(b) The rows of (3.3) are induced by tensoring with R the exact sequences
0→ a
S<∞
Hi(k(v)W,Z)→Hci+1(YW,Z)→Hci+1((SpecOF)W,Z)→0, i= 0,1.
Letχ0 :=χc
(0, H0(k(v)W,Z), H1(k(v)W,Z)),(`ψ)
. Applying Example 3.2 to (3.3), we see that
χc(Y) = ±χc(SpecOF)χ0.
Because Hi(k(v)W,Z) is torsion free, and we has seen in (a) that`ψ sends 1v
tolv :σ7→logN(v), so χ0 =±ΠS<∞logN(v). Thus,
χc(Y) =±Rh w
Y
S<∞
logN(v) =±RShS w
(c.f. [15] Lemma 2.1 for the last equality).
Proof. Consider the spectral sequence
E2pq = ExtpZ(H−q(C•),Z)⇒Hp+q(RHom(C•,Z)),
in which C• =τ≤3RΓc(YW,Z).
It is clear that E2pq = 0 for all p < 0 or q > 0, i.e. the E2 has only non-trivial terms on the 4th quadrant
0 0
XS =XS∨∨ 0
US/tor =US∨∨ Ext1((PicY)D,Z) 0 Ext1(µDF,Z) 0 We have exact sequences
0→PicY →H−1 →XS →0 and 0→µF →H−2 →US/tor →0.
Thus, H−1 = XS ⊕ Pic(Y) and H−2 = (US/tor) ⊕ µF = US. This shows that RHom(τ≤3RΓc(YW,Z),Z)[−2] has only two non-trivial cohomologies H0 = US and H1 =XS⊕PicY.
From now on we assume that S is large enough so that PicY = 0 and is stable under the action of G. We will prove RHom(τ≤3RΓc(YW,Z),Z)[−2] is indeed quasi- isomorphic to ΨS, the canonical Tate sequence. (see the introduction section).
Lemma 3.2. Let A and B ∈ D, the derived category of Z[G]-modules.
a) If Hi(B) areZ[G]-injective for all i, then
HomD(A,B)∼=Y
i
HomG(Hi(A), Hi(B)).
b) If Hi(A) areZ[G]-cohomologically trivial, then we have an exact sequence
0→Y
i
Ext1G(Hi(A), Hi−1(B))→HomD(A,B)→Y
i
HomG(Hi(A), Hi(B))→0.
Proof. a) By [16], there is a spectral sequence
Y
i
ExtpG(Hi(A), Hq+i(B))⇒Hp+q(RHomZ[G](A,B)).
By condition, Extp(Hi(A), Hq+i(B)) is non-trivial only if p = 0, so the above spectral sequence simply shows
HomD(A,B) =H0(RHomZ[G](A,B)) = Y
i
HomG(Hi(A), Hi(B)).
b) In the same spectral sequence, since Hi(A) are cohomologically trivial, Hi(A) are of projective dimension 1, so ExtpG(Hi(A), Hq+i(B)) = 0 for all p ≥ 2.
Therefore, on the position with p+q = 0, we have only 2 non-trivial groups E20,0 andE21,−1, and both the difference maps are 0. This gives the desired short exact sequence.
Lemma 3.3. The canonical exact triangle
RHomZ(RΓc(Yet´,Z),Z)→RHomZ(RΓc(Y´et,Z),Q)→RHomZ(RΓc(Y´et,Z),Q/Z)→
is isomorphic to the exact triangle
Ψ0S[2]→XS⊗ZQ[1]→ΨfS0[3]→
whereΨ0S is the image ofΨS inExt2G(XS,UbS)andΨfS
0 is the image ofΨS inExt3G(XS⊗Z Q/Z,UbS).
Proof. Suppose ΨS is the complex A → B and ΨeS is the complex A → B →
XS ⊗Z Q. Recall that ΨS presents a class of Ext2G(XS,US) and Ψ0S the image of ΨS under the map Ext2G(XS,US) −→∼ Ext2G(XS,UbS). Indeed , Ψ0S is the complex (A⊕UˆS)/US =: A0 → B. Similarly, Ψe0S is the complex A0 → B → XS ⊗Z Q. Re- call that RHom(RΓc(Yet´,Z),Q/Z)−→qis Ψe0S by definition ([2] Prop. 3.1). Also, as Q is injective, Hi(RHomZ(RΓc(Y´et,Z),Q)) = HomZ(Hc−i(Y´et,Z),Q). Thus,
RHomZ(RΓc(Y´et,Z),Q)−→qis XS⊗ZQ[1].
