and right nuclei are defined by
N l ( S ) : = { a ∈ S : a ⋆ ( b ⋆ c ) = ( a ⋆ b ) ⋆ c for all b , c ∈ S } , \mathcal{N}_{l}(\mathbb{S}):=\{a\in\mathbb{S}:a\star(b\star c)=(a\star b)\star c\text{ for all }b,c\in\mathbb{S}\}, N l ( S ) := { a ∈ S : a ⋆ ( b ⋆ c ) = ( a ⋆ b ) ⋆ c for all b , c ∈ S } ,
N m ( S ) : = { b ∈ S : a ⋆ ( b ⋆ c ) = ( a ⋆ b ) ⋆ c for all a , c ∈ S } , \mathcal{N}_{m}(\mathbb{S}):=\{b\in\mathbb{S}:a\star(b\star c)=(a\star b)\star c\text{ for all }a,c\in\mathbb{S}\}, N m ( S ) := { b ∈ S : a ⋆ ( b ⋆ c ) = ( a ⋆ b ) ⋆ c for all a , c ∈ S } ,
and
N r ( S ) : = { c ∈ S : a ⋆ ( b ⋆ c ) = ( a ⋆ b ) ⋆ c for all a , b ∈ S } . \mathcal{N}_{r}(\mathbb{S}):=\{c\in\mathbb{S}:a\star(b\star c)=(a\star b)\star c\text{ for all }a,b\in\mathbb{S}\}. N r ( S ) := { c ∈ S : a ⋆ ( b ⋆ c ) = ( a ⋆ b ) ⋆ c for all a , b ∈ S } .
Their intersection
N ( S ) : = N l ( S ) ∩ N m ( S ) ∩ N r ( S ) \mathcal{N}(\mathbb{S}):=\mathcal{N}_{l}(\mathbb{S})\cap\mathcal{N}_{m}(\mathbb{S})\cap\mathcal{N}_{r}(\mathbb{S}) N ( S ) := N l ( S ) ∩ N m ( S ) ∩ N r ( S )
is the nucleus of S \mathbb{S} S . The center is
Z ( S ) : = { a ∈ N ( S ) : a ⋆ b = b ⋆ a for all b ∈ S } . \mathcal{Z}(\mathbb{S}):=\{a\in\mathcal{N}(\mathbb{S}):a\star b=b\star a\text{ for all }b\in\mathbb{S}\}. Z ( S ) := { a ∈ N ( S ) : a ⋆ b = b ⋆ a for all b ∈ S } .
The nuclei and the center are finite fields.
We shall use left multiplication to associate with S \mathbb{S} S its spread set . Put F : = Z ( S ) \mathbb{F}:=\mathcal{Z}(\mathbb{S}) F := Z ( S ) . Since every left multiplication map is F \mathbb{F} F -linear, we may consider
C ( S ) : = { L a : a ∈ S } ⊆ End F ( S ) . \mathcal{C}(\mathbb{S}):=\{L_a:a\in\mathbb{S}\}\subseteq\operatorname{End}_{\mathbb{F}}(\mathbb{S}). C ( S ) := { L a : a ∈ S } ⊆ End F ( S ) .
The map a ↦ L a a\mapsto L_a a ↦ L a is injective and F \mathbb{F} F -linear. Hence C ( S ) \mathcal{C}(\mathbb{S}) C ( S ) is an F \mathbb{F} F -linear subspace of End F ( S ) \operatorname{End}_{\mathbb{F}}(\mathbb{S}) End F ( S ) , of dimension dim F ( S ) \dim_{\mathbb{F}}(\mathbb{S}) dim F ( S ) , and every nonzero element of C ( S ) \mathcal{C}(\mathbb{S}) C ( S ) is nonsingular. Conversely, let V V V be a finite-dimensional vector space over a finite field F \mathbb{F} F , and let C ⊆ End F ( V ) \mathcal{C}\subseteq\operatorname{End}_{\mathbb{F}}(V) C ⊆ End F ( V ) be an F \mathbb{F} F -subspace such that
dim F ( C ) = dim F ( V ) , det ( T ) ≠ 0 for every 0 ≠ T ∈ C . \dim_{\mathbb{F}}(\mathcal{C})=\dim_{\mathbb{F}}(V),\qquad\det(T)\ne0\quad\text{for every }0\ne T\in\mathcal{C}. dim F ( C ) = dim F ( V ) , det ( T ) = 0 for every 0 = T ∈ C .
After choosing an F \mathbb{F} F -linear isomorphism φ : V ⟶ C \varphi:V\longrightarrow\mathcal{C} φ : V ⟶ C , one obtains a presemifield multiplication on V V V by setting
x ⋆ φ y : = φ ( x ) ( y ) , x , y ∈ V . x\star_{\varphi}y:=\varphi(x)(y),\qquad x,y\in V. x ⋆ φ y := φ ( x ) ( y ) , x , y ∈ V .
Different choices of φ \varphi φ give isotopic presemifields.
The spread set of a semifield can be also represented over its right nucleus. Indeed, S \mathbb{S} S is naturally a right vector space over N r ( S ) \mathcal{N}_{r}(\mathbb{S}) N r ( S ) , with scalar multiplication given by the semifield product. Moreover, for every a , b ∈ S a,b\in\mathbb{S} a , b ∈ S and every α ∈ N r ( S ) \alpha\in\mathcal{N}_{r}(\mathbb{S}) α ∈ N r ( S ) , one has
L a ( b ⋆ α ) = a ⋆ ( b ⋆ α ) = ( a ⋆ b ) ⋆ α = L a ( b ) ⋆ α . L_a(b\star\alpha)=a\star(b\star\alpha)=(a\star b)\star\alpha=L_a(b)\star\alpha. L a ( b ⋆ α ) = a ⋆ ( b ⋆ α ) = ( a ⋆ b ) ⋆ α = L a ( b ) ⋆ α .
Thus every L a L_a L a is N r ( S ) \mathcal{N}_{r}(\mathbb{S}) N r ( S ) -linear, and hence
C ( S ) ⊆ End N r ( S ) ( S ) ⊆ End F ( S ) . \mathcal{C}(\mathbb{S})\subseteq\operatorname{End}_{\mathcal{N}_{r}(\mathbb{S})}(\mathbb{S})\subseteq\operatorname{End}_{\mathbb{F}}(\mathbb{S}). C ( S ) ⊆ End N r ( S ) ( S ) ⊆ End F ( S ) .
We shall use the following standard description of isotopy in terms of spread sets.
Proposition 2.1. [21 Theorem 7] , [34 Proposition 3] Let S \mathbb{S} S and S ′ \mathbb{S}' S ′ be semifields over the same right nucleus E \mathbb{E} E , and let C ( S ) , C ( S ′ ) ⊆ End E ( S ) \mathcal{C}(\mathbb{S}),\mathcal{C}(\mathbb{S}')\subseteq\operatorname{End}_{\mathbb{E}}(\mathbb{S}) C ( S ) , C ( S ′ ) ⊆ End E ( S ) be their associated spread sets. Then S \mathbb{S} S and S ′ \mathbb{S}' S ′ are isotopic if and only if there exist A , B ∈ GL E ( S ) A,B\in\operatorname{GL}_{\mathbb{E}}(\mathbb{S}) A , B ∈ GL E ( S ) and ρ ∈ Aut ( E ) \rho\in\operatorname{Aut}(\mathbb{E}) ρ ∈ Aut ( E ) such that
C ( S ′ ) = A C ( S ) ρ B . \mathcal{C}(\mathbb{S}')=A\mathcal{C}(\mathbb{S})^{\rho}B. C ( S ′ ) = A C ( S ) ρ B . Here
C ( S ) ρ : = { T ρ : T ∈ C ( S ) } , \mathcal{C}(\mathbb{S})^{\rho}:=\{T^{\rho}:T\in\mathcal{C}(\mathbb{S})\}, C ( S ) ρ := { T ρ : T ∈ C ( S )} , where, after fixing an E \mathbb{E} E -basis of S \mathbb{S} S , ρ \rho ρ acts entrywise on matrices.
We now specialize to the case that is relevant for the present paper. Let S \mathbb{S} S be a semifield of order q 2 t q^{2t} q 2 t which is two-dimensional over its right nucleus. We identify
N r ( S ) ≅ F q t \mathrm{N}_{r}(\mathbb{S}) \cong\mathbb{F}_{q^{t}} N r ( S ) ≅ F q t
and, as a right F q t \mathbb{F}_{q^{t}} F q t -vector space,
S ≅ F q 2 t . \mathbb{S} \cong\mathbb{F}_{q^{2t}}. S ≅ F q 2 t .
Therefore the spread set of S \mathbb{S} S may be regarded as a set
C ( S ) ⊆ End F q t ( F q 2 t ) \mathcal{C}(\mathbb{S}) \subseteq\operatorname{End}_{\mathbb{F}_{q^{t}}}\left(\mathbb{F}_{q^{2t}}\right) C ( S ) ⊆ End F q t ( F q 2 t )
of q 2 t q^{2t} q 2 t maps, containing the zero map and the identity map, with the property that every nonzero element is nonsingular. Every F q t \mathbb{F}_{q^{t}} F q t -linear endomorphism of F q 2 t \mathbb{F}_{q^{2t}} F q 2 t has a unique expression of the form
T α , β : z ⟼ α z + β z q t , α , β ∈ F q 2 t . T_{\alpha,\beta}:z \longmapsto\alpha z+\beta z^{q^{t}}, \qquad\alpha,\beta\in\mathbb{F}_{q^{2t}}. T α , β : z ⟼ α z + β z q t , α , β ∈ F q 2 t .
The map T α , β T_{\alpha,\beta} T α , β is nonsingular if and only if
N q 2 t / q t ( α ) ≠ N q 2 t / q t ( β ) . \mathrm{N}_{q^{2t}/q^{t}}(\alpha) \ne\mathrm{N}_{q^{2t}/q^{t}}(\beta). N q 2 t / q t ( α ) = N q 2 t / q t ( β ) .
Indeed, suppose first that β ≠ 0 \beta\ne0 β = 0 . A nonzero element z z z lies in the kernel of T α , β T_{\alpha,\beta} T α , β if and only if
z q t − 1 = − α β . z^{q^{t}-1}=-\frac{\alpha}{\beta}. z q t − 1 = − β α .
The image of the map z ↦ z q t − 1 z \mapsto z^{q^{t}-1} z ↦ z q t − 1 on F q 2 t × \mathbb{F}_{q^{2t}}^{\times} F q 2 t × is precisely the subgroup of elements of norm one over F q t \mathbb{F}_{q^{t}} F q t . Hence T α , β T_{\alpha,\beta} T α , β has a nonzero kernel if and only if
N q 2 t / q t ( α β ) = 1 , \mathrm{N}_{q^{2t}/q^{t}}\left(\frac{\alpha}{\beta}\right)=1, N q 2 t / q t ( β α ) = 1 ,
or, equivalently,
N q 2 t / q t ( α ) = N q 2 t / q t ( β ) . \mathrm{N}_{q^{2t}/q^{t}}(\alpha)=\mathrm{N}_{q^{2t}/q^{t}}(\beta). N q 2 t / q t ( α ) = N q 2 t / q t ( β ) .
If β = 0 \beta=0 β = 0 , then T α , 0 T_{\alpha,0} T α , 0 is nonsingular if and only if α ≠ 0 \alpha\ne0 α = 0 , which is again equivalent to the stated norm inequality, since N q 2 t / q t ( 0 ) = 0 \mathrm{N}_{q^{2t}/q^{t}}(0)=0 N q 2 t / q t ( 0 ) = 0 . Consequently, the spread set of any semifield that is two-dimensional over its right nucleus can be written as
S μ , ζ : = { T μ ( u ) , ζ ( u ) : u ∈ F q 2 t } ⊆ End F q t ( F q 2 t ) , \mathcal{S}_{\mu,\zeta}:=\left\{T_{\mu(u),\zeta(u)}:u\in\mathbb{F}_{q^{2t}}\right\}\subseteq\operatorname{End}_{\mathbb{F}_{q^{t}}}\left(\mathbb{F}_{q^{2t}}\right), S μ , ζ := { T μ ( u ) , ζ ( u ) : u ∈ F q 2 t } ⊆ End F q t ( F q 2 t ) ,
where μ , ζ : F q 2 t → F q 2 t \mu,\zeta:\mathbb{F}_{q^{2t}}\to\mathbb{F}_{q^{2t}} μ , ζ : F q 2 t → F q 2 t are additive maps such that u ↦ ( μ ( u ) , ζ ( u ) ) u\mapsto(\mu(u),\zeta(u)) u ↦ ( μ ( u ) , ζ ( u )) is injective, and each nonzero element of S μ , ζ \mathcal{S}_{\mu,\zeta} S μ , ζ is nonsingular. Equivalently, for every u ∈ F q 2 t u\in\mathbb{F}_{q^{2t}} u ∈ F q 2 t with ( μ ( u ) , ζ ( u ) ) ≠ ( 0 , 0 ) (\mu(u),\zeta(u))\ne(0,0) ( μ ( u ) , ζ ( u )) = ( 0 , 0 ) , one has
N q 2 t / q t ( μ ( u ) ) ≠ N q 2 t / q t ( ζ ( u ) ) . \mathrm{N}_{q^{2t}/q^{t}}(\mu(u))\ne\mathrm{N}_{q^{2t}/q^{t}}(\zeta(u)). N q 2 t / q t ( μ ( u )) = N q 2 t / q t ( ζ ( u )) .
Skew polynomial rings
Let K \mathbb{K} K be a finite field and let L / K \mathbb{L}/\mathbb{K} L / K be a finite field extension with [ L : K ] = m [\mathbb{L}:\mathbb{K}]=m [ L : K ] = m . 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 ] .
Thus, R R R consists of polynomials in the indeterminate X X X with coefficients in L \mathbb{L} L , written on the left, and multiplication is determined 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 as soon as σ ≠ id \sigma\ne\operatorname{id} σ = id . Nevertheless, R R R retains many of the basic structural features of an ordinary polynomial ring. In particular, every nonzero element f ∈ R f\in R f ∈ R has a well-defined degree, denoted by deg ( f ) \deg(f) deg ( f ) , and for all nonzero f , g ∈ R f,g\in R f , g ∈ R one has
deg ( f g ) = deg ( f ) + deg ( g ) , deg ( f + g ) ≤ max { deg ( f ) , deg ( g ) } . \deg(fg)=\deg(f)+\deg(g),\qquad\deg(f+g)\le\max\{\deg(f),\deg(g)\}. deg ( f g ) = deg ( f ) + deg ( g ) , deg ( f + g ) ≤ max { deg ( f ) , deg ( g )} .
Moreover, R R R is both a left and a right Euclidean domain. In particular, given f , g ∈ R f,g\in R f , g ∈ R with g ≠ 0 g\ne0 g = 0 , there exist unique polynomials q , r ∈ R q,r\in R q , r ∈ R such that
f = q g + r , with either r = 0 or deg ( r ) < deg ( g ) . f=qg+r,\qquad\text{with either }r=0\text{ or }\deg(r)<\deg(g). f = q g + r , with either r = 0 or deg ( r ) < deg ( g ) .
If 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 a left multiple of g g g . As a consequence, every left ideal of R R R is principal. Hence, for any two nonzero elements f 1 , f 2 ∈ R f_1,f_2\in R f 1 , f 2 ∈ R , one may define their (monic) greatest common right divisor and least common left multiple, denoted by
gcrd ( f 1 , f 2 ) and lclm ( f 1 , f 2 ) , \operatorname{gcrd}(f_1,f_2)\qquad\text{and}\qquad\operatorname{lclm}(f_1,f_2), gcrd ( f 1 , f 2 ) and lclm ( f 1 , f 2 ) ,
respectively. 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 and R f 1 ∩ R f 2 = R b . Rf_1+Rf_2=Ra\qquad\text{and}\qquad Rf_1\cap Rf_2=Rb. R f 1 + R f 2 = R a and R f 1 ∩ R f 2 = R b .
It is well known that the center of R R R is
Z ( R ) = K [ X m ] . Z(R)=\mathbb{K}[X^m]. Z ( R ) = K [ X m ] .
Now let F ( Y ) ∈ K [ Y ] F(Y)\in\mathbb{K}[Y] F ( Y ) ∈ K [ Y ] be a monic irreducible polynomial of degree s ≥ 1 s\ge1 s ≥ 1 , with F ( Y ) ≠ Y F(Y)\ne Y F ( Y ) = Y . Since F ( X m ) ∈ Z ( R ) F(X^m)\in Z(R) F ( X m ) ∈ Z ( R ) , the ideal R F ( X m ) RF(X^m) R F ( X m ) is two-sided, and we may consider the quotient ring
R F : = R R F ( X m ) . R_F:=\frac{R}{RF(X^m)}. R F := R F ( X m ) R .
Every element of R F R_F R F admits a unique representative of degree strictly smaller than m s ms m s , so we identify each class a ‾ ∈ R F \overline{a}\in R_F a ∈ R F with its canonical representative of degree at most m s − 1 ms-1 m s − 1 . Accordingly, when a ‾ ≠ 0 \overline{a}\ne0 a = 0 , we define deg ( a ‾ ) : = deg ( a ) \deg(\overline{a}):=\deg(a) deg ( a ) := deg ( a ) , where a a a is the canonical representative of a ‾ \overline{a} a .
The center of R F R_F R F will be denoted by E F \mathbb{E}_F E F . Since F ( X m ) F(X^m) F ( X m ) is central, one has
E F = Z ( R F ) ≅ K [ Y ] K [ Y ] F ( Y ) . \mathbb{E}_F=Z(R_F)\cong\frac{\mathbb{K}[Y]}{\mathbb{K}[Y]F(Y)}. E F = Z ( R F ) ≅ K [ Y ] F ( Y ) K [ Y ] .
Because F ( Y ) F(Y) F ( Y ) is irreducible over K \mathbb{K} K , it follows that E F \mathbb{E}_F E F is a field extension of K \mathbb{K} K of degree s s s , and hence
E F ≅ F ∣ K ∣ s . \mathbb{E}_F\cong\mathbb{F}_{|\mathbb{K}|^s}. E F ≅ F ∣ K ∣ s .
Furthermore, the ideal R F ( X m ) RF(X^m) R F ( X m ) is maximal among the two-sided ideals of R R R . Therefore R F R_F R F is a central simple algebra over E F E_F E F , of dimension m 2 m^2 m 2 over E F E_F E F , and thus of dimension m 2 s m^2s m 2 s over K \mathbb{K} K . By the Artin–Wedderburn theorem, one obtains an E F E_F E F -algebra isomorphism
R F ≅ M m ( E F ) . (1) R_F \cong M_m(E_F). \tag*{(1)} R F ≅ M m ( E F ) . (1)
See, for instance, [13 Theorem 1.2.19] and [28 Theorem 20] for further details.
Fix an E F E_F E F -algebra isomorphism
φ F : R F ⟶ M m ( E F ) . \varphi_F : R_F \longrightarrow M_m(E_F). φ F : R F ⟶ M m ( E F ) .
For every a ˉ ∈ R F \bar{a} \in R_F a ˉ ∈ R F , we define
rk ( a ˉ ) : = rank E F ( φ F ( a ˉ ) ) . \operatorname{rk}(\bar{a}) := \operatorname{rank}_{E_F}\left(\varphi_F(\bar{a})\right). rk ( a ˉ ) := rank E F ( φ F ( a ˉ ) ) .
The following result expresses rk ( a ˉ ) \operatorname{rk}(\bar{a}) rk ( a ˉ ) in terms of a greatest common right divisor in R R R .
Theorem 2.2 (see [34 Proposition 7] ). Let a ˉ = a + R F ( X m ) ∈ R F \bar{a}=a+RF(X^m)\in R_F a ˉ = a + R F ( X m ) ∈ R F be nonzero. Then
rk ( a ˉ ) = m − 1 s deg ( gcrd ( a , F ( X m ) ) ) . \operatorname{rk}(\bar{a})=m-\frac{1}{s}\deg\left(\operatorname{gcrd}\left(a,F(X^m)\right)\right). rk ( a ˉ ) = m − s 1 deg ( gcrd ( a , F ( X m ) ) ) .
A lifting construction via skew polynomial rings
In this section we describe the main lifting construction of the paper. We start from a semifield of order q 2 t q^{2t} q 2 t which is two-dimensional over its right nucleus, and we use a skew-polynomial quotient to construct a semifield of order q 2 t s q^{2ts} q 2 t s . The idea is to embed the spread set of the initial semifield into the algebra
R G : = R R G ( X 2 ) . R_G:=\frac{R}{RG(X^2)}. R G := R G ( X 2 ) R .
where G G G is an irreducible polynomial over F q t \mathbb{F}_{q^t} F q t .
Let n = 2 t n=2t n = 2 t , with t ≥ 1 t\geq1 t ≥ 1 , and put
K = F q t , L = F q n = F q 2 t . \mathbb{K}=\mathbb{F}_{q^t},\qquad\mathbb{L}=\mathbb{F}_{q^n}=\mathbb{F}_{q^{2t}}. K = F q t , L = F q n = F q 2 t .
Let σ : L ⟶ L \sigma:\mathbb{L}\longrightarrow\mathbb{L} σ : L ⟶ L be a generator of Gal ( L / K ) \operatorname{Gal}(\mathbb{L}/\mathbb{K}) Gal ( L / K ) . We consider the skew polynomial ring R = L [ X ; σ ] R=\mathbb{L}[X;\sigma] R = L [ X ; σ ] . Since [ L : K ] = 2 [\mathbb{L}:\mathbb{K}]=2 [ L : K ] = 2 , the center of R R R is Z ( R ) = K [ X 2 ] Z(R)=\mathbb{K}[X^2] Z ( R ) = K [ X 2 ] .
Let G ( Y ) ∈ K [ Y ] G(Y)\in\mathbb{K}[Y] G ( Y ) ∈ K [ Y ] be a monic irreducible polynomial of degree s ≥ 1 s\geq1 s ≥ 1 , with G ( Y ) ≠ Y G(Y)\ne Y G ( Y ) = Y . Since G ( X 2 ) ∈ Z ( R ) G(X^2)\in Z(R) G ( X 2 ) ∈ Z ( R ) , the ideal R G ( X 2 ) RG(X^2) R G ( X 2 ) is two-sided, and the quotient
R G = R R G ( X 2 ) R_G=\frac{R}{RG(X^2)} R G = R G ( X 2 ) R
is a central simple algebra over
Z ( R G ) ≅ K [ Y ] K [ Y ] G ( Y ) ≅ F q t s . Z(R_G)\cong\frac{\mathbb{K}[Y]}{\mathbb{K}[Y]G(Y)}\cong\mathbb{F}_{q^{ts}}. Z ( R G ) ≅ K [ Y ] G ( Y ) K [ Y ] ≅ F q t s .
By (1), we have an isomorphism of Z ( R G ) Z(R_G) Z ( R G ) -algebras
R G ≅ M 2 ( F q t s ) . R_G\cong M_2(\mathbb{F}_{q^{ts}}). R G ≅ M 2 ( F q t s ) .
We shall use the following rank-one criterion, which follows from Theorem 2.2 and [34 Theorem 6] .
Theorem 3.1. Let a ‾ = α 0 + α 1 X + ⋯ + α s X s + R G ( X 2 ) ∈ R G \overline{a}=\alpha_0+\alpha_1X+\cdots+\alpha_sX^s+RG(X^2)\in R_G a = α 0 + α 1 X + ⋯ + α s X s + R G ( X 2 ) ∈ R G be nonzero. Then rk ( a ‾ ) ≥ 1 \operatorname{rk}(\overline{a})\ge1 rk ( a ) ≥ 1 . Moreover, if rk ( a ‾ ) = 1 \operatorname{rk}(\overline{a})=1 rk ( a ) = 1 , then deg ( a ‾ ) = s \operatorname{deg}(\overline{a})=s deg ( a ) = s and
N q 2 t / q t ( α 0 ) N q 2 t / q t ( α s ) = ( − 1 ) s G ( 0 ) . \frac{N_{q^{2t}/q^t}(\alpha_0)}{N_{q^{2t}/q^t}(\alpha_s)}=(-1)^sG(0). N q 2 t / q t ( α s ) N q 2 t / q t ( α 0 ) = ( − 1 ) s G ( 0 ) .
The lifted spread set
We now recall the starting point of the construction. Let T ( Y ) = Y − 1 ∈ K [ Y ] T(Y)=Y-1\in\mathbb{K}[Y] T ( Y ) = Y − 1 ∈ K [ Y ] .
Then
R R T ( X 2 ) = R R ( X 2 − 1 ) ≅ M 2 ( F q t ) ≅ End F q t ( F q 2 t ) . \frac{R}{RT(X^2)}=\frac{R}{R(X^2-1)}\cong M_2(\mathbb{F}_{q^t})\cong\operatorname{End}_{\mathbb{F}_{q^t}}(\mathbb{F}_{q^{2t}}). R T ( X 2 ) R = R ( X 2 − 1 ) R ≅ M 2 ( F q t ) ≅ End F q t ( F q 2 t ) .
Under this identification, the class of α + β X + R T ( X 2 ) \alpha+\beta X+RT(X^2) α + β X + R T ( X 2 ) corresponds to the F q t \mathbb{F}_{q^t} F q t -linear map
z ⟼ α z + β z q t . z\longmapsto\alpha z+\beta z^{q^t}. z ⟼ α z + β z q t .
Let S μ , ζ S_{\mu,\zeta} S μ , ζ be the spread set of a semifield of order q 2 t q^{2t} q 2 t which is two-dimensional over its right nucleus. We write
S μ , ζ = { μ ( u ) + ζ ( u ) X + R ( X 2 − 1 ) : u ∈ F q 2 t } ⊆ R R ( X 2 − 1 ) , S_{\mu,\zeta}=\{\mu(u)+\zeta(u)X+R(X^2-1):u\in\mathbb{F}_{q^{2t}}\}\subseteq\frac{R}{R(X^2-1)}, S μ , ζ = { μ ( u ) + ζ ( u ) X + R ( X 2 − 1 ) : u ∈ F q 2 t } ⊆ R ( X 2 − 1 ) R ,
where
μ , ζ : F q 2 t ⟶ F q 2 t \mu,\zeta:\mathbb{F}_{q^{2t}}\longrightarrow\mathbb{F}_{q^{2t}} μ , ζ : F q 2 t ⟶ F q 2 t
are additive maps. Since S μ , ζ S_{\mu,\zeta} S μ , ζ is a spread set, the map
(2)
Δ μ , ζ : F q 2 t ⟶ F q 2 t 2 , u ⟼ ( μ ( u ) , ζ ( u ) ) , (2) \Delta_{\mu,\zeta}:\mathbb{F}_{q^{2t}}\longrightarrow\mathbb{F}_{q^{2t}}^2,\qquad u\longmapsto(\mu(u),\zeta(u)), \tag*{(2)} Δ μ , ζ : F q 2 t ⟶ F q 2 t 2 , u ⟼ ( μ ( u ) , ζ ( u )) , (2)
is injective, and every nonzero element of S μ , ζ S_{\mu,\zeta} S μ , ζ is nonsingular. Equivalently, for every u ∈ F q 2 t u\in\mathbb{F}_{q^{2t}} u ∈ F q 2 t with ζ ( u ) ≠ 0 \zeta(u)\ne0 ζ ( u ) = 0 , one has
(3)
N q 2 t / q t ( μ ( u ) ) ≠ N q 2 t / q t ( ζ ( u ) ) . (3) N_{q^{2t}/q^t}(\mu(u))\ne N_{q^{2t}/q^t}(\zeta(u)). \tag*{(3)} N q 2 t / q t ( μ ( u )) = N q 2 t / q t ( ζ ( u )) . (3)
Let η ∈ F q 2 t \eta\in\mathbb{F}_{q^{2t}} η ∈ F q 2 t be such that
N F q 2 t / F q t ( η ) = ( − 1 ) s G ( 0 ) . N_{\mathbb{F}_{q^{2t}}/\mathbb{F}_{q^t}}(\eta)=(-1)^sG(0). N F q 2 t / F q t ( η ) = ( − 1 ) s G ( 0 ) .
We define the lifted set associated with G G G and η \eta η as
(4)
Φ G , η ( S μ , ζ ) : = { μ ( u 0 ) + ∑ i = 1 s − 1 u i X i + ζ ( u 0 ) η X s + R G ( X 2 ) : u 0 , … , u s − 1 ∈ F q 2 t } ⊆ R G . (4) \Phi_{G,\eta}(S_{\mu,\zeta}):=\left\{\mu(u_0)+\sum_{i=1}^{s-1}u_iX^i+\frac{\zeta(u_0)}{\eta}X^s+RG(X^2):u_0,\ldots,u_{s-1}\in\mathbb{F}_{q^{2t}}\right\}\subseteq R_G. \tag*{(4)} Φ G , η ( S μ , ζ ) := { μ ( u 0 ) + i = 1 ∑ s − 1 u i X i + η ζ ( u 0 ) X s + R G ( X 2 ) : u 0 , … , u s − 1 ∈ F q 2 t } ⊆ R G . (4)
Theorem 3.2. If S μ , ζ S_{\mu,\zeta} S μ , ζ is the spread set of a semifield of order q 2 t q^{2t} q 2 t , then Φ G , η ( S μ , ζ ) \Phi_{G,\eta}(S_{\mu,\zeta}) Φ G , η ( S μ , ζ ) is the spread set of a semifield of order q 2 t s q^{2ts} q 2 t s .
Proof. First, we prove that the lifted set has the correct size. Since
S μ , ζ = { μ ( u ) + ζ ( u ) X + R ( X 2 − 1 ) : u ∈ F q 2 t } S_{\mu,\zeta}=\{\mu(u)+\zeta(u)X+R(X^2-1):u\in\mathbb{F}_{q^{2t}}\} S μ , ζ = { μ ( u ) + ζ ( u ) X + R ( X 2 − 1 ) : u ∈ F q 2 t }
is a spread set, the map Δ μ , ζ : u ⟼ ( μ ( u ) , ζ ( u ) ) \Delta_{\mu,\zeta}:u\longmapsto(\mu(u),\zeta(u)) Δ μ , ζ : u ⟼ ( μ ( u ) , ζ ( u )) is injective. Hence the parameters u 0 , u 1 , … , u s − 1 ∈ F q 2 t u_0,u_1,\ldots,u_{s-1}\in\mathbb{F}_{q^{2t}} u 0 , u 1 , … , u s − 1 ∈ F q 2 t determine uniquely an element of Φ G , η ( S μ , ζ ) \Phi_{G,\eta}(S_{\mu,\zeta}) Φ G , η ( S μ , ζ ) . Therefore
∣ Φ G , η ( S μ , ζ ) ∣ = q 2 t s . \left|\Phi_{G,\eta}(S_{\mu,\zeta})\right|=q^{2ts}. ∣ Φ G , η ( S μ , ζ ) ∣ = q 2 t s .
