Zeta functions of regular arithmetic schemes at s = 0 Baptiste Morin
To cite this version:
Baptiste Morin. Zeta functions of regular arithmetic schemes at s = 0. Duke Mathematical Journal, Duke University Press, 2014, 163 (7), pp.1263-1336. 10.1215/00127094-2681387. hal-02484903
HAL Id: hal-02484903 https://hal.archives-ouvertes.fr/hal-02484903 Submitted on 19 Feb 2020
HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non, lished or not. The documents may come from émanant des établissements d’enseignement et de teaching and research institutions in France or recherche français ou étrangers, des laboratoires abroad, or from public or private research centers. publics ou privés. ZETA FUNCTIONS OF REGULAR ARITHMETIC SCHEMES AT s =0
BAPTISTE MORIN
Abstract. In [35] Lichtenbaum conjectured the existence of a Weil-étale cohomology in or- der to describe the vanishing order and the special value of the Zeta function of an arithmetic scheme at s =0in terms of Euler-Poincaré characteristics. Assuming the (conjectured) X finite generation of some étale motivic cohomology groups we construct such a cohomology theory for regular schemes proper over Spec(Z).Inparticular,weobtain(unconditionally) the right Weil-étale cohomology for geometrically cellular schemes over number rings. We state a conjecture expressing the vanishing order and the special value up to sign of the Zeta function ⇣( ,s) at s =0in terms of a perfect complex of abelian groups R W,c( , Z).Then X X we relate this conjecture to Soulé’s conjecture and to the Tamagawa number conjecture of Bloch-Kato, and deduce its validity in simple cases.
Mathematics Subject Classification (2010): 14F20 14G10 11S40 11G40 19F27 · · · · Keywords: Weil-étale cohomology, special values of Zeta functions, motivic cohomology, regulators
1. Introduction In [35] Lichtenbaum conjectured the existence of a Weil-étale cohomology in order to de- scribe the vanishing order and the special value of the Zeta function of an arithmetic scheme at s =0in terms of Euler-Poincaré characteristics. More precisely, we have the following Conjecture 1.1. (Lichtenbaum) On the category of separated schemes of finite type X! Spec(Z),thereexistsacohomologytheorygivenbyabeliangroupsHi ( , Z) and real vector W,c X spaces Hi ( , R˜) and Hi ( , R˜) such that the following holds. W X W,c X (1) The groups Hi ( , Z) are finitely generated and zero for i large. W,c X (2) The natural map from Z to R˜-coefficients induces isomorphisms i i H ( , Z) R H ( , R˜). W,c X ⌦ ' W,c X (3) There exists a canonical class ✓ H1 ( , R˜) such that cup-product with ✓ turns the 2 W X sequence ✓ i ✓ i+1 ✓ ... [ H ( , R˜) [ H ( , R˜) [ ... ! W,c X ! W,c X ! into a bounded acyclic complex of finite dimensional vector spaces.
B. Morin CNRS Université Paul Sabatier, Institut de Mathématiques de Toulouse, 118 route de Narbonne, 31062 Toulouse cedex 9, France e-mail: [email protected]
1 2BAPTISTEMORIN
(4) The vanishing order of the zeta function ⇣( ,s) at s =0is given by the formula X i i ords=0⇣( ,s)= ( 1) i rank H ( , Z) X · · Z W,c X i 0 X (5) The leading coefficient ⇣ ( , 0) in the Taylor expansion of ⇣( ,s) at s =0is given ⇤ X X up to sign by 1 i ( 1)i Z (⇣⇤( , 0) )= det H ( , Z) · X Z W,c X i Z O2 i ⇠ i ( 1) where : R ( i Z detZHW,c( , Z) ) R is induced by (2) and (3). ! 2 X ⌦ Such a cohomology theoryN for smooth varieties over finite fields was defined in [34], but similar attempts to construct such cohomology groups for flat arithmetic schemes failed. In [35] Lichtenbaum gave the first construction for number rings. He defined a Weil-étale topology which bears the same relation to the usual étale topology as the Weil group does to the Galois group. Under a vanishing statement, he was able to show that his cohomology miraculously yields the value of Dedekind zeta functions at s =0. But this cohomology with coefficients in Z was then shown by Flach to be infinitely generated hence non-vanishing in even degrees i 4 [13]. Consequently, Lichtenbaum’s complex computing the cohomology with Z-coefficients needs to be artificially truncated in the case of number rings, and is not helpful for flat schemes of dimension greater than 1. However, the Weil-étale topology yields the expected cohomology with R˜-coefficients, and this fact extends to higher dimensional arithmetic schemes [14]. The first goal of this paper is to define the right Weil-étale cohomology with Z-coefficients for arithmetic schemes satisfying the following conjecture 1.2 (see Theorem 1.3 below), in order to state a precise version of Conjecture 1.1. We consider a regular scheme proper X over Spec(Z) of pure Krull dimension d, and we denote by Z(d) Bloch’s cycle complex. i Conjecture 1.2. (Lichtenbaum) The étale motivic cohomology groups H ( et, Z(d)) are finitely X generated for 0 i 2d. Using some purity for the cycle complex Z(d) on the étale site (which is the key result of [20]) we construct in the appendix a class (Z) (see Definition 5.9) of separated schemes of L finite type over Spec(Z) satisfying Conjecture 1.2. This class (Z) contains any geometrically L cellular scheme over a number ring (see Definition 5.13), and includes the class A(Fq) of smooth projective varieties over Fq which can be constructed out of products of smooth projective curves by union, base extension, blow-ups and quasi-direct summands in the category of Chow motives (A(Fq) was introduced by Soulé in [43]). In order to state our first main result, we need to fix some notation. For a scheme X separated and of finite type over Spec(Z), we consider the quotient topological space := X1 (C)/GR where (C) is endowed with the complex topology and we set := ( , ). X X X X X1 We denote by the Artin-Verdier étale topos of which comes with a closed embedding X et X u : Sh( ) et in the sense of topos theory, where Sh( ) is the category of sheaves 1 X1 ! X X1 on the topological space (see [14] Section 4). The Weil-étale topos over the archimedean X1 place ,W is associated to the trivial action of the topological group R on the topological X1 space (see [14] Section 6). Our first main result is the following X1 Theorem 1.3. Let be a regular scheme proper over Spec(Z).Assumethat (Z),or X X2L more generally that the connected components of satisfy Conjecture 1.2. Then there exist X complexes R W ( , Z) and R W,c( , Z) such that the following properties hold. X X ZETA FUNCTIONS AT s =0 3
If has pure dimension d,thereisanexacttriangle • X (1) RHom(⌧ 0R ( , Q(d)), Q[ 2d 2]) R ( et, Z) R W ( , Z). X ! X ! X The complex R W ( , Z) is contravariantly functorial. • X There exists a unique morphism i⇤ : R W ( , Z) R ( ,W , Z) which renders the • 1 X ! X1 following square commutative.
