Classi al Motives I: Motivi L-fun tions Minhyong Kim July 17, 2006 1 Variety V ; (V; s), -fun tion of V ; arithmeti invariants of V . This pi ture should be re ned by Variety V ; onstituent motives fMig ; f L(Mi; s)g, L-fun tions of the Mi ; arithmeti invariants of the Mi ; arithmeti invariants of V . 2 Example: E=Q tion ellipti urve with aÆne minimal equa- y 2 + a1xy + a3y = x3 + a2x2 + a4x + a6 S: set of primes of bad redu tion. E : proper smooth model of E over Z [1=S ℄. (E )0: set of losed points. Y 1 S (E; s) = 1 N (x) x2(E )0 s Then there is a de omposition S (E; s) = S (s)S (s 1)=LS (E; s) 3 where S (s) = and Y 1 p2 =S LS (E; s) = Y p2 =S 1 p s Lp(E; s) is the partial L-fun tion of E with fa tors de ned by 1 Lp(E; s) = 1 app s + p1 2s Here ap = p + 1 Np and Np is the number of points on E mod p. 4 Can put in bad Euler fa tors a ording to a re ipe determined by the redu tion of E at p: 8 > < 1=(1 p ss) split multipli ative; Lp(s) = > 1=(1 + p ) non-split multipli ative; : 1 additive: L(E; s) := Y p Lp(E; s) Using thus the breakdown into three fa tors, we an also omplete S (E; s) in a natural way. 5 The estimate japj 2pp implies that the Euler produ t onverges for Re(s) > 3=2. To ontrol the analyti properties, use relation to automorphi L-fun tions. In this ase, an make expli it by omputing the ondu tor NE Here := Y p2 S p fp fp = ordp(E ) + 1 mE where E is the dis riminant of E and mE is the number of omponents over F p of a Neron model of E . 6 Fa t (W, T-W, BCDT): L has an analyti ontinuation to the omplex plane. In fa t, L(E; s) = L(fE ; s) Z1 1 s 1dy = f ( iy ) y E (2)s (s) 0 for a normalized weight 2 new usp form fE of level NE whi h is an eigenve tor for the He ke operators, determined by a q expansion fp = 1 + a1q + a2q 2 + where the ap have to be the same as those for E when p 2= S . Can nd fE and then use this formula for ompute L-values. 7 Conje ture (BSD): ords=1L(E; s) = rankE (Q ) Proved if ords=1L(E; s) 1. (Kolyvagin) 8 Fun tional equation: (E; s) := (2)s (s)NEs=2L(E; s) satis es a fun tional equation (E; 2 s) = E (E; s) where E = 1 depends on the urve E . Can be omputed in a straightforward way as a produ t of lo al terms. Now, if E = 1, then learly L(E; 1) = 0 Suppose you an he k L0(E; 1) 6= 0 using the equality with L0(f; 1), then we on lude E (Q ) has rank one. Thus, analysis of the L-fun tion, in luding the fun tional equation and omputation, gives us the stru ture of E (Q ). 9 Continuation of BSD: If vanishing, then (s 1) = jSha(E )jRE is the order of r js=1 r L(E; s) Y p p= jE (Q )(tor)j2 relating L-values to many other re ned arithmeti invariants of E . General prin iple: L-fun tion en odes Diophantine invariants of E . 10 Brief dis ussion of terms. Important distin tion: Rational terms versus trans endental terms. Rational terms: Sha(E ): The Tate-Shafarevi h group of E , onje tured to be nite. Classi es lo ally trivial torsors for E . Analogous to a lass group. E (Q )(tor ) : p: ( nite) torsion subgroup of E (Q ). Tamagawa number. p = (E (Q p ) : E 0(Q p )) 11 Trans endental terms: RE : Regulator of E omputed using anonial height <; > and basis fP1; P2; : : : ; Pr g for E (Q )=E (Q )(tor). := j det(< Pi; Pj >)j Thus, RE is the ovolume of the MordellWeil group, in a manner analogous to the lassi al regulator of number elds ( ovolume of units). RE : real period = Z j! j where ! and = dx=(2y + a1x + a3) < >= H1(E (C ); Z )+ 12 The known relations between L-fun tions and arithmeti are expe ted to generalize vastly. L-fun tions de ned using Galois a tions on etale ohomology and ompleted using Hodge theory. 13 Conje tures:. (1) Hasse-Weil: analyti ontinuation and fun tional equation, addressed by Langlands' program: `Motivi L-fun tions are automorphi L-fun tions.' (2) Values: (a) Deligne generalizes dis ussion of period (in non-vanishing ase) using omparison of rational De Rham and topologi al ohomologies; (b) Beilinson-Blo h generalizes dis ussion of order of vanishing and regulator using rank and ovolume of motivi ohomology. ( ) Blo h-Kato generalizes dis ussion of rational part using Tamagawa numbers for Galois representations via p-adi Hodge theory. 