Moreover, the set is additive, because μ \mu μ and ζ \zeta ζ are additive and the coefficients u 1 , … , u s − 1 u_1,\ldots,u_{s-1} u 1 , … , u s − 1 run over F q n \mathbb{F}_{q^n} F q n . Since the original spread set contains the identity map, there exists e ∈ F q 2 t e\in\mathbb{F}_{q^{2t}} e ∈ F q 2 t such that μ ( e ) = 1 \mu(e)=1 μ ( e ) = 1 and ζ ( e ) = 0 \zeta(e)=0 ζ ( e ) = 0 . Hence
1 + R G ( X 2 ) ∈ Φ G , η ( S μ , ζ ) . 1+RG(X^2)\in\Phi_{G,\eta}(S_{\mu,\zeta}). 1 + R G ( X 2 ) ∈ Φ G , η ( S μ , ζ ) .
It remains to prove that every nonzero element of Φ G , η ( S μ , ζ ) \Phi_{G,\eta}(S_{\mu,\zeta}) Φ G , η ( S μ , ζ ) is nonsingular in R G ≅ M 2 ( F q t s ) RG\cong M_2(\mathbb{F}_{q^{ts}}) R G ≅ M 2 ( F q t s ) . Suppose, by contradiction, that there exists a nonzero singular element
a ˉ = μ ( u 0 ) + ∑ i = 1 s − 1 u i X i + ζ ( u 0 ) η X s + R G ( X 2 ) \bar{a}=\mu(u_0)+\sum_{i=1}^{s-1}u_iX^i+\frac{\zeta(u_0)}{\eta}X^s+RG(X^2) a ˉ = μ ( u 0 ) + i = 1 ∑ s − 1 u i X i + η ζ ( u 0 ) X s + R G ( X 2 )
in the lifted set. Since a ˉ ≠ 0 \bar{a}\ne0 a ˉ = 0 , we have rk ( a ˉ ) ≥ 1 \operatorname{rk}(\bar{a})\ge1 rk ( a ˉ ) ≥ 1 . As a ˉ \bar{a} a ˉ is singular in M 2 ( F q t s ) M_2(\mathbb{F}_{q^{ts}}) M 2 ( F q t s ) , we must have
rk ( a ˉ ) = 1. \operatorname{rk}(\bar{a})=1. rk ( a ˉ ) = 1.
Therefore, by Theorem 3.1 , the coefficient of X s X^s X s is nonzero, so ζ ( u 0 ) ≠ 0 \zeta(u_0)\ne0 ζ ( u 0 ) = 0 , and
N F q 2 t / F q t ( μ ( u 0 ) ) N F q 2 t / F q t ( ζ ( u 0 ) / η ) = ( − 1 ) s G ( 0 ) . \frac{\mathrm{N}_{\mathbb{F}_{q^{2t}}/\mathbb{F}_{q^t}}(\mu(u_0))}{\mathrm{N}_{\mathbb{F}_{q^{2t}}/\mathbb{F}_{q^t}}(\zeta(u_0)/\eta)}=(-1)^sG(0). N F q 2 t / F q t ( ζ ( u 0 ) / η ) N F q 2 t / F q t ( μ ( u 0 )) = ( − 1 ) s G ( 0 ) .
Since
N F q 2 t / F q t ( ζ ( u 0 ) η ) = N F q 2 t / F q t ( ζ ( u 0 ) ) N F q 2 t / F q t ( η ) , \mathrm{N}_{\mathbb{F}_{q^{2t}}/\mathbb{F}_{q^t}}\left(\frac{\zeta(u_0)}{\eta}\right)=\frac{\mathrm{N}_{\mathbb{F}_{q^{2t}}/\mathbb{F}_{q^t}}(\zeta(u_0))}{\mathrm{N}_{\mathbb{F}_{q^{2t}}/\mathbb{F}_{q^t}}(\eta)}, N F q 2 t / F q t ( η ζ ( u 0 ) ) = N F q 2 t / F q t ( η ) N F q 2 t / F q t ( ζ ( u 0 )) ,
we get
N F q 2 t / F q t ( μ ( u 0 ) ) N F q 2 t / F q t ( ζ ( u 0 ) ) N F q 2 t / F q t ( η ) = ( − 1 ) s G ( 0 ) . \frac{\mathrm{N}_{\mathbb{F}_{q^{2t}}/\mathbb{F}_{q^t}}(\mu(u_0))}{\mathrm{N}_{\mathbb{F}_{q^{2t}}/\mathbb{F}_{q^t}}(\zeta(u_0))}\mathrm{N}_{\mathbb{F}_{q^{2t}}/\mathbb{F}_{q^t}}(\eta)=(-1)^sG(0). N F q 2 t / F q t ( ζ ( u 0 )) N F q 2 t / F q t ( μ ( u 0 )) N F q 2 t / F q t ( η ) = ( − 1 ) s G ( 0 ) .
Using N F q 2 t / F q t ( η ) = ( − 1 ) s G ( 0 ) \mathrm{N}_{\mathbb{F}_{q^{2t}}/\mathbb{F}_{q^t}}(\eta)=(-1)^sG(0) N F q 2 t / F q t ( η ) = ( − 1 ) s G ( 0 ) , we obtain
N F q 2 t / F q t ( μ ( u 0 ) ) = N F q 2 t / F q t ( ζ ( u 0 ) ) . \mathrm{N}_{\mathbb{F}_{q^{2t}}/\mathbb{F}_{q^t}}(\mu(u_0))=\mathrm{N}_{\mathbb{F}_{q^{2t}}/\mathbb{F}_{q^t}}(\zeta(u_0)). N F q 2 t / F q t ( μ ( u 0 )) = N F q 2 t / F q t ( ζ ( u 0 )) .
But ζ ( u 0 ) ≠ 0 \zeta(u_0)\ne0 ζ ( u 0 ) = 0 , and this contradicts the nonsingularity of the original spread set S μ , ζ S_{\mu,\zeta} S μ , ζ . Therefore, every nonzero element of Φ G , η ( S μ , ζ ) \Phi_{G,\eta}(S_{\mu,\zeta}) Φ G , η ( S μ , ζ ) is nonsingular. Thus Φ G , η ( S μ , ζ ) \Phi_{G,\eta}(S_{\mu,\zeta}) Φ G , η ( S μ , ζ ) is an additive spread set of size q 2 t s q^{2ts} q 2 t s , and hence it defines a semifield of order q 2 t s q^{2ts} q 2 t s .
Remark 3.3. The construction also applies to presemifield spread sets in End F q t ( F q 2 t ) \operatorname{End}_{\mathbb{F}_{q^t}}(\mathbb{F}_{q^{2t}}) End F q t ( F q 2 t ) . Indeed, the assumption that the initial spread set contains the identity is used only to ensure that the lifted set also contains the identity. Without this assumption, the same proof shows that the lifted set is an additive spread set of cardinality q 2 t s q^{2ts} q 2 t s , and hence defines a presemifield of order q 2 t s q^{2ts} q 2 t s .
More generally, the additivity of μ \mu μ and ζ \zeta ζ can be dropped provided that the difference of any two distinct elements of S μ , ζ S_{\mu,\zeta} S μ , ζ is nonsingular. Explicitly, suppose that μ , ζ : F q 2 t → F q 2 t \mu,\zeta:\mathbb{F}_{q^{2t}}\to\mathbb{F}_{q^{2t}} μ , ζ : F q 2 t → F q 2 t satisfy μ ( 0 ) = ζ ( 0 ) = 0 \mu(0)=\zeta(0)=0 μ ( 0 ) = ζ ( 0 ) = 0 and
N q 2 t / q t ( μ ( u ) − μ ( v ) ) ≠ N q 2 t / q t ( ζ ( u ) − ζ ( v ) ) for all u ≠ v . \mathrm{N}_{q^{2t}/q^t}(\mu(u)-\mu(v))\ne\mathrm{N}_{q^{2t}/q^t}(\zeta(u)-\zeta(v))\qquad\text{for all }u\ne v. N q 2 t / q t ( μ ( u ) − μ ( v )) = N q 2 t / q t ( ζ ( u ) − ζ ( v )) for all u = v .
*Then the same formula defining Φ G , η ( S μ , ζ ) \Phi_{G,\eta}(S_{\mu,\zeta}) Φ G , η ( S μ , ζ ) produces a set of cardinality q 2 t s q^{2ts} q 2 t s containing zero, in which the difference of any two distinct elements is nonsingular. Thus the construction extends to quasifield spread sets in M 2 ( F q t ) M_2(\mathbb{F}_{q^t}) M 2 ( F q t ) , producing spread sets in M 2 ( F q t s ) M_2(\mathbb{F}_{q^{ts}}) M 2 ( F q t s ) and, after normalization if necessary, quasifields of order q 2 t s q^{2ts} q 2 t s .
Remark 3.4. Notice that, when s s s is odd, the choice of the element η \eta η satisfying the prescribed norm condition does not affect the isotopy class of the resulting semifield. Indeed, let η , η ′ ∈ F q 2 t \eta,\eta' \in\mathbb{F}_{q^{2t}} η , η ′ ∈ F q 2 t be such that
N F q 2 t / F q ( η ) = N F q 2 t / F q ( η ′ ) = ( − 1 ) s G ( 0 ) . \mathrm{N}_{\mathbb{F}_{q^{2t}}/\mathbb{F}_{q}}(\eta)=\mathrm{N}_{\mathbb{F}_{q^{2t}}/\mathbb{F}_{q}}(\eta')=(-1)^sG(0). N F q 2 t / F q ( η ) = N F q 2 t / F q ( η ′ ) = ( − 1 ) s G ( 0 ) .
Then d = η / η ′ d=\eta/\eta' d = η / η ′ has norm one over F q t \mathbb{F}_{q^t} F q t , and hence, by Hilbert’s Theorem 90, there exists c ∈ F q 2 t ∗ c\in\mathbb{F}_{q^{2t}}^{*} c ∈ F q 2 t ∗ such that
d = c − 1 σ ( c ) . d=c^{-1}\sigma(c). d = c − 1 σ ( c ) .
Conjugation by c c c in R G R_G R G fixes the even powers of X X X , while it multiplies the coefficients of the odd powers by d d d . Since s s s is odd, the coefficient of X s X^s X s is therefore transformed as
ζ ( u 0 ) η ⟼ d ζ ( u 0 ) η = ζ ( u 0 ) η ′ . \frac{\zeta(u_0)}{\eta}\longmapsto d\frac{\zeta(u_0)}{\eta}=\frac{\zeta(u_0)}{\eta'}. η ζ ( u 0 ) ⟼ d η ζ ( u 0 ) = η ′ ζ ( u 0 ) .
On the other hand, the coefficients of X , … , X s − 1 X,\ldots,X^{s-1} X , … , X s − 1 range independently over F q 2 t \mathbb{F}_{q^{2t}} F q 2 t , so multiplying the odd-indexed ones by the nonzero scalar d d d does not change their set of possible values. It follows that
c − 1 Φ G , η ( S μ , ζ ) c = Φ G , η ′ ( S μ , ζ ) . c^{-1}\Phi_{G,\eta}(S_{\mu,\zeta})c=\Phi_{G,\eta'}(S_{\mu,\zeta}). c − 1 Φ G , η ( S μ , ζ ) c = Φ G , η ′ ( S μ , ζ ) .
Thus different admissible choices of η \eta η yield equivalent spread sets, and hence isotopic semifields.
The left and middle nuclei
In the following, we compute the left and middle nuclei of the semifields obtained from the lifted spread set Φ G , η ( S μ , ζ ) \Phi_{G,\eta}(S_{\mu,\zeta}) Φ G , η ( S μ , ζ ) . These computations can be performed directly on the spread set by using the standard idealiser description of nuclei.
Theorem 3.5 (see [25 Theorem 1] and [34 Proposition 5] ). Let S S S be a semifield and let C ( S ) ⊆ End F ( S ) \mathcal{C}(S)\subseteq\operatorname{End}_{\mathbb{F}}(S) C ( S ) ⊆ End F ( S ) be its spread set. Then
N l ( S ) ≅ { φ ∈ End F ( S ) : φ ∘ ψ ∈ C ( S ) for every ψ ∈ C ( S ) } , N_l(S)\cong\{\varphi\in\operatorname{End}_{\mathbb{F}}(S):\varphi\circ\psi\in\mathcal{C}(S)\text{ for every }\psi\in\mathcal{C}(S)\}, N l ( S ) ≅ { φ ∈ End F ( S ) : φ ∘ ψ ∈ C ( S ) for every ψ ∈ C ( S )} , and
N m ( S ) ≅ { φ ∈ End F ( S ) : ψ ∘ φ ∈ C ( S ) for every ψ ∈ C ( S ) } . N_m(S)\cong\{\varphi\in\operatorname{End}_{\mathbb{F}}(S):\psi\circ\varphi\in\mathcal{C}(S)\text{ for every }\psi\in\mathcal{C}(S)\}. N m ( S ) ≅ { φ ∈ End F ( S ) : ψ ∘ φ ∈ C ( S ) for every ψ ∈ C ( S )} .
Let
C G , η : = Φ G , η ( S μ , ζ ) . C_{G,\eta}:=\Phi_{G,\eta}(S_{\mu,\zeta}). C G , η := Φ G , η ( S μ , ζ ) .
Thus
C G , η = { μ ( u 0 ) + ∑ i = 1 s − 1 u i X i + ζ ( u 0 ) η X s + R G ( X 2 ) : u 0 , … , u s − 1 ∈ F q 2 t } . C_{G,\eta}=\left\{\mu(u_0)+\sum_{i=1}^{s-1}u_iX^i+\frac{\zeta(u_0)}{\eta}X^s+R_G(X^2):u_0,\ldots,u_{s-1}\in\mathbb{F}_{q^{2t}}\right\}. C G , η = { μ ( u 0 ) + i = 1 ∑ s − 1 u i X i + η ζ ( u 0 ) X s + R G ( X 2 ) : u 0 , … , u s − 1 ∈ F q 2 t } .
Theorem 3.6. Assume s ≥ 2 s\geq2 s ≥ 2 , and let S G , η S_{G,\eta} S G , η be the semifield associated with the spread set C G , η C_{G,\eta} C G , η . Then
N l ( S G , η ) ≅ N l ( μ , ζ ) , N_l(S_{G,\eta})\cong\mathcal{N}_l(\mu,\zeta), N l ( S G , η ) ≅ N l ( μ , ζ ) , where
N l ( μ , ζ ) = { α ∈ F q 2 t : ∀ u ∈ F q 2 t , ∃ v ∈ F q 2 t such that ( μ ( v ) , ζ ( v ) ) = ( α μ ( u ) , α ζ ( u ) ) } . \mathcal{N}_l(\mu,\zeta)=\left\{\alpha\in\mathbb{F}_{q^{2t}}:\forall u\in\mathbb{F}_{q^{2t}},\ \exists v\in\mathbb{F}_{q^{2t}}\text{ such that }(\mu(v),\zeta(v))=(\alpha\mu(u),\alpha\zeta(u))\right\}. N l ( μ , ζ ) = { α ∈ F q 2 t : ∀ u ∈ F q 2 t , ∃ v ∈ F q 2 t such that ( μ ( v ) , ζ ( v )) = ( α μ ( u ) , α ζ ( u )) } . Moreover,
N m ( S G , η ) ≅ N m ( s ) ( μ , ζ ) . N_m(S_{G,\eta})\cong\mathcal{N}_m^{(s)}(\mu,\zeta). N m ( S G , η ) ≅ N m ( s ) ( μ , ζ ) .
where
N m ( s ) ( μ , ζ ) = { α ∈ F q 2 t : ∀ u ∈ F q 2 t , ∃ v ∈ F q 2 t such that ( μ ( v ) , ζ ( v ) ) = ( μ ( u ) α , ζ ( u ) σ s ( α ) ) } . \mathcal{N}_{m}^{(s)}(\mu,\zeta)=\left\{\alpha\in\mathbb{F}_{q^{2t}}:\ \forall u\in\mathbb{F}_{q^{2t}},\ \exists v\in\mathbb{F}_{q^{2t}}\text{ such that }(\mu(v),\zeta(v))=(\mu(u)\alpha,\zeta(u)\sigma^{s}(\alpha))\right\}. N m ( s ) ( μ , ζ ) = { α ∈ F q 2 t : ∀ u ∈ F q 2 t , ∃ v ∈ F q 2 t such that ( μ ( v ) , ζ ( v )) = ( μ ( u ) α , ζ ( u ) σ s ( α )) } .
Proof. We use Theorem 3.5 over the prime field. Since 1 ∈ C G , η 1\in C_{G,\eta} 1 ∈ C G , η , every element of the left idealiser
{ h ∈ R G : h C G , η ⊆ C G , η } \left\{h\in R_G:hC_{G,\eta}\subseteq C_{G,\eta}\right\} { h ∈ R G : h C G , η ⊆ C G , η }
belongs to C G , η C_{G,\eta} C G , η . Hence such an element has a representative of the form
h ‾ = h 0 + h 1 X + ⋯ + h s X s + R G ( X 2 ) . \overline{h}=h_0+h_1X+\cdots+h_sX^s+RG(X^2). h = h 0 + h 1 X + ⋯ + h s X s + R G ( X 2 ) .
We first prove that h ‾ \overline{h} h must be a scalar. Since s ≥ 2 s\ge2 s ≥ 2 , we have
X + R G ( X 2 ) ∈ C G , η . X+RG(X^2)\in C_{G,\eta}. X + R G ( X 2 ) ∈ C G , η .
Thus h ‾ X ∈ C G , η \overline{h}X\in C_{G,\eta} h X ∈ C G , η . Then
h ‾ X = h 0 X + h 1 X 2 + ⋯ + h s X s + 1 ∈ C G , η . \overline{h}X=h_0X+h_1X^2+\cdots+h_sX^{s+1}\in C_{G,\eta}. h X = h 0 X + h 1 X 2 + ⋯ + h s X s + 1 ∈ C G , η .
Since s + 1 < 2 s s+1<2s s + 1 < 2 s , the coefficient of X s + 1 X^{s+1} X s + 1 is not affected by reduction modulo G ( X 2 ) G(X^2) G ( X 2 ) . But every element of C G , η C_{G,\eta} C G , η has degree at most s s s . Therefore h s = 0 h_s=0 h s = 0 . Hence
deg ( h ‾ ) ≤ s − 1. \deg(\overline{h})\le s-1. deg ( h ) ≤ s − 1.
Moreover, X s − 1 + R G ( X 2 ) ∈ C G , η X^{s-1}+RG(X^2)\in C_{G,\eta} X s − 1 + R G ( X 2 ) ∈ C G , η . Therefore
h ‾ X s − 1 ∈ C G , η . \overline{h}X^{s-1}\in C_{G,\eta}. h X s − 1 ∈ C G , η .
The terms
h i X i + s − 1 , 2 ≤ i ≤ s − 1 , h_iX^{i+s-1},\qquad2\le i\le s-1, h i X i + s − 1 , 2 ≤ i ≤ s − 1 ,
have degrees strictly larger than s s s and strictly smaller than 2 s 2s 2 s . Again, they cannot occur in an element of C G , η C_{G,\eta} C G , η . Thus h i = 0 h_i=0 h i = 0 , for every 2 ≤ i ≤ s − 1 2\le i\le s-1 2 ≤ i ≤ s − 1 . Hence
h ‾ = h 0 + h 1 X + R G ( X 2 ) . \overline{h}=h_0+h_1X+RG(X^2). h = h 0 + h 1 X + R G ( X 2 ) .
In the next step we show that h 1 = 0 h_1=0 h 1 = 0 . Since the initial semifield is two-dimensional over its right nucleus, it is not a field. Hence ζ \zeta ζ is not identically zero. Choose u ∈ F q 2 t u\in\mathbb{F}_{q^{2t}} u ∈ F q 2 t such that
ζ ( u ) ≠ 0. \zeta(u)\ne0. ζ ( u ) = 0.
Then
a ‾ = μ ( u ) + ζ ( u ) η X s + R G ( X 2 ) \overline{a}=\mu(u)+\frac{\zeta(u)}{\eta}X^s+RG(X^2) a = μ ( u ) + η ζ ( u ) X s + R G ( X 2 )
belongs to C G , η C_{G,\eta} C G , η . Since h C G , η ⊆ C G , η hC_{G,\eta}\subseteq C_{G,\eta} h C G , η ⊆ C G , η , we have h ‾ a ‾ ∈ C G , η \overline{h}\overline{a}\in C_{G,\eta} h a ∈ C G , η . The coefficient of X s + 1 X^{s+1} X s + 1 in
( h 0 + h 1 X ) ( μ ( u ) + ζ ( u ) η X s ) (h_0+h_1X)\left(\mu(u)+\frac{\zeta(u)}{\eta}X^s\right) ( h 0 + h 1 X ) ( μ ( u ) + η ζ ( u ) X s )
is
h 1 σ ( ζ ( u ) η ) . h_1\sigma\left(\frac{\zeta(u)}{\eta}\right). h 1 σ ( η ζ ( u ) ) .
Since ζ ( u ) ≠ 0 \zeta(u)\ne0 ζ ( u ) = 0 , this coefficient vanishes if and only if
Therefore
h ˉ = h 0 + R G ( X 2 ) , \bar{h}=h_0+RG(X^2), h ˉ = h 0 + R G ( X 2 ) ,
with h 0 ∈ F q 2 t h_0\in\mathbb{F}_{q^{2t}} h 0 ∈ F q 2 t .
Thus the left nucleus is represented by those scalars α ∈ F q 2 t \alpha\in\mathbb{F}_{q^{2t}} α ∈ F q 2 t such that
α a ˉ ∈ C G , η \alpha\bar{a}\in C_{G,\eta} α a ˉ ∈ C G , η
for every a ˉ ∈ C G , η \bar{a}\in C_{G,\eta} a ˉ ∈ C G , η . Let
a ˉ = μ ( u ) + ∑ i = 1 s − 1 u i X i + ζ ( u ) η X s + R G ( X 2 ) ∈ C G , η . \bar{a}=\mu(u)+\sum_{i=1}^{s-1}u_iX^i+\frac{\zeta(u)}{\eta}X^s+RG(X^2)\in C_{G,\eta}. a ˉ = μ ( u ) + i = 1 ∑ s − 1 u i X i + η ζ ( u ) X s + R G ( X 2 ) ∈ C G , η .
Then
α a ˉ = α μ ( u ) + ∑ i = 1 s − 1 α u i X i + α ζ ( u ) η X s + R G ( X 2 ) . \alpha\bar{a}=\alpha\mu(u)+\sum_{i=1}^{s-1}\alpha u_iX^i+\frac{\alpha\zeta(u)}{\eta}X^s+RG(X^2). α a ˉ = α μ ( u ) + i = 1 ∑ s − 1 α u i X i + η α ζ ( u ) X s + R G ( X 2 ) .
Since the coefficients of X , … , X s − 1 X,\ldots,X^{s-1} X , … , X s − 1 are free in the definition of C G , η C_{G,\eta} C G , η , we have α a ˉ ∈ C G , η \alpha\bar{a}\in C_{G,\eta} α a ˉ ∈ C G , η for every a ˉ ∈ C G , η \bar{a}\in C_{G,\eta} a ˉ ∈ C G , η if and only if, for every u ∈ F q 2 t u\in\mathbb{F}_{q^{2t}} u ∈ F q 2 t , there exists v ∈ F q 2 t v\in\mathbb{F}_{q^{2t}} v ∈ F q 2 t such that
μ ( v ) = α μ ( u ) \mu(v)=\alpha\mu(u) μ ( v ) = α μ ( u )
and
ζ ( v ) = α ζ ( u ) . \zeta(v)=\alpha\zeta(u). ζ ( v ) = α ζ ( u ) .
This gives the asserted description of the left nucleus. The proof for the middle nucleus is similar. □
The right nucleus
We now determine the right nucleus of the semifield associated with the lifted spread set C G , η C_{G,\eta} C G , η . Throughout this subsection, we assume that s ≥ 2 s\ge2 s ≥ 2 .
Recall that
E G : = Z ( R G ) ≅ K [ Y ] K [ Y ] G ( Y ) ≅ F q t s . E_G:=Z(R_G)\cong\frac{\mathbb{K}[Y]}{\mathbb{K}[Y]G(Y)}\cong\mathbb{F}_{q^{ts}}. E G := Z ( R G ) ≅ K [ Y ] G ( Y ) K [ Y ] ≅ F q t s .
Put V = E G 2 V=E_G^2 V = E G 2 , and consider an E G E_G E G -algebra isomorphism
φ G : R G ⟶ M 2 ( E G ) ≅ End E G ( V ) . \varphi_G:R_G\longrightarrow M_2(E_G)\cong\operatorname{End}_{E_G}(V). φ G : R G ⟶ M 2 ( E G ) ≅ End E G ( V ) .
In particular, φ G ( λ ) = λ I V \varphi_G(\lambda)=\lambda I_V φ G ( λ ) = λ I V , for every λ ∈ E G \lambda\in E_G λ ∈ E G . By Theorem 3.2 , the set φ G ( C G , η ) \varphi_G(C_{G,\eta}) φ G ( C G , η ) is an additive spread set containing I V I_V I V . As in the previous subsection, let
S G , η = ( V , + , ⋆ G , η ) S_{G,\eta}=(V,+,\star_{G,\eta}) S G , η = ( V , + , ⋆ G , η )
be a semifield with spread set
C ( S G , η ) = φ G ( C G , η ) . \mathcal{C}(S_{G,\eta})=\varphi_G(C_{G,\eta}). C ( S G , η ) = φ G ( C G , η ) .
Denote its multiplicative identity by e e e . Following the notation of Subsection 2.1, write
L x ( y ) = x ⋆ G , η y , R x ( y ) = y ⋆ G , η x ( x , y ∈ V ) . L_x(y)=x\star_{G,\eta}y,\qquad R_x(y)=y\star_{G,\eta}x\qquad(x,y\in V). L x ( y ) = x ⋆ G , η y , R x ( y ) = y ⋆ G , η x ( x , y ∈ V ) .
Theorem 3.7. The right nucleus of S G , η S_{G,\eta} S G , η is
N r ( S G , η ) = { λ e : λ ∈ E G } ≅ E G ≅ F q t s . N_r(S_{G,\eta})=\{\lambda e:\lambda\in E_G\}\cong E_G\cong\mathbb{F}_{q^{ts}}. N r ( S G , η ) = { λ e : λ ∈ E G } ≅ E G ≅ F q t s . Consequently, S G , η S_{G,\eta} S G , η is two-dimensional over its right nucleus.
Proof. Let F p \mathbb{F}_p F p be the prime subfield of K \mathbb{K} K . Consider the centraliser of φ G ( C G , η ) \varphi_G(\mathcal{C}_{G,\eta}) φ G ( C G , η ) in End F p ( V ) \operatorname{End}_{\mathbb{F}_p}(V) End F p ( V ) :
Z : = { D ∈ End F p ( V ) : D ∘ φ G ( a ˉ ) = φ G ( a ˉ ) ∘ D for every a ˉ ∈ C G , η } . \mathcal{Z}:=\{D\in\operatorname{End}_{\mathbb{F}_p}(V):D\circ\varphi_G(\bar{a})=\varphi_G(\bar{a})\circ D\text{ for every }\bar{a}\in\mathcal{C}_{G,\eta}\}. Z := { D ∈ End F p ( V ) : D ∘ φ G ( a ˉ ) = φ G ( a ˉ ) ∘ D for every a ˉ ∈ C G , η } .
The endomorphisms in Z \mathcal{Z} Z are initially assumed to be only F p \mathbb{F}_p F p -linear. By the centraliser description of the right nucleus, applied over F p \mathbb{F}_p F p ,
Z = { R c : c ∈ N r ( S G , η ) } ≅ N r ( S G , η ) ; \mathcal{Z}=\{R_c:c\in\mathrm{N}_r(S_{G,\eta})\}\cong\mathrm{N}_r(S_{G,\eta}); Z = { R c : c ∈ N r ( S G , η )} ≅ N r ( S G , η ) ;
see [34 Proposition 5(iii)] .
We now prove that
Z = { λ I V : λ ∈ E G } . \mathcal{Z}=\{\lambda I_V:\lambda\in E_G\}. Z = { λ I V : λ ∈ E G } .
Since every element of φ G ( C G , η ) \varphi_G(\mathcal{C}_{G,\eta}) φ G ( C G , η ) is E G E_G E G -linear, every scalar endomorphism λ I V \lambda I_V λ I V , with λ ∈ E G \lambda\in E_G λ ∈ E G , belongs to Z \mathcal{Z} Z .
For the reverse inclusion, let D ∈ Z D\in\mathcal{Z} D ∈ Z . For f ∈ R f\in R f ∈ R , write f ˉ : = f + R G ( X 2 ) ∈ R G \bar{f}:=f+RG(X^2)\in R_G f ˉ := f + R G ( X 2 ) ∈ R G . Since s ≥ 2 s\geq2 s ≥ 2 and μ ( 0 ) = ζ ( 0 ) = 0 \mu(0)=\zeta(0)=0 μ ( 0 ) = ζ ( 0 ) = 0 , the definition of C G , η \mathcal{C}_{G,\eta} C G , η gives
a ˉ X ˉ = a X ˉ ∈ C G , η for every a ∈ L . \bar{a}\bar{X}=\bar{aX}\in\mathcal{C}_{G,\eta}\qquad\text{for every }a\in L. a ˉ X ˉ = a X ˉ ∈ C G , η for every a ∈ L .
In particular, X ˉ \bar{X} X ˉ is a nonzero element of C G , η \mathcal{C}_{G,\eta} C G , η , so φ G ( X ˉ ) \varphi_G(\bar{X}) φ G ( X ˉ ) is invertible. As D D D commutes with φ G ( X ˉ ) \varphi_G(\bar{X}) φ G ( X ˉ ) , it also commutes with φ G ( X ˉ ) − 1 \varphi_G(\bar{X})^{-1} φ G ( X ˉ ) − 1 . For every a ∈ L a\in L a ∈ L , we have
φ G ( a ˉ ) = φ G ( a X ‾ ) ∘ φ G ( X ˉ ) − 1 . \varphi_G(\bar{a})=\varphi_G(\overline{aX})\circ\varphi_G(\bar{X})^{-1}. φ G ( a ˉ ) = φ G ( a X ) ∘ φ G ( X ˉ ) − 1 .
Both factors on the right commute with D D D . Consequently,
D ∘ φ G ( a ˉ ) = φ G ( a ˉ ) ∘ D for every a ∈ L . (5) D\circ\varphi_G(\bar{a})=\varphi_G(\bar{a})\circ D\qquad\text{for every }a\in L. \tag*{(5)} D ∘ φ G ( a ˉ ) = φ G ( a ˉ ) ∘ D for every a ∈ L . (5)
Every element f ˉ ∈ R G \bar{f}\in R_G f ˉ ∈ R G has a unique representative of the form
f = ∑ j = 0 2 s − 1 a j X j , a j ∈ L , f=\sum_{j=0}^{2s-1}a_jX^j,\qquad a_j\in L, f = j = 0 ∑ 2 s − 1 a j X j , a j ∈ L ,
and hence
φ G ( f ˉ ) = ∑ j = 0 2 s − 1 φ G ( a j ˉ ) ∘ φ G ( X ˉ ) j . \varphi_G(\bar{f})=\sum_{j=0}^{2s-1}\varphi_G(\bar{a_j})\circ\varphi_G(\bar{X})^j. φ G ( f ˉ ) = j = 0 ∑ 2 s − 1 φ G ( a j ˉ ) ∘ φ G ( X ˉ ) j .