R ( et, Z) / R W ( , Z) X X
u⇤ i⇤ 1 1 ✏ ✏ R ( , Z) / R ( ,W , Z) X1 X1 We define R W,c( , Z) so that there is an exact triangle X i⇤ R W,c( , Z) R W ( , Z) 1 R ( ,W , Z). X ! X ! X1 The cohomology groups Hi ( , Z) and Hi ( , Z) are finitely generated for all i and • W X W,c X zero for i large. The cohomology groups Hi ( , Z) form an integral model for l-adic cohomology: for • W X any prime number l and any i Z there is a canonical isomorphism 2 i i H ( , Z) Zl H ( et, Zl). W X ⌦ ' X If has characteristic p then there is a canonical isomorphism in the derived category • X R ( W , Z) ⇠ R W ( , Z) X ! X where R ( W , Z) is the cohomology of the Weil-étale topos [34] and R W ( , Z) is X X the complex defined in this paper. Moreover the exact triangle (1) is isomorphic to Geisser’s triangle ([18] Corollary 5.2 for · = Z). G If = Spec( ) is the spectrum of a totally imaginary number ring then there is a • X OF canonical isomorphism in the derived category
R W ( , Z) ⇠ ⌧ 3R ( W , Z) X ! X where ⌧ 3R ( W , Z) is the truncation of Lichtenbaum’s complex [35] and R W ( , Z) X X is the complex defined in this paper. Theorem 1.3 is proven in 2.9, 2.10, 2.11, 2.12, 2.13, 2.15 and 3.1. Notice that the same formalism is used to treat flat arithmetic schemes and schemes over finite fields (see [35] Question 1 in the Introduction). The basic idea behind the proof of Theorem 1.3 can be explained as follows. The Weil group is defined as an extension of the Galois group by the idèle class group corresponding to the fundamental class of class field theory (more precisely, as the limit of these group extensions). In this paper we use étale duality for arithmetic schemes rather than class field theory, in order to obtain a canonical "extension" of the étale Z-cohomology by the dual of motivic Q(d)-cohomology. More precisely, our Weil-étale complex is defined as the cone of a map
↵ : RHom(⌧ 0R ( , Q(d)), Q[ 2d 2]) R ( et, Z), X X ! X where ↵ is constructed out of étale duality. Here the scheme is of pure dimension d. This X X idea was suggested by works of Burns [8], Geisser [18] and the author [38]. The techniques involved in this paper rely on results due to Geisser and Levine on Bloch’s cycle complex (see [17], [20], [21], [22] and [32]). 4BAPTISTEMORIN
From now on, Conjecture 1.1 is understood as a list of expected properties for the complex R W,c( , Z) of Theorem 1.3. The second goal of this paper is to relate Conjecture 1.1 to X Soulé’s conjecture [42] and to the Tamagawa number conjecture of Bloch-Kato (in the formu- lation of Fontaine and Perrin-Riou, see [15] and [16]). To this aim, we need to assume the following conjecture, which is a special case of a natural refinement for arithmetic schemes of the classical conjecture of Beilinson relating motivic cohomology to Deligne cohomology. Let be a regular, proper and flat arithmetic scheme of pure dimension d. X Conjecture 1.4. (Beilinson) The Beilinson regulator 2d 1 i 2d 1 i H ( , Q(d))R H ( /R, R(d)) X ! D X is an isomorphism for i 1 and there is an exact sequence 2d 1 2d 1 0 0 H ( , Q(d))R H ( /R, R(d)) CH ( Q)R⇤ 0 ! X ! D X ! X ! Now we can state our second main result. If is defined over a number ring we set X OF F = F and = F . X X⌦OF XF X⌦OF Theorem 1.5. Assume that satisfies Conjectures 1.2 and 1.4. X Statements (1), (2) and (3) of Conjecture 1.1 hold for . • X Assume that is projective over Z.ThenStatement(4) of Conjecture 1.1, i.e. the • X identity i i ords=0⇣( ,s)= ( 1) i rank H ( , Z) X · · Z W,c X i 0 X is equivalent to Soulé’s Conjecture [42] for the vanishing order of ⇣( ,s) at s =0. X Assume that is smooth projective over a number ring F ,andthattherepresenta- • i X 1 i O tions H ( , Ql) of GF satisfies H (F, H ( , Ql)) = 0.ThenStatement(5) of XF,et f XF,et Conjecture 1.1, i.e. the identity 1 i ( 1)i Z (⇣⇤( , 0) )= det H ( , Z) · X Z W,c X i Z O2 2d 2 i is equivalent to the Tamagawa number conjecture [15] for the motive h ( )[ i]. i=0 XF The proof of the last statement of Theorem 1.5 was already given in [14],L assuming expected i ( 1)i properties of Weil-étale cohomology. This proof is based on the fact that i Z detZHW,c( , Z) 2 X provides the fundamental line (in the sense of [15]) with a canonical Z-structure. This result N shows that the Weil-étale point of view is compatible with the Tamagawa number conjecture of Bloch-Kato, answering a question of Lichtenbaum (see [35] Question 2 in the Introduction). We obtain examples of flat arithmetic schemes satisfying Conjecture 1.1. Theorem 1.6. For any number field F ,Conjecture1.1holdsfor =Spec( ). X OF Let be a smooth projective scheme over the number ring ( )= F ,whereF is an X OX X O abelian number field. Assume that admits a smooth cellular decomposition (see Definition XF 5.13) and that (Z) (see Definition 5.9). Then Conjecture 4.2 holds for . X2L X We refer to [11] for a proof of a dynamical system analogue of Conjecture 1.1.
Notations. We denote by Z the ring of integers, and by Q, Qp, R and C the fields of rational, p-adic, real and complex numbers respectively. An arithmetic scheme is a scheme ZETA FUNCTIONS AT s =0 5 which is separated and of finite type over Spec(Z). An arithmetic scheme is said to be X proper (respectively flat) if the map Spec(Z) is proper (respectively flat). X! For a field k, we choose a separable closure k/k¯ and we denote by Gk = Gal(k/k¯ ) the absolute Galois group of k. For a proper scheme over Spec(Z) we denote by := (C)/GR X X1 X the quotient topological space of (C) by G where (C) is given with the complex topology. X R X We consider the Artin-Verdier étale topos given with the open-closed decomposition of X et topoi
' : et et Sh( ):u X ! X X1 1 where is the usual étale topos of the scheme (i.e. the category of sheaves of sets on Xet X the small étale site of )andSh( ) is the category of sheaves of sets on the topological X X1 space (see [14] Section 4). For any abelian sheaf A on et and any n>0 the sheaf X1 X Rn' A is a 2-torsion sheaf concentrated on (R). In particular, if (R)= then ' R' . ⇤ X X ; ⇤ ' ⇤ For (C)= one has et = et.ForT = et, et or Sh( ), or more generally for any X ; X X X X X1 Grothendieck topos T , we denote by (T, ) the global section functor, by R (T, ) its total right derived functor and by Hi(T, ):=Hi(R (T, )) its cohomology. Let Z(n):=zn( , 2n ) be Bloch’s cycle complex (see [3], [17], [19], [31] and [32]), which ⇤ we consider as a complex of abelian sheaves on the small étale (or Zariski) site of .For X an abelian group A we define A(n) to be Z(n) A.NotethatZ(n) is a complex of flat ⌦ sheaves, hence tensor product and derived tensor product with Z(n) agree. We denote by Hi( ,A(n)) the étale hypercohomology of A(n),andbyHi( ,A(n)) := Hi( ,A(n)) its Xet X XZar Zariski hypercohomology. For (R)= , we still denote by A(n)=' A(n) the push-forward X ; ⇤ of the cycle complex A(n) on ,andbyHi( ,A(n)) the étale hypercohomology of A(n) X et X et (in this paper A = Z, Z/mZ, Q or Q/Z). Notice that the complex of étale sheaves Z/mZ(n) n is not in general quasi-isomorphic to µm⌦ [0] (however, see [17] Theorem 1.2(4)). We denote by the derived category of the category of abelian groups. More generally, we D write (R) for the derived category of the category of R-modules, where R is a commutative D ring. An exact (i.e. distinguished) triangle in (R) is (somewhat abusively) depicted as D follows
