scieee AI-readable full text Open interactive document viewer

Galois representations of orthogonal rigid local systems

Schulte, Michael

Full text

Galois representations of orthogonal rigid lo al systems Von der Universität Bayreuth zur Erlangung des Grades eines Doktors der Naturwissenshaften (Dr. rer. nat) genehmigte Abhandlung von Mihael Shulte (geb. Maier) aus Karlsruhe 1. Gutahter: Prof. Dr. Mihael Dettweiler (Universität Bayreuth) 2. Gutahter: Prof. Dr. Stefan Wewers (Universität Ulm) Tag der Einreihung: 17. April 2012 Tag des Kolloquiums: 10. July 2012 Contents 1 Intro dution 5 1.1 Einleitung ......................................... 10 1.2 Aknowledgement ..................................... 15 1.3 Eidesstattlihe Erklärungen ............................... 16 1.4 Notation .......................................... 17 2 Preliminary Results 19 2.1 Galois Representations .................................. 19 2.2 Étale Fundamental Group Funtor π´et 1 ......................... 23 2.3 Weil Conjeture ...................................... 31 2.4 Crystalline Representations ............................... 33 2.5 Semi-Simpliation .................................... 39 3 Intro dution to MCχ 41 3.1 The Middle Convolution ................................. 41 3.2 The Numerology of MCχ ................................ 44 3.3 Constrution of Hm,ℓ ................................... 47 4 Ho dge Strutures and Middle Convolution 51 4.1 Ho dge Strutures ..................................... 51 4.2 Variations of Hodge Struture .............................. 52 4.3 Extensions of Variations of Ho dge Struture . . . . . . . . . . . . . . . . . . . . . . 54 5 Motivi Desription of Hm,ℓ 57 5.1 Setting ........................................... 57 5.2 The Motivi Interpretation of MCχ ........................... 59 5.3 Appliation to Hm,ℓ ................................... 60 5.4 Analytiation of Hm,ℓ .................................. 65 6 Irreduibility of ρm 67 6.1 Lifting Irreduibility ................................... 67 6.2 Serre's Results on Charaters of GQ .......................... 68 6.3 Groups of Lie Type .................................... 69 6.4 Irreduibility of the modℓ Representation ....................... 73 7 Potential Automorphy of Sp eializations 79 7.1 Langlands Corresp ondene ................................ 79 7.2 Mo dular Lifting ...................................... 80 Bibliography 85 3 1 Intro dution For over 5000 years the rational numb ers have b een known and have been a ornerstone of mathematis sine then. The onsideration of p olynomials and the symmetries of their roots by the group of automorphisms was initiated by Galois and marks the starting p oint of mo dern numb er theory. The Galois group of a p olynomial eno des muh far-reahing information. But what ab out the absolute Galois group GQ , whih is the group of automorphisms of the algebrai losure of Q ? Its struture is still a mystery and far from b eing well-understo od. In order to get partial answers, there are various dierent approahes. First we have a lo ok at the fator groups of the absolute Galois group. Esp eially there is the unsolved question, if every nite group is of this type, known as the inverse Galois problem (see [MM99℄). A seond promising approah are Galois representations, i.e. ontinuous homomorphisms of GQ to matrix groups GLn(Qℓ) . Often it is p ossible to onstrut these in a similar way for all ℓ , whih yields under some irumstanes weakly ompatible systems. These are families of ℓ -adi Galois representations for eah prime number ℓ . They have the prop erty that a Frob enius morphism is mapp ed in suh a way that its harateristi p olynomial is rational and indep endent of ℓ for almost all representations. It is p ossible to desribe many arithmeti ob jets by weakly ompatible systems, for example Galois representations on the Tate mo dules of ellipti urves, whih are ruial to the pro of of Fermat's onjeture, f. Theorem of Wiles [Wil95℄. The ℓ -adi Galois representations have interesting onnetions in two diretions. First we an assign to an irreduible, weakly ompatible system of ℓ -adi Galois representations an analyti funtion, known as L -funtion. If the system is automorphi its L -funtion is equal to an analyti L -funtion. Therefore prop erties like meromorphi ontinuation to the whole omplex plane and the funtional equation an b e transferred. The other way around the Langlands orresp ondene is onjetured (f. [BBG + 03℄), whih says that eah L -funtion asso iated to a weakly ompatible system is obtained by suh an automorphi representation. This is still mostly unproven. In [BLGGT10℄ Barnet-Lamb, Gee, Geraghty and Taylor reently provided some interesting to ols for pro ofs in this diretion. The other onnetion is to geometry. In order to ontrol the Galois group, one often ho oses families of Galois representations given by lisse étale sheaves (see Corollary 2.2.12). In suh families it is p ossible to give far reahing information of the absolute Galois group on the stalks by top ologial means, the so alled mono dromy. A very important ase are the motivi families of Galois representations, whih desrib e variations of ℓ -adi ohomology groups for variable ℓ . 5 1. Introdution An example is the variation of H1 ´et of the Legendre family of ellipti urves Eλ given by Y2=X(X−1)(X−λ) on A1 Q\{0,1} . More general for ellipti urves E the orresp onding Galois representation on the rst ohomology ρℓ:GQ−→ GL(H1 ´et(E,Zℓ)) ∼ =GL2(Zℓ)//GL2(Qℓ) is dual to the Galois representation on the Tate mo dule Tℓ(E) . Serre's op en image Theorem gives information on the size of the image of the representation. Théorème ( [Ser72℄, (7) ): If E is an el lipti urve without omplex multipliation dened over a number eld K and ρℓ:GK−→ GL2(Qℓ) the orresponding representation, then for almost al l prime numbers ℓ we have im (ρℓ) = GL2(Zℓ). Let λ b e a geometri p oint whih is dened over Q suh that Eλ has no omplex multipliation. Then the sp eialization H1 ´et(Eλ,Qℓ) is a GQ -mo dule on whih the group GQ ats maximally for almost all ℓ . The seond imp ortant prop erty is that we get a weakly ompatible system of representations (ρℓ)ℓ prime whih is automorphi in the sense of the Langlands program. This was shown in 2001 by Breuil, Conrad, Diamond, Taylor [BCDT01℄ and Wiles [Wil95℄ in the pro of of the Taniyama-Shimura-Weil onjeture, whih is now known as mo dularity theorem, and is true for all ellipti urves. The ambition of this work is to prove similar results for higher dimensional generalizations of the Legendre family. The key observation is that the mono dromy of the Legendre family is given by a rigid loal system whih is the solution of a Piard-Fuhs equation, a sp eial hypergeometri dierential equation: λ(1 −λ)f′′ + (1 −2λ)f′−1 4f= 0. In this ontext, rigid means that the lo al system admits no deformations, whih is equivalent to rig(F) = 2 if the system is irreduible (see Theorem 3.1.5). In general it is p ossible to desrib e rigid lo al systems by Katz' theory of middle onvolution MCχ (Chapters 5,6 of [Kat96℄). Let χ:π´et 1(Gm,K)−→ Qℓ × b e a geometrially non-trivial one dimensional representation and F a lisse étale sheaf on A1 K\S , where S⊆A1 K is nite. Then one obtains a lisse étale sheaf MCχ(F) on A1 K\S . If F is irreduible and rigid, then MCχ(F) is irreduible and rigid as well but usually rk(MCχ(F)) 6= rk(F) . 6 In summary we get the following onstrutive statement, whih is known as Katz algorithm: Theorem ([Kat96℄): Let F be an irreduible rigid loal system of Qℓ -modules on A1 K\S . There exist n∈N0 , loal systems L0,...,Ln of Qℓ -modules of rank one on A1 K\S and representations χ1,...,χn:π´et 1(Gm,K)−→ Qℓ × , suh that F=Ln⊗MCχn(... L2⊗MCχ2(L1⊗MCχ1(L0) ) ... ). For S={0,1} if L0 has monodromy tuple (−1,−1,1) (f. Chapter 2.2) and χ=−1 , the unique quadrati harater, we get the Legendre family MC−1(L0) . This is the starting point of an innite family (Hm,ℓ)m∈N0 of lo al systems of Qℓ -mo duls on A1 K with i∗Hm,ℓ =Ln⊗MC−1(... L2⊗MC−1(L1⊗MC−1(L0) ) ... ), where we have i:A1 K\ {0,1}//A1 K and Lj with mono dromy tuple (1,−1,−1) for j o dd resp etively Lj with (−1,1,−1) for j6= 0 even. This onstrution yields the following result: Theorem (f. 3.3.1) Let ℓ be a prime number and K an algebraial ly losed eld with har (K)∤2ℓ . Then, for any m∈N0 there exists a ohomologial ly rigid Hm,ℓ ∈Tℓ(K) of generi rank m+1 , a Qℓ -sheaf on A1 K whih is lisse on i:A1 K\{0,1}//A1 K . If m is even, then Hm,ℓ has orthogonal monodromy, and if m is odd, Hm,ℓ has sympleti monodromy, i.e. there is an orthogonal respetively sympleti pairing Hm,ℓ ×Hm,ℓ −→ Qℓ. The monodromy tuple of i∗Hm,ℓ has the fol lowing Jordan normal form: at 0 : J1(1)m 2⊕J1(−1)m 2+1 for 2 |m, J2(1)m+1 2for 2 ∤m, at 1 : J2(1)m 2⊕J1(−1) for m≡0 mod 4, J1(1)m−1 2⊕J2(−1) ⊕J1(−1)m−1 2for m≡1 mod 4, J3(1) ⊕J2(1)m 2−1for m≡2 mod 4, J2(1) ⊕J1(1)m−3 2⊕J1(−1)m+1 2for m≡3 mod 4, at ∞ : Jm+1(1). Here Jn(λ) denotes the upp er triangular Jordan blo k of length n and eigenvalue λ . 7 1. Introdution We want to note that the ase rk(H6,ℓ) = 7 is of sp eial interest. Its examination answered a question of Serre on the existene of motivi Galois groups of type G2 (f. [Ser94℄). By this it was possible to onstrut suh motives (f. Dettweiler, Katz and Reiter [DR10℄). The motivi desription of rigid lo al systems by Katz yields lisse étale sheaves Hm,ℓ , whose analytiations ome from variations of Ho dge struture. Theorem (f. 5.4.3) Let Hm,ℓ be as in Theorem 3.3.1 . Then there exists a loal system of Z -modules Gm on A1 C\{0,1} underlying a polarized variation of Z -Hodge struture (Gm,F•,∇) on C\{0,1} pure of weight m suh that (i∗Hm,ℓ)an ∼ =Gm⊗Qℓ. The indued isomorphism on the stalks (i∗Hm,ℓ)an x∼ =Gm⊗Qℓx for x∈C\{0,1} is given by the omparison isomorphism between étale ohomology and singular ohomology: 1 2(1 −σ)ker Hm ´et (Xx,Qℓ)→Hm ´et (Dx,Qℓ)∼ =1 2(1 −σ)ker (Hm B(X(C)x,Z)→Hm B(D(C)x,Z)) ⊗Qℓ. Moreover, the Hodge ltration of Gm has maximal length. In this way we get, as in the Legendre ase, families of weakly ompatible systems of Galois representations ρm of Q by three transformations, namely by tensoring it with its determinant, sp eializing like in Theorem 6.4.1 and nally semi-simpliation: ρm= (ρm,ℓ)ℓ prime := ((ρi∗Hm,ℓ ⊗det(ρi∗Hm,ℓ )) ◦ιx)ss ℓ prime . Here ιx:GK//π´et 1(A1 K\{0,1}) denotes the sp eialization map for a xed x∈A1 K\{0,1} , whih omes from the morphism {x}//A1 K\{0,1} (f. Setion 2.2). These representations ρm have the property that they fator over Zℓ and over a speial orthogonal group for even m or a sympleti group for o dd m . If m is even, by tensoring the initial system of representations with the ylotomi harater χℓ to the p ower m 2 we get systems of weight 0 representations. The redution mo d ℓ is dened as dρm,ℓ := ρi∗Hm,ℓ ⊗det(ρi∗Hm,ℓ )⊗χm 2 ℓ:π1 ´et(A1 Q\{0,1})−→ SOm+1(Fℓ). Theorem (f. 6.4.1) Let x∈A1 Q\ {0,1} , suh that there exist odd prime numbers p, q 6=ℓ satisfying νp(x)<0 but ℓ∤νp(x) and νq(x−1) >0 but ℓ∤νq(x−1) . Then the fol lowing holds: If m∈N0 even and m≥12 then Ωm+1(Fℓ)⊆im ( dρm,ℓ ◦ιx) for almost al l prime numbers ℓ , where dρm,ℓ ◦ιx:GQ−→ SOm+1(Fℓ) is the speialization at x . If m= 6 then for almost al l ℓ , we have im ( dρm,ℓ ◦ιx) = G2(Fℓ) 8 Therefore it is possible to sp eialize in suh a way, that we obtain an irreduible representation. Finally Theorem 7.2.2 and the motivi desription in Chapter 5give the automorphy over a numb er eld. Theorem ([BLGGT10℄, Thm.5.3.1): Suppose that K is a CM (or total ly real) eld and that (ρℓ)ℓ prime is an irreduible, total ly odd, essential ly onjugate self-dual, regular, weakly ompatible system of ℓ -adi representations of K . Then there is a nite, CM (or total ly real), Galois extension L|K suh that the restrition of (ρℓ)ℓ prime to GL is automorphi. The motivi desription of rigid lo al systems by Katz shows that these systems of representations are rystalline for almost all prime numb ers ℓ . Under the assumptions of Theorem 6.4.1 using the work of Barnet-Lamb, Gee, Geraghty and Taylor, we obtain the following result. A weaker statement was proved in [GMHK10℄. Theorem (f. 7.2.4) For m= 6 or m∈N0 even, m≥12 and K=Q the irreduible, weakly ompatible system ρm= (ρm,ℓ)ℓ prime of Galois representations is potential ly automorphi. 