ebook img

Twisted Exponential Sums PDF

0.38 MB·
by  Lei Fu
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview Twisted Exponential Sums

Twisted Exponential Sums Lei Fu 7 0 Chern Instituteof Mathematics and LPMC, NankaiUniversity,Tianjin 300071, P. R.China 0 [email protected] 2 n a J Abstract 0 3 ∗ Letkbeafinitefieldofcharacteristicp,laprimenumberdistincttop,ψ :k →Ql anontrivial ] T additive character, and χ:k∗n →Q∗ a character on k∗n. Then ψ defines an Artin-Schreier sheaf l N L on the affine line A1, and χ defines a Kummer sheaf K on the n-dimensional torus Tn. Let . ψ k χ k h t f ∈ k[X ,X−1,...,X ,X−1] be a Laurent polynomial. It defines a k-morphism f : Tn → A1. a 1 1 n n k k m In this paper, we calculate the dimensions and weights of Hi(Tn,K ⊗f∗L ) under some non- c k¯ χ ψ [ degeneracy conditions on f. Our results can be used to estimate sums of the form 2 v 4 χ (f (x ,...,x ))···χ (f (x ,...,x ))ψ(f(x ,...,x )), 1 1 1 n m m 1 n 1 n 6 1 x1,..X.,xn∈k∗ 7 where χ ,...,χ : k∗ → C∗ are multiplicative characters, ψ : k → C∗ is a nontrivial additive 0 1 m 6 character, and f ,...,f ,f are Laurent polynomials. 0 1 m / h t Key words: Toric scheme, perverse sheaf, weight. a m v: Mathematics Subject Classification: 14G15, 14F20, 11L40. i X r a 0. Introduction Letk beafinitefieldwithq elementsofcharacteristicp,letχ ,...,χ :k∗ →C∗ benontrivial 1 m multiplicative characters,let ψ :k →C∗ be a nontrivial additive character, and let f (X ,...,X ),...,f (X ,...,X ),f(X ,...,X )∈k[X ,X−1,...,X ,X−1] 1 1 n m 1 n 1 n 1 1 n n 1 beLaurentpolynomials. Wemaketheconventionthatχ (0)=0(i=1,...,m). Innumbertheory, i we are often lead to study the sum S = χ (f (x ,...,x ))···χ (f (x ,...,x ))ψ(f(x ,...,x )). 1 1 1 1 n m m 1 n 1 n x1,..X.,xn∈k∗ For this purpose, let’s consider another sum S = χ−1(x )···χ−1(x ) 2 1 n+1 m n+m x1,...,Xxn+m∈k∗ ψ(f(x ,...,x )+x f (x ,...,x )+···+x f (x ,...,x )). 1 n n+1 1 1 n n+m m 1 n We have S 2 = χ−1(x )ψ(x f (x ,...,x )) ··· χ−1(x )ψ(x f (x ,...,x )) 1 n+1 n+1 1 1 n m n+m n+m m 1 n x1,..X.,xn∈k∗xn+1...X,xn+m∈k∗(cid:0) (cid:1) (cid:0) (cid:1) ψ(f(x ,...,x )) 1 n = χ−1(x )ψ(x f (x ,...,x )) ··· χ−1(x )ψ(x f (x ,...,x ))  1 n+1 n+1 1 1 n   m n+m n+m m 1 n  x1,..X.,xn∈k∗ xnX+1∈k∗ xn+Xm∈k∗  ψ(f(x ,...,x )).    1 n For i=1,...,m and x ,...,x ∈k∗, if f (x ,...,x )=0, we have 1 n i 1 n χ−1(x )ψ(x f (x ,...,x ))=0; i n+i n+i i 1 n xnX+i∈k∗ if f (x ,...,x )6=0, we have i 1 n x χ−1(x )ψ(x f (x ,...,x )) = χ−1 ψ(x) i n+i n+i i 1 n i f (x ,...,x ) xnX+i∈k∗ xX∈k∗ (cid:18) i 1 n (cid:19) = χ (f (x ,...,x ))G(χ ,ψ), i i 1 n i where G(χ ,ψ)= χ−1(x)ψ(x) i i x∈k∗ X is the Gauss sum. So in any case, we have χ−1(x )ψ(x f (x ,...,x ))=χ (f (x ,...,x ))G(χ ,ψ). i n+i n+i i 1 n i i 1 n i xnX+i∈k∗ 2 Hence S = χ (f (x ,...,x ))G(χ ,ψ)···χ (f (x ,...,x ))G(χ ,ψ)ψ(f(x ,...,x )) 2 1 1 1 n 1 m m 1 n m 1 n x1,..X.,xn∈k∗ = G(χ ,ψ)···G(χ ,ψ)S . 1 m 1 As the Gauss sums are well-understood, the study of S is reduced to the study of S . 1 2 In this paper, we use l-adic cohomology theory to study sums of the form χ (x )···χ (x )ψ(f(x ,...,x )), 1 1 n n 1 n xXi∈k∗ where χ ,...,χ are multiplicative characters (nontrivial or trivial). Note that S is of this form. 1 n 2 Our results complete those in [DL], where the caseof trivialχ is treated. We follow the approach i initiated by Denef and Loeser. We first associate geometric objects to the above data. The Kummer covering [q−1]:Tn →Tn, x7→xq−1 k k on the torus Tn =Speck[X ,X−1,...,X ,X−1] defines a Tn(k)-torsor k 1 1 n n k 1→Tn(k)→Tn [q→−1]Tn →1, k k k where Tn(k) = Hom (Speck,Tn) is the group of k-rational points in Tn. Fix a prime number l k k k k distinct to p. Let χ : Tn(k) = k∗n → Q∗ be a character. Pushing-forward the above torsor by k l χ−1, we get a lisse Q -sheaf K on Tn of rank 1. We call K the Kummer sheaf associated to χ. l χ k χ For any rational point x in Tn(k′) = Hom (Speck′,Tn) with value in a finite extension k′ of k, k k k we have Tr(Fx,(Kχ)x¯)=χ(Normk′/k(x)), where F is the geometric Frobenius element at x. x The Artin-Schreier covering P :A1 →A1, x7→xq −x k k defines an A1(k)-torsor k 0→A1(k)→A1 →P A1 →0, k k k 3 whereA1(k)=Hom (Speck,A1)isthegroupofk-rationalpointsinA1. Letψ :A1(k)=k →Q∗ k k k k k l beanadditivecharacter. Pushing-forwardthistorsorbyψ−1,wegetalisseQ -sheafL ofrank1on l ψ A1,whichwecalltheArtin-Schreiersheaf. ForanyrationalpointxinA1(k′)=Hom (Speck′,A1) k k k k with value in a finite extension k′ of k, we have Tr(Fx,(Lψ)x¯)=ψ(Trk′/k(x)), where F is the geometric Frobenius element at x. x Let f = a Xi ∈k[X ,X−1,...,X ,X−1] i 1 1 n n i∈Zn X be a Laurent polynomial. The Newton polyhedron ∆ (f) of f at ∞ is the convex hull in Rn of ∞ the set {i∈Zn|a 6=0}∪{0}.We say f is non-degenerate with respect to ∆ (f) if for any face τ i ∞ of ∆ (f) not containing 0, the locus of ∞ ∂f ∂f τ τ =···= =0 ∂X ∂X 1 n in Tn is empty, where k f = a Xi. τ i i∈τ X This is equivalent to saying that the morphism f :Tn =SpecA[X ,X−1,...,X ,X−1]→A1 =Speck[T] τ k 1 1 n n k defined by the k-algebra homomorphism k[T]→k[X ,X−1,...,X ,X−1], T 7→f 1 1 n n τ is smooth. The first main result of this paper is the following theorem. Theorem 0.1. Let f : Tn → A1 be a k-morphism defined by a Laurent polynomial f ∈ k k k[X ,X−1,...,X ,X−1] that is non-degenerate with respect to ∆ (f) and let K be a Kum- 1 1 n n ∞ χ mer sheaf on Tn. Suppose dim(∆ (f))=n. Then k ∞ (i) Hi(Tn,K ⊗f∗L )=0 for i6=n. c k¯ χ ψ (ii) dim(Hn(Tn,K ⊗f∗L ))=n!vol(∆ (f)). c k¯ χ ψ ∞ 4 (iii) If 0 is an interior point of ∆ (f), then Hn(Tn,K ⊗f∗L ) is pure of weight n. ∞ c k¯ χ ψ Herethe conclusionof(iii) meansthatforanyeigenvalueλofthe geometricFrobenius element F inGal(k¯/k)actingonHn(Tn,K ⊗f∗L ),λisanalgebraicnumber,andallthegaloisconjugates c k¯ χ ψ of λ have archimedean absolute value qn2. Note that we have [q−1] [q−1]∗f∗L ∼= (K ⊗f∗L ), ∗ ψ χ ψ χ M where [q−1]:Tn →Tn is the Kummer covering,and in the direct sum on the right-hand side, χ k k goes over the set of all characters χ : Tn(k) → Q . The composition f ◦[q−1] is defined by the k l Laurent polynomial f′(X ,...,X )=f(Xq−1,...,Xq−1). 1 n 1 n Note that f′ is also non-degenerate with respect to its Newton polyhedron at ∞. We have Hi(Tn,f′∗L ) ∼= Hi(Tn,[q−1] [q−1]∗f∗L ) c k¯ ψ c k¯ ∗ ψ ∼= Hi(Tn,K ⊗f∗L ). c k¯ χ ψ χ M So Hi(Tn,K ⊗f∗L ) aredirectfactorsof Hi(Tn,f′∗L ). Hence Theorem0.1(i) and(iii) follow c k¯ χ ψ c k¯ ψ directly from the main theorem 1.3 in [DL] applied to f′. Using [I1] 2.1, one can show χ (Tn,K ⊗f∗L )=χ (Tn,f∗L ), c k¯ χ ψ c k¯ ψ where χ denotes the Euler characteristic for the cohomology with compact support. Hence The- c orem 0.1 (ii) can also be deduced from [DL] 1.3. In this paper, we give a proof of Theorem 0.1 independent of the main theorem of [DL]. On the other hand, oursecondmain Theorem0.4on the weightsof Hn(Tn,K ⊗f∗L ) doesn’t seem c k¯ χ ψ to follow from the corresponding theorem in [DL]. Corollary 0.2. Letf ∈k[X ,X−1,...,X ,X−1]be aLaurentpolynomialthatis non-degenerate 1 1 n n with respect to ∆ (f) and suppose dim(∆ (f)) = n. Then for any multiplicative characters ∞ ∞ χ ,...,χ :k∗ →C∗ and any nontrivial additive character ψ :k →C∗, we have 1 n n | χ1(x1)···χn(xn)ψ(f(x1,...,xn))|≤n!vol(∆∞(f))q2. xXi∈k∗ 5 Proof. Note that the values of χ and ψ are algebraic integers. In particular, we may consider i them to have values in Q . Let χ:k∗n →Q∗ be the character defined by l l χ(x)=χ (x )...χ (x ) 1 1 n n for any x=(x ,...,x )∈k∗n. We then have 1 n χ (x )···χ (x )ψ(f(x ,...,x ))= Tr(F ,(K ⊗f∗L ) ). 1 1 n n 1 n x χ ψ x¯ xXi∈k∗ x∈XTnk(k) By the Grothendieck trace formula ([SGA 41], [Rapport] Th´eor`eme 3.2), we have 2 2n Tr(F ,(K ⊗f∗L ) )= (−1)iTr(F,Hi(Tn,K ⊗f∗L )). x χ ψ x¯ c k¯ χ ψ x∈XTnk(k) Xi=0 By [D] 3.3.1, for any eigenvalue λ of F acting on Hn(Tn,K ⊗f∗L ), λ is an algebraic number c k¯ χ ψ and all its galois conjugates have archimedean absolute value ≤qn2. Combined with Theorem 0.1 (i) and (ii), we get 2n | (−1)iTr(F,Hci(Tnk¯,Kχ⊗f∗Lψ))|≤n!vol(∆∞(f))qn2. i=0 X So we have n | χ1(x1)···χn(xn)ψ(f(x1,...,xn))|≤n!vol(∆∞(f))q2. xXi∈k∗ Combining Corollary 0.2 with the discussion at the beginning, and using the fact that Gauss sums have absolute value q21, we get the following. Corollary0.3. Letf,f ,...,f ∈k[X ,X−1,...,X ,X−1]beLaurentpolynomials,χ ,...,χ : 1 m 1 1 n n 1 m k∗ → C∗ nontrivial multiplicative characters, and ψ : k → C∗ a nontrivial additive character. Suppose the Laurent polynomial F(x ,...,x ,x ,...,x )=f(x ,...,x )+x f (x ,...,x )+...+x f (x ,...,x ) 1 n n+1 n+m 1 n n+1 1 1 n n+m m 1 n is non-degenerate with respect to ∆ (F), and dim(∆ (F))=m+n. Then we have ∞ ∞ | χ (f (x ,...,x ))···χ (f (x ,...,x ))ψ(f(x ,...,x ))|≤(n+m)!vol(∆ (F))qn/2. 1 1 1 n m m 1 n 1 n ∞ xXi∈k∗ 6 Under the assumption of Theorem 0.1, let E(Tn,f,χ)= e Tw, k w w∈Z X where e is the number of eigenvalues counted with multiplicities of the geometric Frobenius w elementF inGal(k¯/k)actingonHn(Tn,K ⊗f∗L )withweightw. Ournextgoalistodetermine c k¯ χ ψ E(Tn,f,χ). k Foranyconvexpolyhedralconeδ inRn with0beingaface,definetheconvexpolytopepoly(δ) to be the intersection of δ with a hyperplane in Rn which does not contain 0 and intersects each one dimensional face of δ. Note that poly(δ) is defined only up to combinatorial equivalence. For any convex polytope ∆ in Rn and any face τ of ∆, define cone (τ) to be the cone generated by ∆ u′−u (u′ ∈∆, u∈τ), and define cone◦(τ) to be the image of cone (τ) in Rn/span(τ −τ). Note ∆ ∆ that 0 is a face of cone◦ (τ). We define polynomials α(δ) and β(∆) in one variable T inductively ∆ by the following formulas: α({0}) = 1, β(∆) = (T2−1)dim(∆)+ (T2−1)dim(τ)α(cone◦(τ)), ∆ τ faceXof ∆,τ6=∆ α(δ) = trunc ((1−T2)β(poly(δ))), ≤dim(δ)−1 where trunc (·) denotes taking the degree ≤d part of a polynomial. These polynomials are first ≤d introducedbyStanley[S].Notethatα(δ)andβ(∆)onlyinvolveevenpowersofT,andtheydepend only on the combinatorial types of δ and ∆. If δ is a simplicial cone, that is, if δ is generated by linearly independent vectors, then using induction on dim(δ), one can verify α(δ)=1. Let χ : Tn(k) → Q∗ be a character. For a rational convex polytope ∆ in Rn of dimension n, k l let T be the set of faces τ of ∆ so that τ 6=∆, 0∈τ, and K ∼=p∗K for a Kummer sheaf K on χ τ τ τ the torus T =Speck[Zn∩span(τ −τ)], where τ p :Tn =Speck[Zn]→T =Speck[Zn∩span(τ −τ)] τ k τ is the morphism defined by the canonical homomorphism k[Zn∩span(τ −τ)]֒→k[Zn]. 7 Define e(∆,χ)=n!vol(∆)+ (−1)n−dim(τ)(dim(τ))!vol(τ)α(cone◦ (τ))(1) ∆ τ∈T X and define a polynomial E(∆,χ) inductively by E(∆,χ)=e(∆,χ)Tn− (−1)n−dim(τ)E(τ,χ )α(cone◦ (τ)). τ ∆ τ∈T X Our second main result is the following. Theorem 0.4. Let f : Tn → A1 be a k-morphism defined by a Laurent polynomial f ∈ k k k[X ,X−1,...,X ,X−1] that is non-degenerate with respect to ∆ (f) and let K be a Kum- 1 1 n n ∞ χ mer sheaf on Tn. Suppose dim(∆ (f))=n. Let k ∞ E(Tn,f,χ)= e Tw, k w w∈Z X where e is the number of eigenvalues counted with multiplicities of the geometric Frobenius w element F in Gal(k¯/k) acting on Hn(Tn,K ⊗ f∗L ) with weight w. Then E(Tn,f,χ) is a c k¯ χ ψ k polynomial of degree ≤n, and E(Tn,f,χ) = E(∆ (f),χ), k ∞ e = e(∆ (f),χ). n ∞ Corollary 0.5. Let f : Tn → A1 be a k-morphism defined by a Laurent polynomial f ∈ k k k[X ,X−1,...,X ,X−1] that is non-degenerate with respect to ∆ (f) and let K be a Kummer 1 1 n n ∞ χ sheaf on Tn. Suppose dim(∆ (f))=n and suppose for any face τ of ∆ (f) of codimension one k ∞ ∞ containing 0, K is not the inverse image of any Kummer sheaf on T =Speck[Zn∩span(τ −τ)] χ τ under the morphism p :Tn =Speck[Zn]→T =Speck[Zn∩span(τ −τ)]. τ k τ Then Hn(Tn,K ⊗f∗L ) is pure of weight n. c k¯ χ ψ Remark 0.6. Note that this corollary together with Theorem 0.1 is Theorem 4.2 in [AS], except that Adolphson and Sperber prove the theorem for almost all p. 8 ProofofCorollary0.5. Underourassumption,thesetT offacesτ of∆ (f)sothatτ 6=∆ (f), ∞ ∞ 0 ∈ τ, and K ∼= p∗K for a Kummer sheaf K on the torus T = Speck[Zn ∩span(τ −τ)] is χ τ τ τ τ empty. Therefore E(∆ (f),χ)=e(∆ (f),χ)Tn. ∞ ∞ By Theorem 0.4, this implies E(Tn,f,χ)=e(∆ (f),χ)Tn, k ∞ and hence Hn(Tn,K ⊗f∗L ) is pure of weight n. c k¯ χ ψ For a rational convex polytope ∆ in Rn with dimension n, recall that T is the set of faces τ of ∆ so that τ 6= ∆, 0 ∈ τ, and K ∼= p∗K for a Kummer sheaf K on the torus T = χ τ τ τ τ Speck[Zn∩span(τ −τ)]. Define V (∆,χ)= vol(τ) i τ∈T,Xdim(τ)=i for 0≤i≤n−1 and define V (∆,χ)=vol(∆). n Let τ be the smallest face of ∆ containing 0. We say ∆ is simplicial at the origin if there are 0 exactly n−dim(τ ) faces of ∆ that have dimension n−1 and contain τ . If ∆ is simplicial at 0 0 the origin and τ is a face containing 0, then the number of faces of ∆ that have dimension k and n−dim(τ) contain τ is . n−k (cid:18) (cid:19) The following corollary is Theorem 4.8 in [AS]. Corollary 0.7. Let f : Tn → A1 be a k-morphism defined by a Laurent polynomial f ∈ k k k[X ,X−1,...,X ,X−1] that is non-degenerate with respect to ∆ (f) and let K be a Kummer 1 1 n n ∞ χ sheaf on Tn. Suppose dim(∆ (f)) = n and suppose ∆ (f) is simplicial at the origin. Let e k ∞ ∞ w be the number of eigenvalues counted with multiplicities of the geometric Frobenius element F in Gal(k¯/k) acting on Hn(Tn,K ⊗f∗L ) with weight w. Then we have c k¯ χ ψ w n−i e = (−1)w−ii! V (∆ (f),χ). w n−w i ∞ i=0 (cid:18) (cid:19) X Proof. The hypothesis that ∆ (f) is simplicial at the origin implies that ∞ α(cone◦ (τ))=1 ∆∞(f) 9 for any face τ of ∆ (f) containing 0. By Theorem 0.4, we have ∞ e = e(∆ (f),χ) n ∞ = n!vol(∆ (f))+ (−1)n−dim(τ)(dim(τ))!vol(τ) ∞ τ∈T X n−1 = n!V (∆ (f),χ)+ (−1)n−ii!V (∆ (f),χ) n ∞ i ∞ i=0 X n = (−1)n−ii!V (∆ (f),χ). i ∞ i=0 X This proves our assertion for w=n. We use induction on n. Under our assumption, we have E(∆ (f),χ)=e(∆ (f),χ)Tn− (−1)n−dim(τ)E(τ,χ ). ∞ ∞ τ τ∈T X For w ≤ n−1, taking the coefficients of Tw on both sides of the above equality and applying Theorem 0.4 and the induction hypothesis to the pairs (τ,χ ) (τ ∈T), we get τ w dim(τ)−i e =− (−1)n−dim(τ) (−1)w−ii! vol(τ′). w dim(τ)−w τ∈T,w≤dim(τ)≤n−1 i=0 (cid:18) (cid:19)τ′≺τ,dim(τ′)=i,τ′∈T X X X So we have w dim(τ)−i e = (−1)w−i+n+1i! (−1)dim(τ) vol(τ′). w dim(τ)−w Xi=0 τ′∈T,Xdim(τ′)=i τ′≺τ,w≤Xdim(τ)≤n−1 (cid:18) (cid:19) By our assumption, for each τ′ containing 0 of dimension i, the number of faces of ∆ (f) that ∞ n−i have dimension k and contain τ′ is . So we have n−k (cid:18) (cid:19) w n−1 k−i n−i e = (−1)w−i+n+1i! (−1)k vol(τ′). w k−w n−k Xi=0 τ′∈T,Xdim(τ′)=ikX=w (cid:18) (cid:19)(cid:18) (cid:19) We have n−1 k−i n−i (−1)k k−w n−k k=w (cid:18) (cid:19)(cid:18) (cid:19) X n−1 n−w n−i = (−1)k k−w n−w k=w (cid:18) (cid:19)(cid:18) (cid:19) X n−1−w n−i n−w = (−1)w (−1)j n−w j (cid:18) (cid:19) j=0 (cid:18) (cid:19) X n−i = (−1)n−1 . n−w (cid:18) (cid:19) 10

See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.