betaFeedback

WorksMathematics

Counting irreducible skew polynomials x2n+axn+1x^{2n}+ax^n+1

Abstract

We give a counting formula for the coefficients a\in\F_{p^h} for which x2n+axn+1x^{2n}+ax^n+1 is irreducible in \F_{p^h}[x;σ], for arbitrary positive integers nn and hh, where σ(a)=apσ(a)=a^p and pp is an arbitrary prime power. The proof reduces the problem to quadratic reduced norms, establishes a uniformity statement for every Frobenius generator, and counts the remaining coefficients by Möbius inversion.

Introduction

Throughout, pp denotes a prime power, not necessarily the characteristic. Let q=phq=p^h, and let σ∈Aut⁡(Fq)\sigma\in\operatorname{Aut}(\mathbb{F}_q) be given by σ(c)=cp\sigma(c)=c^p. We study the number of coefficients a∈Fqa \in\mathbb{F}_q for which

fa(x)=x2n+axn+1f_a(x)=x^{2n}+ax^n+1

is irreducible in the skew polynomial ring Fq[x;σ]\mathbb{F}_q[x;\sigma], whose multiplication is determined by xc=σ(c)xxc=\sigma(c)x. Our aim is an explicit formula valid for every pair of positive integers n,hn,h, including the case n∤hn \nmid h.

The main structural result is a uniformity theorem for quadratic reduced norms on the one-parameter family t2+at+1t^2+at+1. Let PP be any prime power, and let τ(c)=cPr\tau(c)=c^{P^r} act on FPk\mathbb{F}_{P^k}, with gcd⁡(r,k)=1\gcd(r,k)=1. If

Nrd⁡FPk[t;τ](t2+at+1)=Z2+br(a)Z+1,Z=tk,\operatorname{Nrd}_{\mathbb{F}_{P^k}[t;\tau]}(t^2+at+1)=Z^2+b_r(a)Z+1,\qquad Z=t^k,

then for every b∈FPb \in\mathbb{F}_P such that Z2+bZ+1Z^2+bZ+1 is irreducible,

#{a∈FPk:br(a)=b}=Dk(P).\#\{a \in\mathbb{F}_{P^k}:b_r(a)=b\}=D_k(P).

Here Dk(P)D_k(P) is the explicit quantity defined in (2). Thus the norm map has equally sized fibers over all irreducible quadratics with constant term one, and the same fiber size holds for every generator of Gal⁡(FPk/FP)\operatorname{Gal}(\mathbb{F}_{P^k}/\mathbb{F}_P). This separates the count of the original coefficients from the arithmetic restrictions on their norm values.

To state the resulting counting formula, put

d=gcd⁡(n,h),m=n/d,H=h/d,P=pd.d=\gcd(n,h),\qquad m=n/d,\qquad H=h/d,\qquad P=p^d.

Writing Nn(p,h)N_n(p,h) for the number of irreducible members of the family, we prove

Nn(p,h)=DH(P)Bd,m(p).(1)N_n(p,h)=D_H(P)B_{d,m}(p). \tag*{(1)}

2020 Mathematics Subject Classification. Primary 11T06; Secondary 16S36, 11G20.

Key words and phrases. Finite fields, skew polynomial rings, irreducible polynomials, reduced norms, semilinear maps, Drinfeld curves.

The first author was supported by JSPS KAKENHI Grant Number JP24K00630 and JST COI-NEXT Grant Number JPMJPF2221. The second author was supported by JSPS KAKENHI Grant Number JP25K06959.

The factor DH(P)D_H(P) is given by

DH(P)={(PH+1)/(P+1),H odd,(1−(−P)H/2)2/(P+1),H even.D_H(P)=\begin{cases} (P^H+1)/(P+1), & H\ \text{odd},\\ (1-(-P)^{H/2})^2/(P+1), & H\ \text{even}. \end{cases}

The factor Bd,m(p)B_{d,m}(p) counts coefficients b∈Fpdb \in\mathbb{F}_{p^d} for which V2+bV+1V^2+bV+1 is irreducible, bb has degree dd over Fp\mathbb{F}_p, and Xm−αX^m-\alpha is irreducible over Fp2d\mathbb{F}_{p^{2d}} for a root α\alpha of the quadratic. Theorem 2.1 evaluates this factor by an explicit divisor sum. Thus (1) can be evaluated using integer arithmetic, without enumerating the elements of Fq\mathbb{F}_q.

For the standard generator, we express the fiber as a twisted Frobenius fixed-point count on the Drinfeld curve

XYP−XPY=1.XY^P-X^PY=1.

For background on this curve and its group action, see [1 Chapter 2]; the geometric facts and trace calculations needed here are verified in § 4. For a general generator, suitable normal forms for two-dimensional semilinear maps reduce the comparison to power maps having the same image and the same fiber size. Combining this uniformity with the norm reduction and Möbius inversion gives (1).

The distinction from root-count distributions is important here. For the standard Frobenius, the linearized polynomial XP2+aXP+XX^{P^2}+aX^P+X leads, by Y=XP−1Y=X^{P-1} for X≠0X\ne0, to the projective polynomial YP+1+aY+1Y^{P+1}+aY+1 studied in Bluher’s framework [2]. The number of KK-rational roots of this projective polynomial is the number of FP\mathbb{F}_P-lines in the root space invariant under FkF^k, where K=FPkK=\mathbb{F}_{P^k} and FF is the PP-power Frobenius. This number does not, in general, determine the characteristic polynomial of FkF^k. Thus an aggregate root-count distribution does not by itself give the count for each prescribed irreducible quadratic norm. Our curve argument keeps track of that characteristic polynomial.

Skew polynomial rings were introduced by Ore [3] and provide an algebraic setting for division, factorization, and semilinear operators. For finite coefficient fields, Caruso and Le Borgne identify reduced norms with characteristic polynomials of iterated companion semilinear maps and characterize irreducibility through the norm [4 Lemma 2.1.15 and Proposition 2.1.17]. Pumplün and Thompson [5] study reduced norms and their relation to reducibility for a broader class of skew polynomial rings. For more general Ore extensions, Churchill and Zhang [6] obtain irreducibility criteria using noncommutative valuations. In the finite-field setting considered here, the reduced-norm criterion provides the structural reduction. To obtain an exact coefficient count, one must additionally determine the fibers of the norm on the prescribed family and impose the conditions arising from power substitution. It is this restricted fiber count that is supplied by the uniformity theorem above.

Odoni [7 Theorem 1] counts all irreducible additive polynomials of a given degree under composition, using the corresponding skew polynomial ring and its central quotients. Such a total count does not determine how many irreducible elements have specified coefficients or a prescribed sparse form. Our family fixes the leading and constant coefficients at one and allows only the term axnax^n between them. The passage from total enumeration to this constrained count requires the norm fibers on that family. Factorization algorithms over finite fields were developed by Giesbrecht [8] and improved by Caruso and Le Borgne [4]. Skew polynomial rings also occur in the construction of skew-cyclic codes [9]; a survey of the ring-theoretic and coding-theoretic background is given in [10].

A motivation for the prescribed family comes from the second author’s study of self-adjoint irreducible polynomials in skew Laurent polynomial rings over finite fields [11]. That work determines the complete fiber-size distribution of c↦cpj+c−1c \mapsto c^{p^j}+c^{-1} for arbitrary positive integers jj [11 Corollary 3.12], by combining Bluher’s root-counting results [2] with a second-moment argument based on point counts on Edixhoven curves. In particular, [11 Corollary 5.11] gives the number of monic self-adjoint irreducible Laurent polynomials of degree two, equivalently the total number of irreducible polynomials x2+ax+1x^2+ax+1, with Dh(p)D_h(p) as in (2):

N1(p,h)=p−ϵp2Dh(p),ϵp={0,p even,1,p odd.N_1(p,h)=\frac{p-\epsilon_p}{2}D_h(p), \qquad\epsilon_p=\begin{cases}0,&p\text{ even},\\1,&p\text{ odd}.\end{cases}

The same work also gives the total degree-two count for arbitrary Frobenius powers [11 Corollary 4.3]. The present paper refines these total counts by prescribing each irreducible quadratic reduced norm and proving independence of the Frobenius generator for the individual fibers. These refined counts allow us to impose the exact-degree and power-substitution conditions needed to treat x2n+axn+1x^{2n}+ax^n+1 for all n,hn,h. For this same sparse family, [11 Corollary 4.2] gives an upper bound by excluding coefficients that admit a factorization (xn+b)(xn+c)(x^n+b)(x^n+c). Here we obtain the exact count by imposing necessary and sufficient conditions for irreducibility. The general higher-degree bound in [11 Proposition 6.8] concerns the larger family of self-adjoint Laurent polynomials. Indeed, multiplication on the left by x−nx^{-n} gives

x−nfa(x)=xn+σ−n(a)+x−n.x^{-n}f_a(x)=x^n+\sigma^{-n}(a)+x^{-n}.

Since a↦σ−n(a)a\mapsto\sigma^{-n}(a) is a bijection of Fq\mathbb{F}_q, and passing from the polynomial ring to the Laurent ring preserves irreducibility for polynomials with nonzero constant term, our count is equivalently an exact count of the self-adjoint irreducible Laurent polynomials xn+c+x−nx^n+c+x^{-n}, with c∈Fqc\in\mathbb{F}_q.

A further motivation for counting self-adjoint irreducible Laurent polynomials comes from their connection with primitive local Galois representations. In the construction of [12], irreducibility of an associated additive polynomial under composition gives a sufficient condition for the resulting representation to be primitive [12 Corollary 3.28(3)].

The arithmetic part is related to the classical enumeration of self-reciprocal irreducible polynomials over finite fields. Meyn and Götz Meyn and Götz [13] give a Möbius-inversion proof of that enumeration and describe its relation to the groups of roots of unity of order pj+1p^j+1. In our setting, a root α\alpha of an irreducible quadratic V2+bV+1∈Fpd[V]V^2+bV+1\in\mathbb{F}_{p^d}[V] satisfies

αpd+1=1,b=−(α+α−1).\alpha^{p^d+1}=1,\qquad b=-(\alpha+\alpha^{-1}).

This connects exact-degree conditions on the coefficient bb with self-reciprocal polynomials. The remaining power-substitution condition is expressed by the usual finite-field irreducibility criterion for Xm−αX^m-\alpha; see, for example, [14 Theorem 3.75].

When h=nkh=nk, Corollary 7.1 gives

Nn(p,nk)=Cn(p)Dk(pn)=nSp(2n)Dk(pn).N_n(p,nk)=C_n(p)D_k(p^n)=nS_p(2n)D_k(p^n).

where Cn(p)=Bn,1(p)C_n(p)=B_{n,1}(p) and Sp(2n)S_p(2n) denotes the number of monic self-reciprocal irreducible polynomials of degree 2n2n over Fp\mathbb{F}_p. The identity Cn(p)=nSp(2n)C_n(p)=nS_p(2n) is explained in § 7; see [13]. This identifies the classical self-reciprocal count inside the formula and isolates the extension factor Dk(pn)D_k(p^n). The specializations n=2,3,4n=2,3,4 are made explicit at the end of the paper. If nn is a power of two, the formula also gives Nn(p,h)=0N_n(p,h)=0 whenever n∤hn\nmid h.