9 1. Introdution 1.3 Eidesstattlihe Erklärungen Hiermit versihere ih, dass ih die hier vorliegende zur Promotion eingereihte Arb eit mit dem Titel  Galois representations of orthogonal rigid loal systems selbstständig verfasst, nur die angegeb enen Quellen und Hilfsmittel b enutzt und wörtlih o der inhaltlih üb ernommene Stellen als solhe gekennzeihnet hab e. Ih versihere an Eides statt, dass diese Angaben wahr sind und dass ih nihts vershwiegen hab e. Mir ist b ekannt, dass die falshe Abgabe einer Versiherung an Eides statt mit Freiheitsstrafe bis zu drei Jahren o der mit Geldstrafe b estraft wird. Bayreuth, den 16. April 2012 Hiermit erkläre ih, dass ih bisher keine Promotionsversuhe mit dieser o der einer anderen Dissertation unternommen hab e. Die Arb eit wurde bisher weder im Inno h Ausland in gleiher o der ähnliher Form einer anderen Prüfungsb ehörde vorgelegt. Bayreuth, den 16. April 2012 Hiermit b estätige ih, dass ih keinerlei Hilfe von gewerblihen Promotionsb eratern bzw. -vermittlern oder ähnlihen Dienstleistern in Anspruh genommen hab e, no h künftig in Anspruh nehmen werde. Bayreuth, den 16. April 2012 16 1.4. Notation 1.4 Notation N={1,2,...} natural numb ers N0={0,1,2, . . .} natural numb ers with zero K, L elds K algebrai losure of K Ksep separable losure of K in K GK= Gal(Ksep/K) absolute Galois group of K ΣK set of nite plaes of K Kv ompletion of K at v kv residue eld of K at v Cv=ˆ Kv the ompletion of the algebrai losure of Kv Iw inertia group at w∈ΣL\{0} for a eld Extension L/K Itame K:= πtame 1(Gm,K) tame inertia group as in Denition 2.2.4 Knr v maximal unramied extension AK,IK adele ring of K and idele group of K ℓ prime numb er χ one dimensional ℓ -adi Galois representation χℓ ylotomi harater 1,−1 trivial and quadrati rank one representation Lχ Kummer sheaf asso iated to χ L middle extension sheaf on A1 K MCχ middle onvolution funtor as in Denition 3.1.2 MTL middle tensor pro dut as in Denition 3.1.6 Tℓ(K) ategory of sp eial Qℓ -sheaves as in Denition 3.1.3 i:U//A1 K inlusion of an op en dense subset of the ane line R ommutative ring with 1 M(m×n, R)m times n matries over the ring R GL(V) group of invertible endomorphisms of the vetor spae V An K,Ga,K,Gm,K,GLn(K),SLn(K) algebrai groups O n(K) orthogonal group SOn(K) sp eial orthogonal group Ωn(K) derived group of SOn(K) Spn(K) sympleti group G2(K) a sp oradi group Jn(λ) upp er triangular Jordan blo k of length n and eigenvalue λ ιx sp eialization map to x (f. Setion 2.2) Hm,ℓ sp eial Qℓ -sheaf onstruted in Setion 3.3 ρm,ℓ :GK−→ GLm+1(Qℓ)ℓ -adi Galois representation onstruted in Setion 7.2 dρm,ℓ :π´et 1(A1 Q\{0,1})−→ SOm+1(Fℓ) weight 0 representation of π´et 1(A1 Q\{0,1}) (see Setion 6.4) 17 2 Preliminary Results In this hapter, we want to x the notation used in this work. In addition we give an overview of onepts and theorems losely related to Galois representations. 2.1 Galois Representations Let K b e a eld and denote an algebrai losure by K . If L/K is a Galois extension (not neessarily nite), we get the Galois group Gal(L/K) := AutK(L) . The group is equipp ed with a natural top ology, the Krull top ology. This is the ase b eause Gal(L/K) is a topologial group as pro jetive limit of the disrete nite Galois groups of the nite Galois sub extensions. Therefore the absolute Galois group GK:= Gal(Ksep/K) is a pronite group, where Ksep denotes the separable losure of K in K . We will regard all o urring algebrai extensions of K as subelds of K . If K is a p erfet eld, the algebrai and the separable losure oinide. For a xed prime numb er ℓ , we have the ℓ -adi integers Zℓ:= lim ←− Z/ℓnZ and the eld of ℓ -adi rational numb ers Qℓ:= Quot(Zℓ) = Zℓ[1 ℓ] , whih is the ompletion of Q with resp et to the ℓ -adi disrete absolute value. This valuation extends uniquely to the algebrai losure Qℓ . The ℓ -adi distane given by this valuation indues for n∈N a top ology on M(n×n, Qℓ) . Beside the Zariski top ology on GLn(Qℓ) we get thereby another struture as top ologial group, whih we will use in the following denition. This yields a natural ontinuous ation of this top ologial group on Qℓ n equipp ed with any norm, esp eially the ℓ -adi one. This onstrution of the ℓ -adi top ology is suitable for any nite dimensional Qℓ -vetor spae V . For further details on the following denitions see [Ser68℄. Denition 2.1.1 For a eld K an ℓ -adi Galois representation is a homomorphism ρ:GK−→ GL(V) of topologial groups from the absolute Galois group of K to the general linear group of a nite dimensional Qℓ -vetor spae V equipped with the ℓ -adi topology. The dimension of V is al led the rank of ρ . This is the same as a Qℓ -vetor spae V equipp ed with the ℓ -adi top ology and a ontinuous GK -op eration. Two representations ρ, ρ′:GK−→ GL(V) are equivalent , if there exists a linear map φ∈GL(V) suh that φ−1◦ρ(g)◦φ=ρ′(g) for all g∈GK . 19 2. Preliminary Results An imp ortant example of an ℓ -adi Galois representation of GQ of rank one is the ylotomi harater χℓ:GQ−→ GL1(Qℓ) = Qℓ × . More preisely, it maps to Z× ℓ in the following way: For eah n∈N , we have a lo ok at the ylotomi extension Q(ζℓn) for a primitive ℓn -th ro ot of unity ζℓn . Then Gal(Q(ζℓn)/Q)∼ =(Z/ℓnZ)× , whih an b e hosen in suh a way that it ts together with the isomorphism for smaller n . Indep endent of the hoies, we get a ompatible system of ontinuous group homomorphisms whih gives rise to the harater. This onstrution an b e generalized to a eld K with harateristi unequal to ℓ . One of the main prop erties of the Galois representations ρHm,ℓ onstruted in Setion 3.3 and Setion 7.2 is the existene of a long unip otent element in its image. Denition 2.1.2 We say that a representation ρ:G−→ GL(V) for a group G and an n -dimensional vetor spae V over a eld K has a long unip otent element , if there exists an element g∈G suh that the Jordan normal form of ρ(g) is Jn(1) over K , where Jn(1) denotes a Jordan blok of length n to the eigenvalue 1 . For a go o d introdution to the onepts of algebrai numb er theory, have a lo ok at [Neu99℄. If K is a numb er eld, i.e. a nite extension of Q , then ΣK denotes the set of nite plaes , whih is the set of normalized non-arhimedean valuations of K . We identify ΣK\{0} with the set of non-trivial prime ideals of OK . For v∈ΣK\{0} we have two elds: the nite eld kv:= OK/v of harateristi pv and the ompletion via the indued metri Kv:= Quot(lim ←−(OK/vn)) , as eah plae orresp onds to a normalized disrete valuation. The adele ring AK of K is dened as AK:= Y v|∞ Kv |{z } =:AK,∞ ×Y′ v∈ΣK\{0} Kv, where AK,∞ is the produt of the ompletions of K aording to the valuation given by the Arhimedean plaes and Q′ is the restrited pro dut, i.e. almost all entries are in the rings of integers OKv . The idele group IK is the group of units A× K of the adele ring. For a nite Galois extension L/K and w∈ΣL\ {0} suh that w|v , i.e. w⊇vOL , we obtain two anonial subgroups of the Galois group Gal(L/K) , the deomposition group Dw:= {σ∈Gal(L/K)|σw =w} and a normal subgroup of Dw the inertia group Iw:= {σ∈Dw|σ(x)−x∈w∀x∈ OL} . Fixing an embedding of K in Kv , we get a natural emb edding of GKv in GK , whih orresponds to ho osing a nite plae w in K extending v and therefore xing GKv as a sp ei deomp osition group Dw . Setting lw:= OL/w , we have a short exat sequene of nite groups 1−→ Iw−→ Dw−→ Gal(lw/kv)−→ 1. 20 2.1. Galois Representations For a nite plae w6= 0 of L there is a unique nite plae v of K , suh that w|v . The extension L/K is alled unramied at w if [L:K] = [lw:kv] . In this ase we have Iw= 1 . For a nite plae v6= 0 of K multiple nite plaes w of L may exist, suh that w|v . The eld extension L/K is alled unramied at v if [L:K] = [lw:kv] for eah of them (equivalently one of them, as we have a Galois extension). Otherwise the nite plaes are alled ramied and for eah suh extension there is only a nite numb er of them. For a general eld K and L an algebrai extension, L/K is unramied at a non-arhimedean valuation v of K , if for eah nite eld extension L′/K inside L/K and eah valuation w′ of L′ extending v , l′ w′|kv is separable and [L′:K] = [l′ w′:kv] , otherwise L/K is alled ramied at v . In the numb er eld ase, Gal(lw/kv) is a nite yli group generated by the Frob enius. If w∈ΣL\ {0} is unramied, we have Dw∼ =Gal(lw/kv) and we an talk of a Frob enius element in the deomp osition group as well. For v∈ΣK\ {0} unramied and w, w′∈ΣL\ {0} suh that w, w′|v , there is an element σ∈Gal(L/K) mapping one to the other, i.e. σw =w′ . Therefore the orresponding deomposition groups are onjugated , i.e. σDwσ−1=Dw′ , as well as the Frob enius elements. The other way around, for onjugates of Frob enius elements we have orresp onding plaes of L . If we generalize to an arbitrary algebrai Galois extension L/K , the set of nite plaes ΣL is the pro jetive limit of the system of nite plaes of the nite sub extensions of L/K . This is dened via the following onnetion morphisms: whenever we have a sub extension L/L1/L2/K , we map w1∈ΣL1 to w2∈ΣL2 , where w2 is the unique plae suh that w1|w2 . The inertia and deomp osition group an b e dened as pro jetive limits in the same way. Denition 2.1.3 For an ℓ -adi Galois representation ρ of a number eld K , we say that ρ is unramied at v∈ΣK\ {0} , if ρ(Iw) = 1 for any valuation w of Ksep extending v . Let ρ be unramied at v∈ΣK\ {0} , then the Frob enius element Frobv,ρ in the representation ρ at v is the onjugay lass in GL(V) of the images of the Frobenius element in Dw for any w∈ΣKsep \{0} extending v : 1//Iw// ρ|Iw  Dw _  ////Gal(lw/kv)//  1 GK ρ  1//GL(V) . As the Frob enius element Frobv,ρ is a onjugay lass, its harateristi p olynomial fv,ρ(x) := det(1 ·x−Frobv,ρ)∈Qℓ[x] 21 2. Preliminary Results is well-dened. An ℓ -adi Galois representation is rational (resp etively integral ) if at almost all nite plaes v it is unramied, i.e. fv,ρ(x) exists, and the harateristi p olynomial has rational (resp etively integral) o eients. We will keep to the language of Rihard Taylor (f. [BLGGT10℄), for systems of ℓ -adi Galois representations. Denition 2.1.4 a) Let ℓ, ℓ′ be prime numbers. A rational ℓ -adi Galois representation ρ and a rational ℓ′ -adi Galois representation ρ′ of the same number eld K are ompatible at v∈ΣK\ {0} if they are both unramied at v and the harateristi polynomials fv,ρ(x) = fv,ρ′(x)∈Q[x] oinide. b) A weakly ompatible system (ρℓ)ℓ prime of Galois representations of a number eld K onsists of a family of rational, semi-simple ℓ -adi Galois representations ρℓ of K for eah prime number ℓ and a nite set S⊂ΣK , suh that the fol lowing holds: 1. For v∈ΣK\S and prime numbers ℓ, ℓ′ unequal to the harateristi of kv , the representations ρℓ, ρℓ′ are ompatible at v . 2. For v∈ΣK and ℓ equal to the harateristi pv of kv , the representation ρℓ is deRham in v and rystal line in v if v6∈ S (f. Denition 2.4.8 ). 3. For eah embedding τ:K//Q the τ -Hodge-Tate numbers of ρℓ are independent of ℓ (f. Denition 2.4.9 ). ) A weakly ompatible system (ρℓ)ℓ prime is al led irreduible if there is a set P of prime numbers of Dirihlet density 1 , i.e. lim s→1+ |log(s−1)|−1X ℓ∈P ℓ−s= 1, suh that for al l ℓ∈P the representation ρℓ is irreduible. It is also p ossible to extend this denition by ho osing a numb er eld M instead of Q . In this ase the family is indexed by the set of nite plaes of M and harateristi p olynomials in the ring M[x] are allowed. As this is not neessary for this work, we omit this and refer to the more general [BLGGT10℄, Denition 1.1. If ρ= (ρℓ)ℓ prime is a weakly ompatible system and S⊂ΣK the nite exeptional set. For a nite plaes v∈ΣK\S the harateristi p olynomials fv,ρℓ(x) of the Frob enius elements oinide in Q[x] for almost all ℓ , whih will b e alled fv,ρ(x) . This will b e the key ingredient in Setion 7.1 to dene an L -funtion for a sp eial kind of weakly ompatible systems of ℓ -adi Galois representations. 