We claim that the following diagram commutes for a suitable choice of the left vertical quasi-isomorphism.
RHomZ(RΓc(Yet´,Z),Q)[−3] //RHomZ(RΓc(Yet´,Z),Q/Z)[−3]
XS⊗ZQ[−2] //
qis
OO
Ψe0S.
qis
OO (3.4)
Since the both horizontal maps induce the canonical projectionXS⊗ZQ→XS⊗ZQ/Z onH2, the right vertical quasi-isomorphism induces the identity map on H2 and the left vertical one can be induced by any G-automorphism of XS ⊗ZQ, so there is a suitable choice of the quasi-isomorphism so that the above diagram commutes forH2 groups.
Let A be XS⊗ZQ[−2] and B be RHomZ(RΓc(Yet´,Z),Q/Z)[−3]. Clearly, Hi(A) are Z[G]-cohomologically trivial. So we can apply lemma 3.2(b) to A and B, and it induces an exact sequence
0 = Ext1G(XS⊗ZQ, H1(B)(= 0))−→HomD(A,B)−→HomG(H2(A), H2(B))→0.
Thus, HomD(A,B) ∼= HomG(H2(A), H2(B)). Therefore, the commutative on H2 implies the diagram (3.4) indeed commutes. By taking exact triangles of each row, and note that cone(XS⊗ZQ[−2]→Ψe0S)[−1]'Ψ0S[2], we get the desired isomorphism of exact triangles.
Theorem 3.3. Let Y = SpecOF,S, where S is large enough so that PicY = 0 and is
stable under the action of G. Then
RHomZ(τ≤3RΓc(YW,Z),Z)[−2]−→qis ΨS,
where ΨS is the Tate sequence representing the canonical class of Ext2G(XS, US).
Proof. Here, we use the same notation with previous lemma. Clearly, we have a canonical exact triangle
ΨS →Ψ0S →UbS/US[0]→.
On the other hand, by applying RHomZ(−,Z) to exact triangle (3.1), we have another exact triangle
RHomZ(τ≤3RΓc(YW,Z),Z)[−2]→RHomZ(RΓc(Yet´,Z),Z)[−2]→RHomZ(HomZ(US,Q),Z)[−1]
The previous lemma ensures that Ψ0S−→qis RHomZ(RΓc(Yet´,Z),Z)[−2].
Also, observe that RHomZ(HomZ(US,Q),Z)[−1]−→qis UbS/US[0] and we have seen that RHomZ(τ≤3RΓc(YW,Z),Z)[−2] has the same cohomologies as ΨS. We claim that there is an isomorphism between these two exact triangles,
RHomZ(τ≤3RΓc(YW,Z),Z)[−2] //Ψ0S β //RHomZ(HomZ(US,Q),Z)[−1] //
ΨS
δ
OO //Ψ0S γ // bUS/US[0]
α
OO //.
In the top exact triangle, we replaced RHomZ(RΓc(Y´et,Z),Z)[−2] by Ψ0S. By taking the long exact sequence induced by the top exact triangle, we see thatH2(β) = i◦p where p : UbS → UbS/US is the canonical projection and i ∈ Aut(UbS/US). Clearly, H2(γ) = p. As RHomZ(HomZ(US,Q),Z)[−1]−→qis UbS/US[0], we may take α to be the morphism induced by i. Then the right square commutes on H2. By lemma 3.2, becauseUbS/USis uniquely divisible and thus injective,H2(β) =H2(α)◦H2(γ) implies β =α◦γ. Asαis a quasi-isomorphism, so isδ by property of derived categories.
Remark. It is equivalent to say that τ≤3RΓc(YW,Z)−→qis RHom(ΨS,Z)[−2] and this
is an evidence that, at least for open subschemes of spectra of number rings, the Weil-
´
etale cohomology can be obtained from usual cohomology theories without using the Weil groups.
Chapter 4
The Construction of R G m
Throughout this chapter, we assume that F is a totally imaginary number field, X = SpecOF, and U = SpecOL,S is any connected ´etale neighborhood of X. We also assume that K is a subfield of L such that the extension L/K is Galois and G:= Gal(L/K).