Generative AI assistance was used in formulating the conjecture for general nn, exploring and developing proof arguments, and preparing the exposition. It was also used to develop code for supplementary numerical checks, which are not used in the proofs. Further details are provided in the declaration at the end of the paper.

The paper is organized as follows. § 2 states the arithmetic formula. § 3 reduces irreducibility to a quadratic norm and a binomial condition. § 4 and § 5 establish the fiber count for the standard and arbitrary Frobenius generators, respectively. § 6 completes the arithmetic count and proves the main theorem. § 7 records its specializations.

Statement of the formula

Let pp be a prime power, n,h≥1n,h\geq1, and put

Nn(p,h)=#{a∈Fph:x2n+axn+1 is irreducible in Fph[x;σ]},σ(c)=cp.N_n(p,h)=\#\left\{a\in\mathbb{F}_{p^h}:x^{2n}+ax^n+1\text{ is irreducible in }\mathbb{F}_{p^h}[x;\sigma]\right\},\qquad\sigma(c)=c^p.

Set

d=gcd⁡(n,h),m=n/d,H=h/d,P=pd.d=\gcd(n,h),\qquad m=n/d,\qquad H=h/d,\qquad P=p^d.

For any prime power PP define

Dk(P)={(Pk+1)/(P+1),k odd,(1−(−P)k/2)2/(P+1),k even.(2)D_k(P)= \begin{cases} (P^k+1)/(P+1), & k\text{ odd},\\ \left(1-(-P)^{k/2}\right)^2/(P+1), & k\text{ even}. \end{cases} \tag*{(2)}

Write R=rad⁡(m)R=\operatorname{rad}(m), with rad⁡(1)=1\operatorname{rad}(1)=1. If mm is even or R∤P+1R\nmid P+1, define Bd,m(p)=0B_{d,m}(p)=0. Otherwise set

L=∏ℓ∣Rℓvℓ(P+1),ge=gcd⁡(P+1,pe−1),Bd,m(p)=φ(R)2R∑e∣2dL∣geμ(2d/e)ge.(3)L=\prod_{\ell\mid R}\ell^{v_\ell(P+1)},\qquad g_e=\gcd(P+1,p^e-1),\qquad B_{d,m}(p)=\frac{\varphi(R)}{2R}\sum_{\substack{e\mid2d\\L\mid g_e}}\mu(2d/e)g_e. \tag*{(3)}

Empty products are 1; in particular R=L=1R=L=1 when m=1m=1. Here μ\mu and φ\varphi denote the Möbius and Euler functions, rad⁡(m)\operatorname{rad}(m) is the product of the distinct prime divisors of mm, and vℓv_\ell denotes the ℓ\ell-adic valuation.

Theorem 2.1. With the preceding notation,

Nn(p,h)=DH(pd)Bd,m(p).N_n(p,h)=D_H(p^d)B_{d,m}(p).

The two factors in the formula have different roles. The factor Bd,m(p)B_{d,m}(p) counts the quadratic norm coefficients b∈FPb\in\mathbb{F}_P that satisfy the irreducibility conditions obtained in § 3. For each such bb, the factor DH(P)D_H(P) counts the coefficients a∈Fpha\in\mathbb{F}_{p^h} having that norm coefficient. Thus the proof separates into an arithmetic count of admissible norm coefficients and a uniform count of their preimages. The latter is established in § 4 and § 5, and the two counts are combined in § 6.

We fix our convention for reduced norms. Let KK be a finite field, let τ∈Aut⁡(K)\tau\in\operatorname{Aut}(K) have order ss, and put

K0=Kτ,R=K[t;τ],z=ts.K_0=K^\tau,\qquad\mathcal{R}=K[t;\tau],\qquad z=t^s.

The center of R\mathcal{R} is K0[z]K_0[z], and its central localization

A=R⊗K0[z]K0(z)\mathcal{A}=\mathcal{R}\otimes_{K_0[z]}K_0(z)

is a central simple algebra of degree ss over K0(z)K_0(z). For f∈Rf \in\mathcal{R}, we write

Nrd⁡R(f):=Nrd⁡A/K0(z)(f)∈K0[z].\operatorname{Nrd}_{\mathcal{R}}(f) := \operatorname{Nrd}_{\mathcal{A}/K_0(z)}(f) \in K_0[z].

Here the reduced norm is defined by extending scalars to a splitting field, identifying the resulting algebra with a matrix algebra, and taking the determinant. Concretely, A\mathcal{A} is a left K(z)K(z)-vector space with basis {1,t,…,ts−1}\{1,t,\ldots,t^{s-1}\}. Right multiplication by ff is K(z)K(z)-linear, and Nrd⁡R(f)\operatorname{Nrd}_{\mathcal{R}}(f) is the determinant of its matrix in this basis. We always specify the central variable when comparing reduced norms in different skew polynomial rings.

We record explicitly how the determinant convention relates to companion semilinear maps. This is the even-degree form of the standard norm identity and irreducibility criterion; compare [4 Lemma 2.1.15 and Proposition 2.1.17]. Let f=tD+∑i<Dcitif=t^D+\sum_{i<D}c_it^i be monic of even degree DD with c0≠0c_0\ne0. On the left KK-vector space

Mf=R/Rf,M_f=\mathcal{R}/\mathcal{R}f,

left multiplication by tt induces a τ\tau-semilinear map Φ\Phi. In the basis {1,t,…,tD−1}\{1,t,\ldots,t^{D-1}\} its coefficient matrix is the companion matrix CfC_f, with last column (−c0,…,−cD−1)T(-c_0,\ldots,-c_{D-1})^{\mathsf{T}}. In particular,

Φs=Cfτ(Cf)⋯τs−1(Cf)\Phi^s=C_f\tau(C_f)\cdots\tau^{s-1}(C_f)

as a KK-linear map. With z=tsz=t^s, we have

Nrd⁡R(f)=det⁡(zID−Φs),f irreducible⟺Nrd⁡R(f) irreducible in K0[z].\operatorname{Nrd}_{\mathcal{R}}(f)=\det(zI_D-\Phi^s),\qquad f\ \text{irreducible}\Longleftrightarrow\operatorname{Nrd}_{\mathcal{R}}(f)\ \text{irreducible in }K_0[z].

Verification of the conventions and criterion. View R\mathcal{R} as a free left K[z]K[z]-module with basis {1,t,…,ts−1}\{1,t,\ldots,t^{s-1}\}. Right multiplication by ff is injective and has cokernel MfM_f. Smith normal form over K[z]K[z] shows that its determinant, up to a nonzero constant, is the characteristic polynomial of multiplication by zz on MfM_f, namely that of Φs\Phi^s. The leading coefficient of this determinant is (−1)D(s−1)(-1)^{D(s-1)}: its leading term comes from right multiplication by tDt^D, and the matrix for right multiplication by tt has determinant (−1)s−1z(-1)^{s-1}z. More explicitly, a determinant term formed from monomials of degrees j0,…,js−1≤Dj_0,\ldots,j_{s-1}\le D has zz-degree (j0+⋯+js−1)/s(j_0+\cdots+j_{s-1})/s; degree DD requires all ji=Dj_i=D. Since DD is even, the leading coefficient is one. This proves the displayed identity, including the order of the twisted matrix product.

For completeness, the same determinant calculation gives deg⁡zNrd⁡(g)=deg⁡tg\deg_z\operatorname{Nrd}(g)=\deg_tg for any nonzero skew polynomial gg. Multiplicativity therefore shows that an irreducible norm forces ff to be irreducible. Conversely, if ff is irreducible, MfM_f is a simple R\mathcal{R}-module. Its central annihilator is generated by an irreducible polynomial η∈K0[z]\eta\in K_0[z]; this follows since the image of the center in the endomorphism division ring of MfM_f is a finite integral domain. As c0≠0c_0\ne0, Φ\Phi is invertible, so η≠z\eta\ne z. Put E=K0[z]/(η)E=K_0[z]/(\eta) and e=deg⁡ηe=\deg\eta. The central quotient R/ηR\mathcal{R}/\eta\mathcal{R} is a central simple algebra of degree ss over the finite field EE, hence is Ms(E)M_s(E); see also [4 Theorem 2.1.2 and Corollary 2.1.3]. Its simple module has EE-dimension ss. Comparing K0K_0-dimensions gives sD=sesD=se, hence D=eD=e. Since η\eta annihilates Φs\Phi^s, and the characteristic polynomial of Φs\Phi^s belongs to K0[z]K_0[z], all its irreducible factors over K0K_0 equal η\eta. Equality of degrees and monicity therefore give Nrd⁡(f)=η\operatorname{Nrd}(f)=\eta.

For f=t2+at+1f=t^2+at+1, this calculation uses exactly

Cf=Aa=(0−11−a),Φs=Aaτ(Aa)⋯τs−1(Aa).C_f=A_a=\begin{pmatrix}0&-1\\1&-a\end{pmatrix},\qquad \Phi^s=A_a\tau(A_a)\cdots\tau^{s-1}(A_a).

Each factor has determinant one. All reduced norms below use the indicated central variable.

Reduction to a quadratic norm

The aim of this section is to replace irreducibility in the original skew polynomial ring by three conditions over finite fields. We first express the reduced norm in terms of a quadratic polynomial with coefficient ba∈FPb_a \in\mathbb{F}_P. We then show that irreducibility is equivalent to the irreducibility of this quadratic, an exact-degree condition on bab_a, and the irreducibility of a binomial over FP2\mathbb{F}_{P^2}. These are precisely the conditions used later to count the admissible values of bab_a.

Put K=FphK = \mathbb{F}_{p^h} and consider the subrings

S=K[y;σd],y=xd,T=K[t;σn],t=xn=ym.S = K[y;\sigma^d], \qquad y = x^d, \qquad T = K[t;\sigma^n], \qquad t = x^n = y^m.

Both automorphisms on KK have order HH and fixed field FP\mathbb{F}_P. The corresponding central variables satisfy

Z=xh=yH,W=tH=Zm.Z = x^h = y^H, \qquad W = t^H = Z^m.

The inclusions and the corresponding centers can be summarized as

ringcentercentral variable
K[x;σ]K[x;\sigma]FP[Z]\mathbb{F}_P[Z]Z=xhZ = x^h
S=K[y;σd]S = K[y;\sigma^d]FP[Z]\mathbb{F}_P[Z]Z=yHZ = y^H
T=K[t;σn]T = K[t;\sigma^n]FP[W]\mathbb{F}_P[W]W=tH=ZmW = t^H = Z^m

Table 1.

The passage from TT to SS substitutes W=ZmW = Z^m; the passage from SS to the original ring takes the product of the dd coefficient conjugates over FP\mathbb{F}_P. The next lemma makes these two operations precise.