22 2.2. Étale Fundamental Group Funtor π´et 1 2.2 Étale Fundamental Group Funtor π´et 1 This introdution to the étale fundamental funtor π´et 1 from the ategory of No etherian separated onneted shemes to the ategory of groups is as in the rst hapter of [FK88℄. Denition 2.2.1 a) A ring homomorphism f:A−→ B of loal ommutative rings with unit is unramied , if f(mA)·B=mB and the indued eld extension A/mA−→ B/mB is nite and separable. b) Let X,Y be Noetherian separated shemes. The morphism f:Y −→ X is étale , if the fol lowing onditions are satised: 1. f is loal ly of nite type. 2. for every point x∈ X the morphism f♯ x:OY,f(x)−→ OX,x is at, unramied and makes OX,x a nitely generated OY,f(x) -algebra. For a No etherian separated sheme X , we all a No etherian separated sheme Y with an étale morphism X −→ Y an étale extension of X . We denote the full sub ategory of étale extensions of X in the ategory Sch(X) of shemes over X by Ét (X) (then every morphism in Ét (X) is étale, f. [FK88℄, Remark 2.2.). A morphism of Noetherian separated shemes is a overing if it is nite and étale. Again the full sub ategory Cov(X) of overings over X in Ét (X) has only morphisms whih are overings. This is b eause an étale morphism is nite, if and only if it is prop er (see page 282 of [FK88℄ and [Har06℄, Corollary 4.8 (e) ). If we x a geometri p oint s: spec(Ω) −→ X ( Ω separably losed), we get the asso iated funtor of geometri p oints over s Cov(X)−→ Sets,Y 7→ Y(s) := HomX(spec(Ω),Y). A pointed overing of (X, s) is a pair (Y, α) onsisting of Y ∈ Ob (Cov(X)) and an α∈ Y(s) . These form the ategory Cov(X, s) together with the mapping of p ointed overing spaes f: (Y1, α1)−→ (Y2, α2) whih is an X -morphism f:Y1−→ Y2 satisfying f◦α1=α2 . For a onneted Y ∈ Ob (Cov(X)) , we have |AutX(Y)| ≤ |Y(s)|, as there is at most one morphism from a p ointed overing sheme to a onneted p ointed sheme (see [FK88℄, (1)). Now we will have a lo ok at the ase when there exists exatly one morphism. 23 2. Preliminary Results Denition 2.2.2 For a Noetherian separated sheme X and a geometri point s of X a Galois overing is a onneted overing sheme Y over X if |AutX(Y)|=|Y(s)|. This leads to the full sub ategory Gal(X, s) of Galois overings in Cov(X, s) . Sine b etween two ob jets there is at most one morphism, we obtain a total ordering on the isomorphism lasses. Furthermore we get that the ob jets form an inverse system. For an X -morphism f:Z −→ Y b etween two Galois overings and σ∈AutX(Z) , there is exatly one σ′∈AutX(Y) suh that f◦σ=σ′◦f . This mapping denes a surjetive group homomorphism (see [FK88℄, (4)) and hene an inverse system of groups. Denition 2.2.3 For a Noetherian separated sheme X and a geometri point s of X , we dene the étale fundamental group (a pronite group) as the fol lowing inverse limit of nite groups with the disrete topology: π´et 1(X, s) := lim ←− (Y,α)∈ Ob (Gal(X,s)) AutX(Y). Then π´et 1 b eomes a ovariant funtor from the ategory of p ointed shemes to the ategory of pronite groups by onstruting suitable morphisms b etween the inverse systems out of a morphism of shemes (see [FK88℄, A1.3). The tame fundamental group is a fator group of the étale fundamental group. This group will b e of imp ortane b eause ontinuous representations of πtame 1(X, s) give nie ontinuous representations of π´et 1(X, s) . Denition 2.2.4 The tame fundamental group πtame 1(X, s) of a Noetherian separated sheme X and a geometri point s of X is the projetive limit of al l pointed Galois overings whih are tamely ramied i.e. for eah geometri point α: spec(Ω) −→ Y the ardinality |{σ∈AutX(Y)|σ◦α=α}| is invertible in OY,α . If X is onneted and s′ is another geometri p oint of X , the étale fundamental groups are isomorphi: π´et 1(X, s)∼ =π´et 1(X, s′) (see [FK88℄, A1.2). In this ase we write π´et 1(X) := π´et 1(X, s) and view it as a funtor from the ategory of onneted shemes to the ategory of isomorphism lasses of pronite groups. This is valid as well for the tame fundamental group. We dene the tame inertia group Itame K:= πtame 1(Gm,K) . 24 2.2. Étale Fundamental Group Funtor π´et 1 If we x a Noetherian separated onneted sheme and a geometri p oint s of X together with a overing Y , there is a natural ontinuous π´et 1(X, s) -ation on Y(s) . We will now explain this ation in more detail. By [FK88℄, (2) and (3) there is a Galois overing Z of X dominating Y , i.e. suh that there is an X -morphism Y −→ Z . Cho osing an α∈ Z(s) , we get a p ointed Galois overing (Z, α) of (X, s) and a natural bijetion HomX(Z,Y)−→ Y(s), f 7→ f◦α. Therefore the anonial right ation of AutX(Z) on HomX(Z,Y) yields a right ation on Y(s) . As we have a disrete group, this ation is ontinuous and an b e extended to a ontinuous right ation of π´et 1(X) on Y(s) via the anonial pro jetion π´et 1(X)////π´et 1(X)/π´et 1(Z, α) . For a dierent hoie of α we obtain a dierent ation, but this transformation is the same as a onjugation in π´et 1(X, s) . Furthermore it is indep endent of the hoie of Z , as for two hoies there is a third dominating them. Prop osition 2.2.5 Let X be a Noetherian separated onneted sheme. The assignment Y 7→ Y(s) establishes an equivalene between the ategory of overing spaes of X and the ategory of nite ontinuous π´et 1(X) -sets ([FK88℄, A I.5 ). In order to generalize the onept of sheaves the following denition was given by Artin in [Art62℄, Denition 1.1.1. Denition 2.2.6 A Grothendiek top ology onsists of a ategory T and a set Cov T of families {Ui φi −→U}i∈I of maps in T al led overings (where in eah overing the range U of the maps φi is xed) satisfying a) if φ is an isomorphism then {φ} ∈ Cov T ; b) if {Ui→U}i∈I∈Cov T and {Vij →Ui}j∈Ji∈Cov T for eah i then the family {Vij →U}i∈I,j∈J obtained by omposition is in Cov T ; ) if {Ui→U}i∈I∈Cov T and V→U∈ Mor (T) is arbitrary then Ui×UV exists and {Ui×UV→V}i∈I∈Cov T . As Ét (X) fullls all these prop erties for a Noetherian separated sheme X this yields an example of a Grothendiek top ology. 25 2. Preliminary Results Theorem 2.3.2 Let X be a smooth and projetive sheme over the nite eld Fq and X:= X ×spec Fqspec Fq . a) The polynomials Pj(t) = det(1 −t·Frob⋆|Hj(X,Qℓ))∈Qℓ[t] have rational integer oeients. These are independent of ℓ . b) The eigenvalues λ of Frob⋆|Hj(X,Qℓ) , and thus the reiproal roots of Pj(t) , al l have the omplex absolute value |λ|=qj 2. ) There is a funtional equation for ZX(t) = 2 dim X Q j=0 Pj(t)(−1)j+1 , namely ZX1 qjt=ǫ·qj 2χ(X)·tχ(X)·ZX(t). Here χ(X) = 2 dim X P j=0 (−1)jdim Hj(X,Qℓ) is the Euler harateristi of X and ǫ=(1 2 6 | j, (−1)N2|j, where N is the multipliity of the eigenvalue qj 2 of Frob⋆|Hj(X,Qℓ) . Let X b e a nitely generated No etherian separated sheme over the eld Fq and G a onstrutible sheaf of Qℓ -vetor spaes on X . For a geometri p oint α: spec(Fq)−→ X the residue eld κ(α) is nite and beause of that it denes an element in Gal(Fq/κ(α)) , the Frob enius fα:x7→ x|κ(x)| . This element ats on the stalk Gα of G and on the stalk Gα of the sheaf G=G ⊗Fq . Denition 2.3.3 a) We al l the sheaf G puntually pure of weight j if for al l suh geometri points α of X the eigenvalues of f−1 α:Gα−→ Gα are algebrai numbers whose omplex onjugates λ have omplex absolute value |λ|=qj 2d(α), d(α) = [κ(α) : Fq] . b) The sheaf G is al led mixed of weight less or equal to j if G has a ltration 0 = F(0) ⊂ F(1) ⊂...⊂ F(r)=G for whih al l fator sheaves F(ν)/F(ν−1) are puntual ly pure of weight less or equal to j . The following statement is [Del80℄, Théorème 3.3.1. Theorem 2.3.4 Let f:X −→ Y be a morphism of nitely generated shemes over Fq , and let G be a mixed sheaf of weight less or equal j on X . Then the diret image sheaves with ompat support Rnf!G are mixed of weight less or equal j+n . 32 2.4. Crystalline Representations 2.4 Crystalline Representations In order to present the onept of rystalline representations, it is neessary to dene the graded rings BdR, Bcris, Bst and BHT intro dued by Fontaine [Fon02℄ with their natural GKv -ation for a lo al eld Kv . The rst step is to study Witt vetors. 2.4.1 Witt vetors In the artile [Wit37℄, whih was published in 1937, Witt generalized the onstrution of Zℓ out of Fℓ for a given prime numb er ℓ to general ommutative rings. For a xed prime numb er ℓ and n∈N0 we dene the n -th Witt polynomial wn:= n X j=0 ℓjXℓn−j j∈Z[X0,...,Xn]. Denition and remark 2.4.1 Let ℓ be a prime number and A a ommutative ring. Then the fol lowing holds: a) For eah n∈N0 , there exist polynomials sn, mn∈Z[Y0,...,Yn, Z0,...,Zn] , suh that wn(s0,...,sn) = wn(Y0,...,Yn) + wn(Z0,...,Zn) and wn(m0,...,mn) = wn(Y0,...,Yn)·wn(Z0,...,Zn). b) The fol lowing onvention denes a ring struture on AN0 , whih is al led the ring of Witt vetors W(A) (an)n∈N0+ (bn)n∈N0:= ( sn(a0,...,an, b0,...,bn) )n∈N0, (an)n∈N0·(bn)n∈N0:= ( mn(a0,...,an, b0,...,bn) )n∈N0. ) If A is a perfet eld of harateristi ℓ , then W(A) is a omplete disrete valuation ring and its residue eld is A . For n∈N this struture an b e restrited to An by pro jetion on the rst n terms, the Witt vetors of length n . The ring W should b e viewed as the unique ovariant funtor from the ategory of rings to itself, for whih the following map is a homomorphism: W(A)−→ AN0 (an)n∈N07→ (wn(a0,...,an) )n∈N0 . 33 2. Preliminary Results For a more detailed version of the following approah, we refer to [FO08℄. Let Kv b e a lo al eld, that is a omplete disrete valuation eld, whose residue eld kv is p erfet of harateristi ℓ > 0 . The most imp ortant ase is as in Setion 2.1, where K is a numb er eld, v∈ΣK\{0} a plae of K and Kv the ompletion at v . Then the valuation v on Kv an b e uniquely extended to Kv as onsequene of Chevalley's Extension Theorem (f. [EP05℄). Kv might not b e omplete but by Krasner's lemma its ompletion Cv:= ˆ Kv is algebraially losed. For the rings of integers OKv:= x∈Kv|v(x)≥0 and OCv:= {x∈Cv|v(x)≥0} we get ompat rings with the following isomorphi fator rings of harateristi ℓ OKv/ℓOKv∼ =OCv/ℓOCv. Now we sp eialize A as the pro jetive limit of OCv/ℓOCv Frobℓ ←− OCv/ℓOCv Frobℓ ←− OCv/ℓOCv Frobℓ ←− ... i.e. A=n(a(n))n∈N0∈(OCv/ℓOCv)N0(a(n))ℓ=a(n−1) ∀n∈No and get a p erfet ring of harateristi ℓ . The ring A arries a anonial valuation indued by v , whih will not b e disussed in detail here, and is a omplete valuation ring with resp et to it. The ation of GKv on Kv an b e extended ontinuously to Cv and restrits to a lo al ation on OCv . Therefore A is endowed with a natural struture as GKv -mo dule, whih ommutes with the Frob enius on A and gives nally an ation of GKv on W(A) . For a= (a(n))n∈N0∈A we dene ea:= lim n→∞(g a(n)ℓn )∈ OCv , where g a(n)∈ OCv is some lift of a(n)∈ OCv/ℓOCv . It is easy to hek that the limit exists and that it is independent of the hoies. This yields a map θ:W(A)−→ OCv (an)n∈N07→ ∞ P n=0 ℓnfan whih is an epimorphism of GKv -mo dules. This result is obtained by restriting θ to the rst n omp onents of W(A) and then using the pro jetive limit pro ess. For n∈N0 , we suessively hoose a(n)∈ OCv , suh that we get a ompatible system of ℓn+1 -ro ots of ℓ , i.e. a(n) is a zero of Xℓn+1 −ℓ and a(n)= (a(n+1))ℓ . By pro jetion this denes a series of non-zero elements ℓ(n)∈ OCv/ℓOCv and therefore an element ℓ= (ℓ(n))n∈N0∈A for whih eℓ=ℓ . Then the kernel of θ is a prinipal ideal generated by ξ:= (−ℓ, 1,0,...)∈W(A) , whih denes a ξ -adi top ology on W(A) and W(A)[(0,1,0,...)−1] . 34 2.4. Crystalline Representations 2.4.2 The GKv -mo dule BdR : The ompletion of W(A)[(0,1,0,...)−1] in the ξ -adi top ology is the disrete valuation ring B+ dR with maximal ideal (ξ) and residue eld B+ dR/(ξ)∼ =Cv . Denition 2.4.2 The eld B+ dR is dened as the eld of frations of B+ dR : BdR := Quot(B+ dR) = Quot lim ←− n W(A)[(0,1,0,...)−1] ξn!. It has a natural dereasing ltration FilmBdR =ξmB+ dR for m∈Z . 2.4.3 The GKv -mo dule Bcris : We dene Acris to b e the ℓ -adi ompletion of the divided p ower envelop e of W(A) with resp et to (ξ) , i.e.: Acris := (∞ X n=0 wn ξn n!wn∈W(A), wn−→ 0 for n−→ ∞)⊂B+ dR. This is the same as taking the pro jetive limit lim ←− n A0 cris/(0,1,0,...)nA0 cris ∼ =Acris where A0 cris := (N X n=0 wn ξn n!N∈N0, wn∈W(A))⊂W(A)[(0,1,0,...)−1]. In this way we obtain the following subring B+ cris := Acris[(0,1,0,...)