14 X=Q : smooth proje tive variety. Asso iated to X is a olle tion of ohomology groups, the realizations of the motive of X . Q l ) for ea h prime l: the = Hetn (X; Q l - oeÆ ient etale ohomology of degree n. Carries a natural a tion of = Gal(Q =Q ). Hln(X ) := H n(X; :X ): the algebrai De Rham ohomology equipped with a Hodge ltration given by F iH n (X ) = H n(X; i),!H n (X ) n (X ) HDR DR for ea h i. n (X ) HB DR := H n(X (C ); Q ): the Q - oeÆ ient singular ohomology of the omplex manifold X (C ) equipped with a ontinuous a tion F1 of omplex onjugation. 15 The ompleted L-fun tion of all these stru tures. H n(X ) uses 16 Canoni al omparison isomorphisms: n (X ) HB Ql ' Hln(X ) preserving a tion of F1. n (X ) HB C n (X ) ' HDR C This isomorphism endows HBn (X ) with a rational Hodge stru ture of weight n `de ned over R .' 17 That is, we have a dire t sum de omposition n (X ) C ' H p;q (X ) HB where H p;q := F p \ Fq and F1(H p;q ) = H q;p If we denote by the omplex onjugation on C then (HBn (X ) C )F1 n = HDR R 18 At non-ar himedean pla es, there is an important analogue. For any embedding Q ,!Q l , we have DDR(Hln(X )) := (Hln(X ) n (X ) ' HDR BDR ) l Ql where l = Gal(Q l =Q l ), and BDR is Fontaine's ring of p-adi periods. 19 Regardless of its pre ise de nition, a motive M should have asso iated to it a olle tion of obje ts as above that we all a pure system of realizations that make up a ategory R. That is, this is a olle tion R(M ) = ffMlg; MDR; MB g where ea h Ml is a representation of on a ( nite-dimensional) Q l -ve tor spa e, MDR is a ltered Q -ve tor spa e, and MB is a Q ve tor spa e with an involution F1. These ve tor spa es should all have the same dimension and be equipped with a system of omparison isomorphisms as above. This data must be subje t to further onstraints having to do with lo al Galois representations. 20 Re all exa t sequen e: v ^ 0!Ip! p ! Z !0 where Ip is the inertia group and ^ ' Gal(F p=Fp): Z ^ orresponds to the geometri FrobeF rp 2 Z nius, that is, the inverse to the p-power map. For l 6= p, Ip has a tame l-quotient tl : Ip!Ip;l with the stru ture ^l (1) ' lim ln Ip;l ' Z as a module for Gal(F p=Fp). De ne Wp := v 1(Z ) the Weil group at p. p; 21 Convenient to analyze the data of Ml using an asso iated Weil-Deligne (W-D) representation W Dp(Ml ) for ea h p, onsisting of -a representation r of Wp su h that rjIp has nite image, -and a nilpotent operator representation. Np a ting on the These satisfy a ompatiblity r(p)Npr(p 1) = p 1Np for any lift p 2 Wp of F rp. 22 The onstru tion of W Dp(Ml) for p 6= l uses the fa t that the a tion of Ip when restri ted to some nite index subgroup Ip0 is unipotent, and hen e, an be expressed as 7! exp(tl()Np) for a nilpotent Np. Then the representation r is given by r(np ) = np exp( tl ( )N ) For p = l, we use the fa t that any De Rham representation is potentially semistable, and hen e, gives us a ltered (l; Nl) module via Ml 7! (Ml Bst) 0 l whi h is, in any ase, isomorphi to MDR. 23 Remarks: -The point of this onstru tion is that we an pa kage the information of the representation in a form that does not use the topology of Q l . Thereby makes natural the onne tion to omplex automorphi forms. -Creates a pre ise analogy with limit mixed Hodge stru tures. -We an de ne the Frobenius semi-simpli ation W Dp(Ml)ss of W Dp(Ml) by repla ing simple part. p with its semi- 24 Here are the onstraints we impose on our pure system of realizations: -We assume then that there exists a nite set S of primes su h that W Dp(Ml) is unrami ed for all p 2= S , i.e., Np = 0 and Ip a ts trivially. -`Algebrai ity and independen e of l': There exists a Frobenius semi-simple W-D representation W Dp(M ) over Q su h that W Dp(M ) Q l ' W Dpss(Ml ) Q l for any embedding Q ,!Q l Subje t to these onditions, the olle tion fMlg is then referred to as a strongly ompatible system of l-adi representations. 