For a∈Ka \in K, define ba∈FPb_a \in\mathbb{F}_P by

Nrd⁡T(t2+at+1)=Ga(W)=W2+baW+1.(4)\operatorname{Nrd}_T(t^2 + at + 1) = G_a(W) = W^2 + b_a W + 1. \tag*{(4)}

This norm has the stated shape: the companion matrix is

Aa=(0−11−a).A_a = \begin{pmatrix} 0 & -1 \\ 1 & -a \end{pmatrix}.

and the HH-fold iterate of the associated semilinear map is KK-linear with determinant one. Its characteristic polynomial therefore has degree two and constant term one.

Lemma 3.1. In the original skew polynomial ring,

Nrd⁡K[x;σ](x2n+axn+1)=∏i=0d−1(Z2m+bapiZm+1).\operatorname{Nrd}_{K[x;\sigma]}(x^{2n} + ax^n + 1) = \prod_{i=0}^{d-1}(Z^{2m} + b_a^{p^i} Z^m + 1).

Proof. Step 1: compare the norms in TT and SS. We claim that

Nrd⁡S(x2n+axn+1)=Ga(Zm).\operatorname{Nrd}_S(x^{2n} + ax^n + 1) = G_a(Z^m).

Indeed, extend the center of the cyclic algebra obtained by centrally localizing TT from FP(W)\mathbb{F}_P(W) to FP(Z)\mathbb{F}_P(Z) by W↦ZmW \mapsto Z^m. The map fixes KK and sends tt to ymy^m; it respects the skew relation because yma=σn(a)ymy^m a = \sigma^n(a)y^m for a∈Ka \in K. It identifies the resulting cyclic algebra with that of SS. Indeed, since gcd⁡(m,H)=1\gcd(m,H) = 1, choose integers u,vu,v with um+vH=1um + vH = 1. After localization, tt and ZZ are invertible, since tH=Wt^H = W and W,ZW,Z are nonzero elements of the respective centers' fraction fields. Thus negative exponents are allowed, and the image contains (ym)uZv=y(y^m)^u Z^v = y. The map is therefore surjective. Both algebras have dimension H2H^2 over FP(Z)\mathbb{F}_P(Z), so the map is an isomorphism. Reduced norms commute with extension of the center, including inseparable extensions. This proves the first assertion as an identity of rational functions and hence of polynomials.

Step 2: recover the norm in the original ring. Compute this reduced norm using right multiplication in the left K(Z)K(Z)-basis {1,x,…,xh−1}\{1,x,\ldots,x^{h-1}\}. Group the basis vectors according to their exponents modulo dd. Since all exponents of x2n+axn+1x^{2n+ax^n+1} are multiples of dd, the matrix is block diagonal, with dd blocks of size HH. The block indexed by ii is obtained from the matrix over SS by applying σi\sigma^i to its coefficients. Its determinant is Gaσi(Zm)G_a^{\sigma^i}(Z^m). Multiplying the block determinants proves the claim.

Lemma 3.2. For b∈Fpdb \in\mathbb{F}_{p^d} put

Jb(V)=∏i=0d−1(V2+bpiV+1)∈Fp[V].J_b(V)=\prod_{i=0}^{d-1}(V^2+b^{p^i}V+1)\in\mathbb{F}_p[V].

Then Jb(Zm)J_b(Z^m) is irreducible in Fp[Z]\mathbb{F}_p[Z] if and only if

  1. V2+bV+1V^2+bV+1 is irreducible over Fpd\mathbb{F}_{p^d};

  2. [Fp(b):Fp]=d[\mathbb{F}_p(b):\mathbb{F}_p]=d;

  3. for a root α\alpha of this quadratic, Zm−αZ^m-\alpha is irreducible over Fp2d\mathbb{F}_{p^{2d}}.

Proof. Step 1: determine when JbJ_b is irreducible. If bb belongs to a proper subfield Fpe\mathbb{F}_{p^e}, e∣de\mid d, then JbJ_b is a d/ed/e-th power. If the quadratic splits over Fpd\mathbb{F}_{p^d}, taking coefficient norms of its two linear factors gives a factorization of JbJ_b into two polynomials of degree dd. Thus irreducibility of JbJ_b requires the first two conditions.

Conversely, under those two conditions, b=−(α+α−1)b=-(\alpha+\alpha^{-1}) shows that Fp(α)\mathbb{F}_p(\alpha) contains Fpd\mathbb{F}_{p^d}. The quadratic is irreducible there, so [Fp(α):Fp]=2d[\mathbb{F}_p(\alpha):\mathbb{F}_p]=2d. Hence JbJ_b is the minimal polynomial of α\alpha.

Step 2: account for the substitution V=ZmV=Z^m. If JbJ_b is reducible, then substituting ZmZ^m in its factors shows that Jb(Zm)J_b(Z^m) is reducible. Thus it remains to work under the first two conditions. Choose β\beta with βm=α\beta^m=\alpha. Since α∈Fp(β)\alpha\in\mathbb{F}_p(\beta), the tower formula gives

[Fp(β):Fp]=2d[Fp(β):Fp2d].[\mathbb{F}_p(\beta):\mathbb{F}_p]=2d[\mathbb{F}_p(\beta):\mathbb{F}_{p^{2d}}].

The polynomial Jb(Zm)J_b(Z^m) has degree 2dm2dm and has β\beta as a root. It is therefore irreducible exactly when the left-hand side is 2dm2dm, or equivalently when β\beta has degree mm over Fp2d\mathbb{F}_{p^{2d}}. This is exactly the irreducibility of Zm−αZ^m-\alpha.

Uniformity for the standard Frobenius

For this section let PP be any prime power, let K=FPkK=\mathbb{F}_{P^k}, and use the automorphism c↦cPc\mapsto c^P. Put

T1=K[t;τ1],τ1(c)=cP,Z=tk.T_1=K[t;\tau_1],\qquad\tau_1(c)=c^P,\qquad Z=t^k.

Define b1(a)∈FPb_1(a)\in\mathbb{F}_P by

Nrd⁡T1(t2+at+1)=Z2+b1(a)Z+1,wk,1(b)=#{a∈K:b1(a)=b}.\operatorname{Nrd}_{T_1}(t^2+at+1)=Z^2+b_1(a)Z+1,\qquad w_{k,1}(b)=\#\{a\in K:b_1(a)=b\}.

Proposition 4.1. If Z2+bZ+1Z^2+bZ+1 is irreducible over FP\mathbb{F}_P, then wk,1(b)=Dk(P)w_{k,1}(b)=D_k(P).

Proof. We first realize the desired fiber count as a fixed-point count on a curve: equation (6) identifies (P+1)wk,1(b)(P+1)w_{k,1}(b) with the number of fixed points of a twisted Frobenius on the Drinfeld curve YY. We then pass to its smooth projective closure and apply the Grothendieck–Lefschetz fixed-point formula. The key simplification is that the twisted square of Frobenius has only the eigenvalue −P-P on first cohomology. Consequently, all the required traces reduce to the two initial cases of an elliptic group element and such an element composed with Frobenius. This gives the explicit value Dk(P)D_k(P).

Step 1: realize the norm fiber as a fixed-point set. Let FF denote the coordinatewise PP-power Frobenius over FP\mathbb{F}_P, and consider the affine Drinfeld curve

Y={u=(u1,u2)t:det⁡(u,Fu)=1}Y=\left\{u=(u_1,u_2)^t:\det(u,Fu)=1\right\}

over FP\mathbb{F}_P, which admits a natural SL⁡2(FP)\operatorname{SL}_2(\mathbb{F}_P)-action. Define the morphism

Γ:Y⟶AFP1,u⟼−det⁡(u,F2u).\Gamma:Y\longrightarrow\mathbb{A}^1_{\mathbb{F}_P},\qquad u\longmapsto-\det(u,F^2u).

Since

det⁡(Fu,F2u)=det⁡(u,Fu)P=1,\det(Fu,F^2u)=\det(u,Fu)^P=1,

expressing F2uF^2u in the basis (u,Fu)(u,Fu) gives

(F2+Γ(u)F+1)u=0(5)(F^2+\Gamma(u)F+1)u=0 \tag*{(5)}

For a∈FPa\in\mathbb{F}_P, we write Γ−1(a)\Gamma^{-1}(a) for the geometric fiber in Y(FP)Y(\mathbb{F}_P).

Each geometric fiber of Γ\Gamma is a simply transitive SL⁡2(FP)\operatorname{SL}_2(\mathbb{F}_P)-set under the action on column vectors. Indeed, let VaV_a be the root space of

XP2+aXP+X.X^{P^2}+aX^P+X.

This polynomial is separable and FP\mathbb{F}_P-linear, so VaV_a is a two-dimensional FP\mathbb{F}_P-vector space. For an ordered basis (u1,u2)(u_1,u_2) of VaV_a, its Moore determinant

D=u1u2P−u1Pu2D=u_1u_2^P-u_1^Pu_2

is nonzero. The root relation gives DP=DD^P=D, so D∈FP×D\in\mathbb{F}_P^\times. Rescaling one basis vector therefore produces a basis with D=1D=1. By (5), the points of Γ−1(a)\Gamma^{-1}(a) are precisely the column vectors (u1,u2)t(u_1,u_2)^t arising from these normalized ordered bases of VaV_a. Any two such bases are related by a unique element of SL⁡2(FP)\operatorname{SL}_2(\mathbb{F}_P), proving the assertion.

We will also use the identities

Γ(Fku)=Γ(u)Pk,Γ(gu)=Γ(u)(g∈SL⁡2(FP)).\Gamma(F^ku)=\Gamma(u)^{P^k},\qquad\Gamma(gu)=\Gamma(u)\quad(g\in\operatorname{SL}_2(\mathbb{F}_P)).

The first follows because Γ\Gamma is defined over FP\mathbb{F}_P. For the second, F2(gu)=gF2uF^2(gu)=gF^2u and det⁡g=1\det g=1 give

Γ(gu)=−det⁡(gu,gF2u)=Γ(u).\Gamma(gu)=-\det(gu,gF^2u)=\Gamma(u).

Choose C∈SL⁡2(FP)C\in\operatorname{SL}_2(\mathbb{F}_P) with characteristic polynomial

Gb(Z)=Z2+bZ+1.G_b(Z)=Z^2+bZ+1.

For a∈K=FPka\in K=\mathbb{F}_{P^k} and u∈Γ−1(a)u\in\Gamma^{-1}(a), we have

Γ(Fku)=aPk=a.\Gamma(F^ku)=a^{P^k}=a.

Thus FkuF^ku belongs to the same fiber as uu. By simple transitivity, there is a unique Cu∈SL⁡2(FP)C_u\in\operatorname{SL}_2(\mathbb{F}_P) such that

Fku=Cuu.F^ku=C_uu.

Put

B=[u,Fu],Aa=(0−11−a).B=[u,Fu],\qquad A_a=\begin{pmatrix}0&-1\\1&-a\end{pmatrix}.

Equation (5) gives

F(B)=BAa.F(B)=BA_a.

Since FF commutes with CuC_u, iteration yields

Fk(B)=BAaF(Aa)⋯Fk−1(Aa)=CuB.F^k(B)=BA_aF(A_a)\cdots F^{k-1}(A_a)=C_uB.