−1]⊂B+ dR . Again we ho ose suessively a(n)∈ OCv , suh that we get a non-trivial ompatible system of ℓn -th ro ots of unity, i.e. a(n) is a zero of Xℓn−1 , a(1) 6= 1 and a(n)= (a(n+1))ℓ , whih yields an element ε∈A . As (ε−1,0,...)∈ Ker (θ) = (ξ) we have log((ε, 0,...)) := ∞ X n=1 (−1)n+1 (ε−1,0,...)n n∈B+ dR. Denition 2.4.3 By loalizing and taking the subspae grading by BdR , we dene the graded ring Bcris := B+ cris[log((ε, 0,...))−1] = Acris[log((ε, 0,...))−1]⊂BdR. 35 2. Preliminary Results 2.4.4 The GKv -mo dule Bst : We set log((−ℓ, 0,...)) := − ∞ P n=1 ξn n·(0,1,0,...)n∈B+ dR (for a omplete approah to the logarithm see [FO08℄, Setion 6.1.3). This element is transendental over Quot(Bcris) . Denition 2.4.4 The ring Bst is dened as the Bcris -subalgebra of BdR generated by t:= log((−ℓ, 0,...)) : Bst := Bcris[t] = Bcris[log((−ℓ, 0,...))] = Bcris "− ∞ X n=1 ξn n·(0,1,0,...)n#. Remark 2.4.5 We have Bcris ⊆Bst ⊆BdR , whih gives that Bcris and Bst are domains. Eah ring is stable under the ation of GKv on BdR , whih is obtained by onsidering the projetive limit of the ations on W(A)[(0,1,0,...)−1] ξn . As A is of harateristi ℓ we have Fℓ//A , and by the funtoriality of W we get that W(Fℓ) = Zℓ//W(A) , whih leads to Qℓ⊆Bcris ⊆Bst ⊆BdR. 2.4.5 The GKv -mo dule BHT : Now we hoose an emb edding τ:K//Q , whih is the same as a ontinuous inlusion Kv//Qℓ=Kv . This yields a Qℓ -vetor spae struture on Kv and therefore on Cv by ontinuous extension. Q_  oO         //Qℓ _  nN ~~| | | | | | | | K o τ  > > > > > > > > //Kvp A A A A Q//Qℓ The element t has been hosen in suh a way that for an element g∈GKv , we have g·t=χℓ(g)t . Here χℓ:GKv−→ Q× ℓ⊂Bcris is the ylotomi harater, whih is dened as in 2.1 by the op eration on the ℓn -th ro ots of unity. The element t generates the maximal ideal of B+ dR and therefore the grading on BdR . As a eld BdR is isomorphi to Cv((t)) (the isomorphism dep ends on the hoie of ε ). 36 2.4. Crystalline Representations For a j∈Z the j -th Tate Twist Cv(j) of Cv is Cv viewed as GKv -mo dule twisted by the j -th p ower of the ylotomi harater, so that g·c=χj ℓ(g)g(c) for g∈GKv and c∈Cv . This is imp ortant b eause it is isomorphi as GKv -mo dule to the i -th graded omp onent of BdR : grjBdR = FiljBdR/Filj+1BdR =ξjB+ dR/ξj+1B+ dR =tjB+ dR/tj+1B+ dR ∼ =Cv(j). Denition 2.4.6 The Ho dge-Tate ring BHT is dened as the diret sum of al l Tate twists of Cv , whih is BHT =Cv[t, t−1] = M j∈Z grjBdR and the GKv -ation of BdR restrits in the fol lowing way g·X j∈Z cjtj=X j∈Z χℓ(g)jg(cj)tj for g∈GKv and cj∈Cv unequal to 0 only for a nite number of j∈Z . The residue eld kv lies inside OKv/ℓOKv and is xed by the GKv -ation. This denes the eld K0:= Quot(W(kv)) , whih is the xed eld of Bcris and Bst with resp et to the GKv -ation K0=BGKv cris =BGKv st ⊆Kv=BGKv dR =BGKv HT . Then Kv is a totally ramied extension of K0 and b oth elds oinide if K is unramied in v . Denition 2.4.7 For an ℓ -adi Galois representation ρ:GKv−→ GL(V) of Kv , like in Denition 2.1.1 , we dene Qℓ -vetor spaes, the ltered Dieudonné mo dules Dcris(V) := (Bcris ⊗V)GKv, Dst(V) := (Bst ⊗V)GKv, DdR(V) := (BdR ⊗V)GKv and DHT(V) := (BHT ⊗V)GKv as invariants of the tensor produts under the ation of the absolute Galois group. The rst two are free K0⊗Qℓ -mo dules and the seond ones are free Kv⊗Qℓ -mo dules. 37 2. Preliminary Results Then we have the following natural inequalities rank K0⊗QℓDcris(V)≤ rank K0⊗QℓDst(V)≤ rank Kv⊗QℓDdR(V) ≤ rank Kv⊗QℓDHT(V)≤dimQℓV. Denition 2.4.8 Let ρ be an ℓ -adi Galois representation of K on V , v∈ΣK\{0} suh that ℓ=pv= har (kv) . Then we get an ℓ -adi Galois representation ρ|GKv:GKv−→ GL(V) by xing an embedding K//Kv . ρ is al led            Ho dge-Tate at v , if rank Kv⊗QℓDHT(V) = dimQℓV , deRham at v , if rank Kv⊗QℓDdR(V) = dimQℓV , semi-stable at v , if rank K0⊗QℓDst(V) = dimQℓV , rystalline at v , if rank K0⊗QℓDcris(V) = dimQℓV , and we have: ρ rystal line at v⇒ρ semi-stable at v⇒ρ deRham at v⇒ρ Hodge-Tate at v. The grading on DHT(V) is given by the degree in t , so that eah graded piee gr−jDHT(V) = (Cv(−j)⊗V)GKv an b e haraterized by the ylotomi harater ating on it to the j -th p ower. The dimensions of these spaes play an imp ortant role in the lassiation of representations. Denition 2.4.9 Let ρ= (ρℓ)ℓ prime be a system of Galois representations of a number eld K , where ρℓ:GK−→ GL(Vℓ) where Vℓ is a Qℓ -vetor spae. For v∈ΣK\{0} , ℓ=pv= har (kv) and a xed embedding K//Kv a GKv -ation on Vℓ and as before by τ:K//Q a GKv -ation on Cv is given. The τ -Ho dge-Tate numb ers hv,j(ρ)∈N0 for an integer j are dened as hv,j(ρ) = dimQℓ(Cv(−j)⊗Vℓ)GKv. 38 2.5. Semi-Simpliation 2.5 Semi-Simpliation The semi-simpliation of a representation ρ:G−→ GL(V) , where V is a nite-dimensional L -vetor spae, is done by viewing V as an L[G] -mo dule. Then there is a nite, stritly dereasing hain of submo dules V=V0⊃V1⊃...⊃Vo={0} suh that Vj/Vj+1 is a simple L[G] -mo dule. The fators are alled Jordan-Hölder fators and are uniquely dened up to p ermutation. This denes a unique semi-simple L[G] -mo dule o−1 L j=0 Vj/Vj+1 and therefore a semi-simple representation ρss :G−→ o−1 M j=0 Vj/Vj+1. The following part is out of [Wor02℄, 2.1. Statement of the prop osition . Let K b e a numb er eld, v∈ΣK\ {0} a nite plae of K and p= har (kv) . For m∈N oprime to p we get a natural emb edding of the m -th ro ots of unity in the maximal unramied extension Knr v of Kv in kv∼ =Fp : µm={ζ∈Knr v|ζm= 1}// s && M M M M M M M M M M M M MOKv  ⊂Kv kv∼ =Fp . Let π b e a uniformizer of Knr v , then Knr v(π1 m) is a totally ramied extension of Knr v of degree m and we get a pro jetive system of maps e Ψm: Gal(Knr v(π1 m)/Knr v)−→ µm with (e Ψm(σ))(π1 m) = σ(π1 m) for σ∈Gal(Knr v(π1 m)/Knr v) . Its pro jetive limit is Ψ = lim ←− p∤me Ψm:Itame Kv= lim ←− p∤m Gal(Knr v(π1 m)/Knr v)−→ lim ←− p∤m µm As in [Wor02℄ on page 4 we dene its natural pro jetions Ψq−1:Itame Kv−→ µq−1⊆Knr v for a p -p ower q , whih dene Ψ uniquely. 39 2. Preliminary Results Let {m1,...,mr} b e the set of indies where the ltration of Dcris(V) jumps, i.e. grmDcris(V)6= 0 , and with dj:= rank K0⊗QℓgrmjDcris(V)∈N the multipliity of mj , i.e. the rank of the asso iated quotient. Prop osition 2.5.1 ( [Wor02℄, Prop.3 ) Assume that for v∈ΣK\{0} , Kv is absolutely unramied, i.e. Kv/Q is unramied, and let w be the unique extension of v to Kv . If the ℓ -adi Galois representation ρ:GKv−→ GL(V) fators through Zℓ , is rystal line and if the length of the ltration on Dcris(V) is less than ℓ , then the fol lowing holds: a) The semi-simpliation of the modℓ redued Iw -module V , via ρ|Iw:Iw−→ GL(V) , is wel l-dened and the ation of Iw fators through the tame quotient Itame Kv . b) For a simple subquotient W of the Iw -module V of dimension d one has EndFℓ(W)∼ =Fℓd . Fixing an isomorphism gives W the struture of a one dimensional Fℓd -vetor spae on whih Itame Kv ats via multipliation with Ψi0+...+id−1ℓd−1 ℓd−1 , where the indies −ij run through {m1,...,ms} suh that eah omponent of (m1,...,ms) (ounted with multipliities) appears as some index −ij for some subquotient. Corollary 2.5.2 ( [Wor02℄, Cor.4 ) The automorphisms dened by g∈Itame Kv , where the tame fundamental group is viewed as a subgroup (not unique) of the inertia group Iw , satisfy detFℓρ(g) = Ψℓ−1(g)−S, where S:= s P k=1 dkmk . 40 3 Intro dution to MCχ In Setion 2.2, we saw that it is p ossible to onstrut Galois representations by onstruting lisse Qℓ -sheaves. In order to do so, we use the middle onvolution as a geometri op eration. We will follow Chapter 2 in [Kat96℄. 3.1 The Middle Convolution Let K b e a eld and ℓ a prime numb er unequal to the harateristi of K . We x an algebrai group A over K with multipliation map µ:A×A−→ A . We denote by Db c(A,Qℓ) the bounded derived ategory of onstrutible Qℓ -sheaves on A , whih is onstruted by taking the ategory of b ounded hain omplexes and lo alizing the quasi-isomorphisms (for more details see [Kat96℄, Setion 2.2). For two ob jets F,G ∈ Db c(A,Qℓ) , we have the exterior tensor pro dut F⊠G:= pr ∗ 1F ⊗ pr ∗ 2G ∈ Db c(A ×A,Qℓ), and the shift by m∈Z to the left F[m] . As usual the onstrutible Qℓ -sheaves are emb edded as degree 0 ob jets. Denition 3.1.1 The ! -onvolution of F,G is dened as the right derivative of the diret image with ompat support by the multipliation map µ F ∗!G:= Rµ!(F⊠G)∈Db c(A,Qℓ) and their ∗ -onvolution as the right derivative of the diret image by the multipliation map µ F ∗∗G:= Rµ∗(F⊠G)∈Db c(A,Qℓ) of the exterior tensor produt. For eah onstrutible sheaf X on A , the supp ort supp(X) is the losure of the set {a∈A| Xa6= 0} and therefore a variety of some dimension. An element F ∈ Db c(A,Qℓ) is alled a perverse sheaf , 41 3. Introdution to MCχ ohomologially rigid, and dene indutively Hm+1,ℓ := (MTi∗L(1,−1)(MC−1(Hm,ℓ)) 2 |m, MTi∗L(−1,1)(MC−1(Hm,ℓ)) 2 ∤m∈Tℓ(K). By Theorem 3.1.7, we have that Hm,ℓ is ohomologially rigid. That Hm,ℓ resp ets an orthogonal resp etively sympleti form is a onsequene of Poinaré duality (see [DR99℄, Corollary 5.10). The rest of the pro of is an indution on m with the help of Prop osition 3.2.1. For m= 0 the mono dromy tuple of i∗Hm,ℓ is (J1(−1),J1(−1),J1(1)) ∈GL1(Qℓ)3 , whih is of the given form. So we start with Hm,ℓ whose loal mono dromy is of the predited type. Therefore we an determine the rank of Hm+1,ℓ : rk(Hm+1,ℓ) = rk(MC−1(Hm,ℓ)) = 2(m+ 1) −e1(0,−1,Hm,ℓ)−e1(1,1,Hm,ℓ)−e1(∞,−1,Hm,ℓ) =           2(m+ 1) −m 2−m 2−0m≡0 mod 4 2(m+ 1) −m+1 2−m−1 2−0m≡1 mod 4 2(m+ 1) −m 2−m 2−0m≡2 mod 4 2(m+ 1) −m+1 2−m−1 2−0m≡3 mod 4            =m+ 2 For the alulation of the loal mono dromy at 0 , we have two ases. If m is even, the lo al mono dromy of Hm,ℓ is of the form J1(1)m 2⊕J1(−1)m 2+1 : J1(1)m 2⊕J1(−1)m 2+1 MC−1 0⊕J2(1)m 2+1 MTi∗L(1,−1) J2(1)(m+1)+1 2 If m is o dd, then J2(1)m+1 2MC−1 J1(−1)m+1 2⊕J1(1)m+3 2 MTi∗L(−1,1) J1(1)m+1 2⊕J1(−1)m+1 2+1. At 1 we start with the ase m≡0 mod 4 and therefore we get: J2(1)m 2⊕J1(−1) MC−1 J1(−1)m 2⊕J2(1)⊕J1(1)m 2 MTi∗L(1,−1) J1(1)(m+1)−1 2⊕J2(−1)⊕J1(−1)(m+1)−1 2 for m≡1 mod 4 : J1(1)m−1 2⊕J2(−1) ⊕J1(−1)m−1 2MC−1 0⊕J3(1) ⊕J2(1)m−1 2 MTi∗L(−1,1) J3(1) ⊕J2(1)m+1 2−1 for m≡2 mod 4 : J3(1)⊕J2(1)m 2−1MC−1 J2(−1)⊕J1(−1)m 2−1⊕J1(1)m 2+1 MTi∗L(1,−1) J2(1)⊕J1(1)(m+1)−3 2⊕J1(−1)(m+1)+1 2 for m≡3 mod 4 : J2(1) ⊕J1(1)m−3 2⊕J1(−1)m+1 2MC−1 J1(−1) ⊕0⊕J2(1)m+1 2 MTi∗L(−1,1) J1(−1) ⊕J2(1)m+1 2. 48 3.3. Constrution of Hm,ℓ The Jordan normal form of the lo al mono dromy of i∗Hm,ℓ at ∞ is of the form Jm+1(1) . Then we get by the previous remark that the lo al mono dromy of MC−1(Hm,ℓ) is Jm+2(1) . For L(−1,1) and L(1,−1) the lo al mono dromy at ∞ is −1 . Altogether we have: Jm+1(1) MC−1 Jm+2(−1)    MTi∗L(1,−1) MTi∗L(−1,1)   Jm+2(1).  Corresp onding to the onstruted lisse Qℓ -sheaf i∗Hm,ℓ on A1 K\ {0,1} , we have a ontinuous representation ρi∗Hm,ℓ :π´et 1(A1 K\ {0,1})−→ GL((i∗Hm,ℓ)x) for x∈A1 K\ {0,1} (see Corollary 2.2.12). This map an b e tensored by the following ontinuous one dimensional representation det(ρi∗Hm,ℓ ) : π´et 1(A1 K\{0,1})−→ {±1} ⊂ Qℓ ×, γ 7→ det(ρi∗Hm,ℓ (γ)), whih is in our ase det(ρi∗Hm,ℓ (γ0)) = det(ρi∗Hm,ℓ (γ1)) = (1m6≡ 0 mod 4 −1m≡0 mod 4 . We get a representation ρi∗Hm,ℓ ⊗det(ρi∗Hm,ℓ ) : π´et 1(A1 K\{0,1})−→ SL((i∗Hm,ℓ)x), whih fators through SO((i∗Hm,ℓ)x) for even m and through Sp((i∗Hm,ℓ)x) for o dd m . Using the Corollary again, we get the Qℓ -sheaf ] Hm,ℓ := Vρi∗Hm,ℓ ⊗det(ρi∗Hm,ℓ ) on A1 K\{0,1} . This will lead us to the wanted Galois representation in Setion 7.2. Theorem 3.3.2 For m∈N0 and a prime number ℓ , the Zariski losure of the monodromy group of ] Hm,ℓ is as fol lows: a) Sp(W) for m odd, b) G2(W) for m= 6 , ) SO(W) for m even and m6= 6 . 