Since D D D is additive and commutes with φ G ( X ˉ ) \varphi_G(\bar{X}) φ G ( X ˉ ) and with every φ G ( a j ˉ ) \varphi_G(\bar{a_j}) φ G ( a j ˉ ) , it follows from (5) that
D ∘ φ G ( f ˉ ) = φ G ( f ˉ ) ∘ D for every f ˉ ∈ R G . D\circ\varphi_G(\bar{f})=\varphi_G(\bar{f})\circ D\qquad\text{for every }\bar{f}\in R_G. D ∘ φ G ( f ˉ ) = φ G ( f ˉ ) ∘ D for every f ˉ ∈ R G .
In particular, D D D commutes with φ G ( λ ) = λ I V \varphi_G(\lambda)=\lambda I_V φ G ( λ ) = λ I V for every λ ∈ E G \lambda\in E_G λ ∈ E G . Therefore
D ( λ v ) = λ D ( v ) ( λ ∈ E G , v ∈ V ) . D(\lambda v)=\lambda D(v)\qquad(\lambda\in E_G,\ v\in V). D ( λ v ) = λ D ( v ) ( λ ∈ E G , v ∈ V ) .
Thus D D D is E G E_G E G -linear.
We may now use the surjectivity of φ G \varphi_G φ G to write D = φ G ( d ˉ ) D=\varphi_G(\bar{d}) D = φ G ( d ˉ ) for a unique d ˉ ∈ R G \bar{d}\in R_G d ˉ ∈ R G . The commutation relations above imply
φ G ( d ˉ f ˉ ) = φ G ( f ˉ d ˉ ) for every f ˉ ∈ R G . \varphi_G(\bar{d}\bar{f})=\varphi_G(\bar{f}\bar{d})\qquad\text{for every }\bar{f}\in R_G. φ G ( d ˉ f ˉ ) = φ G ( f ˉ d ˉ ) for every f ˉ ∈ R G .
By injectivity of φ G \varphi_G φ G ,
d ˉ f ˉ = f ˉ d ˉ for every f ˉ ∈ R G . \bar{d}\bar{f}=\bar{f}\bar{d}\qquad\text{for every }\bar{f}\in R_G. d ˉ f ˉ = f ˉ d ˉ for every f ˉ ∈ R G .
Hence d ˉ ∈ Z ( R G ) = E G \bar{d} \in Z(R_G) = E_G d ˉ ∈ Z ( R G ) = E G , and therefore D = λ I V D = \lambda I_V D = λ I V for some λ ∈ E G \lambda\in E_G λ ∈ E G . This proves
Z = { λ I V : λ ∈ E G } ≅ E G . \mathcal{Z} = \{\lambda I_V : \lambda\in E_G\} \cong E_G. Z = { λ I V : λ ∈ E G } ≅ E G .
Since dim E G ( V ) = 2 \dim_{E_G}(V) = 2 dim E G ( V ) = 2 , the semifield S G , η \mathbb{S}_{G,\eta} S G , η is two-dimensional over its right nucleus.
Iterative constructions
In this section we show that the lifting construction developed above can be iterated. The main point is that, after one application of the construction, the resulting spread set can again be represented in a form suitable for applying the same procedure. To make this explicit, we use a semilinear transformation model for the quotient algebra arising from the skew polynomial construction.
Irreducible semilinear maps
We now describe a semilinear transformations framework corresponding to the skew-polynomial algebra used in Section 3. We keep the notation
K = F q t , L = F q 2 t . \mathbb{K} = \mathbb{F}_{q^t}, \qquad\mathbb{L} = \mathbb{F}_{q^{2t}}. K = F q t , L = F q 2 t .
and we denote by
σ : L ⟶ L , a ⟼ a q t \sigma: \mathbb{L} \longrightarrow\mathbb{L}, \qquad a \longmapsto a^{q^t} σ : L ⟶ L , a ⟼ a q t
the non-trivial element of Gal ( L / K ) \operatorname{Gal}(\mathbb{L}/\mathbb{K}) Gal ( L / K ) . Thus σ \sigma σ has order 2 2 2 and, for the skew polynomial ring R = L [ X ; σ ] R = \mathbb{L}[X;\sigma] R = L [ X ; σ ] , one has Z ( R ) = K [ X 2 ] Z(R) = \mathbb{K}[X^2] Z ( R ) = K [ X 2 ] .
Let V V V be a vector space over L \mathbb{L} L . A map T : V ⟶ V T : V \longrightarrow V T : V ⟶ V is called σ \sigma σ -semilinear if it is additive and satisfies 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 and v ∈ V v \in V v ∈ V . The study of semilinear maps goes back to Jacobson [12 ] . Of particular relevance in the theory of finite semifields are the irreducible semilinear maps; see, for instance, [4 , 16 , 22 ] . A σ \sigma σ -semilinear map T T T is said to be 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 . In [4 ] , Dempwolff gives a characterization of irreducible semilinear transformations and an explicit description of them. We now recall the special case needed in this paper.
Let s ≥ 3 s \ge3 s ≥ 3 be odd, and let
G ( Y ) ∈ K [ Y ] G(Y) \in\mathbb{K}[Y] G ( Y ) ∈ K [ Y ]
be a monic irreducible polynomial of degree s s s , with G ( Y ) ≠ Y G(Y) \ne Y G ( Y ) = Y . Let u u u be a root of G ( Y ) G(Y) G ( Y ) and put
E = K ( u ) ≅ K [ Y ] K [ Y ] G ( Y ) ≅ F q t s . E = \mathbb{K}(u) \cong\frac{\mathbb{K}[Y]}{\mathbb{K}[Y]G(Y)} \cong\mathbb{F}_{q^{ts}}. E = K ( u ) ≅ K [ Y ] G ( Y ) K [ Y ] ≅ F q t s .
Since gcd ( s , 2 ) = 1 \gcd(s,2) = 1 g cd( s , 2 ) = 1 , the fields L \mathbb{L} L and E E E are linearly disjoint over K \mathbb{K} K . Hence their compositum is
V : = L E = F q 2 t s . V := LE = \mathbb{F}_{q^{2ts}}. V := L E = F q 2 t s .
In particular, V V V is an s s s -dimensional vector space over L \mathbb{L} L and a two-dimensional vector space over E E E .
Let
γ : V ⟶ V , x ⟼ x q s t . \gamma: V \longrightarrow V, \qquad x \longmapsto x^{q^{st}}. γ : V ⟶ V , x ⟼ x q s t .
Then γ \gamma γ is the non-trivial element of Gal ( V / E ) \operatorname{Gal}(V/E) Gal ( V / E ) . Moreover, since s s s is odd, the restriction of γ \gamma γ to L = F q 2 t \mathbb{L} = \mathbb{F}_{q^{2t}} L = F q 2 t is precisely σ \sigma σ . Indeed, s t ≡ t ( m o d 2 t ) st \equiv t \pmod{2t} s t ≡ t ( mod 2 t ) . Thus
γ ∣ L = σ . \gamma|_{\mathbb{L}} = \sigma. γ ∣ L = σ .
Since the norm map N V / E : V × ⟶ E × N_{V/E}: V^{\times} \longrightarrow E^{\times} N V / E : V × ⟶ E × is surjective, we may choose w ∈ V × w \in V^{\times} w ∈ V × such that
N V / E ( w ) = w γ ( w ) = u . N_{V/E}(w) = w\gamma(w) = u. N V / E ( w ) = w γ ( w ) = u .
Define
T w : V ⟶ V , x ⟼ w x γ . T_w: V \longrightarrow V,\qquad x \longmapsto wx^\gamma. T w : V ⟶ V , x ⟼ w x γ .
Then T w T_w T w is σ \sigma σ -semilinear with respect to the L \mathbb{L} L -vector space structure of V V V . Indeed, for every α ∈ L \alpha\in\mathbb{L} α ∈ L and x ∈ V x \in V x ∈ V , we have
T w ( α x ) = w γ ( α ) x γ = σ ( α ) T w ( x ) . T_w(\alpha x) = w\gamma(\alpha)x^\gamma= \sigma(\alpha)T_w(x). T w ( α x ) = w γ ( α ) x γ = σ ( α ) T w ( x ) .
Moreover, T w T_w T w is E \mathbb{E} E -linear, because γ \gamma γ fixes E \mathbb{E} E pointwise. Finally,
T w 2 ( x ) = w γ ( w ) x γ 2 = u x T_w^2(x) = w\gamma(w)x^{\gamma^2} = ux T w 2 ( x ) = w γ ( w ) x γ 2 = ux
for every x ∈ V x \in V x ∈ V . Hence
T w 2 = u id V . T_w^2 = u\,\operatorname{id}_V. T w 2 = u id V .
Since G ( u ) = 0 G(u)=0 G ( u ) = 0 , we get
G ( T w 2 ) = 0. G(T_w^2)=0. G ( T w 2 ) = 0.
Proposition 4.1. The map T w T_w T w is an irreducible σ \sigma σ -semilinear map on the L \mathbb{L} L -vector space V V V .
Proof. Let U U U be a non-zero L \mathbb{L} L -subspace of V V V such that
T w ( U ) ⊆ U . T_w(U) \subseteq U. T w ( U ) ⊆ U . Then
T w 2 ( U ) ⊆ U . T_w^2(U) \subseteq U. T w 2 ( U ) ⊆ U . Since T w 2 T_w^2 T w 2 is multiplication by u u u , the subspace U U U is stable under multiplication by u u u . It is also stable under multiplication by every element of L \mathbb{L} L , because U U U is an L \mathbb{L} L -subspace. Therefore U U U is stable under multiplication by the field generated by L \mathbb{L} L and u u u , namely
L ( u ) = L E = V . \mathbb{L}(u)=\mathbb{L}\mathbb{E}=V. L ( u ) = LE = V . Thus U U U is a non-zero V V V -subspace of the one-dimensional V V V -vector space V V V . Hence U = V U=V U = V . This proves that T w T_w T w has no non-trivial invariant L \mathbb{L} L -subspaces, and so T w T_w T w is irreducible. □
We now connect this semilinear model with the skew-polynomial quotient used in the lifting construction. Recall that
R G = R R G ( X 2 ) . R_G=\frac{R}{R G(X^2)}. R G = R G ( X 2 ) R .
We identify the center of R G R_G R G with E \mathbb{E} E through the isomorphism
K [ Y ] K [ Y ] G ( Y ) ⟶ E , Y + K [ Y ] G ( Y ) ⟼ u . \frac{\mathbb{K}[Y]}{\mathbb{K}[Y]G(Y)} \longrightarrow\mathbb{E}, \qquad Y+\mathbb{K}[Y]G(Y) \longmapsto u. K [ Y ] G ( Y ) K [ Y ] ⟶ E , Y + K [ Y ] G ( Y ) ⟼ u .
Since γ \gamma γ has order 2, we have
T w 2 ( x ) = u x , T_w^2(x)=ux, T w 2 ( x ) = ux ,
and hence, for i ≥ 0 i\geq0 i ≥ 0 ,
T w 2 i ( x ) = u i x , T w 2 i + 1 ( x ) = u i w x γ . T_w^{2i}(x)=u^ix,\qquad T_w^{2i+1}(x)=u^iwx^\gamma. T w 2 i ( x ) = u i x , T w 2 i + 1 ( x ) = u i w x γ .
Theorem 4.3 (see [32 ] ). Let G ( Y ) ∈ K [ Y ] G(Y)\in\mathbb{K}[Y] G ( Y ) ∈ K [ Y ] be a monic irreducible polynomial of odd degree s s s , with G ( Y ) ≠ Y G(Y)\ne Y G ( Y ) = Y . Let u ∈ E u\in\mathbb{E} u ∈ E be a root of G ( Y ) G(Y) G ( Y ) , and let w ∈ V × w\in V^\times w ∈ V × be such that N V / E ( w ) = u N_{V/\mathbb{E}}(w)=u N V / E ( w ) = u . Then the map
Ψ w : R G ⟶ End E ( V ) \Psi_w:\quad R_G\longrightarrow\operatorname{End}_{\mathbb{E}}(V) Ψ w : R G ⟶ End E ( V ) ∑ i = 0 2 s − 1 α i X i + R G ( X 2 ) ⟼ ∑ i = 0 2 s − 1 α i T w i (6) \sum_{i=0}^{2s-1}\alpha_iX^i+RG(X^2)\longmapsto\sum_{i=0}^{2s-1}\alpha_iT_w^i \tag*{(6)} i = 0 ∑ 2 s − 1 α i X i + R G ( X 2 ) ⟼ i = 0 ∑ 2 s − 1 α i T w i (6) is an isomorphism of E \mathbb{E} E -algebras.
We note that if
a ˉ = ∑ i = 0 s − 1 α i X 2 i + ∑ i = 0 s − 1 β i X 2 i + 1 + R G ( X 2 ) , \bar{a}=\sum_{i=0}^{s-1}\alpha_iX^{2i}+\sum_{i=0}^{s-1}\beta_iX^{2i+1}+RG(X^2), a ˉ = i = 0 ∑ s − 1 α i X 2 i + i = 0 ∑ s − 1 β i X 2 i + 1 + R G ( X 2 ) ,
with α i , β i ∈ L \alpha_i,\beta_i\in\mathbb{L} α i , β i ∈ L , then
Ψ w ( a ˉ ) ( x ) = ∑ i = 0 s − 1 α i u i x + ∑ i = 0 s − 1 β i u i w x γ , \Psi_w(\bar{a})(x)=\sum_{i=0}^{s-1}\alpha_i u^ix+\sum_{i=0}^{s-1}\beta_i u^iwx^\gamma, Ψ w ( a ˉ ) ( x ) = i = 0 ∑ s − 1 α i u i x + i = 0 ∑ s − 1 β i u i w x γ ,
for every x ∈ V x\in V x ∈ V . Under the isomorphism Ψ w \Psi_w Ψ w , the lifted spread set Φ G , η ( S μ , ζ ) ⊆ R G \Phi_{G,\eta}(\mathcal{S}_{\mu,\zeta})\subseteq R_G Φ G , η ( S μ , ζ ) ⊆ R G given in (4) can be described explicitly as a set of E \mathbb{E} E -linear endomorphisms of V V V .
Let
For
a ˉ = μ ( u 0 ) + ∑ i = 1 s − 1 u i X i + ζ ( u 0 ) η X s + R G ( X 2 ) , \bar{a}=\mu(u_0)+\sum_{i=1}^{s-1}u_iX^i+\frac{\zeta(u_0)}{\eta}X^s+RG(X^2), a ˉ = μ ( u 0 ) + i = 1 ∑ s − 1 u i X i + η ζ ( u 0 ) X s + R G ( X 2 ) ,
with u 0 , … , u s − 1 ∈ L u_0,\ldots,u_{s-1}\in\mathbb{L} u 0 , … , u s − 1 ∈ L , we have
Ψ w ( a ˉ ) ( x ) = μ ( u 0 ) x + ∑ i = 1 s − 1 u i T w i ( x ) + ζ ( u 0 ) η T w s ( x ) = ( μ ( u 0 ) + ∑ j = 1 r u 2 j u j ) x + ( ∑ j = 0 r − 1 u 2 j + 1 u j + ζ ( u 0 ) η u r ) w x γ . \begin{aligned}
\Psi_w(\bar{a})(x)=\mu(u_0)x+\sum_{i=1}^{s-1}u_iT_w^i(x)+\frac{\zeta(u_0)}{\eta}T_w^s(x) \\
&=\left(\mu(u_0)+\sum_{j=1}^{r}u_{2j}u^j\right)x+\left(\sum_{j=0}^{r-1}u_{2j+1}u^j+\frac{\zeta(u_0)}{\eta}u^r\right)wx^\gamma.
\end{aligned} Ψ w ( a ˉ ) ( x ) = μ ( u 0 ) x + i = 1 ∑ s − 1 u i T w i ( x ) + η ζ ( u 0 ) T w s ( x ) = ( μ ( u 0 ) + j = 1 ∑ r u 2 j u j ) x + ( j = 0 ∑ r − 1 u 2 j + 1 u j + η ζ ( u 0 ) u r ) w x γ .
Therefore
Ψ w ( Φ G , η ( S μ , ζ ) ) = { A ( u 0 , … , u s − 1 ) x + B ( u 0 , … , u s − 1 ) w x γ : u 0 , … , u s − 1 ∈ L } , \Psi_w\bigl(\Phi_{G,\eta}(\mathcal{S}_{\mu,\zeta})\bigr)=\{A(u_0,\ldots,u_{s-1})x+B(u_0,\ldots,u_{s-1})wx^\gamma:u_0,\ldots,u_{s-1}\in\mathbb{L}\}, Ψ w ( Φ G , η ( S μ , ζ ) ) = { A ( u 0 , … , u s − 1 ) x + B ( u 0 , … , u s − 1 ) w x γ : u 0 , … , u s − 1 ∈ L } ,
where
A ( u 0 , … , u s − 1 ) = μ ( u 0 ) + ∑ j = 1 r u 2 j u j A(u_0,\ldots,u_{s-1})=\mu(u_0)+\sum_{j=1}^{r}u_{2j}u^j A ( u 0 , … , u s − 1 ) = μ ( u 0 ) + j = 1 ∑ r u 2 j u j
and
B ( u 0 , … , u s − 1 ) = ∑ j = 0 r − 1 u 2 j + 1 u j + ζ ( u 0 ) η u r . B(u_0,\ldots,u_{s-1})=\sum_{j=0}^{r-1}u_{2j+1}u^j+\frac{\zeta(u_0)}{\eta}u^r. B ( u 0 , … , u s − 1 ) = j = 0 ∑ r − 1 u 2 j + 1 u j + η ζ ( u 0 ) u r .
Thus the lifted spread set is represented, in the semilinear model, by the maps
x ⟼ A x + B w x γ . x\longmapsto Ax+Bw x^\gamma. x ⟼ A x + B w x γ .
where A A A and B B B are constrained by the above parametrization. Now, Theorem 3.2 shows that, whenever S μ , ζ S_{\mu,\zeta} S μ , ζ is the spread set of a semifield, the set
Ψ w ( Φ G , η ( S μ , ζ ) ) \Psi_w\left(\Phi_{G,\eta}(S_{\mu,\zeta})\right) Ψ w ( Φ G , η ( S μ , ζ ) )
is again the spread set of a semifield. Furthermore, by Theorem 3.7 , the resulting semifield is two-dimensional over its right nucleus. Hence the lifting construction can be iterated, and this is the key idea developed in the next subsection.
Iterating the lifting construction
We now explain how the previous construction can be iterated. The main point is that, after one lifting step, the resulting semifield is again two-dimensional over its right nucleus. Hence it can be used as input for the same construction.
Let
t 0 = t , K 0 = F q t 0 , L 0 = F q 2 t 0 . t_0=t,\qquad\mathbb{K}_0=\mathbb{F}_{q^{t_0}},\qquad\mathbb{L}_0=\mathbb{F}_{q^{2t_0}}. t 0 = t , K 0 = F q t 0 , L 0 = F q 2 t 0 .
Let
C 0 = S μ 0 , ζ 0 = { μ 0 ( v ) x + ζ 0 ( v ) x q t 0 : v ∈ L 0 } ⊆ End K 0 ( L 0 ) \mathcal{C}_0=S_{\mu_0,\zeta_0}=\left\{\mu_0(v)x+\zeta_0(v)x^{q^{t_0}}:v\in\mathbb{L}_0\right\}\subseteq\operatorname{End}_{\mathbb{K}_0}(\mathbb{L}_0) C 0 = S μ 0 , ζ 0 = { μ 0 ( v ) x + ζ 0 ( v ) x q t 0 : v ∈ L 0 } ⊆ End K 0 ( L 0 )
be the spread set of a semifield which is two-dimensional over its right nucleus K 0 \mathbb{K}_0 K 0 .
Let h ≥ 1 h\ge1 h ≥ 1 . For each i = 1 , … , h i=1,\ldots,h i = 1 , … , h , choose an odd integer s i ≥ 3 s_i\ge3 s i ≥ 3 and define recursively
t i = t i − 1 s i . t_i=t_{i-1}s_i. t i = t i − 1 s i .
Put
K i = F q t i , L i = F q 2 t i . \mathbb{K}_i=\mathbb{F}_{q^{t_i}},\qquad\mathbb{L}_i=\mathbb{F}_{q^{2t_i}}. K i = F q t i , L i = F q 2 t i .
Thus
K 0 ⊆ K 1 ⊆ ⋯ ⊆ K h , L 0 ⊆ L 1 ⊆ ⋯ ⊆ L h , \mathbb{K}_0\subseteq\mathbb{K}_1\subseteq\cdots\subseteq\mathbb{K}_h,\qquad
\mathbb{L}_0\subseteq\mathbb{L}_1\subseteq\cdots\subseteq\mathbb{L}_h, K 0 ⊆ K 1 ⊆ ⋯ ⊆ K h , L 0 ⊆ L 1 ⊆ ⋯ ⊆ L h ,
and, for every i i i , the extension L i / K i \mathbb{L}_i/\mathbb{K}_i L i / K i is quadratic.
For each i = 1 , … , h i=1,\ldots,h i = 1 , … , h , let
G i ( Y ) ∈ K i − 1 [ Y ] G_i(Y)\in\mathbb{K}_{i-1}[Y] G i ( Y ) ∈ K i − 1 [ Y ]
be a monic irreducible polynomial of degree s i s_i s i , with G i ( Y ) ≠ Y G_i(Y)\ne Y G i ( Y ) = Y . Let u i ∈ K i u_i\in\mathbb{K}_i u i ∈ K i be a root of G i ( Y ) G_i(Y) G i ( Y ) , so that
K i = K i − 1 ( u i ) . \mathbb{K}_i=\mathbb{K}_{i-1}(u_i). K i = K i − 1 ( u i ) .
Since s i s_i s i is odd, the fields L i − 1 \mathbb{L}_{i-1} L i − 1 and K i \mathbb{K}_i K i are linearly disjoint over K i − 1 \mathbb{K}_{i-1} K i − 1 . Hence
L i = L i − 1 K i = L i − 1 ( u i ) . \mathbb{L}_i=\mathbb{L}_{i-1}\mathbb{K}_i=\mathbb{L}_{i-1}(u_i). L i = L i − 1 K i = L i − 1 ( u i ) .
In particular,
{ 1 , u i , … , u i s i − 1 } \{1,u_i,\ldots,u_i^{s_i-1}\} { 1 , u i , … , u i s i − 1 }
is an L i − 1 \mathbb{L}_{i-1} L i − 1 -basis of L i \mathbb{L}_i L i .
Let
γ i : L i ⟶ L i , x ⟼ x q t i , \gamma_i:\mathbb{L}_i\longrightarrow\mathbb{L}_i,\qquad x\longmapsto x^{q^{t_i}}, γ i : L i ⟶ L i , x ⟼ x q t i ,
be the non-trivial element of Gal ( L i / K i ) \operatorname{Gal}(\mathbb{L}_i/\mathbb{K}_i) Gal ( L i / K i ) . Since s i s_i s i is odd, we have
t i = t i − 1 s i ≡ t i − 1 ( m o d 2 t i − 1 ) , t_i=t_{i-1}s_i\equiv t_{i-1}\pmod{2t_{i-1}}, t i = t i − 1 s i ≡ t i − 1 ( mod 2 t i − 1 ) ,
and therefore
γ i ∣ L i − 1 = γ i − 1 , \gamma_i|_{\mathbb{L}_{i-1}}=\gamma_{i-1}, γ i ∣ L i − 1 = γ i − 1 ,
where γ i − 1 \gamma_{i-1} γ i − 1 is the non-trivial element of Gal ( L i − 1 / K i − 1 ) \operatorname{Gal}(\mathbb{L}_{i-1}/\mathbb{K}_{i-1}) Gal ( L i − 1 / K i − 1 ) .
Choose
w i ∈ L i × w_i\in\mathbb{L}_i^{\times} w i ∈ L i ×
such that
N L i / K i ( w i ) = w i γ i ( w i ) = u i . \operatorname{N}_{\mathbb{L}_i/\mathbb{K}_i}(w_i)=w_i\gamma_i(w_i)=u_i. N L i / K i ( w i ) = w i γ i ( w i ) = u i .
This is possible because the norm map N L i / K i : L i × → K i × \operatorname{N}_{\mathbb{L}_i/\mathbb{K}_i}:\mathbb{L}_i^{\times}\to\mathbb{K}_i^{\times} N L i / K i : L i × → K i × is surjective. Also choose
η i ∈ L i − 1 × \eta_i\in\mathbb{L}_{i-1}^{\times} η i ∈ L i − 1 ×
such that
N L i − 1 / K i − 1 ( η i ) = ( − 1 ) s i G i ( 0 ) . \operatorname{N}_{\mathbb{L}_{i-1}/\mathbb{K}_{i-1}}(\eta_i)=(-1)^{s_i}G_i(0). N L i − 1 / K i − 1 ( η i ) = ( − 1 ) s i G i ( 0 ) .
Again, such an element exists by the surjectivity of the norm map.
We now define recursively additive maps
μ i , ζ i : L i ⟶ L i . \mu_i,\zeta_i:\mathbb{L}_i\longrightarrow\mathbb{L}_i. μ i , ζ i : L i ⟶ L i .
Assume that μ i − 1 , ζ i − 1 : L i − 1 → L i − 1 \mu_{i-1},\zeta_{i-1}:\mathbb{L}_{i-1}\to\mathbb{L}_{i-1} μ i − 1 , ζ i − 1 : L i − 1 → L i − 1 have already been defined. Let
r i = s i − 1 2 . r_i=\frac{s_i-1}{2}. r i = 2 s i − 1 .
Every element y ∈ L i y\in\mathbb{L}_i y ∈ L i has a unique expression
y = v 0 + v 1 u i + ⋯ + v s i − 1 u i s i − 1 , v 0 , … , v s i − 1 ∈ L i − 1 . y=v_0+v_1u_i+\cdots+v_{s_i-1}u_i^{s_i-1},\qquad v_0,\ldots,v_{s_i-1}\in\mathbb{L}_{i-1}. y = v 0 + v 1 u i + ⋯ + v s i − 1 u i s i − 1 , v 0 , … , v s i − 1 ∈ L i − 1 .
We define
μ i ( y ) = μ i − 1 ( v 0 ) + ∑ j = 1 r i v 2 j u i j \mu_i(y)=\mu_{i-1}(v_0)+\sum_{j=1}^{r_i}v_{2j}u_i^j μ i ( y ) = μ i − 1 ( v 0 ) + j = 1 ∑ r i v 2 j u i j
and
ζ i ( y ) = ( ∑ j = 0 r i − 1 v 2 j + 1 u i j + ζ i − 1 ( v 0 ) η i u i r i ) w i . \zeta_i(y)=\left(\sum_{j=0}^{r_i-1}v_{2j+1}u_i^j+\frac{\zeta_{i-1}(v_0)}{\eta_i}u_i^{r_i}\right)w_i. ζ i ( y ) = ( j = 0 ∑ r i − 1 v 2 j + 1 u i j + η i ζ i − 1 ( v 0 ) u i r i ) w i .
The maps μ i \mu_i μ i and ζ i \zeta_i ζ i are additive, since the expansion of y y y in the L i − 1 \mathbb{L}_{i-1} L i − 1 -basis
{ 1 , u i , … , u i s i − 1 } \{1,u_i,\ldots,u_i^{s_i-1}\} { 1 , u i , … , u i s i − 1 }
is additive and μ i − 1 , ζ i − 1 \mu_{i-1},\zeta_{i-1} μ i − 1 , ζ i − 1 are additive.
Define
C i = S μ i , ζ i = { μ i ( y ) x + ζ i ( y ) x γ i : y ∈ L i } ⊆ End K i ( L i ) . C_i=S_{\mu_i,\zeta_i}=\{\mu_i(y)x+\zeta_i(y)x^{\gamma_i}:y\in\mathbb{L}_i\}\subseteq\operatorname{End}_{\mathbb{K}_i}(\mathbb{L}_i). C i = S μ i , ζ i = { μ i ( y ) x + ζ i ( y ) x γ i : y ∈ L i } ⊆ End K i ( L i ) .
Theorem 4.4. For every i = 0 , … , h i=0,\ldots,h i = 0 , … , h , the set C i C_i C i is the spread set of a semifield S i S_i S i of order
q 2 t i = q 2 t s 1 ⋯ s i . q^{2t_i}=q^{2t s_1\cdots s_i}. q 2 t i = q 2 t s 1 ⋯ s i . Moreover,
N r ( S i ) ≅ K i = F q t i . \operatorname{N}_r(S_i)\cong K_i=\mathbb{F}_{q^{t_i}}. N r ( S i ) ≅ K i = F q t i . In particular, S i S_i S i is two-dimensional over its right nucleus.
Proof. We argue by induction on i i i . For i = 0 i=0 i = 0 , this is true by assumption.
Assume that C i − 1 = S μ i − 1 , ζ i − 1 C_{i-1}=S_{\mu_{i-1},\zeta_{i-1}} C i − 1 = S μ i − 1 , ζ i − 1 is the spread set of a semifield S i − 1 S_{i-1} S i − 1 of order q 2 t i − 1 q^{2t_{i-1}} q 2 t i − 1 which is two-dimensional over its right nucleus K i − 1 \mathbb{K}_{i-1} K i − 1 .
Let
R i = L i − 1 [ X i ; γ i − 1 ] , R_i=\mathbb{L}_{i-1}[X_i;\gamma_{i-1}], R i = L i − 1 [ X i ; γ i − 1 ] ,
where γ i − 1 \gamma_{i-1} γ i − 1 is the non-trivial element of Gal ( L i − 1 / K i − 1 ) \operatorname{Gal}(\mathbb{L}_{i-1}/\mathbb{K}_{i-1}) Gal ( L i − 1 / K i − 1 ) . Since γ i − 1 \gamma_{i-1} γ i − 1 has order 2, the center of R i R_i R i is
Z ( R i ) = K i − 1 [ X i 2 ] . Z(R_i)=\mathbb{K}_{i-1}[X_i^2]. Z ( R i ) = K i − 1 [ X i 2 ] .
Consider the quotient
( R i ) G i = R i R i G i ( X i 2 ) . (R_i)_{G_i}=\frac{R_i}{R_iG_i(X_i^2)}. ( R i ) G i = R i G i ( X i 2 ) R i .
The lifting construction of Theorem 3.2 gives the spread set
Φ G i , η i ( C i − 1 ) = { μ i − 1 ( v 0 ) + ∑ ℓ = 1 s i − 1 v ℓ X i ℓ + ζ i − 1 ( v 0 ) η i X i s i + R i G i ( X i 2 ) : v 0 , … , v s i − 1 ∈ L i − 1 } ⊆ ( R i ) G i . \Phi_{G_i,\eta_i}(C_{i-1})=
\left\{
\mu_{i-1}(v_0)+\sum_{\ell=1}^{s_i-1}v_\ell X_i^\ell+
\frac{\zeta_{i-1}(v_0)}{\eta_i}X_i^{s_i}+R_iG_i(X_i^2):
v_0,\ldots,v_{s_i-1}\in\mathbb{L}_{i-1}
\right\}\subseteq(R_i)_{G_i}. Φ G i , η i ( C i − 1 ) = { μ i − 1 ( v 0 ) + ℓ = 1 ∑ s i − 1 v ℓ X i ℓ + η i ζ i − 1 ( v 0 ) X i s i + R i G i ( X i 2 ) : v 0 , … , v s i − 1 ∈ L i − 1 } ⊆ ( R i ) G i .