C0 C C00 ! ! where C, C0 and C00 are objects of (R). For an object C of (R) we write C n (respectively D i i D i C n) for the truncated complex so that H (C n)=H (C) for i n and H (C n)=0 i i i otherwise (respectively so that H (C n)=H (C) for i n and H (C n)=0otherwise). Let R be a principal ideal domain. For a finitely generated free R-module L of rank r,we r 1 set detR(L)= R L and detR(L) = HomR(detR(L),R). For a finitely generated R-module M, one may choose a resolution given by an exact sequence 0 L 1 L0 M 0, where V ! ! ! ! L 1 and L0 are finitely generated free R-modules. Then 1 detR(M):=detR(L0) R detR(L 1) ⌦ does not depend on the resolution. If C (R) is a complex such that Hi(C) is finitely gener- i 2D i ( 1)i ated for all i Z and H (C)=0for almost all i, we denote detR(C):= i Z detRH (C) . 2 2 For an abelian group A we write Ator and Adiv for the maximal torsion and the maximal N divisible subgroups of A respectively, and we set Acodiv = A/Adiv and Acotor = A/Ator.We denote by A and A the kernel and the cokernel of the map n : A A (multiplication by n n ! n). We write AD = Hom(A, Q/Z) for the Pontryagin dual of a finite abelian group A. 6BAPTISTEMORIN
Contents 1. Introduction 1 2. Weil-étale cohomology 6 2.1. The conjectures L( et,d) and L( et,d) 0 6 X X 2.2. The morphism ↵ 6 X 2.3. The complex R W ( , Z) 13 X 2.4. Relationship with Lichtenbaum’s definition over finite fields 18 2.5. Relationship with Lichtenbaum’s definition for number rings 22 3. Weil-étale Cohomology with compact support 22 3.1. Cohomology with R˜-coefficients 23 3.2. Cohomology with Z-coefficients 24 3.3. The conjecture B( ,d) and the regulator map 28 X 4. Zeta functions at s =0 30 4.1. The main Conjecture 31 4.2. Relation to Soulé’s conjecture 31 4.3. Relation to the Tamagawa number conjecture 32 4.4. Proven cases 40 4.5. Examples 40 5. Appendix 42 5.1. Motiviccohomologyofnumberrings 43 5.2. Smooth projective varieties over finite fields 45 5.3. The class (Z) 46 L 5.4. Geometrically cellular schemes 48 References 51
2. Weil-étale cohomology
2.1. The conjectures L( et,d) and L( et,d) 0. Let be a proper, regular and connected X X X arithmetic scheme of dimension d. The following conjecture is an étale version of (a special case of) the motivic Bass Conjecture (see [26] Conjecture 37). i Conjecture 2.1. L( et,d) For i 2d,theabeliangroupH ( et, Z(d)) is finitely generated. X X In order to define the Weil-étale cohomology we shall consider schemes satisfying a weak version of the previous conjecture. i Conjecture 2.2. L( et,d) 0 For 0 i 2d,theabeliangroupH ( et, Z(d)) is finitely X X generated. 2.2. The morphism ↵ . Let be a proper regular connected arithmetic scheme. The goal X X of this section is to construct (conditionally) a certain morphism ↵ in the derived category X of abelian groups , in order to define in Section 2.3 the Weil-étale complex R W ( , Z) as D X the cone of ↵ . X 2.2.1. The case (R)= . In this subsection, denotes a proper, regular and connected X ; X arithmetic scheme of dimension d such that (R) is empty. In particular ' is exact (since i i X i ⇤ (R)= ), hence H ( et, Z(d)) := H ( et,' Z(d)) H ( et, Z(d)). X ; X X ⇤ ' X ZETA FUNCTIONS AT s =0 7
Lemma 2.3. We have i H ( et, Z(d)) = 0 for i>2d +2 X Q/Z for i =2d +2 ' =0for i =2d +1 i H ( , Q(d)) for i<0. ' X Proof. For any positive integer n and any i Z,onehas 2 i 1 i 1 (2) H ( et, Z/nZ(d)) = Ext (Z/nZ, Z/nZ(d)) ,Z/nZ X X i (3) Ext (Z/nZ, Z(d)) ' X 2d+2 i D (4) H ( et, Z/nZ) ' X where (3) and (4) are given by ([20] Lemma 2.4) and ([20] Theorem 7.8) respectively, and 2d+2 i D 2d+2 i H ( et, Z/nZ) denotes the Pontryagin dual of the finite group H ( et, Z/nZ). X X The complex Z(d) consists of flat sheaves, hence tensor product and derived tensor product with Z(d) agree. Therefore, applying Z(d) L ( ) to the exact sequence 0 Z Q ⌦ ! ! ! Q/Z 0 yields an exact triangle: ! (5) Z(d) Q(d) Q/Z(d). ! ! i i Moreover, we have H ( et, Q(d)) H ( , Q(d)) for any i (see [17] Proposition 3.6) and X ' X Hi( , Q(d)) = 0 for any i>2d (see [31] Lemma 11.1). Hence (5) gives X i i 1 i 1 H ( et, Z(d)) H ( et, Q/Z(d)) lim H ( et, Z/nZ(d)) X ' X ' ! X for i 2d +2. The result for i 2d +2then follows from (4). By ([24] X.6.2), the scheme has l-cohomological dimension 2d +1for any prime number X i l (note that (R)= ), hence (4) shows that H ( et, Q/Z(d)) = 0 for i<0. The result for X ; i i X i<0 now follows from H ( et, Q(d)) H ( , Q(d)) and from the exact triangle (5). X ' X It remains to treat the case i =2d +1.Assumethat is flat over Z. Then we have (see X [28] Theorem 6.1(2))
2d d (6) H ( , Q(d)) CH ( ) Q =0 X ' X ⌦ i d hence H ( et, Q(d)) = 0 for i 2d. Here CH ( ) denotes the Chow group of cycles of X X codimension d.Weobtain 2d+1 2d H ( et, Z(d)) H ( et, Q/Z(d)) X ' X1 D lim (H ( et, Z/nZ) ) ' X ! D (lim Hom(⇡1( et), Z/nZ)) ' X ab D Hom(⇡1( et) , Z) ' X =0 b where the first isomorphism (respectively the second) is given by (5) (respectively by (4)), and the last isomorphism follows from the fact that the abelian fundamental group ⇡ ( )ab is 1 X et finite (see [28] Theorem 9.10 and [48] Corollary 3). 8BAPTISTEMORIN
Assume now that is a smooth proper scheme over a finite field (hence = ). One has X X X 2d 1 D (7) H ( et, Q/Z(d)) lim (H ( et, Z/nZ) ) X ' X ! D (8) lim (Hom(⇡1( et), Z/nZ) ) ' X ! d (9) lim (CH ( ) Z/nZ) ' X ⌦ ! d (10) CH ( ) Q/Z ' X ⌦ where (7) is given by (4) while (9) follows from class field theory (see [48] Corollary 3). The exact triangle (5) yields an exact sequence 2d p 2d 2d+1 (11) H ( , Q(d)) H ( et, Q/Z(d)) H ( et, Z(d)) 0. X ! X ! X ! Moreover, the isomorphism (10) is the direct limit of the maps d 2d 2d 2d CH ( )n = H ( , Z(d))n H ( , Z/nZ(d)) H ( et, Z/nZ(d)). X X ! X ! X It follows that the map p in the sequence (11) can be identified with the morphism d d CH ( ) Q CH ( ) Q/Z X ⌦ ! X ⌦ induced by the surjection Q Q/Z. Hence p is surjective, since CHd( ) is finitely generated ! 2d+1 X of rank one ([28] Theorem 6.1), so that H ( et, Z(d)) = 0. X ⇤ j j Notation 2.4. We set H ( et, Z(d)) 0 := H (R ( et, Z(d)) 0). X X Lemma 2.5. Assume that satisfies L( et,d) 0.Thenthenaturalmap X X i 2d+2 i H ( et, Z) Hom(H ( et, Z(d)) 0, Q/Z) X ! X is an isomorphism of abelian groups for i 1. Proof. The map of the lemma is given by the pairing i 2d+2 i 2d+2 H ( et, Z) Ext (Z, Z(d)) H ( et, Z(d)) Q/Z. X ⇥ X ! X ' 1 For i =1the result follows from Lemma 2.3, since one has H ( et, Z)=Homcont(⇡1( et,p), Z)= X X 0 because the fundamental group ⇡1( et,p) is profinite (hence compact). X i i The scheme is connected and normal, hence H ( et, Q)=H ( et, Q)=Q, 0 for i =0 X X X and i 1 respectively. We obtain i i 1 i 1 H ( et, Z)=H ( et, Q/Z)=lim H ( et, Z/nZ) X X ! X for i 2, since is quasi-compact and quasi-separated. For any positive integer n the X canonical map 2d+2 i i Ext (Z/nZ, Z(d)) H ( et, Z/nZ) Q/Z X ⇥ X ! is a perfect pairing of finite groups (see [20] Theorem 7.8). The short exact sequence 0 Z Z Z/nZ 0 ! ! ! ! yields an exact sequence j 1 j 1 j H ( et, Z(d)) H ( et, Z(d)) Ext (Z/nZ, Z(d)) X ! X ! X j j H ( et, Z(d)) H ( et, Z(d)). ! X ! X ZETA FUNCTIONS AT s =0 9