25 -`Weil onje ture': There should exist an integer n, alled the weight of M , su h that the eigenvavlues of F rp a ting on W Dp(M ) for p 2= S have all Ar himedean absolute values equal to pn=2. Furthermore, the Hodge stru ture MB should be pure of weight n. -`purity of monodromy ltration': If we denote by Mn: the unique in reasing ltration on W Dp(M ) su h that Mn k = 0, Mnk = W Dp(M ) for suÆ iently large k and N (Mnk ) Mnk 2; then the asso iated graded pie e GrkMn(W Dp(M )) has all Frobenius eigenvalues of ar himedean absolute value p(n+k)=2. 26 Remarks: -In general, need to allow oeÆ ients in E for the representations where E is a number eld and E are ompletions. Arise naturally when onsidering dire t summands or motives with oeÆ ients, e.g., abelian varieties with CM. -The bi-grading MB C ' M p;q whi h is ompatible with the omplex onjugation of oeÆ ients orresponds to a representation of the group ResCR (Gm ) 27 -Together with the a tion of F1 Æ C it an be viewed as a representation of the real Weil group with points given by WR (R ) = C [ C j where j 2 = 1 and jzj 1 = z. Here, C is the Weil operator de ned by C jM pq = iq p -It is onje tured that the realizations n (X ); H n (X )) H n(X ) = (fHln(X )g; HB DR oming from a smooth proje tive variety X satisfy the algebrai ity, independen e of l, and purity onditions even for p 2 S . 28 Category of pure motives should be omprised of obje ts in R of geometri origin, a notion with a rather pre ise interpretation. For example, need to allow duals (homology) and tensor produ ts of all obje ts onsidered. Obje ts that are not generated in an obvious way from those of the form H n(X ) arise via images (or kernels) under pull-ba ks and push-forwards in ohomology indu ed by maps of varieties, as well as Q -linear ombinations of geometri maps. Also should be able to ompose pull-ba ks with pushforwards. 29 Su h ompositions give rise to the idea of using orresponden es modulo homologi al equivalen e as morphisms. On e morphisms are onstru ted in this manner, get naturally new obje ts using the deomposition of End(H n(X )); whi h is a semi-simple Q -algebra subje t to one of the standard onje tures that numeri al equivalen e and homologi al equivalen e oin ide. 30 Can onsider a ategory of mixed systems of realizations by requiring a weight ltration Wn 1M WnM Wn+1M ompatible with all the omparisons and su h that ea h graded quotient n (M ) GrW is a pure system of realizations of weight n. Mixed motives should be those of geometri origin su h as the ohomology of varieties that are not ne essarily smooth or proper. But then, need to in lude obje ts like ( nite-dimensional quotients of) Q [1 ℄ or the ( o)-homology of ( o-)simpli ial varieties. 31 Given a pure system M of realizations we an de ne its L-fun tion L(M; s) as an Euler produ t L(M; s) = with Y p Lp(M; s) Lp(M; s) = det[(1 1 p sF rp)j(W Dp(M ))Ip=1;Np=0℄ Assume M is of weight n, then produ t onverges (and hen e is non-zero) for Re(s) > n=2 + 1: 32 Also a fa tor at 1 depending upon the representation MB C of WR . De ne s=2 R := C := 2(2) (s=2) s (s) := dimM pq hp; := dimM pp;1 hpq where the signs in the supers ript refer to the 1 eigenspa es of the F1-a tion. 33 Then L1(M; s) is de ned by Y p<q C (s p)h for odd n, and Y p<q h C (s p ) pq R (s n=2)h n= pq 2+ R (s n=2+1)h for n even. 34 2 n= It is onje tured that (M; s) has a meromorphi ontinuation to C and satis es a fun tional equation (M; s) = (M; s)(M ; 1 s) where the epsilon fa tor has the form (M; s) = bas. This onje ture should be addressed by the Langlands' program. 