Hence Φ G i , η i ( C i − 1 ) \Phi_{G_i,\eta_i}(C_{i-1}) Φ G i , η i ( C i − 1 ) is the spread set of a semifield of order
q 2 t i − 1 s i = q 2 t i . q^{2t_{i-1}s_i}=q^{2t_i}. q 2 t i − 1 s i = q 2 t i .
By Theorem 3.7, this semifield has right nucleus isomorphic to
K i − 1 ( u i ) = K i . \mathbb{K}_{i-1}(u_i)=\mathbb{K}_i. K i − 1 ( u i ) = K i .
It remains only to identify this spread set with the explicit semilinear model C i C_i C i . We now use the argument of Section 4.1. The map
Ψ w i : ( R i ) G i ⟶ End K i ( L i ) \Psi_{w_i}:(R_i)_{G_i}\longrightarrow\operatorname{End}_{\mathbb{K}_i}(\mathbb{L}_i) Ψ w i : ( R i ) G i ⟶ End K i ( L i )
is an isomorphism of K i \mathbb{K}_i K i -algebras. Moreover, if T w i ( x ) = w i x γ i T_{w_i}(x)=w_ix^{\gamma_i} T w i ( x ) = w i x γ i , then
T w i 2 j ( x ) = u i j x , T w i 2 j + 1 ( x ) = u i j w i x γ i . T_{w_i}^{2j}(x)=u_i^jx,\qquad T_{w_i}^{2j+1}(x)=u_i^jw_ix^{\gamma_i}. T w i 2 j ( x ) = u i j x , T w i 2 j + 1 ( x ) = u i j w i x γ i .
Therefore, for
a ˉ = μ i − 1 ( v 0 ) + ∑ ℓ = 1 s i − 1 v ℓ X i ℓ + ζ i − 1 ( v 0 ) η i X i s i + R i G i ( X i 2 ) , \bar{a}=\mu_{i-1}(v_0)+\sum_{\ell=1}^{s_i-1}v_\ell X_i^\ell+
\frac{\zeta_{i-1}(v_0)}{\eta_i}X_i^{s_i}+R_iG_i(X_i^2), a ˉ = μ i − 1 ( v 0 ) + ℓ = 1 ∑ s i − 1 v ℓ X i ℓ + η i ζ i − 1 ( v 0 ) X i s i + R i G i ( X i 2 ) ,
we get
Ψ w i ( a ˉ ) ( x ) = μ i − 1 ( v 0 ) x + ∑ ℓ = 1 s i − 1 v ℓ T w i ℓ ( x ) + ζ i − 1 ( v 0 ) η i T w i s i ( x ) = ( μ i − 1 ( v 0 ) + ∑ j = 1 r i v 2 j u i j ) x + ( ∑ j = 0 r i − 1 v 2 j + 1 u i j + ζ i − 1 ( v 0 ) η i u i r i ) w i x γ i . \begin{aligned}
\Psi_{w_i}(\bar{a})(x)=\mu_{i-1}(v_0)x+\sum_{\ell=1}^{s_i-1}v_\ell T_{w_i}^{\ell}(x)+\frac{\zeta_{i-1}(v_0)}{\eta_i}T_{w_i}^{s_i}(x) \\
&=\left(\mu_{i-1}(v_0)+\sum_{j=1}^{r_i}v_{2j}u_i^j\right)x+
\left(\sum_{j=0}^{r_i-1}v_{2j+1}u_i^j+
\frac{\zeta_{i-1}(v_0)}{\eta_i}u_i^{r_i}\right)w_ix^{\gamma_i}.
\end{aligned} Ψ w i ( a ˉ ) ( x ) = μ i − 1 ( v 0 ) x + ℓ = 1 ∑ s i − 1 v ℓ T w i ℓ ( x ) + η i ζ i − 1 ( v 0 ) T w i s i ( x ) = ( μ i − 1 ( v 0 ) + j = 1 ∑ r i v 2 j u i j ) x + ( j = 0 ∑ r i − 1 v 2 j + 1 u i j + η i ζ i − 1 ( v 0 ) u i r i ) w i x γ i .
Now let
y = v 0 + v 1 u i + ⋯ + v s i − 1 u i s i − 1 ∈ L i . y=v_0+v_1u_i+\cdots+v_{s_i-1}u_i^{s_i-1}\in\mathbb{L}_i. y = v 0 + v 1 u i + ⋯ + v s i − 1 u i s i − 1 ∈ L i .
By the definitions of μ i \mu_i μ i and ζ i \zeta_i ζ i , the last expression is precisely
μ i ( y ) x + ζ i ( y ) x γ i . \mu_i(y)x+\zeta_i(y)x^{\gamma_i}. μ i ( y ) x + ζ i ( y ) x γ i .
Hence
Ψ w i ( Φ G i , η i ( C i − 1 ) ) = C i . \Psi_{w_i}\left(\Phi_{G_i,\eta_i}(C_{i-1})\right)=C_i. Ψ w i ( Φ G i , η i ( C i − 1 ) ) = C i .
Since Ψ w i \Psi_{w_i} Ψ w i is an algebra isomorphism, it preserves nonsingularity and the spread-set property. Thus C i C_i C i is the spread set of a semifield S i S_i S i of order q 2 t i q^{2t_i} q 2 t i , and its right nucleus is isomorphic to K i \mathbb{K}_i K i . This completes the induction. □
The theorem gives an explicit iterative construction. Starting from a semifield S 0 S_0 S 0 of order q 2 t q^{2t} q 2 t which is two-dimensional over its right nucleus, and choosing odd integers s 1 , … , s h s_1,\ldots,s_h s 1 , … , s h together with irreducible polynomials
G i ( Y ) ∈ F q t i − 1 [ Y ] , deg ( G i ) = s i , G_i(Y)\in\mathbb{F}_{q^{t_{i-1}}}[Y],\qquad\deg(G_i)=s_i, G i ( Y ) ∈ F q t i − 1 [ Y ] , deg ( G i ) = s i ,
one obtains a semifield S h S_h S h of order
q 2 t s 1 ⋯ s h q^{2ts_1\cdots s_h} q 2 t s 1 ⋯ s h
whose right nucleus is
F q t s 1 ⋯ s h . \mathbb{F}_{q^{ts_1\cdots s_h}}. F q t s 1 ⋯ s h .
In particular, every step preserves the property of being two-dimensional over the right nucleus.
Explicit constructions
In this section we apply the iterative lifting construction developed above to the semifields constructed by Ebert, Marino, Polverino and Trombetti [7 ] . We use their spread sets as initial spread sets, but we adopt the convention of the present paper: spread sets are represented over the right nucleus. Equivalently, we work with the opposite semifields of the corresponding EMPT semifields. More precisely, let S = ( S , + , ⋆ ) \mathbb{S}=(\mathbb{S},+,\star) S = ( S , + , ⋆ ) be a semifield. The opposite semifield of S \mathbb{S} S is S o p = ( S , + , ⋆ o p ) \mathbb{S}^{\mathrm{op}}=(\mathbb{S},+,\star^{\mathrm{op}}) S op = ( S , + , ⋆ op ) , where
a ⋆ o p b = b ⋆ a ( a , b ∈ S ) . a\star^{\mathrm{op}} b=b\star a\qquad(a,b\in\mathbb{S}). a ⋆ op b = b ⋆ a ( a , b ∈ S ) .
It has the same multiplicative identity as S \mathbb{S} S , and its left and right multiplication maps satisfy
L a o p = R a , R a o p = L a ( a ∈ S ) . L_a^{\mathrm{op}}=R_a,\qquad R_a^{\mathrm{op}}=L_a\qquad(a\in\mathbb{S}). L a op = R a , R a op = L a ( a ∈ S ) .
It follows directly from the definitions that
N l ( S o p ) = N r ( S ) , N m ( S o p ) = N m ( S ) , N r ( S o p ) = N l ( S ) , N_l(\mathbb{S}^{\mathrm{op}})=N_r(\mathbb{S}),\qquad N_m(\mathbb{S}^{\mathrm{op}})=N_m(\mathbb{S}),\qquad N_r(\mathbb{S}^{\mathrm{op}})=N_l(\mathbb{S}), N l ( S op ) = N r ( S ) , N m ( S op ) = N m ( S ) , N r ( S op ) = N l ( S ) ,
and
Z ( S o p ) = Z ( S ) . Z(\mathbb{S}^{\mathrm{op}})=Z(\mathbb{S}). Z ( S op ) = Z ( S ) .
Hence the distinguished nucleus in what follows is the right nucleus. Throughout this section, let t > 1 t>1 t > 1 be odd and put
K 0 = F q t , L 0 = F q 2 t , γ 0 : x ⟼ x q t . \mathbb{K}_0=\mathbb{F}_{q^t},\qquad\mathbb{L}_0=\mathbb{F}_{q^{2t}},\qquad\gamma_0:x\longmapsto x^{q^t}. K 0 = F q t , L 0 = F q 2 t , γ 0 : x ⟼ x q t .
Let u ∈ K 0 u\in\mathbb{K}_0 u ∈ K 0 be such that
{ 1 , u , u 2 , … , u t − 1 } \{1,u,u^2,\ldots,u^{t-1}\} { 1 , u , u 2 , … , u t − 1 }
is an F q \mathbb{F}_q F q -basis of K 0 \mathbb{K}_0 K 0 . Define
A u = ⟨ 1 , u , … , u t − 1 2 ⟩ F q , B u = ⟨ 1 , u , … , u t − 3 2 ⟩ F q . \mathcal{A}_u=\left\langle1,u,\ldots,u^{\frac{t-1}{2}}\right\rangle_{\mathbb{F}_q},\qquad
\mathcal{B}_u=\left\langle1,u,\ldots,u^{\frac{t-3}{2}}\right\rangle_{\mathbb{F}_q}. A u = ⟨ 1 , u , … , u 2 t − 1 ⟩ F q , B u = ⟨ 1 , u , … , u 2 t − 3 ⟩ F q .
Let a , b , c ∈ L 0 × a,b,c \in\mathbb{L}_0^{\times} a , b , c ∈ L 0 × , with a , c ∉ K 0 a,c \notin\mathbb{K}_0 a , c ∈ / K 0 , and consider the set
S ( u , a , b , c ) = { x ⟼ ( α + a β ) x + b ( γ + c δ ) x q t : α , β ∈ A u , γ , δ ∈ B u } . S(u,a,b,c)=\left\{x\longmapsto(\alpha+a\beta)x+b(\gamma+c\delta)x^{q^t}:\alpha,\beta\in A_u,\ \gamma,\delta\in B_u\right\}. S ( u , a , b , c ) = { x ⟼ ( α + a β ) x + b ( γ + cδ ) x q t : α , β ∈ A u , γ , δ ∈ B u } .
We assume that the parameters u , a , b , c u,a,b,c u , a , b , c satisfy one of the admissible conditions defining the EMPT families F I , … , F V \mathcal{F}_{\mathrm{I}},\ldots,\mathcal{F}_{\mathrm{V}} F I , … , F V in [7 ] . Under these assumptions, S ( u , a , b , c ) S(u,a,b,c) S ( u , a , b , c ) is an additive spread set of K 0 \mathbb{K}_0 K 0 -linear maps of L 0 \mathbb{L}_0 L 0 .
Let
C 0 : = S ( u , a , b , c ) . \mathcal{C}_0:=S(u,a,b,c). C 0 := S ( u , a , b , c ) .
In the right-nucleus convention used in this paper, C 0 \mathcal{C}_0 C 0 defines a semifield S 0 \mathcal{S}_0 S 0 of order q 2 t q^{2t} q 2 t . Moreover,
N r ( S 0 ) ≅ K 0 = F q t , N l ( S 0 ) ≅ N m ( S 0 ) ≅ F q , Z ( S 0 ) ≅ F q . N_r(\mathcal{S}_0)\cong\mathbb{K}_0=\mathbb{F}_{q^t},\qquad N_l(\mathcal{S}_0)\cong N_m(\mathcal{S}_0)\cong\mathbb{F}_q,\qquad Z(\mathcal{S}_0)\cong\mathbb{F}_q. N r ( S 0 ) ≅ K 0 = F q t , N l ( S 0 ) ≅ N m ( S 0 ) ≅ F q , Z ( S 0 ) ≅ F q .
Indeed, in the original EMPT convention the semifields in the families F I , … , F V \mathcal{F}_{\mathrm{I}},\ldots,\mathcal{F}_{\mathrm{V}} F I , … , F V are two-dimensional over their left nucleus and have middle and right nuclei isomorphic to F q \mathbb{F}_q F q . Passing to the opposite semifield interchanges the left and right nuclei and leaves the middle nucleus unchanged.
One lifting step
Let s ≥ 3 s\geq3 s ≥ 3 be odd, and write
Let
G ( Y ) ∈ K 0 [ Y ] G(Y)\in\mathbb{K}_0[Y] G ( Y ) ∈ K 0 [ Y ]
be a monic irreducible polynomial of degree s s s . Let θ ∈ F q t s \theta\in\mathbb{F}_{q^{ts}} θ ∈ F q t s be a root of G ( Y ) G(Y) G ( Y ) , and put
K 1 = K 0 ( θ ) = F q t s , L 1 = L 0 ( θ ) = F q 2 t s . \mathbb{K}_1=\mathbb{K}_0(\theta)=\mathbb{F}_{q^{ts}},\qquad\mathbb{L}_1=\mathbb{L}_0(\theta)=\mathbb{F}_{q^{2ts}}. K 1 = K 0 ( θ ) = F q t s , L 1 = L 0 ( θ ) = F q 2 t s .
Since s s s is odd, the fields L 0 \mathbb{L}_0 L 0 and K 1 \mathbb{K}_1 K 1 are linearly disjoint over K 0 \mathbb{K}_0 K 0 . Hence
L 1 = L 0 K 1 = L 0 ( θ ) , \mathbb{L}_1=\mathbb{L}_0\mathbb{K}_1=\mathbb{L}_0(\theta), L 1 = L 0 K 1 = L 0 ( θ ) ,
and
{ 1 , θ , … , θ s − 1 } \{1,\theta,\ldots,\theta^{s-1}\} { 1 , θ , … , θ s − 1 }
is an L 0 \mathbb{L}_0 L 0 -basis of L 1 \mathbb{L}_1 L 1 .
Choose
η ∈ L 0 × \eta\in\mathbb{L}_0^{\times} η ∈ L 0 ×
such that
N L 0 / K 0 ( η ) = ( − 1 ) s G ( 0 ) . N_{\mathbb{L}_0/\mathbb{K}_0}(\eta)=(-1)^sG(0). N L 0 / K 0 ( η ) = ( − 1 ) s G ( 0 ) .
Also choose
w ∈ L 1 × w\in\mathbb{L}_1^{\times} w ∈ L 1 ×
such that
N L 1 / K 1 ( w ) = w w q t s = θ . N_{\mathbb{L}_1/\mathbb{K}_1}(w)=ww^{q^{ts}}=\theta. N L 1 / K 1 ( w ) = w w q t s = θ .
Both choices are possible because norm maps of finite fields are surjective. Define
C 1 ⊆ End K 1 ( L 1 ) \mathcal{C}_1\subseteq\operatorname{End}_{\mathbb{K}_1}(\mathbb{L}_1) C 1 ⊆ End K 1 ( L 1 )
as the set of all maps
x ⟼ A x + B w x q t s , x\longmapsto Ax+Bwx^{q^{ts}}, x ⟼ A x + B w x q t s ,
where
A = α + α β + ∑ j = 1 r v 2 j θ j A=\alpha+\alpha\beta+\sum_{j=1}^{r}v_{2j}\theta^{j} A = α + α β + j = 1 ∑ r v 2 j θ j
and
B = ∑ j = 0 r − 1 v 2 j + 1 θ j + b ( γ + c δ ) η θ r , B=\sum_{j=0}^{r-1}v_{2j+1}\theta^{j}+\frac{b(\gamma+c\delta)}{\eta}\theta^{r}, B = j = 0 ∑ r − 1 v 2 j + 1 θ j + η b ( γ + cδ ) θ r ,
with α , β ∈ A u \alpha,\beta\in A_u α , β ∈ A u , γ , δ ∈ B u \gamma,\delta\in B_u γ , δ ∈ B u , v 1 , … , v s − 1 ∈ L 0 v_1,\ldots,v_{s-1}\in L_0 v 1 , … , v s − 1 ∈ L 0 .
Theorem 5.1. The set C 1 C_1 C 1 is the spread set of a semifield S 1 S_1 S 1 of order q 2 t s q^{2ts} q 2 t s . Moreover,
N r ( S 1 ) ≅ K 1 = F q t s , N_r(S_1)\cong K_1=\mathbb{F}_{q^{ts}}, N r ( S 1 ) ≅ K 1 = F q t s , and
N l ( S 1 ) ≅ F q , N m ( S 1 ) ≅ F q . N_l(S_1)\cong\mathbb{F}_q,\qquad N_m(S_1)\cong\mathbb{F}_q. N l ( S 1 ) ≅ F q , N m ( S 1 ) ≅ F q .
Proof. The fact that C 1 C_1 C 1 is a semifield spread set of order q 2 t s q^{2ts} q 2 t s follows directly from Theorem 4.4 . The same result also gives
N r ( S 1 ) ≅ K 1 = F q t s . N_r(S_1)\cong K_1=\mathbb{F}_{q^{ts}}. N r ( S 1 ) ≅ K 1 = F q t s .
It remains to determine the left and middle nuclei. By Theorem 3.6 , the left nucleus of S 1 S_1 S 1 is contained in the left nucleus of the initial semifield S 0 S_0 S 0 . Since S 0 S_0 S 0 is the opposite of an EMPT semifield belonging to one of the families F I , … , F V F_{\mathrm{I}},\ldots,F_{\mathrm{V}} F I , … , F V , its left nucleus is isomorphic to F q \mathbb{F}_q F q ; see [7 Theorem 1.1] . Thus, we obtain
F q ⊆ N l ( S 1 ) ⊆ N l ( S 0 ) ≅ F q . \mathbb{F}_q\subseteq N_l(S_1)\subseteq N_l(S_0)\cong\mathbb{F}_q. F q ⊆ N l ( S 1 ) ⊆ N l ( S 0 ) ≅ F q .
Therefore
N l ( S 1 ) ≅ F q . N_l(S_1)\cong\mathbb{F}_q. N l ( S 1 ) ≅ F q .
The argument for the middle nucleus is analogous. □
The iterated construction
We now iterate the previous construction by using the argument developed in § 4.2 .
Let h ≥ 1 h\geq1 h ≥ 1 . For every i = 1 , … , h i=1,\ldots,h i = 1 , … , h , let s i ≥ 3 s_i\geq3 s i ≥ 3 be odd and put
t i = t s 1 ⋯ s i , t 0 = t . t_i=ts_1\cdots s_i,\qquad t_0=t. t i = t s 1 ⋯ s i , t 0 = t .
Set
K i = F q t i , L i = F q 2 t i . K_i=\mathbb{F}_{q^{t_i}},\qquad L_i=\mathbb{F}_{q^{2t_i}}. K i = F q t i , L i = F q 2 t i .
For each i = 1 , … , h i=1,\ldots,h i = 1 , … , h , let
G i ( Y ) ∈ K i − 1 [ Y ] G_i(Y)\in K_{i-1}[Y] G i ( Y ) ∈ K i − 1 [ Y ]
be a monic irreducible polynomial of degree s i s_i s i , and let θ i ∈ K i \theta_i\in K_i θ i ∈ K i be a root of G i ( Y ) G_i(Y) G i ( Y ) . Hence
K i = K i − 1 ( θ i ) . K_i=K_{i-1}(\theta_i). K i = K i − 1 ( θ i ) .
Since s i s_i s i is odd, the fields L i − 1 L_{i-1} L i − 1 and K i K_i K i are linearly disjoint over K i − 1 K_{i-1} K i − 1 . Therefore
L i = L i − 1 K i = L i − 1 ( θ i ) , L_i=L_{i-1}K_i=L_{i-1}(\theta_i), L i = L i − 1 K i = L i − 1 ( θ i ) ,
and
{ 1 , θ i , … , θ i s i − 1 } \{1,\theta_i,\ldots,\theta_i^{s_i-1}\} { 1 , θ i , … , θ i s i − 1 }
is an L i − 1 \mathbb{L}_{i-1} L i − 1 -basis of L i \mathbb{L}_i L i .
Choose elements
η i ∈ L i − 1 × , w i ∈ L i × , \eta_i \in\mathbb{L}_{i-1}^{\times}, \qquad w_i \in\mathbb{L}_i^{\times}, η i ∈ L i − 1 × , w i ∈ L i × ,
such that
N L i − 1 / K i − 1 ( η i ) = ( − 1 ) s i G i ( 0 ) \operatorname{N}_{\mathbb{L}_{i-1}/K_{i-1}}(\eta_i)=(-1)^{s_i}G_i(0) N L i − 1 / K i − 1 ( η i ) = ( − 1 ) s i G i ( 0 )
and
N L i / K i ( w i ) = w i w i q t i = θ i . \operatorname{N}_{\mathbb{L}_i/K_i}(w_i)=w_iw_i^{q^{t_i}}=\theta_i. N L i / K i ( w i ) = w i w i q t i = θ i .
These choices are possible by the surjectivity of the norm maps of finite fields. We now describe explicitly the spread set obtained after h h h lifting steps. Let
P 0 = A u × A u × B u × B u . \mathcal{P}_0=\mathcal{A}_u\times\mathcal{A}_u\times\mathcal{B}_u\times\mathcal{B}_u. P 0 = A u × A u × B u × B u .
For
p 0 = ( α , β , γ , δ ) ∈ P 0 , p_0=(\alpha,\beta,\gamma,\delta)\in\mathcal{P}_0, p 0 = ( α , β , γ , δ ) ∈ P 0 ,
define
A 0 ( p 0 ) = α + a β , B 0 ( p 0 ) = b ( γ + c δ ) . A_0(p_0)=\alpha+a\beta,\qquad B_0(p_0)=b(\gamma+c\delta). A 0 ( p 0 ) = α + a β , B 0 ( p 0 ) = b ( γ + cδ ) .
Thus
C 0 = { x ⟼ A 0 ( p 0 ) x + B 0 ( p 0 ) x q t : p 0 ∈ P 0 } . \mathcal{C}_0=\left\{x\longmapsto A_0(p_0)x+B_0(p_0)x^{q^t}:p_0\in\mathcal{P}_0\right\}. C 0 = { x ⟼ A 0 ( p 0 ) x + B 0 ( p 0 ) x q t : p 0 ∈ P 0 } .
We now define the coefficients recursively. For i = 1 , … , h i=1,\ldots,h i = 1 , … , h , write
r i = s i − 1 2 r_i=\frac{s_i-1}{2} r i = 2 s i − 1
and put
P i = P i − 1 × L i − 1 s i − 1 . \mathcal{P}_i=\mathcal{P}_{i-1}\times\mathbb{L}_{i-1}^{s_i-1}. P i = P i − 1 × L i − 1 s i − 1 .
An element of P i \mathcal{P}_i P i will be written as
p i = ( p i − 1 ; v i , 1 , … , v i , s i − 1 ) , p_i=(p_{i-1};v_{i,1},\ldots,v_{i,s_i-1}), p i = ( p i − 1 ; v i , 1 , … , v i , s i − 1 ) ,
where
p i − 1 ∈ P i − 1 , v i , 1 , … , v i , s i − 1 ∈ L i − 1 . p_{i-1}\in\mathcal{P}_{i-1},\qquad v_{i,1},\ldots,v_{i,s_i-1}\in\mathbb{L}_{i-1}. p i − 1 ∈ P i − 1 , v i , 1 , … , v i , s i − 1 ∈ L i − 1 .
Define
A i ( p i ) = A i − 1 ( p i − 1 ) + ∑ j = 1 r i v i , 2 j θ i j A_i(p_i)=A_{i-1}(p_{i-1})+\sum_{j=1}^{r_i}v_{i,2j}\theta_i^j A i ( p i ) = A i − 1 ( p i − 1 ) + j = 1 ∑ r i v i , 2 j θ i j
and
B i ( p i ) = ( ∑ j = 0 r i − 1 v i , 2 j + 1 θ i j + B i − 1 ( p i − 1 ) η i θ i r i ) w i . B_i(p_i)=\left(\sum_{j=0}^{r_i-1}v_{i,2j+1}\theta_i^j+\frac{B_{i-1}(p_{i-1})}{\eta_i}\theta_i^{r_i}\right)w_i. B i ( p i ) = ( j = 0 ∑ r i − 1 v i , 2 j + 1 θ i j + η i B i − 1 ( p i − 1 ) θ i r i ) w i .
Then the spread set obtained after i i i steps is
C i = { x ⟼ A i ( p i ) x + B i ( p i ) x q t i : p i ∈ P i } ⊆ End K i ( L i ) . \mathcal{C}_i=\left\{x\longmapsto A_i(p_i)x+B_i(p_i)x^{q^{t_i}}:p_i\in\mathcal{P}_i\right\}\subseteq\operatorname{End}_{K_i}(\mathbb{L}_i). C i = { x ⟼ A i ( p i ) x + B i ( p i ) x q t i : p i ∈ P i } ⊆ End K i ( L i ) .
In particular, the final iterated spread set is
C h = { x ⟼ A h ( p h ) x + B h ( p h ) x q t h : p h ∈ P h } ⊆ End K h ( L h ) . \mathcal{C}_h=\left\{x\longmapsto A_h(p_h)x+B_h(p_h)x^{q^{t_h}}:p_h\in\mathcal{P}_h\right\}\subseteq\operatorname{End}_{K_h}(\mathbb{L}_h). C h = { x ⟼ A h ( p h ) x + B h ( p h ) x q t h : p h ∈ P h } ⊆ End K h ( L h ) .
Theorem 5.2. The set C h C_h C h is the spread set of a semifield S h S_h S h of order
q 2 t h = q 2 t s 1 ⋯ s h . q^{2t_h}=q^{2ts_1\cdots s_h}. q 2 t h = q 2 t s 1 ⋯ s h . Moreover,
N r ( S h ) ≅ K h = F q t s 1 ⋯ s h , N_r(S_h)\cong K_h=\mathbb{F}_{q^{ts_1\cdots s_h}}, N r ( S h ) ≅ K h = F q t s 1 ⋯ s h , and
N l ( S h ) ≅ F q , N m ( S h ) ≅ F q , Z ( S h ) ≅ F q . N_l(S_h)\cong\mathbb{F}_q,\qquad N_m(S_h)\cong\mathbb{F}_q,\qquad Z(S_h)\cong\mathbb{F}_q. N l ( S h ) ≅ F q , N m ( S h ) ≅ F q , Z ( S h ) ≅ F q .
Proof. The recursive description above is exactly the construction of § 4.2 , applied to the initial spread set
C 0 = S ( u , a , b , c ) . C_0=S(u,a,b,c). C 0 = S ( u , a , b , c ) .
Hence, by Theorem 4.4 , C h C_h C h is a semifield spread set of order
q 2 t h = q 2 t s 1 ⋯ s h , q^{2t_h}=q^{2ts_1\cdots s_h}, q 2 t h = q 2 t s 1 ⋯ s h ,
and its right nucleus is
N r ( S h ) ≅ K h = F q t h . N_r(S_h)\cong K_h=\mathbb{F}_{q^{t_h}}. N r ( S h ) ≅ K h = F q t h .
It remains to determine the left and middle nuclei. At each lifting step, Theorem 3.6 shows that the new left and middle nuclei are contained in the corresponding nuclei of the previous semifield. Then arguing as in the proof of Theorem 5.1 , we get the assertion. Finally, the center is contained in the intersection of the three nuclei and contains the central scalar maps induced by F q \mathbb{F}_q F q . Thus Z ( S h ) ≅ F q Z(S_h)\cong\mathbb{F}_q Z ( S h ) ≅ F q .
The isotopy issue
Linear sets and semifields
In this section we recall the geometric language of linear sets associated with spread sets of semifields. This will be used in the next subsection to distinguish the semifields arising from different lifting polynomials.
Let
K = F q t and L = F q 2 t . \mathbb{K}=\mathbb{F}_{q^t}\qquad\text{and}\qquad\mathbb{L}=\mathbb{F}_{q^{2t}}. K = F q t and L = F q 2 t .
We view L \mathbb{L} L as a two-dimensional vector space over K \mathbb{K} K . Every K \mathbb{K} K -linear endomorphism of L \mathbb{L} L has the form
φ A , B : x ⟼ A x + B x q t , A , B ∈ L . \varphi_{A,B}:x\longmapsto Ax+Bx^{q^t},\qquad A,B\in\mathbb{L}. φ A , B : x ⟼ A x + B x q t , A , B ∈ L .
Thus End K ( L ) \operatorname{End}_{\mathbb{K}}(\mathbb{L}) End K ( L ) is naturally identified with the 4-dimensional K \mathbb{K} K -vector space
L × L , \mathbb{L}\times\mathbb{L}, L × L ,
by sending φ A , B \varphi_{A,B} φ A , B to the pair ( A , B ) (A,B) ( A , B ) . We shall therefore write
Σ = PG ( End K ( L ) , K ) ≅ PG ( 3 , K ) . \Sigma=\operatorname{PG}(\operatorname{End}_{\mathbb{K}}(\mathbb{L}),\mathbb{K})\cong\operatorname{PG}(3,\mathbb{K}). Σ = PG ( End K ( L ) , K ) ≅ PG ( 3 , K ) .
Let C ⊆ End K ( L ) C\subseteq\operatorname{End}_{\mathbb{K}}(\mathbb{L}) C ⊆ End K ( L ) be an F q \mathbb{F}_q F q -subspace of dimension 2 t 2t 2 t defining a semifield spread set. Equivalently, C C C contains the zero map and every nonzero element of C C C is nonsingular. Hence, we may define the associated F q \mathbb{F}_q F q -linear set by
L ( C ) = { ⟨ φ ⟩ K : 0 ≠ φ ∈ C } ⊆ Σ . L(C)=\{\langle\varphi\rangle_{\mathbb{K}}:0\ne\varphi\in C\}\subseteq\Sigma. L ( C ) = {⟨ φ ⟩ K : 0 = φ ∈ C } ⊆ Σ.
Using the identification End K ( L ) ≅ L × L \operatorname{End}_{\mathbb{K}}(\mathbb{L})\cong\mathbb{L}\times\mathbb{L} End K ( L ) ≅ L × L , this can also be written as
L ( C ) = { ⟨ ( A , B ) ⟩ K : 0 ≠ ( A , B ) ∈ C } ⊆ PG ( L × L , K ) . L(C)=\{\langle(A,B)\rangle_{\mathbb{K}}:0\ne(A,B)\in C\}\subseteq\operatorname{PG}(\mathbb{L}\times\mathbb{L},\mathbb{K}). L ( C ) = {⟨( A , B ) ⟩ K : 0 = ( A , B ) ∈ C } ⊆ PG ( L × L , K ) .