49 3. Introdution to MCχ Proof: Sine Hm,ℓ ∈Tℓ(K) , the Qℓ -sheaf i∗] Hm,ℓ is irreduible on A1 K\{0,1} . Therefore we have that H=ρi∗Hm,ℓ ⊗det(ρi∗Hm,ℓ )(γ0), ρi∗Hm,ℓ ⊗det(ρi∗Hm,ℓ )(γ1) is irreduible and onneted by the presene of the long unip otent element. By the onsiderations ab ove the group H leaves a sympleti form invariant if m is o dd, i.e. H≤Sp(W) , and an orthogonal form if m is even, i.e. H≤SO(W) . Moreover the presene of the long unip otent element, given by the mono dromy at ∞ , implies that H is tensor indeomp osable and by [SS97℄, Theorem B, we have the following p ossibilities for maximal losed redutive subgroups of Sp(W) resp etively SO(W) ontaining H : (a) A1<Sp(W) resp etively SO(W) (for p= 0 or p > h ), (b) SO(W).2<Sp(W) (for p= 2 ), () G2<SO7 (resp etively Sp6 if p= 2 ), (d) A2.2<Sp8 (for p= 2 ), (e) B3<SO8 . Here p= har (Qℓ) = 0 , whih ontradits ase (b) and (d). The ase (e) is not p ossible b eause for m= 7 a sympleti form is resp eted. For m= 6 the laim was proved in [DR10℄, Theorem 1. For m6= 6 the presene of the unip otent mono dromy element at 1 rules out the ase H≤A1= PSL2(W) . This proves the laim.  50 4 Ho dge Strutures and Middle Convolution The following intro dution to Ho dge theory is taken out of [PS08℄. 4.1 Ho dge Strutures Let b e m∈Z and R⊆R a No etherian ring, suh that R⊗Q is a eld and VR a nitely generated R -mo dule. Denition 4.1.1 A (pure) R -Ho dge struture of weight m on VR is a diret sum deomposition VC:= VR⊗C=M p+q=m Vp,q with Vp,q =Vq,p omplex vetor spaes for eah p, q . The numbers hp,q(V) := dimCVp,q are al led the Ho dge numbers of the Hodge struture. A morphism of Ho dge strutures f:VR−→ WR is an R -linear map suh that its omplexiation fC=f⊗ id C preserves types, i.e. fC(Vp,q)⊆Wp,q . Dening a Ho dge struture of weight m on a nite dimensional omplex vetor spae V is the same as giving a Hodge ltration F• of V . That is a dereasing ltration of omplex vetor spaes, suh that Fp∩Fq={0} for p+q=m+ 1 . A Ho dge ltration is asso iated to a Ho dge struture by Fp:= M r≥p Vr,s and vie versa Vp,q := Fp∩Fq. The free rank one R -mo dules whih arry a Ho dge struture are all of even weight and up to isomorphism of the following form: Denition 4.1.2 A Ho dge struture of Tate , denoted by Z(n) , for n∈Z is the Z -module (2πi)nZ⊂C with sum deomposition Z(n)⊗C=V−n,−n , whih has therefore weight −2n . If we have an R -Hodge struture VR of weight m , the Tate twist VR(n) is an R -Hodge struture of weight m−2n . It has VR⊗(2πi)nZ as underlying R -module, while VR(n)p,q =Vp−n,q−n R . 51 4. Ho dge Strutures and Middle Convolution 4.2 Variations of Ho dge Struture Ho dge strutures o ur quite naturally on the lo al systems onstruted by middle onvolution. In this ase, one gets a whole sheaf of Ho dge strutures tting together, whih is alled a variation of Ho dge struture (VHS). Denition 4.2.1 Let X be a omplex manifold. A variation of R -Ho dge struture (VR,F•,∇) of weight m on X onsists of the fol lowing data:  a loal system VR of nitely generated R -modules on X ,  a nite dereasing ltration F• of the holomorphi vetor bund le V:= VR⊗ OX by holomorphi subbund les (the Ho dge ltration ). These data should satisfy the fol lowing onditions: a) for eah x∈ X the ltration F• x of Vx∼ =VR,x ⊗C denes a Hodge struture of weight m on the nitely generated R -module VR,x , b) the onnetion ∇:V−→ V⊗Ω1 X , whose sheaf of horizontal setions is VC , satises the Griths' transversality ondition ∇(Fp)⊆ Fp−1⊗Ω1 X. A morphism of variations of Ho dge struture is a morphism of loal systems whih preserves types, i.e. agrees with the ltrations. We x a omplex manifold X , a base p oint x∈ X and a Ho dge struture V . For eah group homomorphism ρ:πtop 1(X, x)−→ Aut(V) , we get a lo ally onstant variation of Ho dge struture. This is the same metho d as in Corollary 2.2.10, gluing V as stalk in eah p oint and gluing the lo al Ho dge ltration. This prop erty haraterizes the lo al system obtained by representation preserving types. Therefore we get ∇(Fp)⊆ Fp⊗Ω1 X . By VX we denote the variation for the trivial representation. For a xed Ho dge struture V , the Weil operator C is the C -linear automorphism of V , suh that for all v∈Vp,q we have C(v) = ip−q·v . Denition 4.2.2 a) A p olarization of an R -Hodge struture VR of weight m is an R -valued bilinear form Q:VR⊗VR−→ R whih is (−1)m -symmetri and suh that 1. The orthogonal omplement of Fn is Fm−n+1 for al l n∈Z , 2. The hermitian form Q(C(·),·) : VC⊗VC−→ R on VC is positive-denite. 52 4.2. Variations of Ho dge Struture b) A p olarization of a variation of R -Hodge struture V of weight m on X is a morphism of variations Q:V⊗V−→ R(−m)X whih indues on eah bre a polarization of the orresponding R -Hodge struture of weight m . Theorem 4.2.3 Let X be a ompat Kähler manifold. Let Hp,q(X) be the spae of ohomology lasses whose harmoni representative is of type (p, q) . There is a diret sum deomposition Hm dR (X, C) := Hm dR (X)⊗C=M p+q=m Hp,q(X). Moreover Hp,q(X) = Hq,p(X) . If we denote the losed Kähler form of X with ω and the dimension of X with n , then the Ho dge-Riemann form on Hm dR (X, C) is the bilinear form Q(α, β) = (−1)m(m−1) 2ZX α∧β∧ωn−m, whih is a p olarization of the previously dened pure R -Ho dge struture on Hm dR (X, C) of weight m (f. [PS08℄, Theorem 1.33). The standard examples are geometri variations of Ho dge struture (see page 507-508 in [SZ85℄). Remark 4.2.4 a) Given a smooth, proper holomorphi mapping f:X−→ S with X as above a Kähler manifold. Then Rmf∗Q is the underlying system of a variation of Hodge struture of weight m , dened over Q , in whih F•(s) is the usual Hodge ltration of the ohomology of the bre Hm dR (Xs,C) . By adjusting the up-produt on ohomology by the use of the Kähler lass and its (at) primitive deomposition, one obtains a polarization over R for Rmf∗R in the geometri ase. If X is a family of algebrai varieties, then the polarization is in fat dened over Q . b) By [SZ85℄, Remark 3.3 , the polarized struture passes on to subvariations, kernel and images of funtorial morphism of ohomology. Given a variation of Hodge struture, it is in some ases possible to extend the struture to a punture of the underlying spae. This struture is not pure any more, but onsists of the sum of multiple Ho dge strutures of dierent weights, hene it is a mixed Ho dge struture (MHS). 53 4. Ho dge Strutures and Middle Convolution Denition 4.2.5 An R -mixed Ho dge struture on VR onsists of two ltrations:  an inreasing ltration by rational vetor spaes on VR⊗Q , the weight ltration W• and  a dereasing ltration F• by omplex vetor spaes on VC=VR⊗C , the Ho dge ltration . The Hodge ltration indues a pure K:= (R⊗Q) -Hodge struture of weight m on eah graded piee GrW m(VR⊗Q) = Wm/Wm−1 by Fp(GrW m(VR⊗Q)⊗C) = (Fp∩Wm⊗C+Wm−1⊗C)/(Wm−1⊗C). The Ho dge numb ers are the dimensions of the graded piees of this indued grading: hp,q(V) := dimCGrp F(GrW p+q(VR⊗Q)⊗C) = dimCFp(GrW p+q(VR⊗Q)⊗C)/Fp+1(GrW p+q(VR⊗Q)⊗C). For two nitely generated R -modules VR, VR ′ with R -mixed Hodge strutures a morphism f:VR−→ VR ′ (of weight 0 ) is an R -linear map, whih indues for m∈Z morphisms of Hodge strutures by GrW m(f) : GrW m(VR⊗Q)−→ GrW m(VR ′⊗Q). A graded p olarization on an R -mixed Hodge struture is a polarization of eah GrW m(VR⊗Q) , whih has a pure (R⊗Q) -Hodge struture. 4.3 Extensions of Variations of Ho dge Struture In order to get some information on the Ho dge struture onstruted by middle onvolution, whih will b e introdued in the next setion, it is helpful to have a lo ok at the limit struture at the singularities. This is p ossible as the struture an b e extended in some ases by the work of Shmid. We will summarize the results of [Sh73℄. Every omplex lo al system an b e seen as a holomorphi vetor bundle with a at (and therefore integrable) onnetion. In fat, if the ground spae is omplex analyti b oth ategories are equivalent. Let (V,∇) b e a holomorphi vetor bundle on the puntured disk D∗ equipp ed with an integrable onnetion. An extension (e V,e ∇) of the bundle to D is said to b e logarithmi at 0 if ∇ extends to a morphism e ∇:e V −→ e V ⊗Ω1 D(log z) whih satises Leibniz' rule, i.e. e ∇(fs) = fe ∇(s) + s⊗ d f for every lo al setion f of OD and s of e V . The Poinaré residue map an b e dened as R: Ω1 D(log z)−→ Oz∼ =C, ω =η∧ d z z+η′7→ η(0), suh that z= 0 is an equation for D and η, η′ not ontaining d z . This indues a C -linear endomorphism res 0(e ∇) of e V0 , the residue at 0 . 54 4.3. Extensions of Variations of Ho dge Struture If we x the anonial ontinuous setion τ:C/Z−→ [0,1) + iR⊂C , we get the following prop osition due to Manin (see [Del70℄). Prop osition 4.3.1 Let (V,∇) be a holomorphi vetor bund le on the puntured disk D∗ equipped with an integrable onnetion. There exists a unique extension e V of V , al led the anonial extension , to a vetor bund le on D suh that ∇ extends to a logarithmi onnetion e ∇ on e V whose residue at 0 has its eigenvalues in the image of τ , i.e. their real part is greater or equal to 0 and less than 1 . Let V b e a p olarized variation of C -Ho dge strutures of weight m on D∗ . Supp ose that the lo al mono dromy op erator is T∈GLn(C) , where T is unip otent and we have a dereasing ltration of holomorphi vetor bundles F• . Now we want to extend this Ho dge ltration F• to D so that we get something lose to a VHS. This extended ltration gives a mixed Ho dge struture in 0 , where the weight ltration an b e desrib ed very expliitly in the following way. As T is unip otent there is a nilpotent matrix N , suh that T= exp N . On a nite dimensional vetor spae every nilpotent endomorphism has a Jordan deomp osition and therefore an b e written as sum of Jordan blo ks to the eigenvalue 0 . There is an appropriate basis (v1,...,vj) of length j for eah Jordan blo k Jj(0) in the Jordan normal form of N . We an dene an inreasing ltration on the vetor spae by putting Wo:=      {0}o≤ −j, hv1,...,v⌊o+j+1 2⌋i −j < o < j −1, hv1,...,vjij−1≤o. By adding these for the dierent blo ks and shifting it by an integer m , we get the following prop erties of the weight ltration, whih desrib e it uniquely. Denition and remark 4.3.2 Given a nilpotent endomorphism N of a nite dimensional vetor spae V , there exists a unique inreasing ltration W•=W•(N, m) of V , al led the weight ltration of N entred at m , with the properties a) N(Wo+2)⊆Wo, o ∈N0 b) the map No: GrW m+oV−→ GrW m−oV is an isomorphism for al l o∈N0 . Moreover, there is a Lefshetz-typ e deomp osition GrWV= m M o=0 o M r=0 NrPVm+o with PVm+o:= Ker (No+1 : GrW m+oV−→ GrW m−o−2V) and the endomorphism N has dimCPVm+o Jordan blo ks of size o+ 1, o = 0,...,m . 55 4. Ho dge Strutures and Middle Convolution As [Sh73℄, Theorem 6.16 we nd the following result: Theorem 4.3.3 Let V be a polarized variation of C -Hodge strutures of weight m on D∗ with loal monodromy operator T∈GLn(C) , suh that T is unipotent. Choose N∈GLn(C) nilpotent suh that exp N=T . The Hodge bund les F• of V extend to holomorphi subbund les e F• of the anonial extension e V , and the triple V Hdg 0:= (e V0, W•(N, m),e F• 0) is a mixed Hodge struture, al led the anonial bre . This is an important tool for the determination of the original variation of C -Ho dge strutures. For a unip otent T a nilp otent matrix N with exp N=T has the same Jordan blo k struture but eigenvalue 0 instead of 1 . Therefore the rational dimension of GrW m±o(e V0⊗Q) for o∈N0 is the numb er of Jordan blo ks of o dd length whose length is greater than o , if o is even, and the numb er of Jordan blo ks of even length whose length is greater than o , if o is o dd. Let us assume from now on that T is a long unip otent element, i.e. has Jordan normal form Jn(1) , where n is the rank of V . By the last theorem, we an extend a polarized variation V of C -Ho dge struture of weight m on P1 C\S ( S nite) to a xed s∈S by a mixed Ho dge struture V Hdg s= (e Vs, W•(1 2πi log T, m),e F• s). The weight ltration W•:= W•(1 2πi log T, m) on Vs indues therefore the following dimensions of the grading: dimQGrW j(Vs) = (1−n < j −m < n and 2|j−m−(n−1), 0 else . In general this do es not dene e F• s uniquely. But exatly in this stritly one and zero dimensional ase the dimensions of the Ho dge ltration are determined by the dimensions of the weight ltration. This is done by alulating the Ho dge numb ers of the mixed Ho dge strutures V Hdg s . As eah graduated piee is of dimension either 0 or 1 , there is exatly one Ho dge struture over Q of a given even weight m , namely Q(−m 2) with Ho dge type (m 2,m 2) . Sine N lowers the weight stritly by 2 , the Ho dge ltration has maximal length and the mixed Ho dge struture is uniquely dened by the weight ltration. Summarizing we obtain the following result: Corollary 4.3.4 Let V be a polarized variation of C -Hodge struture of weight m and rank n on P1 C\S ( S nite) and s∈S suh that the monodromy operator at s is a long unipotent element. Then the Hodge ltration of V has maximal length. 56 5 Motivi Desription of Hm,ℓ For a general intro dution to motives, have a look at [Jan94℄ or [And04℄. The following part is a lose adaption to hapter 8 of [Kat96℄. 5.1 Setting Let K b e an algebraially losed eld and ho ose o≥2 distint p oints s1,...,so∈A1 K . For a prime numb er ℓ and an integer N∈N , suh that har (K)∤Nℓ , we x a primitive N -th ro ot of unity in K and a primitive N -th ro ot of unity ζN in Q . We dene the rings RN,ℓ := Z[ζN,(Nℓ)−1]⊂Q and SN,o,ℓ := RN,ℓ[T1,...,To][∆−1]⊂Q(T1,...,To), where ∆ := Q i<j (Ti−Tj) . By the xation ab ove, we get a unique ring homomorphism ϕ:SN,o,ℓ −→ K with the prop erty, that ϕ(Ti) = si for i= 1,...,o and that ζN is mapp ed to the hosen N -th ro ot of unity in K . For m∈N0 we onsider the following ane spaes A(o, m + 1)RN,ℓ := spec RN,ℓ[T1,...,To, X1,...,Xm+1][∆−1 m+1] where ∆m+1 := Y i<j (Ti−Tj)· m+1 Y i=1 o Y j=1 (Xi−Tj)· m Y i=1 (Xi+1 −Xi) with natural pro jetions pr i:A(o, m + 1)RN,ℓ ////A1 SN,o,ℓ \ {T1,...,To} indued by the emb edding SN,o,ℓ[X]//RN,ℓ[T1,...,To, X1,...,Xm+1][∆−1 m+1] with X7→ Xi . On Gm,RN,ℓ with o ordinate Z , one has the Kummer overing of degree N of the equation YN=Z . Let µN(RN,ℓ) denote the N -th ro ots of unity, then ho osing an N -th primitive ro ot of unity in Qℓ is the same as ho osing an emb edding χ:µN(RN,ℓ)//Qℓ × . Then the overing denes a onneted µN(RN,ℓ) -torsor, whih gives a representation π´et 1(Gm,RN,ℓ )ε////µN(RN,ℓ)χ //Qℓ × and therefore the orresp onding Kummer sheaf Lχ . For any sheme G and any morphism f:G −→ Gm,RN,ℓ , we dene Lχ(f):= f∗Lχ . Let f:A(o, 2)RN,ℓ −→ Gm,RN,ℓ b e indued by the ring homomorphism X7→ X2−X1 . 57 5. Motivi Desription of Hm,ℓ The funtoriality of ( −1 )-omp onent of the higher diret image in the sense of Setion 5.2 (the notion extends in an obvious way to X and to D ) and again the exatness of the sequene yield the following hain of isomorphisms K=m= (Rm(φA)!Qℓ)−1 =m∼ =im Rm(φA)!Qℓ→Rm(φXA)∗Qℓ−1 ∼ =ker Rm(φXA)∗Qℓ→Rm(φDA)∗Qℓ−1. By Corollary 5.3.1 the sheaf K=m is lisse and the isomorphisms imply that ker Rm(φX)∗Qℓ→Rm(φD)∗Qℓ−1 is lisse to o. It follows from prop er base hange that ker Rm(φXA)∗Qℓ→Rm(φDA)∗Qℓ−1 is lisse, where DA:= ` i∈I Di,A and φDA=` i∈I φXA|Di,A . We laim that the natural map ψ: ker Rm(φXA)∗Qℓ→Rm(φDA)∗Qℓ−1−→ ker Rm(φXA)∗Qℓ→Rm(φDA)∗Qℓ−1, where φDA:= ` i∈I φDi,A :DA−→ A1 A\{0,1} , is an isomorphism. In order to prove that by the Sp eialization Theorem (see [Kat90℄, 8.18.2), it sues to show this for any losed geometri p oint x of HypA . As (Rm(φA)!Qℓ)−1 =m∼ =ker Rm(φXA)∗Qℓ→Rm(φDA)∗Qℓ−1 , we have to show that (Hm c(HypA,x,Qℓ))−1 =m−→ ker Hm ´et (XA,x,Qℓ)→Hm ´et (DA,x,Qℓ)−1 is an isomorphism for DA,x := ` i∈I Di,A,x . We dene the following sequene of stalks X0 A,x := XA,x , and for natural numb ers i , let Xi A,x denote the disjoint union of the irreduible omponents of the lo us, where i pairwise dierent omp onents of DA,x meet. It follows from the Weil onjetures [Del74℄ that the sp etral sequene E1=Hj ´et(Xi A,x,Qℓ)x⇒Hi+j c(UA,x,Qℓ) degenerates at E2 . Consequently, we have (Hm c(HypA,x,Qℓ))=m∼ =ker Hm ´et (XA,x,Qℓ)→Hm ´et (DA,x,Qℓ), whih proves that the map ψ is an isomorphism as laimed. So, (Rm(φHypA)∗Qℓ)−1 =m∼ =ker Rm(φXA)∗Qℓ→Rm(φDA)∗Qℓ−1 =1 2(1 −σ)ker Rm(φXA)∗Qℓ→Rm(φDA)∗Qℓ, where the last equality is by using representation theory of nite (yli) groups. It follows that φ∗K=m=φ∗(Rm(φHyp)!Qℓ)−1 =m∼ =1 2(1 −σ)ker Rm(φX)∗Qℓ→Rm(φD)∗Qℓ, as laimed.  64 5.4. Analytiation of Hm,ℓ 5.4 Analytiation of Hm,ℓ Let K b e a number eld and S⊆K a nite set. We x an emb edding K//C . This yields a ontinuous morphism ι:πtop 1(C\S)//\ πtop 1(C\S) = π´et 1(A1 C\S)−→ π´et 1(A1 K\S) . A lisse Qℓ -sheaf V on A1 K\S orresp onds by Corollary 2.2.12 to a ontinuous representation ρV:π´et 1(A1 K\S)−→ GLn(Qℓ) . Denition 5.4.1 The analytiation Van of V is the loal system of Qℓ -modules VρV◦ι on C\S orresponding to the representation ρV◦ι:πtop 1(C\S)−→ GLn(Qℓ) by Corol lary 2.2.10 . The omparison isomorphism b etween étale and singular ohomology implies (after xing an isomorphism C∼ =Qℓ of elds), that (φ∗K=m)an ∼ =1 2(1 −σ)an(ker Rm(φX)an ∗Qℓ→Rm(φD)an ∗Qℓ), as 1 2(1 −σ) is an algebrai pro jetor and hene deRham. Further we have a morphism b etween smo oth pro jetive varieties on the right hand side, whih is given by restrition of inlusions of the smo oth divisors Di . Applying [Del87℄, Prop osition 1.13. to the irreduible sheaf (φ∗K=m)an , it omes from a variation of Ho dge struture in the following sense: Remark 5.4.2 The loal system of Q -vetor spaes Gm:= 1 2(1 −σ)an |{z } =1 2(1−σan) ker (Rm(φX)an ∗Q→Rm(φD)an ∗Q) on A1 C\{0,1} is a polarized variation of Hodge struture, whih is pure of weight m , sine it is a subvariation of Rm(φX)∗Z by Remark 4.2.4 b). Moreover by Corol lary 5.3.1 and Corol lary 5.3.2 , we have Gm⊗Qℓ∼ =(φ∗K=m)an = (Hm,ℓ|A1 C\{0,1})an. Conluding we have: Theorem 5.4.3 Let Hm,ℓ be as in Theorem 3.3.1 . Then there exists a loal system of Z -modules Gm on A1 C\{0,1} underlying a polarized variation of Z -Hodge struture (Gm,F•,∇) on C\{0,1} pure of weight m suh that (i∗Hm,ℓ)an ∼ =Gm⊗Qℓ. The indued isomorphism on the stalks (i∗Hm,ℓ)an x∼ =Gm⊗Qℓx for x∈C\{0,1} is given by the 65 5. Motivi Desription of Hm,ℓ omparison isomorphism between étale ohomology and singular ohomology 1 2(1 −σ)ker Hm ´et (Xx,Qℓ)→Hm ´et (Dx,Qℓ)∼ =1 2(1 −σ)ker (Hm B(X(C)x,Z)→Hm B(D(C)x,Z)) ⊗Qℓ. Moreover, the Hodge ltration of Gm has maximal length. Proof: All laims but the last follow from Remark 5.4.2. The last laim follows from the long unip otent lo al mono dromy of Hm,ℓ using Theorem 3.3.1 and Corollary 4.3.4.  In [Fal88℄, Setion 4(a) Faltings gives the onstrution of natural isomorphisms of ℓ -adi étale ohomology and deRham ohomology. Remark 5.4.4 Let v∈ΣK\{0} and ℓ= har (kv) and X a proper at OKv -sheme, then we have an isomorphism Hm ´et (X⊗Kv,Qℓ)⊗Cv ∼ −→ M p+q=m Hq(X, Ωp X/OKv)⊗Cv(−q), whih preserve up produts, GKv -ation, harateristi lasses of yles and Chern lasses of vetor bund les. This remark of Faltings has imp ortant onsequenes for the onnetion b etween τ -Ho dge-Tate numb ers of an ℓ -adi representation ρℓ and Ho dge numbers of XC , related by the ation of GK on X . Illusie arries out the essential part of the exat onnetion after [Ill94℄, Theorem 3.1.2: hj,m−j(XC) = dimQℓ(Cv⊗Hm ´et (XQℓ,Qℓ)(j))GKv. Therefore we have for Vℓ:= Hm ´et (XQℓ,Qℓ)⊗Qℓ that hv,j(ρℓ) = dimQℓ(Cv(j)⊗Vℓ)GKv=h−j,m+j(XC). Corollary 5.4.5 For X:= Gm,x as in Theorem 5.4.3 endowed with a GKv -ation, orresponding to the representation ρℓ:GKv−→ GL(Gm,x) , we get the fol lowing result: hv,j(ρℓ) = (1−m≤j≤0, 0 else . Moreover ρ= (ρℓ)ℓ prime full ls regularity in the sense of Denition 7.2.1 . By Theorem 5.4.3 these τ -Ho dge-Tate numb ers oinide with the τ -Ho dge-Tate numb ers for the system of ℓ -adi representations ρm= (ρm,ℓ)ℓ prime dened in Setion 7.2. 66 6 Irreduibility of ρm 6.1 Lifting Irreduibility Let G b e a group, n∈N and ℓ a prime numb er. Lemma 6.1.1 Let ρFℓ:G−→ GLn(Fℓ) be an irreduible representation with a long unipotent element. Then the extension ρFℓ:G−→ GLn(Fℓ), g 7→ ρFℓ(g) is irreduible (i.e. ρFℓ is absolutely irreduible). Proof: We take a minimal ρFℓ -invariant subspae {0} 6=W⊆Fℓ n . We have the omp onent-wise Galois ation of GFℓ on Fℓ n , whih maps W to an orbit of subspaes. As the representation is dened over Fℓ b oth ations ommute and therefore eah subspae in the orbit is ρFℓ -invariant. By the minimality of W these are all linearly disjoint and on the other hand xed by the long unip otent element. Eah spae xed by this element inludes the eigenvetor and therefore the orbit has just one element. As W is invariant under b oth ations, there is a subspae U⊆Fn ℓ whih is ρFℓ -invariant and for whih W=Fℓ⊗U . As ρFℓ is irreduible and {0} 6=W , we have U=Fn ℓ and therefore W=Fℓ n .  The lo al ring Zℓ=x∈Qℓ|v(x)≥0 has the maximal ideal x∈Zℓ|v(x)>0 with residue eld Fℓ . In the ase of the lemma ab ove, we have hene that ρZℓ:G−→ GLn(Zℓ), g 7→ ρZℓ(g) is irreduible as well. Here we all a Zℓ[G] -mo dule irreduible if and only if it has no non-trivial Zℓ[G] -submo dules. At this p oint I want to thank Stefan Reiter and Andreas Maurishat for telling me ab out the next well-known statement. Lemma 6.1.2 If ρZℓ:G−→ GLn(Zℓ) is an irreduible representation, then the extension ρQℓ:G−→ GLn(Qℓ), g 7→ ρZℓ(g) is irreduible. Proof: Assume that ρQℓ is reduible and has the invariant subspae 06=V6=Qℓ n . Therefore the Zℓ -mo dule W:= V∩Zn ℓ is invariant and W6=Zn ℓ , b eause V6=Qℓ n . Additionally it is non-trivial as for all v∈Qℓ n there exists a λ∈Qℓ × , suh that λv ∈Zn ℓ , and we have 06=W . This is a ontradition sine ρZℓ is irreduible.  This lemma an easily b e adapted to rings and their quotient elds, esp eially to valuation rings. 67 6. Irreduibility of ρm 6.2 Serre's Results on Charaters of GQ In the next part, we restate some results of Serre. The main one is that a system of homomorphisms (θℓ:GQ−→ F× ℓ)ℓ∈L fullling some ompatibility relations is the redution of the pro dut of a nite harater and a p ower of the ylotomi harater. This will b e another ingredient for the pro of of Theorem 6.4.1. Let K b e a number eld and IK the idele group of K . For a nite set S of plaes of K , we dene a modulus m with support S as a family m:= (mv)v∈S∈NS . Then we get an op en subgroup Um for eah mo dulus m , by Um:= Q v plae of K Um,v ⊆IK , where Um,v is as follows: onneted omp onent of 1 in K× v for an innite plae v6∈ S , K× v for a nite plae v6∈ S , {x∈K× v suh that v(1 −x)≥mv} for v∈S . On the other hand, we have two asso iated algebrai groups Tm and Sm over Q with an algebrai morphism Tm//Sm (for more details see [Ser68℄, Setion 2.2.). Let E b e a number eld. It is shown in [Ser68℄ how to attah a stritly ompatible system (ψℓ:GQ−→ E× λ)λ∈ΣE\{0} of one dimensional λ -adi Q -rational Galois representations to any harater ψ:Sm−→ E×. Prop osition 6.2.1 ( [Ser72℄, Prop.20 ) Let L⊆ΣQ∪ {∞} \ {0} be an innite set and let θℓ:GQ−→ F× ℓ, ℓ ∈L, be a ol letion of homomorphisms. Assume that there exists a modulus m and j∈Z suh that for al l ℓ∈L and for al l a∈Um one has θℓ(cf−1[a]) ≡a−j ℓmod ℓ, where cf : Gab Q−→ IQ/Q× is the lass eld isomorphism and aℓ is the omponent of a at ℓ. Then there exists a number eld E and a Heke harater ψ:Sm−→ E× suh that ψλ=θℓ for innitely many ℓ∈L and λ a nite plae of E above ℓ . The following result is a onsequene of [Sh88℄, Prop osition 1.4, and Serre's theory of ab elian representations [Ser68℄: Prop osition 6.2.2 Let (ψλ:GQ−→ E× λ)λ∈ΣE\{0} be a stritly ompatible system of one dimensional λ -adi E -rational Galois representations whih are assoiated to a Heke harater ψ:Sm−→ E×. Then there exists a nite harater ǫ:GQ−→ E× and an integer k∈Z suh that ψλ=ǫ·χk ℓ, where λ|ℓ. Combining b oth results, we get the following: 68 6.3. Groups of Lie Type Corollary 6.2.3 Let L⊆ΣQ∪ {∞} \ {0} be an innite set and let θℓ:GQ−→ F× ℓ, ℓ ∈L, be a ol letion of homomorphisms. Assume that there exists a modulus m and j∈Z suh that for al l ℓ∈L and for al l a∈Um one has θℓ(cf−1[a]) ≡a−j ℓmod ℓ, where cf : Gab Q−→ IQ/Q× is the lass eld isomorphism and aℓ is the omponent of a at ℓ. Then there exists a number eld E , a nite harater ǫ:GQ−→ E× and an integer k∈Z suh that ǫ·χk ℓ=θℓ for innitely many ℓ∈L. 6.3 Groups of Lie Typ e In the year 1972 Gorenstein announed a program for the omplete lassiation of nite simple groups. Many mathematiians worked on it and the last gap was lled 2004 by Ashbaher and Smith (f. [As04℄). Beside the well-known abelian nite simple groups, i.e. yli groups of prime order, there are the following p ossibilities:  the alternating groups An ( n≥5 ),  the nite lassial groups - that is, the linear, sympleti, unitary and orthogonal groups of nite vetor spaes,  the exeptional groups of Lie type,  the 26 sp oradi groups (see [Gor85℄). This shows that groups of Lie typ e play an important role in the understanding of groups and their subgroup struture. Let K b e a eld of harateristi p , q a p -p ower and n∈N , then we have the homomorphism Frobq: GLn(K)−→ GLn(K),(aij)7→ (aq ij) . If K is algebraially losed and G is a linear algebrai group over K a standard Frobenius F:G−→ G is a map suh that there exists an n∈N , an inlusion ι:G//GLn(K) and a p ower q of p suh that for all g∈G we have ι(F(g)) = Frobq(ι(g)). A homomorphism F:G−→ G is alled a Frobenius morphism , if some p ower of F is a standard Frob enius. Denition 6.3.1 Let G be a onneted redutive algebrai group over Fℓ and let F:G−→ G be a Frobenius morphism. The nite group of xed points GF is al led group of Lie typ e and sometimes also its ommutator subgroup (GF)′:= [GF, GF] and its entral quotient G/Z(G) , where Z(G) is the enter. 