We obtain a short exact sequence j 1 j j 0 H ( et, Z(d))n Ext (Z/nZ, Z(d)) nH ( et, Z(d)) 0 ! X ! X ! X ! for any j. By left exactness of projective limits the sequence j 1 j j 0 lim H ( et, Z(d))n lim Ext (Z/nZ, Z(d)) lim nH ( et, Z(d)) ! X ! X ! X j j is exact. The module lim nH ( et, Z(d)) vanishes for 0 j 2d +1since H ( et, Z(d)) is X X j assumed to be finitely generated for such j.WehavelimnH ( et, Z(d)) = 0 for j<0 since X j H ( et, Z(d)) is uniquely divisible for j<0 (see Lemma 2.3). This yields an isomorphism of X profinite groups j 1 j 2d+2 j D lim H ( et, Z(d))n ⇠ lim Ext (Z/nZ, Z(d)) ⇠ (lim H ( et, Z/nZ)) , X ! X ! X ! j where the last isomorphism follows from the duality above (and from the fact that Ext (Z/nZ, Z(d)) is finite for any n). This gives isomorphisms of torsion groups X 2d+2 (j 1) 2d+2 j DD H ( et, Z) ⇠ (lim H ( et, Z/nZ)) X ! ! X j 1 D j 1 ⇠ (lim H ( et, Z(d))n) ⇠ Hom(H ( et, Z(d)) 0, Q/Z) ! X ! X for any j 2d +1 (note that 2d +2 (j 1) 2 j 2d +1). The last isomorphism above j , follows from the fact that H ( et, Z(d)) is finitely generated for 0 j 2d and uniquely X divisible for j<0. Hence for any i 2 the natural map i 2d+2 i H ( et, Z) Hom(H ( et, Z(d)) 0, Q/Z) X ! X is an isomorphism. ⇤ Recall that an abelian group A is of cofinite type if A Hom (B,Q/Z) where B is a ' Z finitely generated abelian group. If A is of cofinite type then there exists an isomorphism r A (Q/Z) T , where r N and T is a finite abelian group. If satisfies L( et,d) 0 then i' 2 X X H ( et, Z) is of cofinite type for i 1 by Lemma 2.5. X Theorem 2.6. Assume that satisfies L( et,d) 0.Thereexistsauniquemorphismin X X D ↵ : RHom(R ( , Q(d)) 0, Q[ 2d 2]) R ( et, Z) X X ! X such that Hi(↵ ) is the following composite map X 2d+2 i 2d+2 i Hom(H ( , Q(d)) 0, Q) ⇠ Hom(H ( et, Z(d)) 0, Q) X ! X 2d+2 i i Hom(H ( et, Z(d)) 0, Q/Z) ⇠ H ( et, Z) ! X X for any i 2. Proof. In order to ease the notation, we set D := RHom(R ( , Q(d)) 0, Q[ 2d 2]).We X X consider the spectral sequence (see [45] III, Section 4.6.10) p,q p i q i p+q (12) E2 = Ext (H (D ),H ( et, Z)) H (RHom(D , R ( et, Z))). X X ) X X i Z Y2 An edge morphism from (12) gives a morphism 0 0 i i (13) H (RHom(D , R ( et, Z))) Ext (H (D ),H ( et, Z)) X X ! X X i Z Y2 10 BAPTISTE MORIN such that the composite map 0 0 i i Hom (D , R ( et, Z))) ⇠ H (RHom(D , R ( et, Z))) Ext (H (D ),H ( et, Z)) D X X ! X X ! X X i Z Y2 is the obvious one. i By Lemma 2.5, H ( et, Z) is of cofinite type for i =0. It follows that, if both i =0 pX i q i 6 i q 6 and p =0, then Ext (H (D ),H ( et, Z)) = 0, since H (D ) is uniquely divisible. 6 p i q rX X r X Indeed, Ext (H (D ), (Q/Z) )=0for p 1, since (Q/Z) is divisible. Moreover, if T is X p i q p i q a finite abelian group of order n then Ext (H (D ),T)=0since Ext (H (D ),T) is i q X X both uniquely divisible (since H (D ) is) and killed by n (since T is). For i =0,onehas p 0 q 0 X 0 Ext (H (D ),H ( et, Z)) = 0 as long as p 2, because H ( et, Z)=Z has an injective X X p,q X resolution of length one. In particular, E =0for p =0, 1, hence the spectral sequence (12) 2 6 degenerates at E2. Moreover, one has 1, 1 1 i+1 i 1 1 0 E2 = Ext (H (D ),H ( et, Z)) = Ext (H (D ),H ( et, Z)) = 0 X X X X i Z Y2 since H1(D )=0. It follows that the edge morphism (13) is an isomorphism. X i i i i For i 1,anymapH (↵ ):H (D ) H ( et, Z) must be trivial since H (D )=0.For X X ! X X any i 2, we consider the morphism i i 2d+2 i 2d+2 i ↵ : H (D )=Hom(H ( , Q(d)) 0, Q) ⇠ Hom(H ( et, Z(d)) 0, Q) X X X ! X 2d+2 i i Hom(H ( et, Z(d)) 0, Q/Z) ⇠ H ( et, Z) ! X X where the last isomorphism is given by Lemma 2.5. But (13) is an isomorphism, hence there exists a unique map in D ↵ : RHom(R ( , Q(d)) 0, Q[ 2d 2]) R ( et, Z) X X ! X i i such that H (↵ )=↵ . ⇤ X X 2.2.2. The general case. In this subsection, we allow (R) = .So denotes a proper, X 6 ; X regular and connected arithmetic scheme of dimension d. It is possible to define a dualizing complex Z(d)X on et, to prove Artin-Verdier duality in this setting and to define ↵ in the X X very same way as in Section 2.2.1. For the sake of conciseness, we shall use a somewhat trickier construction, which is based on Milne’s cohomology with compact support [37]. We denote by ˆ ˆ (14) R c( et, ):=R c(Spec(Z)et, Rf ) X F ⇤F ˆ Milne’s cohomology with compact support, where R c(Spec(Z)et, ) is defined as in ([37] Section II.2) and f : Spec(Z) is the structure map. Milne’s definition yields a canonical ˆ X! map R c(Spec(Z)et, ) R (Spec(Z)et, ) inducing ! ˆ (15) R c( et, Z) R ( et, Z). X ! X We need the following lemma. ˆ c Lemma 2.7. There is a morphism R c( et, Z) R ( et, Z) satisfying the following prop- X ! X erties. The composite map ˆ c '⇤ R c( et, Z) R ( et, Z) R ( et, Z) X ! X ! X coincides with (15), Hi(c) is an isomorphism for i large and Hi(c) has finite 2-torsion kernel and cokernel for any i Z. 2 ZETA FUNCTIONS AT s =0 11
Proof. Recall from ([14] Section 4) that there is an isomorphism of functors u⇤ ' ⇡ ↵⇤ 1 ⇤ ' ⇤ where ↵ : Sh(GR, (C)) et and ⇡ : Sh(GR, (C)) Sh( ) are the canonical maps X !X X ! X1 while u : Sh( ) et and ' : et et are complementary closed and open embeddings 1 X1 ! X X ! X (see [14] Section 4). Here Sh(GR, (C)) (respectively Sh( )) denotes the topos of GR- X X1 equivariant sheaves of sets on (C) (respectively the topos of sheaves on ). It is easy to X X1 see that ↵⇤ sends injective abelian sheaves to ⇡ -acyclics ones. We obtain ⇤ u⇤ R' R(u⇤ ' ) R(⇡ ↵⇤) (R⇡ )↵⇤. 1 ⇤ ' 1 ⇤ ' ⇤ ' ⇤ Consider the map >0 R' Z u , u⇤ R' Z u , R⇡ Z u , ⌧ R⇡ Z. ⇤ ! 1 ⇤ 1 ⇤ ' 1 ⇤ ⇤ ! 1 ⇤ ⇤ 0 Since ⌧ R⇡ Z is the constant sheaf Z = u⇤ Z put in degree 0, we obtain an exact triangle ⇤ 1 >0 (16) ZX R' Z u , ⌧ R⇡ Z. ! ⇤ ! 1 ⇤ ⇤ >0 where ZX denotes the constant sheaf Z on et. Moreover, we have R ( et,u , ⌧ R⇡ Z) >0 >0 X >0X 1 ⇤ ⇤ ' R ( ,⌧ R⇡ Z) R ( (R),⌧ R⇡ Z) because u , is exact and ⌧ R⇡ Z is concentrated X1 ⇤ ' X ⇤ 1 ⇤ ⇤ on (R) . Therefore, applying R ( et, ) to (16), we get an exact triangle X ⇢X1 X >0 (17) R ( et, Z) R ( et, Z) R ( (R),⌧ R⇡ Z). X ! X ! X ⇤ On the other hand there is an exact triangle ˆ ˆ (18) R c( et, Z) R ( et, Z) R (G , (R), Z) X ! X ! R X where R ˆ(G , (R), ):=R ˆ(G , (R ( (R), )) denotes equivariant Tate cohomology (which R X R X was introduced in [44]). Notice that the canonical map R ˆ(G , (C), Z) R ˆ(G , (R), Z) R X ! R X is an isomorphism [44]. ˆ >0 We shall define below a canonical map c : R (GR, (R), Z) R ( (R),⌧ R⇡ Z) such 1 X ! X ⇤ that Hi(c ) is an isomorphism for i large and Hi(c ) has finite 2-torsion kernel and cokernel 1 1 for any i Z. The map c is compatible (in the obvious sense) with the identity map of 2 1 ˆ R ( et, Z), hence there exists a morphism c : R c( et, Z) R ( et, Z) such that (c, Id, c ) X X ! X 1 is a morphism of exact triangles from (18) to (17). Hence the result will follow. It remains to define c . We denote by R (GR, (R), ):=R (GR, (R ( (R), )) usual 1 X X equivariant cohomology. We have isomorphisms ˆ (19) R (G , (R), Z)>d R (G , (R), Z)>d R X ' R X (20) R ( (R),R⇡ Z)>d ' X ⇤ (21) R ( (R), (R⇡ Z)>0)>d ' X ⇤ (22) R ( (R), R (G , Z)>0)>d ' X R ˆ (23) R ( (R), R (G , Z)>0)>d ' X R ˆ (24) R ( (R), R (G , Z))>d ' X R where we consider R ˆ(G , Z) and R (G , Z) as complexes of constant sheaves on (R). R R X The canonical map R (G , (R), Z) R ˆ(G , (R), Z) and the corresponding morphism of R X ! R X hypercohomology spectral sequences give the isomorphism (19), since (R) is of topological X dimension d. Similarly (21) and (24) can be deduced from the corresponding morphisms of 12 BAPTISTE MORIN spectral sequences. Finally, (22) is given by (R⇡ Z) (R) R (GR, Z).SinceGR is cyclic, we ⇤ |X ' have an isomorphism R ˆ(G , Z) ⇠ R ˆ(G , Z)[2] in .ApplyingR ( (R), ) we obtain R ! R D X R ( (R), R ˆ(G , Z)) ⇠ R ( (R), R ˆ(G , Z))[2]. X R ! X R The same is true for equivariant Tate cohomology: we have an isomorphism R ˆ(G , (R), Z) ⇠ R ˆ(G , (R), Z)[2]. R X ! R X Hence for any integer k 1, (24) induces an isomorphism ˆ ˆ R (GR, (R), Z)>d 2k R ( (R), R (GR, Z))>d 2k. X ' X Taking k large enough, we obtain a canonical isomorphism ˆ ˆ (25) R (G , (R), Z)>0 R ( (R), R (G , Z))>0. R X ' X R ˆ ˆ The map R (G , Z) R (G , Z)>0 induces R ! R ˆ ˆ (26) R ( (R), R (G , Z)) R ( (R), R (G , Z)>0). X R ! X R ˆ Since R ( (R), R (G , Z)>0) is acyclic in degrees 0, (26) is induced by a unique map X R ˆ ˆ (27) R ( (R), R (G , Z))>0 R ( (R), R (G , Z)>0) X R ! X R and c is defined as the composition 1 ˆ ˆ (25) ˆ R (G , (R), Z) R (G , (R), Z)>0 R ( (R), R (G , Z))>0 R X ! R X ! X R (27) ˆ >0 R ( (R), R (GR, Z)>0) R ( (R),⌧ R⇡ Z). ! X ' X ⇤ ⇤
If (R)= , then the maps c and '⇤ of Lemma 2.7 are isomorphisms in (this follows X ; D from (18) and from the fact that ' is exact in this case). Therefore, the following result ⇤ generalizes Theorem 2.6.