If P ∈ Σ P \in\Sigma P ∈ Σ , the weight of P P P with respect to C \mathcal{C} C is
wt C ( P ) = dim F q ( C ∩ P ) . \operatorname{wt}_{\mathcal{C}}(P)=\dim_{\mathbb{F}_q}(\mathcal{C}\cap P). wt C ( P ) = dim F q ( C ∩ P ) .
We shall use repeatedly the following standard criterion, see e.g. [29 Property 2.3] . Let
ℓ = PG ( R , K ) \ell=\operatorname{PG}(R,\mathbb{K}) ℓ = PG ( R , K )
be a line of Σ \Sigma Σ , where R R R is a two-dimensional K \mathbb{K} K -subspace of End K ( L ) \operatorname{End}_{\mathbb{K}}(L) End K ( L ) . Then
ℓ ⊆ L ( C ) ⟺ dim F q ( C ∩ R ) ≥ t + 1. (7) \ell\subseteq L(\mathcal{C})\Longleftrightarrow\dim_{\mathbb{F}_q}(\mathcal{C}\cap R)\geq t+1. \tag*{(7)} ℓ ⊆ L ( C ) ⟺ dim F q ( C ∩ R ) ≥ t + 1. (7)
The nonsingular condition for the maps in C \mathcal{C} C has a useful geometric interpretation. Recall that the map φ A , B : x ↦ A x + B x q t \varphi_{A,B}:x\mapsto Ax+Bx^{q^t} φ A , B : x ↦ A x + B x q t is singular if and only if N L / K ( A ) = N L / K ( B ) N_{L/K}(A)=N_{L/K}(B) N L / K ( A ) = N L / K ( B ) . Consequently, the singular maps define the hyperbolic quadric
Q = { ⟨ ( A , B ) ⟩ K : N L / K ( A ) = N L / K ( B ) } ⊆ Σ . \mathcal{Q}=\{\langle(A,B)\rangle_{\mathbb{K}}:N_{L/K}(A)=N_{L/K}(B)\}\subseteq\Sigma. Q = {⟨( A , B ) ⟩ K : N L / K ( A ) = N L / K ( B )} ⊆ Σ.
Since every nonzero element of a semifield spread set is nonsingular, we have
L ( C ) ∩ Q = ∅ . L(\mathcal{C})\cap\mathcal{Q}=\varnothing. L ( C ) ∩ Q = ∅ .
The polarity associated with Q \mathcal{Q} Q will be denoted by ⊥ \perp ⊥ . In particular, the two coordinate lines
ℓ = { ⟨ ( A , 0 ) ⟩ K : A ∈ L × } \ell=\{\langle(A,0)\rangle_{\mathbb{K}}:A\in L^{\times}\} ℓ = {⟨( A , 0 ) ⟩ K : A ∈ L × }
and
ℓ ⊥ = { ⟨ ( 0 , B ) ⟩ K : B ∈ L × } \ell^{\perp}=\{\langle(0,B)\rangle_{\mathbb{K}}:B\in L^{\times}\} ℓ ⊥ = {⟨( 0 , B ) ⟩ K : B ∈ L × }
are polar to each other.
We shall also use the following standard consequence of the spread-set description of isotopy. If two semifield spread sets C , C ′ ⊆ End K ( L ) \mathcal{C},\mathcal{C}'\subseteq\operatorname{End}_{\mathbb{K}}(L) C , C ′ ⊆ End K ( L ) define isotopic semifields with the same right nucleus K \mathbb{K} K , then their associated linear sets are equivalent under a semilinear collineation of Σ = PG ( 3 , K ) \Sigma=\operatorname{PG}(3,\mathbb{K}) Σ = PG ( 3 , K ) which preserves the hyperbolic quadric Q \mathcal{Q} Q , see [2 Theorem 2.1] and [7 Theorem 5.3] . In particular, such a collineation preserves the geometric properties of L ( C ) L(\mathcal{C}) L ( C ) , such as the weights of points, the lines contained in the linear set, and the polarity induced by Q \mathcal{Q} Q .
For later use we record the form of the action on the coordinate line ℓ \ell ℓ , see [7 Proposition 5.2] . Let Φ \Phi Φ be a semilinear collineation of Σ \Sigma Σ preserving Q \mathcal{Q} Q , and suppose that
Φ ( ℓ ) = ℓ . \Phi(\ell)=\ell. Φ ( ℓ ) = ℓ .
Then the induced map on ℓ \ell ℓ has one of the following two forms:
⟨ ( A , 0 ) ⟩ K ⟼ ⟨ ( λ A ρ , 0 ) ⟩ K \langle(A,0)\rangle_{\mathbb{K}}\longmapsto\langle(\lambda A^{\rho},0)\rangle_{\mathbb{K}} ⟨( A , 0 ) ⟩ K ⟼ ⟨( λ A ρ , 0 ) ⟩ K
or
⟨ ( A , 0 ) ⟩ K ⟼ ⟨ ( λ ( A ρ ) q t , 0 ) ⟩ K , \langle(A,0)\rangle_{\mathbb{K}}\longmapsto\langle(\lambda(A^{\rho})^{q^t},0)\rangle_{\mathbb{K}}, ⟨( A , 0 ) ⟩ K ⟼ ⟨( λ ( A ρ ) q t , 0 ) ⟩ K ,
where
λ ∈ L × and ρ ∈ Aut ( L ) . \lambda\in L^{\times}\qquad\text{and}\qquad\rho\in\operatorname{Aut}(L). λ ∈ L × and ρ ∈ Aut ( L ) .
Since x ↦ x q t x\mapsto x^{q^t} x ↦ x q t is the nontrivial automorphism of L / K L/K L / K , both cases can be written, after replacing ρ \rho ρ if necessary, in the form
⟨ ( A , 0 ) ⟩ K ⟼ ⟨ ( λ A ρ , 0 ) ⟩ K . \langle(A,0)\rangle_{\mathbb{K}}\longmapsto\langle(\lambda A^{\rho},0)\rangle_{\mathbb{K}}. ⟨( A , 0 ) ⟩ K ⟼ ⟨( λ A ρ , 0 ) ⟩ K .
For further details on linear sets and their connections with semifields, we refer the reader to [2 , 7 , 29 ] .
We finally recall the field-theoretic lemma that will be used to convert equivalence of linear sets into a relation between the roots of the lifting polynomials.
Lemma 6.1 (see [5 Lemma 2.4] ). Let K ′ / K K^{\prime}/K K ′ / K be a field extension of odd degree s s s . Let u , w ∈ K ′ u,w \in K^{\prime} u , w ∈ K ′ be such that
K ′ = K ( u ) = K ( w ) . K^{\prime}=K(u)=K(w). K ′ = K ( u ) = K ( w ) . For 1 ≤ ℓ ≤ ( s − 1 ) / 2 1 \le\ell\le(s-1)/2 1 ≤ ℓ ≤ ( s − 1 ) /2 , set U = ⟨ 1 , u , u 2 , … , u ℓ ⟩ K U=\langle1,u,u^2,\ldots,u^\ell\rangle_K U = ⟨ 1 , u , u 2 , … , u ℓ ⟩ K and W = ⟨ 1 , w , w 2 , … , w ℓ ⟩ K W=\langle1,w,w^2,\ldots,w^\ell\rangle_K W = ⟨ 1 , w , w 2 , … , w ℓ ⟩ K . If there exists λ ∈ K ′ × \lambda\in K^{\prime\times} λ ∈ K ′ × such that U = λ W U=\lambda W U = λW , then there exist
a , b , c , d ∈ K , a d − b c ≠ 0 , a,b,c,d\in K,\qquad ad-bc\ne0, a , b , c , d ∈ K , a d − b c = 0 , such that
w = a + b u c + d u . w=\frac{a+bu}{c+du}. w = c + d u a + b u . Conversely, if w w w has this form, then U U U is a scalar multiple of W W W .
Isotopy of the lifted semifields
We now apply the geometric machinery introduced in § 6.1 to the semifields obtained from the lifting construction. We first treat one lifting step and then derive the corresponding counting result.
Throughout this section we fix the initial EMPT spread set
C 0 = S ( u , a , b , c ) ⊆ End K 0 ( L 0 ) , \mathcal{C}_0=S(u,a,b,c)\subseteq\operatorname{End}_{K_0}(L_0), C 0 = S ( u , a , b , c ) ⊆ End K 0 ( L 0 ) ,
where
K 0 = F q t , L 0 = F q 2 t , K_0=\mathbb{F}_{q^t},\qquad L_0=\mathbb{F}_{q^{2t}}, K 0 = F q t , L 0 = F q 2 t ,
and t ≥ 3 t\ge3 t ≥ 3 is odd. We write
A u = ⟨ 1 , u , … , u ( t − 1 ) / 2 ⟩ F q \mathcal{A}_u=\langle1,u,\ldots,u^{(t-1)/2}\rangle_{\mathbb{F}_q} A u = ⟨ 1 , u , … , u ( t − 1 ) /2 ⟩ F q
and
B u = ⟨ 1 , u , … , u ( t − 3 ) / 2 ⟩ F q . \mathcal{B}_u=\langle1,u,\ldots,u^{(t-3)/2}\rangle_{\mathbb{F}_q}. B u = ⟨ 1 , u , … , u ( t − 3 ) /2 ⟩ F q .
Thus
C 0 = { x ↦ ( α + a β ) x + b ( γ + c δ ) x q t : α , β ∈ A u , γ , δ ∈ B u } . \mathcal{C}_0=\left\{x\mapsto(\alpha+a\beta)x+b(\gamma+c\delta)x^{q^t}:\alpha,\beta\in\mathcal{A}_u,\ \gamma,\delta\in\mathcal{B}_u\right\}. C 0 = { x ↦ ( α + a β ) x + b ( γ + cδ ) x q t : α , β ∈ A u , γ , δ ∈ B u } .
Let
s = 2 r + 1 ≥ 3 s=2r+1\ge3 s = 2 r + 1 ≥ 3
be odd, and let
G ( Y ) ∈ K 0 [ Y ] G(Y)\in K_0[Y] G ( Y ) ∈ K 0 [ Y ]
be monic irreducible of degree s s s . Let θ \theta θ be a root of G G G , and put
K 1 = K 0 ( θ ) = F q t s , L 1 = L 0 ( θ ) = F q 2 t s . K_1=K_0(\theta)=\mathbb{F}_{q^{ts}},\qquad L_1=L_0(\theta)=\mathbb{F}_{q^{2ts}}. K 1 = K 0 ( θ ) = F q t s , L 1 = L 0 ( θ ) = F q 2 t s .
Choose
η ∈ L 0 × and w ∈ L 1 × \eta\in L_0^\times\qquad\text{and}\qquad w\in L_1^\times η ∈ L 0 × and w ∈ L 1 ×
such that
N L 0 / K 0 ( η ) = ( − 1 ) s G ( 0 ) N_{L_0/K_0}(\eta)=(-1)^sG(0) N L 0 / K 0 ( η ) = ( − 1 ) s G ( 0 )
and
N L 1 / K 1 ( w ) = w w q t s = θ . N_{L_1/K_1}(w)=ww^{q^{ts}}=\theta. N L 1 / K 1 ( w ) = w w q t s = θ .
Let C θ C_\theta C θ be the spread set obtained from C 0 C_0 C 0 by the first lifting step. Hence C θ C_\theta C θ consists of the maps
x ↦ A x + B w x q t s , x \mapsto Ax + Bwx^{q^{ts}}, x ↦ A x + B w x q t s ,
where
A = α + a β + ∑ j = 1 r v 2 j θ j A = \alpha+ a\beta+ \sum_{j=1}^{r} v_{2j}\theta^j A = α + a β + j = 1 ∑ r v 2 j θ j
and
B = ∑ j = 0 r − 1 v 2 j + 1 θ j + b ( γ + c δ ) η θ r , B = \sum_{j=0}^{r-1} v_{2j+1}\theta^j + \frac{b(\gamma+c\delta)}{\eta}\theta^r, B = j = 0 ∑ r − 1 v 2 j + 1 θ j + η b ( γ + cδ ) θ r ,
with α , β ∈ A u \alpha,\beta\in\mathcal{A}_u α , β ∈ A u , γ , δ ∈ B u \gamma,\delta\in\mathcal{B}_u γ , δ ∈ B u , v 1 , … , v s − 1 ∈ L 0 v_1,\ldots,v_{s-1} \in\mathbb{L}_0 v 1 , … , v s − 1 ∈ L 0 .
The corresponding semifield has order
and right nucleus isomorphic to
K 1 = F q t s . \mathbb{K}_1 = \mathbb{F}_{q^{ts}}. K 1 = F q t s .
We identify the map
x ↦ A x + B w x q t s x \mapsto Ax+Bwx^{q^{ts}} x ↦ A x + B w x q t s
with its coefficient pair
( A , B w ) ∈ L 1 × L 1 . (A,Bw) \in\mathbb{L}_1 \times\mathbb{L}_1. ( A , B w ) ∈ L 1 × L 1 .
Thus the associated linear set is
L θ = L ( C θ ) ⊆ PG ( L 1 × L 1 , K 1 ) ≅ PG ( 3 , K 1 ) . L_\theta=L(C_\theta)\subseteq\operatorname{PG}(\mathbb{L}_1\times\mathbb{L}_1,\mathbb{K}_1)\cong\operatorname{PG}(3,\mathbb{K}_1). L θ = L ( C θ ) ⊆ PG ( L 1 × L 1 , K 1 ) ≅ PG ( 3 , K 1 ) .
We denote by
ℓ = { ⟨ ( A , 0 ) ⟩ K 1 : A ∈ L 1 × } \ell=\{\langle(A,0)\rangle_{\mathbb{K}_1}: A\in\mathbb{L}_1^\times\} ℓ = {⟨( A , 0 ) ⟩ K 1 : A ∈ L 1 × }
the first coordinate line and by
ℓ ⊥ = { ⟨ ( 0 , B w ) ⟩ K 1 : B ∈ L 1 × } \ell^\perp=\{\langle(0,Bw)\rangle_{\mathbb{K}_1}: B\in\mathbb{L}_1^\times\} ℓ ⊥ = {⟨( 0 , B w ) ⟩ K 1 : B ∈ L 1 × }
the second coordinate line. These two lines are polar with respect to the hyperbolic quadric of singular maps.
We first describe the intersections of the defining F q \mathbb{F}_q F q -space of L θ L_\theta L θ with the two coordinate lines. Put
U θ = A u + a A u + ∑ j = 1 r L 0 θ j ⊆ L 1 (8) U_\theta=\mathcal{A}_u+a\mathcal{A}_u+\sum_{j=1}^{r}\mathbb{L}_0\theta^j\subseteq\mathbb{L}_1 \tag*{(8)} U θ = A u + a A u + j = 1 ∑ r L 0 θ j ⊆ L 1 (8)
and
W θ = w ( ∑ j = 0 r − 1 L 0 θ j + b η ( B u + c B u ) θ r ) ⊆ L 1 . (9) W_\theta=w\left(\sum_{j=0}^{r-1}\mathbb{L}_0\theta^j+\frac{b}{\eta}(\mathcal{B}_u+c\mathcal{B}_u)\theta^r\right)\subseteq\mathbb{L}_1. \tag*{(9)} W θ = w ( j = 0 ∑ r − 1 L 0 θ j + η b ( B u + c B u ) θ r ) ⊆ L 1 . (9)
Then
C θ ∩ ( L 1 × { 0 } ) = U θ × { 0 } C_\theta\cap(\mathbb{L}_1\times\{0\})=U_\theta\times\{0\} C θ ∩ ( L 1 × { 0 }) = U θ × { 0 }
and
C θ ∩ ( { 0 } × L 1 ) = { 0 } × W θ . C_{\theta} \cap(\{0\} \times L_{1}) = \{0\} \times W_{\theta}. C θ ∩ ({ 0 } × L 1 ) = { 0 } × W θ .
Lemma 6.2. The line ℓ \ell ℓ is contained in L θ L_{\theta} L θ , whereas ℓ ⊥ \ell^{\perp} ℓ ⊥ is not contained in L θ L_{\theta} L θ . Moreover, every point of L θ ∩ ℓ L_{\theta} \cap\ell L θ ∩ ℓ has weight at most
t s + 1 2 , \frac{ts+1}{2}, 2 t s + 1 , and every point of L θ ∩ ℓ ⊥ L_{\theta} \cap\ell^{\perp} L θ ∩ ℓ ⊥ has weight at most
t s − 1 2 . \frac{ts-1}{2}. 2 t s − 1 .
Proof. Since a ∉ K 0 a \notin K_{0} a ∈ / K 0 , we have
L 0 = K 0 ⊕ a K 0 . L_{0}=K_{0}\oplus aK_{0}. L 0 = K 0 ⊕ a K 0 .
Therefore
U θ = P θ ⊕ a P θ , U_{\theta}=P_{\theta}\oplus aP_{\theta}, U θ = P θ ⊕ a P θ ,
where
P θ = A u + K 0 θ + K 0 θ 2 + ⋯ + K 0 θ r ⊆ K 1 . P_{\theta}=\mathcal{A}_{u}+K_{0}\theta+K_{0}\theta^{2}+\cdots+K_{0}\theta^{r}\subseteq K_{1}. P θ = A u + K 0 θ + K 0 θ 2 + ⋯ + K 0 θ r ⊆ K 1 .
Indeed, for j ≥ 1 j\geq1 j ≥ 1 ,
L 0 θ j = K 0 θ j ⊕ a K 0 θ j . L_{0}\theta^{j}=K_{0}\theta^{j}\oplus aK_{0}\theta^{j}. L 0 θ j = K 0 θ j ⊕ a K 0 θ j .
Hence
dim F q P θ = dim F q A u + r t = t + 1 2 + r t = t s + 1 2 , \dim_{\mathbb{F}_{q}}P_{\theta}=\dim_{\mathbb{F}_{q}}\mathcal{A}_{u}+rt=\frac{t+1}{2}+rt=\frac{ts+1}{2}, dim F q P θ = dim F q A u + r t = 2 t + 1 + r t = 2 t s + 1 ,
and consequently
dim F q U θ = t s + 1. \dim_{\mathbb{F}_{q}}U_{\theta}=ts+1. dim F q U θ = t s + 1.
Therefore, by (7) , we get
ℓ ⊆ L θ . \ell\subseteq L_{\theta}. ℓ ⊆ L θ .
We now bound the weights of the points on ℓ \ell ℓ . Let
P = ⟨ ( ρ , 0 ) ⟩ K 1 ∈ ℓ . P=\langle(\rho,0)\rangle_{K_{1}}\in\ell. P = ⟨( ρ , 0 ) ⟩ K 1 ∈ ℓ .
Write
ρ = x + a y , x , y ∈ K 1 , \rho=x+ay,\qquad x,y\in K_{1}, ρ = x + a y , x , y ∈ K 1 ,
using
L 1 = K 1 ⊕ a K 1 . L_{1}=K_{1}\oplus aK_{1}. L 1 = K 1 ⊕ a K 1 .
Then
K 1 ρ ∩ U θ K_{1}\rho\cap U_{\theta} K 1 ρ ∩ U θ
is identified with the set of all z ∈ K 1 z\in K_{1} z ∈ K 1 such that
z ρ ∈ P θ ⊕ a P θ . z\rho\in P_{\theta}\oplus aP_{\theta}. z ρ ∈ P θ ⊕ a P θ .
Since
z ρ = z x + a ( z y ) , z\rho=zx+a(zy), z ρ = z x + a ( z y ) ,
this is equivalent to
z x ∈ P θ and z y ∈ P θ . zx\in P_{\theta}\qquad\text{and}\qquad zy\in P_{\theta}. z x ∈ P θ and z y ∈ P θ .
At least one of x , y x,y x , y is nonzero. Therefore the set of such z z z has F q \mathbb{F}_q F q -dimension at most
dim F q P θ = t s + 1 2 . \dim_{\mathbb{F}_q} P_\theta= \frac{ts+1}{2}. dim F q P θ = 2 t s + 1 .
Hence every point of L θ ∩ ℓ L_\theta\cap\ell L θ ∩ ℓ has weight at most
t s + 1 2 . \frac{ts+1}{2}. 2 t s + 1 .
We now consider ℓ ⊥ \ell^\perp ℓ ⊥ . Since c ∉ K 0 c \notin\mathbb{K}_0 c ∈ / K 0 , we have
L 0 = b η K 0 ⊕ b η c K 0 . L_0 = \frac{b}{\eta}\mathbb{K}_0 \oplus\frac{b}{\eta}c\mathbb{K}_0. L 0 = η b K 0 ⊕ η b c K 0 .
Thus
w − 1 W θ = b η Q θ ⊕ b η c Q θ , w^{-1}W_\theta= \frac{b}{\eta}Q_\theta\oplus\frac{b}{\eta}cQ_\theta, w − 1 W θ = η b Q θ ⊕ η b c Q θ ,
where
Q θ = K 0 + K 0 θ + ⋯ + K 0 θ r − 1 + B u θ r ⊆ K 1 . Q_\theta= \mathbb{K}_0+\mathbb{K}_0\theta+\cdots+\mathbb{K}_0\theta^{r-1}+\mathcal{B}_u\theta^r \subseteq K_1. Q θ = K 0 + K 0 θ + ⋯ + K 0 θ r − 1 + B u θ r ⊆ K 1 .
Indeed, for 0 ≤ j ≤ r − 1 0 \le j \le r-1 0 ≤ j ≤ r − 1 ,
L 0 θ j = b η K 0 θ j ⊕ b η c K 0 θ j , L_0\theta^j=\frac{b}{\eta}\mathbb{K}_0\theta^j\oplus\frac{b}{\eta}c\mathbb{K}_0\theta^j, L 0 θ j = η b K 0 θ j ⊕ η b c K 0 θ j ,
and the highest term is
b η ( B u + c B u ) θ r . \frac{b}{\eta}(\mathcal{B}_u+c\mathcal{B}_u)\theta^r. η b ( B u + c B u ) θ r .
Therefore
dim F q Q θ = r t + t − 1 2 = t s − 1 2 , \dim_{\mathbb{F}_q} Q_\theta=rt+\frac{t-1}{2}=\frac{ts-1}{2}, dim F q Q θ = r t + 2 t − 1 = 2 t s − 1 ,
and so
dim F q W θ = t s − 1. \dim_{\mathbb{F}_q} W_\theta=ts-1. dim F q W θ = t s − 1.
Again (7) implies that
ℓ ⊥ ⊈ L θ . \ell^\perp\not\subseteq L_\theta. ℓ ⊥ ⊆ L θ .
Arguing as before, we get that every point of L θ ∩ ℓ ⊥ L_\theta\cap\ell^\perp L θ ∩ ℓ ⊥ has weight at most
t s − 1 2 . \frac{ts-1}{2}. 2 t s − 1 .
Proposition 6.3. The line ℓ \ell ℓ is the unique line of PG ( L 1 × L 1 , K 1 ) \operatorname{PG}(L_1 \times L_1,K_1) PG ( L 1 × L 1 , K 1 ) contained in L θ L_\theta L θ .
Proof. By Lemma 6.2, we know that ℓ ⊆ L θ \ell\subseteq L_\theta ℓ ⊆ L θ . Suppose, by contradiction, that there exists another line
such that ℓ ′ ⊆ L θ \ell' \subseteq L_\theta ℓ ′ ⊆ L θ .
Let R R R and S S S be the two-dimensional K 1 K_1 K 1 -subspaces of L 1 × L 1 L_1 \times L_1 L 1 × L 1 defining ℓ \ell ℓ and ℓ ′ \ell' ℓ ′ , respectively. Since both ℓ \ell ℓ and ℓ ′ \ell' ℓ ′ are contained in L θ L_\theta L θ , by (7), we have
dim F q ( C θ ∩ R ) ≥ t s + 1 \dim_{\mathbb{F}_q}(C_\theta\cap R) \ge ts+1 dim F q ( C θ ∩ R ) ≥ t s + 1 and
dim F q ( C θ ∩ S ) ≥ t s + 1. \dim_{\mathbb{F}_q}(C_\theta\cap S) \ge ts+1. dim F q ( C θ ∩ S ) ≥ t s + 1.
Since dim F q C θ = 2 t s \dim_{\mathbb{F}_q} C_\theta= 2ts dim F q C θ = 2 t s , we get
( C θ ∩ R ) ∩ ( C θ ∩ S ) ≠ { 0 } . (C_\theta\cap R) \cap(C_\theta\cap S) \ne\{0\}. ( C θ ∩ R ) ∩ ( C θ ∩ S ) = { 0 } .
Therefore R ∩ S ≠ { 0 } R \cap S \ne\{0\} R ∩ S = { 0 } , and the projective lines ℓ \ell ℓ and ℓ ′ \ell' ℓ ′ meet in a point. Let
T = ℓ ∩ ℓ ′ . T = \ell\cap\ell'. T = ℓ ∩ ℓ ′ .
Let
π = ⟨ ℓ , ℓ ′ ⟩ \pi= \langle\ell, \ell' \rangle π = ⟨ ℓ , ℓ ′ ⟩
be the plane generated by ℓ \ell ℓ and ℓ ′ \ell' ℓ ′ , and let Π \Pi Π be the corresponding three-dimensional K 1 \mathbb{K}_1 K 1 -subspace of L 1 × L 1 L_1 \times L_1 L 1 × L 1 . Then
dim F q ( C θ ∩ Π ) ≥ dim F q ( C θ ∩ R ) + dim F q ( C θ ∩ S ) − wt C θ ( T ) . \dim_{\mathbb{F}_q}(C_\theta\cap\Pi) \ge\dim_{\mathbb{F}_q}(C_\theta\cap R) + \dim_{\mathbb{F}_q}(C_\theta\cap S) - \operatorname{wt}_{C_\theta}(T). dim F q ( C θ ∩ Π ) ≥ dim F q ( C θ ∩ R ) + dim F q ( C θ ∩ S ) − wt C θ ( T ) .
By Lemma 6.2,
wt C θ ( T ) ≤ t s + 1 2 . \operatorname{wt}_{C_\theta}(T) \le\frac{ts+1}{2}. wt C θ ( T ) ≤ 2 t s + 1 .
Therefore
dim F q ( C θ ∩ Π ) ≥ ( t s + 1 ) + ( t s + 1 ) − t s + 1 2 = 3 t s + 3 2 . \dim_{\mathbb{F}_q}(C_\theta\cap\Pi) \ge(ts+1) + (ts+1) - \frac{ts+1}{2} = \frac{3ts+3}{2}. dim F q ( C θ ∩ Π ) ≥ ( t s + 1 ) + ( t s + 1 ) − 2 t s + 1 = 2 3 t s + 3 .
Now ℓ ⊥ \ell^\perp ℓ ⊥ meets the plane π \pi π in a unique point, say P P P . Let R ⊥ R^\perp R ⊥ be the two-dimensional K 1 \mathbb{K}_1 K 1 -subspace defining ℓ ⊥ \ell^\perp ℓ ⊥ . Since
C θ = ( C θ ∩ R ) + ( C θ ∩ R ⊥ ) C_\theta= (C_\theta\cap R) + (C_\theta\cap R^\perp) C θ = ( C θ ∩ R ) + ( C θ ∩ R ⊥ )
and R ⊆ Π R \subseteq\Pi R ⊆ Π , we have
C θ = ⟨ C θ ∩ Π , C θ ∩ R ⊥ ⟩ F q . C_\theta= \langle C_\theta\cap\Pi, C_\theta\cap R^\perp\rangle_{\mathbb{F}_q}. C θ = ⟨ C θ ∩ Π , C θ ∩ R ⊥ ⟩ F q .
Hence
2 t s = dim F q C θ ≥ dim F q ( C θ ∩ Π ) + dim F q ( C θ ∩ R ⊥ ) − wt C θ ( P ) . 2ts = \dim_{\mathbb{F}_q} C_\theta\ge\dim_{\mathbb{F}_q}(C_\theta\cap\Pi) + \dim_{\mathbb{F}_q}(C_\theta\cap R^\perp) - \operatorname{wt}_{C_\theta}(P). 2 t s = dim F q C θ ≥ dim F q ( C θ ∩ Π ) + dim F q ( C θ ∩ R ⊥ ) − wt C θ ( P ) .
Using Lemma 6.2, we have
dim F q ( C θ ∩ R ⊥ ) = t s − 1 \dim_{\mathbb{F}_q}(C_\theta\cap R^\perp) = ts - 1 dim F q ( C θ ∩ R ⊥ ) = t s − 1
and
wt C θ ( P ) ≤ t s − 1 2 . \operatorname{wt}_{C_\theta}(P) \le\frac{ts-1}{2}. wt C θ ( P ) ≤ 2 t s − 1 .
Thus
2 t s ≥ 3 t s + 3 2 + ( t s − 1 ) − t s − 1 2 = 2 t s + 1 , 2ts \ge\frac{3ts+3}{2} + (ts-1) - \frac{ts-1}{2} = 2ts+1, 2 t s ≥ 2 3 t s + 3 + ( t s − 1 ) − 2 t s − 1 = 2 t s + 1 ,
a contradiction. Therefore ℓ \ell ℓ is the unique line contained in L θ L_\theta L θ . □
Let H ( Y ) ∈ K 0 [ Y ] H(Y) \in\mathbb{K}_0[Y] H ( Y ) ∈ K 0 [ Y ] be another monic irreducible polynomial of degree s s s , and let θ ′ \theta' θ ′ be a root of H H H . Let C θ ′ C_{\theta'} C θ ′ be the lifted spread set obtained from the same initial spread set C 0 C_0 C 0 , using θ ′ \theta' θ ′ instead of θ \theta θ .
Theorem 6.4. Assume that the semifields defined by C θ C_\theta C θ and C θ ′ C_{\theta'} C θ ′ are isotopic. Then there exist
κ ∈ K 0 × \kappa\in\mathbb{K}_0^\times κ ∈ K 0 × and an automorphism ρ ∈ Aut ( K 1 ) \rho\in\operatorname{Aut}(\mathbb{K}_1) ρ ∈ Aut ( K 1 ) such that
θ ′ = κ θ ρ . \theta' = \kappa\theta^\rho. θ ′ = κ θ ρ .
Proof. The two semifields have the same order and their right nuclei are both isomorphic to K 1 = F q t s \mathbb{K}_1=\mathbb{F}_{q^{ts}} K 1 = F q t s . By the geometric description of isotopy recalled in Subsection 6.1, the isotopy induces a semilinear collineation Φ \Phi Φ such that
Φ ( L θ ) = L θ ′ . \Phi(L_\theta)=L_{\theta'}. Φ ( L θ ) = L θ ′ .
Moreover, Φ \Phi Φ preserves the hyperbolic quadric of singular maps. By Proposition 6.3, the line ℓ \ell ℓ is the unique line contained in L θ L_\theta L θ , and the same is true for L θ ′ L_{\theta'} L θ ′ . Hence
Φ ( ℓ ) = ℓ . \Phi(\ell)=\ell. Φ ( ℓ ) = ℓ .