35 Notation: Q: trivial system of realizations. Q (1) := H 2(P1) Q (i) = Q (1) i for i 0 and Q (i) = Hom(Q ( i); Q ) for i < 0. For a system M of realizations, M (i) := M Q (i) Then for any smooth proje tive variety of dim d, we have H 2d(X ) ' Q ( d) and a perfe t pairing H i(X ) H 2d i(X )!H 2d(X ) Cup produ t with the ohomology lass of a hyperplane gives us H i(X ) ' H 2d i(X )(d i) 36 Some simple properties of twisting: M (n)l is the tensor produ t of Ml with the n-th power of the Q l y lotomi hara ter. F i(M (n)DR) = F n+iMDR with a orresponding shift in Hodge numbers hpq . F1jM (n)B = (F1jMB ) ( 1)n Finally, L(M (n); s) = L(M; s + n) 37 Hen eforward, we will fo us on the ase where M is H n(X ) for a smooth proje tive variety X of dimension d and assume that - H n(X ) is a pure system of realizations; - the analyti ontinuation and fun tional equation hold true. 38 Conje tures on orders. We have H n(X ) ' H 2d n(X )(d) ' H n(X )(n) Thus, the fun tional equation relates L(H n(X ); s) and L(H n(X )(n); 1 s) = L(H n(X ); n + 1 s) with enter of re e tion (n + 1)=2 Thus, for the most part, we an on ne interest to m (n + 1)=2 or, equivalently, n+1 m (n + 1)=2: 39 Brief reminder on two simple ase: n odd: BSD ords=1L(H 1(E ); s) = rankE (Q ) Now, an element x 2 E (Q ) gives rise to an extension in the ategory R of realizations Æ (x) 2 Ext1R(Q ; H 1(E )(1)) It is onje tured that when R is repla ed by a suitable ategory of motives, this is the only way to onstru t su h extensions. 40 n even: F=Q Galois extension and : Gal(F=Q )!Aut(V ) a nite-dimensional representation. L(; s) Artin L-fun tion. Then ords=1L(; s) = dimHomRep(Q ; V ) 41 The general onje ture is ords=n+1 mL(H n(X ); s) = dimExt1MotZ (Q ; H n(X )(m)) dimHomMotZ (Q ; H n(X )(m)) The Hom and Ext should o ur inside a onje tural ategory of mixed motives over Z with Q - oeÆ ients. For weight reasons, the Hom term vanishes unless n = 2m in whi h ase the Ext term vanishes. That is, in the pure situation we are onsidering, only one term or the other o urs. This is the prototype of the sort of statement that should hold for an arbitrary (mixed) motive. 42 So when n = 2m, this be omes ords=m+1L(H 2m (X ); s) = dimHomMotZ (Q ; H 2m(X )(m)) generalizing he pole of the Artin L-fun tion (m = 0). It is expe ted that HomMotZ (Q ; H 2m(X )(m)) ' [CH m(X )=CH m(X )0℄ Q Of ourse the isomorphism should arise via a y le map CH m(X )!H 2m(X )(m) killing the y les CH m(X )0 homologi ally equivalent to zero. 43 When n + 1 = 2m, the onje ture predi ts the order of vanishing at the entral riti al point: ords=mL(H 2m 1(X ); s) = dimExt1MotZ (Q ; H 2m 1(X )(m)) It is then onje tured that dimExt1MotZ (Q ; H 2m 1(X )(m)) ' CH m(X )0 44 Q The map from y les to extensions goes as follows: given a representative Z for a lass in CH m(X )0, we get an exa t sequen e 0!H 2m 1(X )(m)!H 2m 1(X n Z )(m) !Æ HZ2m(X )(m)!H 2m(X )(m) There is a lo al y le lass l(Z ) 2 HZ2m(X )(m) that maps to zero in H 2m(X )(m), giving rise to the desired extension: 0!H 2m 1(X )(m)!Æ 1( l(Z ))!Q !0 45 These two lassi al points, entral riti al: n+1 m = m = (n + 1)=2; n odd; and just right of it: n+1 m = n=2 + 1; even; are somewhat ex eptional. In all other ases, one expe ts n n+1 Ext1MotZ (Q ; H n(X )(m)) = HM; Z (X; Q (m)) with the last group, often referred to as motivi ohomology, de ned using K -theory : Im[(K2m n 1(X ))(m) !(K2m n 1(X ))(m) ℄ (X is a proper at regular Z -model for or Blo h's higher Chow groups Im[CH n+1(X ; 2m !CH n+1(X; 2m n 1) X) Q 1) Q ℄ Latter interpretation more popular lately. n 46 However, intrinsi interpretation in terms of the ategory of motives should be kept in mind in all onstru tions. In fa t, when m > n=2 + 1, the onje tured fun tional equation implies ords=n+1 mL(H n(X ); s) = dimExt1MHS R (R ; HBn (X )(m) R) R where the extension o urs inside the ategory of real mixed Hodge stru tures de ned over R . So the onje ture on order of vanishing follows from the onje ture that the Hodge realization fun tor indu es an isomorphism Ext1MotZ (Q ; H n(X )(m)) ' Ext1MHSR (R ; HBn (X )(m) R R R) 47 In general, onje ture should be on eptualized in two parts: (1) Relation between L fun tions and groups in ategory of motives. Ext (2) Geometri interpretation of Ext groups. Provides unity to a wide range of related issues in Diophantine geometry. 