The matrix BB is invertible because det⁡(u,Fu)=1\det(u,Fu)=1. Consequently, CuC_u is conjugate to AaF(Aa)⋯Fk−1(Aa)A_aF(A_a)\cdots F^{k-1}(A_a). By the companion description of the reduced norm, its characteristic polynomial is

Z2+b1(a)Z+1.Z^2+b_1(a)Z+1.

Matrices in SL⁡2(FP)\operatorname{SL}_2(\mathbb{F}_P) with irreducible characteristic polynomial GbG_b form one conjugacy class. Their centralizer in SL⁡2(FP)\operatorname{SL}_2(\mathbb{F}_P) is isomorphic to the norm-one subgroup of FP2×\mathbb{F}_{P^2}^{\times} and therefore has order P+1P+1. For g∈SL⁡2(FP)g\in\operatorname{SL}_2(\mathbb{F}_P), the identity

Fk(gu)=gCuu=(gCug−1)(gu)F^k(gu)=gC_u u=(gC_u g^{-1})(gu)

shows that

Cgu=gCug−1.C_{gu}=gC_u g^{-1}.

It follows that, in a fiber with b1(a)=bb_1(a)=b, exactly P+1P+1 points uu satisfy Cu=CC_u=C, or equivalently C−1Fku=uC^{-1}F^k u=u. Fibers with b1(a)≠bb_1(a)\ne b contribute no such points.

Conversely, if u∈Y(F‾P)u\in Y(\overline{\mathbb{F}}_P) satisfies C−1Fku=uC^{-1}F^k u=u, then

Γ(u)Pk=Γ(Fku)=Γ(Cu)=Γ(u).\Gamma(u)^{P^k}=\Gamma(F^k u)=\Gamma(Cu)=\Gamma(u).

Hence Γ(u)∈K\Gamma(u)\in K, so every fixed point belongs to one of the fibers considered above. Therefore

(P+1)wk,1(b)=#YC−1Fk.(6)(P+1)w_{k,1}(b)=\#Y^{C^{-1}F^k}. \tag*{(6)}

where the right-hand side denotes the number of geometric fixed points of C−1FkC^{-1}F^k on YY. In particular, for k=1k=1 this number is P+1P+1, since b1(a)=ab_1(a)=a and hence w1,1(b)=1w_{1,1}(b)=1.

Step 2: compute the action of a twisted square of Frobenius. The smooth projective closure is

Y‾:STP−SPT=UP+1.\overline{Y}: \quad ST^P-S^PT=U^{P+1}.

The affine curve YY is identified with the chart U≠0U\ne0 via u1=S/Uu_1=S/U and u2=T/Uu_2=T/U. The partial derivatives of its defining equation are TPT^P, −SP-S^P, −UP-U^P, which have no common projective zero. It is thus a smooth plane curve of degree P+1P+1, with genus g=P(P−1)/2g=P(P-1)/2. Setting U=0U=0 identifies its boundary with P1(FP)\mathbb{P}^1(\mathbb{F}_P). Write ε=−I2\varepsilon=-I_2 and H=Heˊt1(Y‾F‾P,Qℓ)\mathcal{H}=H^1_{\mathrm{\acute et}}(\overline{Y}_{\overline{\mathbb{F}}_P},\mathbb{Q}_{\ell}), where ℓ\ell is a prime different from the characteristic. Here FF is the curve morphism raising coordinates to the PP-th power, and its action on cohomology is the pullback F∗F^*, with Weil eigenvalues of absolute value P1/2P^{1/2} on H\mathcal{H}. We use the same notation for commuting curve endomorphisms and their pullback actions on cohomology. For h∈SL⁡2(FP)h\in\operatorname{SL}_2(\mathbb{F}_P) and j≥1j\ge1, the map hFjhF^j has zero differential, so its fixed points are simple. Its traces on H0H^0 and H2H^2 are 11 and PjP^j, respectively. Thus the trace formula reads

#Y‾hFj=1+Pj−Tr⁡(hFj∣H).\#\overline{Y}^{hF^j}=1+P^j-\operatorname{Tr}(hF^j\mid\mathcal{H}).

There are P3−PP^3-P fixed points of εF2\varepsilon F^2 on YY. Indeed, u1P2=−u1u_1^{P^2}=-u_1 has P2−1P^2-1 nonzero solutions. For each such u1u_1, the equation u1u2P−u1Pu2=1u_1u_2^P-u_1^Pu_2=1 is a degree-PP polynomial equation in u2u_2 with nonzero derivative −u1P-u_1^P, so it has exactly PP distinct solutions over F‾P\overline{\mathbb{F}}_P. Raising this equation to the PP-th power and using u1P2=−u1u_1^{P^2}=-u_1 gives u1Pu2P2+u1u2P=1u_1^Pu_2^{P^2}+u_1u_2^P=1. Subtracting the original equation yields u1P(u2P2+u2)=0u_1^P(u_2^{P^2}+u_2)=0, hence u2P2=−u2u_2^{P^2}=-u_2. No solution has u1=0u_1=0. Adding the P+1P+1 boundary points gives P3+1P^3+1. The Grothendieck–Lefschetz trace formula gives

Tr⁡(εF2∣H)=1+P2−(P3+1)=−P2(P−1)=−2gP.\operatorname{Tr}(\varepsilon F^2\mid\mathcal{H})=1+P^2-(P^3+1)=-P^2(P-1)=-2gP.

To determine the absolute values of its eigenvalues, put Ψ=εF2\Psi=\varepsilon F^2. Since ε\varepsilon commutes with FF and ε2=1\varepsilon^2=1, we have

Ψ2=F4.\Psi^2=F^4.

By the Weil conjectures for smooth projective curves, every eigenvalue of F4F^4 on H\mathcal{H} has complex absolute value P2P^2. Hence every eigenvalue of Ψ\Psi has complex absolute value PP. There are 2g2g such eigenvalues, counted with multiplicity, and their sum is −2gP-2gP. Equality in the triangle inequality therefore forces every eigenvalue of Ψ\Psi to be −P-P. In particular, for every operator AA commuting with εF2\varepsilon F^2,

Tr⁡((εF2)sA∣H)=(−P)sTr⁡(A∣H).(7)\operatorname{Tr}\left((\varepsilon F^2)^sA\mid\mathcal{H}\right)=(-P)^s\operatorname{Tr}(A\mid\mathcal{H}). \tag*{(7)}

Indeed, after extending the coefficient field to an algebraic closure, the commuting operators Ψ\Psi and AA can be simultaneously upper triangularized. Every diagonal entry of Ψ\Psi is −P-P, so the diagonal entries of ΨsA\Psi^sA are (−P)s(-P)^s times those of AA. No semisimplicity assumption is needed.

Step 3: obtain the two initial traces. If g0∈SL⁡2(FP)g_0\in\operatorname{SL}_2(\mathbb{F}_P) has irreducible characteristic polynomial, it has no fixed points on Y‾\overline{Y}. On the affine curve a fixed vector would give eigenvalue one, and on the boundary a fixed point would give an FP\mathbb{F}_P-rational eigenline. Since g0g_0 has order prime to the characteristic and has no fixed points on Y‾\overline{Y}, the Grothendieck–Lefschetz trace formula gives

0=1−Tr⁡(g0∣H)+1,0=1-\operatorname{Tr}(g_0\mid\mathcal{H})+1,

and hence Tr⁡(g0∣H)=2\operatorname{Tr}(g_0\mid\mathcal{H})=2. Equation (6) for k=1k=1 gives #Yg0F=P+1\#Y^{g_0F}=P+1. There are no fixed boundary points, so

Tr⁡(g0F∣H)=0.\operatorname{Tr}(g_0F\mid\mathcal{H})=0.

Both statements remain true after multiplying g0g_0 by ε\varepsilon.

Step 4: evaluate the trace for every kk. Put Ψ=εF2\Psi=\varepsilon F^2. Since ε2=1\varepsilon^2=1 and all the operators in question commute, we have

C−1F2s=Ψs(C−1εs),C−1F2s+1=Ψs(C−1εsF).C^{-1}F^{2s}=\Psi^s(C^{-1}\varepsilon^s),\qquad C^{-1}F^{2s+1}=\Psi^s(C^{-1}\varepsilon^sF).

The element C−1εsC^{-1}\varepsilon^s again has irreducible characteristic polynomial. Hence Step 3 and (7) give trace 2(−P)s2(-P)^s in the even case and zero in the odd case. The action on the boundary is that of C−1C^{-1}, since Frobenius fixes P1(FP)\mathbb{P}^1(\mathbb{F}_P) pointwise; it has no fixed points there. Applying the Grothendieck–Lefschetz formula on Y‾\overline{Y} therefore gives

#YC−1Fk={Pk+1,k odd,Pk+1−2(−P)k/2,k even.\#Y^{C^{-1}F^k}= \begin{cases} P^k+1, & k\text{ odd},\\ P^k+1-2(-P)^{k/2}, & k\text{ even}. \end{cases}

Division by P+1P+1 in (6) proves the proposition. □

Changing the Frobenius generator

Let K=FpkK=\mathbb{F}_{p^k} and let

τ(c)=cpr,gcd⁡(r,k)=1.\tau(c)=c^{p^r},\qquad\gcd(r,k)=1.

For k>1k>1, we may and do assume 1≤r<k1\le r<k. Put Tr=K[t;τ]T_r=K[t;\tau] and Z=tkZ=t^k. Define br(a)∈Fpb_r(a)\in\mathbb{F}_p by

Nrd⁡Tr(t2+at+1)=Z2+br(a)Z+1,\operatorname{Nrd}_{T_r}(t^2+at+1)=Z^2+b_r(a)Z+1,

and set

wk,r(b)=#{a∈K:br(a)=b}.w_{k,r}(b)=\#\{a\in K:b_r(a)=b\}.

The purpose of this section is to prove that changing the generator of Gal⁡(K/Fp)\operatorname{Gal}(K/\mathbb{F}_p) does not change the fiber sizes over irreducible quadratics. More precisely, we establish the following reduction to § 4.

Proposition 5.1. For any rr relatively prime to kk, and any b∈Fpb\in\mathbb{F}_p for which Z2+bZ+1Z^2+bZ+1 is irreducible over Fp\mathbb{F}_p,

wk,r(b)=wk,1(b)=Dk(P).w_{k,r}(b)=w_{k,1}(b)=D_k(P).

For k=1k=1 this is already Proposition 4.1; henceforth assume k>1k>1. We first establish the conjugacy classification and normal forms, then express a fiber size as a count of normalized cyclic vectors. Finally, we compare that count with the standard-Frobenius count, using power maps with the same image and the same number of preimages.

Conjugacy and normal forms

In standard coordinates, a τ\tau-semilinear map has the form F(v)=Aτ(v)F(v)=A\tau(v), where τ\tau acts coordinatewise. We call AA its coefficient matrix. Its iterates satisfy

Fi(v)=Aτ(A)⋯τi−1(A)τi(v).F^i(v)=A\tau(A)\cdots\tau^{i-1}(A)\tau^i(v).

In particular, FkF^k is KK-linear because τk=1\tau^k=1. A τ\tau-semilinear map FF on K2K^2 is said to have determinant one if

ω(Fv,Fw)=τ(ω(v,w)),ω(v,w)=det⁡(v,w).\omega(Fv,Fw)=\tau(\omega(v,w)),\qquad\omega(v,w)=\det(v,w).

Equivalently, the coefficient matrix of FF has determinant one.

For b∈Fpb\in\mathbb{F}_p such that

Gb(Z)=Z2+bZ+1G_b(Z)=Z^2+bZ+1

is irreducible, let Fk,r(b)\mathcal{F}_{k,r}(b) denote the set of determinant-one τ\tau-semilinear maps FF on K2K^2 for which FkF^k has characteristic polynomial GbG_b. Here SL⁡2(K)\operatorname{SL}_2(K) acts on semilinear maps by conjugation: two maps FF and F′F' are conjugate if

F′=gFg−1F'=gFg^{-1}

for some g∈SL⁡2(K)g\in\operatorname{SL}_2(K).

Equivalently, if F=AτF=A\tau and F′=A′τF'=A'\tau, then

A′=gAτ(g)−1,A'=gA\tau(g)^{-1},

where τ(g)\tau(g) is obtained by applying τ\tau to each entry of gg.

The conjugacy question is a twisted conjugacy problem: it belongs to the framework of Lang–Steinberg theory and Shintani descent; see, for example, [15 Section 2.2] for the correspondence of classes and centralizers. We give an explicit two-dimensional proof to obtain the normal forms used in the counting argument. The classification and centralizer calculation are structural ingredients; the additional task is to count normalized cyclic vectors, or equivalently intersections with the companion family {Aa:a∈K}\{A_a:a\in K\}, and compare these counts for different generators.

Lemma 5.2. The set Fk,r(b)\mathcal{F}_{k,r}(b) consists of a single SL⁡2(K)\operatorname{SL}_{2}(K)-conjugacy class. The centralizer in SL⁡2(K)\operatorname{SL}_{2}(K) of any F∈Fk,r(b)F \in\mathcal{F}_{k,r}(b) has order P+1P+1.

Proof. We prove the conjugacy assertion in three steps. First, we derive normal forms and use them to show that Fk,r(b)\mathcal{F}_{k,r}(b) is nonempty. Second, we show that any two such maps are GL⁡2(K)\operatorname{GL}_{2}(K)-conjugate and compute their invertible self-intertwiners. Here an invertible self-intertwiner of FF means an element g∈GL⁡2(K)g \in\operatorname{GL}_{2}(K) satisfying gF=FggF=Fg. Finally, we use the determinants of these self-intertwiners to obtain SL⁡2(K)\operatorname{SL}_{2}(K)-conjugacy and determine the centralizer in SL⁡2(K)\operatorname{SL}_{2}(K).

Let α\alpha be a root of GbG_b in FP2\mathbb{F}_{P^2}. Then

αP=α−1,αP+1=1,\alpha^P=\alpha^{-1}, \qquad\alpha^{P+1}=1,

and α∉FP\alpha\notin\mathbb{F}_P.

We classify the semilinear maps in Fk,r(b)\mathcal{F}_{k,r}(b).

Case 1: k=2sk=2s is even.

Since kk is even and gcd⁡(r,k)=1\gcd(r,k)=1, the integer rr is odd. Thus τ\tau acts on FP2\mathbb{F}_{P^2} as the PP-power Frobenius, and in particular

τ(α)=αP=α−1.\tau(\alpha)=\alpha^P=\alpha^{-1}.

The two eigenvalues of FkF^k belong to KK and are distinct. Order the eigenlines so that the first eigenvalue is α\alpha. The identity FkF=FFkF^kF=FF^k and the τ\tau-semilinearity of FF show that FF exchanges these eigenlines: if Fkv=αvF^kv=\alpha v, then

Fk(Fv)=F(Fkv)=τ(α)Fv=α−1Fv.F^k(Fv)=F(F^kv)=\tau(\alpha)Fv=\alpha^{-1}Fv.

Choose eigenvectors e1,e2e_1,e_2 with

Fke1=αe1,Fke2=α−1e2,det⁡(e1,e2)=1.F^ke_1=\alpha e_1,\qquad F^ke_2=\alpha^{-1}e_2,\qquad\det(e_1,e_2)=1.

Such a normalization is possible by rescaling one eigenvector. Since FF is invertible and exchanges the eigenlines, write

F(e1)=de2,F(e2)=ce1,c,d∈K×.F(e_1)=de_2,\qquad F(e_2)=ce_1,\qquad c,d\in K^\times.

Semilinearity now gives

F(ue1+ve2)=cvτe1+duτe2.F(ue_1+ve_2)=cv^\tau e_1+du^\tau e_2.

In this determinant-one basis, the coefficient matrix of FF is (0cd0)\begin{pmatrix}0&c\\d&0\end{pmatrix}. Its determinant is −cd=1-cd=1, so d=−c−1d=-c^{-1}. Thus, in the resulting coordinates,

F(u,v)=(cvτ,−c−1uτ),c∈K×.(8)F(u,v)=(cv^\tau,-c^{-1}u^\tau),\qquad c\in K^\times. \tag*{(8)}

Put

N=NK/FP2(c).N=N_{K/\mathbb{F}_{P^2}}(c).

A direct iteration gives

F2(u,v)=(−cτ(c)−1uτ2,−c−1τ(c)vτ2).F^2(u,v)=\left(-c\tau(c)^{-1}u^{\tau^2},-c^{-1}\tau(c)v^{\tau^2}\right).

Since τ2\tau^2 generates Gal⁡(K/FP2)\operatorname{Gal}(K/\mathbb{F}_{P^2}), iterating this identity ss times yields

Fk=diag⁡((−1)sN1−P,(−1)sNP−1).F^k=\operatorname{diag}\left((-1)^sN^{1-P},(-1)^sN^{P-1}\right).

Thus FkF^k has characteristic polynomial GbG_b precisely when its eigenvalues are α\alpha and α−1\alpha^{-1}.

The norm map K×→FP2×K^{\times} \to\mathbb{F}_{P^2}^{\times} is surjective, and the map

N⟼N1−PN \longmapsto N^{1-P}

has image μP+1\mu_{P+1}, the group of (P+1)(P+1)-st roots of unity in FP2×\mathbb{F}_{P^2}^{\times}. Consequently, we may choose NN so that

N1−P=(−1)sα.N^{1-P}=(-1)^s\alpha.

This proves existence.

Next consider two maps of the form (8), with parameters c,c′c,c', whose first eigenvalues agree. Set

δ=c′/c.\delta=c'/c.

The equality of their first eigenvalues gives

N⁡K/FP2(δ)P−1=1,\operatorname{N}_{K/\mathbb{F}_{P^2}}(\delta)^{P-1}=1,

hence

N⁡K/FP2(δ)∈FP×.\operatorname{N}_{K/\mathbb{F}_{P^2}}(\delta)\in\mathbb{F}_{P}^{\times}.

It follows that

N⁡K/FP2(τ(δ)δ)=1.\operatorname{N}_{K/\mathbb{F}_{P^2}}\left(\frac{\tau(\delta)}{\delta}\right)=1.

By Hilbert’s Theorem 90 for the cyclic extension K/FP2K/\mathbb{F}_{P^2}, whose Galois group is generated by τ2\tau^2, there is u∈K×u\in K^{\times} such that

τ2(u)u=τ(δ)δ.\frac{\tau^2(u)}{u}=\frac{\tau(\delta)}{\delta}.

Define

g=diag⁡(u,δ−1τ(u)).g=\operatorname{diag}\left(u,\delta^{-1}\tau(u)\right).

The defining equation for uu gives gFc=Fc′ggF_c=F_{c'}g. The two maps are therefore GL⁡2(K)\operatorname{GL}_2(K)-conjugate.

Any invertible self-intertwiner gg of FcF_c also commutes with FckF_c^k. Since FckF_c^k is diagonal with distinct eigenvalues, gg must be diagonal. Writing g=diag⁡(u,v)g=\operatorname{diag}(u,v), the identity gFc=FcggF_c=F_cg is equivalent to

u=τ(v),v=τ(u).u=\tau(v),\qquad v=\tau(u).

Thus τ2(u)=u\tau^2(u)=u. Since the fixed field of τ2\tau^2 is FP2\mathbb{F}_{P^2}, the invertible self-intertwiners are exactly

diag⁡(u,τ(u)),u∈FP2×.\operatorname{diag}(u,\tau(u)),\qquad u\in\mathbb{F}_{P^2}^{\times}.

Their determinants are

uτ(u)=uP+1.u\tau(u)=u^{P+1}.

Hence the determinant map on the self-intertwiner group is the surjective norm

N⁡FP2/FP:FP2×⟶FP×.\operatorname{N}_{\mathbb{F}_{P^2}/\mathbb{F}_{P}}:\mathbb{F}_{P^2}^{\times}\longrightarrow\mathbb{F}_{P}^{\times}.

Case 2: kk is odd.

Put

Q=Pk,E=FP2k.Q=P^k,\qquad E=\mathbb{F}_{P^{2k}}.

Since kk is odd, K=FQK=\mathbb{F}_{Q} does not contain FP2\mathbb{F}_{P^2}. Therefore GbG_b remains irreducible over KK.

We may identify the two-dimensional KK-vector space K2K^2 with EE so that FkF^k acts as multiplication by α\alpha.

Let jj be the unique even integer satisfying

0≤j<2k,j≡r(modk),0 \le j < 2k,\qquad j \equiv r \pmod{k},

and put

ρ(z)=zPj.\rho(z)=z^{P^j}.

Then ρ∣K=τ\rho|_K=\tau, and ρ\rho fixes FP2\mathbb{F}_{P_2} pointwise. Moreover,

gcd⁡(j/2,k)=1,\gcd(j/2,k)=1,

so ρ\rho generates Gal⁡(E/FP2)\operatorname{Gal}(E/\mathbb{F}_{P_2}).

Because FF commutes with FkF^k, it commutes with multiplication by α\alpha. As E=K(α)E=K(\alpha), the map FF is ρ\rho-semilinear on the one-dimensional EE-vector space EE. Consequently,

F(z)=cρ(z)(9)F(z)=c\rho(z) \tag*{(9)}

for some c∈E×c\in E^\times.

Since ρ\rho generates Gal⁡(E/FP2)\operatorname{Gal}(E/\mathbb{F}_{P_2}), iteration gives

Fk(z)=NE/FP2(c)z.F^k(z)=\mathrm{N}_{E/\mathbb{F}_{P_2}}(c)z.

Thus the condition Fk(z)=αzF^k(z)=\alpha z for every z∈Ez\in E is equivalent to

NE/FP2(c)=α.\mathrm{N}_{E/\mathbb{F}_{P_2}}(c)=\alpha.