When P ic(U) = 0 and S is a G-stable set containing all the archimedean places and those ramified in K(U)/Q, there exists a canonical Tate sequence:
ΨU : Ψ0S →Ψ1S,
which represents a canonical class of Ext2G(XS, US). By the construction of Tate sequences, if U0 = SpecOL0,S0 is ´etale over U, then there is a morphism from ΨS to ΨS0 that fits into the following morphism of complexes
0 //US
//Ψ0S
//Ψ1S
//XS //
β
0
0 //US0 //Ψ0S0 //Ψ1S0 //XS0 //0, where β((av)v∈S) = (aw := [Lw :Fv]·av)w|v,w∈S0
.
In fact {XS} and {β} define an Xet´-presheaf X, and we denote its associated sheaf by Xf. In section 4.1, we construct a complex RGm ∈ Sh(X´et) which rep- resents a canonical class of Ext2X
et(Xf,Gm) = Zb (see below). The cohomologies of RΓ(U´et, RGm) and RHomZ τ≤3RΓc(UW,Z), Z
[−2] are the same for arbitrary U.
Furthermore, in section 4.2, we prove that
RΓ(U´et, RGm)−→qis RHom τ≤3RΓc(UW,Z),Z
[−2], (4.1)
inD(Z[G]), when S is stable under the action of G.
This implies RΓ(Uet´, RGm) and RHomZ τ≤3RΓc(UW,Z), Z
[−2] are canonically quasi-isomorphic in D(Z) for any U. Also, whenU is small enough so that the Tate sequence exists, then by (4.1) and by the help of Theorem 3.3, one can see that
RΓ(Uet´, RGm)'ΨU
inD(Z[G]).
Thus, one can conclude that the complex RΓ(U´et, RGm) generalizes the Tate sequences and itsZ-dual defines Weil-´etale cohomology ofS-integers without trunca- tions.
4.1 The Definition
LetM be the sheaf sending a connected ´etaleU →X to the group L
w∈UAw, where Aw is Q (resp. Z) if w -∞ (resp. w | ∞), and the transition map M(U)→ M(V) for an X-morphism V →U is
M
v∈U
Av −→ M
w∈V
Aw
av 7→ X
w|v
[K(V)w :K(U)v]av.
It is clear that the projectionsL
v∈UAv L
v∈UAv, for any ´etaleU →X, define an epimorphism of sheaves M →L
v∈Xiv,∗Av.
We define F• to be the complex
F0
P //F1
Q,
where F0 =ker(M →L
v∈Xiv,∗Av) is the X-´etale sheaf U 7→ M
w∈U−U
Aw,
for any connected U →X and P
(U)((aw)) =P
w 1
[K(U)w:Fv]aw. Note that if U doesn’t contain all the finite places of F, then P
(U) is surjective.
This implies P
is an epimorphism in the category Sh(X´et) as any U → X can be covered by some {Ui}, where Ui doesn’t contain all the finite places ofK(Ui), for all i. Consequently, F• '(kerP
)[0] inD(Sh(X´et)).
Remark. a. The way that we define the transition maps of the sheaf F0 makes X a sub-presheaf of kerP
. In fact, we have the following exact sequence of presheaves
0→X →kerX
→Brf →0,
where Brf is the presheaf U 7→H2(U,Gm). By taking associated sheaves, we see that Xf∼= kerP
as a(Hi(−,G)) = 0 for any sheaf G wheni≥1.
b. Let Mv = lim−→LML,v where L runs over all finite extension of F and ML,v = L
wAw where the sum runs over all placeswofLlying overv and the transition map is defined similarly to those of X. We denote Mv the sheaf associated to the Galois module Mv. Equivalently, in the level of sheaves, Mv = lim−→LπL,∗Av
where πL : SpecLv −→ SpecF is the canonical morphism and Lv is the fixed field of Dv ∩Gal(L/F). Clearly, M = L
v∈Xα∗Mv, where α : SpecF → X.
Moreover, consider the Cartesian diagram of schemes
SpecLv
πL
αL //u(Lv)
πL0
SpecF α //X,
whereu(Lv) = SpecOLv. Henceα∗lim−→LπL,∗Av = lim−→Lα∗πL,∗Av = lim−→LπL,∗0 αL,∗Av = lim−→Lπ0L,∗Av. (Note that αL,∗Av is the constant Av on Spec(Lv).)
Lemma 4.1. The sheaf F0 is cohomologically trivial.
Proof. Since F0 fits into the short exact sequence
0→F0 →M
v∈X
α∗Mv →M
v∈X
iv,∗Av →0, (4.2)
which is derived from the short exact sequence of sections:
0→ M
v∈U−U
Av →M
v∈U
Av →M
v∈U
Av →0.
Since there is no real place and Av = Q when v - ∞, the cohomology of iv,∗Av concentrates at degree zero. Also, the above exact sequence remains exact when we pass to global sections. Thus, we only need to show that α∗Mv is cohomologically trivial for any v.
Recall that α∗Mv = lim−→LπL,∗0 Av. Whenv is finite,α∗Mv is acyclic asAv =Qand inductive limit and πL,∗0 both preserve acyclic sheaves (as π0L is finite).
Whenv|∞,Iv is trivial, soMv =IndG1FAv and thenMv is cohomologically trivial.
Therefore i∗w(Rqα∗Mv) =Hq(Iw, Mv) = 0 for q >0. The Leray spectral sequence for α,
Hp(U´et, Rqα∗Mv)⇒Hp+q(F, Mv),
for any ´etale U over X, is degenerated, i.e. Hp(U´et, α∗Mv) = Hp(F, Mv) = 0 for p >0. Thus the result follows.
Proposition 4.1. For any connected ´etale U →X,
Hp(Uet´,F•) =
ker Σ(U) p= 0,
coker Σ(U) = Q/Im(Σ(U)) p= 1,
0 p≥2.
In particular, H0(X´et,F•) = 0, H1(X´et,F•) = Q, H0((X − v)´et,F•) = 0 and H1((X−v)´et,F•) = Q/Av.
Corollary 4.1. The canonical morphism Hv1(X´et,F•) → H1(X´et,F•) is an inclu- sion of Av into Q.
Now we define a complex RGm, up to quasi-isomorphism, ofSh(X´et) by an exact triangle
Gm→RGm →F•[−1]→, or equivalently an exact triangle
RGm →F•[−1]−→γ Gm[1]→.
One can choose carefully a morphismγ ∈HomD(Sh(X
´
et))(F•[−1],Gm[1]) ∼= Ext2X
´
et(F•,Gm) so that RGm has the cohomologies that we expected. Before computing the group Ext2X
´
et(F•,Gm), we need to derive the following lemmas.
Lemma 4.2. Let I be an inductive system of index that can be refined by an in- dex which is isomorphic to Z, then the canonical morphism ExtsC(lim−→iAi, B) −→
lim←−iExtsC(Ai, B) is an isomorphism if any of the following conditions holds a. the projective system {Exts−1C (Ai, B)} satisfies the Mittag-Leffler condition, b. Exts−1C (Ai, B) is a finite group for all i.
Proof. Because I can be refined by Z, so lim−→IAi = lim−→ZAi and lim←−IExtpC(Ai, B) = lim←−ZExtpC(Ai, B). Thus, we only need to consider the case that I ∼=Z. When I ∼=Z,
by [13], there is a spectral sequence
lim←−rIExtsC(Ai, B)⇒Extr+sC (lim−→IAi, B).
This spectral sequence is certainly degenerated as we know that lim←−
r is vanishing for any inductive system of abelian groups when r ≥ 2 (cf. [17] 3.5). Therefore, we get exact sequences
0→lim←−1IExts−1C (Ai, B)→ExtsC(lim−→IAi, B)→lim←−1IExtsC(Ai, B)→0,
for all s≥0.
The Mittag-Leffler condition implies lim←−1Exts−1C (Ai, B) = 0 and condition (b) also implies the vanishing of lim←−
1Exts−1C (Ai, B) by [7] 2.3. Thus, the isomorphism is established under both conditions.
Lemma 4.3. a.
Extp
Xet´(Q,Gm) =
0 p= 2,
ZbN
ZQ=Q0
qQq p= 3, where Q0
qQq = AQ/R is the restricted product of the Qp’s with respect to the Zp’s.
b.
Extp
Xet´(α∗Mv,Gm) =
0 p= 2,
H3(X´et,Gm) = Q/Z p= 3and v | ∞, lim←−nH3(X,Gm) =ZbN
ZQ p= 3and v -∞.
Proof. a. Since Q= lim−→nZ andHi(X´et,Gm) are finite groups fori= 1,2, we have
Extp
Xet´(Q,Gm) = lim←−nHp(Xet´,Gm).
AsH2(X´et,Gm) = 0, one sees that Ext2X
´
et(Q,Gm) = 0. Also, lim←−nH3(Xet´,Gm) =
lim←−nQ/Z= Hom(Q,Q/Z) =ZbN
ZQ. b. As Mv = lim−→LπL,∗0 Av,
Extp
Xet´(α∗Mv,Gm) = Extp
X´et(lim−→Lπ0L,∗Av,Gm).
When v is an infinite place, Av =Z, and by the norm theorem,
Extp
X´et(π0L,∗Z,Gm) = Hp(u(Lv)et´,Gm),
for all p. Note that F is totally imaginary and u(Lv) contains all the fi- nite places of Lv, so H1(u(Lv)´et,Gm) = P ic(u(Lv)), H2(u(Lv)´et,Gm) = 0, and H3(u(Lv)´et,Gm) = Q/Z. Since H1(u(Lv)et´,Gm) and H2(u(Lv)´et,Gm) are both finite groups, by Lemma 4.2, we actually get
Extp
X´et(α∗Mv,Gm) = lim←−LHp(u(Lv)et´,Gm) for p = 2,3. Clearly, Ext2X
´
et(α∗Mv,Gm) = H2(u(Lv)´et,Gm) = 0. Moreover, for any finite extension K over L, the transition map H3(u(Kv)et´,Gm) → H3(u(Lv)´et,Gm) is an isomorphism. So, we conclude that Ext3X
´
et(α∗Mv,Gm) = H3(X´et,Gm) = Q/Z.
For the case v is a finite place of F, there is a slightly difference, as nowAv = Q= lim−→ 1nZ and so α∗Mv = lim−→L,nπ0L,∗Z. We claim that
Extp
X´et(lim−→L,nπ0L,∗Z,Gm) = lim←−L,nHp(u(Lv)et´,Gm),
for p = 2,3. Indeed, because Ext1u(Lv)et´(πL,∗0 Z,Gm) = H1(u(Lv)´et,Gm) = P ic(u(Lv)) is a finite groups, and Ext2u(Lv)et´(πL,∗0 Z,Gm) = H2(u(Lv)´et,Gm) = 0, by Lemma 4.2, one gets Extp
Xet´(α∗Mv,Gm) = lim←−L,nHp(u(Lv)et´,Gm) for p = 2,3. Again, by Lemma 4.2, one get ExtpX
´
et(Mv,Gm) = lim←−L,nExtpu(Lv)(Z,Gm)
for p= 2,3. Hence
Extp
Xet´ (Mv,Gm) =
0 p= 2,
lim←−nH3(Xet´,Gm) p= 3.
Proposition 4.2.
ExtpX
´
et(F0,Gm) =
0 p= 2,
Q
v|∞H3(X´et,Gm) = Q
X∞Q/Z p= 3.
Proof. LetFv0 :=ker(α∗Mv →iv,∗Av). Then it arises a long exact sequence
∂p−1
−−−→ExtpX
´
et(α∗Mv,Gm)→ExtpX
´
et(Fv0,Gm)→ExtpX
´
et(iv,∗Av,Gm)−→∂p Extp+1X
´
et(α∗Mv,Gm)→. Since Ext2X
´
et(α∗Mv,Gm) = 0 by the previous lemma, and it is easy to see that Ext4X
´
et(iv,∗Av,Gm) = 0, we have the following exact sequence, 0→Ext2X
´
et(Fv0,Gm)→Ext3X
´
et(iv,∗Av,Gm)→Ext3X
´
et(α∗Mv,Gm)→Ext3X
´
et(Fv0,Gm)→0.
By the adjunction of iv,∗ and i!v and Lemma 4.2 (asHv2(Xet´,Gm) = 0),
Ext3X
´
et(iv,∗Av,Gm) =
Hv3(Xet´,Gm) = 0 v | ∞, lim←−nHv3(Xet´,Gm) =Q0
qQq v -∞.
Therefore, Ext2X
´
et(Fv0,Gm) = 0 and Ext3X
´
et(Fv0,Gm) = Ext3X
´
et(α∗Mv,Gm) =Q/Z when v|∞.
We claim that the morphism Ext3X
´
et(iv,∗Av,Gm) → Ext3X
´
et(α∗Mv,Gm) is an iso- morphism when v - ∞. For this, observe that there is a commutative diagram of
sheaves
α∗Mv = lim−→LπL,∗0 Av //iv,∗Av
Av =πF,∗0 Av
OO 66nnnnnnnnnnnnn
where Av → iv,∗Av is the canonical morphism induced by adjunction of i∗v and iv,∗. Applying Ext3(−,Gm) to the above diagram, we get another commutative diagram :
Ext3(α∗Mv,Gm)
lim←−nHv3(X´et,Gm)
uukkkkkkkkihkkkkkk
oo
lim←−nH3(X´et,Gm).
All the groups in the above diagram are isomorphic to ZbN
ZQ = Q
qQq. Note the h is an isomorphism as the natural morphismHv3(Xet´,Gm)→H3(Xet´,Gm) is an isomorphism ([4] Prop.3.2). Therefore, i is an injection from Q
qQq to itself. Since HomZ(Qp,Qq) = 0 isp6=qandQqare fields, one sees thatihas to be an isomorphism.
It follows that Ext2X
´
et(Fv0,Gm) and Ext3X
´
et(Fv0,Gm) are vanishing.
Consequently, Ext2X
´
et(F0,Gm) = 0 and Ext3X
´
et(F0,Gm) = M
v∈X
Ext3X
´
et(Fv0,Gm) =M
v|∞
Q/Z
Proposition 4.3. There is a canonical isomorphism Ext2X
´
et(F•,Gm)∼=Zb. Proof. The exact sequence
0−→kerX
−→F0 −→Q−→0,
induces a long exact sequence
Ext2X
´
et(F0,Gm)(= 0)→Ext2X
´
et(kerX
,Gm)→Ext3X
´
et(Q,Gm)−→∆ M
v|∞
H3(X´et,Gm)→,
where ∆ is a copies of the canonical map π: Hom(Q,Q/Z)→Hom(Z,Q/Z).
Clearly, ker ∆ = kerπ = HomZ(Q/Z,Q/Z) =Zb. As a consequence Ext2X
´
et(kerP
,Gm) = ker∆ = ker(π) = Zb.
Let γ be the class in Ext2X
´
et(F•,Gm) that corresponds to the generator 1 of Zb, and RGm is the complex decided by the exact triangle
RGm →F•[−1]−→γ Gm[1]→.
To compute the ´etale cohomology of the complex RGm, we need to determine the coboundary morphisms ∂Ui :Hi(U´et,F•)−−→^γ Hi+2(Uet´,Gm), for any connected ´etale U →X.
Lemma 4.4. For any connected ´etale U →X, let S :=U −U, we have a. ∂U0 is the canonical projection
(M
v∈S
Av)Σ=0 →(M
v∈S
Av/Z)Σ=0,
b. ∂U1 is an isomorphism if S is non-empty, and it is the canonical projection Q→Q/Z otherwise.
c. ∂Ui is isomorphic for all i≥2.
In particular, all the ∂i are surjective for all i≥0.
Proof. We first deal with the second part of (b). Consider the following commutative
diagram
H1(Xet´, kerP
) × Ext2X
´
et(kerP ,Gm)
^ //H3(X´et,Gm)
H0(Xet´,Q)
o
OO
× Ext3X
´
et(Q,Gm) ^ //H3(X´et,Gm) lim−→H0(Xet´,n1Z) × lim←−Ext3X
´
et(1nZ,Gm)^ //H3(X´et,Gm)
It is easy to see that the bottom cup product sends pair (a,1) to (a mod Z) and thus a ^ γ = a ^ 1 = a modZ. This shows ∂1 : H1(X´et,F•) → H3(Xet´,Gm) is the canonical projection Q → Q/Z. The general case that U = U follows from the commutative diagram
H1(X´et,F•)
=
^γ modZ
//H3(X´et,Gm)
=
H1(U´et,F•) ^γ //H3(U´et,Gm).
For a Zariski open subset j :U ,→X, set i: X−U ,→X, we have the following commutative diagram
0 //H0(U´et,F•) //
^γ
L
v /∈UHv1(X´et,F•)
^γ
P //H1(X´et,F•)
^γ
//H1(U´et,F•)
^γ
//0
0 //H2(Uet´,Gm) //L
v /∈UHv3(X´et,Gm)
P //H3(Xet´,Gm) //H3(Uet´,Gm) //0,
where the rows are exact sequences induced by the short exact sequence
0−→j!j∗F −→F −→i∗i∗F −→0, for any F ∈Sh(X).
The morphism H0(Uet´, kerP
) −−→^γ H2(U´et,Gm) is completely determined by H1(X´et, kerP
)−−→^γ H3(X´et,Gm) because Σ restricts to an inclusion onHv1(Xet´,F•) = Av. (This can be seen by putU to beX−v and use the factH0((X−v)et´, kerP
) =
0). Therefore, by the commutativity, ^ γ : H0(Uet´,F•) → H2(U´et,Gm) maps (L
v∈U−UAv)Σ=0 to (L
v∈U−UAv/Z)Σ=0 by taking modular by Z componentwise.
For general i : U → X, set Ua to be the largest open subscheme of U such that for any v ∈ Ua, all the valuations of K(U) above i(v) is contained in Ua. We have the commutative diagram
H0(U´et _,F•) ^γ //
H2(U´et _,Gm)
H0(U´eta,F•) ^γ //H2(U´eta,Gm)
H0(i(Ua)et´,F•) ^γ //
OO
H2(i(Ua)et´,Gm)
OO
More precisely,
L
w∈U−U
AwΣ=0
^γ //
_
L
w∈U−U
Aw/Z Σ=0
_
L
w∈U−Ua
Aw Σ=0
^γ //
L
w∈U−Ua
Aw/Z Σ=0
L
v∈X−i(Ua)
AvΣ=0
^γ//
OO
L
v∈X−i(Ua)
Av/Z Σ=0
OO
Note that the cup products are canonical on v and w, and we have seen above that the bottom cup product is canonical projection. By chasing diagram, one sees that the middle one is also canonical projection, and so is the top one as the injection is canonical. It follows that ∂U0 is the canonical projection.
To see the first part of part (b), we assume S is non-empty. When Uf * U, H3(U´et,Gm) = 0 and H1(Uet´,F•) = 0 (Prop. 4.1). Therefore, we have ∂U1 = 0. For the case Uf ⊆ U, note that U is then ´etale over X as all the infinite places are
complex. Therefore, we have the following commutation diagram
Q=H1(Uet´,F•) π //
∂1
U
H1(U´et,F•) =Q/Z
∂U1
Q/Z=H3(U´et,Gm) H3(U´et,Gm) = Q/Z.
As we have seen, in the very beginning of the proof, that ∂U1 and π are the canonical projections, ∂U1 has to be the identity map on Q/Z by the commutative.
Part (c) is trivial because Hi(U´et,F•) =Hi+2(Uet´,Gm) = 0 for all i≥2.
Now, we are ready to compute the ´etale cohomology of the complex RGm. Proposition 4.4. Let U be ´etale over X, S =U −U and L=K(U), then
Hp(U´et, RGm) =
O×L,S p= 0, Pic(OL,S)⊕XS p= 1,
0 p= 2,
Z(resp.0) p= 3 and S =∅(resp. S 6=∅),
0 p≥4.
Hence, these cohomologies coincide with those of theZ-dual ofτ≤3RΓc(UW,Z)[2] (c.f.
Prop. 3.5).
Proof. Consider the usual long exact sequence on cohomology induced by the exact triangle RGm → F•[−1] −→γ Gm[1] → and use the result that ∂i are surjective (Lemma 4.4), we get
H0(Uet´,Gm)−→∼ H0(U´et, RGm),
0→H1(Uet´,Gm)→H1(Uet´, RGm)→ker(∂0)→0 is exact, Hi(Uet´, RGm) = ker(∂i−1).
Thus,H0(Uet´, RGm) = O×L,S. Because ker∂0 = (⊕v∈SZ)Σ=0 =XS isZ-free, the mid- dle exact sequence splits and soH1(Uet´, RGm) = P ic(OL,S)⊕XS. Also,H3(Uet´, RGm) =
0 (resp. Z) when S 6= ∅ (resp. S = ∅) and Hi(U´et, RGm) = 0 for all i ≥ 4, by Lemma 4.4 (b) and (c).
The computations of Prop.4.4 and 3.5 suggest that
Theorem 4.1. RΓ(U´et, RGm)−→qis RHom τ≤3RΓc(UW,Z),Z
[−2] in D(Z).
In the next section, we shall prove the following more general quasi-isomorphism
RΓ(U´et, RGm)−→qis RHom τ≤3RΓc(UW,Z),Z [−2],
in D(Z[G]), when S is stable under the action of G. Note that when G is the trivial group, this is just Theorem 4.1