69 6. Irreduibility of ρm In this setting, a maximal losed onneted solvable algebrai subgroup is alled Borel subgroup and an algebrai subgroup whih ontains a Borel subgroup is alled paraboli . Remark 6.3.2 a) The standard Frobenius morphism Frobq:Fq−→ Fq, α 7→ αq , gives rise to the so al led Chevalley groups (untwisted groups of Lie type), whereas the produt of some Frobq with other automorphisms leads to twisted groups of Lie type (e.g. SUn(Fq2) = SLn(Fq)δ◦Frobq , where δ is the inverse transpose map). b) Eah group of Lie type is the quotient of Fq -points G(Fq) of an algebrai group sheme over Fq . (For the derived group of an orthogonal group, we take the spin group, f. [Wil09 ℄). Denition and remark 6.3.3 To any onneted Dynkin diagram there is an assoiated simple algebrai group G over Fq (with the exeptions A1(F2), A1(F3), A2(F3), B2(F2) and G2(F2) ). Cartan typ e Chevalley group Dynkin diagram Al(Fq) SLl+1,PGLl+1 Bl(Fq) SO2l+1 Cl(Fq) Sp2l Dl(Fq) SO2l E6(Fq) E7(Fq) E8(Fq) F4(Fq) G2(Fq) For the lassiation it is neessary to take a Dynkin diagram together with an automorphism of the graph, whih then is in orrespondene to simple group of Lie type. This leads to the fol lowing types 2Al(Fq2),2B2(F22n+1 ),2Dl(Fq2),3D4(Fq3),2E6(Fq2),2F4(F22n+1 ),2G2(F32n+1 ) . 70 6.3. Groups of Lie Type Next we will need the struture of the maximal subgroups of the ommutator subgroup Ωn(Fq) := [SOn(Fq),SOn(Fq)] for n∈N o dd and q an o dd prime p ower. In [KL90℄, we have the following denition of sets S,C of maximal subgroups of Ωn(Fq) : Denition of S A maximal subgroup H of Ωn(Fq) lies in S:= S(Ωn(Fq)) if and only if the following holds. a) The so le S of H , that is the subgroup generated by the minimal non-trivial normal subgroups of H , is a non-ab elian simple group - i.e. H has a unique minimal non-trivial normal subgroup, whih is non-ab elian and simple. b) If L is the full overing group of S , and if ρ:L−→ GL(V) is a representation of L suh that ρ(L) = S , then ρ is absolutely irreduible. ) ρ(L) annot b e realized over a prop er subeld of Fq . (rough) Denition of C For the subgroup Ωn(Fq) of SOn(Fq) we dene Ci(Ωn(Fq)) := {C∩Ωn(Fq)|C∈ Ci(SOn(Fq))} for i= 1,...,8 , and let C:= C(Ωn(Fq)) := 8 [ i=1 Ci(Ωn(Fq)). C1: stabilizers of totally singular or non-singular subspaes C2: stabilizers of deomp ositions V= t L j=1 Vj with tdimFqVj= dimFqV=n C3: stabilizers of extension elds of Fq of prime index C4: stabilizers of tensor pro dut deomp ositions V=V1⊗V2 C5: stabilizers of subelds of Fq of prime index C6: normalizers of sympleti-typ e r -groups ( r prime) in absolutely irreduible representations C7: stabilizers of deomp ositions V= t N j=1 Vj with (dimFqVj)t= dimFqV=n C8: lassial subgroups 71 6. Irreduibility of ρm The [KL90℄, Main Theorem (C) and [KL90℄, Table 3.5.D reveals the struture of Ωn(Fq) for n, q o dd: Theorem 6.3.4 ( [KL90℄, Main Theorem (C) ) Assume that n > 12 is odd an q an odd prime power. For a member H∈ C , the preise onditions under whih H is maximal in Ωn(Fq) are determined by the fol lowing table. Moreover, this table also determines the set of overgroups of H lying in C ∪S . Ci typ e onditions C1Pm1≤m≤n−1 2 Om(Fq)⊥Oǫ n−m(Fq) 1 ≤m < n , m o dd, ǫ=± C2Om(Fq)≀Stn=mt, m, t ≥2 O1(Fq)≀Snq prime C3On r(Fqr)r|n , r prime, r6=n C4Om(Fq)⊗On m(Fq)m|n , m < √n C5On(Fq0)q=qr 0 , r prime C6 do es not o ur C7Om(Fq)≀Stn=mt , (q, m)6= (3,3) C8 do es not o ur Table 6.1: The maximal subgroups of Ωn(Fq) for n, q o dd ([KL90℄, Table 3.5.D) Here Pm is a stabilizer of an m -dimensional total ly singular spae, i.e. a paraboli subgroup, Om and Oǫ m are orthogonal groups respeting some symmetri bilinear form and Sm is a symmetri group. Lemma 6.3.5 Let n∈N be a xed odd integer. For almost al l prime numbers ℓ let G(Fq)⊆SOn(Fℓ) be a group of Lie type for q a power of ℓ ontaining a long unipotent element, i.e. an element of Jordan normal form Jn(1) , and one non-trivial unipotent element with dierent Jordan normal form. Then for almost al l ℓ as above, we have the inlusion Ωn(Fℓ)⊆G(Fq) if n6= 7 and G2(Fℓ)⊆G(Fq) if n= 7 . Proof: There exists a simply onneted group of Lie typ e G(Fq) and an epimorphism G(Fq)////G(Fq) . By [Ste63℄ we have a morphism L(λ) : G(Fq)−→ Ωn(Fq)⊆GLn(Fq) of Fq -p oints of algebrai group shemes on Fq dened over Fq (where λ denotes the highest weight of the representation), if ℓ is large ompared to n . The long unip otent element u is asso iated to a ertain ro ot α in the ro ot system of G . The transp osed element ut is then asso iated to the negative ro ot −α of G . Then the group H=hu, uti is an irreduible subgroup of G of type A1 , again if ℓ is large enough. 72 6.4. Irreduibility of the mo dℓ Representation Now the existene of the non-trivial unip otent element with Jordan normal form dierent from Jn(1) implies that H is prop erly ontained in G(Fq) and that G is of dierent type then A1 (if ℓ is large enough). By Steinberg's Tensor Pro dut Theorem (f. [MT11℄, Theorem 29.6), if ℓ is large ompared to n any irreduible representation of G is given by a highest weight representation and hene was indued by a morphism of onneted group shemes over Fq . Base hange of L(λ) to Fq denes a morphism of algebrai groups and hene a representation L(λ)⊗Fq:G(Fq)−→ GLn(Fq) , whih fators over SOn(Fq) . By [SS97℄, Theorem B, any algebrai group ontaining a long unip otent element inside an underlying general linear group GL(W) dierent from GL(W) is either of typ e A1,SO(W),Sp(W), G2 or B3 . As G(Fq) denes G uniquely if the harateristi ℓ and therefore q is large enough and sine there are no non-trivial twists of o dd dimensional orthogonal groups, this implies that G lies b etween Ωn(Fℓ) and SOn(Fℓ) if n6= 7 . In the hase n= 7 , hene we have G2(Fℓ)⊆G(Fq) .  6.4 Irreduibility of the mo dℓ Representation In this setion we have K=Q , m∈N0 even and ℓ a prime numb er. Then we get the lisse Qℓ -sheaf i∗Hm,ℓ of rank m+ 1 for i:A1 Q\ {0,1}//A1 Q and Hm,ℓ like in Setion 3.3. By xing an s∈A1 Q\ {0,1} this orresp onds to a ontinuous representation of rank m+ 1 ρi∗Hm,ℓ :π1 ´et(A1 K\{0,1})−→ GL((Hm,ℓ)s) , whih fators through Zℓ and resp ets a symmetri bilinear form. This representation an b e tensored with the determinant as in Theorem 3.3.2, to obtain a ontinuous representation ρi∗Hm,ℓ ⊗det(ρi∗Hm,ℓ ) : π1 ´et(A1 Q\{0,1})−→ GL((Hm,ℓ)s), fatoring over SL and respeting a symmetri bilinear form. This representation is of weight m by Delignes' work (f. Theorem 2.3.4), i.e. maps Frobq to q−m 2 . Then we get a weight 0 representation by tensoring with the m 2 th p ower of the ylotomi harater χℓ . Now we will show that the redution mo d ℓ dρm,ℓ := ρi∗Hm,ℓ ⊗det(ρi∗Hm,ℓ )⊗χm 2 ℓ:π1 ´et(A1 Q\{0,1})−→ SOm+1(Fℓ) is irreduible for almost all ℓ . This will lead us in Setion 7.2 to the fat that ρm,ℓ is irreduible as a weakly ompatible system of Galois representations of Q . For any x∈A1 Q\{0,1} , we get the sp eialization map ιx:π´et 1({x})∼ =GQ//π´et 1(A1 Q\{0,1}) as dened in Setion 2.2. 73 7. Potential Automorphy of Sp eializations where pv is the harateristi of the residue eld kv as b efore. For v∈S the denition of Lv(π, s) is not so straightforward, but an b e obtained by the lo al Langlands orrespondene [HT01℄ and [Hen00℄. It is known that these analyti L -funtions satisfy favorable prop erties like meromorphi (mostly even holomorphi) ontinuation to the whole omplex plane and fulll funtional equations et.. On the other hand, for any irreduible, weakly ompatible system ρ= (ρℓ)ℓ prime of Galois representations ρℓ:GK−→ GL(Vℓ) (see Denition 2.1.4), we an also dene the restrited L -funtion of ρ by L(ρ, s) := Y v∈S Lv(ρ, s)·Y v∈ΣK\S det(1−p−s vFrobv,ρℓ)−1:= Y v∈S p−snv v˜ fv,ρℓ(ps v)·Y v∈ΣK\S psnv vfv,ρℓ(ps v)−1, where n is the rank for the representation and fv,ρℓ(x)∈Q[x]⊂Qℓ[x] the harateristi p olynomial of the Frob enius (see page 22) for v|ℓ . Stritly sp eaking Frobv,ρℓ is not well-dened and should b e seen as symbol for the very last pro dut on the right. In the ase v∈S , we have a preimage of the Frob enius in the deomp osition group for eah element of the inertia group Iv:= lim ←− Iw (for the system w|v ). The ation of these preimages via the Galois representation ρℓ is unique up to onjugation on the spae Vρℓ(Iv) ℓ xed by ρℓ(Iv) . Therefore the harateristi p olynomial ˜ fv,ρℓ(x) := det(1 ·x−ρℓ(Frobv)|Vρℓ(Iv) ℓ )∈Qℓ[x] is well-dened and an b e interpreted as a omplex p olynomial by ho osing an emb edding of elds ιℓ:Qℓ//C . We dene Lv(ρ, s) := det(1 −p−s vρℓ(Frobv)|Vρℓ(Iv) ℓ ) := p−snv v˜ fv,ρℓ(ps v), where nv is the dimension of Vρℓ(Iv) ℓ . Denition 7.1.1 Let ρ= (ρℓ)ℓ prime be an irreduible, weakly ompatible system of ℓ -adi representations of a number eld K , then ρ is al led automorphi , if there exists an irreduible omplex subrepresentation π of GLn(AK) , suh that L(ρ, s) = L(π, s). In this ase, we say L(ρ, s) is automorphi as wel l. If there is a nite Galois extension L|K suh that the restrition (ρℓ|GL)ℓ prime is automorphi, ρ is al led p otentially automorphi . An imp ortant onjeture whih is a part of the famous Langlands program [Lan79℄, states that any irreduible uspidal system of Galois representations is automorphi. 7.2 Mo dular Lifting Due to Barnet-Lamb, Gee, Geraghty and Taylor (f. [BLGGT10℄) one has a riterion for automorphy for whih we will have to intro due some more notation. 80 7.2. Mo dular Lifting For a numb er eld K , we have the maximal totally real subeld K+ . The innite plaes of K+ orresp ond to the emb eddings K+//R . If we x suh an innite plae v , we get an embedding of GR={1, c}//GK+ . The image of the omplex onjugation c will b e denoted by cv . Denition 7.2.1 Let K be a number eld and (ρℓ)ℓ prime a weakly ompatible system of Galois representations for K of rank n (see Denition 2.1.4 ). The system (ρℓ)ℓ prime is al led a) totally o dd, essentially onjugate self-dual in the ase K is total ly real or CM, if a weakly ompatible system (εℓ)ℓ prime of Galois representations for K+ of rank one with the fol lowing property exists. For al l prime numbers ℓ there is a non-degenerate, symmetri pairing h·,·iℓ on Qℓ n , suh that for al l σ∈GK and x, y ∈Qℓ n we have hρℓ(σ)x, ρℓ(cℓσcℓ)yiℓ=εℓ(σ)hx, yiℓ. b) regular , if for eah τ:K//Q we have n distint τ -Hodge-Tate numbers (with multipliity one). A p owerful to ol for proving the p otential automorphy of systems of ℓ -adi Galois representations is the following theorem. Theorem 7.2.2 ( [BLGGT10℄, Thm.5.3.1 ) Suppose that K is a CM (or total ly real) eld and that (ρℓ)ℓ prime is an irreduible, total ly odd, essential ly onjugate self-dual, regular, weakly ompatible system of ℓ -adi representations of K . Then there is a nite, CM (or total ly real), Galois extension L|K suh that the restrition of (ρℓ)ℓ prime to GL is automorphi. For m∈N0 and a prime numb er ℓ the Qℓ -sheaf Hm,ℓ on A1 K onstruted in 3.3 (page 47), we get a lisse Qℓ -sheaf i∗Hm,ℓ on A1 K\{0,1} by pulling bak along the inlusion i:A1 K\{0,1}//A1 K . It is by denition the tensor pro dut of a lisse Zℓ -sheaf and Qℓ over Zℓ . This orresp onds by Corollary 2.2.12 to a ontinuous representation ρi∗Hm,ℓ :π´et 1(A1 K\{0,1})−→ GL(W) on the stalk W of i∗Hm,ℓ at a hosen p oint in A1 K\{0,1} , whih fators through Zℓ . The tensor pro dut of this representation with the one dimensional representation det(ρi∗Hm,ℓ ) : π´et 1(A1 K\{0,1})−→ {±1} ⊂ Qℓ ×, whih is obtained by taking the omp osition of the representation and the determinant fators as well through SL(W) . By a hoie of x∈A1 K\{0,1} aording to the restritions of Theorem 6.4.1 81 7. Potential Automorphy of Sp eializations and sp eializing ρi∗Hm,ℓ ⊗det(ρi∗Hm,ℓ ) to x (see page 29) ρi∗Hm,ℓ ⊗det(ρi∗Hm,ℓ )◦ιx:GK ιx //π´et 1(A1 K\{0,1})−→ GL(W), we get an ℓ -adi Galois representation of K for eah prime numb er ℓ . Out of these maps we want to onstrut a weakly ompatible system by semi-simpliation (see page 39) of this system. Lemma 7.2.3 For m∈N0 , K=Q and x∈A1 Q\{0,1} , the system ρm= (ρm,ℓ)ℓ prime := ((ρι∗Hm,ℓ ⊗det(ρι∗Hm,ℓ )) ◦ιx)ss ℓ prime is a weakly ompatible system of ℓ -adi Galois representations, whih respets an orthogonal respetively sympleti form if m is even respetively odd. Let m= 6 or m even and m≥12 for x∈A1 Q\ {0,1} , suh that there exist odd prime numbers p, q satisfying νp(x)<0 but ℓ∤νp(x) and νq(x−1) >0 but ℓ∤νq(x−1) (see Theorem 6.4.1 ), this system is irreduible. Proof: By Setion 3.3 we get representations ρm,ℓ :GK−→ SOm+1(Qℓ) for m even and ρm,ℓ :GK−→ Spm+1(Qℓ) for m o dd. The rationality and ompatibility of the system ρm is a diret onsequene of [Kat96℄, Theorem 5.5.4. From Corollary 5.3.1 and Corollary 5.3.2 we have that i∗Hm,ℓ ∼ =1 2(1 −σ) ker Rm(φX)∗Qℓ→Rm(φD)∗Qℓ. We have go o d redution for almost all ℓ . The derived funtors are rystalline as ohomology and therefore the kernel is rystalline as this ommutes with morphisms. As the pro jetor 1 2(1 −σ) is algebrai this is rystalline. The laim follows from the rystalline omparison isomorphism sine for ℓ large enough, the bre W is smo oth over Zℓ if ℓ is large enough (f. [Fal89℄). Hene for v∈ΣK\{0} and ℓ equal to the harateristi kv , the representation ρℓ is deRham in v and for almost all v even rystalline. By Corollary 5.4.5 we see that the τ -Ho dge-Tate numb ers are independent of ℓ for any emb edding τ:K//Q . Combining the results of this hapter, we see that for every even m∈N0 the set of prime numb ers for whih ρm,ℓ is reduible is nite. Therefore the Dirihlet density of the other prime numb ers is 1 and the system ρm= (ρm,ℓ)ℓ prime of ℓ -adi Galois representations, dened in 7.2, is irreduible in the sense of Denition 2.1.4.  82 Theorem 7.2.4 For m= 6 or m∈N0 even, m≥12 and K=Q the irreduible, weakly ompatible system ρm= (ρm,ℓ)ℓ prime of Galois representations is potential ly automorphi. Proof: Q is a totally real eld and has exatly one innite plae ∞ , the usual absolute value, with c∞∈GQ the usual omplex onjugation. We dene (εℓ)ℓ prime as the trivial system, whih is weakly ompatible. As b efore, we have im (ρm,ℓ)⊆SOm+1(Qℓ) . The pairing hx, yi∞:= xtρm,ℓ(c∞)y for x, y ∈Qℓ m+1 is non-degenerate, symmetri and satises ρm,ℓ(c∞)t=ρm,ℓ(c∞)−1=ρm,ℓ(c−1 ∞) = ρm,ℓ(c∞). For σ∈GQ , we have hρm,ℓ(σ)x, ρm,ℓ(c∞σc∞)yi∞=xtρm,ℓ(σ)tρm,ℓ(c∞)ρm,ℓ(c∞) |{z } =1 ρm,ℓ(σ)ρm,ℓ(c∞)y =xtρm,ℓ(c∞)y= 1 ·hx, yi∞ and ρm is totally o dd, essentially onjugate self-dual. The regularity is a diret onsequene of Corollary 5.4.5. This shows that for m even we are in the situation of Theorem 7.2.2 and proves that ρm is p otentially automorphi.  83 Bibliography [And04℄ Yves André. Une introdution aux motifs . Panoramas et synthèses 17. So iété Mathématique de Frane (2004) [Art62℄ Mihael Artin. Grothendiek topologies : notes on a seminar . Harvard University, Department of Mathematis (1962) [As04℄ Mihael Ashbaher. The Status of the Classiation of the Finite Simple Groups . Noties of the Amerian Mathematial So iety 51 (2004), 736740 [BBG + 03℄ Daniel Bump, Joseph [Hrsg.℄ Bernstein, Stephen S. [Hrsg.℄ Gelbart, James W. Cogdell, Ehud DeShalit, D. Gaitsgory, Emmanuel Kowalski and Stephen S. Kudla, editors. An introdution to the Langlands program . Birkhäuser (2003) [BCDT01℄ Christophe Breuil, Brian Conrad, Fred Diamond and Rihard Taylor. On the modularity of el lipti urves over Q: wild 3-adi exerises . Journal of the Amerian Mathematial So iety 14 (2001), 843939 [Be91℄ Sybilla Bekmann. On extensions of number elds obtained by speializing branhed overings . Journal für die reine und angewandte Mathematik (Crelles series) 419 (1991), 2753 [Ber97℄ Pierre Berthelot. Finitude et pureté ohomologique en ohomologie rigide . Inventiones Mathematiae 128 (1997), 329377 [BLGGT10℄ Thomas Barnet-Lamb, Toby Gee, David Geraghty and Rihard Taylor. Potential automorphy and hange of weight . ArXiv e-prints (2010) [Bum97℄ Daniel Bump. Automorphi forms and representations . Cambridge studies in advaned mathematis 55. Cambridge University Press (1997) [Car85℄ Roger W. Carter. Finite groups of Lie type: onjugay lasses and omplex haraters . A Wiley library edition. Wiley (1985) [Chê04℄ Gabriel Chênevert. Some remarks on Frobenius and Lefshetz in étale ohomology . ReCALL (2004) [Del70℄ Pierre Deligne. Equations diérentiel les à points singuliers réguliers . Leture Notes in Mathematis 163. Springer (1970) [Del74℄ Pierre Deligne. La onjeture de Weil : I . Publiations Mathématiques de l'IHÉS 43 (1974), 273307 85 [Del80℄ Pierre Deligne. La onjeture de Weil : II . Publiations Mathématiques de l'IHÉS 52 (1980), 137252 [Del87℄ Pierre Deligne. Un théorème de nitude pour la monodromie . Progress in mathematis 67. Birkhäuser (1987), 119 [DR99℄ Mihael Dettweiler and Stefan Reiter. An algorithm of Katz and its appliation to the inverse Galois problem . Journal of Symb oli Computation 30 (1999), 761798 [DR10℄ Mihael Dettweiler and Stefan Reiter. Rigid loal systems and motives of type G2 With an appendix by Mihael Dettweiler and Niholas M. Katz . Comp ositio Mathematia 146 (2010), 929963 [EP05℄ Antonio J. Engler and Alexander Prestel. Valued elds . Springer Monographs in Mathematis. Springer (2005) [ESV92℄ Hélène Esnault, Vadim Shehtman and Ekart Viehweg. Cohomology of loal systems on the omplement of hyperplanes . Inventiones Mathematiae 109 (1992), 557561 [Fal88℄ Gerd Faltings. p -adi Hodge Theory . Journal of the Amerian Mathematial So iety 1 (1988), 255299 [Fal89℄ Gerd Faltings. Crystal line ohomology and p -adi Galois representations . In  Algebrai Analysis, Geometry and Numb er Theory (J.I.Igusa ed), John Hopkins Universtity Press, Baltimore (1989), 2580 [FK88℄ Eb erhard Freitag and Reinhardt Kiehl. Étale ohomology and the Weil onjeture . Ergebnisse der Mathematik und ihrer Grenzgebiete Folge 3 Band 13. Springer (1988) [FO08℄ Jean-Mar Fontaine and Yi Ouyang. Theory of p-adi Galois Representations . preprint (2008) [Fon02℄ Jean-Mar Fontaine. Analyse p -adiques et Représentations Galoisiennes . Pro eedings of the International Congress of Mathematiians 2 (2002), 139148 [GMHK10℄ Rob ert M. Guralnik, Mihael Mihael Harris and Niholas M. Katz. Automorphi realization of residual Galois representations . Journal of the Europ ean Mathematial So iety 12 (2010), 915937 [Gor85℄ Daniel Gorenstein. Finite simple groups . The University Series in Mathematis. Plenum Press (1985) [GR05℄ Alexander Grothendiek and Mihèle Raynaud. Revêtements étales et groupe fondamental (SGA 1) . So ieté Mathematique de Frane 3 (2005) [Gro98℄ Benedit H. Gross. On the Satake isomorphism . London Mathematial So iety Leture Note 254. Cambridge University Press (1998), 223238 86 [Har06℄ Robin Hartshorne. Algebrai geometry . Graduate texts in mathematis 52. Springer (2006) [Hen00℄ Guy Henniart. Une preuve simple des onjetures de Langlands pour GL(n) sur un orps p-adique . Inventiones Mathematiae 139 (2000), 439455 [HT01℄ Mihael Harris and Rihard Taylor. The geometry and ohomology of some simple Shimura varieties . Annals of Mathematis Studies 151. Prineton University Press (2001), 276 [Ill94℄ Lu Illusie. Crystal line Cohomology . Pro eedings of Symp osia in Pure Mathematis 55. Amerian Mathematial Soiety (1994), 4370 [Jan94℄ Uwe Jannsen. Motives . Pro eedings of Symposia in Pure Mathematis 55. Amerian Mathematial So iety (1994), 747/676 [JL70℄ Hervé Jaquet and Rob ert P. Langlands. Automorphi Forms on GL(2) . Leture Notes in Mathematis 114 (1970) [Kat90℄ Niholas M. Katz. Exponential sums and dierential equations . Prineton University Press (1990) [Kat96℄ Niholas M. Katz. Rigid loal systems . Annals of Mathematis Studies 139. Prineton University Press (1996), 223 [KL90℄ Peter Kleidman and Martin W. Lieb ek. The subgroup struture of the nite lassial groups . London Mathematial So iety Leture Note Series 129. Cambridge University Press (1990) [Kle88℄ Peter B. Kleidman. The maximal subgroups of the heval ley groups G2(q) with q odd, the Ree groups 2G2(q) , and their automorphism groups . Journal of Algebra 117 (1988), 3071 [KW03℄ Mark Kisin and Sigrid Wortmann. A Note on Artin Motives . Mathematial Researh Letters 10 (2003), 375389 [Lan79℄ Rob ert P. Langlands. Automorphi representations, Shimura varieties, and motives. Ein Märhen . Pro eedings of Symposia in Pure Mathematis 33. Amerian Mathematial So iety (1979), 205246 [LS74℄ Viente Landazuri and Gary M. Seitz. On the Minimal Degrees of Projetive Representations of the Finite Cheval ley Groups . Journal of Algebra 32 (1974), 418443 [Mil80℄ James S. Milne. Étale ohomology . Prineton Mathematial Series 33. University Press (1980) [MM99℄ Gunter Malle and Bernd H. Matzat. Inverse Galois theory . Monographs in Mathematis. Springer (1999) 87 [MT11℄ Gunter Malle and Donna M. Testerman. Linear algebrai groups and nite groups of Lie type . Cambridge Studies in Advaned Mathematis 133. Cambridge Univ. Press (2011) [Mur67℄ Jaob P. Murre. Letures on an introdution to Grothendieks theory of the fundamental group . Letures on Mathematis and Physis: Mathematis 40. Tata Institute of Fundamental Researh (1967) [MV04℄ Alex Martsinkovsky and Anatoly Vlassov. The representation rings of k[X] . preprint (2004) [Neu99℄ Jürgen Neukirh. Algebrai number theory . Die Grundlehren der mathematishen Wissenshaften in Einzeldarstellungen 322. Springer (1999) [PS08℄ Chris A. M. Peters and Joseph H. M. Steenbrink, editors. Mixed Hodge Strutures . Ergebnisse der Mathematik und ihrer Grenzgebiete. 3. Folge / A Series of Mo dern Surveys in Mathematis 52. Springer (2008) [Sh73℄ Wilfried Shmid. Variation of Hodge struture: The singularities of the period mapping . Inventiones Mathematiae 22. Springer (1973), 211319 [Sh74℄ Barbara Shiht. Verfahren höherer Ordnung zur Einshlieÿung der Inversen einer komplexen Matrix . preprint (1974) [Sh88℄ Norb ert Shappaher. Periods of Heke haraters . Leture Notes in Mathematis 1301. Springer (1988) [Ser68℄ Jean-Pierre Serre. Abelian l-adi representations and el lipti urves . MGill University Leture Notes. Benjamin (1968) [Ser72℄ Jean-Pierre Serre. Propriétés galoisiennes des points d'ordre ni des ourbes el liptiques . Inventiones Mathematiae 15 (1971/1972), 259331 [Ser94℄ Jean-Pierre Serre. Propriétés onjeturales des groupes de Galois motiviques et des représentations ℓ -adiques . Pro eedings of Symp osia in Pure Mathematis 55. Amerian Mathematial So iety (1994), 377400 [SS97℄ Jan Saxl and Gary M. Seitz. Subgroups of Algebrai Groups ontaining Regular Unipotent Elements . Journal of the London Mathematial So iety 55 (1997), 370386 [Ste63℄ Rob ert Steinberg. Representations of algebrai groups . Nagoya Mathemati Journal 22 (1963), 3356 [SV99℄ Karl Strambah and Helmut Völklein. On linearly rigid tuples . Journal für die reine und angewandte Mathematik (Crelles series) 510 (1999), 5762 [SZ85℄ Joseph Steenbrink and Steven Zuker. Variation of mixed Hodge struture I . Inventiones Mathematiae 80 (1985), 489542 88 [Wil95℄ Andrew Wiles. Modular el lipti urves and Fermat's last theorem . Annals of Mathematis 142 (1995), 443551 [Wil09℄ Rob ert Wilson. The nite simple groups . Graduate Texts in Mathematis 251. Springer (2009) [Wit37℄ Ernst Witt. Zyklishe Körper und Algebren der Charakteristik p vom Grad pn , Struktur diskret bewerteter perfekter Körper mit vol lkommenem Restklassenkörper der Charakteristik p . Journal für die reine und angewandte Mathematik (Crelles series) 176 (1937), 126140 [Wor02℄ Sigrid Wortmann. Galois representations of three-dimensional orthogonal motives . Manusripta Mathematia 109. Springer (2002), 128 Contat: E-Mail: mihael.maieriwr.uni-heidelb erg.de 89