We next express the determinant-one condition in these coordinates, before verifying that both conditions on cc can be satisfied simultaneously. Choose θ∈FP2×\theta\in\mathbb{F}_{P_2}^{\times} such that θP=−θ\theta^P=-\theta; in characteristic two, take θ=1\theta=1. Define

ωθ(z,w)=θ(zwQ−zQw).\omega_\theta(z,w)=\theta(zw^Q-z^Qw).

This is a nondegenerate, alternating, KK-valued bilinear form on EE. Indeed, θQ=−θ\theta^Q=-\theta, since kk is odd, and therefore

ωθ(z,w)Q=ωθ(z,w).\omega_\theta(z,w)^Q=\omega_\theta(z,w).

Let ϕ:E⟶K2\phi:E\longrightarrow K^2 be an identification for which ϕ−1Fkϕ\phi^{-1}F^k\phi is multiplication by α\alpha. The space of alternating KK-bilinear forms on the two-dimensional space EE is one-dimensional. Since both forms are nondegenerate, there is λ∈K×\lambda\in K^\times such that

det⁡(ϕ(z),ϕ(w))=λωθ(z,w).\det(\phi(z),\phi(w))=\lambda\omega_\theta(z,w).

Multiplication by u∈E×u\in E^\times satisfies

ωθ(uz,uw)=uQ+1ωθ(z,w)=NE/K(u)ωθ(z,w).\omega_\theta(uz,uw)=u^{Q+1}\omega_\theta(z,w)=\mathrm{N}_{E/K}(u)\omega_\theta(z,w).

Choose uu with NE/K(u)=λ−1\mathrm{N}_{E/K}(u)=\lambda^{-1}, which is possible by surjectivity of the norm, and replace ϕ(z)\phi(z) by ϕ(uz)\phi(uz). The transported determinant form is then exactly ωθ\omega_\theta. This change preserves the description of FkF^k as multiplication by α\alpha, since multiplication by uu commutes with multiplication by α\alpha. In these normalized coordinates, write F=cρF=c\rho again.

As ρ(θ)=θ\rho(\theta)=\theta and ρ\rho commutes with the QQ-power Frobenius, we have

ωθ(Fz,Fw)=cQ+1ρ(ωθ(z,w)).\omega_\theta(Fz,Fw)=c^{Q+1}\rho(\omega_\theta(z,w)).

Thus FF has determinant one if and only if

NE/K(c)=cQ+1=1.\mathrm{N}_{E/K}(c)=c^{Q+1}=1.

We now verify that both norm conditions can be satisfied simultaneously.

Let

U=ker⁡(NE/K).U=\ker(\mathrm{N}_{E/K}).

This is a cyclic group of order Q+1Q+1. Set

A=Q+1P+1,B=Q−1P−1.A=\frac{Q+1}{P+1},\qquad B=\frac{Q-1}{P-1}.

The exponent of NE/FP2\mathrm{N}_{E/\mathbb{F}_{P^2}} is

P2k−1P2−1=AB.\frac{P^{2k}-1}{P^2-1}=AB.

Since kk is odd,

B=1+P+⋯+Pk−1≡1(modP+1).B=1+P+\cdots+P^{k-1}\equiv1\pmod{P+1}.

Hence

gcd⁡(B,P+1)=1,gcd⁡(Q+1,AB)=A.\gcd(B,P+1)=1,\qquad\gcd(Q+1,AB)=A.

It follows that the image of the norm map on UU has order

Q+1A=P+1.\frac{Q+1}{A}=P+1.

This image is contained in μP+1⊂FP2×\mu_{P+1}\subset\mathbb{F}_{P^2}^{\times}. Therefore

NE/FP2(U)=μP+1.\mathrm{N}_{E/\mathbb{F}_{P^2}}(U)=\mu_{P+1}.

In particular, there exists c∈Uc\in U with

NE/FP2(c)=α.\mathrm{N}_{E/\mathbb{F}_{P^2}}(c)=\alpha.

This proves existence.

Suppose now that Fc=cρF_c=c\rho and Fc′=c′ρF_{c'}=c'\rho have the same kk-th power. Then

NE/FP2(c′/c)=1.\mathrm{N}_{E/\mathbb{F}_{P^2}}(c'/c)=1.

Hilbert’s Theorem 90 gives u∈E×u\in E^\times such that

ρ(u)u=cc′.\frac{\rho(u)}{u}=\frac{c}{c'}.

Multiplication by uu intertwines FcF_c and Fc′F_{c'}, proving GL2(K)\mathrm{GL}_2(K)-conjugacy.

Any self-intertwiner commutes with multiplication by α\alpha. Since E=K(α)E=K(\alpha), it is therefore EE-linear and hence is multiplication by some u∈Eu\in E. Commutation with cρc\rho then requires ρ(u)=u\rho(u)=u. Consequently, the invertible self-intertwiners are exactly the multiplications by

u∈Eρ,×=FP2×.u\in E^{\rho,\times}=\mathbb{F}_{P^2}^{\times}.

Their determinants as KK-linear maps are

NE/K(u)=uQ+1=uP+1,\mathrm{N}_{E/K}(u)=u^{Q+1}=u^{P+1},

where the last equality follows because kk is odd. Thus, also in this case, the determinant map on the self-intertwiner group is the surjective norm

NFP2/FP.\mathrm{N}_{\mathbb{F}_{P^2}/\mathbb{F}_P}.

Passage to SL2(K)\mathrm{SL}_2(K)

In either case, let g∈GL2(K)g\in\mathrm{GL}_2(K) intertwine two determinant-one semilinear maps. Taking determinants in the intertwining identity gives

τ(det⁡g)=det⁡g.(10)\tau(\det g)=\det g. \tag*{(10)}

Since the fixed field of τ\tau is FP\mathbb{F}_P, we have

det⁡g∈FP×.\det g \in\mathbb{F}_P^{\times}.

More explicitly, let F,F′∈Fk,r(b)F,F' \in\mathcal{F}_{k,r}(b) and choose g∈GL2(K)g \in\mathrm{GL}_2(K) with gF=F′ggF=F'g, as established above. Put δ=det⁡g∈FP×\delta=\det g \in\mathbb{F}_P^{\times}. In either parity, the determinant map on the self-intertwiner group of FF is surjective onto FP×\mathbb{F}_P^{\times}. We may therefore choose a KK-linear self-intertwiner uu of FF with det⁡u=δ−1\det u=\delta^{-1}. Then

(gu)F=g(Fu)=F′(gu),det⁡(gu)=1.(gu)F=g(Fu)=F'(gu), \qquad\det(gu)=1.

Thus any two elements of Fk,r(b)\mathcal{F}_{k,r}(b) are SL2(K)\mathrm{SL}_2(K)-conjugate. Since the constructions above also prove that this set is nonempty, it is a single SL2(K)\mathrm{SL}_2(K)-conjugacy class.

In either parity, the self-intertwiner group is isomorphic to FP2×\mathbb{F}_{P^2}^{\times}, and its determinant map is the norm to FP×\mathbb{F}_P^{\times}. Consequently, its intersection with SL2(K)\mathrm{SL}_2(K) has order

P2−1P−1=P+1.\frac{P^2-1}{P-1}=P+1.

From norm fibers to cyclic vectors

For F∈Fk,r(b)F \in\mathcal{F}_{k,r}(b), put

C(F)={v∈K2:ω(v,Fv)=1}.C(F)=\{v \in K^2:\omega(v,Fv)=1\}.

We claim that

wk,r(b)=#C(F)P+1.(11)w_{k,r}(b)=\frac{\#C(F)}{P+1}. \tag*{(11)}

Proof. Let F∈Fk,r(b)F \in\mathcal{F}_{k,r}(b). If v∈C(F)v \in C(F), then (v,Fv)(v,Fv) is a determinant-one basis of K2K^2. Since FF is τ\tau-semilinear and has determinant one, there is a unique a∈Ka \in K such that

F2v=−v−aFv.F^2v=-v-aFv.

Indeed, write F2v=λv+μFvF^2v=\lambda v+\mu Fv. The determinant-one condition gives

1=ω(Fv,F2v)=−λ,1=\omega(Fv,F^2v)=-\lambda,

so λ=−1\lambda=-1; setting a=−μa=-\mu gives the assertion. Thus, relative to the basis (v,Fv)(v,Fv), the map FF has coefficient matrix

Aa=(0−11−a).A_a= \begin{pmatrix} 0 & -1\\ 1 & -a \end{pmatrix}.

The KK-linear map FkF^k has characteristic polynomial

Z2+br(a)Z+1.Z^2+b_r(a)Z+1.

Since this polynomial is GbG_b, we obtain br(a)=bb_r(a)=b.

Conversely, if br(a)=bb_r(a)=b, the semilinear map AaτA_a\tau belongs to Fk,r(b)\mathcal{F}_{k,r}(b) and hence is SL2(K)\mathrm{SL}_2(K)-conjugate to FF.

The determinant-one bases that identify FF with AaτA_a\tau form a torsor under the centralizer of FF in SL2(K)\mathrm{SL}_2(K). More explicitly, if B0B_0 is one such basis matrix, then all the others are uniquely of the form gB0gB_0 with g∈ZSL2(K)(F)g \in Z_{\mathrm{SL}_2(K)}(F), where ZSL2(K)(F)Z_{\mathrm{SL}_2(K)}(F) denotes this centralizer. There are therefore exactly P+1P+1 such bases.

Each such basis is uniquely determined by its first vector vv, since the second vector is FvFv.

It follows that

#C(F)=(P+1)wk,r(b),\#C(F)=(P+1)w_{k,r}(b),

which proves (11).

Comparison with the standard Frobenius

Proof of Proposition 5.1. By Lemma 5.2, it suffices to compare the cyclic-vector counts for FF and a standard-Frobenius map F0F_0 such that both FkF^k and F0kF_0^k have characteristic polynomial GbG_b.

Case 1: k=2sk=2s is even.

Use the normal form (8):

F(u,v)=(cvP,−c−1uP).F(u,v)=(cv^P,-c^{-1}u^P).

Define the standard-Frobenius semilinear map

F0(u,v)=(cvP,−c−1uP).F_0(u,v)=(cv^P,-c^{-1}u^P).

Since rr is odd, the action of τ\tau on FP2\mathbb{F}_{P^2} is the PP-power Frobenius. Consequently, both FkF^k and F0kF_0^k have eigenvalues

(−1)sN1−P,(−1)sNP−1,N=NK/FP2(c).(-1)^sN^{1-P},\qquad(-1)^sN^{P-1},\qquad N=N_{K/\mathbb{F}_{P^2}}(c).

Thus F∈Fk,r(b)F\in\mathcal{F}_{k,r}(b) and F0∈Fk,1(b)F_0\in\mathcal{F}_{k,1}(b).

Their cyclic-vector equations are respectively

−c−1uPr+1−cvPr+1=1-c^{-1}u^{P^r+1}-cv^{P^r+1}=1

and

−c−1uP+1−cvP+1=1.-c^{-1}u^{P+1}-cv^{P+1}=1.

Since rr is odd and gcd⁡(r,k)=1\gcd(r,k)=1, the standard gcd identity for powers gives

gcd⁡(Pr+1,Pk−1)=P+1.\gcd(P^r+1,P^k-1)=P+1.

The multiplicative group K×K^\times is cyclic. Therefore the power maps

x↦xPr+1,x↦xP+1x\mapsto x^{P^r+1},\qquad x\mapsto x^{P+1}

have the same kernel size, namely P+1P+1, and hence the same image. Moreover, every nonzero element of their common image has exactly P+1P+1 preimages under either map, while zero has exactly one preimage. It follows that the two cyclic-vector equations have the same number of solutions.

Case 2: kk is odd.

We identify the two-dimensional KK-vector space K2K^2 with the quadratic extension E/KE/K. We retain the notation

Q=Pk,E=FP2k,j≡r(modk),j even,0≤j<2k.Q=P^k,\qquad E=\mathbb{F}_{P^{2k}},\qquad j\equiv r\pmod{k},\qquad j\text{ even},\qquad0\leq j<2k.

Use the normal form (9), with the normalized alternating form ωθ\omega_\theta:

F(z)=czPj,NE/K(c)=1,NE/FP2(c)=α.F(z)=cz^{P^j},\qquad N_{E/K}(c)=1,\qquad N_{E/\mathbb{F}_{P^2}}(c)=\alpha.

Define

ρ0(z)=zPk+1,F0(z)=cρ0(z).\rho_0(z)=z^{P^{k+1}},\qquad F_0(z)=c\rho_0(z).

Because kk is odd, k+1k+1 is even and

gcd⁡(k+1,2k)=2.\gcd(k+1,2k)=2.

Thus ρ0\rho_0 generates Gal⁡(E/FP2)\operatorname{Gal}(E/\mathbb{F}_{P^2}), and its restriction to KK is the standard PP-power Frobenius.

Both FkF^{k} and F0kF_{0}^{k} are multiplication by α\alpha. Furthermore, both maps have determinant one with respect to ωθ\omega_{\theta}, because NE/K(c)=1N_{E/K}(c)=1.

Put

l=j+k.l=j+k.

Since jj is even and kk is odd, ll is odd. Also

gcd⁡(l,k)=gcd⁡(j,k)=1,\gcd(l,k)=\gcd(j,k)=1,

so

gcd⁡(l,2k)=1.\gcd(l,2k)=1.

For z∈Ez\in E, direct calculation gives

ωθ(z,Fz)=θ(cQzPl+1+Pj+k−czPk+Pj)=θ(cQs−csQ),\omega_{\theta}(z,Fz)=\theta\left(c^{Q}z^{P^{l+1}+P^{j+k}}-cz^{P^{k}+P^{j}}\right) =\theta(c^{Q}s-cs^{Q}),

where

s=zPl+1.s=z^{P^{l+1}}.

Here we used

(zPl+1)Q=zPj+Pk,\left(z^{P^{l+1}}\right)^{Q}=z^{P^{j}+P^{k}},

which holds because zP2k=zz^{P^{2k}}=z.

For F0F_{0}, the same calculation gives

ωθ(z,F0z)=θ(cQs0−cs0Q),s0=zP+1.\omega_{\theta}(z,F_{0}z)=\theta(c^{Q}s_{0}-cs_{0}^{Q}),\qquad s_{0}=z^{P+1}.

Since ll is odd and gcd⁡(l,2k)=1\gcd(l,2k)=1, we have

gcd⁡(Pl+1,P2k−1)=P+1.\gcd(P^{l}+1,P^{2k}-1)=P+1.

Consequently, the power maps

z⟼zPl+1,z⟼zP+1z\longmapsto z^{P^{l+1}},\qquad z\longmapsto z^{P+1}

on E×E^{\times} have the same image and the same kernel size P+1P+1. As in the even case, zero has a unique preimage under either power map. Thus, for every s∈Es\in E, the numbers of solutions of

zPl+1=sandzP+1=sz^{P^{l+1}}=s\qquad\text{and}\qquad z^{P+1}=s

are equal.

Since the cyclic-vector conditions for FF and F0F_{0} are respectively

θ(cQs−csQ)=1\theta(c^{Q}s-cs^{Q})=1

with s=zPl+1s=z^{P^{l+1}} and s=zP+1s=z^{P+1}, their numbers of solutions coincide.

In both cases, the cyclic-vector count for the generator τ\tau agrees with that for the standard Frobenius.

By Lemma 5.2,

wk,r(b)=wk,1(b).w_{k,r}(b)=w_{k,1}(b).

Proposition 4.1 gives

wk,1(b)=Dk(P),w_{k,1}(b)=D_{k}(P),

and the assertion follows.

Remark 5.3. The fiber counts satisfy ∑b∈FPwk,r(b)=Pk\sum_{b\in\mathbb{F}_P} w_{k,r}(b)=P^k, since each a∈Ka\in K has exactly one norm coefficient. There are (P−ϵP)/2(P-\epsilon_P)/2 irreducible quadratics Z2+bZ+1Z^2+bZ+1, where ϵP\epsilon_P is zero for even PP and one for odd PP. Consequently, the total contribution of irreducible norm values is (P−ϵP)Dk(P)/2(P-\epsilon_P)D_k(P)/2, and that of reducible norm values is Pk−(P−ϵP)Dk(P)/2P^k-(P-\epsilon_P)D_k(P)/2. The proposition concerns the individual fibers over irreducible norm values.

Arithmetic count and conclusion

We now count the admissible norm coefficients. These satisfy the three conditions of Lemma 3.2. We do this by counting roots α\alpha in the norm-one group

{α∈FP2×:αP+1=1}.\left\{\alpha\in\mathbb{F}_{P^2}^{\times}:\alpha^{P+1}=1\right\}.

The binomial irreducibility criterion imposes conditions on the prime parts of ord⁡(α)\operatorname{ord}(\alpha), and Möbius inversion selects the elements of exact degree 2d2d over Fp\mathbb{F}_p. Finally, we divide by two because α\alpha and α−1\alpha^{-1} give the same coefficient b=−(α+α−1)b=-(\alpha+\alpha^{-1}). This yields Bd,m(p)B_{d,m}(p); multiplying by the common fiber size from § 5 completes the proof of the main theorem.

Lemma 6.1. The number of b∈Fpdb\in\mathbb{F}_{p^d} satisfying the three conditions of Lemma 3.2 is Bd,m(p)B_{d,m}(p) of (3).

Proof. The first two conditions are equivalent to choosing

αP+1=1,[Fp(α):Fp]=2d,\alpha^{P+1}=1,\qquad[\mathbb{F}_p(\alpha):\mathbb{F}_p]=2d,

and setting b=−(α+α−1)b=-(\alpha+\alpha^{-1}). Each bb corresponds to exactly two such elements, α\alpha and α−1\alpha^{-1}. They are distinct, since an element equal to its inverse has degree one over Fp\mathbb{F}_p.

If mm is even, α\alpha is a square in FP2\mathbb{F}_{P^2}: this is automatic in characteristic two, while for odd PP one has α(P2−1)/2=1\alpha^{(P^2-1)/2}=1. Writing α=β2\alpha=\beta^2, we obtain

Xm−α=(Xm/2−β)(Xm/2+β),X^m-\alpha=(X^{m/2}-\beta)(X^{m/2}+\beta),

so Xm−αX^m-\alpha is reducible. This proves the assertion in this case.

Suppose m>1m>1 is odd. The finite-field binomial criterion [14 Theorem 3.75] says that Xm−αX^m-\alpha is irreducible if and only if every prime ℓ∣m\ell\mid m divides ord⁡(α)\operatorname{ord}(\alpha) but does not divide (P2−1)/ord⁡(α)(P^2-1)/\operatorname{ord}(\alpha); the additional condition for multiples of four is irrelevant here. Since ord⁡(α)∣P+1\operatorname{ord}(\alpha)\mid P+1 and gcd⁡(P−1,P+1)∣2\gcd(P-1,P+1)\mid2, this is exactly

0<vℓ(ord⁡(α))=vℓ(P+1)(ℓ∣m).(12)0<v_\ell(\operatorname{ord}(\alpha))=v_\ell(P+1)\qquad(\ell\mid m). \tag*{(12)}

For m=1m=1, the binomial is linear and this condition is empty. In particular, R∤P+1R\nmid P+1 implies that there are no such elements.

Assume R∣P+1R\mid P+1 and let LL be as in (3). For each e∣2de\mid2d, the group

{α∈Fp2d×:αP+1=1}∩Fpe×\left\{\alpha\in\mathbb{F}_{p^{2d}}^\times:\alpha^{P+1}=1\right\}\cap\mathbb{F}_{p^e}^\times

is cyclic of order ge=gcd⁡(P+1,pe−1)g_e=\gcd(P+1,p^e-1). It contains no elements satisfying (12) unless L∣geL\mid g_e.

Suppose L∣geL\mid g_e. For each prime ℓ∣R\ell\mid R, the ℓ\ell-primary component of this cyclic group has order ℓvℓ(P+1)\ell^{v_\ell(P+1)}. The proportion of elements whose orders have maximal ℓ\ell-part is 1−1/ℓ1-1/\ell. The conditions for distinct primes are independent under the decomposition into primary components. Hence the number of elements satisfying (12) is

ge∏ℓ∣R(1−ℓ−1)=geφ(R)R.g_e\prod_{\ell\mid R}(1-\ell^{-1})=g_e\frac{\varphi(R)}{R}.

If f=[Fp(α):Fp]f=[\mathbb{F}_p(\alpha):\mathbb{F}_p], then α∈Fpe\alpha\in\mathbb{F}_{p^e} if and only if f∣ef\mid e. For α∈Fp2d\alpha\in\mathbb{F}_{p^{2d}}, the indicator that its degree over Fp\mathbb{F}_p is exactly 2d2d equals

∑e∣2dμ(2d/e)1α∈Fpe.\sum_{e\mid2d}\mu(2d/e)1_{\alpha\in\mathbb{F}_{p^e}}.

Summing this indicator over the elements satisfying (12), and dividing by two for inversion, gives

Bd,m(p)=φ(R)2R∑e∣2dL∣geμ(2d/e)ge.B_{d,m}(p)=\frac{\varphi(R)}{2R}\sum_{\substack{e\mid2d\\L\mid g_e}}\mu(2d/e)g_e.

as required. □

Proof of Theorem 2.1. By Lemmas 3.1 and 3.2, the admissible norm coefficients are precisely the Bd,m(p)B_{d,m}(p) elements counted in Lemma 6.1. On K=FPHK=\mathbb{F}_{P^H} the automorphism σn\sigma^n is c↦cpmc\mapsto c^{p^m}, with gcd⁡(m,H)=1\gcd(m,H)=1. Proposition 5.1, applied with k=Hk=H and r=mr=m, shows that each admissible coefficient has exactly DH(P)D_H(P) preimages. Multiplication yields the asserted formula. □

Specializations

Recall that ϵp=0\epsilon_p=0 for even pp and ϵp=1\epsilon_p=1 for odd pp. For m=1m=1 the arithmetic count simplifies to

Cd(p):=Bd,1(p)=12∑e∣de oddμ(e)(pd/e−ϵp).(13)C_d(p):=B_{d,1}(p)=\frac{1}{2}\sum_{\substack{e\mid d\\e\ \mathrm{odd}}}\mu(e)(p^{d/e}-\epsilon_p). \tag*{(13)}

One direct derivation is as follows. Put

Uj={c∈Fp2j×:cpj+1=1}.U_j=\left\{c\in\mathbb{F}_{p^{2j}}^\times:c^{p^j+1}=1\right\}.

The coefficients of irreducible quadratics X2+bX+1X^2+bX+1 over Fpj\mathbb{F}_{p^j} are precisely the values b=−(c+c−1)b=-(c+c^{-1}) with c∈Ujc\in U_j and c≠c−1c\ne c^{-1}, each obtained from exactly two such elements. The group UjU_j has pj+1p^j+1 elements, of which 1+ϵp1+\epsilon_p satisfy c=c−1c=c^{-1}. Hence the number of these quadratics is (pj−ϵp)/2(p^j-\epsilon_p)/2. An irreducible quadratic over a subfield remains irreducible precisely under an extension of odd degree. Partitioning coefficients by their exact degree gives (pj−ϵp)/2=∑e∣j, j/e oddCe(p)(p^j-\epsilon_p)/2=\sum_{e\mid j,\ j/e\ \mathrm{odd}}C_e(p); Möbius inversion gives (13).

Let Sp(2d)S_p(2d) denote the number of monic self-reciprocal irreducible polynomials of degree 2d2d over Fp\mathbb{F}_p. Then

Cd(p)=dSp(2d).(14)C_d(p)=dS_p(2d). \tag*{(14)}

Indeed, the admissible coefficients counted by Cd(p)C_d(p) correspond to the inverse pairs {α,α−1}\{\alpha,\alpha^{-1}\} satisfying

αpd+1=1,[Fp(α):Fp]=2d,\alpha^{p^d+1}=1,\qquad[\mathbb{F}_p(\alpha):\mathbb{F}_p]=2d,

via b=−(α+α−1)b=-(\alpha+\alpha^{-1}). The minimal polynomial of such an α\alpha is self-reciprocal, since α−1=αpd\alpha^{-1}=\alpha^{p^d} is a Frobenius conjugate. Conversely, if α\alpha is a root of a self-reciprocal irreducible polynomial of degree 2d2d, inversion permutes its Frobenius orbit and is a nontrivial involution commuting with Frobenius. It must therefore act on that orbit as the dd-th power of Frobenius, so α−1=αpd\alpha^{-1}=\alpha^{p^d}. Each such polynomial has 2d2d distinct roots, grouped into dd inverse pairs. Distinct inverse pairs give distinct values of bb, since their elements are precisely the roots of X2+bX+1X^2+bX+1. Thus each polynomial contributes exactly dd coefficients to Cd(p)C_d(p), proving (14). Combining (13) and (14) recovers the classical enumeration of monic self-reciprocal irreducible polynomials; see [13 Theorem 3].

Corollary 7.1. For all n,k≥1n,k \ge1,

Nn(p,nk)=Cn(p)Dk(pn)=nSp(2n)Dk(pn).N_n(p,nk)=C_n(p)D_k(p^n)=nS_p(2n)D_k(p^n).

If nn is a power of two, then

Nn(p,h)={0,n∤h,pn−ϵp2Dh/n(pn),n∣h.N_n(p,h)=\begin{cases} 0, & n\nmid h, \\ \dfrac{p^n-\epsilon_p}{2}D_{h/n}(p^n), & n\mid h. \end{cases}

In particular, this includes the counting formulas for n=2n=2 and n=4n=4.

Proof. If h=nkh=nk, then

d=gcd⁡(n,h)=n,m=1,H=k,P=pn.d=\gcd(n,h)=n,\qquad m=1,\qquad H=k,\qquad P=p^n.

Theorem 2.1 therefore gives

Nn(p,nk)=Dk(pn)Bn,1(p)=Dk(pn)Cn(p).N_n(p,nk)=D_k(p^n)B_{n,1}(p)=D_k(p^n)C_n(p).

Equation (14) gives the expression in terms of Sp(2n)S_p(2n). For a power of two, n∤hn\nmid h implies that m=n/gcd⁡(n,h)m=n/\gcd(n,h) is even, so the count is zero by Theorem 2.1. If n∣hn\mid h, the sum in (13) for Cn(p)C_n(p) has only the odd divisor e=1e=1, which gives the stated formula. □

Corollary 7.2. For n=3n=3,

N3(p,h)={0,3∤h, p≡0,1(mod3),p+13Dh(p),3∤h, p≡2(mod3),p3−p2Dh/3(p3),3∣h.N_3(p,h)=\begin{cases} 0, & 3\nmid h,\ p\equiv0,1\pmod3,\\ \dfrac{p+1}{3}D_h(p), & 3\nmid h,\ p\equiv2\pmod3,\\ \dfrac{p^3-p}{2}D_{h/3}(p^3), & 3\mid h. \end{cases}

Proof. If 3∤h3\nmid h, then d=1d=1, m=3m=3. The arithmetic count vanishes unless 3∣p+13\mid p+1. In that case the norm-one group has p+1p+1 elements, and precisely 2(p+1)/32(p+1)/3 have maximal 33-part in their order. These elements have degree two, since the elements of the norm-one group in Fp\mathbb{F}_p have order at most two. Dividing by two gives B1,3(p)=(p+1)/3B_{1,3}(p)=(p+1)/3. If 3∣h3\mid h, use C3(p)=(p3−p)/2C_3(p)=(p^3-p)/2. □

Remark 7.3. For an example with n∤hn\nmid h, take (p,n,h)=(2,3,5)(p,n,h)=(2,3,5). Then d=1d=1, m=3m=3, H=5H=5, and

B1,3(2)=1,D5(2)=25+13=11,N3(2,5)=11.B_{1,3}(2)=1,\qquad D_5(2)=\frac{2^5+1}{3}=11,\qquad N_3(2,5)=11.

The automorphism in the quadratic reduction is c↦c23c\mapsto c^{2^3} on F25\mathbb{F}_{2^5}, a nonstandard Frobenius generator. This illustrates the use of § 5.

Declarations

Competing interests

Competing interests. The authors declare no competing interests.

Data availability

No external datasets were used.

Declaration of generative AI and AI-assisted technologies in the manuscript preparation process

Before using generative AI in this work, the authors had proved the counting formula for n=2n=2 by a different method and had formulated conjectures for the cases n=3n=3 and n=4n=4 based on computations in Python. These computations were performed on the computing cluster at RIKEN iTHEMS. They subsequently used OpenAI’s Codex with the GPT-6.1 Sol model to help formulate the conjectured counting formula for general nn, explore proof strategies, develop proof arguments, write code for computational checks, and assist with drafting and revising the manuscript. The authors reviewed and edited all AI-assisted material and take full responsibility for the mathematical content, references, and final presentation.

(Tomoya Hayata) Keio University School of Medicine, 4-1-1 Hiyoshi, Kohoku-ku, Yokohama, 223-8521, Japan

(Tomoya Hayata) RIKEN Center for Interdisciplinary Theoretical and Mathematical Sciences (iTHEMS), RIKEN, Wako 351-0198, Japan

Email address: [email protected]

(Takahiro Tsushima) Keio University School of Medicine, 4-1-1 Hiyoshi, Kohoku-ku, Yokohama, 223-8521, Japan

Email address: [email protected]

References

  1. [1]C. Bonnafé, Representations of SL₂(F_q), Algebra and Applications 13, Springer, 2011, Chapter 2. https://doi.org/10.1007/978-0-85729-157-8_2.
  2. [2]A. W. Bluher, On x^{q+1} + ax + b, Finite Fields Appl. 10 (2004), 285–305. https://doi.org/10.1016/j.ffa.2003.08.004.
  3. [3]O. Ore, Theory of non-commutative polynomials, Ann. of Math. (2) 34 (1933), no. 3, 480–508. https://doi.org/10.2307/1968173.
  4. [4]X. Caruso and J. Le Borgne, A new faster algorithm for factoring skew polynomials over finite fields, J. Symbolic Comput. 79 (2017), 411–443. https://xavier.caruso.ovh/papers/publis/fastfactor.pdf.
  5. [5]S. Plumpton and D. Thompson, The norm of a skew polynomial, Algebr. Represent. Theory 25 (2022), 869–887. https://doi.org/10.1007/s10468-021-10051-z.
  6. [6]R. C. Churchill and Y. Zhang, Irreducibility criteria for skew polynomials, J. Algebra 322 (2009), no. 11, 3797–3822. https://doi.org/10.1016/j.jalgebra.2009.08.012.
  7. [7]R. W. K. Odoni, On additive polynomials over a finite field, Proc. Edinburgh Math. Soc. (2) 42 (1999), no. 1, 1–16. https://doi.org/10.1017/S0013091500019970.
  8. [8]M. Giesbrecht, Factoring in skew-polynomial rings over finite fields, J. Symbolic Comput. 26 (1998), no. 4, 463–486. https://doi.org/10.1006/jsco.1998.0224.
  9. [9]D. Boucher, W. Geiselmann and F. Ulmer, Skew-cyclic codes, Appl. Algebra Engrg. Comm. Comput. 18 (2007), 379–389. https://doi.org/10.1007/s00200-007-0043-z.
  10. [10]H. Gluesing-Luerssen, Introduction to skew-polynomial rings and skew-cyclic codes, in Concise Encyclopedia of Coding Theory, CRC Press, Boca Raton, FL, 2021, pp. 151–180. Preprint available at https://arxiv.org/abs/1902.03516.
  11. [11]T. Tsushima, The Distribution of x^{pⁿ⁻¹} + x^{pⁿ⁻¹} over Finite Fields and Self-Adjoint Irreducible Polynomials, preprint, version 2, Zenodo, 2026. https://doi.org/10.5281/zenodo.2387833.
  12. [12]T. Tsushima, Local Galois representations associated to additive polynomials, Manuscripta Math. 174 (2024), no. 3–4, 1151–1182. https://doi.org/10.1007/s00229-024-01550-6.
  13. [13]H. Meyn and W. Götz, Self-reciprocal polynomials over finite fields, Sém. Lothar. Combin. 21 (1989), Art. B21d, 8 pp. https://www.mat.univie.ac.at/~slc/s/s21meyn.html.
  14. [14]R. Lidl and H. Niederreiter, Finite Fields, second edition, Encyclopedia of Mathematics and its Applications 20, Cambridge University Press, Cambridge, 1997.
  15. [15]S. Harper, Totally deranged elements of almost simple groups and invariable generating sets, J. Lond. Math. Soc. (2) 109 (2024), no. 6, e12935. https://doi.org/10.1112/jlms.12935.

Paper details

vlnc:W_hk4z3rzy
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 6

downloads · 1

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Codex

View statement
Looking for related papers…
Contents