By [7 Proposition 5.2] , the action induced by Φ \Phi Φ on ℓ \ell ℓ is of the form
⟨ ( A , 0 ) ⟩ K 1 ⟼ ⟨ ( λ A ρ , 0 ) ⟩ K 1 \langle(A,0)\rangle_{\mathbb{K}_1}\longmapsto\langle(\lambda A^\rho,0)\rangle_{\mathbb{K}_1} ⟨( A , 0 ) ⟩ K 1 ⟼ ⟨( λ A ρ , 0 ) ⟩ K 1
for some
λ ∈ L 1 × and ρ ∈ Aut ( L 1 ) . \lambda\in\mathbb{L}_1^\times\qquad\text{and}\qquad\rho\in\operatorname{Aut}(\mathbb{L}_1). λ ∈ L 1 × and ρ ∈ Aut ( L 1 ) .
Since finite fields have a unique subfield of each order, ρ \rho ρ stabilizes K 1 \mathbb{K}_1 K 1 , L 0 \mathbb{L}_0 L 0 , and K 0 \mathbb{K}_0 K 0 . We shall denote again by ρ \rho ρ its restriction to K 1 \mathbb{K}_1 K 1 .
We have
L 1 = L 0 ( θ ρ ) = L 0 ( θ ′ ) . \mathbb{L}_1=\mathbb{L}_0(\theta^\rho)=\mathbb{L}_0(\theta'). L 1 = L 0 ( θ ρ ) = L 0 ( θ ′ ) .
because L 0 \mathbb{L}_0 L 0 and K 1 \mathbb{K}_1 K 1 are linearly disjoint over K 0 \mathbb{K}_0 K 0 . The collineation Φ \Phi Φ maps the defining F q \mathbb{F}_q F q -subspace of L θ L_\theta L θ onto the defining F q \mathbb{F}_q F q -subspace of L θ ′ L_{\theta'} L θ ′ . Since Φ ( ℓ ) = ℓ \Phi(\ell)=\ell Φ ( ℓ ) = ℓ , it maps the intersection with the first coordinate space onto the corresponding intersection for θ ′ \theta' θ ′ . Therefore
λ U θ ρ = U θ ′ . \lambda U_\theta^\rho=U_{\theta'}. λ U θ ρ = U θ ′ .
For z ∈ { θ ρ , θ ′ } z\in\{\theta^\rho,\theta'\} z ∈ { θ ρ , θ ′ } , set
S z = L 0 + L 0 z + ⋯ + L 0 z r , S_z=\mathbb{L}_0+\mathbb{L}_0z+\cdots+\mathbb{L}_0z^r, S z = L 0 + L 0 z + ⋯ + L 0 z r ,
J z = L 0 z + ⋯ + L 0 z r , J_z=\mathbb{L}_0z+\cdots+\mathbb{L}_0z^r, J z = L 0 z + ⋯ + L 0 z r ,
and
R z = L 0 + L 0 z + ⋯ + L 0 z r − 1 . R_z=\mathbb{L}_0+\mathbb{L}_0z+\cdots+\mathbb{L}_0z^{r-1}. R z = L 0 + L 0 z + ⋯ + L 0 z r − 1 .
Recalling the definition of U θ U_\theta U θ in (8) , we claim that J θ ′ J_{\theta'} J θ ′ is the largest L 0 \mathbb{L}_0 L 0 -subspace contained in U θ ′ U_{\theta'} U θ ′ , and similarly J θ ρ J_{\theta^\rho} J θ ρ is the largest L 0 \mathbb{L}_0 L 0 -subspace contained in U θ ρ U_\theta^\rho U θ ρ . Indeed, the L 0 \mathbb{L}_0 L 0 -component of an element of U θ ′ U_{\theta'} U θ ′ , with respect to the decomposition S θ ′ = L 0 ⊕ J θ ′ S_{\theta'}=\mathbb{L}_0\oplus J_{\theta'} S θ ′ = L 0 ⊕ J θ ′ , lies in
A u + a A u , \mathcal{A}u+a\mathcal{A}u, A u + a A u ,
which has F q \mathbb{F}_q F q -dimension t + 1 t+1 t + 1 . Since t ≥ 3 t\geq3 t ≥ 3 , this is strictly smaller than
dim F q L 0 = 2 t . \dim_{\mathbb{F}_q}\mathbb{L}_0=2t. dim F q L 0 = 2 t .
Hence any L 0 \mathbb{L}_0 L 0 -subspace contained in U θ ′ U_{\theta'} U θ ′ must have zero L 0 \mathbb{L}_0 L 0 -component, and is therefore contained in J θ ′ J_{\theta'} J θ ′ . The same argument applies to U θ ρ U_\theta^\rho U θ ρ , because ρ \rho ρ stabilizes L 0 \mathbb{L}_0 L 0 and preserves F q \mathbb{F}_q F q -dimensions. Consequently, from λ U θ ρ = U θ ′ \lambda U_\theta^\rho=U_{\theta'} λ U θ ρ = U θ ′ we get
λ S θ ρ = S θ ′ and λ J θ ρ = J θ ′ . \lambda S_{\theta^\rho}=S_{\theta'}\qquad\text{and}\qquad\lambda J_{\theta^\rho}=J_{\theta'}. λ S θ ρ = S θ ′ and λ J θ ρ = J θ ′ .
Applying Lemma 6.1 to the equality λ S θ ρ = S θ ′ \lambda S_{\theta^\rho}=S_{\theta'} λ S θ ρ = S θ ′ , with base field L 0 \mathbb{L}_0 L 0 , gives
θ ′ = A + B θ ρ C + D θ ρ . \theta'=\frac{A+B\theta^\rho}{C+D\theta^\rho}. θ ′ = C + D θ ρ A + B θ ρ .
for some A , B , C , D ∈ L 0 A,B,C,D \in\mathbb{L}_0 A , B , C , D ∈ L 0 , A D − B C ≠ 0 AD-BC \ne0 A D − B C = 0 .
We now show that A = 0 A=0 A = 0 . Put
Δ = ( C + D θ ρ ) r . \Delta=(C+D\theta^\rho)^r. Δ = ( C + D θ ρ ) r .
Then
Δ S θ ′ = S θ ρ . (10) \Delta S_{\theta'}=S_{\theta\rho}. \tag*{(10)} Δ S θ ′ = S θ ρ . (10)
Since λ S θ ρ = S θ ′ \lambda S_{\theta\rho}=S_{\theta'} λ S θ ρ = S θ ′ , we have
Δ λ S θ ρ = S θ ρ . \Delta\lambda S_{\theta\rho}=S_{\theta\rho}. Δ λ S θ ρ = S θ ρ .
We shall use the following elementary observation: if α S θ ρ = S θ ρ \alpha S_{\theta\rho}=S_{\theta\rho} α S θ ρ = S θ ρ for some α ∈ L 1 × \alpha\in\mathbb{L}_1^\times α ∈ L 1 × , then α ∈ L 0 × \alpha\in\mathbb{L}_0^\times α ∈ L 0 × . Indeed, S θ ρ S_{\theta\rho} S θ ρ is then a vector space over L 0 ( α ) \mathbb{L}_0(\alpha) L 0 ( α ) . Hence [ L 0 ( α ) : L 0 ] [\mathbb{L}_0(\alpha):\mathbb{L}_0] [ L 0 ( α ) : L 0 ] divides dim L 0 S θ ρ = r + 1 \dim_{\mathbb{L}_0}S_{\theta\rho}=r+1 dim L 0 S θ ρ = r + 1 . On the other hand, [ L 0 ( α ) : L 0 ] [\mathbb{L}_0(\alpha):\mathbb{L}_0] [ L 0 ( α ) : L 0 ] divides [ L 1 : L 0 ] = 2 r + 1 [\mathbb{L}_1:\mathbb{L}_0]=2r+1 [ L 1 : L 0 ] = 2 r + 1 . Since
gcd ( r + 1 , 2 r + 1 ) = 1 , \gcd(r+1,2r+1)=1, g cd( r + 1 , 2 r + 1 ) = 1 ,
we get α ∈ L 0 × \alpha\in\mathbb{L}_0^\times α ∈ L 0 × . It follows that
Δ λ ∈ L 0 × . (11) \Delta\lambda\in\mathbb{L}_0^\times. \tag*{(11)} Δ λ ∈ L 0 × . (11)
Therefore
Δ J θ ′ = Δ λ J θ ρ = J θ ρ . \Delta J_{\theta'}=\Delta\lambda J_{\theta\rho}=J_{\theta\rho}. Δ J θ ′ = Δ λ J θ ρ = J θ ρ .
In particular,
Δ ( θ ′ ) r ∈ J θ ρ . \Delta(\theta')^r\in J_{\theta\rho}. Δ ( θ ′ ) r ∈ J θ ρ .
But
Δ ( θ ′ ) r = ( C + D θ ρ ) r ( A + B θ ρ C + D θ ρ ) r = ( A + B θ ρ ) r . \Delta(\theta')^r=(C+D\theta^\rho)^r\left(\frac{A+B\theta^\rho}{C+D\theta^\rho}\right)^r=(A+B\theta^\rho)^r. Δ ( θ ′ ) r = ( C + D θ ρ ) r ( C + D θ ρ A + B θ ρ ) r = ( A + B θ ρ ) r .
Since
J θ ρ = L 0 θ ρ + ⋯ + L 0 ( θ ρ ) r J_{\theta\rho}=\mathbb{L}_0\theta^\rho+\cdots+\mathbb{L}_0(\theta^\rho)^r J θ ρ = L 0 θ ρ + ⋯ + L 0 ( θ ρ ) r
consists of the elements of S θ ρ S_{\theta\rho} S θ ρ with zero constant coefficient, the constant coefficient of ( A + B θ ρ ) r (A+B\theta^\rho)^r ( A + B θ ρ ) r must be zero. This coefficient is A r A^r A r , and so A = 0 A=0 A = 0 . Thus
θ ′ = B θ ρ C + D θ ρ , B C ≠ 0. \theta'=\frac{B\theta^\rho}{C+D\theta^\rho},\qquad BC\ne0. θ ′ = C + D θ ρ B θ ρ , B C = 0.
It remains to prove that D = 0 D=0 D = 0 . Since Φ \Phi Φ preserves the hyperbolic quadric and Φ ( ℓ ) = ℓ \Phi(\ell)=\ell Φ ( ℓ ) = ℓ , it also preserves the polar line ℓ ⊥ \ell^\perp ℓ ⊥ . Hence
Φ ( ℓ ⊥ ) = ℓ ⊥ . \Phi(\ell^\perp)=\ell^\perp. Φ ( ℓ ⊥ ) = ℓ ⊥ .
Using again the form of the action on a coordinate line, there exists μ ∈ L 1 × \mu\in\mathbb{L}_1^\times μ ∈ L 1 × such that the induced action on ℓ ⊥ \ell^\perp ℓ ⊥ gives
μ W θ ρ = W θ ′ . \mu W_{\theta^\rho}=W_{\theta'}. μ W θ ρ = W θ ′ .
Write
W θ = w T θ , W_\theta=wT_\theta, W θ = w T θ ,
where
T θ = ∑ j = 0 r − 1 L 0 θ j + b η ( B u + c B u ) θ r . T_\theta=\sum_{j=0}^{r-1}\mathbb{L}_0\theta^j+\frac{b}{\eta}(B_u+cB_u)\theta^r. T θ = j = 0 ∑ r − 1 L 0 θ j + η b ( B u + c B u ) θ r .
The L 0 \mathbb{L}_0 L 0 -span of T θ ′ T_{\theta}^{\prime} T θ ′ is S θ ′ S_{\theta}^{\prime} S θ ′ , and the largest L 0 \mathbb{L}_0 L 0 -subspace contained in T θ ′ T_{\theta}^{\prime} T θ ′ is R θ ′ R_{\theta}^{\prime} R θ ′ . Indeed, the coefficient of θ r \theta^{r} θ r belongs to
b η r ( B u + c B u ) , \frac{b}{\eta^{r}}(\mathcal{B}_{u}+c\mathcal{B}_{u}), η r b ( B u + c B u ) ,
which has F q \mathbb{F}_q F q -dimension t − 1 < 2 t t-1<2t t − 1 < 2 t . The same argument applies to T θ ρ T_{\theta}^{\rho} T θ ρ . Therefore, after removing the nonzero scalar factors w ρ w^{\rho} w ρ and the corresponding choice w ′ w^{\prime} w ′ for θ ′ \theta^{\prime} θ ′ , the equality μ W θ ρ = W θ ′ \mu W_{\theta}^{\rho}=W_{\theta^{\prime}} μ W θ ρ = W θ ′ gives a scalar ν ∈ L × \nu\in\mathbb{L}^{\times} ν ∈ L × such that
ν S θ ρ = S θ ′ and ν R θ ρ = R θ ′ . \nu S_{\theta^{\rho}}=S_{\theta^{\prime}}\qquad\text{and}\qquad\nu R_{\theta^{\rho}}=R_{\theta^{\prime}}. ν S θ ρ = S θ ′ and ν R θ ρ = R θ ′ .
By (10), we obtain
Δ ν S θ ρ = S θ ρ . \Delta_{\nu}S_{\theta^{\rho}}=S_{\theta^{\rho}}. Δ ν S θ ρ = S θ ρ .
By the same computation for (11),
Δ ν ∈ L 0 × . \Delta_{\nu}\in\mathbb{L}_{0}^{\times}. Δ ν ∈ L 0 × .
Therefore
Δ R θ ′ = Δ ν R θ ρ = R θ ρ . \Delta R_{\theta^{\prime}}=\Delta_{\nu}R_{\theta^{\rho}}=R_{\theta^{\rho}}. Δ R θ ′ = Δ ν R θ ρ = R θ ρ .
Since 1 ∈ R θ ′ 1\in R_{\theta^{\prime}} 1 ∈ R θ ′ , this implies
Δ = ( C + D θ ρ ) r ∈ R θ ρ . \Delta=(C+D\theta^{\rho})^{r}\in R_{\theta^{\rho}}. Δ = ( C + D θ ρ ) r ∈ R θ ρ .
But
R θ ρ = L 0 + L 0 θ ρ + ⋯ + L 0 ( θ ρ ) r − 1 R_{\theta^{\rho}}=\mathbb{L}_{0}+\mathbb{L}_{0}\theta^{\rho}+\cdots+\mathbb{L}_{0}(\theta^{\rho})^{r-1} R θ ρ = L 0 + L 0 θ ρ + ⋯ + L 0 ( θ ρ ) r − 1
consists of the elements of S θ ρ S_{\theta^{\rho}} S θ ρ with zero coefficient of ( θ ρ ) r (\theta^{\rho})^{r} ( θ ρ ) r . The coefficient of ( θ ρ ) r (\theta^{\rho})^{r} ( θ ρ ) r in ( C + D θ ρ ) r (C+D\theta^{\rho})^{r} ( C + D θ ρ ) r is D r D^{r} D r . Hence D = 0 D=0 D = 0 . Thus
θ ′ = B C θ ρ . \theta^{\prime}=\frac{B}{C}\theta^{\rho}. θ ′ = C B θ ρ .
Set
κ = B C ∈ L 0 × . \kappa=\frac{B}{C}\in\mathbb{L}_{0}^{\times}. κ = C B ∈ L 0 × .
Since θ ρ , θ ′ ∈ K 1 \theta^{\rho},\theta^{\prime}\in K_{1} θ ρ , θ ′ ∈ K 1 , we also have
κ = θ ′ θ ρ ∈ K 1 × . \kappa=\frac{\theta^{\prime}}{\theta^{\rho}}\in K_{1}^{\times}. κ = θ ρ θ ′ ∈ K 1 × .
Finally,
K 1 ∩ L 0 = K 0 , K_{1}\cap\mathbb{L}_{0}=K_{0}, K 1 ∩ L 0 = K 0 ,
because [ K 1 : K 0 ] = s [K_{1}:K_{0}]=s [ K 1 : K 0 ] = s is odd and [ L 0 : K 0 ] = 2 [\mathbb{L}_{0}:K_{0}]=2 [ L 0 : K 0 ] = 2 . Hence
κ ∈ K 0 × . \kappa\in K_{0}^{\times}. κ ∈ K 0 × .
This proves the theorem. □
We now count the number of isotopy classes obtained by varying the lifting polynomial G G G . Let
Q = ∣ K 0 ∣ = q t . Q=|K_{0}|=q^{t}. Q = ∣ K 0 ∣ = q t .
Denote by
I Q ( s ) = 1 s ∑ d ∣ s μ ( d ) Q s / d I_{Q}(s)=\frac{1}{s}\sum_{d\mid s}\mu(d)Q^{s/d} I Q ( s ) = s 1 d ∣ s ∑ μ ( d ) Q s / d
the number of monic irreducible polynomials of degree s s s over K 0 K_{0} K 0 .
Theorem 6.5. Fix the initial spread set C 0 = S ( u , a , b , c ) \mathcal{C}_0=S(u,a,b,c) C 0 = S ( u , a , b , c ) and fix an odd integer s ≥ 3 s \ge3 s ≥ 3 . Let F ~ s \widetilde{\mathcal{F}}_s F s be the family of semifields arising from the first lifting step. Then F ~ s \widetilde{\mathcal{F}}_s F s contains at least
⌈ I Q ( s ) ∣ Aut ( K 0 ) ∣ ( Q − 1 ) ⌉ \left\lceil\frac{I_Q(s)}{|\operatorname{Aut}(\mathbb{K}_0)|(Q-1)} \right\rceil ⌈ ∣ Aut ( K 0 ) ∣ ( Q − 1 ) I Q ( s ) ⌉ pairwise non-isotopic semifields of order
q 2 t s = Q 2 s . q^{2ts}=Q^{2s}. q 2 t s = Q 2 s .
Proof. By Theorem 6.4, two such polynomials can give isotopic semifields only if their roots are related by
θ ′ = κ θ ρ , \theta'=\kappa\theta^\rho, θ ′ = κ θ ρ ,
with
κ ∈ K 0 × and ρ ∈ Aut ( K 1 ) . \kappa\in\mathbb{K}_0^\times\qquad\text{and}\qquad\rho\in\operatorname{Aut}(\mathbb{K}_1). κ ∈ K 0 × and ρ ∈ Aut ( K 1 ) .
Let τ = ρ ∣ K 0 ∈ Aut ( K 0 ) \tau=\left.\rho\right|_{\mathbb{K}_0}\in\operatorname{Aut}(\mathbb{K}_0) τ = ρ ∣ K 0 ∈ Aut ( K 0 ) . If G ( Y ) G(Y) G ( Y ) is the minimal polynomial of θ \theta θ over K 0 \mathbb{K}_0 K 0 , then θ ρ \theta^\rho θ ρ is a root of the polynomial obtained by applying τ \tau τ to the coefficients of G G G , which we denote by G τ ( Y ) G^\tau(Y) G τ ( Y ) . Therefore κ θ ρ \kappa\theta^\rho κ θ ρ is a root of
G κ , τ ( Y ) = κ s G τ ( Y κ ) . G_{\kappa,\tau}(Y)=\kappa^sG^\tau\left(\frac{Y}{\kappa}\right). G κ , τ ( Y ) = κ s G τ ( κ Y ) .
For a fixed G G G , the number of polynomials that can arise in this way is at most
∣ K 0 × ∣ ∣ Aut ( K 0 ) ∣ = ( Q − 1 ) ∣ Aut ( K 0 ) ∣ . |\mathbb{K}_0^\times||\operatorname{Aut}(\mathbb{K}_0)|=(Q-1)|\operatorname{Aut}(\mathbb{K}_0)|. ∣ K 0 × ∣∣ Aut ( K 0 ) ∣ = ( Q − 1 ) ∣ Aut ( K 0 ) ∣.
The automorphisms of K 1 \mathbb{K}_1 K 1 which fix K 0 \mathbb{K}_0 K 0 only replace a root by another root of the same irreducible polynomial, and hence do not produce further polynomials. Therefore each isotopy class contains at most
( Q − 1 ) ∣ Aut ( K 0 ) ∣ (Q-1)|\operatorname{Aut}(\mathbb{K}_0)| ( Q − 1 ) ∣ Aut ( K 0 ) ∣
choices of the polynomial G G G . It follows that the number of isotopy classes is at least
⌈ I Q ( s ) ∣ Aut ( K 0 ) ∣ ( Q − 1 ) ⌉ , \left\lceil\frac{I_Q(s)}{|\operatorname{Aut}(\mathbb{K}_0)|(Q-1)} \right\rceil, ⌈ ∣ Aut ( K 0 ) ∣ ( Q − 1 ) I Q ( s ) ⌉ ,
as claimed. □
Theorem 6.5 shows that, for fixed odd integers t , s ≥ 3 t,s\ge3 t , s ≥ 3 and q = p j q=p^j q = p j , where p p p is prime, the first lifting step gives at least
Ω t , s ( q t ( s − 1 ) s f ) \Omega_{t,s}\left(\frac{q^{t(s-1)}}{sf}\right) Ω t , s ( s f q t ( s − 1 ) )
pairwise non-isotopic semifields of order
which are two-dimensional over their right nucleus of order q t s q^{ts} q t s , as q → + ∞ q\to+\infty q → + ∞ .
Isotopy for the iteration
We now explain how the isotopy criterion obtained for the first lifting step extends to every level of the iterative construction. The point is that, at each step, the new spread set has the same coordinate-line structure as in the first lift. Hence one can vary the polynomial used in the quotient construction at any prescribed level and still obtain many pairwise non-isotopic semifields.
We keep the notation of Section 5. Thus
K i = F q t i , L i = F q 2 t i , t i = t s 1 ⋯ s i , \mathbb{K}_i = \mathbb{F}_{q^{t_i}}, \qquad\mathbb{L}_i = \mathbb{F}_{q^{2t_i}}, \qquad t_i = t s_1 \cdots s_i, K i = F q t i , L i = F q 2 t i , t i = t s 1 ⋯ s i ,
and
C i = { x ↦ A i ( p i ) x + B i ( p i ) x q t i : p i ∈ P i } ⊆ End K i ( L i ) \mathcal{C}_i = \left\{x \mapsto A_i(p_i)x + B_i(p_i)x^{q^{t_i}} : p_i \in\mathcal{P}_i\right\} \subseteq\operatorname{End}_{\mathbb{K}_i}(\mathbb{L}_i) C i = { x ↦ A i ( p i ) x + B i ( p i ) x q t i : p i ∈ P i } ⊆ End K i ( L i )
is the spread set obtained after i i i lifting steps.
Let
ℓ i = { ⟨ ( A , 0 ) ⟩ K i : A ∈ L i × } \ell_i = \left\{\langle(A,0)\rangle_{\mathbb{K}_i} : A \in\mathbb{L}_i^{\times}\right\} ℓ i = { ⟨( A , 0 ) ⟩ K i : A ∈ L i × }
and
ℓ i ⊥ = { ⟨ ( 0 , B ) ⟩ K i : B ∈ L i × } \ell_i^\perp= \left\{\langle(0,B)\rangle_{\mathbb{K}_i} : B \in\mathbb{L}_i^{\times}\right\} ℓ i ⊥ = { ⟨( 0 , B ) ⟩ K i : B ∈ L i × }
be the two coordinate lines in
PG ( L i × L i , K i ) ≅ PG ( 3 , K i ) . \operatorname{PG}(\mathbb{L}_i \times\mathbb{L}_i,\mathbb{K}_i) \cong\operatorname{PG}(3,\mathbb{K}_i). PG ( L i × L i , K i ) ≅ PG ( 3 , K i ) .
We denote by
U i = { A ∈ L i : ( A , 0 ) ∈ C i } U_i = \left\{A \in\mathbb{L}_i : (A,0) \in\mathcal{C}_i\right\} U i = { A ∈ L i : ( A , 0 ) ∈ C i }
and
W i = { B ∈ L i : ( 0 , B ) ∈ C i } . W_i = \left\{B \in\mathbb{L}_i : (0,B) \in\mathcal{C}_i\right\}. W i = { B ∈ L i : ( 0 , B ) ∈ C i } .
We first record the recursive form of these two coordinate sections.
Lemma 6.6. For every i ≥ 0 i \ge0 i ≥ 0 , there exist F q \mathbb{F}_q F q -subspaces
P i , Q i ⊆ K i \mathcal{P}_i,\mathcal{Q}_i \subseteq\mathbb{K}_i P i , Q i ⊆ K i and an element d i ∈ L i × d_i \in\mathbb{L}_i^{\times} d i ∈ L i × such that
U i = P i ⊕ a P i U_i = \mathcal{P}_i \oplus a\mathcal{P}_i U i = P i ⊕ a P i and
W i = d i ( Q i ⊕ c Q i ) . W_i = d_i(\mathcal{Q}_i \oplus c\mathcal{Q}_i). W i = d i ( Q i ⊕ c Q i ) . Moreover,
dim F q P i = t i + 1 2 , dim F q Q i = t i − 1 2 . \dim_{\mathbb{F}_q}\mathcal{P}_i = \frac{t_i+1}{2}, \qquad\dim_{\mathbb{F}_q}\mathcal{Q}_i = \frac{t_i-1}{2}. dim F q P i = 2 t i + 1 , dim F q Q i = 2 t i − 1 .
Proof. We argue by induction on i i i , extending the decomposition used in the proof of Lemma 6.2 to every step of the iteration. For i = 0 i=0 i = 0 , we have
U 0 = A u + a A u U_0 = \mathcal{A}_u + a\mathcal{A}_u U 0 = A u + a A u
and
W 0 = b ( B u + c B u ) . W_0 = b(\mathcal{B}_u + c\mathcal{B}_u). W 0 = b ( B u + c B u ) .
Hence we may take
P 0 = A u , Q 0 = B u , d 0 = b . \mathcal{P}_0 = \mathcal{A}_u, \qquad\mathcal{Q}_0 = \mathcal{B}_u, \qquad d_0 = b. P 0 = A u , Q 0 = B u , d 0 = b .
The asserted dimensions are precisely
dim F q A u = t + 1 2 , dim F q B u = t − 1 2 . \dim_{\mathbb{F}_q}\mathcal{A}_u=\frac{t+1}{2},\qquad\dim_{\mathbb{F}_q}\mathcal{B}_u=\frac{t-1}{2}. dim F q A u = 2 t + 1 , dim F q B u = 2 t − 1 .
Assume now that the assertion holds for i − 1 i-1 i − 1 . Write
s i = 2 r i + 1. s_i=2r_i+1. s i = 2 r i + 1.
By the recursive definition of C i C_i C i , an element of C i C_i C i has first coordinate
A i = A i − 1 + ∑ j = 1 r i v i , 2 j θ i j A_i=A_{i-1}+\sum_{j=1}^{r_i}v_{i,2j}\theta_i^j A i = A i − 1 + j = 1 ∑ r i v i , 2 j θ i j
and second coordinate
B i = ( ∑ j = 0 r i − 1 v i , 2 j + 1 θ i j + B i − 1 η i θ i r i ) w i . B_i=\left(\sum_{j=0}^{r_{i-1}}v_{i,2j+1}\theta_i^j+\frac{B_{i-1}}{\eta_i}\theta_i^{r_i}\right)w_i. B i = ( j = 0 ∑ r i − 1 v i , 2 j + 1 θ i j + η i B i − 1 θ i r i ) w i .
Therefore
U i = U i − 1 + L i − 1 θ i + ⋯ + L i − 1 θ i r i . U_i=U_{i-1}+L_{i-1}\theta_i+\cdots+L_{i-1}\theta_i^{r_i}. U i = U i − 1 + L i − 1 θ i + ⋯ + L i − 1 θ i r i .
Since a ∉ K 0 a\notin K_0 a ∈ / K 0 , and since
K i − 1 ∩ L 0 = K 0 , K_{i-1}\cap L_0=K_0, K i − 1 ∩ L 0 = K 0 ,
we have a ∉ K i − 1 a\notin K_{i-1} a ∈ / K i − 1 . Hence
L i − 1 = K i − 1 ⊕ a K i − 1 . L_{i-1}=K_{i-1}\oplus aK_{i-1}. L i − 1 = K i − 1 ⊕ a K i − 1 .
Using the induction hypothesis
U i − 1 = P i − 1 ⊕ a P i − 1 , U_{i-1}=P_{i-1}\oplus aP_{i-1}, U i − 1 = P i − 1 ⊕ a P i − 1 ,
we get
U i = P i ⊕ a P i , U_i=P_i\oplus aP_i, U i = P i ⊕ a P i ,
where
P i = P i − 1 + K i − 1 θ i + ⋯ + K i − 1 θ i r i ⊆ K i . P_i=P_{i-1}+K_{i-1}\theta_i+\cdots+K_{i-1}\theta_i^{r_i}\subseteq K_i. P i = P i − 1 + K i − 1 θ i + ⋯ + K i − 1 θ i r i ⊆ K i .
Thus
dim F q P i = dim F q P i − 1 + r i t i − 1 = t i − 1 + 1 2 + s i − 1 2 t i − 1 = t i + 1 2 . \dim_{\mathbb{F}_q}P_i=\dim_{\mathbb{F}_q}P_{i-1}+r_it_{i-1}
=\frac{t_{i-1}+1}{2}+\frac{s_i-1}{2}t_{i-1}
=\frac{t_i+1}{2}. dim F q P i = dim F q P i − 1 + r i t i − 1 = 2 t i − 1 + 1 + 2 s i − 1 t i − 1 = 2 t i + 1 .
Similarly,
W i = w i ( L i − 1 + L i − 1 θ i + ⋯ + L i − 1 θ i r i − 1 + W i − 1 η i θ i r i ) . W_i=w_i\left(L_{i-1}+L_{i-1}\theta_i+\cdots+L_{i-1}\theta_i^{r_i-1}+\frac{W_{i-1}}{\eta_i}\theta_i^{r_i}\right). W i = w i ( L i − 1 + L i − 1 θ i + ⋯ + L i − 1 θ i r i − 1 + η i W i − 1 θ i r i ) .
By the induction hypothesis,
W i − 1 = d i − 1 ( Q i − 1 ⊕ c Q i − 1 ) . W_{i-1}=d_{i-1}(Q_{i-1}\oplus cQ_{i-1}). W i − 1 = d i − 1 ( Q i − 1 ⊕ c Q i − 1 ) .
Set
d i = w i η i − 1 d i − 1 . d_i=w_i\eta_i^{-1}d_{i-1}. d i = w i η i − 1 d i − 1 .
Since c ∉ K i − 1 c\notin K_{i-1} c ∈ / K i − 1 , we have
L i − 1 = K i − 1 ⊕ c K i − 1 . L_{i-1}=K_{i-1}\oplus cK_{i-1}. L i − 1 = K i − 1 ⊕ c K i − 1 .
Therefore
W i = d i ( Q i ⊕ c Q i ) , W_i=d_i(Q_i\oplus cQ_i), W i = d i ( Q i ⊕ c Q i ) ,
where
Q i = K i − 1 + K i − 1 θ i + ⋯ + K i − 1 θ i r i − 1 + Q i − 1 θ i r i ⊆ K i . Q_i=K_{i-1}+K_{i-1}\theta_i+\cdots+K_{i-1}\theta_i^{r_i-1}+Q_{i-1}\theta_i^{r_i}\subseteq K_i. Q i = K i − 1 + K i − 1 θ i + ⋯ + K i − 1 θ i r i − 1 + Q i − 1 θ i r i ⊆ K i .