Proposition 2.8. Assume that satisfies L( et,d) 0.Thenthereexistsauniquemorphism X X in D ↵ : RHom(R ( , Q(d)) 0, Q[ 2d 2]) R ( et, Z) X X ! X such that Hi(↵ ) is the following composite map X 2d+2 i 2d+2 i Hom(H ( , Q(d)) 0, Q) ⇠ Hom(H ( et, Z(d)) 0, Q) X ! X i 2d+2 i ˆ i H (c) i Hom(H ( et, Z(d)) 0, Q/Z) ⇠ Hc( et, Z) H ( et, Z) ! X X ! X for any i 2. Proof. We consider the composite map p RHom (Z, Z(d)) RHomSpec( )(Rf Z, Rf Z(d)) RHomSpec( )(Rf Z, Z(1)[ 2d + 2]) X ! Z ⇤ ⇤ ! Z ⇤ q ˆ ˆ RHom (R c(Spec(Z)et, Rf Z), Q/Z[ 2d 2]) = RHom (R c( et, Z), Q/Z[ 2d 2]) ! D ⇤ D X where p is induced by the push-forward map of ([20] Corollary 7.2 (b)) and q is given by 1-dimensional Artin-Verdier duality (see [37] II.2.6) since Z(1) Gm[ 1] (see [20] Lemma ' 7.4). This composite map induces ˆ R ( et, Z(d)) 0 RHom(R c( et, Z), Q/Z[ 2d 2]) X ! X ZETA FUNCTIONS AT s =0 13
ˆ ˆ i since R c( et, Z) is acyclic in degrees > 2d +2. Indeed, H (Spec(Z)et,F)=0for any i>3 X c and any torsion abelian sheaf F (see [37] Theorem II.3.1(b)); hence by proper base change the ˆ i hypercohomology spectral sequence associated to (14) gives H ( et, Q/Z)=0for i>2d +1. c X By adjunction, we obtain the product map L ˆ (28) R ( et, Z(d)) 0 R c( et, Z) Q/Z[ 2d 2]. X ⌦Z X ! For any positive integer n and any i Z, the canonical map 2 2d+2 i ˆ i (29) Ext (Z/nZ, Z(d)) Hc( et, Z/nZ) Q/Z X ⇥ X ! is a perfect pairing of finite groups (see [20] Theorem 7.8). Note that (29) is compatible with (28). The arguments of the proof of Lemma 2.5 show that the morphism ˆ i 2d+2 i (30) Hc( et, Z) ⇠ Hom(H ( et, Z(d)) 0, Q/Z), X ! X induced by (28), is an isomorphism of abelian groups for i 2. The argument of the proof of Theorem 2.6 provides us with a unique map ˆ ↵˜ : RHom(R ( , Q(d)) 0, Q[ 2d 2]) R c( et, Z) X X ! X such that Hi(˜↵ ) is the following composite map X 2d+2 i 2d+2 i Hom(H ( , Q(d)) 0, Q) ⇠ Hom(H ( et, Z(d)) 0, Q) X ! X 2d+2 i ˆ i Hom(H ( et, Z(d)) 0, Q/Z) ⇠ Hc( et, Z) ! X X for any i 2. Then we define ↵ as the composite map X ↵˜ ˆ c RHom(R ( , Q(d)) 0, Q[ ]) X R c( et, Z) R ( et, Z) X ! X ! X where c is the map of Lemma 2.7. It remains to show that ↵ does not depend on the X choice of c (note that c is not uniquely defined). The map ' c is well defined (since it ⇤ coincides with (15) by Lemma 2.7), hence so is '⇤ ↵ : D R ( et, Z), where D := X X ! X X RHom(R ( , Q(d)) 0, Q[ 2d 2]). By (17) and using the fact that D is a complex of X X Q-vector spaces while R ( (R),⌧>0R⇡ Z) is a complex of Z/2Z-vector spaces, we see that X ⇤ composition with '⇤ induces an isomorphism
'⇤ : Hom (D , R ( et, Z)) ⇠ Hom (D , R ( et, Z)) D X X ! D X X so that ↵ does not depend on the choice of c. The result follows. ⇤ X
2.3. The complex R W ( , Z). Throughout this section denotes a proper regular con- X X nected arithmetic scheme of dimension d satisfying L( et,d) 0. X Definition 2.9. There exists an exact triangle ↵ RHom(R ( , Q(d)) 0, Q[ 2d 2]) X R ( et, Z) R W ( , Z) X ! X ! X well defined up to a unique isomorphism in . We define D i i H ( , Z):=H (R W ( , Z)). W X X The existence of such an exact triangle follows from axiom TR1 of triangulated categories. Its uniqueness is given by Theorem 2.11 and can be stated as follows. If we have two objects 14 BAPTISTE MORIN
R W ( , Z) and R W ( , Z)0 given with exact triangles as in Definition 2.9, then there exists X X a unique map f : R W ( , Z) R W ( , Z)0 in which yields a morphism of exact triangles X ! X D ↵ RHom(R ( , Q(d)) 0, Q[ 2d 2]) X / R ( et, Z) / R W ( , Z) X X X Id Id f ✏ ↵ ✏ ✏ RHom(R ( , Q(d)) 0, Q[ 2d 2]) X / R ( et, Z) / R W ( , Z)0 X X X Proposition 2.10. The following assertions are true. (1) The group Hi ( , Z) is finitely generated for any i Z.OnehasHi ( , Z)=0for W X 2 W X i<0, H0 ( , Z)=Z and Hi ( , Z)=0for i large. W X W X (2) If (R)= then there is an exact sequence X ; i i 2d+2 (i+1) 0 H ( et, Z)codiv HW ( , Z) HomZ(H ( et, Z(d)) 0, Z) 0 ! X ! X ! X ! for any i Z. 2 (3) For any and any i Z,thereisanexactsequence X 2 i i i+1 0 H ( et, Z)codiv HW ( , Z) Ker(H (↵ )) 0 ! X ! X ! X ! i+1 2d+2 (i+1) where Ker(H (↵ )) HomZ(H ( et, Q(d)) 0, Q) is a Z-lattice. X ✓ X Proof. In order to ease the notation we set := 2d +2. The exact triangle of Defintion 2.9 yields a long exact sequence i i i ... Hom(H ( , Q(d)) 0, Q) H ( et, Z) HW ( , Z) ! X ! X ! X (i+1) i+1 Hom(H ( , Q(d)) 0, Q) H ( et, Z) ... ! X ! X ! and a short exact sequence i i i+1 (31) 0 CokerH (↵ ) HW ( , Z) KerH (↵ ) 0. ! X ! X ! X ! Assume for the moment that (R)= .Fori 1,themorphism X ; i i i H (↵ ):Hom(H ( , Q(d)) 0, Q) H ( et, Z) X X ! X is the following composite map: i i Hom(H ( , Q(d)) 0, Q) ⇠ Hom(H ( et, Z(d)) 0, Q) X ! X i i Hom(H ( et, Z(d)) 0, Q/Z) H ( et, Z) ! X ' X i Since H ( et, Z(d)) 0 is assumed to be finitely generated, the image of the morphism i X i i H (↵ ) is the maximal divisible subgroup of H ( et, Z), which we denote by H ( et, Z)div. X X X For i =0one has 0 0 H (↵ ):0=Hom(H ( , Q(d)), Q) H ( et, Z)=Z X X ! X i i and H ( , Q(d)) = H ( et, Z)=0for i<0.Weobtain X X i i i i CokerH (↵ )=H ( et, Z)/H ( et, Z)div =: H ( et, Z)codiv X X X X for any i Z. By definition of ↵ , the kernel of Hi(↵ ) can be identified with the kernel of 2 i X i X the map Hom(H ( et, Z(d)) 0, Q) Hom(H ( et, Z(d)) 0, Q/Z). In other words, one X ! X has i i KerH (↵ ) Hom(H ( et, Z(d)) 0, Z). X ' X ZETA FUNCTIONS AT s =0 15
