Introduction
Let P P P be an odd prime power and set Q = P H Q=P^H Q = P H . In the skew polynomial ring F Q [ t ; σ ] \mathbb{F}_Q[t;\sigma] F Q [ t ; σ ] , where t c = c P t tc=c^Pt t c = c P t , consider the one-parameter self-adjoint family
f a ( t ) = t 4 + t 3 + a t 2 + t + 1 , a ∈ F Q . f_a(t)=t^4+t^3+at^2+t+1,\qquad a\in\mathbb{F}_Q. f a ( t ) = t 4 + t 3 + a t 2 + t + 1 , a ∈ F Q .
The Laurent polynomial f a t − 2 f_at^{-2} f a t − 2 is fixed by the anti-involution that fixes coefficients and sends t t t to t − 1 t^{-1} t − 1 . We write A H ( P ) A_H(P) A H ( P ) for the number of parameters for which f a f_a f a is irreducible. Our aim is to determine these counts over extensions of the parameter field and to identify the geometric and representation-theoretic structures responsible for them.
The starting point is that the self-adjoint coefficient pattern gives the roots of the associated linearized polynomial a nondegenerate alternating pairing. Irreducibility in the Ore ring is then equivalent to irreducibility of Frobenius on this four-dimensional symplectic root space. Thus the counting problem selects particular Frobenius classes in a finite symplectic group. Expanding their indicator in irreducible characters turns the count over each extension field into a sum of cohomological traces.
The first result is a complete evaluation at P = 3 P=3 P = 3 . Theorem 2.4 gives A H ( 3 ) A_H(3) A H ( 3 ) for all H H H in terms of explicit polynomials N ( u ) N(u) N ( u ) and D ( u ) D(u) D ( u ) of degrees 12 and 10. The former is the Steinberg factor and the latter is the regular cuspidal factor. We compute the ramification of the root local system at infinity using the Weil local systems of Katz–Tiep [10 ] and local Fourier transformation [11 , 1 ] . The resulting conductor degrees and Frobenius duality give a finite system of equations for the coefficients. The counts through H = 11 H=11 H = 11 determine both factors, and the count at H = 12 H=12 H = 12 checks the reconstruction. Proposition 6.1 then proves that the minimal recurrence has order 24.
The second result gives a geometric interpretation of the Steinberg factor for P = p f P=p^f P = p f with p ∤ f p\nmid f p ∤ f . Theorem 9.2 constructs a smooth projective curve B P \mathcal{B}_P B P of genus P ( P + 1 ) / 2 P(P+1)/2 P ( P + 1 ) /2 from flags consisting of a root line in a Lagrangian root plane. Its zeta numerator is the Steinberg Frobenius polynomial S P ( u ) S_P(u) S P ( u ) . In particular, S 3 ( u ) = N ( u ) S_3(u)=N(u) S 3 ( u ) = N ( u ) identifies the numerator factor already computed in Part I with the zeta numerator of B 3 \mathcal{B}_3 B 3 . The construction of the flag curve uses the same monodromy and ramification calculation as Part I.
Key words and phrases. Finite fields, skew polynomials, irreducibility, local Fourier transform, Deligne–Lusztig characters, zeta functions.
With m = ( P 2 + 1 ) / 2 m=(P^2+1)/2 m = ( P 2 + 1 ) /2 and κ = ( P 2 − 1 ) / 8 \kappa=(P^2-1)/8 κ = ( P 2 − 1 ) /8 , the general counting formula is
m A H ( P ) = κ # B P ( F P H ) + ∑ λ λ H mA_H(P)=\kappa\#\mathcal{B}_P(\mathbb{F}_{P^H})+\sum_{\lambda}\lambda^H m A H ( P ) = κ # B P ( F P H ) + λ ∑ λ H
where λ \lambda λ runs through the inverse roots of the regular cuspidal factor C P ( u ) C_P(u) C P ( u ) , with multiplicity. The degrees of S P S_P S P and C P C_P C P are P 2 + P P^2+P P 2 + P and κ ( P 2 + P − 2 ) \kappa(P^2+P-2) κ ( P 2 + P − 2 ) , respectively. Part II gives explicit formulas for the first two coefficients of each factor, including in characteristics 3 and 5, and relates the full counting series to a quotient of covering-curve zeta functions.
Ore’s theory [13 ] and the reduced-norm methods of Caruso–Le Borgne [4 ] relate skew-polynomial factorization to semilinear algebra. The self-adjoint trinomials x 2 n + a x n + 1 x^{2n}+ax^n+1 x 2 n + a x n + 1 studied by Hayata and Tsushima [8 ] give rise to quadratic reduced norms. For the present quartic, the symplectic character expansion [5 ] instead isolates a Steinberg term and regular cuspidal terms. The appendix records the alternating product of flag-cover zeta functions associated with the Steinberg representation in higher rank.
Part I. Exact enumeration over extensions of F 3 \mathbb{F}_3 F 3
Let P P P be an odd prime power, put Q = P H Q=P^H Q = P H , and let R = F Q [ t ; σ ] R=\mathbb{F}_Q[t;\sigma] R = F Q [ t ; σ ] , where σ ( c ) = c P \sigma(c)=c^P σ ( c ) = c P and
t c = c P t ( c ∈ F Q ) . tc=c^Pt \qquad(c\in\mathbb{F}_Q). t c = c P t ( c ∈ F Q ) .
We write A H ( P ) A_H(P) A H ( P ) for the number of parameters a ∈ F Q a\in\mathbb{F}_Q a ∈ F Q for which
f a ( t ) = t 4 + t 3 + a t 2 + t + 1 f_a(t)=t^4+t^3+at^2+t+1 f a ( t ) = t 4 + t 3 + a t 2 + t + 1
is an atom of R R R . We first work with an arbitrary odd P P P , since the character identity underlying the enumeration is uniform in P P P ; we then specialize to P = 3 P=3 P = 3 . The units of R R R are exactly the nonzero constants. A nonconstant polynomial is an atom if it cannot be written as a product of two nonunits. This is irreducibility in the Ore ring, not in the commutative polynomial ring F Q [ t ] \mathbb{F}_Q[t] F Q [ t ] .
Companion modules and the root space
Companion modules and the root space. The quotient R / R f a R/Rf_a R / R f a , with R f a Rf_a R f a the left ideal generated by f a f_a f a , is a left R R R -module of dimension four over F Q \mathbb{F}_Q F Q . Left multiplication by t t t is σ \sigma σ -semilinear; in the basis 1 , t , t 2 , t 3 1,t,t^2,t^3 1 , t , t 2 , t 3 its companion matrix is
C a = ( 0 0 0 − 1 1 0 0 − 1 0 1 0 − a 0 0 1 − 1 ) . C_a=
\begin{pmatrix}
0&0&0&-1\\
1&0&0&-1\\
0&1&0&-a\\
0&0&1&-1
\end{pmatrix}. C a = 0 1 0 0 0 0 1 0 0 0 0 1 − 1 − 1 − a − 1 .
Right factors of f a f_a f a correspond to submodules of this cyclic left module. Consequently, f a f_a f a is an atom if and only if R / R f a R/Rf_a R / R f a is simple.
Consider also the linearized polynomial and its root space
L a ( X ) = X P 4 + X P 3 + a X P 2 + X P + X , V a = ker ( L a ) ⊂ F ‾ P . L_a(X)=X^{P^4}+X^{P^3}+aX^{P^2}+X^P+X,\qquad V_a=\ker(L_a)\subset\overline{\mathbb{F}}_P. L a ( X ) = X P 4 + X P 3 + a X P 2 + X P + X , V a = ker ( L a ) ⊂ F P .
Because the coefficient of X X X is nonzero, L a L_a L a is separable, and V a V_a V a is a four-dimensional F P \mathbb{F}_P F P -vector space. The linearization map ∑ i c i t i ↦ ∑ i c i X P i \sum_i c_i t^i\mapsto\sum_i c_iX^{P^i} ∑ i c i t i ↦ ∑ i c i X P i takes skew multiplication to composition. A monic proper right factor g g g of f a f_a f a has nonzero constant coefficient, so its linearization has a nonzero proper Frobenius-stable root space in V a V_a V a . Conversely, let W ⊂ V a W\subset V_a W ⊂ V a be a nonzero proper F Q \mathbb{F}_Q F Q -stable F P \mathbb{F}_P F P -subspace. Its monic root polynomial ∏ w ∈ W ′ ( X − w ) \prod_{w\in W'}(X-w) ∏ w ∈ W ′ ( X − w ) is linearized. Indeed, starting with X X X , adjoining a basis vector h h h to a subspace with root polynomial D W ′ ( X ) D_W'(X) D W ′ ( X ) replaces that polynomial by D W ′ ( X ) P − D W ′ ( h ) P − 1 D W ′ ( X ) D_W'(X)^P-D_W'(h)^{P-1}D_W'(X) D W ′ ( X ) P − D W ′ ( h ) P − 1 D W ′ ( X ) . Frobenius permutes its roots, so its coefficients lie in F Q \mathbb{F}_Q F Q . Let g g g be the associated monic skew polynomial. Right division f a = q g + r f_a=qg+r f a = q g + r gives deg r < deg g \deg r<\deg g deg r < deg g . The linearization of r r r vanishes on W ′ W' W ′ , which has P deg g P^{\deg g} P d e g g elements. Its degree if nonzero is at most P deg r P^{\deg r} P d e g r , so r = 0 r=0 r = 0 . Thus f a f_a f a is an atom exactly when F Q \mathbb{F}_Q F Q has no nonzero proper invariant subspace in V a V_a V a . For a linear operator this is equivalent to having irreducible characteristic polynomial, by viewing the space as an F P [ Z ] \mathbb{F}_{P}[Z] F P [ Z ] -module with Z Z Z acting by F Q F_{Q} F Q . Therefore
f a is an atom ⟺ det ( Z − F Q ∣ V a ) is irreducible over F P . (1) f_a\text{ is an atom}\quad\Longleftrightarrow\quad\det(Z-F_Q\mid V_a)\text{ is irreducible over }\mathbb{F}_P. \tag*{(1)} f a is an atom ⟺ det ( Z − F Q ∣ V a ) is irreducible over F P . (1)
Here and below, “irreducible” for the Frobenius action means irreducibility of its four-dimensional characteristic polynomial.
For clarity, put v ( x ) = ( x , x P , x P 2 , x P 3 ) t \mathbf{v}(x)=(x,x^{P},x^{P^2},x^{P^3})^{\mathrm{t}} v ( x ) = ( x , x P , x P 2 , x P 3 ) t . The equation L a ( x ) = 0 L_a(x)=0 L a ( x ) = 0 gives
v ( x P ) = C a t v ( x ) . \mathbf{v}(x^P)=C_a^{\mathrm{t}}\mathbf{v}(x). v ( x P ) = C a t v ( x ) .
Successive application of Frobenius therefore yields
v ( x Q ) = M a , H t v ( x ) , M a , H = C a C a P ⋯ C a P H − 1 . \mathbf{v}(x^Q)=M_{a,H}^{\mathrm{t}}\mathbf{v}(x),\qquad M_{a,H}=C_aC_a^P\cdots C_a^{P^{H-1}}. v ( x Q ) = M a , H t v ( x ) , M a , H = C a C a P ⋯ C a P H − 1 .
If x 1 , … , x 4 x_1,\ldots,x_4 x 1 , … , x 4 is an F P \mathbb{F}_P F P -basis of V a V_a V a , the matrix with columns v ( x i ) \mathbf{v}(x_i) v ( x i ) is a nonsingular Moore matrix. The preceding identity shows that the matrix of F Q F_Q F Q on V a V_a V a is similar over F P \mathbb{F}_P F P to M a , H t M_{a,H}^{\mathrm{t}} M a , H t . In particular, the two operators have the same characteristic polynomial in F P [ Z ] \mathbb{F}_P[Z] F P [ Z ] . This gives the concrete companion-matrix test for (1) .
The symplectic structure
A direct computation gives
C a t J C a = J , J = ( 0 0 1 − 1 0 0 0 1 − 1 0 0 0 1 − 1 0 0 ) , det J = 1. C_a^{\mathrm{t}}JC_a=J,\qquad J=\begin{pmatrix}0&0&1&-1\\0&0&0&1\\-1&0&0&0\\1&-1&0&0\end{pmatrix},\qquad\det J=1. C a t J C a = J , J = 0 0 − 1 1 0 0 0 − 1 1 0 0 0 − 1 1 0 0 , det J = 1.
This is a common alternating form for the companion matrices. For the root-space realization, it is convenient to use the following (equivalent) symplectic structure:
B ( x , y ) = x P 2 y − x y P 2 + x P 2 y P − x P y P 2 + x P 3 y P − x P y P 3 . B(x,y)=x^{P^2}y-xy^{P^2}+x^{P^2}y^P-x^Py^{P^2}+x^{P^3}y^P-x^Py^{P^3}. B ( x , y ) = x P 2 y − x y P 2 + x P 2 y P − x P y P 2 + x P 3 y P − x P y P 3 .
Its defining identity is
B ( x , y ) P − B ( x , y ) = L a ( x ) y P 2 − x P 2 L a ( y ) . (2) B(x,y)^P-B(x,y)=L_a(x)y^{P^2}-x^{P^2}L_a(y). \tag*{(2)} B ( x , y ) P − B ( x , y ) = L a ( x ) y P 2 − x P 2 L a ( y ) . (2)
Thus B ( x , y ) ∈ F P B(x,y)\in\mathbb{F}_P B ( x , y ) ∈ F P for x , y ∈ V a x,y\in V_a x , y ∈ V a . In Moore coordinates it is represented by the alternating matrix
K = ( 0 0 − 1 0 0 0 − 1 − 1 1 1 0 0 0 1 0 0 ) , det K = 1. K=\begin{pmatrix}0&0&-1&0\\0&0&-1&-1\\1&1&0&0\\0&1&0&0\end{pmatrix},\qquad\det K=1. K = 0 0 1 0 0 0 1 1 − 1 − 1 0 0 0 − 1 0 0 , det K = 1.
Since the Moore matrix of any F P \mathbb{F}_P F P -basis of V a V_a V a is invertible, B B B is nondegenerate on V a V_a V a . Furthermore, B ( x Q , y Q ) = B ( x , y ) Q = B ( x , y ) B(x^Q,y^Q)=B(x,y)^Q=B(x,y) B ( x Q , y Q ) = B ( x , y ) Q = B ( x , y ) for x , y ∈ V a x,y\in V_a x , y ∈ V a . Hence the arithmetic Frobenius acts through Sp ( V a , B ) ≃ Sp 4 ( F P ) \operatorname{Sp}(V_a,B)\simeq\operatorname{Sp}_4(\mathbb{F}_P) Sp ( V a , B ) ≃ Sp 4 ( F P ) .
As a a a varies, the root spaces form a rank-four étale F P \mathbb{F}_P F P -local system V \mathcal{V} V on A F P 1 \mathbb{A}^1_{\mathbb{F}_P} A F P 1 , and B B B defines a perfect alternating pairing on it. Fix a geometric generic point η ˉ \bar{\eta} η ˉ and a symplectic basis of V η ˉ \mathcal{V}_{\bar{\eta}} V η ˉ . We obtain a monodromy representation
ρ : π 1 ( A F P 1 , η ˉ ) ⟶ G : = Sp 4 ( F P ) . \rho:\pi_1(\mathbb{A}^1_{\mathbb{F}_P},\bar{\eta})\longrightarrow G:=\operatorname{Sp}_4(\mathbb{F}_P). ρ : π 1 ( A F P 1 , η ˉ ) ⟶ G := Sp 4 ( F P ) .
Changing the basis conjugates ρ \rho ρ , so its Frobenius conjugacy classes and all the character values below are well defined. For g ∈ G g\in G g ∈ G define 1 i r r ( g ) = 1 1_{\mathrm{irr}}(g)=1 1 irr ( g ) = 1 when its natural four-dimensional representation is irreducible, and 0 0 0 otherwise. By (1) ,
A H ( P ) = ∑ a ∈ F P H 1 i r r ( ρ ( Frob a ) ) . (3) A_H(P)=\sum_{a\in\mathbb{F}_{P^H}}1_{\mathrm{irr}}\bigl(\rho(\operatorname{Frob}_a)\bigr). \tag*{(3)} A H ( P ) = a ∈ F P H ∑ 1 irr ( ρ ( Frob a ) ) . (3)
with the arithmetic Frobenius convention. Since g g g is irreducible if and only if − g -g − g is irreducible, 1 i r r 1_{\mathrm{irr}} 1 irr descends to S = G / { ± I } = PSp 4 ( F P ) S=G/\{\pm I\}=\operatorname{PSp}_4(\mathbb{F}_P) S = G / { ± I } = PSp 4 ( F P ) .
The irreducibility indicator and a Coxeter-torus identity
Let T ⊂ G T \subset G T ⊂ G be a Coxeter torus. Concretely, on the natural four-dimensional module it is realized by the norm-one subgroup
T ≃ { z ∈ F P 4 × : z P 2 + 1 = 1 } , ∣ T ∣ = P 2 + 1. T \simeq\{z \in\mathbb{F}_{P^{4}}^{\times}: z^{P^{2}+1}=1\}, \qquad|T|=P^{2}+1. T ≃ { z ∈ F P 4 × : z P 2 + 1 = 1 } , ∣ T ∣ = P 2 + 1.
The Frobenius orbit of a noncentral element of T T T has length four: indeed, if z ∈ T z \in T z ∈ T and z P 2 = z − 1 z^{P^{2}}=z^{-1} z P 2 = z − 1 , then an orbit of length at most two forces z = ± 1 z=\pm1 z = ± 1 . Thus every t ∈ T ∖ { ± I } t \in T \setminus\{\pm I\} t ∈ T ∖ { ± I } is regular semisimple and acts irreducibly on the natural module. Conversely, any irreducible g ∈ G g \in G g ∈ G is regular semisimple and belongs to a conjugate of T T T : its eigenvalues are z , z P , z P 2 , z P 3 z,z^{P},z^{P^{2}},z^{P^{3}} z , z P , z P 2 , z P 3 , and symplectic self-duality forces z P 2 = z − 1 z^{P^{2}}=z^{-1} z P 2 = z − 1 . To see the conjugacy assertion, identify the natural module with F P 4 \mathbb{F}_{P^{4}} F P 4 so that g g g acts by multiplication by z z z . The adjoint on F P [ g ] = F P 4 \mathbb{F}_{P}[g]=\mathbb{F}_{P^{4}} F P [ g ] = F P 4 is x ‾ = x P 2 \overline{x}=x^{P^{2}} x = x P 2 because it sends z z z to z − 1 z^{-1} z − 1 . By nondegeneracy of the trace pairing, every compatible alternating form is Tr F P 4 / F P ( λ x y ‾ ) \operatorname{Tr}_{\mathbb{F}_{P^{4}}/\mathbb{F}_{P}}(\lambda x\overline{y}) Tr F P 4 / F P ( λ x y ) with λ ‾ = − λ \overline{\lambda}=-\lambda λ = − λ and λ ≠ 0 \lambda\ne0 λ = 0 . The ratio of any two such λ \lambda λ belongs to F P 2 × \mathbb{F}_{P^{2}}^{\times} F P 2 × , and multiplication by an element of F P 4 × \mathbb{F}_{P^{4}}^{\times} F P 4 × of that norm transports one form to the other. This gives a symplectic conjugacy of the norm-one tori.
Put
T 0 = T / { ± I } ⊂ S , m = ∣ T 0 ∣ = P 2 + 1 2 , κ = m − 1 4 = P 2 − 1 8 . T_{0}=T/\{\pm I\}\subset S,\qquad m=|T_{0}|=\frac{P^{2}+1}{2},\qquad\kappa=\frac{m-1}{4}=\frac{P^{2}-1}{8}. T 0 = T / { ± I } ⊂ S , m = ∣ T 0 ∣ = 2 P 2 + 1 , κ = 4 m − 1 = 8 P 2 − 1 .
As P P P is odd, gcd ( m , P 2 − 1 ) = 1 \gcd(m,P^{2}-1)=1 g cd( m , P 2 − 1 ) = 1 . The rational Weyl group of T T T is the finite quotient N G ( T ) / T N_{G}(T)/T N G ( T ) / T , where N G ( T ) N_{G}(T) N G ( T ) is the normalizer of T T T in G G G . For this Coxeter torus it is cyclic of order four; its action on T T T and on the character group T ^ \widehat{T} T is induced by P P P -power Frobenius. Every nontrivial character of T 0 T_{0} T 0 therefore has an orbit of length four. Let O \mathcal{O} O be the set of these orbits; then ∣ O ∣ = κ |\mathcal{O}|=\kappa ∣ O ∣ = κ . For each orbit choose a representative θ \theta θ and inflate it to T T T , so that θ ( − I ) = 1 \theta(-I)=1 θ ( − I ) = 1 . Write ρ θ = R T G ( θ ) \rho_{\theta}=R_{T}^{G}(\theta) ρ θ = R T G ( θ ) for its Deligne–Lusztig character. Here R T G ( θ ) R_{T}^{G}(\theta) R T G ( θ ) denotes the (a priori virtual) character obtained by Deligne–Lusztig induction from θ \theta θ ; in the regular cases used below it is an actual irreducible character because the Deligne–Lusztig sign is ε G ε T = ( − 1 ) 2 + 0 = + 1 \varepsilon_{G}\varepsilon_{T}=(-1)^{2+0}=+1 ε G ε T = ( − 1 ) 2 + 0 = + 1 , using the split ranks of G G G and T T T [5 ] .
We use Srinivasan’s labels [16 ] for the unipotent characters
η + = θ 9 , η − = θ 10 , γ = θ 11 , ξ = θ 12 , S t = θ 13 . \eta_{+}=\theta_{9},\qquad\eta_{-}=\theta_{10},\qquad\gamma=\theta_{11},\qquad\xi=\theta_{12},\qquad\mathrm{St}=\theta_{13}. η + = θ 9 , η − = θ 10 , γ = θ 11 , ξ = θ 12 , St = θ 13 .
Their degrees, in this order, are P ( P + 1 ) 2 / 2 P(P+1)^{2}/2 P ( P + 1 ) 2 /2 , P ( P − 1 ) 2 / 2 P(P-1)^{2}/2 P ( P − 1 ) 2 /2 , P ( P 2 + 1 ) / 2 P(P^{2}+1)/2 P ( P 2 + 1 ) /2 , P ( P 2 + 1 ) / 2 P(P^{2}+1)/2 P ( P 2 + 1 ) /2 , and P 4 P^{4} P 4 . For each orbit representative θ \theta θ above, ρ θ \rho_{\theta} ρ θ is a regular cuspidal character of degree ( P 2 − 1 ) 2 (P^{2}-1)^{2} ( P 2 − 1 ) 2 .
Proposition 2.1 (Uniform irreducibility indicator). As class functions on G G G (and, equivalently, on S S S ),
m 1 i r r = κ ( 1 − η + + η − + S t ) − ∑ θ ∈ O ρ θ . (4) m1_{\mathrm{irr}}=\kappa(1-\eta_{+}+\eta_{-}+\mathrm{St})-\sum_{\theta\in\mathcal{O}}\rho_{\theta}. \tag*{(4)} m 1 irr = κ ( 1 − η + + η − + St ) − θ ∈ O ∑ ρ θ . (4)
Proof. The Coxeter-torus Deligne–Lusztig decomposition is
R T G ( 1 ) = 1 − η + + η − + S t ; R_{T}^{G}(1)=1-\eta_{+}+\eta_{-}+\mathrm{St}; R T G ( 1 ) = 1 − η + + η − + St ;
see [9 Table 2.8] and compare [16 ] . Set
Φ = κ R T G ( 1 ) − ∑ θ ∈ O R T G ( θ ) . \Phi=\kappa R_{T}^{G}(1)-\sum_{\theta\in\mathcal{O}}R_{T}^{G}(\theta). Φ = κ R T G ( 1 ) − θ ∈ O ∑ R T G ( θ ) .
We show that Φ ( g ) = m 1 i r r ( g ) \Phi(g)=m1_{\mathrm{irr}}(g) Φ ( g ) = m 1 irr ( g ) by examining the semisimple part of g ∈ G g\in G g ∈ G .
If the semisimple part is not conjugate into T T T , the Deligne–Lusztig character formula gives Φ ( g ) = 0 \Phi(g)=0 Φ ( g ) = 0 . Such an element cannot be irreducible on the natural module.
Suppose next that the semisimple part is central. Write g = z u g=zu g = z u with z ∈ { ± I } z\in\{\pm I\} z ∈ { ± I } and u u u unipotent. The Jordan-decomposition formula expresses R T G ( θ ) ( z u ) R_{T}^{G}(\theta)(zu) R T G ( θ ) ( z u ) as the same Green-function value multiplied by θ ( z ) \theta(z) θ ( z ) . As all characters used here are trivial on the center, the κ \kappa κ terms in the sum cancel the first term. Hence Φ ( z u ) = 0 \Phi(zu)=0 Φ ( z u ) = 0 ; this includes g = ± I g=\pm I g = ± I . No such g g g acts irreducibly.
Finally, suppose the semisimple part is conjugate to an element t ∈ T ∖ { ± I } t\in T\setminus\{\pm I\} t ∈ T ∖ { ± I } . Such an element is regular, so its centralizer is T T T , and g g g itself is conjugate to t t t . The rational Weyl group has order four. Hence
R T G ( 1 ) ( t ) = 4 , ∑ θ ∈ O R T G ( θ ) ( t ) = ∑ ϑ ∈ T ^ 0 ϑ ≠ 1 ϑ ( t 0 ) = − 1. R_T^G(1)(t)=4,\qquad\sum_{\theta\in\mathcal{O}}R_T^G(\theta)(t)=\sum_{\substack{\vartheta\in\widehat{T}_0\\ \vartheta\ne1}}\vartheta(t_0)=-1. R T G ( 1 ) ( t ) = 4 , θ ∈ O ∑ R T G ( θ ) ( t ) = ϑ ∈ T 0 ϑ = 1 ∑ ϑ ( t 0 ) = − 1.
where t 0 t_0 t 0 is the image of t t t in T 0 T_0 T 0 . The final equality is the orthogonality relation for characters of the cyclic group T 0 T_0 T 0 , evaluated at t 0 ≠ 1 t_0\ne1 t 0 = 1 . It follows that
Φ ( t ) = 4 κ + 1 = m . \Phi(t)=4\kappa+1=m. Φ ( t ) = 4 κ + 1 = m .
These, and only these, are the irreducible elements of G G G . Thus Φ = m 1 i r r \Phi=m1_{\mathrm{irr}} Φ = m 1 irr on every conjugacy class.
Remark 2.2 (The density of irreducible elements). As a consistency check, every irreducible element is conjugate to one of the P 2 − 1 P^2-1 P 2 − 1 noncentral elements of T T T . Since each such element has centralizer T T T and rational Weyl group of order four,
∣ { g ∈ G : 1 i r r ( g ) = 1 } ∣ ∣ G ∣ = P 2 − 1 4 ( P 2 + 1 ) = κ m . \frac{|\{g\in G:1_{\mathrm{irr}}(g)=1\}|}{|G|}=\frac{P^2-1}{4(P^2+1)}=\frac{\kappa}{m}. ∣ G ∣ ∣ { g ∈ G : 1 irr ( g ) = 1 } ∣ = 4 ( P 2 + 1 ) P 2 − 1 = m κ .
This is also the coefficient of the trivial character in (4) .
Remark 2.3. For P = 3 P=3 P = 3 , one has m = 5 m=5 m = 5 , κ = 1 \kappa=1 κ = 1 , and ∣ O ∣ = 1 |\mathcal{O}|=1 ∣ O ∣ = 1 . The characters specialize to
η − = χ 6 , η + = χ 24 , ρ ϑ = χ 64 , S t = χ 81 , \eta_-=\chi_6,\qquad\eta_+=\chi_{24},\qquad\rho_\vartheta=\chi_{64},\qquad\mathrm{St}=\chi_{81}, η − = χ 6 , η + = χ 24 , ρ ϑ = χ 64 , St = χ 81 ,
where the subscripts indicate degrees. Consequently,
5 1 i r r = 1 + χ 6 − χ 24 − χ 64 + χ 81 . (5) 5\,1_{\mathrm{irr}}=1+\chi_6-\chi_{24}-\chi_{64}+\chi_{81}. \tag*{(5)} 5 1 irr = 1 + χ 6 − χ 24 − χ 64 + χ 81 . (5)
In particular, the character identity used in the explicit enumeration for P = 3 P=3 P = 3 follows from the uniform Coxeter-torus formula, rather than requiring an independent table calculation. Summing it over the Frobenius classes of the parameters a ∈ F 3 H a\in\mathbb{F}_{3^H} a ∈ F 3 H expresses 5 A H ( 3 ) 5A_H(3) 5 A H ( 3 ) as four character sums. Their cohomological evaluation below yields the explicit formula in Theorem 2.4.
Let F a , H F_{a,H} F a , H denote the geometric Frobenius at the parameter a ∈ F P H a\in\mathbb{F}_{P^H} a ∈ F P H on the root local system. The fixed subspaces and the irreducibility test are unchanged on passing from arithmetic Frobenius to its inverse. We use geometric Frobenius throughout the trace formulas. Define the formal power series (without assuming rationality at this stage)
R P ( u ) = exp ( 4 m ∑ H ≥ 1 A H ( P ) u H H ) , m = P 2 + 1 2 . R_P(u)=\exp\left(4m\sum_{H\geq1}A_H(P)\frac{u^H}{H}\right),\qquad m=\frac{P^2+1}{2}. R P ( u ) = exp ( 4 m H ≥ 1 ∑ A H ( P ) H u H ) , m = 2 P 2 + 1 .
Theorem 2.4. Put
N ( u ) = 1 + u + 2 u 2 + u 3 + 2 u 4 + 17 u 5 + 26 u 6 + 51 u 7 + 18 u 8 + 27 u 9 + 162 u 10 + 243 u 11 + 729 u 12 , N(u)=1+u+2u^2+u^3+2u^4+17u^5+26u^6
+51u^7+18u^8+27u^9+162u^{10}+243u^{11}+729u^{12}, N ( u ) = 1 + u + 2 u 2 + u 3 + 2 u 4 + 17 u 5 + 26 u 6 + 51 u 7 + 18 u 8 + 27 u 9 + 162 u 10 + 243 u 11 + 729 u 12 ,
D ( u ) = 1 − u 2 − 3 u 3 + 3 u 4 + 18 u 5 + 9 u 6 − 27 u 7 − 27 u 8 + 243 u 10 . D(u)=1-u^2-3u^3+3u^4+18u^5+9u^6-27u^7-27u^8+243u^{10}. D ( u ) = 1 − u 2 − 3 u 3 + 3 u 4 + 18 u 5 + 9 u 6 − 27 u 7 − 27 u 8 + 243 u 10 .
Then
R 3 ( u ) = ( N ( u ) ( 1 − u ) ( 1 − 3 u ) D ( u ) ) 4 . R_3(u)=\left(\frac{N(u)}{(1-u)(1-3u)D(u)}\right)^4. R 3 ( u ) = ( ( 1 − u ) ( 1 − 3 u ) D ( u ) N ( u ) ) 4 .
Equivalently, if N ( u ) = ∏ i = 1 12 ( 1 − α i u ) N(u)=\prod_{i=1}^{12}(1-\alpha_i u) N ( u ) = ∏ i = 1 12 ( 1 − α i u ) and D ( u ) = ∏ j = 1 10 ( 1 − β j u ) D(u)=\prod_{j=1}^{10}(1-\beta_j u) D ( u ) = ∏ j = 1 10 ( 1 − β j u ) , then for every H ≥ 1 H\geq1 H ≥ 1 ,
A H ( 3 ) = 3 H + 1 − ∑ i α i H + ∑ j β j H 5 . A_H(3)=\frac{3^H+1-\sum_i\alpha_i^H+\sum_j\beta_j^H}{5}. A H ( 3 ) = 5 3 H + 1 − ∑ i α i H + ∑ j β j H .
The cohomological input is established in § 2.2 and § 2.3 . The exact computation in § 2.4 then determines the coefficients uniquely, using the exact finite-field counts stated below.
Notation
The following symbols recur across the two parts.
Monodromy and ramification at infinity
The Katz–Tiep construction and its relation to the root space
We work for P = p f P=p^f P = p f , p p p odd, p ∤ f p\nmid f p ∤ f , and fix an auxiliary prime ℓ ≠ p \ell\ne p ℓ = p . We briefly recall the particular part of the construction of Katz–Tiep [10 ] needed here. Their local systems on the affine parameter line are obtained from exponential sums and realize finite Weil representations of symplectic groups. For n = 2 n=2 n = 2 the total Weil representation of Sp 4 ( F P ) \operatorname{Sp}_4(\mathbb{F}_P) Sp 4 ( F P ) has dimension P 2 P^2 P 2 and is the direct sum of its odd- and even-dimensional constituents, of dimensions ( P 2 + 1 ) / 2 (P^2+1)/2 ( P 2 + 1 ) /2 and ( P 2 − 1 ) / 2 (P^2-1)/2 ( P 2 − 1 ) /2 . The term “total” is important: using both constituents retains the action of the central involution, which will be needed to identify the kernel of the monodromy representation.
Here is the elementary bridge from the exponential sum to our skew quartic. For a finite extension k / F P k/\mathbb{F}_P k / F P and b ∈ k b\in k b ∈ k , consider the quadratic form on the F p \mathbb{F}_p F p -vector space underlying k k k given by
q b ( x ) = Tr k / F p ( x P 2 + 1 + x P + 1 + b x 2 ) . q_b(x)=\operatorname{Tr}_{k/\mathbb{F}_p}\left(x^{P^2+1}+x^{P+1}+bx^2\right). q b ( x ) = Tr k / F p ( x P 2 + 1 + x P + 1 + b x 2 ) .
Its polar form has a radical that can be described by an additive polynomial. More precisely, for each finite extension k / F P k/\mathbb{F}_P k / F P , the radical consists of the elements of k k k annihilated by this polynomial: the polar equation, raised to the P 2 P^2 P 2 -th power, becomes the equation displayed below. In the normalization used below this polynomial is
R b ( X ) = X P 4 + X P 3 + 2 b P 2 X P 2 + X P + X . R_b(X)=X^{P^4}+X^{P^3}+2b^{P^2}X^{P^2}+X^P+X. R b ( X ) = X P 4 + X P 3 + 2 b P 2 X P 2 + X P + X .
Thus the purely inseparable change of parameter
identifies R b R_b R b with L a L_a L a from Section 2. This is the reason Katz–Tiep’s sheaves are relevant to the present counting problem: the quadratic form appearing in their trace function has exactly the root space whose Frobenius action tests whether f a f_a f a is an atom.
The second ingredient is the quadratic Gauss-sum identity. We record its elementary content, to make the normalization of the subsequent comparison transparent. Let k / F P k/\mathbb{F}_P k / F P be finite, let ψ k = ψ ∘ Tr k / F p \psi_k=\psi\circ\operatorname{Tr}_{k/\mathbb{F}_p} ψ k = ψ ∘ Tr k / F p , and set
S b ( k ) = ∑ x ∈ k ψ k ( x P 2 + 1 + x P + 1 + b x 2 ) . S_b(k)=\sum_{x\in k}\psi_k\left(x^{P^2+1}+x^{P+1}+bx^2\right). S b ( k ) = x ∈ k ∑ ψ k ( x P 2 + 1 + x P + 1 + b x 2 ) .
Writing q b q_b q b for the corresponding F p \mathbb{F}_p F p -valued quadratic form, translation ( x , y ) ↦ ( x + y , y ) (x,y)\mapsto(x+y,y) ( x , y ) ↦ ( x + y , y ) and additive-character orthogonality give
∣ S b ( k ) ∣ 2 = ∣ k ∣ ∑ v ∈ rad ( q b ) ψ ( q b ( v ) ) = ∣ k ∣ # rad ( q b ) . \lvert S_b(k)\rvert^2=\lvert k\rvert\sum_{v\in\operatorname{rad}(q_b)}\psi(q_b(v))=\lvert k\rvert\,\#\operatorname{rad}(q_b). ∣ S b ( k ) ∣ 2 = ∣ k ∣ v ∈ rad ( q b ) ∑ ψ ( q b ( v )) = ∣ k ∣ # rad ( q b ) .
In the last equality, q b q_b q b vanishes on its radical, since p p p is odd. The radical is exactly ker L 2 b P 2 / k \ker L_{2b^{P^2}}/k ker L 2 b P 2 / k . Thus, for the normalization A k = A F P [ k : F P ] A_k=A_{\mathbb{F}_P}^{[k:\mathbb{F}_P]} A k = A F P [ k : F P ] of [10 §3] , where A F P = − ∑ r ∈ F P × ψ ( 2 r ) χ 2 , F P ( r ) A_{\mathbb{F}_P}=-\sum_{r\in\mathbb{F}_P^\times}\psi(2r)\chi_{2,\mathbb{F}_P}(r) A F P = − ∑ r ∈ F P × ψ ( 2 r ) χ 2 , F P ( r ) because P 2 ≡ 1 ( m o d 4 ) P^2\equiv1\pmod4 P 2 ≡ 1 ( mod 4 ) , we have ∣ A k ∣ 2 = ∣ k ∣ \lvert A_k\rvert^2=\lvert k\rvert ∣ A k ∣ 2 = ∣ k ∣ and
∣ − A k − 1 S b ( k ) ∣ 2 = # { x ∈ k : L 2 b P 2 ( x ) = 0 } . (6) \left\lvert-A_k^{-1}S_b(k)\right\rvert^2=\#\{x\in k:L_{2b^{P^2}}(x)=0\}. \tag*{(6)} − A k − 1 S b ( k ) 2 = # { x ∈ k : L 2 b P 2 ( x ) = 0 } . (6)
This is the specialization of Katz–Tiep [10 Lemma 5.2] to ( n , s , t , q ) = ( 2 , 1 , b , P ) (n,s,t,q)=(2,1,b,P) ( n , s , t , q ) = ( 2 , 1 , b , P ) . The arithmetic normalization of the total Weil trace function is that of [10 §§3,5] ; the absolute-square identity does not depend on the choice of phase of A k A_k A k . Since the trace of the endomorphism representation of a finite unitary representation is the squared absolute value of its character, the Frobenius trace on the root permutation representation agrees with that on the endomorphism of the total Weil sheaf. The latter is insensitive to scalar operators, so the kernel comparison below is needed to pass to full symplectic monodromy.
Full monodromy and the kernel comparison
The relevant result is Katz–Tiep [10 Theorem 3.2(iii)] , not a theorem about the natural four-dimensional representation. For their specialization s = 1 s=1 s = 1 , take n = 2 n=2 n = 2 and q = P = p f q=P=p^f q = P = p f . For odd p p p , the theorem applies when n ≥ 2 n\ge2 n ≥ 2 , p ∤ n p\nmid n p ∤ n , and q = p f q=p^f q = p f with p ∤ f p\nmid f p ∤ f . The parameter dictionary is ( n , s , t , q ) = ( 2 , 1 , b , P ) (n,s,t,q)=(2,1,b,P) ( n , s , t , q ) = ( 2 , 1 , b , P ) : s s s is fixed as the coefficient of x P + 1 x^{P+1} x P + 1 , and t = b t=b t = b varies as the coefficient of x 2 x^2 x 2 in the total trace − S b ( k ) / A k -S_b(k)/A_k − S b ( k ) / A k . The theorem gives geometric monodromy Sp ( 2 n , q ) \operatorname{Sp}(2n,q) Sp ( 2 n , q ) on the direct sum of the two matched normalized Fourier constituents, in its total Weil representation. The system in the theorem starts on the parameter line over F p \mathbb{F}_p F p ; on its pullback to the line over F q \mathbb{F}_q F q , the arithmetic and geometric groups agree. The two constituents must be kept together in order to detect the central involution.
The change a = 2 b P 2 a=2b^{P^2} a = 2 b P 2 is radicial and induces an equivalence of the corresponding geometric finite étale sites. Under this change the polar radical of the quadratic phase is the root space of L a L_a L a , as made explicit in [10 §5, especially Lemma 5.2] . It is important, however, that the monodromy theorem concerns the Weil sheaf, whereas our counting problem concerns the natural four-dimensional root-space representation. We therefore give the additional kernel-comparison argument rather than identifying these representations without proof.
Lemma 3.1 (Kernel comparison in the monodromy argument). Let Γ \Gamma Γ be the geometric fundamental group of the parameter line, let ω : Γ ↠ G \omega:\Gamma\twoheadrightarrow G ω : Γ ↠ G be the Katz–Tiep monodromy quotient, and let
Ω : G ⟶ GL ( W ) \Omega:G\longrightarrow\operatorname{GL}(\mathcal{W}) Ω : G ⟶ GL ( W ) be the total Weil representation, where dim W = P 2 \dim\mathcal{W}=P^2 dim W = P 2 . Let ρ : Γ ⟶ Sp ( V ) \rho:\Gamma\longrightarrow\operatorname{Sp}(V) ρ : Γ ⟶ Sp ( V ) denote the action on the four-dimensional root space of L a L_a L a . Assume that, for every closed point and every finite constant-field extension, the Frobenius traces of the associated arithmetic representations satisfy
Tr ( Frob ∣ Q ‾ ℓ [ V ] ) = Tr ( Frob ∣ End ( W ) ) , \operatorname{Tr}(\operatorname{Frob}\mid\overline{\mathbb{Q}}_\ell[V])=\operatorname{Tr}(\operatorname{Frob}\mid\operatorname{End}(\mathcal{W})), Tr ( Frob ∣ Q ℓ [ V ]) = Tr ( Frob ∣ End ( W )) , where the latter is acted on by conjugation through Ω ∘ ω \Omega\circ\omega Ω ∘ ω . Then ρ \rho ρ factors through G G G , and the induced map G ⟶ Sp ( V ) G\longrightarrow\operatorname{Sp}(V) G ⟶ Sp ( V ) is an isomorphism.
Proof. The permutation representation on the finite set V V V has Frobenius trace equal to the number of fixed roots. The action of g ∈ G g\in G g ∈ G on End ( W ) \operatorname{End}(\mathcal{W}) End ( W ) is
T ⟼ Ω ( g ) T Ω ( g ) − 1 . T\longmapsto\Omega(g)T\Omega(g)^{-1}. T ⟼ Ω ( g ) T Ω ( g ) − 1 .
Via End ( W ) ≃ W ⊗ W ∨ \operatorname{End}(\mathcal{W})\simeq\mathcal{W}\otimes\mathcal{W}^{\vee} End ( W ) ≃ W ⊗ W ∨ , its trace is
Tr ( Ω ( g ) ) Tr ( Ω ( g ) − 1 ) = ∣ Tr ( Ω ( g ) ) ∣ 2 , \operatorname{Tr}(\Omega(g))\operatorname{Tr}(\Omega(g)^{-1})=\lvert\operatorname{Tr}(\Omega(g))\rvert^2, Tr ( Ω ( g )) Tr ( Ω ( g ) − 1 ) = ∣ Tr ( Ω ( g )) ∣ 2 ,
where the final equality uses a complex realization of the finite Weil representation. The quadratic Gauss-sum identity [10 Lemma 5.2] , after a = 2 b P 2 a=2b^{P^2} a = 2 b P 2 , identifies this with the number of Frobenius-fixed roots. The finite-image normalization and arithmetic extensions are those in [10 §§3,5] : one must compare traces of the same normalized total Weil sheaf in every finite extension. This avoids inferring equality of arithmetic representations from a comparison of geometric traces alone.
More formally, apply the comparison first to the arithmetic fundamental group. Chebotarev density gives equality of characters from equality of closed-point Frobenius traces. Both representations have finite image and are semisimple in characteristic zero; thus Brauer–Nesbitt makes them isomorphic. Restricting to Γ \Gamma Γ gives
Q ‾ ℓ [ V ] ≃ End ( W ) \overline{\mathbb{Q}}_\ell[V]\simeq\operatorname{End}(\mathcal{W}) Q ℓ [ V ] ≃ End ( W )
as Γ \Gamma Γ -representations.
The kernel of the first representation is ker ρ \ker\rho ker ρ , since the permutation action on the basis indexed by V V V is faithful. The kernel of the second is the inverse image, under ω \omega ω , of the projective kernel of Ω \Omega Ω , because conjugation on all endomorphisms is trivial exactly for scalar operators.
The total Weil representation is projectively faithful. By its projective kernel we mean
ker ( G ⟶ PGL ( W ) ) = { g ∈ G : Ω ( g ) acts as a scalar on W } . \ker(G\longrightarrow\operatorname{PGL}(\mathcal{W}))=\{g\in G:\Omega(g)\text{ acts as a scalar on }\mathcal{W}\}. ker ( G ⟶ PGL ( W )) = { g ∈ G : Ω ( g ) acts as a scalar on W } .
It is precisely the kernel of the conjugation action on End ( W ) \operatorname{End}(W) End ( W ) , and it is a normal subgroup of G G G . The central quotient PSp 4 ( F P ) \operatorname{PSp}_{4}(\mathbb{F}_{P}) PSp 4 ( F P ) is simple for odd P P P ; together with the perfectness of G G G , this shows that a proper normal subgroup of G G G must be central. The kernel is proper since the projective Weil representation is nontrivial. Moreover − I -I − I acts by opposite scalars on the two nonzero even and odd Weil constituents, so it is not scalar on their direct sum W W W . Thus the projective kernel is trivial. Consequently
ker ρ = ker ω . \ker\rho= \ker\omega. ker ρ = ker ω .
Thus ρ \rho ρ and ω \omega ω have the same geometric kernel, and ρ \rho ρ descends through the quotient Γ / ker ω ≃ G \Gamma/\ker\omega\simeq G Γ/ ker ω ≃ G . The resulting homomorphism G ↪ Sp ( V ) G \hookrightarrow\operatorname{Sp}(V) G ↪ Sp ( V ) is injective. Both groups have order P 4 ( P 2 − 1 ) ( P 4 − 1 ) P^{4}(P^{2}-1)(P^{4}-1) P 4 ( P 2 − 1 ) ( P 4 − 1 ) , so it is an isomorphism. In particular, the natural root representation and the total Weil representation have the same monodromy quotient, up to this identification. □
The small Weil constituent and local Fourier transformation
To compute the local ramification it is enough to work with one faithful constituent of the total Weil representation. We spell out why this is legitimate and why the binomial Artin–Schreier sheaf appears.
Set
d = P 2 + 1 2 , e = P + 1 2 , k = d − 1 = P 2 − 1 2 , F ( x ) = x d + x e . d = \frac{P^{2}+1}{2}, \qquad e = \frac{P+1}{2}, \qquad k = d-1 = \frac{P^{2}-1}{2}, \qquad F(x) = x^{d}+x^{e}. d = 2 P 2 + 1 , e = 2 P + 1 , k = d − 1 = 2 P 2 − 1 , F ( x ) = x d + x e .
The total Weil representation decomposes into constituents of dimensions d d d and k k k . In the Fourier–Deligne realization of Katz–Tiep [10 ] , the rank-k k k (small) constituent is, up to a constant-field rank-one normalization, the Fourier transform of the rank-one Artin–Schreier sheaf
L ψ ( F ( x ) ) . \mathcal{L}_{\psi}(F(x)). L ψ ( F ( x )) .
The complementary constituent is obtained by inserting the quadratic Kummer character in the input variable before Fourier transformation. At the level of trace functions, for a finite extension K / F P K/\mathbb{F}_{P} K / F P , these two transforms involve respectively the sums
∑ x ∈ K ψ K ( F ( x ) + b x ) , ∑ x ∈ K × χ 2 , K ( x ) ψ K ( F ( x ) + b x ) . \sum_{x\in K}\psi_{K}(F(x)+bx), \qquad\sum_{x\in K^{\times}}\chi_{2,K}(x)\psi_{K}(F(x)+bx). x ∈ K ∑ ψ K ( F ( x ) + b x ) , x ∈ K × ∑ χ 2 , K ( x ) ψ K ( F ( x ) + b x ) .
Indeed the elementary square-map identity
∑ y ∈ K h ( y 2 ) = ∑ x ∈ K ( 1 + χ 2 , K ( x ) ) h ( x ) , χ 2 , K ( 0 ) = 0 , \sum_{y\in K}h(y^{2})=\sum_{x\in K}(1+\chi_{2,K}(x))h(x), \qquad\chi_{2,K}(0)=0, y ∈ K ∑ h ( y 2 ) = x ∈ K ∑ ( 1 + χ 2 , K ( x )) h ( x ) , χ 2 , K ( 0 ) = 0 ,
explains why their sum is the total Weil trace, whose phase is
F ( y 2 ) + b y 2 = y P 2 + 1 + y P + 1 + b y 2 . F(y^{2})+by^{2}=y^{P^{2}+1}+y^{P+1}+by^{2}. F ( y 2 ) + b y 2 = y P 2 + 1 + y P + 1 + b y 2 .
Thus the binomial F F F is not an auxiliary polynomial introduced for the ramification calculation: it is the defining phase of the small constituent in Katz–Tiep’s construction.
The small Weil representation Ω s m : G → GL ( W s m ) \Omega_{\mathrm{sm}}:G\to\operatorname{GL}(W_{\mathrm{sm}}) Ω sm : G → GL ( W sm ) is faithful. One can see this from the normal-subgroup structure of G G G : its kernel is proper and hence central, whereas the central involution acts nontrivially on the small constituent.
We use the upper numbering of ramification groups in the convention of Serre. For v ≥ 0 v\geq0 v ≥ 0 , let I ∞ v I_{\infty}^{v} I ∞ v denote the upper-numbering subgroup of geometric inertia at infinity; for a finite-image representation, its positive upper breaks are the numbers b > 0 b>0 b > 0 at which the image filtration jumps, or equivalently where its image J b J^{b} J b differs from the image at b + ϵ b+\epsilon b + ϵ (the subgroup for v > b v>b v > b sufficiently close to b b b ). The slopes of an individual representation need not detect every break of the image filtration. Write
J v = ω ( I ∞ v ) ⊂ G , J b + = ⋃ ϵ > 0 J b + ϵ . J^{v}=\omega(I_{\infty}^{v})\subset G, \qquad J^{b+}=\bigcup_{\epsilon>0}J^{b+\epsilon}. J v = ω ( I ∞ v ) ⊂ G , J b + = ϵ > 0 ⋃ J b + ϵ .
The filtration is decreasing; the union defines the right-hand subgroup because its image is finite. Faithfulness gives
J u → ∼ Ω s m ( J u ) ( u ≥ 0 ) , J^{u}\xrightarrow{\sim}\Omega_{\mathrm{sm}}(J^{u}) \qquad(u\geq0), J u ∼ Ω sm ( J u ) ( u ≥ 0 ) ,
compatibly with subgroup inclusions. By the preceding kernel comparison the same J u J^{u} J u acts faithfully on the natural root space. Hence the ranks, orders, and jumps of the images of upper ramification groups may be calculated on the small Weil sheaf. This does not by itself locate their images inside a specified Siegel parabolic; that identification will be made using the roots.
The next calculation is the main local input to the paper. It is useful to separate three steps: first compute the polar phases after a tame pullback, then recover the two filtered pieces of the abelian wild image from their characters, and finally identify that image with the Siegel radical by an independent Newton-polygon argument. The last step is necessary because local Fourier analysis alone gives an abstract filtered group, not its embedding in Sp 4 ( F P ) \operatorname{Sp}_4(\mathbb{F}_P) Sp 4 ( F P ) .
We use Laumon’s local Fourier transform and stationary-phase formalism [11 ] and the explicit local Fourier formula of Abbes–Saito [1 Theorem 3.9] . The latter, under its Legendre conditions, describes the transformed local representation as the pushforward of a rank-one Artin–Schreier character (with tame and unramified corrections). It is this explicit pushforward description, rather than the mere existence of local Fourier transformation, that allows us to read the ramification filtration.
Recall that a Singer subgroup of GL 2 ( F P ) \operatorname{GL}_2(\mathbb{F}_P) GL 2 ( F P ) is the image of F P 2 × \mathbb{F}_{P^2}^{\times} F P 2 × acting by multiplication on the 2-dimensional F P \mathbb{F}_P F P -space F P 2 \mathbb{F}_{P^2} F P 2 . It is cyclic of order P 2 − 1 P^2-1 P 2 − 1 , and a generator is called a Singer cycle. We will use this terminology for its conjugates in GL ( W ) \operatorname{GL}(W) GL ( W ) as well.
Lemma 3.3 (Direct calculation of tame inertia on the roots). Let K = F ‾ P ( ( u ) ) K=\overline{\mathbb{F}}_P((u)) K = F P (( u )) , where u = a − 1 u=a^{-1} u = a − 1 , and let V V V be the four-dimensional root space of L a L_a L a . The roots of strictly positive valuation, together with zero, form an inertia-stable 2-dimensional F P \mathbb{F}_P F P -subspace W W W . Wild inertia fixes W W W pointwise, and the inertia image on W W W is a Singer subgroup of order exactly P 2 − 1 P^2-1 P 2 − 1 . The plane W W W is Lagrangian, and the prime-to-p p p quotient of the full inertia image on V V V has order P 2 − 1 P^2-1 P 2 − 1 .
Proof. With v ( u ) = 1 v(u)=1 v ( u ) = 1 , the polynomial whose nonzero roots we study is
L a ( X ) X = X P 4 − 1 + X P 3 − 1 + u − 1 X P 2 − 1 + X P − 1 + 1. \frac{L_a(X)}{X}=X^{P^4-1}+X^{P^3-1}+u^{-1}X^{P^2-1}+X^{P-1}+1. X L a ( X ) = X P 4 − 1 + X P 3 − 1 + u − 1 X P 2 − 1 + X P − 1 + 1.
Its lower Newton polygon has two segments, joining ( 0 , 0 ) (0,0) ( 0 , 0 ) to ( P 2 − 1 , − 1 ) (P^2-1,-1) ( P 2 − 1 , − 1 ) and then to ( P 4 − 1 , 0 ) (P^4-1,0) ( P 4 − 1 , 0 ) . Their respective slopes are − 1 / ( P 2 − 1 ) -1/(P^2-1) − 1/ ( P 2 − 1 ) and 1 / ( P 4 − P 2 ) 1/(P^4-P^2) 1/ ( P 4 − P 2 ) . Thus precisely P 2 − 1 P^2-1 P 2 − 1 nonzero roots have valuation 1 / ( P 2 − 1 ) 1/(P^2-1) 1/ ( P 2 − 1 ) ; the remaining P 4 − P 2 P^4-P^2 P 4 − P 2 nonzero roots have valuation − 1 / ( P 4 − P 2 ) -1/(P^4-P^2) − 1/ ( P 4 − P 2 ) . Because L a L_a L a is additive, the zero root together with the roots of positive valuation is closed under addition and multiplication by F P \mathbb{F}_P F P . It therefore forms an inertia-stable subspace W ⊂ V W\subset V W ⊂ V of cardinality P 2 P^2 P 2 .
Choose λ \lambda λ with λ P 2 − 1 = u \lambda^{P^2-1}=u λ P 2 − 1 = u and substitute X = λ Y X=\lambda Y X = λY . After division by λ \lambda λ we obtain
Y + Y P 2 + λ P 2 − 1 Y P + λ P 3 − 1 Y P 3 + λ P 4 − 1 Y P 4 = 0. (7) Y+Y^{P^2}+\lambda^{P^2-1}Y^P+\lambda^{P^3-1}Y^{P^3}+\lambda^{P^4-1}Y^{P^4}=0. \tag*{(7)} Y + Y P 2 + λ P 2 − 1 Y P + λ P 3 − 1 Y P 3 + λ P 4 − 1 Y P 4 = 0. (7)
The reduction modulo λ \lambda λ is Y P 2 + Y = 0 Y^{P^2}+Y=0 Y P 2 + Y = 0 . Its derivative is 1 1 1 , so its P 2 P^2 P 2 roots in the algebraically closed residue field are simple. Hensel’s lemma lifts each root uniquely to F ‾ P ( ( λ ) ) \overline{\mathbb{F}}_P((\lambda)) F P (( λ )) . Every element of W W W is among these lifts, because its rescaled coordinate Y = X / λ Y=X/\lambda Y = X / λ has nonnegative valuation. Thus W W W is pointwise fixed by wild inertia: the extension K ( λ ) / K K(\lambda)/K K ( λ ) / K is tame of degree P 2 − 1 P^2-1 P 2 − 1 .
The residual root space
W r e d = { y ∈ F ‾ P : y P 2 + y = 0 } W_{\mathrm{red}}=\{y\in\overline{\mathbb{F}}_P:y^{P^2}+y=0\} W red = { y ∈ F P : y P 2 + y = 0 }
is one-dimensional over F P 2 \mathbb{F}_{P^2} F P 2 . A tame inertia element taking λ \lambda λ to ζ λ \zeta\lambda ζ λ , with ζ ∈ μ P 2 − 1 \zeta\in\mu_{P^2-1} ζ ∈ μ P 2 − 1 , acts on this residual root space by multiplication by ζ \zeta ζ . Indeed, if g ( λ ) = ζ λ g(\lambda)=\zeta\lambda g ( λ ) = ζ λ , then
g ( X ) λ = g ( λ ) λ g ( X λ ) = ζ X λ , \frac{g(X)}{\lambda}=\frac{g(\lambda)}{\lambda}g\left(\frac{X}{\lambda}\right)=\zeta\frac{X}{\lambda}, λ g ( X ) = λ g ( λ ) g ( λ X ) = ζ λ X ,
because geometric inertia acts trivially on the residue field. These multiplications give the entire F P 2 × \mathbb{F}_{P^2}^{\times} F P 2 × -action, and hence an injective cyclic subgroup of GL F P ( W ) \operatorname{GL}_{\mathbb{F}_P}(W) GL F P ( W ) of order P 2 − 1 P^2-1 P 2 − 1 (a Singer subgroup). The inertia action on W W W factors through this group because all elements of W W W are defined over K ( λ ) K(\lambda) K ( λ ) .
The alternating form on V V V is inertia invariant. If its restriction to the two-dimensional space W W W were nonzero, it would be nondegenerate, so the inertia image on W W W would lie in Sp ( W ) = SL 2 ( F P ) \operatorname{Sp}(W)=\operatorname{SL}_2(\mathbb{F}_P) Sp ( W ) = SL 2 ( F P ) . But the determinant of multiplication by ζ ∈ F P 2 × \zeta\in\mathbb{F}_{P^2}^{\times} ζ ∈ F P 2 × is its norm to F P × \mathbb{F}_P^{\times} F P × , and this norm is surjective. Hence the restriction of the alternating form must vanish. As dim W = 1 2 dim V \dim W=\frac{1}{2}\dim V dim W = 2 1 dim V , the plane W W W is Lagrangian.
Consequently the full inertia image I I I lies in the Siegel parabolic P W = Stab Sp ( V ) ( W ) P_W=\operatorname{Stab}_{\operatorname{Sp}(V)}(W) P W = Stab Sp ( V ) ( W ) . The restriction map P W → GL ( W ) P_W\to\operatorname{GL}(W) P W → GL ( W ) has kernel the Siegel unipotent radical U S i e g U_{\mathrm{Sieg}} U Sieg . The kernel of its restriction to I I I is I ∩ U S i e g I\cap U_{\mathrm{Sieg}} I ∩ U Sieg , a p p p -group. The image of I I I in GL ( W ) \operatorname{GL}(W) GL ( W ) is the Singer group of order P 2 − 1 P^2-1 P 2 − 1 , while the wild inertia image is contained in U S i e g U_{\mathrm{Sieg}} U Sieg . Since the kernel of I → GL ( W ) I\to\operatorname{GL}(W) I → GL ( W ) is a p p p -group and the quotient has order prime to p p p , this kernel is exactly the wild inertia image (the normal Sylow p p p -subgroup of I I I ). Thus the tame quotient has order P 2 − 1 P^2-1 P 2 − 1 , as claimed. □
For subsequent use, let
Sym 2 ( F P ) = { B ∈ M 2 ( F P ) : B t = B } . \operatorname{Sym}_2(\mathbb{F}_P)=\{B\in M_2(\mathbb{F}_P):B^{\mathsf{t}}=B\}. Sym 2 ( F P ) = { B ∈ M 2 ( F P ) : B t = B } .
This is a three-dimensional F P \mathbb{F}_P F P -vector space, considered as an additive group. In a symplectic basis adapted to a Lagrangian plane, the Siegel unipotent radical is
U S i e g = { ( I 2 B 0 I 2 ) : B ∈ Sym 2 ( F P ) } . U_{\mathrm{Sieg}}=\left\{\begin{pmatrix} I_2 & B \\ 0 & I_2 \end{pmatrix}:B\in\operatorname{Sym}_2(\mathbb{F}_P)\right\}. U Sieg = { ( I 2 0 B I 2 ) : B ∈ Sym 2 ( F P ) } .
The multiplication law is u B u C = u B + C u_Bu_C=u_{B+C} u B u C = u B + C ; thus this is an isomorphism of additive groups, not an assertion about matrix multiplication in Sym 2 ( F P ) \operatorname{Sym}_2(\mathbb{F}_P) Sym 2 ( F P ) .
Lemma 3.4 (Wild inertia and its breaks). The wild inertia image U U U on the natural root space is elementary abelian of order P 3 P^3 P 3 . Its upper breaks are b 1 = 1 / ( P − 1 ) b_1=1/(P-1) b 1 = 1/ ( P − 1 ) and b 2 = ( P 2 + 1 ) / ( P 2 − 1 ) b_2=(P^2+1)/(P^2-1) b 2 = ( P 2 + 1 ) / ( P 2 − 1 ) . Between these breaks its image is a subgroup U 0 U_0 U 0 of order P 2 P^2 P 2 .
Proof. By the preceding discussion, the rank-k k k small Weil sheaf is the Fourier–Deligne transform of L ψ ( F ( x ) ) \mathcal{L}_{\psi}(F(x)) L ψ ( F ( x )) up to an unramified constant-field normalization. Its trace over K K K is
− A K − 1 ∑ x ∈ K ψ K ( F ( x ) + b x ) , A K = A F P [ K : F P ] . -A_K^{-1}\sum_{x\in K}\psi_K(F(x)+bx),\qquad A_K=A_{\mathbb{F}_P}^{[K:\mathbb{F}_P]}. − A K − 1 x ∈ K ∑ ψ K ( F ( x ) + b x ) , A K = A F P [ K : F P ] .
The normalization does not change geometric inertia. The parameter change a = 2 b P 2 a=2b^{P^2} a = 2 b P 2 is a scalar change followed by an iterate of relative Frobenius. On the completed local fields at infinity, it identifies the ramification filtrations: the scalar change preserves the normalized valuation, and the purely inseparable part induces the same map on finite separable extensions as the Frobenius identification, which preserves normalized valuations with the corresponding uniformizers. The local Fourier calculation therefore applies to the inertia quotient in Lemma 3.1 .
Apply Abbes–Saito [1 Theorem 3.9] at infinity to the rank-one sheaf G = L ψ ( F ( x ) ) \mathcal{G}=\mathcal{L}_{\psi}(F(x)) G = L ψ ( F ( x )) . Their hypotheses in this case are (i) the Legendre triple condition, (ii) finite source and target maps that are étale on generic points, (iii) a source point at infinity, and (iv′) a target point at infinity and all slopes of the source pushforward strictly greater than one. We work over the henselization at infinity with perfect residue field F ‾ P \overline{\mathbb{F}}_P F P , in characteristic p > 2 p > 2 p > 2 , and use its completion for the valuation calculation.
In the uniformizer v = 1 / x v = 1/x v = 1/ x , set
A = v − d + v − e , b = v − 1 , c = − 1 2 ( v − k + v − ( e − 1 ) ) , α = A ′ = − d v − d − 1 − e v − e − 1 , A = v^{-d} + v^{-e}, \qquad b = v^{-1}, \qquad c = -\frac{1}{2}(v^{-k} + v^{-(e-1)}), \qquad\alpha= A' = -dv^{-d-1} - ev^{-e-1}, A = v − d + v − e , b = v − 1 , c = − 2 1 ( v − k + v − ( e − 1 ) ) , α = A ′ = − d v − d − 1 − e v − e − 1 ,
with primes denoting d / d v d/dv d / d v . Here d = e = 1 / 2 d = e = 1/2 d = e = 1/2 and k = − 1 / 2 k = -1/2 k = − 1/2 as elements of the residue field. In Definition 2.16 of [1 ] , take the tame factor to be trivial and the wild factor to be the Artin–Schreier sheaf of A A A . It is trivialized by a cyclic extension of degree p p p , since A A A has pole order d d d prime to p p p . For this length-one Witt vector, the conditions (2.13.1)–(2.13.3) of [1 ] are
A ∈ fil n , 2 ord v ( α + c b ′ ) ≥ − n + ν ( c ) , 2 ν ( b ) + p ν ( c ) < ( p − 2 ) n , A \in\operatorname{fil}_{n}, \qquad2\operatorname{ord}_{v}(\alpha+ cb') \geq-n + \nu(c), \qquad2\nu(b) + p\nu(c) < (p - 2)n, A ∈ fil n , 2 ord v ( α + c b ′ ) ≥ − n + ν ( c ) , 2 ν ( b ) + p ν ( c ) < ( p − 2 ) n ,
where n = − ord v ( v α ) n = -\operatorname{ord}_{v}(v\alpha) n = − ord v ( v α ) and ν ( h ) = ord v ( v h ′ / h ) \nu(h) = \operatorname{ord}_{v}(vh'/h) ν ( h ) = ord v ( v h ′ / h ) . The pole orders give n = d n = d n = d and A ∈ fil d A \in\operatorname{fil}_{d} A ∈ fil d . We have α + c b ′ = 0 \alpha+ cb' = 0 α + c b ′ = 0 and ν ( b ) = ν ( c ) = 0 \nu(b) = \nu(c) = 0 ν ( b ) = ν ( c ) = 0 , since the leading exponents − 1 -1 − 1 and − k -k − k are prime to p p p . Hence the last inequality is 0 < ( p − 2 ) d 0 < (p - 2)d 0 < ( p − 2 ) d . The source map b = x b = x b = x has degree one and the target map c = − F ′ ( x ) c = -F'(x) c = − F ′ ( x ) has local degree k k k prime to p p p . Both are finite on the henselian traits, generically étale, and have poles at v = 0 v = 0 v = 0 . The source pushforward has the single slope d > 1 d > 1 d > 1 . The formula therefore gives the pushforward along c c c of the character with phase A + b c = F − x F ′ A + bc = F - xF' A + b c = F − x F ′ up to tame factors.
Choose a tame coordinate z z z with c = − z k / 2 c = -z^k/2 c = − z k /2 . Writing h = d − e = P ( P − 1 ) / 2 h = d - e = P(P - 1)/2 h = d − e = P ( P − 1 ) /2 , we have z k = x k ( 1 + x − h ) z^k = x^k(1 + x^{-h}) z k = x k ( 1 + x − h ) , hence z = x ( 1 + x − h ) 1 / k z = x(1 + x^{-h})^{1/k} z = x ( 1 + x − h ) 1/ k and x = z + 2 z 1 − h + O ( z 1 − 2 h ) x = z + 2z^{1-h} + O(z^{1-2h}) x = z + 2 z 1 − h + O ( z 1 − 2 h ) ; here 1 / k = − 2 1/k = -2 1/ k = − 2 in characteristic p p p . Since d − 2 h < 0 d - 2h < 0 d − 2 h < 0 and e − h < 0 e - h < 0 e − h < 0 for P ≥ 3 P \geq3 P ≥ 3 ,
x d = z d + z e + O ( z d − 2 h ) , x e = z e + O ( z e − h ) . x^d = z^d + z^e + O(z^{d-2h}), \qquad x^e = z^e + O(z^{e-h}). x d = z d + z e + O ( z d − 2 h ) , x e = z e + O ( z e − h ) .
As 1 − d = 1 − e = 1 / 2 1 - d = 1 - e = 1/2 1 − d = 1 − e = 1/2 in characteristic p p p , F − x F ′ = 1 2 ( x d + x e ) F - xF' = \frac{1}{2}(x^d + x^e) F − x F ′ = 2 1 ( x d + x e ) , and therefore its polar part is exactly
F ( x ) − x F ′ ( x ) = 1 2 z d + z e ( m o d z − 1 F ‾ P [ [ z − 1 ] ] ) . F(x) - xF'(x) = \frac{1}{2}z^d + z^e \pmod{z^{-1}\overline{\mathbb{F}}_P[[z^{-1}]]}. F ( x ) − x F ′ ( x ) = 2 1 z d + z e ( mod z − 1 F P [[ z − 1 ]]) .
After the tame degree-k k k pullback, its conjugate wild characters are
L ψ ( 1 2 ζ j z d + ζ e j z e ) , ζ ∈ μ k . \mathcal{L}_{\psi}\left(\frac{1}{2}\zeta^jz^d + \zeta^{ej}z^e\right), \qquad\zeta\in\mu_k. L ψ ( 2 1 ζ j z d + ζ e j z e ) , ζ ∈ μ k .
We now prove the linear independence assertion used to determine the wild image, rather than only citing the two coefficient fields. Fix a primitive k k k th root of unity ζ \zeta ζ and recall that k = e ( P − 1 ) k = e(P - 1) k = e ( P − 1 ) and d = k + 1 d = k + 1 d = k + 1 . The conjugate Artin–Schreier classes have pole terms
1 2 ζ j z d + ζ e j z e , j ∈ Z / k Z . \frac{1}{2}\zeta^jz^d + \zeta^{ej}z^e, \qquad j \in\mathbb{Z}/k\mathbb{Z}. 2 1 ζ j z d + ζ e j z e , j ∈ Z / k Z .
Because p ∤ d e p \nmid de p ∤ d e and d ≠ e d \ne e d = e , a nonzero linear combination of z d z^d z d and z e z^e z e cannot be an Artin–Schreier coboundary: the pole order of a nontrivial coboundary with a pole is divisible by p p p . Thus their F p \mathbb{F}_p F p -linear relations are exactly the relations between the coefficient pairs
( ζ j , ζ e j ) ∈ F P 2 ⊕ F P . (\zeta^j,\zeta^{ej}) \in\mathbb{F}_P^2 \oplus\mathbb{F}_P. ( ζ j , ζ e j ) ∈ F P 2 ⊕ F P .
The scalar 1 / 2 1/2 1/2 in the first component is invertible and hence irrelevant to their span. The order of p p p modulo k k k is 2 f 2f 2 f : it divides 2 f 2f 2 f because k ∣ ( P 2 − 1 ) k \mid(P^2 - 1) k ∣ ( P 2 − 1 ) , and it cannot be a proper divisor of 2 f 2f 2 f , since every proper divisor is at most f f f and k > P = p f k > P = p^f k > P = p f . Similarly, the order of p p p modulo P − 1 P - 1 P − 1 is f f f . Consequently, the minimal polynomials m d ( T ) m_d(T) m d ( T ) and m e ( T ) m_e(T) m e ( T ) of ζ \zeta ζ and ζ e \zeta^e ζ e over F p \mathbb{F}_p F p are distinct irreducibles of degrees 2 f 2f 2 f and f f f . By the Chinese remainder theorem, evaluation induces a surjection
F p [ T ] ⟶ F P 2 ⊕ F P , h ( T ) ⟼ ( h ( ζ ) , h ( ζ e ) ) , \mathbb{F}_p[T] \longrightarrow\mathbb{F}_{P^2} \oplus\mathbb{F}_P, \qquad h(T) \longmapsto(h(\zeta),h(\zeta^e)), F p [ T ] ⟶ F P 2 ⊕ F P , h ( T ) ⟼ ( h ( ζ ) , h ( ζ e )) ,
whose kernel is ( m d m e ) (m_dm_e) ( m d m e ) . Since ζ k = 1 \zeta^k = 1 ζ k = 1 , every evaluation is in the span of the k k k pairs displayed above. Their span therefore has dimension $2f + f = 3f$, and, as a module for the tame cyclic quotient, splits into the two distinct irreducible summands coming from m d m_d m d and $m_e.
After the tame degree-k k k pullback, the wild inertia representation is the direct sum of the above rank-one characters. Let U U U be its finite wild image and let M ⊆ U ∨ M \subseteq U^\vee M ⊆ U ∨ be the subgroup generated by these characters. They separate points of U U U because U U U is defined as the image on the small Weil constituent; hence M = U ∨ M = U^\vee M = U ∨ . The calculation just made gives
M = M d ⊕ M e , dim F p M d = 2 f , dim F p M e = f . M = M_d \oplus M_e, \qquad\dim_{\mathbb{F}_p} M_d = 2f, \qquad\dim_{\mathbb{F}_p} M_e = f. M = M d ⊕ M e , dim F p M d = 2 f , dim F p M e = f .
In particular, U ∨ U^\vee U ∨ is elementary abelian of order p 3 f p^{3f} p 3 f , so U U U is elementary abelian of order P 3 P^3 P 3 .
The conductor of a nontrivial class in M e M_e M e is e e e , while a class with nonzero M d M_d M d -component has conductor d d d : the nonzero leading pole cannot be removed by an Artin–Schreier coboundary, since p ∤ d e p\nmid de p ∤ d e . Thus the lower-break characters form exactly M e M_e M e .
For a finite abelian p p p -group U U U and any subgroup H ⊂ U H\subset U H ⊂ U , Pontryagin duality recovers H H H as the intersection of the kernels of all characters that are trivial on H H H . The subgroup M e M_e M e consists exactly of the characters whose restriction has conductor at most the smaller of the two breaks. Its common kernel therefore detects the part of U U U still visible after the first break. Applying this to the upper ramification groups, the subgroup surviving between the two breaks is
U 0 = ⋂ χ ∈ M e ker χ , ( U / U 0 ) ∨ = M e . U_0=\bigcap_{\chi\in M_e}\ker\chi,\qquad(U/U_0)^\vee=M_e. U 0 = χ ∈ M e ⋂ ker χ , ( U / U 0 ) ∨ = M e .
In particular
∣ U ∣ = ∣ M ∣ = p 3 f = P 3 , ∣ U / U 0 ∣ = ∣ M e ∣ = p f = P , ∣ U 0 ∣ = P 2 . |U|=|M|=p^{3f}=P^3,\qquad|U/U_0|=|M_e|=p^f=P,\qquad|U_0|=P^2. ∣ U ∣ = ∣ M ∣ = p 3 f = P 3 , ∣ U / U 0 ∣ = ∣ M e ∣ = p f = P , ∣ U 0 ∣ = P 2 .
One should distinguish this from the fact that each individual rank-one summand can have conductor d d d : the smaller break is detected by products of diagonal characters in which the highest pole cancels. Finally, upper breaks are divided by k k k when descending from the tame degree-k k k extension. Thus the two positive breaks on the original parameter line are
e k = 1 P − 1 , d k = P 2 + 1 P 2 − 1 . \frac{e}{k}=\frac{1}{P-1},\qquad\frac{d}{k}=\frac{P^2+1}{P^2-1}. k e = P − 1 1 , k d = P 2 − 1 P 2 + 1 .
and the successive positive-level images are U U U , U 0 U_0 U 0 , and 1 1 1 . The faithfulness argument preceding this lemma now transfers the filtration of the image of wild inertia to the natural root space. In particular, it transfers the intermediate subgroup and both breaks, not just the order of the full wild image.
Lemma 3.5 (Inertia in the Siegel parabolic). The inertia image is I = U ⋊ ⟨ s ⟩ I=U\rtimes\langle s\rangle I = U ⋊ ⟨ s ⟩ , where U = U S i e g U=U_{\mathrm{Sieg}} U = U Sieg and s s s acts on W W W as a Singer cycle of order P 2 − 1 P^2-1 P 2 − 1 . Under the Singer action on U U U , the subgroup U 0 U_0 U 0 of Lemma 3.4 is the two-dimensional summand of Sym 2 ( F P ) \operatorname{Sym}_2(\mathbb{F}_P) Sym 2 ( F P ) .
Proof. The Fourier-transform computation above, together with the faithfulness and kernel comparison, shows that the image U U U of wild inertia on the natural root space has order P 3 P^3 P 3 . Lemma 3.3 places it inside the Siegel radical U S i e g U_{\mathrm{Sieg}} U Sieg , which also has order P 3 P^3 P 3 because
U S i e g ≃ Sym 2 ( F P ) U_{\mathrm{Sieg}}\simeq\operatorname{Sym}_2(\mathbb{F}_P) U Sieg ≃ Sym 2 ( F P )
as an additive group. Therefore U = U S i e g U=U_{\mathrm{Sieg}} U = U Sieg . The tame Singer group has order prime to p p p ; the resulting extension splits by Schur–Zassenhaus. In particular, the full inertia image is, up to conjugacy in G G G ,
I = U ⋊ ⟨ s ⟩ , ∣ U ∣ = P 3 , ∣ s ∣ = P 2 − 1. I=U\rtimes\langle s\rangle,\qquad|U|=P^3,\qquad|s|=P^2-1. I = U ⋊ ⟨ s ⟩ , ∣ U ∣ = P 3 , ∣ s ∣ = P 2 − 1.
This identifies the inertia subgroup as a concrete subgroup of Sp 4 ( F P ) \operatorname{Sp}_4(\mathbb{F}_P) Sp 4 ( F P ) , rather than only determining its abstract order.
The character calculation in Lemma 3.4 gives the tame-stable filtration of the wild image
U ⊃ U 0 ⊃ { 1 } , ∣ U 0 ∣ = P 2 , U\supset U_0\supset\{1\},\qquad|U_0|=P^2, U ⊃ U 0 ⊃ { 1 } , ∣ U 0 ∣ = P 2 ,
where U 0 U_0 U 0 is the annihilator of the lower-break character space M e ⊂ U ∨ M_e\subset U^\vee M e ⊂ U ∨ . In particular, U 0 U_0 U 0 is not obtained from the Newton polygon alone: its order and break come from the local Fourier computation. Under the Singer action on U ≃ Sym 2 ( F P ) U\simeq\operatorname{Sym}_2(\mathbb{F}_P) U ≃ Sym 2 ( F P ) , the three-dimensional F P \mathbb{F}_P F P -space decomposes into a one-dimensional norm-character summand and a two-dimensional summand, the latter being irreducible under the Singer cycle. These summands are irreducible and non-isomorphic even over F P \mathbb{F}_P F P . Indeed, if the Singer generator has eigenvalue β \beta β on W W W over F P 2 \mathbb{F}_{P^2} F P 2 , the eigenvalues on symmetric forms are β P + 1 \beta^{P+1} β P + 1 for the norm summand and β 2 , β 2 P \beta^2,\beta^{2P} β 2 , β 2 P for the other summand. Their respective minimal-polynomial degrees over F P \mathbb{F}_P F P are f f f and 2 f 2f 2 f by the same order calculation used above for P − 1 P-1 P − 1 and k k k . As the Singer group has order prime to p p p , its only invariant subgroup of order P 2 P^2 P 2 in U U U is the latter summand, which therefore equals U 0 U_0 U 0 . This explains the identification of U 0 U_0 U 0 used in subsequent character averages.
For a subgroup J J J and a representation E E E of G G G , write E J E^J E J for its fixed subspace and Sw ∞ ( E ) \operatorname{Sw}_\infty(E) Sw ∞ ( E ) for the Swan conductor at infinity.
Lemma 3.6 (Swan conductor). For every characteristic-zero representation E E E of G G G ,
Sw ∞ ( E ) = dim E − dim E U P − 1 + P P + 1 ( dim E − dim E U 0 ) . (8) \operatorname{Sw}_\infty(E)=\frac{\dim E-\dim E^U}{P-1}+\frac{P}{P+1}(\dim E-\dim E^{U_0}). \tag*{(8)} Sw ∞ ( E ) = P − 1 dim E − dim E U + P + 1 P ( dim E − dim E U 0 ) . (8)
Proof. By Lemma 3.4 , the successive positive-level images are U U U , U 0 U_0 U 0 , and 1 1 1 . Integrating dim E − dim E J \dim E-\dim E^J dim E − dim E J over positive upper levels gives b 1 ( dim E − dim E U ) + ( b 2 − b 1 ) ( dim E − dim E U 0 ) b_1(\dim E-\dim E^U)+(b_2-b_1)(\dim E-\dim E^{U_0}) b 1 ( dim E − dim E U ) + ( b 2 − b 1 ) ( dim E − dim E U 0 ) . Here b 1 = 1 / ( P − 1 ) b_1=1/(P-1) b 1 = 1/ ( P − 1 ) and b 2 − b 1 = P / ( P + 1 ) b_2-b_1=P/(P+1) b 2 − b 1 = P / ( P + 1 ) . □
Cohomological factors
The purpose of the preceding inertia computation is to determine the cohomological degrees of the trace factors used below. Every representation of S = G / { ± I } S=G/\{\pm I\} S = G / { ± I } , pulled back along the monodromy quotient, gives an ℓ \ell ℓ -adic local system on the parameter line. For a character χ \chi χ of S S S , denote it by E χ \mathcal{E}_\chi E χ , with coefficients in Q ℓ \mathbb{Q}_\ell Q ℓ for ℓ ≠ p \ell\ne p ℓ = p , and set
L χ ( u ) = det ( 1 − u F ∣ H c 1 ( A F p 1 , E χ ) ) . L_\chi(u)=\det\left(1-uF\mid H^1_c\left(\mathbb{A}^1_{\mathbb{F}_p},\mathcal{E}_\chi\right)\right). L χ ( u ) = det ( 1 − u F ∣ H c 1 ( A F p 1 , E χ ) ) .
where F F F is the geometric Frobenius of the base field. For a nontrivial irreducible χ \chi χ , full geometric monodromy implies H c 0 = H c 2 = 0 H^0_c=H^2_c=0 H c 0 = H c 2 = 0 . The Grothendieck trace formula then gives
∑ a ∈ F p H χ ( F a , H ) = − Tr ( F H ∣ H c 1 ( A F p 1 , E χ ) ) . \sum_{a\in\mathbb{F}_p^H}\chi(F_{a,H})=-\operatorname{Tr}\left(F^H\mid H^1_c\left(\mathbb{A}^1_{\mathbb{F}_p},\mathcal{E}_\chi\right)\right). a ∈ F p H ∑ χ ( F a , H ) = − Tr ( F H ∣ H c 1 ( A F p 1 , E χ ) ) .
Thus the finite-field character sums are encoded by the Frobenius polynomial L χ L_\chi L χ . The Swan conductor formula (8) , together with Grothendieck–Ogg–Shafarevich, will determine the degrees of these polynomials.
Cohomological degrees and duality for P = 3 P=3 P = 3
In this section P = 3 P=3 P = 3 , so S ≃ PSU 4 ( 2 ) S\simeq\operatorname{PSU}_4(2) S ≃ PSU 4 ( 2 ) . Denote by χ 6 , χ 24 , χ 64 , χ 81 \chi_6,\chi_{24},\chi_{64},\chi_{81} χ 6 , χ 24 , χ 64 , χ 81 its irreducible characters with the indicated degrees. The last two are the Steinberg characters in the unitary and symplectic descriptions respectively. By the specialization of the uniform Deligne–Lusztig identity (4) (see Remark 2.3),
5 1 i r r = 1 + χ 6 − χ 24 − χ 64 + χ 81 . 5\,1_{\mathrm{irr}}=1+\chi_6-\chi_{24}-\chi_{64}+\chi_{81}. 5 1 irr = 1 + χ 6 − χ 24 − χ 64 + χ 81 .
In S S S , irreducibility is equivalent to membership in its unique order-five class.
The purpose of this identity is to convert the condition that a Frobenius element act irreducibly into character sums. Indeed,
A H ( 3 ) = ∑ a ∈ F 3 H 1 i r r ( F a , H ) , A_H(3)=\sum_{a\in\mathbb{F}_3^H}1_{\mathrm{irr}}(F_{a,H}), A H ( 3 ) = a ∈ F 3 H ∑ 1 irr ( F a , H ) ,
so the right-hand side of (5) can be treated by the Grothendieck trace formula. We first compute the dimensions of the inertia-invariant subspaces needed for this passage.
Identify U U U with symmetric 2 × 2 2\times2 2 × 2 matrices B B B over F 3 \mathbb{F}_3 F 3 via
u B = ( 1 B 0 1 ) . u_B=\begin{pmatrix}1&B\\0&1\end{pmatrix}. u B = ( 1 0 B 1 ) .
Choose the Singer cycle on W W W as A = ( 1 2 1 1 ) A=\begin{pmatrix}1&2\\1&1\end{pmatrix} A = ( 1 1 2 1 ) ; A T A = 2 I A^{\mathsf{T}}A=2I A T A = 2 I and U 0 U_0 U 0 is the trace-zero plane. The needed values and numbers of elements are
Table 1.
The rank-two class labels can be checked without a naming convention: the trace on the small Weil representation is ( S B − 1 ) / 2 (S_B-1)/2 ( S B − 1 ) /2 , where S B S_B S B is the two-variable quadratic Gauss sum. It is $-2 for det B = 1 \det B = 1 det B = 1 and 1 1 1 for det B = 2 \det B = 2 det B = 2 . The character of its exterior square, χ 6 \chi_6 χ 6 , is therefore 3 3 3 and 0 0 0 respectively; for rank one it is − 3 -3 − 3 .
The rank-one entries aggregate two congruence classes: for 2 × 2 2 \times2 2 × 2 symmetric matrices of odd characteristic, congruence classes are distinguished by rank and, in each positive rank, by the discriminant square class. The two rank-one classes give the same displayed character values, so they have been combined in one column. The two nonsingular columns are distinguished by the determinant square class in F 3 × \mathbb{F}_3^\times F 3 × .
For a representation E E E of a finite group J J J , averaging the operators in J J J gives the projector onto E J E^J E J . Hence
dim E J = 1 ∣ J ∣ ∑ g ∈ J χ E ( g ) . \dim E^J = \frac{1}{|J|} \sum_{g \in J} \chi_E(g). dim E J = ∣ J ∣ 1 g ∈ J ∑ χ E ( g ) .
For example, the first two rows of the table yield
dim E 6 U = 6 + 8 ( − 3 ) + 6 ( 3 ) + 12 ( 0 ) 27 = 0 , dim E 6 U 0 = 6 + 4 ( 3 ) + 4 ( 0 ) 9 = 2. \dim E^U_6 = \frac{6 + 8(-3) + 6(3) + 12(0)}{27} = 0, \qquad\dim E^{U_0}_6 = \frac{6 + 4(3) + 4(0)}{9} = 2. dim E 6 U = 27 6 + 8 ( − 3 ) + 6 ( 3 ) + 12 ( 0 ) = 0 , dim E 6 U 0 = 9 6 + 4 ( 3 ) + 4 ( 0 ) = 2.
The remaining invariant dimensions follow by the same averaging calculation. We then insert them into the Swan conductor formula (8) .
We write E j \mathcal{E}_j E j and L j L_j L j for the objects attached to χ j \chi_j χ j . For these nontrivial characters, full geometric monodromy implies H c 0 = H c 2 = 0 H^0_c = H^2_c = 0 H c 0 = H c 2 = 0 . Since the local systems are lisse on the affine line, Grothendieck–Ogg–Shafarevich gives
χ c ( A 1 , E j ) = dim E j − Sw ∞ ( E j ) . \chi_c(\mathbb{A}^1,\mathcal{E}_j) = \dim E_j - \operatorname{Sw}_{\infty}(E_j). χ c ( A 1 , E j ) = dim E j − Sw ∞ ( E j ) .
Consequently,
dim H c 1 ( A 1 , E j ) = Sw ∞ ( E j ) − dim E j . \dim H^1_c(\mathbb{A}^1,\mathcal{E}_j) = \operatorname{Sw}_{\infty}(E_j) - \dim E_j. dim H c 1 ( A 1 , E j ) = Sw ∞ ( E j ) − dim E j .
For instance, when j = 64 j = 64 j = 64 the table above gives dim E 64 U = 0 \dim E^{U}_{64} = 0 dim E 64 U = 0 and dim E 64 U 0 = 8 \dim E^{U_0}_{64} = 8 dim E 64 U 0 = 8 , whence
Sw ∞ ( E 64 ) = 64 2 + 3 4 ( 64 − 8 ) = 74 , dim H c 1 ( A 1 , E 64 ) = 10. \operatorname{Sw}_{\infty}(E_{64}) = \frac{64}{2} + \frac{3}{4}(64 - 8) = 74, \qquad\dim H^1_c(\mathbb{A}^1,\mathcal{E}_{64}) = 10. Sw ∞ ( E 64 ) = 2 64 + 4 3 ( 64 − 8 ) = 74 , dim H c 1 ( A 1 , E 64 ) = 10.
The same calculation for the other three characters yields
Table 4.1.
The last column determines the degrees of the cohomological Frobenius polynomials; it does not yet determine their coefficients. Thus L 6 = 1 L_6 = 1 L 6 = 1 . At H = 1 H = 1 H = 1 , a = 0 , 1 , 2 a = 0, 1, 2 a = 0 , 1 , 2 give projective orders 9 , 5 , 12 9, 5, 12 9 , 5 , 12 , and the χ 24 \chi_{24} χ 24 values are 0 , − 1 , 0 0, -1, 0 0 , − 1 , 0 . The trace formula gives L 24 = 1 − u L_{24} = 1 - u L 24 = 1 − u .
We next justify the reciprocal symmetries used to determine L 64 L_{64} L 64 and L 81 L_{81} L 81 . These symmetries reduce the number of coefficients that must be recovered from the exact finite-field counts in the next section. For E 64 E_{64} E 64 , (4.1) already implies E 64 I = 0 E^I_{64} = 0 E 64 I = 0 . For E 81 E_{81} E 81 , the U U U -invariants of the symplectic Steinberg representation are the Steinberg representation of the Levi GL 2 ( 3 ) \operatorname{GL}_2(3) GL 2 ( 3 ) . The Singer subgroup of order eight has trace 3 3 3 at its two scalar elements and − 1 -1 − 1 at its other six elements. Its invariant dimension is ( 2 ⋅ 3 − 6 ) / 8 = 0 (2 \cdot3 - 6)/8 = 0 ( 2 ⋅ 3 − 6 ) /8 = 0 . Equivalently, this follows by averaging the Steinberg character on the four cosets in I / { ± 1 } U I/\{\pm1\}U I / { ± 1 } U . Hence both local systems have no inertia invariants at infinity. Their compactly supported first cohomology equals their middle-extension cohomology and is pure of weight one. The Steinberg representation has a symmetric invariant form, for example from its real realization in the building. The regular cuspidal character χ 64 \chi_{64} χ 64 is self-dual: the contragredient of R T G ( θ ) R^G_T(\theta) R T G ( θ ) has parameter θ − 1 \theta^{-1} θ − 1 , and θ − 1 = θ P 2 \theta^{-1} = \theta^{P^2} θ − 1 = θ P 2 is in the same rational Weyl orbit. Poincaré duality therefore pairs the Frobenius eigenvalues of each cohomology group by α ↦ 3 / α \alpha\mapsto3/\alpha α ↦ 3/ α . For the Steinberg factor, the induced cohomology pairing is alternating with Frobenius multiplier three, so the determinant has positive sign. For the cuspidal factor we retain both signs initially. Writing L 81 = ∑ n i u i L_{81} = \sum n_i u^i L 81 = ∑ n i u i and L 64 = ∑ d i u i L_{64} = \sum d_i u^i L 64 = ∑ d i u i gives
n 12 − i = 3 6 − i n i ( 0 ≤ i ≤ 6 ) , d 10 − i = ε 3 5 − i d i ( 0 ≤ i ≤ 5 ) , ε ∈ { 1 , − 1 } . (9) n_{12-i} = 3^{6-i} n_i \quad(0 \leq i \leq6), \qquad d_{10-i} = \varepsilon3^{5-i} d_i \quad(0 \leq i \leq5), \qquad\varepsilon\in\{1,-1\}. \tag*{(9)} n 12 − i = 3 6 − i n i ( 0 ≤ i ≤ 6 ) , d 10 − i = ε 3 5 − i d i ( 0 ≤ i ≤ 5 ) , ε ∈ { 1 , − 1 } . (9)
The finite reconstruction in § 2.4 will also determine ε \varepsilon ε .
Finally, sum (5) over all parameters a ∈ F 3 H a \in\mathbb{F}_{3^H} a ∈ F 3 H . The trivial character contributes 3 H 3^H 3 H , while L 6 = 1 L_6 = 1 L 6 = 1 makes the χ 6 \chi_6 χ 6 -sum vanish. Since L 24 = 1 − u L_{24} = 1-u L 24 = 1 − u , the χ 24 \chi_{24} χ 24 -sum equals − 1 -1 − 1 . For j = 64 , 81 j = 64,81 j = 64 , 81 , the trace formula identifies the corresponding character sum with the negative of the Frobenius trace on H c 1 H^1_c H c 1 . We obtain
5 A H ( 3 ) = 3 H + 1 − Tr ( F H ∣ H c 1 ( A 1 , E 81 ) ) + Tr ( F H ∣ H c 1 ( A 1 , E 64 ) ) . (10) 5A_H(3) = 3^H + 1 - \operatorname{Tr}(F^H \mid H^1_c(\mathbb{A}^1,\mathcal{E}_{81})) + \operatorname{Tr}(F^H \mid H^1_c(\mathbb{A}^1,\mathcal{E}_{64})). \tag*{(10)} 5 A H ( 3 ) = 3 H + 1 − Tr ( F H ∣ H c 1 ( A 1 , E 81 )) + Tr ( F H ∣ H c 1 ( A 1 , E 64 )) . (10)
In particular the degrees 12 and 10 in this calculation are theoretical dimensions, not degrees selected by a fitting procedure.
Exact reconstruction for P = 3 P = 3 P = 3
The preceding section determines the degrees and reciprocal symmetries of L 81 L_{81} L 81 and L 64 L_{64} L 64 , but not their coefficients. We now explain how a finite collection of exact atom counts determines both polynomials uniquely. The role of the computation is not to guess a recurrence: the cohomological dimension and duality arguments already supply the finite-dimensional constraints that make the reconstruction rigorous.
Exact enumeration of atoms
For each H H H and each a ∈ F 3 H a \in\mathbb{F}_{3^H} a ∈ F 3 H , consider the companion product
M a , H = C a C a 3 ⋯ C a 3 H − 1 M_{a,H} = C_a C_{a^3} \cdots C_{a^{3^{H-1}}} M a , H = C a C a 3 ⋯ C a 3 H − 1
from Section 2. By the companion-module criterion, f a f_a f a is an atom if and only if the characteristic polynomial of M a , H M_{a,H} M a , H is irreducible over F 3 \mathbb{F}_3 F 3 . Since M a , H M_{a,H} M a , H is symplectic, its characteristic polynomial has the self-reciprocal form
T 4 + c 1 T 3 + c 2 T 2 + c 1 T + 1. T^4 + c_1T^3 + c_2T^2 + c_1T + 1. T 4 + c 1 T 3 + c 2 T 2 + c 1 T + 1.
The only irreducible self-reciprocal quartics over F 3 \mathbb{F}_3 F 3 are
T 4 + T 3 + T 2 + T + 1 , T 4 + 2 T 3 + T 2 + 2 T + 1. T^4 + T^3 + T^2 + T + 1,\qquad T^4 + 2T^3 + T^2 + 2T + 1. T 4 + T 3 + T 2 + T + 1 , T 4 + 2 T 3 + T 2 + 2 T + 1.
Thus the trace and second characteristic coefficient suffice to recognize the two permissible characteristic polynomials. In particular, the test can be performed using finite-field arithmetic alone, without factoring a skew polynomial for every parameter.
For reproducibility, use F 3 H = F 3 [ X ] / ( X H + ∑ i = 0 H − 1 b i X i ) \mathbb{F}_{3^H} = \mathbb{F}_3[X]/(X^H + \sum_{i=0}^{H-1} b_iX^i) F 3 H = F 3 [ X ] / ( X H + ∑ i = 0 H − 1 b i X i ) . The second row below records the integers ∑ i = 0 H − 1 b i 3 i \sum_{i=0}^{H-1} b_i3^i ∑ i = 0 H − 1 b i 3 i , with b i ∈ { 0 , 1 , 2 } b_i \in\{0,1,2\} b i ∈ { 0 , 1 , 2 } . These moduli are irreducible over F 3 \mathbb{F}_3 F 3 .
Table 2.
Loop over the field elements a a a , form M a , H M_{a,H} M a , H using the successive cubes of a a a , and count an element when
Tr ( M a , H ) ∈ { 1 , 2 } , Tr ( M a , H ) 2 − Tr ( M a , H 2 ) 2 = 1. \operatorname{Tr}(M_{a,H}) \in\{1,2\},\qquad\frac{\operatorname{Tr}(M_{a,H})^2-\operatorname{Tr}(M_{a,H}^2)}{2}=1. Tr ( M a , H ) ∈ { 1 , 2 } , 2 Tr ( M a , H ) 2 − Tr ( M a , H 2 ) = 1.
Equivalently, one may take one representative per cubing orbit and weight each accepted representative by its orbit size. The companion products for consecutive elements of a cubing orbit are conjugate by cyclic rotation of the invertible factors.
From point counts to a determinant quotient
Put
b H = 3 H + 1 − 5 A H ( 3 ) . b_H = 3^H + 1 - 5A_H(3). b H = 3 H + 1 − 5 A H ( 3 ) .
By the trace identity (10) ,
b H = Tr ( F H ∣ H c 1 ( A 1 , E 81 ) ) − Tr ( F H ∣ H c 1 ( A 1 , E 64 ) ) . b_H = \operatorname{Tr}(F^H \mid H^1_c(\mathbb{A}^1,\mathcal{E}_{81})) - \operatorname{Tr}(F^H \mid H^1_c(\mathbb{A}^1,\mathcal{E}_{64})). b H = Tr ( F H ∣ H c 1 ( A 1 , E 81 )) − Tr ( F H ∣ H c 1 ( A 1 , E 64 )) .
Recall the formal trace–determinant identity
det ( 1 − u T ) = exp ( − ∑ H ≥ 1 Tr ( T H ) H u H ) (11) \det(1-uT) = \exp\left(-\sum_{H\geq1}\frac{\operatorname{Tr}(T^H)}{H}u^H\right) \tag*{(11)} det ( 1 − u T ) = exp ( − H ≥ 1 ∑ H Tr ( T H ) u H ) (11)
for any endomorphism T T T of a finite-dimensional vector space in characteristic zero. Consequently the formal series
R ( u ) : = exp ( − ∑ H ≥ 1 b H H u H ) = L 81 ( u ) L 64 ( u ) (12) R(u) := \exp\left(-\sum_{H\geq1}\frac{b_H}{H}u^H\right)=\frac{L_{81}(u)}{L_{64}(u)} \tag*{(12)} R ( u ) := exp ( − H ≥ 1 ∑ H b H u H ) = L 64 ( u ) L 81 ( u ) (12)
is determined by the atom counts. We write R ( u ) = ∑ h ≥ 0 r h u h R(u)=\sum_{h\geq0}r_hu^h R ( u ) = ∑ h ≥ 0 r h u h with r 0 = 1 r_0=1 r 0 = 1 . Logarithmic differentiation gives
u R ′ ( u ) = − R ( u ) ∑ j ≥ 1 b j u j uR'(u)=-R(u)\sum_{j\geq1}b_ju^j u R ′ ( u ) = − R ( u ) j ≥ 1 ∑ b j u j
and hence the recursion
h r h = − ∑ j = 1 h b j r h − j ( h ≥ 1 ) . (13) hr_h=-\sum_{j=1}^{h}b_jr_{h-j}\qquad(h\geq1). \tag*{(13)} h r h = − j = 1 ∑ h b j r h − j ( h ≥ 1 ) . (13)
For instance, b 1 = − 1 b_1=-1 b 1 = − 1 and b 2 = − 5 b_2=-5 b 2 = − 5 , so that r 1 = 1 r_1=1 r 1 = 1 and r 2 = 3 r_2=3 r 2 = 3 . All calculations take place over Q \mathbb{Q} Q ; no floating-point approximation enters the reconstruction.
Uniqueness from reciprocal duality
Write
L 81 ( u ) = ∑ i = 0 12 n i u i , L 64 ( u ) = ∑ i = 0 10 d i u i , n 0 = d 0 = 1. L_{81}(u)=\sum_{i=0}^{12}n_iu^i,\qquad L_{64}(u)=\sum_{i=0}^{10}d_iu^i,\qquad n_0=d_0=1. L 81 ( u ) = i = 0 ∑ 12 n i u i , L 64 ( u ) = i = 0 ∑ 10 d i u i , n 0 = d 0 = 1.
Use the relations (9) . For ε = 1 \varepsilon=1 ε = 1 , the unknowns are n 1 , … , n 6 n_1,\ldots,n_6 n 1 , … , n 6 and d 1 , … , d 5 d_1,\ldots,d_5 d 1 , … , d 5 . For ε = − 1 \varepsilon=-1 ε = − 1 , the middle coefficient d 5 d_5 d 5 vanishes, leaving ten unknowns. By (12) ,
L 64 ( u ) R ( u ) = L 81 ( u ) . L_{64}(u)R(u)=L_{81}(u). L 64 ( u ) R ( u ) = L 81 ( u ) .
Comparison of coefficients of u h u^h u h , for 1 ≤ h ≤ 11 1\leq h\leq11 1 ≤ h ≤ 11 , yields
∑ i = 0 min ( h , 10 ) d i r h − i = n h , 1 ≤ h ≤ 11. (14) \sum_{i=0}^{\min(h,10)}d_ir_{h-i}=n_h,\qquad1\leq h\leq11. \tag*{(14)} i = 0 ∑ m i n ( h , 10 ) d i r h − i = n h , 1 ≤ h ≤ 11. (14)
where coefficients with indices greater than half the respective degrees are replaced using reciprocity. The r h r_h r h occurring here are obtained solely from A 1 ( 3 ) , … , A 11 ( 3 ) A_1(3),\ldots,A_{11}(3) A 1 ( 3 ) , … , A 11 ( 3 ) by (13) . For ε = − 1 \varepsilon=-1 ε = − 1 , order the ten unknowns as ( n 1 , … , n 6 , d 1 , … , d 4 ) (n_1,\ldots,n_6,d_1,\ldots,d_4) ( n 1 , … , n 6 , d 1 , … , d 4 ) and write the equations with left-hand side ∑ i d i r h − i − n h \sum_i d_ir_{h-i}-n_h ∑ i d i r h − i − n h . The 11 × 11 11\times11 11 × 11 matrix formed by appending the right-hand side to the coefficient matrix has determinant 461 761 074 461\,761\,074 461 761 074 , so that system is inconsistent. Hence ε = 1 \varepsilon=1 ε = 1 .
For this sign, in the ordering ( n 1 , … , n 6 , d 1 , … , d 5 ) (n_1,\ldots,n_6,d_1,\ldots,d_5) ( n 1 , … , n 6 , d 1 , … , d 5 ) , the coefficient matrix has determinant − 6 975 000 -6\,975\,000 − 6 975 000 , which is nonzero. Therefore these counts determine the two polynomials uniquely. Solving gives
( n 1 , … , n 6 ) = ( 1 , 2 , 1 , 2 , 17 , 26 ) , ( d 1 , … , d 5 ) = ( 0 , − 1 , − 3 , 3 , 18 ) . (n_1,\ldots,n_6)=(1,2,1,2,17,26),\qquad(d_1,\ldots,d_5)=(0,-1,-3,3,18). ( n 1 , … , n 6 ) = ( 1 , 2 , 1 , 2 , 17 , 26 ) , ( d 1 , … , d 5 ) = ( 0 , − 1 , − 3 , 3 , 18 ) .
The remaining coefficients are fixed by reciprocity, yielding L 81 = N L_{81}=N L 81 = N and L 64 = D L_{64}=D L 64 = D as in Theorem 2.4. Substitution into (12) then proves the formula for A H ( 3 ) A_H(3) A H ( 3 ) for every H ≥ 1 H\geq1 H ≥ 1 , not merely for the extension degrees in the table.
The value at H = 12 H=12 H = 12 is an additional consistency check: it is not needed for uniqueness, since the eleven coefficients were already determined from H ≤ 11 H\leq11 H ≤ 11 . Substituting the reconstructed polynomials in (12) recovers the twelfth count.
Code and data availability
The table and the exact reconstruction can be reproduced with the companion file verify_counts.py. With NumPy and SymPy installed, run python3 verify_counts.py. The defining field moduli and the count data are also recorded above.
The exact minimal recurrence order for P = 3 P=3 P = 3
Proposition 6.1. The polynomials N N N and D D D in Theorem 2.4 are coprime and separately squarefree over Q \mathbb{Q} Q . The reduced ordinary generating series for A H ( 3 ) A_H(3) A H ( 3 ) has denominator
Q 3 ( u ) = ( 1 − u ) ( 1 − 3 u ) N ( u ) D ( u ) . (15) Q_3(u)=(1-u)(1-3u)N(u)D(u). \tag*{(15)} Q 3 ( u ) = ( 1 − u ) ( 1 − 3 u ) N ( u ) D ( u ) . (15) In particular, A H ( 3 ) A_H(3) A H ( 3 ) has a homogeneous constant-coefficient linear recurrence of minimal order 24 over Q \mathbb{Q} Q , including among recurrences valid only for sufficiently large H H H .
Proof. The Euclidean algorithm over F 7 \mathbb{F}_7 F 7 gives
gcd ( N ‾ , D ‾ ) = 1 , (16) \operatorname{gcd}(\overline{N},\overline{D})=1, \tag*{(16)} gcd ( N , D ) = 1 , (16) and over F 5 \mathbb{F}_5 F 5 it gives
gcd ( N ‾ , N ‾ ′ ) = gcd ( D ‾ , D ‾ ′ ) = 1. (17) \operatorname{gcd}(\overline{N},\overline{N}')=\operatorname{gcd}(\overline{D},\overline{D}')=1. \tag*{(17)} gcd ( N , N ′ ) = gcd ( D , D ′ ) = 1. (17) In addition, direct multiplication gives
N ( u ) = ( 1 + u + 3 u 2 ) N 10 ( u ) , N(u)=(1+u+3u^2)N_{10}(u), N ( u ) = ( 1 + u + 3 u 2 ) N 10 ( u ) , N 10 ( u ) = 1 − u 2 + 2 u 3 + 3 u 4 + 8 u 5 + 9 u 6 + 18 u 7 − 27 u 8 + 243 u 10 . N_{10}(u)=1-u^2+2u^3+3u^4+8u^5+9u^6+18u^7-27u^8+243u^{10}. N 10 ( u ) = 1 − u 2 + 2 u 3 + 3 u 4 + 8 u 5 + 9 u 6 + 18 u 7 − 27 u 8 + 243 u 10 . The polynomial N 10 N_{10} N 10 is irreducible modulo 17: after reducing modulo N 10 N_{10} N 10 over F 17 \mathbb{F}_{17} F 17 one has u 17 10 = u u^{17^{10}}=u u 1 7 10 = u , and the gcds of N 10 N_{10} N 10 with u 17 5 − u u^{17^5}-u u 1 7 5 − u and u 17 2 − u u^{17^2}-u u 1 7 2 − u are both 1. Its degree is preserved modulo 17, so it is irreducible over Q \mathbb{Q} Q .
The leading coefficients of N N N and D D D , respectively, 729 and 243, are nonzero modulo both 5 and 7. Thus these modular computations prove coprimality and squarefreeness over Q \mathbb{Q} Q .
The 12 inverse roots of N N N and the 10 inverse roots of D D D are now distinct within and across the two sets. They all have absolute value 3 \sqrt{3} 3 ; in particular they cannot equal 1 or 3. By Theorem 2.4, 5 A H ( 3 ) 5A_H(3) 5 A H ( 3 ) is a sum of 24 exponential sequences with distinct nonzero bases and nonzero coefficients. Each contributes a simple pole at the reciprocal of its base, giving (15) . If a polynomial in the shift annihilates A H ( 3 ) A_H(3) A H ( 3 ) for all sufficiently large H H H , the Vandermonde determinant applied to 24 consecutive values forces the polynomial to vanish at every base. Conversely, the polynomial with precisely these 24 roots annihilates the sequence and has rational coefficients by (15) . □
Remark 6.2 (Explicit recurrence coefficients). Writing
Q 3 ( u ) = 1 + q 1 u + ⋯ + q 24 u 24 , Q_3(u)=1+q_1u+\cdots+q_{24}u^{24}, Q 3 ( u ) = 1 + q 1 u + ⋯ + q 24 u 24 ,
the recurrence takes the form
A H + 24 ( 3 ) + q 1 A H + 23 ( 3 ) + ⋯ + q 24 A H ( 3 ) = 0 ( H ≥ 1 ) . A_{H+24}(3)+q_1A_{H+23}(3)+\cdots+q_{24}A_H(3)=0 \qquad(H\ge1). A H + 24 ( 3 ) + q 1 A H + 23 ( 3 ) + ⋯ + q 24 A H ( 3 ) = 0 ( H ≥ 1 ) .
The coefficients q i q_i q i are obtained by multiplying the four explicit polynomials in (15) ; no additional counting or cohomological calculation is required. This expresses all further counts from 24 consecutive initial values. The order 24 is a structural statement, rather than merely an upper bound obtained from the cohomological degrees.
Part II. The flag curve and the general counting series
Throughout Part II, assume that P = p f P=p^f P = p f is odd and p ∤ f p\nmid f p ∤ f , except where a direct finite-field identity is stated for every odd prime power. The two factors computed in Part I come from the Steinberg representation and a regular cuspidal representation. We now determine their general counterparts and realize the Steinberg factor as the zeta numerator of a flag curve.
Cohomological degrees and factorization for general P P P
The conductor data of Lemmas 3.5 and 3.6 can be computed from four unipotent character values. Write r + , r − r_{+},r_{-} r + , r − for the two rank-one classes in the Siegel radical and s + , s − s_{+},s_{-} s + , s − for its split and anisotropic rank-two classes. For a representation of dimension v v v and character χ \chi χ , put
a χ = 1 P 3 [ v + P 2 − 1 2 ( χ ( r + ) + χ ( r − ) ) + P ( P 2 − 1 ) 2 χ ( s + ) + P ( P − 1 ) 2 2 χ ( s − ) ] , a_{\chi}=\frac{1}{P^{3}}\left[v+\frac{P^{2}-1}{2}\left(\chi(r_{+})+\chi(r_{-})\right)+\frac{P(P^{2}-1)}{2}\chi(s_{+})+\frac{P(P-1)^{2}}{2}\chi(s_{-})\right], a χ = P 3 1 [ v + 2 P 2 − 1 ( χ ( r + ) + χ ( r − ) ) + 2 P ( P 2 − 1 ) χ ( s + ) + 2 P ( P − 1 ) 2 χ ( s − ) ] ,
b χ = 1 P 2 [ v + P 2 − 1 2 ( χ ( s + ) + χ ( s − ) ) ] . b_{\chi}=\frac{1}{P^{2}}\left[v+\frac{P^{2}-1}{2}\left(\chi(s_{+})+\chi(s_{-})\right)\right]. b χ = P 2 1 [ v + 2 P 2 − 1 ( χ ( s + ) + χ ( s − ) ) ] .
Then a χ = dim E U a_{\chi}=\dim E^{U} a χ = dim E U , b χ = dim E U 0 b_{\chi}=\dim E^{U_{0}} b χ = dim E U 0 . To justify evaluating characters on congruence classes of symmetric forms, note that the Levi subgroup acts by actual conjugation in G G G :
( A 0 0 ( A T ) − 1 ) u B ( A − 1 0 0 A T ) = u A B A T . \begin{pmatrix} A & 0 \\ 0 & (A^{\mathsf T})^{-1} \end{pmatrix}u_{B}\begin{pmatrix} A^{-1} & 0 \\ 0 & A^{\mathsf T} \end{pmatrix}=u_{ABA^{\mathsf T}}. ( A 0 0 ( A T ) − 1 ) u B ( A − 1 0 0 A T ) = u A B A T .
Thus any character is constant on each congruence class. Here s + s_{+} s + and s − s_{-} s − denote the split and anisotropic rank-two forms, respectively; the split condition is that − det B -\det B − det B be a square, rather than necessarily that det B \det B det B be a square. To see the second count, identify the two-dimensional summand of symmetric forms with x ↦ Tr F P 2 / F P ( b x 2 ) x\mapsto\operatorname{Tr}_{\mathbb F_{P^{2}}/\mathbb F_{P}}(bx^{2}) x ↦ Tr F P 2 / F P ( b x 2 ) , b ∈ F P 2 b\in\mathbb F_{P^{2}} b ∈ F P 2 . Every nonzero such form is nonsingular, since its polar form is ( x , y ) ↦ 2 Tr F P 2 / F P ( b x y ) (x,y)\mapsto2\operatorname{Tr}_{\mathbb F_{P^{2}}/\mathbb F_{P}}(bxy) ( x , y ) ↦ 2 Tr F P 2 / F P ( b x y ) and the trace pairing is nondegenerate. Up to a fixed nonzero square class, the determinant of this bilinear form is N F P 2 / F P ( b ) N_{\mathbb F_{P^{2}}/\mathbb F_{P}}(b) N F P 2 / F P ( b ) . Since the norm map is surjective, its two square classes each occur ( P 2 − 1 ) / 2 (P^{2}-1)/2 ( P 2 − 1 ) /2 times. The first count is the elementary congruence classification of symmetric two-by-two matrices. In particular, rank-one matrices have two congruence classes according to the nonzero square class of their diagonal coefficient; their cardinalities are added when the character values coincide. Nondegenerate forms have two classes distinguished by their determinant square class (or equivalently by split versus anisotropic type, with the convention just specified).
For nontrivial irreducible characters of S = PSp 4 ( F P ) S=\operatorname{PSp}_{4}(\mathbb F_{P}) S = PSp 4 ( F P ) , the exact degree of the associated compactly supported factor is
δ χ = v − a χ P − 1 + P P + 1 ( v − b χ ) − v . (18) \delta_{\chi}=\frac{v-a_{\chi}}{P-1}+\frac{P}{P+1}(v-b_{\chi})-v. \tag*{(18)} δ χ = P − 1 v − a χ + P + 1 P ( v − b χ ) − v . (18)
In particular, this is an exact degree formula, not merely an upper bound. Recall the cyclic torus T 0 ⊂ S T_{0}\subset S T 0 ⊂ S of order m = ( P 2 + 1 ) / 2 m=(P^{2}+1)/2 m = ( P 2 + 1 ) /2 introduced in Section 2. For each irreducible character χ \chi χ of S S S , put K χ = m dim E χ T 0 − dim E χ K_{\chi}=m\dim E_{\chi}^{T_{0}}-\dim E_{\chi} K χ = m dim E χ T 0 − dim E χ . The elementary induction identity
4 m 1 i r r = m Ind T 0 S 1 − Reg S = ∑ χ K χ χ 4m\,\mathbf{1}_{\mathrm{irr}}=m\operatorname{Ind}_{T_{0}}^{S}\mathbf{1}-\operatorname{Reg}_{S}=\sum_{\chi}K_{\chi}\chi 4 m 1 irr = m Ind T 0 S 1 − Reg S = χ ∑ K χ χ
then gives the uniform finite factorization
R P ( u ) = ( 1 − P u ) − ( m − 1 ) ∏ χ ≠ 1 L χ ( u ) K χ , deg L χ = δ χ . (19) R_{P}(u)=(1-Pu)^{-(m-1)}\prod_{\chi\ne1}L_{\chi}(u)^{K_{\chi}},\qquad\deg L_{\chi}=\delta_{\chi}. \tag*{(19)} R P ( u ) = ( 1 − P u ) − ( m − 1 ) χ = 1 ∏ L χ ( u ) K χ , deg L χ = δ χ . (19)
The identity follows because every nonidentity element of T 0 T_{0} T 0 is regular and irreducible, with centralizer T 0 T_{0} T 0 and Weyl-group orbit of size four. Indeed, if a lift t ∈ T ∖ { ± I } t\in T\setminus\{\pm I\} t ∈ T ∖ { ± I } were conjugate to − t -t − t , then t P j = − t t^{P^{j}}=-t t P j = − t for some j j j by the description of its eigenvalues in the norm-one torus. Squaring would give ( t 2 ) P j = t 2 (t^{2})^{P^{j}}=t^{2} ( t 2 ) P j = t 2 . The order of t 2 t^{2} t 2 divides m m m , and gcd ( m , P 2 − 1 ) = 1 \gcd(m,P^{2}-1)=1 g cd( m , P 2 − 1 ) = 1 , so for t 2 ≠ 1 t^{2}\ne1 t 2 = 1 its P P P -Frobenius orbit has length four. This forces 4 ∣ j 4\mid j 4 ∣ j , which would give t = − t t=-t t = − t . Thus the preimage of C S ( t 0 ) C_{S}(t_{0}) C S ( t 0 ) is C G ( t ) = T C_{G}(t)=T C G ( t ) = T . The same orbit argument gives the four conjugates of t 0 t_{0} t 0 in T 0 T_{0} T 0 . At the identity the two permutation terms cancel. Moreover each factor is determined by a finite, explicitly specified calculation: if
S χ , h = ∑ a ∈ F P h χ ( F a , h ) S_{\chi,h}=\sum_{a\in\mathbb F_{P^{h}}}\chi(F_{a,h}) S χ , h = a ∈ F P h ∑ χ ( F a , h )
then
L χ ( u ) = [ exp ( ∑ h = 1 δ χ S χ , h u h h ) ] ≤ δ χ . L_{\chi}(u)=\left[\exp\left(\sum_{h=1}^{\delta_{\chi}}S_{\chi,h}\frac{u^h}{h}\right)\right]_{\leq\delta_{\chi}}. L χ ( u ) = exp h = 1 ∑ δ χ S χ , h h u h ≤ δ χ .
Thus there are no unknown extension-dependent degrees. We now evaluate the unipotent contributions and make the remaining family explicit. The monodromy theorem used here is invoked only under p ∤ f p\nmid f p ∤ f ; no assertion based on that theorem is made for p ∣ f p\mid f p ∣ f .
The irreducibility indicator and its character constituents
The Coxeter-torus identity (4) , already proved in Section 2, isolates the unipotent characters η ± \eta_{\pm} η ± , the Steinberg character, and the regular cuspidal ρ θ \rho_{\theta} ρ θ . We now evaluate their local conductor contributions. The two additional unipotent characters γ \gamma γ and ξ \xi ξ will be used for permutation representations of lines and flags.
Here are all unipotent values needed to evaluate the conductors. The two rank-one classes have the same value in each row.
Table 8.1.
The five unipotent rows agree with Srinivasan’s original character table for Sp 4 ( F P ) \operatorname{Sp}_{4}(\mathbb{F}_{P}) Sp 4 ( F P ) [16 §8] , with the class labels matched by rank and split type. For a complementary classification in the conformal symplectic group, see Shinoda’s original character table for CSp ( 4 , P ) \operatorname{CSp}(4,P) CSp ( 4 , P ) [14 ] ; compare also [2 Table 20] and the generic unipotent table [7 ] . The latter is based on Shinoda’s work, but we retain Srinivasan’s labels for characters of Sp 4 ( F P ) \operatorname{Sp}_{4}(\mathbb{F}_{P}) Sp 4 ( F P ) . For the last row, restrictions of R T G ( θ ) R_{T}^{G}(\theta) R T G ( θ ) to unipotent elements are independent of θ \theta θ . The Coxeter Green function, equivalently obtained from [6 Examples 5.2(a), 5.8] , has the values shown. These also follow from R T G ( 1 ) = 1 − η + + η − + S t R_{T}^{G}(1)=1-\eta_{+}+\eta_{-}+\mathrm{St} R T G ( 1 ) = 1 − η + + η − + St . Substitution in (18) gives exact, uniform identities:
Table 8.2.
In particular L η − = L ξ = 1 L_{\eta_{-}}=L_{\xi}=1 L η − = L ξ = 1 .
Lemma 8.1. For every H ≥ 1 H\geq1 H ≥ 1 one has
∑ a ∈ F P H η + ( F a , H ) = − 1 , L η + ( u ) = 1 − u . \sum_{a\in\mathbb{F}_{P^H}}\eta_{+}(F_{a,H})=-1,\qquad L_{\eta_{+}}(u)=1-u. a ∈ F P H ∑ η + ( F a , H ) = − 1 , L η + ( u ) = 1 − u .
Proof. The permutation character on F P \mathbb{F}_{P} F P -lines in the natural space is 1 + η + + ξ 1+\eta_{+}+\xi 1 + η + + ξ ; this is the Klingen permutation character, see [2 Table 15] . A line spanned by x ≠ 0 x\neq0 x = 0 is parametrized by z = x P − 1 z=x^{P-1} z = x P − 1 . Dividing L a ( x ) = 0 L_a(x)=0 L a ( x ) = 0 by x x x gives
a = − z n 4 + z n 3 + z + 1 z P + 1 , n j = P j − 1 P − 1 . a=-\frac{z^{n_4}+z^{n_3}+z+1}{z^{P+1}},\qquad n_j=\frac{Pj-1}{P-1}. a = − z P + 1 z n 4 + z n 3 + z + 1 , n j = P − 1 P j − 1 .
The formula describes the root-line cover by the parameter z ∈ G m z\in\mathbb{G}_{m} z ∈ G m . Indeed, two nonzero roots have the same z z z precisely when they differ by an element of F P × \mathbb{F}_{P}^{\times} F P × . Conversely, if z ∈ F P H z\in\mathbb{F}_{P^H} z ∈ F P H and x P − 1 = z x^{P-1}=z x P − 1 = z , then
x P H x = z 1 + P + ⋯ + P H − 1 = N F P H / F P ( z ) ∈ F P × , \frac{x^{P^H}}{x}=z^{1+P+\cdots+P^{H-1}}=N_{\mathbb{F}_{P^H}/\mathbb{F}_P}(z)\in\mathbb{F}_P^\times, x x P H = z 1 + P + ⋯ + P H − 1 = N F P H / F P ( z ) ∈ F P × ,
so its root line is Frobenius-stable. Moreover direct differentiation in characteristic p p p gives
d a d z = z − ( P + 2 ) , \frac{da}{dz}=z^{-(P+2)}, d z d a = z − ( P + 2 ) ,
which is nonzero on G m \mathbb{G}_m G m , confirming that this description is an étale root-line cover. Consequently its F P H \mathbb{F}_{P^H} F P H -points count all Frobenius-stable root lines over parameters a ∈ F P H a\in\mathbb{F}_{P^H} a ∈ F P H , and the total is P H − 1 P^H-1 P H − 1 . Since L ξ = 1 L_\xi=1 L ξ = 1 , its character sum is zero for every H H H . The stated identity follows, and the rank-one cohomological factor is 1 − u 1-u 1 − u . □
The line-cover computation determines one of the degree-one factors. The remaining geometric input is the compactification of the covers parametrizing Lagrangian planes and isotropic flags, treated in the next section; the associated boundary points are counted by inertia orbits rather than by the character table.
The flag curve and the Steinberg factor
Why full flags appear
A full isotropic flag in a four-dimensional symplectic F P \mathbb{F}_P F P -space V V V is a pair ℓ ⊂ M \ell\subset M ℓ ⊂ M where ℓ \ell ℓ is a one-dimensional subspace and M M M is a Lagrangian two-plane. The finite G G G -set of such pairs will be denoted by Flags, and we likewise write Lines and Lag for the sets of lines and Lagrangian planes. The latter are finite sets of F P \mathbb{F}_P F P -rational subspaces, not varieties whose geometric points are being counted here.
The Steinberg factor in the preceding character identity has a natural geometric origin. The spherical Tits building of G = Sp 4 ( F P ) G=\operatorname{Sp}_4(\mathbb{F}_P) G = Sp 4 ( F P ) is the simplicial complex whose simplices encode proper nonzero isotropic flags. In rank two it is a bipartite graph: its vertices are the isotropic lines and the Lagrangian planes in a four-dimensional symplectic space, and its edges are the incident pairs ℓ ⊂ M \ell\subset M ℓ ⊂ M . Its first reduced homology is the Steinberg representation. In particular, the augmented chain complex gives an exact sequence of G G G -modules
0 ⟶ St ⟶ Q ‾ ℓ [ F l a g s ] ⟶ Q ‾ ℓ [ L i n e s ] ⊕ Q ‾ ℓ [ L a g ] ⟶ Q ‾ ℓ ⟶ 0. (20) 0\longrightarrow\operatorname{St}\longrightarrow\overline{\mathbb{Q}}_\ell[\mathrm{Flags}]\longrightarrow\overline{\mathbb{Q}}_\ell[\mathrm{Lines}]\oplus\overline{\mathbb{Q}}_\ell[\mathrm{Lag}]\longrightarrow\overline{\mathbb{Q}}_\ell\longrightarrow0. \tag*{(20)} 0 ⟶ St ⟶ Q ℓ [ Flags ] ⟶ Q ℓ [ Lines ] ⊕ Q ℓ [ Lag ] ⟶ Q ℓ ⟶ 0. (20)
Here the last map takes the difference of the two augmentation maps. Consequently the corresponding permutation characters satisfy
1 F l a g s = 1 L i n e s + 1 L a g − 1 + St . (21) \mathbf{1}_{\mathrm{Flags}}=\mathbf{1}_{\mathrm{Lines}}+\mathbf{1}_{\mathrm{Lag}}-1+\operatorname{St}. \tag*{(21)} 1 Flags = 1 Lines + 1 Lag − 1 + St . (21)
For the root spaces V a V_a V a of Section 2, this suggests studying the finite cover of the parameter line that classifies isotropic full flags ℓ ⊂ M ⊂ V a \ell\subset M\subset V_a ℓ ⊂ M ⊂ V a . Although the flags form part of a positive-dimensional flag variety before taking F P \mathbb{F}_P F P -points, their set in each V a V_a V a is finite; therefore the resulting cover of the one-dimensional parameter line is itself a curve.
We shall write down this flag cover explicitly and show that its smooth projective model realizes the Steinberg L L L -polynomial. The lower-level line and Lagrangian-plane covers account for the other constituents of (21) . The same building-theoretic construction in higher symplectic rank is explained in Appendix § 3.8 ; the special feature here is that a single smooth projective curve suffices.
Inertia orbits and the boundary
The inertia subgroup computed in § 2.2 controls how these finite covers compactify. Indeed, geometric points over infinity in the normalization of a finite cover of the affine line correspond to inertia orbits in its geometric generic fibre. The following calculation therefore determines the number of boundary points; the assertion about rationality uses the arithmetic decomposition group, not just geometric inertia.
Lemma 9.1 (Inertia orbits on isotropic flags ). Let V = W ⊕ W ∗ V=W\oplus W^* V = W ⊕ W ∗ be symplectic of dimension four, and let I = U ⋊ Σ I=U\rtimes\Sigma I = U ⋊ Σ , where U U U is the Siegel radical of the stabilizer of W W W and Σ ≃ F P 2 × \Sigma\simeq\mathbb{F}_{P^2}^{\times} Σ ≃ F P 2 × acts on W W W as a Singer group. Then I I I has three orbits on Lagrangian planes M M M , indexed by
dim ( M ∩ W ) = 2 , 1 , 0 \dim(M \cap W) = 2, 1, 0 dim ( M ∩ W ) = 2 , 1 , 0 , and four orbits on flags ℓ ⊂ M \ell\subset M ℓ ⊂ M with dim ℓ = 1 \dim\ell= 1 dim ℓ = 1 and M M M Lagrangian. The plane orbit sizes are 1 1 1 , P ( P + 1 ) P(P+1) P ( P + 1 ) , and P 3 P^3 P 3 , respectively. The flag orbits are
M = W M = W M = W ; dim ( M ∩ W ) = 1 \dim(M \cap W) = 1 dim ( M ∩ W ) = 1 , ℓ = M ∩ W \ell= M \cap W ℓ = M ∩ W ; dim ( M ∩ W ) = 1 \dim(M \cap W) = 1 dim ( M ∩ W ) = 1 , ℓ ≠ M ∩ W \ell\ne M \cap W ℓ = M ∩ W ; M ∩ W = 0 M \cap W = 0 M ∩ W = 0 .
Their sizes in this order are P + 1 P+1 P + 1 , P ( P + 1 ) P(P+1) P ( P + 1 ) , P 2 ( P + 1 ) P^2(P+1) P 2 ( P + 1 ) , and P 3 ( P + 1 ) P^3(P+1) P 3 ( P + 1 ) . For the root-space family, the positive-valuation plane W W W is preserved by the arithmetic decomposition group at infinity. Consequently these orbits are Frobenius-stable and correspond to rational boundary points in the smooth projective compactifications of the respective connected covers.
Proof. The Singer subgroup is transitive on the lines in W W W and W ∗ W^* W ∗ . The Lagrangian complements to W W W are graphs of symmetric maps W ∗ → W W^* \to W W ∗ → W , on which U U U acts simply transitively; the stabilizer of one complement contains Σ \Sigma Σ , transitive on its lines. This proves the orbit assertion when M ∩ W = 0 M \cap W = 0 M ∩ W = 0 , and M = W M = W M = W is immediate. For dim ( M ∩ W ) = 1 \dim(M \cap W) = 1 dim ( M ∩ W ) = 1 , fix L = M ∩ W L = M \cap W L = M ∩ W . The planes M M M correspond to lines in the symplectic plane L ⊥ / L L^\perp/L L ⊥ / L other than W / L W/L W / L . The unipotent radical acts transitively on these P P P lines. Within such an M M M , the stabilizer of M M M in U U U is transitive on the lines ℓ ≠ L \ell\ne L ℓ = L . To see this, write M = L + F P ( w + λ ) M = L + \mathbb{F}_P(w + \lambda) M = L + F P ( w + λ ) , where λ ∈ W ∗ \lambda\in W^* λ ∈ W ∗ is nonzero and annihilates L L L . For any h ∈ L h \in L h ∈ L , a symmetric map W ∗ → W W^* \to W W ∗ → W can send λ \lambda λ to h h h . Indeed, in a basis of W ∗ W^* W ∗ starting with λ \lambda λ , the first column of its symmetric matrix can be prescribed arbitrarily. The corresponding element of U U U sends w + λ w + \lambda w + λ to w + λ + h w + \lambda+ h w + λ + h , giving all translations on the P P P lines of M M M other than L L L . Since Σ \Sigma Σ is transitive on possible L L L , there are precisely the two stated flag orbits in this case. Finally the boundary geometric points of a finite étale cover of A 1 \mathbb{A}^1 A 1 are the inertia orbits on its generic fiber. The intersection conditions with W W W are preserved by the whole decomposition group, so each boundary orbit is individually stable under Frobenius and hence rational over F P \mathbb{F}_P F P .
An explicit equation and its cohomology
We now express the full-flag moduli problem by two coordinates and compare the cohomology of its compactification with the Steinberg factor.
Theorem 9.2. Assume P = p f P = p^f P = p f is odd and p ∤ f p \nmid f p ∤ f . Let B P B_P B P be the smooth projective model of the affine curve
z P 2 + 1 s P + 1 + ( 1 + z + z P 2 + 1 ) s + 1 = 0 , z , s ≠ 0. (22) z^{P^2+1}s^{P+1} + (1+z+z^{P^2+1})s + 1 = 0,\qquad z,s \ne0. \tag*{(22)} z P 2 + 1 s P + 1 + ( 1 + z + z P 2 + 1 ) s + 1 = 0 , z , s = 0. (22) Then B P B_P B P is geometrically connected, has genus P ( P + 1 ) / 2 P(P+1)/2 P ( P + 1 ) /2 , and the complement in B P B_P B P of the displayed affine curve consists of four F P \mathbb{F}_P F P -rational points. Its zeta numerator is exactly L S t L_{\mathrm{St}} L St :
Z ( B P , u ) = L S t ( u ) ( 1 − u ) ( 1 − P u ) . Z(B_P,u) = \frac{L_{\mathrm{St}}(u)}{(1-u)(1-Pu)}. Z ( B P , u ) = ( 1 − u ) ( 1 − P u ) L St ( u ) . The compactification of the Lagrangian-plane cover is a genus-zero curve whose affine part has three rational points removed. In particular,
∑ a ∈ F P H # { Frobenius-stable Lagrangian planes in V a } = P H − 2 ( H ≥ 1 ) . \sum_{a \in\mathbb{F}_{P^H}} \#\{\text{Frobenius-stable Lagrangian planes in } V_a\} = P^H - 2 \qquad(H \ge1). a ∈ F P H ∑ # { Frobenius-stable Lagrangian planes in V a } = P H − 2 ( H ≥ 1 ) .
Proof. The flag coordinates. The coordinates below have an intrinsic meaning: z z z records the root line ℓ \ell ℓ , while s s s records the additional choice of Lagrangian plane M M M containing it. More explicitly, for a nonzero root x x x and z = x P − 1 z = x^{P-1} z = x P − 1 , the equation L a ( x ) = 0 L_a(x) = 0 L a ( x ) = 0 implies
a = − z 1 + P + P 2 + P 3 + z 1 + P + P 2 + z + 1 z 1 + P . (23) a = -\frac{z^{1+P+P^2+P^3} + z^{1+P+P^2} + z + 1}{z^{1+P}}. \tag*{(23)} a = − z 1 + P z 1 + P + P 2 + P 3 + z 1 + P + P 2 + z + 1 . (23)
Thus the z z z -coordinate determines both the line and its parameter, and the two-coordinate equation below is to be viewed as a finite cover of the a a a -line through (23) . Both are invariant under changing the chosen generators of the flag. The resulting plane is required to lie in the root space V a V_a V a and to be isotropic for the pairing B B B . Using the form B B B of Section 2, fix a root line, choose x ≠ 0 x \ne0 x = 0 on it, and put
y = x v , w = v P − v , z = x P − 1 . y = xv,\qquad w = v^P - v,\qquad z = x^{P-1}. y = xv , w = v P − v , z = x P − 1 .
Dividing the expression for B ( x , x v ) B(x,xv) B ( x , xv ) by − x P 2 + 1 -x^{P^{2}+1} − x P 2 + 1 gives
B ( x , x v ) x P 2 + 1 = ( v P 2 − v ) + z ( v P 2 − v P ) + z P 2 + 1 ( v P 3 − v P ) = w + w P + z w P + z P 2 + 1 ( w P 2 + w P ) . \begin{aligned}
\frac{B(x,xv)}{x^{P^{2}+1}}=(v^{P^{2}}-v)+z(v^{P^{2}}-v^{P})+z^{P^{2}+1}(v^{P^{3}}-v^{P}) \\
&= w+w^{P}+zw^{P}+z^{P^{2}+1}(w^{P^{2}}+w^{P}).
\end{aligned} x P 2 + 1 B ( x , xv ) = ( v P 2 − v ) + z ( v P 2 − v P ) + z P 2 + 1 ( v P 3 − v P ) = w + w P + z w P + z P 2 + 1 ( w P 2 + w P ) .
Thus B ( x , y ) = 0 B(x,y)=0 B ( x , y ) = 0 is equivalent to
z P 2 + 1 w P 2 + ( 1 + z + z P 2 + 1 ) w P + w = 0. z^{P^{2}+1}w^{P^{2}}+(1+z+z^{P^{2}+1})w^{P}+w=0. z P 2 + 1 w P 2 + ( 1 + z + z P 2 + 1 ) w P + w = 0.
The nonvanishing of w w w expresses the linear independence of x , y x,y x , y . Putting s = w P − 1 s=w^{P-1} s = w P − 1 and dividing the equation by w w w gives (22) , since w P 2 − 1 = s P + 1 w^{P^{2}-1}=s^{P+1} w P 2 − 1 = s P + 1 . Changing the basis of the flag preserves z , s z,s z , s : replacing x x x by λ x \lambda x λ x with λ ∈ F P × \lambda\in\mathbb{F}_{P}^{\times} λ ∈ F P × fixes z z z , and replacing y y y by μ y + ν x \mu y+\nu x μ y + ν x with μ ∈ F P × \mu\in\mathbb{F}_{P}^{\times} μ ∈ F P × sends w = ( y / x ) P − y / x w=(y/x)^{P}-y/x w = ( y / x ) P − y / x to ( μ / λ ) w (\mu/\lambda)w ( μ / λ ) w , leaving w P − 1 w^{P-1} w P − 1 fixed. Conversely, z z z specifies the root line, while s s s specifies the F P \mathbb{F}_{P} F P -line spanned by w w w in the Artin–Schreier quotient; choosing a lift v v v with v P − v = w v^{P}-v=w v P − v = w reconstructs the two-plane ⟨ x , x v ⟩ F P \langle x,xv\rangle_{\mathbb{F}_{P}} ⟨ x , xv ⟩ F P . Two lifts v v v of the same w w w differ by an element of F P \mathbb{F}_{P} F P , so the corresponding roots x v xv xv differ by an element of F P x \mathbb{F}_{P}x F P x and give the same plane. The coordinate map and its inverse on geometric points commute with F Q F_{Q} F Q : applying Q Q Q -power Frobenius to x , y x,y x , y sends the coordinates to ( z Q , s Q ) (z^{Q},s^{Q}) ( z Q , s Q ) . By the bijection, a flag is F Q F_{Q} F Q -stable precisely when both z z z and s s s belong to F Q \mathbb{F}_{Q} F Q . The alternating-form equation is exactly the displayed equation in w w w ; the identity
B ( x , y ) P − B ( x , y ) = L a ( x ) y P 2 − x P 2 L a ( y ) B(x,y)^{P}-B(x,y)=L_{a}(x)y^{P^{2}}-x^{P^{2}}L_{a}(y) B ( x , y ) P − B ( x , y ) = L a ( x ) y P 2 − x P 2 L a ( y )
shows that B ( x , y ) = 0 B(x,y)=0 B ( x , y ) = 0 and L a ( x ) = 0 L_{a}(x)=0 L a ( x ) = 0 force L a ( y ) = 0 L_{a}(y)=0 L a ( y ) = 0 . Thus (22) describes the full-flag cover.
Smoothness and connectedness. If Φ ( z , s ) \Phi(z,s) Φ ( z , s ) denotes the left-hand side of (22) , then on the curve
∂ Φ ∂ s = z P 2 + 1 s P + 1 + z + z P 2 + 1 = − s − 1 . \frac{\partial\Phi}{\partial s}=z^{P^{2}+1}s^{P}+1+z+z^{P^{2}+1}=-s^{-1}. ∂ s ∂ Φ = z P 2 + 1 s P + 1 + z + z P 2 + 1 = − s − 1 .
Thus the projection to the root-line coordinate z z z is étale on the displayed open set. Together with the root-line cover this gives the requisite finite étale full-flag cover over the parameter line. Full geometric monodromy, which acts transitively on full flags, proves geometric connectedness.
The Lagrangian-plane cover. By Lemma 9.1, the plane cover has three F P \mathbb{F}_{P} F P -rational boundary points. Its permutation character is 1 + η + + γ 1+\eta_{+}+\gamma 1 + η + + γ , and the dimensions of the corresponding affine cohomology groups in (8.2) give dim H c 1 = 2 \dim H^{1}_{c}=2 dim H c 1 = 2 . For a connected smooth affine curve obtained by removing r r r geometric points from a projective curve of genus g g g , one has
dim H c 1 = 2 g + r − 1. \dim H^{1}_{c}=2g+r-1. dim H c 1 = 2 g + r − 1.
Hence 2 = 2 g + 3 − 1 2=2g+3-1 2 = 2 g + 3 − 1 , and g = 0 g=0 g = 0 . Since all three boundary points are rational, the affine plane cover has P H − 2 P^{H}-2 P H − 2 rational points over F P H \mathbb{F}_{P^{H}} F P H . The resulting character sum gives L γ ( u ) = 1 − u L_{\gamma}(u)=1-u L γ ( u ) = 1 − u , alongside the earlier L η + ( u ) = 1 − u L_{\eta_{+}}(u)=1-u L η + ( u ) = 1 − u and L ξ ( u ) = 1 L_{\xi}(u)=1 L ξ ( u ) = 1 .
The full-flag cover. The flag permutation character is
1 + 2 η + + γ + ξ + S t , 1+2\eta_{+}+\gamma+\xi+\mathrm{St}, 1 + 2 η + + γ + ξ + St ,
in agreement with (21) and the line and plane decompositions. Let Y Y Y be its affine cover. The preceding linear factors and the permutation-sheaf decomposition yield the Frobenius characteristic polynomial
det ( 1 − u F ∣ H c 1 ( Y F ‾ P ) ) = ( 1 − u ) 3 L S t ( u ) . (24) \det(1-uF\mid H^{1}_{c}(Y_{\overline{\mathbb{F}}_{P}}))=(1-u)^{3}L_{\mathrm{St}}(u). \tag*{(24)} det ( 1 − u F ∣ H c 1 ( Y F P )) = ( 1 − u ) 3 L St ( u ) . (24)
In particular its first compactly supported cohomology has dimension P 2 + P + 3 P^{2}+P+3 P 2 + P + 3 . The Steinberg factor has degree P 2 + P P^{2}+P P 2 + P by (8.2).
Removing the boundary contribution. Let D = B P ∖ Y D=\mathcal{B}_{P}\setminus Y D = B P ∖ Y . By the inertia-orbit lemma, D D D consists of four F P \mathbb{F}_{P} F P -rational points. The localization sequence gives a Frobenius-equivariant exact sequence
0 ⟶ H 0 ( B P ) ⟶ H 0 ( D ) ⟶ H c 1 ( Y ) ⟶ H 1 ( B P ) ⟶ 0 , 0\longrightarrow H^{0}(\mathcal{B}_{P})\longrightarrow H^{0}(D)\longrightarrow H^{1}_{c}(Y)\longrightarrow H^{1}(\mathcal{B}_{P})\longrightarrow0, 0 ⟶ H 0 ( B P ) ⟶ H 0 ( D ) ⟶ H c 1 ( Y ) ⟶ H 1 ( B P ) ⟶ 0 ,
where all cohomology is taken after base change to F ‾ P \overline{\mathbb{F}}_P F P with coefficients in Q ‾ ℓ \overline{\mathbb{Q}}_\ell Q ℓ . The cokernel of H 0 ( B P ) → H 0 ( D ) H^0(B_P) \to H^0(D) H 0 ( B P ) → H 0 ( D ) is three-dimensional with trivial Frobenius action. Therefore
det ( 1 − u F ∣ H c 1 ( Y ) ) = ( 1 − u ) 3 det ( 1 − u F ∣ H 1 ( B P ) ) . \det(1-uF \mid H^1_c(Y))=(1-u)^3\det(1-uF \mid H^1(B_P)). det ( 1 − u F ∣ H c 1 ( Y )) = ( 1 − u ) 3 det ( 1 − u F ∣ H 1 ( B P )) .
Comparing with (24) identifies the latter determinant with L S t ( u ) L_{\mathrm{St}}(u) L St ( u ) . Its degree is P 2 + P = 2 g ( B P ) P^2+P=2g(B_P) P 2 + P = 2 g ( B P ) , proving the genus formula and the asserted zeta function. □
Remark 9.3 (A direct low-degree check at P = 3 P=3 P = 3 ). The affine equation (22) , with z , s ≠ 0 z,s\ne0 z , s = 0 , can be counted independently over small fields. For H = 1 , 2 , 3 , 4 , 5 H=1,2,3,4,5 H = 1 , 2 , 3 , 4 , 5 , its numbers of points over F 3 H \mathbb{F}_{3^H} F 3 H are
1 , 9 , 22 , 81 , 321. 1,\qquad9,\qquad22,\qquad81,\qquad321. 1 , 9 , 22 , 81 , 321.
Adding the four rational boundary points gives 5 , 13 , 26 , 85 , 325 5,13,26,85,325 5 , 13 , 26 , 85 , 325 , respectively. These agree with 3 H + 1 − ∑ i α i H 3^H+1-\sum_i\alpha_i^H 3 H + 1 − ∑ i α i H calculated from the explicit polynomial N ( u ) N(u) N ( u ) of Theorem 2.4. This is a consistency check, not a substitute for the geometric proof of the flag-curve identification.
The counting series for general P P P
The character identity of Proposition 2.1 reduces the atom-counting problem to the Steinberg and regular cuspidal constituents. The point of this section is to assemble these contributions into one formula, while keeping distinct the geometrically realized factor and the factors not yet explicitly evaluated for arbitrary P P P .
The two cohomological factors
Throughout this section let P = p f P=p^f P = p f be odd, with p ∤ f p\nmid f p ∤ f , and put
m = P 2 + 1 2 , κ = P 2 − 1 8 . m=\frac{P^2+1}{2},\qquad\kappa=\frac{P^2-1}{8}. m = 2 P 2 + 1 , κ = 8 P 2 − 1 .
Recall the set O \mathcal{O} O of Frobenius orbits of nontrivial characters of the Coxeter torus T 0 T_0 T 0 from Section 2. Set
S P ( u ) = L S t ( u ) , C P ( u ) = ∏ θ ∈ O L ρ θ ( u ) . (25) S_P(u)=L_{\mathrm{St}}(u),\qquad C_P(u)=\prod_{\theta\in\mathcal{O}}L_{\rho_\theta}(u). \tag*{(25)} S P ( u ) = L St ( u ) , C P ( u ) = θ ∈ O ∏ L ρ θ ( u ) . (25)
Here L χ ( u ) L_\chi(u) L χ ( u ) denotes the Frobenius characteristic polynomial on H c 1 ( A F ‾ P 1 , E χ ) H^1_c(\mathbb{A}^1_{\overline{\mathbb{F}}_P},\mathcal{E}_\chi) H c 1 ( A F P 1 , E χ ) , with constant term one. The conductor computation in § 3.1 –§ 3.2 gives
deg S P = P 2 + P , deg C P = κ ( P 2 + P − 2 ) . (26) \deg S_P=P^2+P,\qquad\deg C_P=\kappa(P^2+P-2). \tag*{(26)} deg S P = P 2 + P , deg C P = κ ( P 2 + P − 2 ) . (26)
Both polynomials have integral coefficients. For S P S_P S P this follows also from Theorem 9.2 , since it is the zeta numerator of the curve B P B_P B P . For C P C_P C P , the roots are algebraic integers and the full collection of regular cuspidal constituents is stable under Galois conjugation; hence the product has rational algebraic-integer coefficients. In the present normalization their inverse roots have complex absolute value P \sqrt{P} P . In particular, none of these inverse roots equals 1 1 1 or P P P .
Remark 10.1 (Purity of the two factors). Here purity means that every inverse root α \alpha α of the relevant Frobenius polynomial satisfies ∣ ι ( α ) ∣ = P 1 / 2 |\iota(\alpha)|=P^{1/2} ∣ ι ( α ) ∣ = P 1/2 for every embedding ι : Q ‾ ↪ C \iota:\overline{\mathbb{Q}}\hookrightarrow\mathbb{C} ι : Q ↪ C . For S P S_P S P , this follows from Theorem 9.2 : it is the characteristic polynomial of Frobenius on H 1 H^1 H 1 of the smooth projective curve B P B_P B P , which is pure of weight one by the Weil conjectures. For each regular cuspidal factor L ρ θ L_{\rho_\theta} L ρ θ , the corresponding local system has finite monodromy and is therefore pointwise pure of weight zero. Moreover, E ρ θ I ∞ = 0 E_{\rho_\theta}^{I_\infty}=0 E ρ θ I ∞ = 0 , since E ρ θ U = 0 E_{\rho_\theta}^{U}=0 E ρ θ U = 0 by (8.2) and U ⊂ I ∞ U\subset I_\infty U ⊂ I ∞ . To see why this yields purity of its compactly supported cohomology, trivialize the local system on a finite Galois cover of A 1 \mathbb{A}^1 A 1 and take a smooth projective compactification X X X of that cover. The boundary contribution to H c 1 H^1_c H c 1 is built from inertia-invariant stalks; it vanishes in the ρ θ \rho_\theta ρ θ -isotypic component by the preceding equality. Consequently, H c 1 ( A F ‾ P 1 , E ρ θ ) H^1_c(\mathbb{A}^1_{\overline{\mathbb{F}}_P},\mathcal{E}_{\rho_\theta}) H c 1 ( A F P 1 , E ρ θ ) is a Frobenius-stable direct summand of H 1 ( X F ‾ P , Q ‾ ℓ ) H^1(X_{\overline{\mathbb{F}}_P},\overline{\mathbb{Q}}_\ell) H 1 ( X F P , Q ℓ ) . This proves purity of weight one for the regular cuspidal factors as well. The vanishing of inertia invariants is essential here: finite monodromy alone would in general give only mixed weights for cohomology of an open curve.
For a polynomial E ( u ) = ∏ i ( 1 − λ i u ) E(u)=\prod_i(1-\lambda_i u) E ( u ) = ∏ i ( 1 − λ i u ) , write p H ( E ) = ∑ i λ i H p_H(E)=\sum_i\lambda_i^H p H ( E ) = ∑ i λ i H . We use the notation
s H = p H ( S P ) , c H = p H ( C P ) . (27) s_H=p_H(S_P),\qquad c_H=p_H(C_P). \tag*{(27)} s H = p H ( S P ) , c H = p H ( C P ) . (27)
These are algebraic integers and, as the coefficients of the two polynomials are integers, in fact ordinary integers.
The general counting formula
Theorem 10.2. For every H ≥ 1 H\geq1 H ≥ 1 , the number A H ( P ) A_H(P) A H ( P ) of atoms in the one-parameter family satisfies
m A H ( P ) = κ ( P H + 1 − s H ) + c H . (28) mA_H(P)=\kappa(P^H+1-s_H)+c_H. \tag*{(28)} m A H ( P ) = κ ( P H + 1 − s H ) + c H . (28) Equivalently, in terms of the flag curve of Theorem 9.2 ,
m A H ( P ) = κ # B P ( F P H ) + c H . (29) mA_H(P)=\kappa\#\mathcal{B}_P(\mathbb{F}_{P^H})+c_H. \tag*{(29)} m A H ( P ) = κ # B P ( F P H ) + c H . (29) The exponential counting series defined in Section 2 therefore has the rational expression
R P ( u ) = [ S P ( u ) κ ( 1 − u ) κ ( 1 − P u ) κ C P ( u ) ] 4 . (30) \mathcal{R}_P(u)=\left[\frac{S_P(u)^\kappa}{(1-u)^\kappa(1-Pu)^\kappa C_P(u)}\right]^4. \tag*{(30)} R P ( u ) = [ ( 1 − u ) κ ( 1 − P u ) κ C P ( u ) S P ( u ) κ ] 4 . (30) These identities are understood as identities of rational functions, or equivalently of formal series at u = 0 u=0 u = 0 .
Proof. Evaluate the character identity (4) on the Frobenius elements attached to parameters a ∈ F P H a\in\mathbb{F}_{P^H} a ∈ F P H and sum over a a a . The trivial character contributes P H P^H P H . The trace formulas established earlier give, in the same normalization,
∑ a η + ( F a , H ) = − 1 , ∑ a η − ( F a , H ) = 0 , ∑ a St ( F a , H ) = − s H , \sum_a\eta_+(F_{a,H})=-1,\qquad\sum_a\eta_-(F_{a,H})=0,\qquad\sum_a\operatorname{St}(F_{a,H})=-s_H, a ∑ η + ( F a , H ) = − 1 , a ∑ η − ( F a , H ) = 0 , a ∑ St ( F a , H ) = − s H , while the sum of the ρ θ \rho_\theta ρ θ -contributions is − c H -c_H − c H . The signs in the indicator identity thus yield
m A H ( P ) = κ ( P H − ( − 1 ) + 0 − s H ) − ( − c H ) , mA_H(P)=\kappa(P^H-(-1)+0-s_H)-(-c_H), m A H ( P ) = κ ( P H − ( − 1 ) + 0 − s H ) − ( − c H ) , which is (28) . By Theorem 9.2 , # B P ( F P H ) = P H + 1 − s H \#\mathcal{B}_P(\mathbb{F}_{P^H})=P^H+1-s_H # B P ( F P H ) = P H + 1 − s H , giving (29) . Finally, for any E ( u ) = ∏ i ( 1 − λ i u ) E(u)=\prod_i(1-\lambda_i u) E ( u ) = ∏ i ( 1 − λ i u ) ,
− log E ( u ) = ∑ H ≥ 1 p H ( E ) H u H . -\log E(u)=\sum_{H\geq1}\frac{p_H(E)}{H}u^H. − log E ( u ) = H ≥ 1 ∑ H p H ( E ) u H . Multiply (28) by 4 u H / H 4u^H/H 4 u H / H , sum over H ≥ 1 H\geq1 H ≥ 1 , and exponentiate, using the definition of R P \mathcal{R}_P R P . This proves (30) . □
Remark 10.3 (What is and is not explicit). Equation (29) splits the atom count into a term given by rational points on an explicit curve and a regular cuspidal trace term. It is not, by itself, a closed numerical formula for all P P P , since the coefficients of C P C_P C P have not been fully evaluated. For P = 3 P=3 P = 3 , κ = 1 \kappa=1 κ = 1 , S 3 = N S_3=N S 3 = N , and C 3 = D C_3=D C 3 = D ; thus Theorem 10.2 specializes to the completely explicit formula of Theorem 2.4. For general P P P , § 3.6 evaluates the initial coefficients of the two factors. The degrees in (26) are the degrees before possible common factors of S P S_P S P and C P C_P C P are cancelled.
Poles and minimal recurrences
The rational expression in Theorem 10.2 does more than encode the counts: it implies that the sequence A H ( P ) A_H(P) A H ( P ) satisfies a homogeneous linear recurrence with constant coefficients. A precise minimal order, however, depends on which Frobenius inverse roots actually survive in the virtual quotient of the Steinberg and cuspidal factors.
The ordinary generating series and its poles
Set
Q P ( u ) = S P ( u ) κ C P ( u ) . Q_P(u)=\frac{S_P(u)^{\kappa}}{C_P(u)}. Q P ( u ) = C P ( u ) S P ( u ) κ .
Taking the logarithmic derivative of (30) gives the ordinary , rather than exponential, counting series:
m ∑ H ≥ 1 A H ( P ) u H = κ P u 1 − P u + κ u 1 − u + u Q P ′ ( u ) Q P ( u ) . (31) m\sum_{H\ge1}A_H(P)u^H=\frac{\kappa Pu}{1-Pu}+\frac{\kappa u}{1-u}+u\frac{Q_P'(u)}{Q_P(u)}. \tag*{(31)} m H ≥ 1 ∑ A H ( P ) u H = 1 − P u κ P u + 1 − u κ u + u Q P ( u ) Q P ′ ( u ) . (31)
Indeed, u R P ′ ( u ) / R P ( u ) = 4 m ∑ H ≥ 1 A H ( P ) u H uR_P'(u)/R_P(u)=4m\sum_{H\ge1}A_H(P)u^H u R P ′ ( u ) / R P ( u ) = 4 m ∑ H ≥ 1 A H ( P ) u H . After cancellation, write
Q P ( u ) = ∏ α ∈ Σ P ( 1 − α u ) e α , e α ∈ Z ∖ { 0 } , (32) Q_P(u)=\prod_{\alpha\in\Sigma_P}(1-\alpha u)^{e_\alpha},\qquad e_\alpha\in\mathbb{Z}\setminus\{0\}, \tag*{(32)} Q P ( u ) = α ∈ Σ P ∏ ( 1 − α u ) e α , e α ∈ Z ∖ { 0 } , (32)
where the numbers α \alpha α are distinct. Then
m A H ( P ) = κ P H + κ − ∑ α ∈ Σ P e α α H . (33) mA_H(P)=\kappa P^H+\kappa-\sum_{\alpha\in\Sigma_P}e_\alpha\alpha^H. \tag*{(33)} m A H ( P ) = κ P H + κ − α ∈ Σ P ∑ e α α H . (33)
Thus each surviving inverse root gives a simple pole of the ordinary generating series at u = α − 1 u=\alpha^{-1} u = α − 1 , regardless of its multiplicity in Q P Q_P Q P . Multiplicity changes the residue, not the pole order. A common inverse root cancels only when its signed multiplicity in (32) is zero.
Proposition 11.1 (Minimal-order principle). Let r P = ∣ Σ P ∣ r_P=|\Sigma_P| r P = ∣ Σ P ∣ be the number of distinct inverse roots surviving in the reduced quotient Q P Q_P Q P . Then A H ( P ) A_H(P) A H ( P ) satisfies a homogeneous rational constant-coefficient recurrence of minimal order
Its characteristic polynomial can be taken to be
( T − 1 ) ( T − P ) ∏ α ∈ Σ P ( T − α ) . (34) (T-1)(T-P)\prod_{\alpha\in\Sigma_P}(T-\alpha). \tag*{(34)} ( T − 1 ) ( T − P ) α ∈ Σ P ∏ ( T − α ) . (34) This order is minimal even if a recurrence is required only for all sufficiently large H H H .
Proof. By the purity argument in Remark 10.1, all α ∈ Σ P \alpha\in\Sigma_P α ∈ Σ P satisfy ∣ ι ( α ) ∣ = P |\iota(\alpha)|=\sqrt{P} ∣ ι ( α ) ∣ = P for every complex embedding ι \iota ι ; in particular, none equals 1 1 1 or P P P . The coefficients of P H P^H P H , 1 H 1^H 1 H , and each α H \alpha^H α H in (33) are nonzero, and all bases are distinct. Their powers are linearly independent over characteristic zero by the Vandermonde determinant. Hence any eventual constant-coefficient recurrence must have each of these bases as a root of its characteristic polynomial. Conversely, the polynomial (34) annihilates every summand. It has rational coefficients: Q P ∈ Q ( u ) Q_P\in\mathbb{Q}(u) Q P ∈ Q ( u ) , so the set of its poles and zeros, including the nonzero signed multiplicities, is stable under Galois conjugation. □ \square □
The indicator identity has already eliminated the elementary non-Steinberg factors. For general P P P , it does not rule out cancellation between S P S_P S P and C P C_P C P or repeated inverse roots within either factor, and we make no uniform assertion of minimality without checking those possibilities.
The first two coefficients for general P P P
We now turn from the general cohomological description of the counting series to explicit coefficient calculations. There are two rather different phenomena. Over F P \mathbb{F}_P F P , the equation of the flag curve specializes to a quadratic equation whose discriminant defines an auxiliary genus-one curve; this determines the linear coefficients by the trace of a fixed elliptic curve. Over F P 2 \mathbb{F}_{P^2} F P 2 , trace–norm coordinates instead reduce the point counts to explicit two-variable quadratic-character sums. An identity between these sums permits a common formula for both quadratic coefficients.
The cohomological factors S P S_P S P and C P C_P C P are used under the standing hypothesis P = p f P=p^f P = p f with p ∤ f p\nmid f p ∤ f . The direct arithmetic counts A 1 ( P ) , A 2 ( P ) A_1(P),A_2(P) A 1 ( P ) , A 2 ( P ) and the finite-field character-sum identities below make sense for every odd prime power P P P , without that monodromy hypothesis. Throughout, χ \chi χ denotes the quadratic character of F P \mathbb{F}_P F P , extended by χ ( 0 ) = 0 \chi(0)=0 χ ( 0 ) = 0 .
The linear coefficients: an auxiliary elliptic curve
Write
S P ( u ) = 1 + a 1 u + a 2 u 2 + ⋯ , C P ( u ) = 1 + b 1 u + b 2 u 2 + ⋯ . S_P(u) = 1 + a_1u + a_2u^2 + \cdots,\qquad C_P(u) = 1 + b_1u + b_2u^2 + \cdots. S P ( u ) = 1 + a 1 u + a 2 u 2 + ⋯ , C P ( u ) = 1 + b 1 u + b 2 u 2 + ⋯ .
The first coefficient of a Frobenius polynomial is minus its first power trace. The point-count calculation below makes this trace accessible without computing a character table.
Theorem 12.1. For p ≠ 3 , 5 p \ne3,5 p = 3 , 5 , let
τ P = P + 1 − # E ( F P ) , E : y 2 = x ( x − 16 ) ( x + 9 ) . \tau_P = P + 1 - \#E(\mathbb{F}_P),\qquad E:y^2=x(x-16)(x+9). τ P = P + 1 − # E ( F P ) , E : y 2 = x ( x − 16 ) ( x + 9 ) . Then
a 1 = − τ P , b 1 = − P + 1 − τ P 4 . (35) a_1=-\tau_P,\qquad b_1=-\frac{P+1-\tau_P}{4}. \tag*{(35)} a 1 = − τ P , b 1 = − 4 P + 1 − τ P . (35) For P = 3 f P=3^f P = 3 f put τ P = ( − 1 ) f \tau_P=(-1)^f τ P = ( − 1 ) f , and for P = 5 f P=5^f P = 5 f put τ P = 1 \tau_P=1 τ P = 1 . In these two cases,
a 1 = − τ P , b 1 = κ − P − τ P 4 . (36) a_1=-\tau_P,\qquad b_1=\kappa-\frac{P-\tau_P}{4}. \tag*{(36)} a 1 = − τ P , b 1 = κ − 4 P − τ P . (36) For good primes, the sequence τ P \tau_P τ P can also be evaluated from the weight-two newform
η ( z ) η ( 3 z ) η ( 5 z ) η ( 15 z ) = ∑ n ≥ 1 e n q n . \eta(z)\eta(3z)\eta(5z)\eta(15z)=\sum_{n\geq1}e_nq^n. η ( z ) η ( 3 z ) η ( 5 z ) η ( 15 z ) = n ≥ 1 ∑ e n q n . Namely τ p = e p \tau_p=e_p τ p = e p , and for f ≥ 2 f\geq2 f ≥ 2 , τ p f = e p τ p f − 1 − p τ p f − 2 \tau_{p^f}=e_p\tau_{p^{f-1}}-p\tau_{p^{f-2}} τ p f = e p τ p f − 1 − p τ p f − 2 , with τ p 0 = 2 \tau_{p^0}=2 τ p 0 = 2 . These power traces are not, in general, the newform coefficients e p f e_{p^f} e p f .
Proof. At F P \mathbb{F}_P F P -points, (22) reduces to a quadratic in s s s :
z 2 s 2 + ( z 2 + z + 1 ) s + 1 = 0. z^2s^2+(z^2+z+1)s+1=0. z 2 s 2 + ( z 2 + z + 1 ) s + 1 = 0.
Its discriminant is Δ ( z ) = ( z 2 − z + 1 ) ( z 2 + 3 z + 1 ) \Delta(z)=(z^2-z+1)(z^2+3z+1) Δ ( z ) = ( z 2 − z + 1 ) ( z 2 + 3 z + 1 ) . Adding the four boundary points gives
# B P ( F P ) = P + 2 + ∑ z ∈ F P χ ( Δ ( z ) ) , a 1 = 1 + ∑ z χ ( Δ ( z ) ) . (37) \#\mathcal{B}_P(\mathbb{F}_P)=P+2+\sum_{z\in\mathbb{F}_P}\chi(\Delta(z)),\qquad a_1=1+\sum_z\chi(\Delta(z)). \tag*{(37)} # B P ( F P ) = P + 2 + z ∈ F P ∑ χ ( Δ ( z )) , a 1 = 1 + z ∑ χ ( Δ ( z )) . (37)
If p ≠ 3 , 5 p\ne3,5 p = 3 , 5 , the quartic Δ \Delta Δ is squarefree, and the smooth projective model of F : w 2 = Δ ( z ) F:w^2=\Delta(z) F : w 2 = Δ ( z ) has genus one and two rational points at infinity. Consequently its point count equals that of B P \mathcal{B}_P B P over F P \mathbb{F}_P F P ; this is a point-count identity, not a morphism from the full flag curve to F F F . We next show that F F F is isogenous to E E E . The involution ( z , w ) ↦ ( 1 / z , − w / z 2 ) (z,w)\mapsto(1/z,-w/z^2) ( z , w ) ↦ ( 1/ z , − w / z 2 ) is fixed-point-free when p ≠ 3 , 5 p\ne3,5 p = 3 , 5 : a fixed point would require z = ± 1 z=\pm1 z = ± 1 and w = 0 w=0 w = 0 , which can only happen in the excluded characteristics. Its quotient has equation
r = z + z − 1 , v = ( z − z − 1 ) w / z , v 2 = ( r 2 − 4 ) ( r − 1 ) ( r + 3 ) . r=z+z^{-1},\qquad v=(z-z^{-1})w/z,\qquad v^2=(r^2-4)(r-1)(r+3). r = z + z − 1 , v = ( z − z − 1 ) w / z , v 2 = ( r 2 − 4 ) ( r − 1 ) ( r + 3 ) .
The substitution X = 5 ( r + 2 ) / ( r − 2 ) X=5(r+2)/(r-2) X = 5 ( r + 2 ) / ( r − 2 ) , Y = v ( X − 5 ) 2 / 20 Y=v(X-5)^2/20 Y = v ( X − 5 ) 2 /20 identifies the quotient with E ′ : Y 2 = X ( X − 1 ) ( X + 15 ) E':Y^2=X(X-1)(X+15) E ′ : Y 2 = X ( X − 1 ) ( X + 15 ) . On the other hand, translating x x x on E E E by u = x + 9 u=x+9 u = x + 9 gives y 2 = u 3 − 34 u 2 + 225 u y^2=u^3-34u^2+225u y 2 = u 3 − 34 u 2 + 225 u . The standard two-isogeny with kernel generated by ( 0 , 0 ) (0,0) ( 0 , 0 ) has target η 2 = ξ 3 + 68 ξ 2 + 256 ξ \eta^2=\xi^3+68\xi^2+256\xi η 2 = ξ 3 + 68 ξ 2 + 256 ξ ; writing ξ = 4 ( X − 1 ) \xi=4(X-1) ξ = 4 ( X − 1 ) and η = 8 Y \eta=8Y η = 8 Y gives exactly E ′ E' E ′ . Thus E E E and F F F have identical Frobenius traces, proving the formula for a 1 a_1 a 1 .
We also obtain the atom count A 1 ( P ) A_1(P) A 1 ( P ) directly, independently of the flag cover. Because H = 1 H=1 H = 1 , the automorphism σ \sigma σ of F P \mathbb{F}_P F P is the identity, so the skew polynomial becomes an ordinary reciprocal quartic. Substitute Y = t + t − 1 Y=t+t^{-1} Y = t + t − 1 in this quartic. The necessary and sufficient conditions for irreducibility are
χ ( 9 − 4 a ) = − 1 , χ ( a ( a + 4 ) ) = − 1. (38) \chi(9-4a)=-1,\qquad\chi(a(a+4))=-1. \tag*{(38)} χ ( 9 − 4 a ) = − 1 , χ ( a ( a + 4 )) = − 1. (38)
Indeed the first condition makes Y 2 + Y + a − 2 Y^2+Y+a-2 Y 2 + Y + a − 2 irreducible over F P \mathbb{F}_P F P ; the second says that the norm of Y 2 − 4 Y^2-4 Y 2 − 4 is a nonsquare, so the two remaining roots have degree four. When p ≠ 3 , 5 p\ne3,5 p = 3 , 5 , expansion of (38) has no zero-discriminant correction and gives
A 1 ( P ) = P + 1 + ∑ a χ ( ( 9 − 4 a ) a ( a + 4 ) ) 4 = # E ( F P ) 4 . A_1(P)=\frac{P+1+\sum_a\chi((9-4a)a(a+4))}{4}=\frac{\#E(\mathbb{F}_P)}{4}. A 1 ( P ) = 4 P + 1 + ∑ a χ (( 9 − 4 a ) a ( a + 4 )) = 4 # E ( F P ) .
Under x = − 4 a x=-4a x = − 4 a , the cubic x ( x − 16 ) ( x + 9 ) x(x-16)(x+9) x ( x − 16 ) ( x + 9 ) becomes 16 a ( a + 4 ) ( 9 − 4 a ) 16a(a+4)(9-4a) 16 a ( a + 4 ) ( 9 − 4 a ) . Since 16 = 4 2 16=4^2 16 = 4 2 , the quadratic character of the cubic equals that of a ( a + 4 ) ( 9 − 4 a ) a(a+4)(9-4a) a ( a + 4 ) ( 9 − 4 a ) . The coefficient identity from (28) is
b 1 = κ # B P ( F P ) − m A 1 ( P ) . b_1=\kappa\#\mathcal{B}_P(\mathbb{F}_P)-mA_1(P). b 1 = κ # B P ( F P ) − m A 1 ( P ) .
Since m = 4 κ + 1 m=4\kappa+1 m = 4 κ + 1 , this proves (35) .
For p = 3 p=3 p = 3 , Δ ( z ) = ( z + 1 ) 2 ( z 2 + 1 ) \Delta(z)=(z+1)^2(z^2+1) Δ ( z ) = ( z + 1 ) 2 ( z 2 + 1 ) ; for p = 5 p=5 p = 5 , Δ ( z ) = ( z − 1 ) 2 ( z 2 − z + 1 ) \Delta(z)=(z-1)^2(z^2-z+1) Δ ( z ) = ( z − 1 ) 2 ( z 2 − z + 1 ) . These factorizations explain why the genus-one argument above must not be applied in these two characteristics. The quadratic character sums in (37) give a 1 = − ( − 1 ) f a_1=-(-1)^f a 1 = − ( − 1 ) f and − 1 -1 − 1 , respectively. In (38) exactly one common zero requires correction; direct evaluation gives A 1 = ( P − τ P ) / 4 A_1=(P-\tau_P)/4 A 1 = ( P − τ P ) /4 . This proves (36) .
Finally consider the integral model
E min : y 2 + x y + y = x 3 + x 2 − 10 x − 10. E_{\min}: \quad y^2+xy+y=x^3+x^2-10x-10. E m i n : y 2 + x y + y = x 3 + x 2 − 10 x − 10.
The change of coordinates
( x , y ) ⟼ ( 4 x + 4 , 8 y + 4 x + 4 ) (x,y)\longmapsto(4x+4,8y+4x+4) ( x , y ) ⟼ ( 4 x + 4 , 8 y + 4 x + 4 )
from E min E_{\min} E m i n to E E E gives Y 2 = X ( X − 16 ) ( X + 9 ) Y^2=X(X-16)(X+9) Y 2 = X ( X − 16 ) ( X + 9 ) , as is checked by direct substitution. This model belongs to isogeny class 15a, with the newform stated above; see [12 ] . The prime-power recursion is the characteristic-polynomial identity for its two Frobenius eigenvalues. □
The quadratic coefficients: trace–norm descent
For extension degree H = 2 H=2 H = 2 the preceding one-variable elliptic reduction is no longer available in the same form. The useful replacement is the trace–norm map
s ⟼ ( T , N ) = ( s + s P , s P + 1 ) s\longmapsto(T,N)=(s+s^P,s^{P+1}) s ⟼ ( T , N ) = ( s + s P , s P + 1 )
from F P 2 \mathbb{F}_{P^2} F P 2 to F P 2 \mathbb{F}_P^2 F P 2 . The polynomial X 2 − T X + N X^2-TX+N X 2 − T X + N has discriminant T 2 − 4 N T^2-4N T 2 − 4 N ; the number of s ∈ F P 2 s\in\mathbb{F}_{P^2} s ∈ F P 2 with the given trace and norm is 1 − χ ( T 2 − 4 N ) 1-\chi(T^2-4N) 1 − χ ( T 2 − 4 N ) . This multiplicity will be built into both character sums.
Define the following sums over the base field. For T , N ∈ F P T,N\in\mathbb{F}_P T , N ∈ F P set
W ( T , N ) = 1 − χ ( T 2 − 4 N ) , B ( T , N ) = 3 T + 4 N + 4 , W(T,N)=1-\chi(T^2-4N),\qquad B(T,N)=3T+4N+4, W ( T , N ) = 1 − χ ( T 2 − 4 N ) , B ( T , N ) = 3 T + 4 N + 4 ,
and put
U P = ∑ T , N ∈ F P W ( T , N ) χ ( N ( B ( T , N ) 2 + T B ( T , N ) + N ) ) , U_P=\sum_{T,N\in\mathbb{F}_P}W(T,N)\chi\bigl(N(B(T,N)^2+TB(T,N)+N)\bigr), U P = T , N ∈ F P ∑ W ( T , N ) χ ( N ( B ( T , N ) 2 + T B ( T , N ) + N ) ) ,
V P = ∑ T , N ∈ F P W ( T , N ) ν ( ( T − 1 ) 2 − 4 N + 8 ) ν ( ( N + 2 ) 2 − 4 ( T − 1 ) 2 ) , V_P=\sum_{T,N\in\mathbb{F}_P}W(T,N)\nu\bigl((T-1)^2-4N+8\bigr)\nu\bigl((N+2)^2-4(T-1)^2\bigr), V P = T , N ∈ F P ∑ W ( T , N ) ν ( ( T − 1 ) 2 − 4 N + 8 ) ν ( ( N + 2 ) 2 − 4 ( T − 1 ) 2 ) ,
where ν ( x ) = ( χ ( x ) 2 − χ ( x ) ) / 2 \nu(x)=(\chi(x)^2-\chi(x))/2 ν ( x ) = ( χ ( x ) 2 − χ ( x )) /2 is exactly the indicator of nonzero nonsquares. All constants in these sums are interpreted in F P \mathbb{F}_P F P ; there is no restriction excluding p = 3 , 5 p=3,5 p = 3 , 5 .
Theorem 12.2. The second coefficients are
a 2 = τ P 2 + U P 2 , b 2 = b 1 2 − m V P + κ ( P 2 + 1 + U P ) 2 . (39) a_2=\frac{\tau_P^2+U_P}{2},\qquad b_2=\frac{b_1^2-mV_P+\kappa(P^2+1+U_P)}{2}. \tag*{(39)} a 2 = 2 τ P 2 + U P , b 2 = 2 b 1 2 − m V P + κ ( P 2 + 1 + U P ) . (39) In particular these formulas contain no unevaluated finite-group characters or unknown Frobenius eigenvalues; the two remaining sums are explicit quadratic-character sums over the base field.
Proof. Over F P 2 \mathbb{F}_{P^2} F P 2 , regard (22) as a quadratic in z z z :
( s P + 1 + s ) z 2 + s z + s + 1 = 0. (s^{P+1}+s)z^2+sz+s+1=0. ( s P + 1 + s ) z 2 + sz + s + 1 = 0.
The exceptional values are s = 0 , − 1 s=0,-1 s = 0 , − 1 . The first gives no affine points, and the second gives only the excluded z = 0 z=0 z = 0 . For every other s s s , the constant and leading coefficients are nonzero, so its number of solutions is 1 + χ 2 ( D ( s ) ) 1+\chi_2(D(s)) 1 + χ 2 ( D ( s )) , where
D ( s ) = s 2 − 4 ( s P + 1 + s ) ( s + 1 ) . D(s)=s^2-4(s^{P+1}+s)(s+1). D ( s ) = s 2 − 4 ( s P + 1 + s ) ( s + 1 ) .
Since D ( 0 ) = 0 D(0)=0 D ( 0 ) = 0 and D ( − 1 ) = 1 D(-1)=1 D ( − 1 ) = 1 , the projective point count is P 2 + 1 + ∑ s χ 2 ( D ( s ) ) P^2+1+\sum_s\chi_2(D(s)) P 2 + 1 + ∑ s χ 2 ( D ( s )) . Writing T = s + s P T=s+s^P T = s + s P and N = s P + 1 N=s^{P+1} N = s P + 1 gives
D ( s ) = − ( 3 T + 4 N + 4 ) s − N , Norm ( D ( s ) ) = N ( ( 3 T + 4 N + 4 ) 2 + T ( 3 T + 4 N + 4 ) + N ) . D(s)=-(3T+4N+4)s-N,\qquad\operatorname{Norm}(D(s))=N\left((3T+4N+4)^2+T(3T+4N+4)+N\right). D ( s ) = − ( 3 T + 4 N + 4 ) s − N , Norm ( D ( s )) = N ( ( 3 T + 4 N + 4 ) 2 + T ( 3 T + 4 N + 4 ) + N ) .
There are W ( T , N ) W(T,N) W ( T , N ) elements with this trace and norm, and χ 2 ( x ) = χ ( Norm ( x ) ) \chi_2(x)=\chi(\operatorname{Norm}(x)) χ 2 ( x ) = χ ( Norm ( x )) . Thus # B P ( F P 2 ) = P 2 + 1 + U P \#B_P(\mathbb{F}_{P^2})=P^2+1+U_P # B P ( F P 2 ) = P 2 + 1 + U P , and the second Steinberg power trace is − U P -U_P − U P .
For the atom count at H = 2 H=2 H = 2 , direct multiplication of the two companion matrices shows that, for T = a + a P T=a+a^P T = a + a P and N = a P + 1 N=a^{P+1} N = a P + 1 ,
det ( X I 4 − C a C a P ) = X 4 + ( T − 1 ) X 3 + N X 2 + ( T − 1 ) X + 1. \det(XI_4-C_aC_{a^P})=X^4+(T-1)X^3+NX^2+(T-1)X+1. det ( X I 4 − C a C a P ) = X 4 + ( T − 1 ) X 3 + N X 2 + ( T − 1 ) X + 1.
The same reciprocal-quartic test used in (38) gives precisely the two nonsquare indicators defining V P V_P V P . The multiplicity of ( T , N ) (T,N) ( T , N ) is again W ( T , N ) W(T,N) W ( T , N ) , so A 2 ( P ) = V P A_2(P)=V_P A 2 ( P ) = V P . Equation (28) identifies the second cuspidal power trace as m V P − κ ( P 2 + 1 + U P ) mV_P-\kappa(P^2+1+U_P) m V P − κ ( P 2 + 1 + U P ) . Newton’s identity [ u 2 ] ∏ i ( 1 − α i u ) = ( ( ∑ i α i ) 2 − ∑ i α i 2 ) / 2 [u^2]\prod_i(1-\alpha_i u)=((\sum_i\alpha_i)^2-\sum_i\alpha_i^2)/2 [ u 2 ] ∏ i ( 1 − α i u ) = (( ∑ i α i ) 2 − ∑ i α i 2 ) /2 proves (39) . □
For example, exact evaluations give
Table 3.
Theorem 12.1 evaluates the linear coefficients through a fixed elliptic curve (or its explicit modular form). Theorem 12.2 gives exact base-field character sums for the quadratic coefficients. A uniform modular or other closed evaluation of these sums, and evaluations of all higher coefficients for general P P P , are not asserted.
Simplifying the degree-two atom count
The sum V P V_P V P is an explicit answer, but it involves two nonsquare indicators. To simplify it, we expand these indicators and evaluate the six lower-order character sums separately. The only remaining three-factor sum is J P J_P J P below. This calculation is valid over every finite field of odd order, without the monodromy restriction.
K = T 2 − 4 N , D = ( T − 1 ) 2 − 4 N + 8 , F = ( N + 2 ) 2 − 4 ( T − 1 ) 2 , J P = ∑ T , N ∈ F P χ ( K D F ) . K=T^2-4N,\qquad D=(T-1)^2-4N+8,\qquad F=(N+2)^2-4(T-1)^2,\qquad J_P=\sum_{T,N\in\mathbb{F}_P}\chi(KDF). K = T 2 − 4 N , D = ( T − 1 ) 2 − 4 N + 8 , F = ( N + 2 ) 2 − 4 ( T − 1 ) 2 , J P = T , N ∈ F P ∑ χ ( K D F ) .
Theorem 12.3. For every odd prime power P P P ,
4 A 2 ( P ) = 4 V P = P 2 + 1 + P χ ( − 3 ) + χ ( − 15 ) − J P . (40) 4A_2(P)=4V_P=P^2+1+P\chi(-3)+\chi(-15)-J_P. \tag*{(40)} 4 A 2 ( P ) = 4 V P = P 2 + 1 + P χ ( − 3 ) + χ ( − 15 ) − J P . (40) In particular, under the hypotheses defining C P C_P C P above,
8 b 2 = 4 b 1 2 − P 2 − 1 − m P χ ( − 3 ) − m χ ( − 15 ) + m J P + 4 κ U P . 8b_2=4b_1^2-P^2-1-mP\chi(-3)-m\chi(-15)+mJ_P+4\kappa U_P. 8 b 2 = 4 b 1 2 − P 2 − 1 − m P χ ( − 3 ) − m χ ( − 15 ) + m J P + 4 κ U P .
Proof. All sums in this proof are over ( T , N ) ∈ F P 2 (T,N)\in\mathbb{F}_P^2 ( T , N ) ∈ F P 2 . We evaluate
∑ χ K = ∑ χ D = ∑ χ F = ∑ χ ( K D ) = 0 , (41) \sum\chi K=\sum\chi D=\sum\chi F=\sum\chi(KD)=0, \tag*{(41)} ∑ χK = ∑ χD = ∑ χ F = ∑ χ ( K D ) = 0 , (41)
∑ χ ( D F ) = 1 , ∑ χ ( K F ) = 1 + P χ ( − 3 ) . \sum\chi(DF)=1,\qquad\sum\chi(KF)=1+P\chi(-3). ∑ χ ( D F ) = 1 , ∑ χ ( K F ) = 1 + P χ ( − 3 ) .
The first two sums vanish by summing in N N N . For F F F , put v = T − 1 v=T-1 v = T − 1 : the quadratic sum in N N N is − 1 -1 − 1 for v ≠ 0 v\ne0 v = 0 and P − 1 P-1 P − 1 for v = 0 v=0 v = 0 . For K D KD K D , the two linear factors in N N N coincide exactly at T = 9 / 2 T=9/2 T = 9/2 ; the same quadratic-sum rule proves vanishing, including in characteristic three. For the fifth sum, adjoin z z z with z 2 = D z^2=D z 2 = D . Then N = ( v 2 + 8 − z 2 ) / 4 N=(v^2+8-z^2)/4 N = ( v 2 + 8 − z 2 ) /4 , and the invertible change x = v − z x=v-z x = v − z , y = v + z y=v+z y = v + z gives
F = ( x 2 − 16 ) ( y 2 − 16 ) 16 . F=\frac{(x^2-16)(y^2-16)}{16}. F = 16 ( x 2 − 16 ) ( y 2 − 16 ) .
Consequently
∑ x ( 1 + χ D ) χ F = ( ∑ x χ ( x 2 − 16 ) ) 2 = 1. \sum_x(1+\chi_D)\chi_F=\left(\sum_x\chi(x^2-16)\right)^2=1. x ∑ ( 1 + χ D ) χ F = ( x ∑ χ ( x 2 − 16 ) ) 2 = 1.
Since ∑ x χ F = 0 \sum_x\chi_F=0 ∑ x χ F = 0 , this proves the fifth identity.
For the last sum, instead adjoin z 2 = K z^2=K z 2 = K and put x = T − z x=T-z x = T − z , y = T + z y=T+z y = T + z , so N = x y / 4 N=xy/4 N = x y /4 . Now
F = ( x − 4 ) ( y − 4 ) ( x y + 4 x + 4 y ) 16 . F=\frac{(x-4)(y-4)(xy+4x+4y)}{16}. F = 16 ( x − 4 ) ( y − 4 ) ( x y + 4 x + 4 y ) .
Translating X = x − 4 X=x-4 X = x − 4 , Y = y − 4 Y=y-4 Y = y − 4 yields
∑ X , Y ( 1 + χ K ) χ F = ∑ X , Y χ ( X Y ( ( X + 8 ) Y + 8 X + 48 ) ) . \sum_{X,Y}(1+\chi_K)\chi_F=\sum_{X,Y}\chi\bigl(XY((X+8)Y+8X+48)\bigr). X , Y ∑ ( 1 + χ K ) χ F = X , Y ∑ χ ( X Y (( X + 8 ) Y + 8 X + 48 ) ) .
For X ≠ − 8 , − 6 X\ne-8,-6 X = − 8 , − 6 , the sum over Y Y Y is − χ ( X ( X + 8 ) ) -\chi(X(X+8)) − χ ( X ( X + 8 )) . At X = − 8 X=-8 X = − 8 it is zero; at X = − 6 X=-6 X = − 6 its correction to this generic expression is P χ ( − 12 ) P\chi(-12) P χ ( − 12 ) . This assertion also holds in characteristic three, when the correction is zero because X = − 6 = 0 X=-6=0 X = − 6 = 0 . Since ∑ X χ ( X ( X + 8 ) ) = − 1 \sum_X\chi(X(X+8))=-1 ∑ X χ ( X ( X + 8 )) = − 1 and χ ( − 12 ) = χ ( − 3 ) \chi(-12)=\chi(-3) χ ( − 12 ) = χ ( − 3 ) , the last identity in (41) follows.
It remains to account for zero discriminants before expanding the nonsquare indicators. Define
R = ∑ ( 1 − χ K ) ( 1 − χ D ) ( 1 − χ F ) − 4 V P . R=\sum(1-\chi_K)(1-\chi_D)(1-\chi_F)-4V_P. R = ∑ ( 1 − χ K ) ( 1 − χ D ) ( 1 − χ F ) − 4 V P .
On D = 0 D=0 D = 0 one has F = ( N − 6 ) 2 F=(N-6)^2 F = ( N − 6 ) 2 . On the two lines F = 0 F=0 F = 0 , namely N = 2 T − 4 N=2T-4 N = 2 T − 4 and N = − 2 T N=-2T N = − 2 T , one has respectively D = ( T − 5 ) 2 D=(T-5)^2 D = ( T − 5 ) 2 and D = ( T + 3 ) 2 D=(T+3)^2 D = ( T + 3 ) 2 . Thus only the two intersections ( T , N ) = ( 5 , 6 ) , ( − 3 , 6 ) (T,N)=(5,6),(-3,6) ( T , N ) = ( 5 , 6 ) , ( − 3 , 6 ) contribute to R R R . Their contributions are 1 − χ ( 1 ) = 0 1-\chi(1)=0 1 − χ ( 1 ) = 0 and 1 − χ ( − 15 ) 1-\chi(-15) 1 − χ ( − 15 ) . They remain distinct in every odd characteristic. Therefore R = 1 − χ ( − 15 ) R=1-\chi(-15) R = 1 − χ ( − 15 ) . Expanding the product and using (41) gives
4 V P = P 2 + ( 1 + P χ ( − 3 ) ) + 1 − J P − ( 1 − χ ( − 15 ) ) , 4V_P=P^2+(1+P\chi(-3))+1-J_P-(1-\chi(-15)), 4 V P = P 2 + ( 1 + P χ ( − 3 )) + 1 − J P − ( 1 − χ ( − 15 )) ,
which is (40) . The coefficient formula follows from (39) and m = 4 κ + 1 m=4\kappa+1 m = 4 κ + 1 .
Eliminating the second character sum
The apparently different sums J P J_P J P and U P U_P U P are not independent. The following identity is proved by counting a double cover of an affine surface in two different sets of coordinates. In particular, it is an algebraic character-sum identity, not an appeal to a hypothetical elliptic-curve interpretation of the degree-two count.
Theorem 12.4. For every finite field of odd order P P P ,
J P = 1 + P χ ( − 3 ) − U P . (42) J_P=1+P\chi(-3)-U_P. \tag*{(42)} J P = 1 + P χ ( − 3 ) − U P . (42)
Consequently
A 2 ( P ) = P 2 + χ ( − 15 ) + U P 4 . (43) A_2(P)=\frac{P^2+\chi(-15)+U_P}{4}. \tag*{(43)} A 2 ( P ) = 4 P 2 + χ ( − 15 ) + U P . (43)
Under the hypotheses for S P S_P S P , C P C_P C P above, the quadratic coefficients are
a 2 = τ P 2 + U P 2 , b 2 = 4 b 1 2 − P 2 − 1 − U P + m ( 1 − χ ( − 15 ) ) 8 . (44) a_2=\frac{\tau_P^2+U_P}{2},\qquad
b_2=\frac{4b_1^2-P^2-1-U_P+m(1-\chi(-15))}{8}. \tag*{(44)} a 2 = 2 τ P 2 + U P , b 2 = 8 4 b 1 2 − P 2 − 1 − U P + m ( 1 − χ ( − 15 )) . (44)
Thus only one character sum, U P U_P U P , is needed for both quadratic coefficients. Equivalently, they satisfy the evaluated relation
8 b 2 + 2 a 2 = 4 b 1 2 + τ P 2 − P 2 − 1 + m ( 1 − χ ( − 15 ) ) . (45) 8b_2+2a_2=4b_1^2+\tau_P^2-P^2-1+m(1-\chi(-15)). \tag*{(45)} 8 b 2 + 2 a 2 = 4 b 1 2 + τ P 2 − P 2 − 1 + m ( 1 − χ ( − 15 )) . (45)
Proof. Put ϵ = χ ( − 3 ) \epsilon=\chi(-3) ϵ = χ ( − 3 ) and
Q ( T , N ) = 12 T 2 + 28 ( N + 1 ) T + 16 ( N + 1 ) 2 + N , S 0 = ∑ T , N χ ( N Q ( T , N ) ) . Q(T,N)=12T^2+28(N+1)T+16(N+1)^2+N,\qquad S_0=\sum_{T,N}\chi(NQ(T,N)). Q ( T , N ) = 12 T 2 + 28 ( N + 1 ) T + 16 ( N + 1 ) 2 + N , S 0 = T , N ∑ χ ( N Q ( T , N )) .
We first evaluate S 0 = P ( 1 + ϵ ) S_0=P(1+\epsilon) S 0 = P ( 1 + ϵ ) . If p ≠ 3 p\ne3 p = 3 , the discriminant of Q Q Q as a quadratic in T T T is 16 ( N 2 − N + 1 ) 16(N^2-N+1) 16 ( N 2 − N + 1 ) . Summing in T T T and using ∑ N χ N = 0 \sum_N\chi N=0 ∑ N χ N = 0 gives
S 0 = P ∑ N 2 − N + 1 = 0 χ ( 12 N ) = P ( 1 + ϵ ) . S_0=P\sum_{N^2-N+1=0}\chi(12N)=P(1+\epsilon). S 0 = P N 2 − N + 1 = 0 ∑ χ ( 12 N ) = P ( 1 + ϵ ) .
Indeed, ( N + 1 ) 2 = 3 N (N+1)^2=3N ( N + 1 ) 2 = 3 N at each such root, so every summand equals one. If p = 3 p=3 p = 3 , then Q = ( N + 1 ) T + N 2 + 1 Q=(N+1)T+N^2+1 Q = ( N + 1 ) T + N 2 + 1 . All inner sums vanish except the one at N = − 1 N=-1 N = − 1 , where N Q = 1 NQ=1 N Q = 1 and the contribution is P P P . This proves the same evaluation in characteristic three.
Define the polynomial
G 0 ( a , b ) = a b ( 4 a b + 4 a + 3 b + 4 ) ( 4 a b + 3 a + 4 b + 4 ) , Σ 0 = ∑ a , b χ ( G 0 ( a , b ) ) . G_0(a,b)=ab(4ab+4a+3b+4)(4ab+3a+4b+4),\qquad\Sigma_0=\sum_{a,b}\chi(G_0(a,b)). G 0 ( a , b ) = ab ( 4 ab + 4 a + 3 b + 4 ) ( 4 ab + 3 a + 4 b + 4 ) , Σ 0 = a , b ∑ χ ( G 0 ( a , b )) .
The change T = a + b T=a+b T = a + b , N = a b N=ab N = ab has 1 + χ ( T 2 − 4 N ) 1+\chi(T^2-4N) 1 + χ ( T 2 − 4 N ) ordered preimages. The polynomial identity
N Q ( T , N ) ∣ T = a + b , N = a b = G 0 ( a , b ) NQ(T,N)\big|_{T=a+b,N=ab}=G_0(a,b) N Q ( T , N ) T = a + b , N = ab = G 0 ( a , b )
therefore gives, directly from the definition of U P U_P U P ,
Σ 0 = 2 S 0 − U P = 2 P ( 1 + ϵ ) − U P . (46) \Sigma_0=2S_0-U_P=2P(1+\epsilon)-U_P. \tag*{(46)} Σ 0 = 2 S 0 − U P = 2 P ( 1 + ϵ ) − U P . (46)
Next count the affine scheme
M : x 2 = K ( T , N ) , y 2 = D ( T , N ) , z 2 = F ( T , N ) . \mathcal{M}: \quad x^2=K(T,N),\quad y^2=D(T,N),\quad z^2=F(T,N). M : x 2 = K ( T , N ) , y 2 = D ( T , N ) , z 2 = F ( T , N ) .
Expansion of its fiber sizes, followed by (41) , gives
# M ( F P ) = P 2 + 2 + P ϵ + J P . \#\mathcal{M}(\mathbb{F}_P)=P^2+2+P\epsilon+J_P. # M ( F P ) = P 2 + 2 + P ϵ + J P .
Conversely, x , y x,y x , y determine T , N T,N T , N uniquely:
T = x 2 − y 2 + 9 2 , N = T 2 − x 2 4 . T=\frac{x^2-y^2+9}{2},\qquad N=\frac{T^2-x^2}{4}. T = 2 x 2 − y 2 + 9 , N = 4 T 2 − x 2 .
Put u = x − y u=x-y u = x − y , v = x + y v=x+y v = x + y . Substitution yields the polynomial identity
256 F ( T , N ) = ( u 2 − 1 ) ( v 2 − 1 ) ( ( u − 1 ) ( v + 1 ) + 16 ) ( ( u + 1 ) ( v − 1 ) + 16 ) . 256F(T,N)=(u^2-1)(v^2-1)\bigl((u-1)(v+1)+16\bigr)\bigl((u+1)(v-1)+16\bigr). 256 F ( T , N ) = ( u 2 − 1 ) ( v 2 − 1 ) ( ( u − 1 ) ( v + 1 ) + 16 ) ( ( u + 1 ) ( v − 1 ) + 16 ) .
Writing the product on the right as G ( u , v ) G(u,v) G ( u , v ) , we obtain
Z 0 : = ∑ u , v χ ( G ( u , v ) ) = 2 + P ϵ + J P . Z_0:=\sum_{u,v}\chi(G(u,v))=2+P\epsilon+J_P. Z 0 := u , v ∑ χ ( G ( u , v )) = 2 + P ϵ + J P .
On u , v ≠ − 1 u,v\ne-1 u , v = − 1 , make the invertible substitutions
u = 1 − a 1 + a , v = 1 − b 1 + b , a , b ≠ − 1. u=\frac{1-a}{1+a},\qquad v=\frac{1-b}{1+b},\qquad a,b\ne-1. u = 1 + a 1 − a , v = 1 + b 1 − b , a , b = − 1.
They satisfy
G ( u , v ) = 256 G 0 ( a , b ) ( 1 + a ) 4 ( 1 + b ) 4 . G(u,v)=\frac{256G_0(a,b)}{(1+a)^4(1+b)^4}. G ( u , v ) = ( 1 + a ) 4 ( 1 + b ) 4 256 G 0 ( a , b ) .
Every factor removed from the quadratic character is a nonzero square. The excluded lines u = − 1 u=-1 u = − 1 or v = − 1 v=-1 v = − 1 contribute zero to Z 0 Z_0 Z 0 . On the other side, G 0 ( − 1 , b ) = b 2 G_0(-1,b)=b^2 G 0 ( − 1 , b ) = b 2 and G 0 ( a , − 1 ) = a 2 G_0(a,-1)=a^2 G 0 ( a , − 1 ) = a 2 ; their character sums are each P − 1 P-1 P − 1 , and their intersection contributes one. Consequently the exact boundary correction is
Z 0 = Σ 0 − ( 2 P − 3 ) . Z_0=\Sigma_0-(2P-3). Z 0 = Σ 0 − ( 2 P − 3 ) .
Together with (46) this proves (42) , including characteristics three and five. Substituting (42) into (40) proves (43) , and substitution into (39) gives (44) and (45) . □
Remark 12.5 (Elliptic factors and point counts). At P = 3 P=3 P = 3 , the factor 1 + u + 3 u 2 1+u+3u^2 1 + u + 3 u 2 in S 3 = N S_3=N S 3 = N is the Frobenius polynomial of the elliptic curve y 2 = x 3 + 2 x 2 + 1 y^2=x^3+2x^2+1 y 2 = x 3 + 2 x 2 + 1 over F 3 \mathbb{F}_3 F 3 , which has five rational points. By semisimplicity and Tate’s isogeny theorem for abelian varieties over finite fields [15 ] , this elliptic curve is an isogeny factor of Jac ( B 3 ) \operatorname{Jac}(B_3) Jac ( B 3 ) over F 3 \mathbb{F}_3 F 3 .
For p ≠ 3 , 5 p\ne3,5 p = 3 , 5 , the equality # B P ( F P ) = # E ( F P ) \#B_P(\mathbb{F}_P)=\#E(\mathbb{F}_P) # B P ( F P ) = # E ( F P ) compares the first Frobenius traces over F P \mathbb{F}_P F P . A comparison over extension fields would require higher Frobenius power traces. The quadratic-coefficient identities above express the next coefficients of S P S_P S P and C P C_P C P in terms of the finite-field sum U P U_P U P .
A quotient of covering-curve zeta functions
Remark 13.1. Let X X X be the symplectic frame torsor of V \mathcal{V} V , let Z = { ± I } Z=\{\pm I\} Z = { ± I } , and let X Z X_Z X Z and X T X_T X T be the smooth projective compactifications of X / Z X/Z X / Z and X / T X/T X / T , respectively. The latter quotient uses the Coxeter torus T T T lifting T 0 T_0 T 0 . For each boundary point of X T X_T X T , the decomposition group of X Z → X T X_Z\to X_T X Z → X T is the image in T / Z T/Z T / Z of the intersection of T T T with a conjugate of a decomposition group of X X X over infinity. Each such conjugate preserves a Lagrangian plane: the original group preserves the positive-valuation plane. Every element of T ∖ Z T\setminus Z T ∖ Z acts irreducibly, so this intersection is contained in Z Z Z . Thus every boundary point splits completely in the degree-m m m map X Z → X T X_Z\to X_T X Z → X T .
On the affine line, Frobenius acts on the fibres of X / T X/T X / T and X / Z X/Z X / Z with permutation characters Ind T 0 S 1 \operatorname{Ind}_{T_0}^{S}1 Ind T 0 S 1 and Reg S \operatorname{Reg}_S Reg S , respectively. Summing their difference m Ind T 0 S 1 − Reg S = 4 m 1 i r r m\operatorname{Ind}_{T_0}^{S}1-\operatorname{Reg}_S=4m\mathbf{1}_{\mathrm{irr}} m Ind T 0 S 1 − Reg S = 4 m 1 irr over F P H \mathbb{F}_{P^H} F P H gives the affine point-count difference. Complete splitting makes the corresponding boundary contributions cancel, so
4 m A H ( P ) = m # X T ( F P H ) − # X Z ( F P H ) , R P ( u ) = Z ( X T , u ) m Z ( X Z , u ) . 4mA_H(P)=m\#X_T(\mathbb{F}_{P^H})-\#X_Z(\mathbb{F}_{P^H}),\qquad R_P(u)=\frac{Z(X_T,u)^m}{Z(X_Z,u)}. 4 m A H ( P ) = m # X T ( F P H ) − # X Z ( F P H ) , R P ( u ) = Z ( X Z , u ) Z ( X T , u ) m .
The second identity follows by exponentiating the first with weights u H / H u^H/H u H / H .
Appendix A. Higher-rank symplectic flag covers and Steinberg factors
The flag curve of § 3.3 is an instance of a construction which applies to symplectic local systems of arbitrary even rank. The underlying reason for considering full flags is the Solomon–Tits realization of the Steinberg representation in the top reduced homology of the spherical building. We record the general construction and its cohomological consequence, without assuming that the required higher-rank local systems arise from the skew polynomials considered in this paper.
Let k = F P k=\mathbb{F}_P k = F P , let X X X be a smooth geometrically connected affine curve over k k k , and let V \mathcal{V} V be a lisse sheaf of 2 n 2n 2 n -dimensional F P \mathbb{F}_P F P -vector spaces on X X X , equipped with a nondegenerate alternating pairing. Assume that its geometric monodromy is G = Sp 2 n ( F P ) G=\operatorname{Sp}_{2n}(\mathbb{F}_P) G = Sp 2 n ( F P ) ; in particular the arithmetic monodromy is also G G G . Fix a symplectic fibre V V V . Write Δ \Delta Δ for the set of n n n simple roots of G G G , and for J ⊆ Δ J\subseteq\Delta J ⊆ Δ let P J P_J P J be a standard parabolic subgroup, with P ∅ = B P_{\varnothing}=B P ∅ = B a Borel subgroup and P Δ = G P_\Delta=G P Δ = G . The character Ind P J G 1 \operatorname{Ind}_{P_J}^{G}1 Ind P J G 1 is the permutation character of G G G on the coset set G / P J G/P_J G / P J . The finite transitive G G G -set G / P J G/P_J G / P J consists of isotropic partial flags of the corresponding type. By the monodromy action it defines a finite étale cover
π J : Y J ⟶ X . \pi_J:Y_J\longrightarrow X. π J : Y J ⟶ X .
Every Y J Y_J Y J is geometrically connected. The cover Y ∅ Y_{\varnothing} Y ∅ parametrizes complete isotropic flags
0 = V 0 ⊂ V 1 ⊂ ⋯ ⊂ V n ⊂ V , dim F P V i = i , V n = V n ⊥ , 0=V_0\subset V_1\subset\cdots\subset V_n\subset V,\qquad\dim_{\mathbb{F}_P}V_i=i,\qquad V_n=V_n^\perp, 0 = V 0 ⊂ V 1 ⊂ ⋯ ⊂ V n ⊂ V , dim F P V i = i , V n = V n ⊥ ,
and Y Δ = X Y_\Delta=X Y Δ = X . Although the flags have length n n n , each Y J Y_J Y J remains a curve, since it is finite over X X X .
Proposition A.1 (The building identity for flag covers). Let St G \operatorname{St}_G St G be the Steinberg character of G G G , and let F x , h ∈ G F_{x,h}\in G F x , h ∈ G denote the Frobenius conjugacy class of the fibre at x ∈ X ( F P h ) x\in X(\mathbb{F}_{P^h}) x ∈ X ( F P h ) . Then
St G = ∑ J ⊆ Δ ( − 1 ) ∣ J ∣ Ind P J G 1 , (47) \operatorname{St}_G=\sum_{J\subseteq\Delta}(-1)^{|J|}\operatorname{Ind}_{P_J}^{G}1, \tag*{(47)} St G = J ⊆ Δ ∑ ( − 1 ) ∣ J ∣ Ind P J G 1 , (47)
and, for every h ≥ 1 h \ge1 h ≥ 1 ,
∑ x ∈ X ( F p h ) St G ( F x , h ) = ∑ J ⊆ Δ ( − 1 ) ∣ J ∣ # Y J ( F p h ) . (48) \sum_{x \in X(\mathbb{F}_{p^h})} \operatorname{St}_{G}(F_{x,h}) = \sum_{J \subseteq\Delta} (-1)^{|J|} \#Y_J(\mathbb{F}_{p^h}). \tag*{(48)} x ∈ X ( F p h ) ∑ St G ( F x , h ) = J ⊆ Δ ∑ ( − 1 ) ∣ J ∣ # Y J ( F p h ) . (48)
If E S t \mathcal{E}_{\mathrm{St}} E St is the lisse Q ‾ ℓ \overline{\mathbb{Q}}_\ell Q ℓ -sheaf associated with St G \operatorname{St}_{G} St G (ℓ ≠ p \ell\ne p ℓ = p ), then its L L L -function satisfies
L ( X , E S t , u ) = ∏ J ⊆ Δ Z ( Y J , u ) ( − 1 ) ∣ J ∣ . (49) L(X,\mathcal{E}_{\mathrm{St}},u) = \prod_{J \subseteq\Delta} Z(Y_J,u)^{(-1)^{|J|}}. \tag*{(49)} L ( X , E St , u ) = J ⊆ Δ ∏ Z ( Y J , u ) ( − 1 ) ∣ J ∣ . (49)
Proof. The simplices of the spherical building of G G G are the proper isotropic flags; its augmented simplicial chain complex has, in each degree, the permutation modules on flags of the relevant types. The Solomon–Tits theorem (see, e.g., [3 ] ) identifies its sole nonzero reduced homology with St G \operatorname{St}_{G} St G , giving (A.1) by the Euler characteristic. The value of Ind P J G 1 \operatorname{Ind}_{P_J}^{G}1 Ind P J G 1 at F x , h F_{x,h} F x , h counts F x , h F_{x,h} F x , h -fixed flags in G / P J G/P_J G / P J , equivalently the F p h \mathbb{F}_{p^h} F p h -points of the fibre of Y J Y_J Y J above x x x . Summing over x x x gives (A.2). Finally, apply the exponential definition of the L L L -function to the trace identity for all h h h ; the zeta function of Y J Y_J Y J is the L L L -function of the permutation sheaf ( π J ) ∗ Q ‾ ℓ (\pi_J)_*\overline{\mathbb{Q}}_\ell ( π J ) ∗ Q ℓ .
For comparison with projective curves, write Y ‾ J \overline{Y}_J Y J for the smooth projective compactification of Y J Y_J Y J and D J = Y ‾ J ∖ Y J D_J=\overline{Y}_J\setminus Y_J D J = Y J ∖ Y J for its reduced geometric boundary. Let
P J g e o m ( u ) = det ( 1 − u F ∣ H 1 ( Y ‾ J , k ‾ , Q ‾ ℓ ) ) , B J ( u ) = det ( 1 − u F ∣ H 0 ( D J , k ‾ , Q ‾ ℓ ) ) . P_J^{\mathrm{geom}}(u)=\det(1-uF\mid H^1(\overline{Y}_{J,\overline{k}},\overline{\mathbb{Q}}_\ell)),\qquad B_J(u)=\det(1-uF\mid H^0(D_{J,\overline{k}},\overline{\mathbb{Q}}_\ell)). P J geom ( u ) = det ( 1 − u F ∣ H 1 ( Y J , k , Q ℓ )) , B J ( u ) = det ( 1 − u F ∣ H 0 ( D J , k , Q ℓ )) .
The finite set D J D_J D J carries its natural Frobenius permutation action. The localization sequence gives
det ( 1 − u F ∣ H c 1 ( Y J , k ‾ ) ) = P J g e o m ( u ) B J ( u ) 1 − u . \det(1-uF\mid H^1_c(Y_{J,\overline{k}}))=P_J^{\mathrm{geom}}(u)\frac{B_J(u)}{1-u}. det ( 1 − u F ∣ H c 1 ( Y J , k )) = P J geom ( u ) 1 − u B J ( u ) .
Since Z ( Y J , u ) = Z ( Y ‾ J , u ) / Z ( D J , u ) Z(Y_J,u)=Z(\overline{Y}_J,u)/Z(D_J,u) Z ( Y J , u ) = Z ( Y J , u ) / Z ( D J , u ) and ∑ J ( − 1 ) ∣ J ∣ = 0 \sum_J(-1)^{|J|}=0 ∑ J ( − 1 ) ∣ J ∣ = 0 , equation (A.3) becomes
L ( X , E S t , u ) = ∏ J ⊆ Δ ( P J g e o m ( u ) B J ( u ) ) ( − 1 ) ∣ J ∣ . (50) L(X,\mathcal{E}_{\mathrm{St}},u)=\prod_{J\subseteq\Delta}\left(P_J^{\mathrm{geom}}(u)B_J(u)\right)^{(-1)^{|J|}}. \tag*{(50)} L ( X , E St , u ) = J ⊆ Δ ∏ ( P J geom ( u ) B J ( u ) ) ( − 1 ) ∣ J ∣ . (50)
In particular, the boundary correction is indispensable in general. If X = A 1 X=\mathbb{A}^1 X = A 1 , the absence of compactly supported cohomology in degrees 0 and 2 for the nontrivial Steinberg local system gives
L ( X , E S t , u ) = det ( 1 − u F ∣ H c 1 ( X k ‾ , E S t ) ) . L(X,\mathcal{E}_{\mathrm{St}},u)=\det(1-uF\mid H^1_c(X_{\overline{k}},\mathcal{E}_{\mathrm{St}})). L ( X , E St , u ) = det ( 1 − u F ∣ H c 1 ( X k , E St )) .
Declaration of generative AI assistance
OpenAI ChatGPT (GPT-6) and Codex were used to assist with mathematical exploration, reviewing arguments, checking calculations and references, preparing and running finite-field verification programs, and improving the exposition and language of this manuscript. The authors independently reviewed the AI-assisted material and take full responsibility for the correctness of the mathematical statements, proofs, computational results, references, and final manuscript.
References
(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: hayata@keio.jp
(Takahiro Tsushima) Keio University School of Medicine, 4-1-1 Hiyoshi, Kohoku-ku, Yokohama, 223-8521, Japan
Email address: tsushima@keio.jp