Hence
dim F q Q i = r i t i − 1 + dim F q Q i − 1 = s i − 1 2 t i − 1 + t i − 1 − 1 2 = t i − 1 2 . \dim_{\mathbb{F}_q} Q_i = r_i t_{i-1} + \dim_{\mathbb{F}_q} Q_{i-1} = \frac{s_i-1}{2}t_{i-1}+\frac{t_{i-1}-1}{2}=\frac{t_i-1}{2}. dim F q Q i = r i t i − 1 + dim F q Q i − 1 = 2 s i − 1 t i − 1 + 2 t i − 1 − 1 = 2 t i − 1 .
This completes the induction.
Corollary 6.7. For every i ≥ 1 i \ge1 i ≥ 1 , the line ℓ i \ell_i ℓ i is the unique line of
PG ( L i × L i , K i ) \operatorname{PG}(\mathbb{L}_i \times\mathbb{L}_i,\mathbb{K}_i) PG ( L i × L i , K i ) contained in the linear set L ( C i ) L(C_i) L ( C i ) .
Proof. By Lemma 6.6, we have
dim F q U i = t i + 1. \dim_{\mathbb{F}_q} U_i=t_i+1. dim F q U i = t i + 1. Hence, by (7) , we have
ℓ i ⊆ L ( C i ) . \ell_i \subseteq L(C_i). ℓ i ⊆ L ( C i ) . Moreover, every point of L ( C i ) ∩ ℓ i L(C_i) \cap\ell_i L ( C i ) ∩ ℓ i has weight at most
dim F q P i = t i + 1 2 . \dim_{\mathbb{F}_q} P_i=\frac{t_i+1}{2}. dim F q P i = 2 t i + 1 . Indeed, write
ρ = x + a y , x , y ∈ K i , \rho=x+ay,\qquad x,y\in\mathbb{K}_i, ρ = x + a y , x , y ∈ K i , where ρ ≠ 0 \rho\ne0 ρ = 0 . Since U i = P i ⊕ a P i U_i=P_i\oplus aP_i U i = P i ⊕ a P i , we have
K i ρ ∩ U i = { z ρ : z ∈ K i , z x ∈ P i , z y ∈ P i } . \mathbb{K}_i\rho\cap U_i=\{z\rho:z\in\mathbb{K}_i,\ zx\in P_i,\ zy\in P_i\}. K i ρ ∩ U i = { z ρ : z ∈ K i , z x ∈ P i , z y ∈ P i } . If x ≠ 0 x\ne0 x = 0 , the map
K i ρ ∩ U i ⟶ P i , z ρ ⟼ z x , \mathbb{K}_i\rho\cap U_i\longrightarrow P_i,\qquad z\rho\longmapsto zx, K i ρ ∩ U i ⟶ P i , z ρ ⟼ z x , is F q \mathbb{F}_q F q -linear and injective. If x = 0 x=0 x = 0 , then y ≠ 0 y\ne0 y = 0 , and the same conclusion holds using z ρ ⟼ z y z\rho\longmapsto zy z ρ ⟼ z y instead. Therefore
dim F q ( K i ρ ∩ U i ) ≤ dim F q P i = t i + 1 2 . \dim_{\mathbb{F}_q}(\mathbb{K}_i\rho\cap U_i)\le\dim_{\mathbb{F}_q}P_i=\frac{t_i+1}{2}. dim F q ( K i ρ ∩ U i ) ≤ dim F q P i = 2 t i + 1 . The left-hand side is precisely the weight of ⟨ ( ρ , 0 ) ⟩ K i \langle(\rho,0)\rangle_{\mathbb{K}_i} ⟨( ρ , 0 ) ⟩ K i in L ( C i ) L(C_i) L ( C i ) , which proves the claimed bound.
Similarly,
dim F q W i = t i − 1 , \dim_{\mathbb{F}_q}W_i=t_i-1, dim F q W i = t i − 1 , so ℓ i ⊥ \ell_i^\perp ℓ i ⊥ is not contained in L ( C i ) L(C_i) L ( C i ) , and every point of L ( C i ) ∩ ℓ i ⊥ L(C_i)\cap\ell_i^\perp L ( C i ) ∩ ℓ i ⊥ has weight at most
dim F q Q i = t i − 1 2 . \dim_{\mathbb{F}_q}Q_i=\frac{t_i-1}{2}. dim F q Q i = 2 t i − 1 . The same dimension argument used in Proposition 6.3 now applies verbatim and shows that no line other than ℓ i \ell_i ℓ i can be contained in L ( C i ) L(C_i) L ( C i ) .
We can now repeat the root-obstruction argument at an arbitrary step of the iteration.
Theorem 6.8. Fix i ≥ 1 i \ge1 i ≥ 1 , and fix the spread set C i − 1 \mathcal{C}_{i-1} C i − 1 obtained after the first i − 1 i-1 i − 1 steps. Let
s i = 2 r i + 1 ≥ 3 s_i = 2r_i + 1 \ge3 s i = 2 r i + 1 ≥ 3 be odd. Let
G i ( Y ) , H i ( Y ) ∈ K i − 1 [ Y ] G_i(Y), H_i(Y) \in\mathbb{K}_{i-1}[Y] G i ( Y ) , H i ( Y ) ∈ K i − 1 [ Y ] be monic irreducible polynomials of degree s i s_i s i . Let θ i \theta_i θ i be a root of G i G_i G i , and let θ i ′ \theta_i^{\prime} θ i ′ be a root of H i H_i H i . Let
C i ( θ i ) and C i ( θ i ′ ) \mathcal{C}_i(\theta_i) \qquad\text{and} \qquad\mathcal{C}_i(\theta_i^{\prime}) C i ( θ i ) and C i ( θ i ′ ) be the spread sets obtained from C i − 1 \mathcal{C}_{i-1} C i − 1 by the i i i -th lifting step using θ i \theta_i θ i and θ i ′ \theta_i^{\prime} θ i ′ , respectively. If the corresponding semifields are isotopic, then there exist
κ ∈ K i − 1 × \kappa\in\mathbb{K}_{i-1}^{\times} κ ∈ K i − 1 × and
ρ ∈ Aut ( K i ) \rho\in\operatorname{Aut}(\mathbb{K}_i) ρ ∈ Aut ( K i ) such that
θ i ′ = κ θ i ρ . \theta_i^{\prime} = \kappa\theta_i^{\rho}. θ i ′ = κ θ i ρ .
Proof. The proof is the same as that of Theorem 6.4, with
K 0 , L 0 , K 1 , L 1 \mathbb{K}_0,\mathbb{L}_0,\mathbb{K}_1,\mathbb{L}_1 K 0 , L 0 , K 1 , L 1
replaced by
K i − 1 , L i − 1 , K i , L i . \mathbb{K}_{i-1},\mathbb{L}_{i-1},\mathbb{K}_i,\mathbb{L}_i. K i − 1 , L i − 1 , K i , L i .
We next show that the liftings of non-isotopic semifields remain non-isotopic.
Theorem 6.9. Fix i ≥ 1 i \ge1 i ≥ 1 , and put K = K i − 1 \mathbb{K} = \mathbb{K}_{i-1} K = K i − 1 , L = L i − 1 \mathbb{L} = \mathbb{L}_{i-1} L = L i − 1 , K ′ = K i \mathbb{K}^{\prime} = \mathbb{K}_i K ′ = K i , L ′ = L i \mathbb{L}^{\prime} = \mathbb{L}_i L ′ = L i . Let further s = s i = 2 r + 1 ≥ 3 s = s_i = 2r+1 \ge3 s = s i = 2 r + 1 ≥ 3 , and C , C ′ ⊆ End K ( L ) \mathcal{C},\mathcal{C}^{\prime} \subseteq\operatorname{End}_{\mathbb{K}}(\mathbb{L}) C , C ′ ⊆ End K ( L ) be two spread sets occurring after i − 1 i-1 i − 1 steps of the construction, possibly arising from different choices at the preceding steps. Choose monic irreducible polynomials G ( Y ) , H ( Y ) ∈ K [ Y ] G(Y), H(Y) \in\mathbb{K}[Y] G ( Y ) , H ( Y ) ∈ K [ Y ] of degree s s s , roots θ , θ ′ ∈ K ′ \theta,\theta^{\prime} \in\mathbb{K}^{\prime} θ , θ ′ ∈ K ′ , and admissible elements η , η ′ ∈ L × \eta,\eta^{\prime} \in\mathbb{L}^{\times} η , η ′ ∈ L × , w , w ′ ∈ L ′ × w,w^{\prime} \in\mathbb{L}^{\prime\times} w , w ′ ∈ L ′ × , so that N L / K ( η ) = ( − 1 ) s G ( 0 ) N_{\mathbb{L}/\mathbb{K}}(\eta) = (-1)^sG(0) N L / K ( η ) = ( − 1 ) s G ( 0 ) , N L / K ( η ′ ) = ( − 1 ) s H ( 0 ) N_{\mathbb{L}/\mathbb{K}}(\eta^{\prime}) = (-1)^sH(0) N L / K ( η ′ ) = ( − 1 ) s H ( 0 ) , N L ′ / K ′ ( w ) = θ N_{\mathbb{L}^{\prime}/\mathbb{K}^{\prime}}(w) = \theta N L ′ / K ′ ( w ) = θ , N L ′ / K ′ ( w ′ ) = θ ′ N_{\mathbb{L}^{\prime}/\mathbb{K}^{\prime}}(w^{\prime}) = \theta^{\prime} N L ′ / K ′ ( w ′ ) = θ ′ .
Let C ~ θ \widetilde{\mathcal{C}}_{\theta} C θ and C ~ θ ′ ′ \widetilde{\mathcal{C}}^{\prime}_{\theta^{\prime}} C θ ′ ′ be the corresponding lifted spread sets. If the semifields defined by these two lifted spread sets are isotopic, then the semifields defined by C \mathcal{C} C and C ′ \mathcal{C}^{\prime} C ′ are isotopic.
Proof. We identify the map
x ↦ A x + B x q t i − 1 x \mapsto Ax + Bx^{q^{t_{i-1}}} x ↦ A x + B x q t i − 1 with the pair ( A , B ) ∈ L × L (A,B) \in\mathbb{L} \times\mathbb{L} ( A , B ) ∈ L × L . In this notation the first lifted spread set has the form
C ~ θ = { ( A + ∑ j = 1 r a j θ j , ( ∑ j = 0 r − 1 b j θ j + B η θ r ) w ) : ( A , B ) ∈ C , a j , b j ∈ L } . \widetilde{\mathcal{C}}_{\theta}
=
\left\{
\left(
A+\sum_{j=1}^{r}a_j\theta^j,
\left(\sum_{j=0}^{r-1}b_j\theta^j+\frac{B}{\eta}\theta^r\right)w
\right)
:
(A,B)\in\mathcal{C},\ a_j,b_j\in\mathbb{L}
\right\}. C θ = { ( A + j = 1 ∑ r a j θ j , ( j = 0 ∑ r − 1 b j θ j + η B θ r ) w ) : ( A , B ) ∈ C , a j , b j ∈ L } . and we can find a similar representation for C ~ θ ′ ′ \widetilde{\mathcal{C}}^{\prime}_{\theta^{\prime}} C θ ′ ′ .
Suppose that the lifted semifields are isotopic. By Corollary 6.7, the first coordinate line is the unique line contained in each of the associated linear sets. Hence the semilinear collineation induced by the isotopy fixes this line. Using the spread-set form of isotopy
from Proposition 2.1 , and absorbing the non-trivial automorphism of L ′ / K ′ \mathbb{L}'/\mathbb{K}' L ′ / K ′ into the field automorphism if necessary, the action on coefficient pairs is then of the form
( A , B ) ↦ ( λ A ρ , μ B ρ ) (12) (A,B) \mapsto(\lambda A^\rho,\mu B^\rho) \tag*{(12)} ( A , B ) ↦ ( λ A ρ , μ B ρ ) (12)
for some λ , μ ∈ L × \lambda,\mu\in\mathbb{L}^{\times} λ , μ ∈ L × and ρ ∈ Aut ( L ′ ) \rho\in\operatorname{Aut}(\mathbb{L}') ρ ∈ Aut ( L ′ ) with
N L ′ / K ′ ( λ ) = N L ′ / K ′ ( μ ) . (13) N_{\mathbb{L}'/\mathbb{K}'}(\lambda)=N_{\mathbb{L}'/\mathbb{K}'}(\mu). \tag*{(13)} N L ′ / K ′ ( λ ) = N L ′ / K ′ ( μ ) . (13)
For convenience, put
τ = ρ ∣ K , x = θ ρ , y = θ ′ . \tau=\rho|_{\mathbb{K}},\ x=\theta^\rho,\ y=\theta'. τ = ρ ∣ K , x = θ ρ , y = θ ′ .
The argument in the proof of Theorem 6.4 applies also when the two original spread sets are different. Indeed, by Lemma 6.6 , the first coordinate section of each original spread set has F q \mathbb{F}_q F q -dimension t i − 1 + 1 < 2 t i − 1 t_{i-1}+1<2t_{i-1} t i − 1 + 1 < 2 t i − 1 , whereas the second coordinate section has dimension t i − 1 − 1 < 2 t i − 1 t_{i-1}-1<2t_{i-1} t i − 1 − 1 < 2 t i − 1 . These are the only properties of the spread set used to identify the largest L \mathbb{L} L -subspaces in the two lifted coordinate sections. Applying Lemma 6.1 to their L \mathbb{L} L -spans, and then using both coordinate sections as in the proof of Theorem 6.4 , gives
y = κ x (14) y=\kappa x \tag*{(14)} y = κ x (14)
for some
κ ∈ K × . \kappa\in\mathbb{K}^{\times}. κ ∈ K × .
For z ∈ { x , y } z\in\{x,y\} z ∈ { x , y } , put
S z = L + L z + ⋯ + L z r . S_z=\mathbb{L}+\mathbb{L}z+\cdots+\mathbb{L}z^r. S z = L + L z + ⋯ + L z r .
The equality of the L \mathbb{L} L -spans of the first coordinate sections gives
λ S x = S y . \lambda S_x=S_y. λ S x = S y .
Since y = κ x y=\kappa x y = κ x with κ ∈ K × \kappa\in\mathbb{K}^{\times} κ ∈ K × , we have S y = S x S_y=S_x S y = S x . The stabilizer computation used in the proof of Theorem 6.4 therefore yields that λ ∈ L × \lambda\in\mathbb{L}^{\times} λ ∈ L × .
Similarly, after removing the factors w ρ w^\rho w ρ and w ′ w' w ′ from the second coordinate sections, put
ν = w ρ w ′ . \nu=\frac{w^\rho}{w'}. ν = w ′ w ρ .
Their L \mathbb{L} L -spans then satisfy ν S x = S y = S x \nu S_x=S_y=S_x ν S x = S y = S x , and we conclude that ν ∈ L × \nu\in\mathbb{L}^{\times} ν ∈ L × .
We now pass from the lifted spread sets to the original spread sets. Since y = κ x y=\kappa x y = κ x , the free-coefficient subspaces in the two lifts agree after the transformation in Eq. (12) . Taking the quotient by these free coefficients, or equivalently comparing the constant coefficient in the first coordinate and the coefficient of x r x^r x r in the second coordinate, gives
C ′ = { ( λ A ρ , β B ρ ) : ( A , B ) ∈ C } , (15) \mathcal{C}'=\{(\lambda A^\rho,\beta B^\rho):(A,B)\in\mathcal{C}\}, \tag*{(15)} C ′ = {( λ A ρ , β B ρ ) : ( A , B ) ∈ C } , (15)
where
β = ν η ′ η ρ κ − r ∈ L × . (16) \beta=\nu\frac{\eta'}{\eta^\rho}\kappa^{-r}\in\mathbb{L}^{\times}. \tag*{(16)} β = ν η ρ η ′ κ − r ∈ L × . (16)
It remains to show that the two coordinate factors in Eq. (15) have the same norm over K \mathbb{K} K . By Eqs. (13) , (14) and the definitions of ν , x , y \nu,x,y ν , x , y , we have
N L / K ( ν ) = N L ′ / K ′ ( ν ) = N L ′ / K ′ ( μ ) x y = N L / K ( λ ) κ . N_{\mathbb{L}/\mathbb{K}}(\nu)=N_{\mathbb{L}'/\mathbb{K}'}(\nu)=N_{\mathbb{L}'/\mathbb{K}'}(\mu)\frac{x}{y}=\frac{N_{\mathbb{L}/\mathbb{K}}(\lambda)}{\kappa}. N L / K ( ν ) = N L ′ / K ′ ( ν ) = N L ′ / K ′ ( μ ) y x = κ N L / K ( λ ) .
Moreover, x x x is a root of G τ G^\tau G τ , while y = κ x y=\kappa x y = κ x is a root of H H H . Since both polynomials are monic and irreducible of degree s s s ,
H ( Y ) = κ s G τ ( Y κ ) . H(Y)=\kappa^s G^\tau\left(\frac{Y}{\kappa}\right). H ( Y ) = κ s G τ ( κ Y ) .
Consequently,
N L / K ( η ′ ) N L / K ( η ρ ) = H ( 0 ) G τ ( 0 ) = κ s . \frac{N_{\mathbb{L}/\mathbb{K}(\eta')}}{N_{\mathbb{L}/\mathbb{K}(\eta^\rho)}}=\frac{H(0)}{G^\tau(0)}=\kappa^s. N L / K ( η ρ ) N L / K ( η ′ ) = G τ ( 0 ) H ( 0 ) = κ s .
Since κ ∈ K \kappa\in\mathbb{K} κ ∈ K and L / K \mathbb{L}/\mathbb{K} L / K is quadratic, we also have
N L / K ( κ − r ) = κ − 2 r . N_{\mathbb{L}/\mathbb{K}}(\kappa^{-r})=\kappa^{-2r}. N L / K ( κ − r ) = κ − 2 r .
Using Eq. (16) and s = 2 r + 1 s=2r+1 s = 2 r + 1 , we obtain
N L / K ( β ) = N L / K ( ν ) N L / K ( η ′ ) N L / K ( η ρ ) N L / K ( κ − r ) = N L / K ( λ ) κ κ s κ − 2 r = N L / K ( λ ) . \begin{aligned}
N_{\mathbb{L}/\mathbb{K}}(\beta)=N_{\mathbb{L}/\mathbb{K}}(\nu)\frac{N_{\mathbb{L}/\mathbb{K}(\eta')}}{N_{\mathbb{L}/\mathbb{K}(\eta^\rho)}}N_{\mathbb{L}/\mathbb{K}}(\kappa^{-r}) \\
&=\frac{N_{\mathbb{L}/\mathbb{K}}(\lambda)}{\kappa}\kappa^s\kappa^{-2r}=N_{\mathbb{L}/\mathbb{K}}(\lambda).
\end{aligned} N L / K ( β ) = N L / K ( ν ) N L / K ( η ρ ) N L / K ( η ′ ) N L / K ( κ − r ) = κ N L / K ( λ ) κ s κ − 2 r = N L / K ( λ ) .
Thus N L / K ( β / λ ) = 1 N_{\mathbb{L}/\mathbb{K}}(\beta/\lambda)=1 N L / K ( β / λ ) = 1 . Let γ \gamma γ be the non-trivial automorphism of L / K \mathbb{L}/\mathbb{K} L / K . So there exists d ∈ L × d\in\mathbb{L}^{\times} d ∈ L × such that β / λ = d γ − 1 \beta/\lambda=d^{\gamma-1} β / λ = d γ − 1 . Put c = λ / d c=\lambda/d c = λ / d . This means λ = c d \lambda=cd λ = c d , β = c d γ \beta=cd^\gamma β = c d γ .
If
T A , B : z ↦ A z + B z γ , T_{A,B}:z\mapsto Az+Bz^\gamma, T A , B : z ↦ A z + B z γ ,
and M c , M d M_c,M_d M c , M d denote scalar multiplication by c , d c,d c , d , respectively, then
M c ∘ T A ρ , B ρ ∘ M d = T λ A ρ , β B ρ . M_c\circ T_{A\rho,B\rho}\circ M_d=T_{\lambda A\rho,\beta B\rho}. M c ∘ T A ρ , B ρ ∘ M d = T λ A ρ , β B ρ .
Now, it follows from Eq. (15) and Proposition 2.1 that C \mathcal{C} C and C ′ \mathcal{C}' C ′ define isotopic semifields. □
In particular, Theorem 6.9 shows that lifts of non-isotopic semifields are non-isotopic.
We now give a concrete consequence of the previous criterion. The idea is that, at each level of the iteration, once the previous semifield has been fixed, the polynomial used in the quotient construction can be varied in many essentially different ways.
Let
Q i − 1 = ∣ K i − 1 ∣ = q t i − 1 . Q_{i-1}=\lvert\mathbb{K}_{i-1}\rvert=q^{t_{i-1}}. Q i − 1 = ∣ K i − 1 ∣ = q t i − 1 .
For an odd integer s i ≥ 3 s_i\geq3 s i ≥ 3 , denote by
I Q i − 1 ( s i ) = 1 s i ∑ d ∣ s i μ ( d ) Q i − 1 s i / d I_{Q_{i-1}}(s_i)=\frac{1}{s_i}\sum_{d\mid s_i}\mu(d)Q_{i-1}^{s_i/d} I Q i − 1 ( s i ) = s i 1 d ∣ s i ∑ μ ( d ) Q i − 1 s i / d
the number of monic irreducible polynomials of degree s i s_i s i over K i − 1 \mathbb{K}_{i-1} K i − 1 . By Theorem 6.8 , if two choices
G i ( Y ) , H i ( Y ) ∈ K i − 1 [ Y ] G_i(Y),H_i(Y)\in\mathbb{K}_{i-1}[Y] G i ( Y ) , H i ( Y ) ∈ K i − 1 [ Y ]
give isotopic semifields at the i i i -th step, then their roots satisfy
θ i ′ = κ θ i ρ \theta'_i=\kappa\theta_i^\rho θ i ′ = κ θ i ρ
for some
κ ∈ K i − 1 × \kappa\in\mathbb{K}_{i-1}^{\times} κ ∈ K i − 1 ×
and some
ρ ∈ Aut ( K i ) . \rho\in\operatorname{Aut}(K_i). ρ ∈ Aut ( K i ) .
Therefore, for a fixed polynomial G i G_i G i , the number of polynomials that can possibly give an isotopic semifield is at most
∣ K i − 1 × ∣ ∣ Aut ( K i − 1 ) ∣ = ( Q i − 1 − 1 ) ∣ Aut ( K i − 1 ) ∣ . |\mathbb{K}_{i-1}^{\times}|\,|\operatorname{Aut}(K_{i-1})|=(Q_{i-1}-1)|\operatorname{Aut}(K_{i-1})|. ∣ K i − 1 × ∣ ∣ Aut ( K i − 1 ) ∣ = ( Q i − 1 − 1 ) ∣ Aut ( K i − 1 ) ∣.
Indeed, the automorphisms of K i K_i K i fixing K i − 1 K_{i-1} K i − 1 only replace a root by another conjugate root of the same irreducible polynomial over K i − 1 K_{i-1} K i − 1 , and hence do not produce a new polynomial. Consequently, after fixing the first i − 1 i-1 i − 1 steps, the i i i -th lifting step produces at least
M i : = ⌈ I Q i − 1 ( s i ) ( Q i − 1 − 1 ) ∣ Aut ( K i − 1 ) ∣ ⌉ M_i:=\left\lceil\frac{I_{Q_{i-1}}(s_i)}{(Q_{i-1}-1)|\operatorname{Aut}(K_{i-1})|}\right\rceil M i := ⌈ ( Q i − 1 − 1 ) ∣ Aut ( K i − 1 ) ∣ I Q i − 1 ( s i ) ⌉
pairwise non-isotopic semifields of order
q 2 t i = q 2 t i − 1 s i . q^{2t_i}=q^{2t_{i-1}s_i}. q 2 t i = q 2 t i − 1 s i .
We now specialize this to the simplest iterative choice.
Example 6.10. Assume that
for every i = 1 , … , k i=1,\ldots,k i = 1 , … , k , where e e e is a fixed odd prime. Then
t i = e t i − 1 , t_i=et_{i-1}, t i = e t i − 1 , and hence
The semifields obtained after k k k steps have order
N k = q 2 t k = q 2 t e k . N_k=q^{2t_k}=q^{2te^k}. N k = q 2 t k = q 2 t e k . At the i i i -th step we have
Q i − 1 = q t i − 1 = q t e i − 1 . Q_{i-1}=q^{t_{i-1}}=q^{te^{i-1}}. Q i − 1 = q t i − 1 = q t e i − 1 . Moreover, since e e e is prime, the number of monic irreducible polynomials of degree e e e over K i − 1 K_{i-1} K i − 1 is
I Q i − 1 ( e ) = Q i − 1 e − Q i − 1 e . I_{Q_{i-1}}(e)=\frac{Q_{i-1}^{e}-Q_{i-1}}{e}. I Q i − 1 ( e ) = e Q i − 1 e − Q i − 1 . By Theorem 6.8 , two choices θ i , θ i ′ \theta_i,\theta_i' θ i , θ i ′ can give rise to isotopic semifields only if
θ i ′ = κ θ i ρ , κ ∈ K i − 1 × , ρ ∈ Aut ( K i ) . \theta_i'=\kappa\theta_i^\rho,\qquad\kappa\in\mathbb{K}_{i-1}^{\times},\qquad\rho\in\operatorname{Aut}(K_i). θ i ′ = κ θ i ρ , κ ∈ K i − 1 × , ρ ∈ Aut ( K i ) . Therefore the number of pairwise non-isotopic choices at the i i i -th step is at least
M i = ⌈ Q i − 1 e − Q i − 1 e ( Q i − 1 − 1 ) ∣ Aut ( K i − 1 ) ∣ ⌉ . M_i=\left\lceil\frac{Q_{i-1}^{e}-Q_{i-1}}{e(Q_{i-1}-1)|\operatorname{Aut}(K_{i-1})|}\right\rceil. M i = ⌈ e ( Q i − 1 − 1 ) ∣ Aut ( K i − 1 ) ∣ Q i − 1 e − Q i − 1 ⌉ . Write q = p f q=p^f q = p f , where p p p is prime. Since
∣ Aut ( K i − 1 ) ∣ = f t e i − 1 , |\operatorname{Aut}(K_{i-1})|=fte^{i-1}, ∣ Aut ( K i − 1 ) ∣ = f t e i − 1 , we have
M i = ⌈ Q i − 1 + Q i − 1 2 + ⋯ + Q i − 1 e − 1 f t e i ⌉ ≥ q t e i − 1 ( e − 1 ) f t e i . M_i=\left\lceil\frac{Q_{i-1}+Q_{i-1}^{2}+\cdots+Q_{i-1}^{e-1}}{fte^i}\right\rceil\geq\frac{q^{te^{i-1}(e-1)}}{fte^i}. M i = ⌈ f t e i Q i − 1 + Q i − 1 2 + ⋯ + Q i − 1 e − 1 ⌉ ≥ f t e i q t e i − 1 ( e − 1 ) .
By Theorem 6.9 , liftings of non-isotopic semifields remain non-isotopic. Hence the total number of pairwise non-isotopic semifields obtained after k k k steps is at least
∏ i = 1 k M i ≥ q t ( e k − 1 ) ( f t ) k e k ( k + 1 ) / 2 . \prod_{i=1}^{k} M_i \ge\frac{q^{t(e^k-1)}}{(ft)^k e^{k(k+1)/2}}. i = 1 ∏ k M i ≥ ( f t ) k e k ( k + 1 ) /2 q t ( e k − 1 ) .
Since
N k = q 2 t e k , N_k=q^{2te^k}, N k = q 2 t e k ,
we get
q t ( e k − 1 ) = N k e k − 1 2 e k = N k 1 2 ( 1 − e − k ) . q^{t(e^k-1)}=N_k^{\frac{e^k-1}{2e^k}}=N_k^{\frac{1}{2}(1-e^{-k})}. q t ( e k − 1 ) = N k 2 e k e k − 1 = N k 2 1 ( 1 − e − k ) .
Thus, for fixed e , t , k e,t,k e , t , k , as q → + ∞ q\to+\infty q → + ∞ , the construction gives at least
N k 1 2 ( 1 − e − k ) + o ( 1 ) N_k^{\frac{1}{2}(1-e^{-k})+o(1)} N k 2 1 ( 1 − e − k ) + o ( 1 )
pairwise non-isotopic semifields of order N k N_k N k , which are two-dimensional over their right nucleus. Finally, letting k → + ∞ k\to+\infty k → + ∞ , the exponent
1 2 ( 1 − e − k ) \frac{1}{2}(1-e^{-k}) 2 1 ( 1 − e − k )
tends to 1 / 2 1/2 1/2 .
Newness of the construction
We now compare the semifields in Theorem 5.2 with the known constructions. Our aim is to show that the family contains semifields which are not isotopic to any previously known example. We retain the notation of § 5 . The orders of the nuclei provide useful invariants for distinguishing semifields up to isotopy, although they will not suffice for all comparisons below. We also recall the Knuth orbit [17 ] . Choose a basis { e 1 , … , e n } \{e_1,\ldots,e_n\} { e 1 , … , e n } of S S S over its center and write
e i ⋆ e j = ∑ k = 1 n c i j k e k . e_i\star e_j=\sum_{k=1}^{n}c_{ijk}e_k. e i ⋆ e j = k = 1 ∑ n c ij k e k .
Permuting the three indices of the structure constants c i j k c_{ijk} c ij k gives presemifields S π S^\pi S π , with π ∈ S 3 \pi\in S_3 π ∈ S 3 , called Knuth derivatives of S S S . Their isotopy classes form the Knuth orbit
K ( S ) = { [ S π ] : π ∈ S 3 } , \mathcal{K}(S)=\{[S^\pi]:\pi\in S_3\}, K ( S ) = {[ S π ] : π ∈ S 3 } ,
which contains at most six classes; here [ T ] [T] [ T ] denotes the isotopy class of T T T [19 Section 2.1] . Knuth operations preserve the order and the center up to isomorphism, while permuting the three nuclei up to isomorphism, computed in semifield representatives of these classes. For example, the opposite multiplication x ⋆ d y = y ⋆ x x\star^d y=y\star x x ⋆ d y = y ⋆ x interchanges the left and right nuclei and preserves the middle nucleus [25 Proposition 2.4] . Consequently, the multiset
{ ∣ N l ( S ) ∣ , ∣ N m ( S ) ∣ , ∣ N r ( S ) ∣ } \{|N_l(S)|,|N_m(S)|,|N_r(S)|\} { ∣ N l ( S ) ∣ , ∣ N m ( S ) ∣ , ∣ N r ( S ) ∣ }
is invariant throughout the Knuth orbit. Distinct such multisets therefore exclude isotopy even after applying any Knuth operation.
The comparison with EMPT uses the coordinate sections and the unique-line result established in § 6 .