The exact sequence (31) therefore takes the form i i 2d+2 (i+1) 0 H ( et, Z)codiv HW ( , Z) HomZ(H ( et, Z(d)) 0, Z) 0. ! X ! X ! X ! We obtain the second claim of the proposition. Assume now that (R) = . Then for i 2,themorphismHi(˜↵ ) is the map X 6 ; X i i Hom(H ( , Q(d)) 0, Q) ⇠ Hom(H ( et, Z(d)) 0, Q) X ! X i ˆ i Hom(H ( et, Z(d)) 0, Q/Z) Hc( et, Z) ! X ' X and Hi(↵ ) is defined as the composite map X Hi(˜↵ ) i i X ˆ i i H (↵ ):Hom(H ( , Q(d)) 0, Q) Hc( et, Z) H ( et, Z). X X ! X ! X i ˆ i The same argument as in the case (R)= shows Im(H (˜↵ )) = Hc( et, Z)div. Let us ˆ i i X ; X X show that h : Hc( et, Z)div H ( et, Z)div is surjective for any i Z. This is obvious for i X ! X ˆ i 2 i i 1, since H ( et, Z)div =0for i 1.Fori 2,themapsHc( et, Z) H ( et, Z) and ˆi Xˆ i ˆ iX ! X Hc( et, Z)div Hc( et, Z) have both finite cokernels (note that Hc( et, Z) is of cofinite type X ! X ˆ i X i for i 2 by (30)). It follows immediately that the cokernel of h : H ( et, Z)div H ( et, Z)div c X ! X is of finite exponent and divisible, hence vanishes. So h is indeed surjective. We obtain i i Im(H (↵ )) = H ( et, Z)div hence X X i i CokerH (↵ )=H ( et, Z)codiv X X for any i Z. We now compute KerHi(↵ ). By definition of ↵˜ , the kernel of Hi(˜↵ ) can be 2 X i X i X identified with the kernel of the map Hom(H ( et, Z(d)) 0, Q) Hom(H ( et, Z(d)) 0, Q/Z). X ! X Hence one has i i (32) KerH (˜↵ ) Hom(H ( et, Z(d)) 0, Z). X ' X ˆ i i We denote by Ti the kernel of the surjective map h : H ( et, Z)div H ( et, Z)div so that c X ! X there is an exact sequence ˆ i h i (33) 0 Ti H ( et, Z)div H ( et, Z)div 0. ! ! c X ! X ! Consider the obvious injective map f : KerHi(˜↵ ) KerHi(↵ ), and the following morphism X ! X of exact sequences:
Hi(˜↵ ) i i X ˆ i 0 / KerH (˜↵ ) / Hom(H ( , Q(d)) 0, Q) / Hc( et, Z)div / 0 X X X f = h ✏ ✏ Hi(↵ ) ✏ i i i 0 / KerH (↵ ) / Hom(H ( , Q(d)) 0, Q) X / H ( et, Z)div / 0 X X X The snake lemma then yields an isomorphism T = Ker(h) Coker(f).Using(32),weobtain i ' an exact sequence i i 0 Hom(H ( et, Z(d)) 0, Z) KerH (↵ ) Ti 0. ! X ! X ! ! Moreover, Ti is finite and 2-torsion. We obtain the third claim of the proposition. It follows from the third claim that Hi ( , Z) is finitely generated for any i Z. Indeed, W X 2 KerHi+1(↵ ) is finitely generated since it is a Z-lattice in a finite dimensional Q-vector space. Xˆ i Moreover, Hc( et, Z) is of cofinite type for i 2 and finitely generated for i 1. Hence ˆ i X i H ( et, Z)codiv is finite for any i Z and so is H ( et, Z)codiv by (33). c X 2 X 16 BAPTISTE MORIN
i 0 Moreover, one has H ( et, Z)=0for i<0 and H ( et, Z)=Z. Moreover one has i ˆ i X ˆ i X H ( et, Z) Hc( et, Z) for i large (by Lemma 2.7) and Hc( et, Z)=0for i>2d +2 by (30), X i ' X i+1 X 2d+2 (i+1) hence H ( et, Z)=0for i large. Finally, KerH (↵ ) Hom(H ( , Q(d)) 0, Q)= X X ⇢ X 0 for i 0 as well as for i 2d +2. The first claim of the proposition follows. ⇤ Theorem 2.11. Let f : be a morphism of relative dimension c between proper X!Y regular connected arithmetic schemes, such that L( et,d ) 0 and L( et,d ) 0 hold, where X X Y Y d (respectively d )denotesthedimensionof (respectively of ). We choose complexes X Y X Y R W ( , Z) and R W ( , Z) as in Definition 2.9. Assume either that c 0 or that L( et,d ) X Y X X holds. Then there exists a unique map in D f ⇤ : R W ( , Z) R W ( , Z) W Y ! X which sits in a morphism of exact triangles
RHom(R ( , Q(d )) 0, Q[ 2d 2]) / R ( et, Z) / R W ( , Z) X XO X XO O X
fet⇤ fW⇤
RHom(R ( , Q(d )) 0, Q[ 2d 2]) / R ( et, Z) / R W ( , Z) Y Y Y Y Y In particular, if satisfies L( et,d ) 0 then R W ( , Z) is well defined up to a unique X X X X isomorphism in . D Proof. Let and be connected, proper and regular arithmetic schemes of dimension d and X Y X d respectively. We set := 2d +2and := 2d +2. We choose complexes R W ( , Z) Y X X Y Y X and R W ( , Z) as in Definition 2.9. Let f : be a morphism of relative dimension Y X!Y c = d d . The morphism f is proper and the map X Y n n c z ( , ) z ( , ) X ⇤ ! Y ⇤ induces a morphism f Q(d ) Q(d ))[ 2c] ⇤ X ! Y of complexes of abelian Zariski sheaves on . We need to see that Y (34) f Q(d ) Rf Q(d ). ⇤ X ' ⇤ X Localizing over the base, it is enough to show this fact for f a proper map over a discrete valuation ring. Over a discrete valuation ring, Zariski hypercohomology of the cycle complex coincides with its cohomology as a complex of abelian group (see [32] and [17] Theorem 3.2). This yields (34) and a morphism of complexes R ( , Q(d )) R ( ,f Q(d )) R ( , Q(d ))[ 2c]. X X ' Y ⇤ X ! Y Y If c 0 this induces a morphism R ( , Q(d )) 0 R ( , Q(d )) 0[ 2c]. X X ! Y Y i If c<0 then L( et,d ) holds by assumption. It follows that H ( , Q(d )) = 0 for i<0. Xi X i X X Indeed, we have H ( et, Z(d)) H ( , Q(d )) for i<0 (by Lemma 2.3 for (R)= and by X ' X X i i X ; (29) for the general case). Hence for i<0, H ( , Q(d )) H ( et, Z(d)) is both uniquely X X ' X ZETA FUNCTIONS AT s =0 17 divisible and finitely generated by L( et,d ), hence must be trivial. So we may consider the X X map
R ( , Q(d )) 0 ⇠ R ( , Q(d )) R ( , Q(d ))[ 2c] R ( , Q(d )) 0[ 2c]. X X X X ! Y Y ! Y Y In both cases we get a morphism
RHom(R ( , Q(d )) 0, Q[ ]) RHom(R ( , Q(d )) 0, Q[ ]) Y Y Y ! X X X such that the following square is commutative: ↵ RHom(R ( , Q(d )) 0, Q[ ]) Y / R ( et, Z) Y Y Y Y
✏ ↵ ✏ RHom(R ( , Q(d )) 0, Q[ ]) X / R ( et, Z) X X X X Showing that the diagram above is indeed commutative is tedious but straightforward, using the push-forward map defined in [20] Corollary 7.2 (b). Hence there exists a morphism
f ⇤ : R W ( , Z) R W ( , Z) W Y ! X inducing a morphism of exact triangles. We claim that such a morphism fW⇤ is unique. In order to ease the notations, we set
D := RHom(R ( , Q(d )) 0, Q[ ]) and D := RHom(R ( , Q(d )) 0, Q[ ]). X X X X Y Y Y Y The complexes R W ( , Z) and R W ( , Z) are both perfect complexes of abelian groups, X Y since they are bounded complexes with finitely generated cohomology groups. Applying the functor Hom ( , R W ( , Z)) to the exact triangle D X D R ( et, Z) R W ( , Z) D [1] Y ! Y ! Y ! Y we obtain an exact sequence of abelian groups:
Hom (D [1], R W ( , Z)) Hom (R W ( , Z), R W ( , Z)) D Y X ! D Y X Hom (R ( et, Z), R W ( , Z)). ! D Y X On the one hand, Hom (D [1], R W ( , Z)) is uniquely divisible since D [1] is a complex D Y X Y of Q-vector spaces. On the other hand, the abelian group Hom (R W ( , Z), R W ( , Z)) is D Y X finitely generated as it follows from the spectral sequence p i q+i p+q Ext (H ( , Z),H ( , Z)) H (RHom(R W ( , Z), R W ( , Z))) W Y W X ) Y X i Z Y2 since R W ( , Z) and R W ( , Z) are both perfect. Hence the morphism X Y Hom (R W ( , Z), R W ( , Z)) Hom (R ( et, Z), R W ( , Z)) D Y X ! D Y X is injective, which implies the uniqueness of the morphism fW⇤ sitting in the morphism of exact triangles of the theorem. Assume now that satisfies L( et,d ) 0, and let R W ( , Z) and R W ( , Z)0 be two X X X X X complexes given with exact triangles as in Definition 2.9. Then the identity map Id : X!X induces a unique isomorphism R W ( , Z) R W ( , Z)0 in . X ' X D ⇤ i i ⌫ We denote by Hcont( et, Zl):=H (RlimR ( et, Z/l Z)) continuous l-adic cohomology. X X 18 BAPTISTE MORIN
Corollary 2.12. Let be a proper regular connected arithmetic scheme of dimension d sat- X isfying L( et,d) 0.Foranyprimenumberl and any i Z,thereisanisomorphism X 2 i i H ( , Z) Zl H ( et, Zl). W X ⌦ ' cont X Proof. Consider the exact triangle
RHom(R ( , Q(d)) 0, Q[ 2d 2]) R ( et, Z) R W ( , Z) X ! X ! X Let n be any positive integer. Applying the functor L Z/nZ we obtain an isomorphism ⌦Z L L R ( et, Z/nZ) R ( et, Z) Z/nZ ⇠ R W ( , Z) Z/nZ X ' X ⌦Z ! X ⌦Z L since RHom(R ( , Q(d )) 0, Q[ 2d 2]) Z/nZ 0. Let l be a prime number and let ⌫ X X ⌦Z ' be a positive integer. We obtain a short exact sequence i i ⌫ i+1 (35) 0 H ( , Z)l⌫ H ( et, Z/l Z) l⌫ H ( , Z) 0 ! W X ! X ! W X ! for any i Z. By left exactness of projective limits we get 2 i i ⌫ i+1 0 lim HW ( , Z)l⌫ lim H ( et, Z/l Z) lim l⌫ HW ( , Z). ! X ! X ! X i+1 i+1 But lim l⌫ HW ( , Z)=0since HW ( , Z) is finitely generated. Moreover, we have X X i i ⌫ Hcont( et, Zl) lim H ( et, Z/l Z) X ' X i ⌫ since H ( et, Z/l Z) is finite by (35). X ⇤ 2.4. Relationship with Lichtenbaum’s definition over finite fields. Let Y be a scheme of finite type over a finite field k. We denote by Gk and Wk the Galois group and the Weil sm group of k respectively. The (small) Weil-étale topos YW is the category of Wk-equivariant sheaves of sets on the étale site of Y k. The big Weil-étale topos is defined [14] as the fiber ⌦k product Y := Y ( ) Y sm B W et Spec Z W et BG Wk ⇥Spec(Z)et ' ⇥ k Bsm B G where Gk (resp. Wk ) denotes the small classifying topos of k (resp. the big classifying sm topos of Wk). The topoi YW and YW are cohomologically equivalent (see [14] Corollary 2). Therefore, by [18] one has an exact triangle in the derived category of abelian sheaves on Yet Z R Z Q[ 1] Z[1] ! ⇤ ! ! where : Y Y is the first projection. Applying R (Y , ) and rotating, we get W ! et et aY R (Yet, Q[ 2]) R (Yet, Z) R (YW , Z) R (Yet, Q[ 2])[1]. ! ! ! Theorem 2.13. Let Y be a d-dimensional connected projective smooth scheme over k satis- fying L(Yet,d) 0.Thenthereisanisomorphismin D (36) R (YW , Z) ⇠ R W (Y,Z) ! where R (YW , Z) is the cohomology of the Weil-étale topos and R W (Y,Z) is the complex defined in this paper. Moreover, the exact triangle of Definition 2.9 is isomorphic to Geisser’s triangle ([18] Corollary 5.2). ZETA FUNCTIONS AT s =0 19
Proof. We shall define a commutative square in D aY R (Yet, Q[ 2]) / R (Yet, Z) Id ' ✏ ↵Y ✏ RHom(R (Y,Q(d)) 0, Q[ 2d 2]) / R (Yet, Z) where the vertical maps are isomorphisms. The existence of the isomorphism (36) and the compatibility with Geisser’s triangle ([18] Corollary 5.2 for · = Z) will immediately follow. G Moreover, the uniqueness of (36) will follow from the argument of the proof of Theorem 2.11. One is therefore reduced to define the commutative square above. Replacing k with a finite extension if necessary, one may suppose that Y is geometrically connected over k.Onehas H2d(Y,Q(d)) = CHd(Y ) Q ([28] Theorem 6.1) and Hi(Y,Q(d)) = 0 for i>2d. This Q ' yields a map R (Y,Q(d)) Q[ 2d]. The morphism ! RHomY (Q, Q(d)) RHom(R (Y,Q), R (Y,Q(d))) RHom(R (Y,Q), Q[ 2d]) ! ! induces a morphism
(37) R (Yet, Q) R (Y,Q) RHom(R (Y,Q(d)) 0, Q[ 2d]). ' ! i It follows from Conjecture L(Yet,d) 0, Lemma 2.5 and from the fact that H (Yet, Z) is finite i for i =0, 2,thatthegroupH (Yet, Z(d)) 0 is finite for i =2d, 2d+2 and torsion for i =2d+2. 6 6 6 It follows easily that (37) is a quasi-isomorphism. It remains to check the commutativity of the square above. The complex
DY := RHom(R (Y,Q(d)) 0, Q[ 2d 2]) R (Yet, Q[ 2]) ' is concentrated in degree 2. It follows that both aY and ↵Y uniquely factor through the truncated complex R (Yet, Z) 2. It is therefore enough to show that the square
R (Yet, Q[ 2]) / R (Yet, Z) 2 Id ✏ ✏ RHom( (Y,Q(d)) 0, Q[ 2d 2]) / R (Yet, Z) 2 commutes. The exact triangle D Z[0] R (Yet, Z) 2 ⇡1(Yet) [ 2] Z[1] ! ! ! induces an exact sequence of abelian groups D Hom (DY , Z[0]) Hom (DY , R (Yet, Z) 2) Hom (DY ,⇡1(Yet) [ 2]) D ! D ! D which shows that the map D Hom (DY , R (Yet, Z) 2) Hom (DY ,⇡1(Yet) [ 2]) D ! D is injective. Indeed, Z[0] [Q Q/Z] has an injective resolution of length one and DY is ' ! concentrated in degree 2, hence Hom (DY , Z[0]) = 0. In view of the quasi-isomorphism D 2d 0 DY Hom(H (Y,Q(d)), Q)[ 2] H (Y,Q)[ 2], ' ' 20 BAPTISTE MORIN one is reduced to show the commutativity of the following square (of abelian groups):
d0,1 0 2 2 H (Yet, Q) / H (Yet, Z)
Id ✏ H2(↵ ) ✏ 2d Y 2 Hom(H (Y,Q(d)), Q) / H (Yet, Z)