48 There is a onstru tion, onvenient in pra ti e, of the real Ext group via Deligne ohomology: n (X )(m) Ext1MHS R (R ; HB R R) ' HDn+1(XR ; R (m)) and using properties of Deligne ohomology, one an onstru t regulator maps n+1 HM; Z (X; Q (m)) !Ext1MHSR (R ; HBn (X )(m) R R) that an be studied independently of a ategory of motives. For example, an onstru t subgroups n+1 L HM; Z (X; Q (m)); that should onje turally be of full rank, and study their image. 49 Conje tures on trans endental part of values. Central riti al values (Blo h-Beilinson): s = m, n = 2m 1. We have an isomorphism 2m 1(X ) F m HDR ' [HB2m 1(X )(m 1)℄( R 1)m 1 R This is then realized as an isomorphism 2m 1(X ))℄ 1 [^top(F mHDR ^ top[[H 2m B 1(X )(m m 1 ( 1) ℄ 1)℄ R 'R 50 Choosing bases for the two Q -lines determines a period p(H 2m 1 (X )(m)) 2 R =Q Get additional trans endental ontribution by onsidering a height pairing, onje tured to be non-degenerate: CH m(X )0 CH dim(X )+1 m(X )0!R whose determinant gives us a regulator r(H 2m 1(X )(m)) 2 R =Q Re all that onje turally dm := ords=mL(H 2m 1(X ); s) = dimCH m(X )0 Q 51 As for the value then, it is onje tured that L(H 2m 1(X ); m) := slim !m(s m) dm L(H 2m 1(X ); s) = p(H 2m 1(X )(m))r(H 2m 1 (X )(m)) in R =Q . 52 Values at n + 1 m < n=2. Note that this is equivalent to m > n=2 + 1; the region of onvergen e for the Euler produ t. Hen e, we are skipping the lassi ally interesting ase of m = n=2 + 1 (n + 1 m = n=2) for n even. Instead of a period isomorphism, there is then an exa t sequen e: m 1 ( 1) n m n R℄ 0!F H (X ) R ![H (X )(m 1) B DR !Ext1MHSR (R ; HBn (X )(m) R R) !0 53 Thus, the trans endental part should in orporate a Q -stru ture on n (X )(m) Ext1MHS R (R ; HB R R) oming from the onje tured isomorphism n+1 HM; Z (X; Q (m))) R ' Ext1MHSR (R ; HBn (X )(m) R R) Assuming this, we are led to a trivialization n (X )℄ 1 ^top([H n (X )(m 1)℄( 1)m 1 ) [^topF mHDR B n+1 1 ( X; Q ( m )))℄ [^top(HM; Z R 'R 54 Thus, hoosing bases for the three Q -lines determines a number (H n(X )(m)) 2 R =Q : Beilinson's onje ture is that L(H n(X ); n + 1 m) = (H n(X )(m)): 55 For the value at m = n=2 (n + 1 m = n=2 + 1), the regulator in orporates maps both from motivi ohomology and n+1 HM; Z (X; Q (m + 1)) CH m(X )0: 56 In the Blo h-Kato onje ture isomorphisms are normalized more arefully, omparing ertain integral stru tures one prime at a time. Thereby, it onstru ts a lift of (H n(X )(m)) to R and interprets q (H n(X )(m)) := L(H n(X ); n + 1 m)= (H n(X )(m)) in terms of arithmeti invariants arising from Galois ohomology. 57 Extra tion of the rational part is supposed to lead eventually to a p-adi L-fun tion L(p)(H n(X )) that exer ises ontrol over Galois ohomology (i.e., Selmer groups) and Diophantine invariants. Best strategy so far for `dire t appli ation' of L-fun tions to the elu idation of Diophantine stru tures. 58 Warning: onspi uous de ien y in theory of motives: Even in the best of possible worlds, only abelian invariants are a essible, su h as CH m(X ): Does not yield information about X (Q ) unless X is an abelian variety. In fa t, theory of motives is impli itly modelled after the theory of abelian varieties and H1. 59 Attempts to address this de ien y for ertain varieties are ontained in Grothendie k's anabelian program that on erns itself with the theory of pro- nite 1's: Also an interesting role for the intermediate theory of motivi fundamental groups, where Ext groups are repla ed by lassifying spa es for non-abelian torsors. 60