Comparison with EMPT semifields
Proposition 7.1. Let h ≥ 1 h \ge1 h ≥ 1 . No semifield S h \mathbb{S}_h S h in Theorem 5.2 is isotopic to an admissible instance of either construction (3) or (9) of [7 ] , or to a semifield in their Knuth orbit.
Proof. Put m = t h m=t_h m = t h , d = t h − 1 d=t_{h-1} d = t h − 1 and s = s h s=s_h s = s h , so that m = d s m=ds m = d s and d , s ≥ 3 d,s\ge3 d , s ≥ 3 are odd. Set E = F q d ⊆ K = F q m \mathbb{E}=\mathbb{F}_{q^d}\subseteq\mathbb{K}=\mathbb{F}_{q^m} E = F q d ⊆ K = F q m and L = F q 2 m \mathbb{L}=\mathbb{F}_{q^{2m}} L = F q 2 m . We represent the spreads over their large right nuclei, as in § 5 .
We first record an elementary observation. Suppose that K = F q ( v ) \mathbb{K}=\mathbb{F}_q(v) K = F q ( v ) and put
P k ( v ) = ⟨ 1 , v , … , v k − 1 ⟩ F q . P_k(v)=\langle1,v,\ldots,v^{k-1}\rangle_{\mathbb{F}_q}. P k ( v ) = ⟨ 1 , v , … , v k − 1 ⟩ F q .
If k + s ≤ m k+s\le m k + s ≤ m , then x E ⊆ P k ( v ) x\mathbb{E}\subseteq P_k(v) x E ⊆ P k ( v ) implies x = 0 x=0 x = 0 . Indeed, let π : K → F q \pi:\mathbb{K}\to\mathbb{F}_q π : K → F q extract the coefficient of v m − 1 v^{m-1} v m − 1 in the power basis. For c ∈ E c\in\mathbb{E} c ∈ E and 0 ≤ j < s 0\le j<s 0 ≤ j < s , the inclusion x E ⊆ P k ( v ) x\mathbb{E}\subseteq P_k(v) x E ⊆ P k ( v ) gives
x c v j ∈ ⟨ 1 , v , … , v k + s − 2 ⟩ F q , π ( x c v j ) = 0. xcv^j\in\langle1,v,\ldots,v^{k+s-2}\rangle_{\mathbb{F}_q},\qquad\pi(xcv^j)=0. x c v j ∈ ⟨ 1 , v , … , v k + s − 2 ⟩ F q , π ( x c v j ) = 0.
Since 1 , v , … , v s − 1 1,v,\ldots,v^{s-1} 1 , v , … , v s − 1 form an E \mathbb{E} E -basis of K \mathbb{K} K , it follows that π ( x K ) = 0 \pi(x\mathbb{K})=0 π ( x K ) = 0 . This is impossible for x ≠ 0 x\ne0 x = 0 . Consequently, if a ∈ L ∖ K a\in\mathbb{L}\setminus\mathbb{K} a ∈ L ∖ K , the space P k ( v ) ⊕ a P k ( v ) P_k(v)\oplus aP_k(v) P k ( v ) ⊕ a P k ( v ) contains no nonzero one-dimensional E \mathbb{E} E -subspace: writing z = x + a y z=x+ay z = x + a y , the inclusion E z ⊆ P k ( v ) ⊕ a P k ( v ) \mathbb{E}z\subseteq P_k(v)\oplus aP_k(v) E z ⊆ P k ( v ) ⊕ a P k ( v ) would imply E x , E y ⊆ P k ( v ) \mathbb{E}x,\mathbb{E}y\subseteq P_k(v) E x , E y ⊆ P k ( v ) .
For an EMPT semifield of Type (3) isotopic to S h \mathbb{S}_h S h , invariance of the order and center forces its defining base field to be F q \mathbb{F}_q F q and its extension parameter to be m m m ; see [7 Proposition 3.1] . Its defining section on the first coordinate line is
U E M P T = P ⊕ a P , P = ⟨ 1 , v , … , v ( m − 1 ) / 2 ⟩ F q . U_{\mathrm{EMPT}}=P\oplus aP,\qquad P=\langle1,v,\ldots,v^{(m-1)/2}\rangle_{\mathbb{F}_q}. U EMPT = P ⊕ a P , P = ⟨ 1 , v , … , v ( m − 1 ) /2 ⟩ F q .
where K = F q ( v ) \mathbb{K}=\mathbb{F}_q(v) K = F q ( v ) and a ∉ K a\notin\mathbb{K} a ∈ / K . With k = ( m + 1 ) / 2 k=(m+1)/2 k = ( m + 1 ) /2 , we have
k + s = d s + 1 2 + s ≤ d s = m . k+s=\frac{ds+1}{2}+s\le ds=m. k + s = 2 d s + 1 + s ≤ d s = m .
Thus U E M P T U_{\mathrm{EMPT}} U EMPT contains no nonzero one-dimensional E \mathbb{E} E -subspace. In contrast, Lemma 6.6 gives
E θ h ⊆ L h − 1 θ h ⊆ U h \mathbb{E}\theta_h\subseteq\mathbb{L}_{h-1}\theta_h\subseteq U_h E θ h ⊆ L h − 1 θ h ⊆ U h
for the first coordinate section of the lifted spread.
Suppose that these two semifields are isotopic. Their exact right nuclei are then both K \mathbb{K} K . By Proposition 2.1 , the isotopy gives a K \mathbb{K} K -semilinear map
T : End K ( L ) ⟶ End K ( L ) , M ⟼ A M ρ B , T:\operatorname{End}_{\mathbb{K}}(\mathbb{L})\longrightarrow\operatorname{End}_{\mathbb{K}}(\mathbb{L}),\qquad M\longmapsto AM^\rho B, T : End K ( L ) ⟶ End K ( L ) , M ⟼ A M ρ B ,
which maps one defining spread subspace exactly onto the other. By Corollary 6.7 and [7 Proposition 3.2(iii)] , the first coordinate line is the unique projective line contained in each associated linear set. Hence T T T maps their defining sections on these lines onto each other. Every automorphism of K \mathbb{K} K stabilizes its unique subfield E \mathbb{E} E , so this semilinear bijection preserves the existence of a nonzero one-dimensional E \mathbb{E} E -subspace. This contradicts the preceding paragraph.
The EMPT construction (9) has order r 2 n r^{2n} r 2 n and center F r \mathbb{F}_r F r , where n n n is even; see the paragraph preceding Proposition 3.5 of [7 ] . Isotopy would force r = q r=q r = q and n = m n=m n = m , contrary to the oddness of m m m .
The nuclear multiset of S h S_h S h has a unique large member. After placing it on the right, the only Knuth possibilities are represented by a spread and its adjoint. For the trace pairing Tr L / K ( x y ) \operatorname{Tr}_{L/K}(xy) Tr L / K ( x y ) , the adjoint of φ A , B ( x ) = A x + B x q m \varphi_{A,B}(x)=Ax+Bx^{q^m} φ A , B ( x ) = A x + B x q m is
φ A , B ∗ ( x ) = A x + B q m x q m . \varphi^*_{A,B}(x)=Ax+B^{q^m}x^{q^m}. φ A , B ∗ ( x ) = A x + B q m x q m .
Thus the first coordinate section is unchanged, and the same obstruction applies.
We finally check the translation dual. Put r = ( m − 1 ) / 2 r=(m-1)/2 r = ( m − 1 ) /2 and R = ⟨ 1 , v , … , v r − 1 ⟩ F q R=\langle1,v,\ldots,v^{r-1}\rangle_{\mathbb{F}_q} R = ⟨ 1 , v , … , v r − 1 ⟩ F q , so that P = ⟨ 1 , v , … , v r ⟩ F q P=\langle1,v,\ldots,v^r\rangle_{\mathbb{F}_q} P = ⟨ 1 , v , … , v r ⟩ F q . If f f f is the minimal polynomial of v v v over F q \mathbb{F}_q F q , then orthogonality with respect to Tr K / F q ( x y ) \operatorname{Tr}_{K/\mathbb{F}_q}(xy) Tr K / F q ( x y ) gives
R ⊥ = f ′ ( v ) − 1 P , P ⊥ = f ′ ( v ) − 1 R . R^\perp=f'(v)^{-1}P,\qquad P^\perp=f'(v)^{-1}R. R ⊥ = f ′ ( v ) − 1 P , P ⊥ = f ′ ( v ) − 1 R .
These identities follow from Tr K / F q ( v j / f ′ ( v ) ) = 0 \operatorname{Tr}_{K/\mathbb{F}_q}(v^j/f'(v))=0 Tr K / F q ( v j / f ′ ( v )) = 0 for 0 ≤ j ≤ m − 2 0\leq j\leq m-2 0 ≤ j ≤ m − 2 and comparison of dimensions. The coefficient space of Type (3) is C = ( P ⊕ a P ) × b ( R ⊕ c R ) C=(P\oplus aP)\times b(R\oplus cR) C = ( P ⊕ a P ) × b ( R ⊕ c R ) . The translation dual is obtained by taking its orthogonal complement under the polar form of the determinant, followed by the trace to F q \mathbb{F}_q F q ; see [21 ] . In these coordinates this form is
β ( ( A , B ) , ( A ′ , B ′ ) ) = Tr K / F q ( A A ′ q m + A q m A ′ − B B ′ q m − B q m B ′ ) . \beta((A,B),(A',B'))=\operatorname{Tr}_{K/\mathbb{F}_q}\left(AA'^{q^m}+A^{q^m}A'-BB'^{q^m}-B^{q^m}B'\right). β (( A , B ) , ( A ′ , B ′ )) = Tr K / F q ( A A ′ q m + A q m A ′ − B B ′ q m − B q m B ′ ) .
The two coordinate spaces are orthogonal. Using dual K K K -bases of L L L and the preceding trace identities, their complements have the forms
( P ⊕ a P ) ⊥ = μ ( R ⊕ c ′ R ) , ( b ( R ⊕ c R ) ) ⊥ = λ ( P ⊕ a ′ P ) , (P\oplus aP)^\perp=\mu(R\oplus c'R),\qquad(b(R\oplus cR))^\perp=\lambda(P\oplus a'P), ( P ⊕ a P ) ⊥ = μ ( R ⊕ c ′ R ) , ( b ( R ⊕ c R ) ) ⊥ = λ ( P ⊕ a ′ P ) ,
for some λ , μ ∈ L × \lambda,\mu\in L^\times λ , μ ∈ L × and a ′ , c ′ ∉ K a',c'\notin K a ′ , c ′ ∈ / K . Swapping the two coordinates and multiplying both by λ − 1 \lambda^{-1} λ − 1 gives another admissible spread of Type (3). These operations are spread equivalences: they correspond to right composition by x ↦ x q m x\mapsto x^{q^m} x ↦ x q m and left multiplication by λ − 1 \lambda^{-1} λ − 1 . Type (3) is therefore closed under translation duality up to isotopy. For Type (9), both power spaces have dimension n / 2 n/2 n /2 , and the same trace identity shows that their orthogonal complements are scalar multiples of themselves. The dual-basis argument therefore also proves closure of Type (9) under translation duality up to isotopy. The preceding exclusions apply to both translation duals and their Knuth derivatives. □
Comparison with cyclic and twisted cyclic semifields
We now compare the semifields in Theorem 5.2 with cyclic semifields and the twisted cyclic semifields of [34 ] , using the invariance of the multiset of nuclear orders under Knuth operations. Put m = t h m=t h m = t h . By Theorem 5.2 , m > 1 m>1 m > 1 is odd and
∣ S h ∣ = q 2 m , ∣ Z ( S h ) ∣ = q , ( ∣ N l ( S h ) ∣ , ∣ N m ( S h ) ∣ , ∣ N r ( S h ) ∣ ) = ( q , q , q m ) . (17) |S_h|=q^{2m},\qquad|Z(S_h)|=q,\qquad(|N_l(S_h)|,|N_m(S_h)|,|N_r(S_h)|)=(q,q,q^m). \tag*{(17)} ∣ S h ∣ = q 2 m , ∣ Z ( S h ) ∣ = q , ( ∣ N l ( S h ) ∣ , ∣ N m ( S h ) ∣ , ∣ N r ( S h ) ∣ ) = ( q , q , q m ) . (17)
Proposition 7.2. No semifield S h S_h S h in Theorem 5.2 is isotopic to a cyclic semifield or to a Knuth derivative of a cyclic semifield.
Proof. By [22 Theorem 1] , a cyclic semifield is isotopic to a Petit semifield obtained from an irreducible polynomial of degree d d d in F r n [ X ; σ ] \mathbb{F}_{r^n}[X;\sigma] F r n [ X ; σ ] , where Fix ( σ ) = F r \operatorname{Fix}(\sigma)=\mathbb{F}_r Fix ( σ ) = F r . The cases n = 1 n=1 n = 1 or d = 1 d=1 d = 1 give fields, whose nuclear parameters differ from (17) . For a proper cyclic semifield T \mathbb{T} T , we have n , d > 1 n,d>1 n , d > 1 and
( ∣ Z ( T ) ∣ , ∣ N l ( T ) ∣ , ∣ N m ( T ) ∣ , ∣ N r ( T ) ∣ ) = ( r , r n , r n , r d ) (|Z(\mathbb{T})|,|N_l(\mathbb{T})|,|N_m(\mathbb{T})|,|N_r(\mathbb{T})|)=(r,r^n,r^n,r^d) ( ∣ Z ( T ) ∣ , ∣ N l ( T ) ∣ , ∣ N m ( T ) ∣ , ∣ N r ( T ) ∣ ) = ( r , r n , r n , r d )
by [22 Theorem 9] . Hence all three nuclei have order strictly larger than the center. This remains true after any Knuth operation, whereas two of the nuclei in (17) have the same order as the center.
Proposition 7.3. No semifield S h \mathrm{S}_{h} S h in Theorem 5.2 is isotopic to a twisted cyclic semifield of [34 ] or to a Knuth derivative of such a semifield.
Proof. Write q = p f q=p^{f} q = p f , where p p p is prime. Suppose that S h \mathrm{S}_{h} S h is isotopic to a Knuth derivative of a twisted cyclic semifield T \mathrm{T} T . Then T \mathrm{T} T has order q 2 m q^{2m} q 2 m , center of order q q q , and multiset of nuclear orders { q , q , q m } \{q,q,q^{m}\} { q , q , q m } .
We use the notation of the skew-polynomial construction, keeping its base field distinct from the center of T \mathrm{T} T . Let r = p e r=p^{e} r = p e , L = F r n L=\mathbb{F}_{r^{n}} L = F r n , and let σ ∈ Aut ( L ) \sigma\in\operatorname{Aut}(L) σ ∈ Aut ( L ) have order n n n and fixed field F r \mathbb{F}_{r} F r . Let F ( Y ) ∈ F r [ Y ] F(Y)\in\mathbb{F}_{r}[Y] F ( Y ) ∈ F r [ Y ] be monic irreducible of degree d d d , with F ( 0 ) ≠ 0 F(0)\ne0 F ( 0 ) = 0 . The spread set is the image of
T = { ∑ j = 0 d − 1 a j X j + η a 0 ρ X d : a 0 , … , a d − 1 ∈ L } . \mathcal{T}=\left\{\sum_{j=0}^{d-1}a_{j}X^{j}+\eta a_{0}^{\rho}X^{d}:a_{0},\ldots,a_{d-1}\in L\right\}. T = { j = 0 ∑ d − 1 a j X j + η a 0 ρ X d : a 0 , … , a d − 1 ∈ L } .
in
A = L [ X ; σ ] / ( F ( X n ) ) ≅ M n ( F r d ) . \mathcal{A}=L[X;\sigma]/(F(X^{n}))\cong M_{n}(\mathbb{F}_{r^{d}}). A = L [ X ; σ ] / ( F ( X n )) ≅ M n ( F r d ) .
where ρ ∈ Aut ( L ) \rho\in\operatorname{Aut}(L) ρ ∈ Aut ( L ) and the parameters satisfy the hypotheses of [34 ] . In particular, ∣ T ∣ = r n d |\mathcal{T}|=r^{nd} ∣ T ∣ = r n d . If η = 0 \eta=0 η = 0 , the construction is cyclic, so Proposition 7.2 applies. If n = 1 n=1 n = 1 , the ambient algebra is a field and the spread set has the same cardinality as this field, so it defines a field isotope. We may therefore assume that η ≠ 0 \eta\ne0 η = 0 and n > 1 n>1 n > 1 .
Suppose first that d ≥ 3 d\ge3 d ≥ 3 , and write ρ : x ↦ x p i \rho:x\mapsto x^{p^{i}} ρ : x ↦ x p i . The nuclear formulas in [34 Section 5.1] give
N l ( T ) ≅ Fix ( ρ ) , N m ( T ) ≅ Fix ( σ d ρ − 1 ) , N r ( T ) ≅ F r d , N_{l}(\mathrm{T})\cong\operatorname{Fix}(\rho),\qquad N_{m}(\mathrm{T})\cong\operatorname{Fix}(\sigma^{d}\rho^{-1}),\qquad N_{r}(\mathrm{T})\cong\mathbb{F}_{r^{d}}, N l ( T ) ≅ Fix ( ρ ) , N m ( T ) ≅ Fix ( σ d ρ − 1 ) , N r ( T ) ≅ F r d ,
and Z ( T ) ≅ Fix ( ρ ∣ F r ) Z(\mathrm{T})\cong\operatorname{Fix}(\rho|_{\mathbb{F}_{r}}) Z ( T ) ≅ Fix ( ρ ∣ F r ) . Thus f = gcd ( e , i ) f=\gcd(e,i) f = g cd( e , i ) and q ≤ r q\le r q ≤ r . Since r d > r ≥ q r^{d}>r\ge q r d > r ≥ q , the right nucleus must have order q m q^{m} q m . Comparing this with ∣ T ∣ = q 2 m |\mathrm{T}|=q^{2m} ∣ T ∣ = q 2 m yields n = 2 n=2 n = 2 and e d = f m ed=fm e d = f m . Consequently σ : x ↦ x p e \sigma:x\mapsto x^{p^{e}} σ : x ↦ x p e , and
∣ N l ( T ) ∣ = p gcd ( 2 e , i ) , ∣ N m ( T ) ∣ = p gcd ( 2 e , e d − i ) . |N_{l}(\mathrm{T})|=p^{\gcd(2e,i)},\qquad|N_{m}(\mathrm{T})|=p^{\gcd(2e,ed-i)}. ∣ N l ( T ) ∣ = p g c d ( 2 e , i ) , ∣ N m ( T ) ∣ = p g c d ( 2 e , e d − i ) .
Write e = f u e=fu e = f u and i = f v i=fv i = f v . Then gcd ( u , v ) = 1 \gcd(u,v)=1 g cd( u , v ) = 1 and u d = m ud=m u d = m , so u u u and d d d are odd. Since the left nucleus must have order q q q , we have gcd ( 2 u , v ) = 1 \gcd(2u,v)=1 g cd( 2 u , v ) = 1 , and hence v v v is odd. It follows that u d − v ud-v u d − v is even and
gcd ( 2 u , u d − v ) = 2 , \gcd(2u,ud-v)=2, g cd( 2 u , u d − v ) = 2 ,
because gcd ( u , u d − v ) = gcd ( u , v ) = 1 \gcd(u,ud-v)=\gcd(u,v)=1 g cd( u , u d − v ) = g cd( u , v ) = 1 . Therefore ∣ N m ( T ) ∣ = q 2 |N_{m}(\mathrm{T})|=q^{2} ∣ N m ( T ) ∣ = q 2 , contrary to the required nuclear orders.
We treat d = 2 d=2 d = 2 directly. Write F ( Y ) = Y 2 + c 1 Y + c 0 F(Y)=Y^{2}+c_{1}Y+c_{0} F ( Y ) = Y 2 + c 1 Y + c 0 . The element X X X is invertible in A \mathcal{A} A , and right multiplication of the spread set by X − 1 X^{-1} X − 1 gives the equivalent spread set
D = { b + c X − 1 + η c ρ X : b , c ∈ L } . \mathcal{D}=\{b+cX^{-1}+\eta c^{\rho}X:b,c\in L\}. D = { b + c X − 1 + η c ρ X : b , c ∈ L } .
It contains the identity and L L L . A central element is represented by an element z = b + c X − 1 + η c ρ X z=b+cX^{-1}+\eta c^{\rho}X z = b + c X − 1 + η c ρ X of D \mathcal{D} D which commutes with every element of D \mathcal{D} D . In A \mathcal{A} A we have
X − 1 = − c 0 − 1 ( X 2 n − 1 + c 1 X n − 1 ) . X^{-1}=-c_{0}^{-1}(X^{2n-1}+c_{1}X^{n-1}). X − 1 = − c 0 − 1 ( X 2 n − 1 + c 1 X n − 1 ) .
For v ∈ L v \in L v ∈ L , the coefficient of X 2 n − 1 X^{2n-1} X 2 n − 1 in z v − v z zv-vz z v − v z is − c 0 − 1 c ( v σ − 1 − v ) -c_{0}^{-1}c(v^{\sigma^{-1}}-v) − c 0 − 1 c ( v σ − 1 − v ) . Since n > 1 n>1 n > 1 , commutation with every v ∈ L v \in L v ∈ L forces c = 0 c=0 c = 0 . Commuting b b b with X − 1 + η X ∈ D X^{-1}+\eta X \in D X − 1 + η X ∈ D then gives b σ = b b^{\sigma}=b b σ = b . Thus the center is represented by a subfield of F r \mathbb{F}_r F r , and hence f ∣ e f \mid e f ∣ e .
On the other hand, the scalar field F r 2 \mathbb{F}_{r^2} F r 2 of M n ( F r 2 ) M_n(\mathbb{F}_{r^2}) M n ( F r 2 ) centralizes D D D and therefore embeds in its right nucleus; see [34 Proposition 5] . The degree of this right nucleus over the center is consequently divisible by 2 e / f 2e/f 2 e / f , and in particular is even. This contradicts the fact that the three nuclear degrees over the center are 1 , 1 , m 1,1,m 1 , 1 , m , all of which are odd.
Finally, when d = 1 d=1 d = 1 , the construction gives a generalized twisted field. In the proper case it is isotopic to the presemifield with multiplication
x ∗ y = x y − η x p i y p j , x , y ∈ F p 2 f m , x * y = xy-\eta x^{p^i}y^{p^j}, \qquad x,y \in\mathbb{F}_{p^{2fm}}, x ∗ y = x y − η x p i y p j , x , y ∈ F p 2 f m ,
where the two Frobenius maps are nonidentity and distinct. If either map is the identity, or if the two maps coincide, the multiplication is isotopic to that of a field. By the nuclear formulas recalled in [34 Section 5.1] , the three nuclear degrees over F p \mathbb{F}_p F p , up to order, are
gcd ( 2 f m , i ) , gcd ( 2 f m , j ) , gcd ( 2 f m , j − i ) . \gcd(2fm,i), \qquad\gcd(2fm,j), \qquad\gcd(2fm,j-i). g cd( 2 f m , i ) , g cd( 2 f m , j ) , g cd( 2 f m , j − i ) .
To obtain the multiset { f , f , f m } \{f,f,fm\} { f , f , f m } , we must have f ∣ i , j f \mid i,j f ∣ i , j . Writing i = f u i=fu i = f u and j = f v j=fv j = f v , the integers
gcd ( 2 m , u ) , gcd ( 2 m , v ) , gcd ( 2 m , v − u ) \gcd(2m,u), \qquad\gcd(2m,v), \qquad\gcd(2m,v-u) g cd( 2 m , u ) , g cd( 2 m , v ) , g cd( 2 m , v − u )
would all be odd. Since 2 m 2m 2 m is even, this would require u u u , v v v , and v − u v-u v − u all to be odd, which is impossible. □
Proof of newness
To prove that our construction gives a new family, it is enough to find one member which is not isotopic to any previously known semifield.
Theorem 7.4. The construction in Theorem 5.2 contains a semifield of order 2 18 2^{18} 2 18 which is not isotopic to any previously known semifield.
Proof. Choose an admissible EMPT seed in F V \mathbb{F}_V F V with q = 2 q=2 q = 2 and t = 3 t=3 t = 3 , which exists by [7 Theorem 4.5] . Apply Theorem 5.1 with s = 3 s=3 s = 3 and any monic irreducible cubic G ( Y ) ∈ F 8 [ Y ] G(Y) \in\mathbb{F}_8[Y] G ( Y ) ∈ F 8 [ Y ] . The resulting semifield S 1 S_1 S 1 satisfies
∣ S 1 ∣ = 2 18 , Z ( S 1 ) = N l ( S 1 ) = N m ( S 1 ) ≅ F 2 , N r ( S 1 ) ≅ F 2 9 . |S_1|=2^{18}, \qquad Z(S_1)=N_l(S_1)=N_m(S_1)\cong\mathbb{F}_2, \qquad N_r(S_1)\cong\mathbb{F}_{2^9}. ∣ S 1 ∣ = 2 18 , Z ( S 1 ) = N l ( S 1 ) = N m ( S 1 ) ≅ F 2 , N r ( S 1 ) ≅ F 2 9 . Any comparison semifield has nuclear multiset { 2 , 2 , 2 9 } \{2,2,2^9\} { 2 , 2 , 2 9 } and hence has dimension 2 2 2 over a nucleus.
Propositions 7.2 , 7.3 and 7.1 exclude cyclic, twisted cyclic and EMPT semifields, together with their Knuth derivatives. The cyclic and twisted cyclic comparisons also cover finite fields, generalized twisted fields, Sandler and Hughes–Kleinfeld semifields, Knuth Types II–IV, Bierbrauer’s construction and its characteristic-two generalization, and Dempwolff’s construction, including its even-characteristic extension; see [18 Sections 4.1 and 5.2–5.4] .
Every HMO semifield has order r 4 r^4 r 4 for a prime power r r r [10 Section 2 and Remark 2.4] , so its order cannot be 2 18 2^{18} 2 18 . A commutative semifield has equal left and right nuclei. Hence a commutative representative in the Knuth orbit would have middle nucleus of order 2 9 2^9 2 9 and left and right nuclei of order 2 2 2 . It would be a proper commutative semifield of characteristic 2 2 2 and dimension 2 2 2 over its middle nucleus, contrary to [3 Theorem 3.1] .
We check the remaining noncommutative families. A Knuth semifield of Type I over K = F 2 9 \mathbb{K} = \mathbb{F}_{2^9} K = F 2 9 has three equal nuclei: a nonidentity defining automorphism has odd order, and all three nuclei equal its fixed field [26 p. 283] . The identity-automorphism case is a field. The Johnson–Marino–Polverino–Trombetti constructions and the related Marino–Polverino–Trombetti and Marino–Trombetti examples have two nuclei of order r 2 r^2 r 2 when the center has order r r r [20 , 26 ] ; they cannot have two nuclei of order 2 2 2 .
Generalized Dickson semifields over K = F 2 9 \mathbb{K} = \mathbb{F}_{2^9} K = F 2 9 require a nonzero parameter which is not a ( σ + 1 ) (\sigma+ 1) ( σ + 1 ) st power, where σ : x ↦ x 2 k \sigma: x \mapsto x^{2^k} σ : x ↦ x 2 k is an automorphism of K \mathbb{K} K [20 Section 3] and [18 Section 5.1] . Kölsch’s explicit family in [18 Section 6] requires α ∉ ( K × ) σ + 1 ( K × ) τ − 1 \alpha\notin(\mathbb{K}^{\times})^{\sigma+1}(\mathbb{K}^{\times})^{\tau-1} α ∈ / ( K × ) σ + 1 ( K × ) τ − 1 for automorphisms σ , τ \sigma,\tau σ , τ ; see [18 Proposition 4.4] . Both conditions are impossible, since
gcd ( 2 k + 1 , 2 9 − 1 ) = 1 ( 0 ≤ k < 9 ) . \gcd(2^k+1,2^9-1)=1 \qquad(0 \leq k < 9). g cd( 2 k + 1 , 2 9 − 1 ) = 1 ( 0 ≤ k < 9 ) .
For a Taniguchi semifield, the nonzero parameter η \eta η is required not to be a ( 2 k − 1 ) (2^k-1) ( 2 k − 1 ) st power in K \mathbb{K} K [9 Section 2] . Consequently, d = gcd ( k , 9 ) > 1 d = \gcd(k,9) > 1 d = g cd( k , 9 ) > 1 . Its multiplication is bilinear over Fix ( σ ) = F 2 d \operatorname{Fix}(\sigma) = \mathbb{F}_{2^d} Fix ( σ ) = F 2 d , so its unital isotope has center containing F 2 d \mathbb{F}_{2^d} F 2 d , which excludes center F 2 \mathbb{F}_2 F 2 . The specialization η = 0 \eta= 0 η = 0 lies in the Knuth orbit of Type I [18 Section 5.6] .
The field-switching construction of [11 ] is bilinear over its base field, which therefore must be F 2 \mathbb{F}_2 F 2 . In this case, Theorem 4.1 of that paper forces the switching map to be scalar, since the dimension is at least 3. The product is then an invertible additive image of the field product and defines a field isotope.
The other characteristic-two rank-two families of order r 6 r^6 r 6 in the classical list, including the extensions of the Huang–Johnson examples, have center containing F r \mathbb{F}_r F r [24 Section 3.3.3] . Order 2 18 2^{18} 2 18 would force r = 8 r = 8 r = 8 . The remaining sporadic orders in that list differ from 2 18 2^{18} 2 18 , and the constructions confined to odd characteristic do not occur here. The new family of Lobillo, Santonastaso and Sheekey in [23 Theorem 6.4] has odd characteristic, and their other finite-field specialization is twisted cyclic. The semifields in [27 Corollary 5.2] have three equal nuclei, while those classified in [8 ] have order 256.
The explicit finite instances in [31 Section 7] have order 81, order r 4 r^4 r 4 , or order r 9 r^9 r 9 with multiplication bilinear over F r \mathbb{F}_r F r . Only the last order can equal 2 18 2^{18} 2 18 , but then r = 4 r = 4 r = 4 and the center contains F 4 \mathbb{F}_4 F 4 . Finally, the sufficient norm condition in [30 Theorem 1] admits no nonzero twisting parameter over F 2 \mathbb{F}_2 F 2 , since both sides of the required inequality are 1. A larger base field is contained in the center of the normalized semifield.
These numerical obstructions persist under Knuth operations and translation duality. In particular, two nuclei of order 2 force the center to have order 2, so the central-subfield comparisons are also nuclear obstructions. The EMPT translation duals are covered by Proposition 7.1. This proves the result. □
Declaration of AI use
The majority of this paper was developed and written without the use of generative AI. Generative AI (ChatGPT 6) was only used for investigation purposes in the chapter studying newness (§ 7 of the paper). Outside of that, we used AI only for finding typos, clearing up notational issues, and minor editorial assistance.
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) and the Natural Science Foundation of Hunan Province (No. 2023RC1003).
Paolo Santonastaso and Yue Zhou are deeply grateful to the Kaiyuan International Mathematical Sciences Institute (KIMSI) in Changsha, China, for hosting their visit and providing office space and facilities, where part of this work was completed.
References
Lukas Kölsch, Department of Mathematics and Statistics, University of South Florida, 140 7th Ave. South, St. Petersburg, FL 33701, USA, Email address: lukas.koelsch.math@gmail.com
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@outlook.de