2 By construction the map H (↵Y ) is the following composition 2d d d D 2 Hom(H (Y,Q(d)), Q) Hom(CH (Y ), Q) Hom(CH (Y ), Q/Z) ⇠ ⇡1(Yet) H (Yet, Z) ' ! ' where the isomorphism CHd(Y )D ⇠ ⇡ (Y )D is the dual of the map CHd(Y ) ⇡ (Y )ab 1 et ! 1 et given by class field theory, which is injective with dense image (see [48] Corollary 3). The top 0,1 horizontal map in the last commutative square is the differential d2 of the spectral sequence i j i+j H (Yet,R ( )Z) H (YW , Z). ⇤ ) There is a canonical isomorphism 0 1 H (Yet,R ( )Z)=lim Hom(Wk , Z)=Hom(Wk, Q), k0/k 0 ⇤ ! k0/k runs over the finite extensions of k, as it follows from the isomorphism of pro-discrete groups ⇡ (Y ,p) ⇡ (Y ,p) W , which is valid for any Y connected étale over Y . Then 1 W0 ' 1 et0 ⇥Gk k 0 the left vertical map in the last square above is the map d deg⇤ : Hom(Wk, Q) Hom(CH (Y ), Q) ! induced by the degree map d deg : CH (Y ) Z Wk ! ' 0,1 and the differential d2 is the following map: D Hom(Wk, Q)=Hom(⇡1(YW ,p), Q) Homc(⇡1(YW ,p), Q/Z) ⇡1(Yet,p) ! ' where Hom ( , ) denotes the group of continuous morphisms. One is therefore reduced to c observe that the square d0,1 2 D Hom(Wk, Q) / ⇡1(Yet)
deg⇤ Id ✏ H2(↵ ) ✏ d Y D Hom(CH (Y ), Q) / ⇡1(Yet) commutes. ⇤ We shall need the following result in Section 4.3. Proposition 2.14. Let f : Y be a morphism of proper regular arithmetic schemes, such !X that is flat over Spec(Z) and Y has characteristic p.Assumethat has pure dimension d X X and that L( et,d) 0 holds. Then there exists a unique map in X D ˜ f ⇤ : R W ( , Z) R (YW , Z) W X ! ZETA FUNCTIONS AT s =0 21 which renders the following square commutative
R (Yet, Z) / R (YW , Z) O O ˜ fet⇤ fW⇤
R ( et, Z) / R W ( , Z) X X where f ⇤ is induced by the map Yet et and R (YW , Z) is the cohomology of the Weil-étale et ! X topos. Proof. We shall define a morphism of exact triangles
aY R (Y,Q)[ 2] / R (Yet, Z) / R (YW , Z) O O O
0 fet⇤ ↵ RHom(R ( , Q(d)) 0, Q[ ]) X / R ( et, Z) / R W ( , Z) X X X ˜ where the left vertical map is the zero map and := 2d +2. The existence of fW⇤ will follow ˜ from the commutativity of the left square and the uniqueness of fW⇤ will follow from the facts that R (YW , Z) and R W ( , Z) are both perfect (by [34] Theorem 7.4 and Proposition 2.10 X respectively) and that RHom(R ( , Q(d)) 0, Q[ ]) is a complex of Q-vector spaces, as in X the proof of Theorem 2.11. In order to show that such a morphism of exact triangles does exist, we only need to check that the composite map
↵ fet⇤ (38) RHom(R ( , Q(d)) 0, Q[ ]) X R ( et, Z) R (Yet, Z) X ! X ! is the zero map. The complex of Q-vector spaces D := RHom(R ( , Q(d)) 0, Q[ ]) is X X 0 concentrated in degrees 3, because is flat (see (6)). Moreover, we have H (Yet, Z)= ⇡0(Y ) 1 2 X ⇡0(Y ) Z , H (Yet, Z)=0, H (Yet, Z) (Q/Z) A is the direct sum of a divisible group i ' and a finite group A,andH (Yet, Z) is finite for i>2. The spectral sequence p i q+i p+q Ext (H (D ),H (Yet, Z)) H (RHom(D , R (Yet, Z))) X ) X i Z Y2 then shows that 0 Hom (D , R (Yet, Z)) = H (RHom(D , R (Yet, Z))) = 0. D X X Hence (38) must be the zero map, and the result follows. ⇤ In cases where Theorem 2.11 and Proposition 2.14 both apply, we have a commutative diagram: ˜ fW⇤ R W ( , Z) / R (YW , Z) X
f ' W⇤ ' ✏ R W (Y,Z) ˜ where fW⇤ is the map defined in Theorem 2.11, fW⇤ is the map defined in Proposition 2.14 and the vertical isomorphism is defined in Theorem 2.13. Indeed, up to the identification given by Theorem 2.13, the map fW⇤ sits in the commutative square of Proposition 2.14, hence must ˜ coincide with fW⇤ . 22 BAPTISTE MORIN
2.5. Relationship with Lichtenbaum’s definition for number rings. In this section we consider a totally imaginary number field F and we set = Spec( ). The complex X OF R W ( , Z) is well defined since L( et, Z(1)) holds (see Theorem 5.1). X X Theorem 2.15. There is a canonical isomorphism in D R W ( , Z) ⇠ R ( W , Z) 3 X ! X where R ( W , Z) 3 is the truncation of Lichtenbaum’s complex [35] and R W ( , Z) is the X X complex defined in this paper. Proof. By [38] Theorem 9.5, we have a quasi-isomorphism
R ( et,RZ) ⇠ R ( W , Z) X ! X inducing
(39) R ( et,RW Z) ⇠ R ( W , Z) 3 X ! X where RZ is the complex defined in [38] Theorem 8.5 and RW Z := RZ 2. The complex R ( W , Z), defined in [35], is the cohomology of the Weil-étale topos W which is defined in X X [39]. We have an exact triangle 2 Z[0] RW Z R Z[ 2] ! ! W in the derived category of étale sheaves on .RotatingandapplyingR ( , ) we get an X X et exact triangle 2 R ( et,R Z)[ 3] R ( et, Z) R ( et,RW Z). X W ! X ! X We have canonical quasi-isomorphisms (see [38] Theorem 9.4 and [38] Isomorphism (35)) 2 R ( et,R Z)[ 3] Hom ( ⇥, Q)[ 3] RHom(R ( , Q(1)), Q[ 4]). X W ' Z OF ' X 2 It follows that the morphism R ( et,R Z)[ 3] R ( et, Z) in the triangle above is deter- X W ! X mined by the induced map 0 2 3 Hom ( ⇥, Q)=H ( et,R Z) H ( et, Z)=Hom ( ⇥, Q/Z) Z OF X W ! X Z OF and so is the morphism ↵ . In both cases this map is the obvious one. Hence the square X 2 R ( et,RW Z)[ 3] / R ( et, Z) X O XO Id '
RHom(R ( , Q(1)), Q[ 4]) / R ( et, Z) X X is commutative. Hence there exists an isomorphism R W ( , Z) R ( et,RW Z). The X ' X uniqueness of this isomorphism can be shown as in the proof of Theorem 2.11. Composing this isomorphism with (39), we obtain the result. ⇤
3. Weil-étale Cohomology with compact support We recall below the definition given in [14] of the Weil-étale topos and some results concern- ing its cohomology with R˜-coefficients. Then we define Weil-étale cohomology with compact support and Z-coefficients and we study the expected map from Z to R-coefficients. ZETA FUNCTIONS AT s =0 23
3.1. Cohomology with R˜-coefficients. Let be any proper regular connected arithmetic X scheme. The Weil-étale topos is defined as a 2-fiber product of topoi
W := et Spec(Z)W . X X ⇥Spec(Z)et There is a canonical morphism
f : W Spec(Z) B X ! W ! R where BR is Grothendieck’s classifying topos of the topological group R (see [24] or [14]). Consider the sheaf yR on BR represented by R with the standard topology and trivial R- action. Then one defines the sheaf R˜ := f⇤(yR) on . By [14], the following diagram consists of two pull-back squares of topoi, and the X W rows give open-closed decompositions: