This paper introduces a framework for non-commutative schemes based on prime ideals, explores their cohomology theories, and establishes finiteness and comparison theorems, extending classical algebraic geometry concepts to non-commutative settings.
Contribution
It defines non-commutative schemes via prime ideals, develops their cohomology theories, and proves key finiteness and comparison theorems for these schemes.
Findings
01
Finiteness theorem for higher direct images in étale cohomology.
02
Comparison theorem between étale and Betti cohomology.
03
Expression of L-functions via étale cohomology with compact supports.
Abstract
We define non-commutative schemes by using prime ideals of non-commutative rings, and discuss the \'etale cohomology, the Betti cohomology, and the fundamental groups of non-commutative schemes. For non-commutative schemes which are finite over centers, we prove the finiteness theorem for the higher direct images in \'etale cohomology theory, and the comparison theorem between \'etale cohomology and Betti cohomology. In Appendix, for non-commutative schemes over finite fields which are finite over centers and satisfy a certain condition, L-functions are expressed by using \'etale cohomology with compact supports.
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsHomotopy and Cohomology in Algebraic Topology · Algebraic structures and combinatorial models · Algebraic Geometry and Number Theory
Full text
Attempts on SGA for non-commutative rings. By Kazuya Kato.
With Appendix by Takako Fukaya and Kazuya Kato.
Abstract
We define non-commutative schemes by using prime ideals of non-commutative rings, and discuss the étale cohomology, the Betti cohomology, and the fundamental groups of non-commutative schemes. For non-commutative schemes which are finite over centers, we prove the finiteness theorem for the higher direct images in étale cohomology theory, and the comparison theorem between étale cohomology and Betti cohomology. In Appendix, for non-commutative schemes over finite fields which are finite over centers and satisfy a certain condition, L-functions are expressed by using étale cohomology with compact supports.
§9. Appendix: Zeta and L-functions of non-commutative schemes By Takako Fukaya
0 Introduction
0.1**.**
For a ring A which need not be commutative, let Spec(A) be the set of all prime ideals of A.
Then for ring homomorphism h:A→B and p∈Spec(B), h−1(p) need not be a prime ideal of A, unlike the case of commutative rings.
This is probably the reason of the fact that
in non-commutative algebraic geometry, people do not use the space Spec(A) so much.
However, Procesi [16] shows that if h satisfies the condition
(*) As a ring, B is generated by h(A) and the centralizer C_{B}(A)=\{b\in B\;|\;h(a)b=bh(a)\;\text{for all a\in A}\} of A in B,
then we have a map
[TABLE]
We say B is an A-algebra if this condition (*) is satisfied. We will show that if B and C are A-algebras, the tensor product B⊗AC has a unique ring structure such that we have (b1⊗c1)(b2⊗c2)=(b1b2)⊗(c1c2) if b1,b2∈B and c1,c2∈C and if either b1,b2∈CB(A) or c1,c2∈CC(A) (Theorem. 1.5). Many things about algebras over commutative rings are generalized to A-algebras in this sense.
Consider the category (Alg) of rings in which a morphism is a ring homomorphism A→B such that B is an A-algebra. Using prime ideals of rings, we will define a category of non-commutative schemes which are endowed with sheaves of (not necessarily commutative) rings, and we will have a contra-variant functor
[TABLE]
from the full subcategory of (Alg) consisting of A which is finitely generated as an algebra over the center Z(A), to the category of non-commutative schemes. See Section 2.
0.2**.**
For a non-commutative scheme X,
we define in Section 5 the étale cohomology groups of X and in Section 7 the fundamental group of X, imitating SGA 4 and SGA 1, respectively, generalizing these groups for schemes.
The main results of this paper are the following (1)–(5).
We consider a condition (F) on a non-commutative scheme and a condition (FC) on a non-commutative scheme over C (cf. 4.10).
(i) If R is an excellent commutative ring (resp. finitely generated commutative C-algebra) and A is an R-algebra such that A is finitely generated as an R-module, every open subspace U of Spec(A) satisfies (F) (resp. (FC)).
(ii) For a non-commutative scheme X (resp. X over C), the condition (F) (resp. (FC)) is that X has an open covering whose every member is isomorphic to some U in (i).
The following (1) and (2) are the non-commutative versions of the finiteness theorem of Gabber (([10] Exp. XIII, Exp. XXI) and the comparison theorem of Artin (([1] vol. 3, Exp. XVI, 4)), and in fact proved by the reductions to their theorems.
(1) The finiteness theorem.
Theorem (Theorem 5.21). Let Xand Y be quasi-compact non-commutative schemes satisfying (F) and let f:X→Y be a morphism which is locally of finite presentation (3.4). Let F be a constructible sheaf (5.19) of abelian groups (resp. sets, resp. groups) on the étale site of X. Then Rf∗F (resp. f∗F, resp. R1f∗F) is constructible.
(2) Comparison theorem.
For a non-commutative scheme X over C satisfying (FC), we define a topological space Xcl (6.4). In the case X is a scheme of finite type over C, Xcl is the set X(C) of C-valued points of X endowed with the classical topology.
Theorem (Theorem 6.11). Let X and Y be quasi-compact non-commutative schemes over C satisfying (FC) and let f:X→Y be a morphism which is locally of finite presentation. Let F be a constructible sheaf of abelian groups (resp. sets, resp. groups) on the étale site of X. Then we have an isomorphism
[TABLE]
for every m (resp. for m=0, resp. for m=1), where fcl denotes the continuous map Xcl→Ycl induced by f and ()cl for sheaves denote the pullbacks to these spaces.
As a corollary of this theorem, for non-commutative schemes over C satisfying (FC),
we have a comparison theorem (Theorem 7.4) for fundamental groups.
(3) Proper base change theorem.
We also prove a kind of proper base change theorem (Theorem 5.27, Theorem 5.29) for non-commutative schemes satisfying (F) and for morphisms satisfying a certain condition (5.28.1).
(4) Chow groups.
In Section 8, we consider Chow groups of non-commutative schemes satisfying (F), and consider a relation of CH0 with class field theory (Theorem 8.13).
(5) Zeta and L-functions.
In Appendix with Takako Fukaya, basing on the study of Hasse zeta functions of non-commutative rings [5], the theory of zeta functions and L-functions for non-commutative schemes is given. It is shown that for a non-commutative scheme X over a finite field satisfying (F) and a certain condition, the zeta function and L-functions of X are expressed by using the compact support étale cohomology (Theorem 9.23). This should be a fragment of SGA 5 for non-commutative schemes.
Discussion of the authors on the zeta and L-functions for non-commutative rings was the motivation of the study of this paper.
Our hope is
to develop theories as in SGA (Séminaire de Géométrie Algébrique du Bois Marie) for non-commutative schemes.
However we have to remark a negative present aspect of our work. Strong results are obtained in this paper only for non-commutative schemes satisfying (F). It seems that more ideas are needed to treat non-commutative rings with small centers nicely.
0.3**.**
The authors thank Benson Farb and Nobushige Kurokawa for encouragements.
The authors are partially supported by NSF Award 1601861.
We assume a ring has the identity element written by 1 and ring homomorphisms respect 1.
1 Algebras over non-commutative rings
Let A be a (not necessarily commutative) ring.
1.1**.**
By an A-algebra, we mean a ring B endowed with a ring homomorphism A→B satisfying the following equivalent conditions (i)–(iv). Let CB(A) be the centralizer \{b\in B\;|\;ab=ba\;\text{in B for all}\;a\in A\} of A in B.
(i) As a ring, B is generated by the image of A and CB(A).
(ii) As a left A-module, B is generated by CB(A).
(iii) As a right A-module, B is generated by CB(A).
(iv) As a ring endowed with a ring homomorphism from A, B is isomorphic to A⟨Tλ(λ∈Λ)⟩/I for some index set Λ and
some two-sided ideal I of the non-commutative polynomial ring A⟨Tλ(λ∈Λ)⟩.
Here Tλ are indeterminates which do not commute with each other but commute with elements of A.
If A is commutative, this terminology A-algebra coincides with the usual terminology A-algebra.
1.2**.**
By a relatively commutative A-algebra (r.c. A-algebra for short), we mean a ring B endowed with a ring homomorphism A→B satisfying the following equivalent conditions (i)–(iv).
(i) As a ring, B is generated by the image of A and the center Z(B) of B.
(ii) As a left A-module, B is generated by Z(B).
(iii) As a right A-module, B is generated by Z(B).
(iv) As a ring endowed with a ring homomorphism from A, B is isomorphic to A[Tλ;λ∈Λ]/I for some index set Λ and for some two-sided ideal I of A[Tλ;λ∈Λ]. Here Tλ are indeterminates which commute with each other
and with elements of A.
If A is commutative, an r.c. A-algebra is nothing but a commutative A-algebra.
Proposition 1.3**.**
(1) If B is an A-algebra and C is a B-algebra, C is an A-algebra.
(2) If B is an r.c. A-algebra and C is an r.c. B-algebra, C is an r.c. A-algebra.
Proof.
These are seen by using the condition (iv) in 1.1 or 1.2.
∎
1.4**.**
We denote by (Rings) the usual category of rings. We denote by (Alg) (resp. (Algr.c) the category of rings for which a morphism is a ring homomorphism A→B such that B is an A-algebra (resp. an r.c. A-algebra).
For a ring A,
we denote by (Rings/A) the category of rings endowed with a ring homomorphism from A, in which a morphism is a ring homomorphism which commutes with the homomorphisms from A. Similarly, (Alg/A) (resp. (Algr.c/A)) denotes the category of objects over A in the category (Alg) (resp. (Algr,c)).
The following tensor product is related, through 1.12, to the base change (3.6, 3.10) in the category of non-commutative schemes.
Proposition 1.5**.**
Let B and C be A-algebras.
(1) There is a unique ring structure on B⊗AC such that (b1⊗c1)(b2⊗c2)=b1b2⊗c1c2 if b1,b2∈B and c1,c2∈C and if either b1,b2∈CB(A) or c1,c2∈CC(A).
(2) For an object D of (Rings/A), we have a bijection
[TABLE]
[TABLE]
[TABLE]
where an element h of the first set corresponds to an element (f,g) of the second set by
[TABLE]
Note that in the definition of the second set, we can replace “b∈CB(A),c∈CC(A)” by “b∈B,c∈CC(A)”, and also by “b∈CB(A),c∈C”, not changing the second set.
(3)
With the ring homomorphism B→B⊗AC;b↦b⊗1, B⊗AC is a B-algebra.
With the ring homomorphism C→B⊗AC;c↦1⊗c, B⊗AC is a C-algebra.
With the ring homomorphism A→B⊗AC;a↦a⊗1=1⊗a, B⊗AC is an A-algebra.
Proof.
Take an isomorphism B≅A⟨Tλ(λ∈Λ)⟩/I over A as in (iv) of 1.1. This isomorphism induces B⊗AC≅C⟨Tλ(λ∈Λ)⟩/IC where IC denotes the right ideal of C⟨Tλ(λ∈Λ)⟩ generated by the image of I. In the ring C⟨Tλ(λ∈Λ)⟩, elements of CC(A) commute with all elements in the image of B because they commute with Tλ and also with elements of A. Since C is generated by CC(A) and A as a ring, this shows that IC is a two-sided ideal of C⟨Tλ(λ∈Λ)⟩. Hence B⊗AC has a ring structure. By the definition of this ring structure, we have (b1⊗c1)(b2⊗c2)=b1b2⊗c1c2 for b1,b2∈B and c1,c2∈CC(A). Since C is generated by CC(A) and A as a ring, this shows that (b1⊗c1)(b2⊗c2)=b1b2⊗c1c2 for b1,b2∈CB(A) and c1,c2∈C. This proves (1). This also proves that we have the map h↦(f,g) from the first set to the second set in (2). If we have an element (f,g) of the second set
in (2), the map B⊗ZC→D;b⊗c↦f(b)g(c) induces a map h:B⊗AC→D. The induced map C⟨Tλ(λ∈Λ)⟩→D is a ring homomorphism because f(b)g(c)=g(c)f(b) for all b∈CB(A) and c∈C. Hence we have (2).
(3) is easily deduced from (1), (2).
∎
Remark 1.6**.**
We do not have (b1⊗c1)(b2⊗c2)=(b1b2⊗c1c2) for b1,b2∈B and c1,c2∈C in general. If this holds in the case A=B=C, then for all a1,a2∈A, we have in B⊗AC
[TABLE]
that is, a1a2=a2a1 in A=A⊗AA and this would imply that A is commutative.
Proposition 1.7**.**
Let the notation be as in 1.5. Then there is a unique isomorphism B⊗AC≅C⊗AB of A-algebras which sends b⊗c to c⊗b for all b∈CB(A) and c∈CC(A). For b∈B and c∈C, this sends b⊗c to c⊗b if either b∈CB(A) or c∈CC(A).
Proof.
By 1.5 (2), the functor (Rings/A)→ (Sets) represented by B⊗AC and that represented by C⊗AB are identified, and the isomorphism
B⊗AC≅C⊗AB which gives the identification is described as in 1.7.
∎
Remark 1.8**.**
The isomorphism in 1.7 does not send b⊗c to c⊗b for b∈B and c∈C in general. If this holds in the case A=B=C, then for a1,a2∈A, the element a1a2⊗1=a1⊗a2 of B⊗AC would be sent to the element a2⊗a1=a2a1⊗1 of C⊗AB, but
since B⊗AC=A=C⊗AB, this would imply that a1a2=a2a1 in A and hence that A is commutative.
Proposition 1.9**.**
If B is an A-algebra, the image of the center Z(A) of A in B is contained in Z(B).
Proof.
The image of an element of Z(A) in B commutes with elements of CB(A) and also with the images of all element of A. Hence it commutes with all elements of B.
∎
Corollary 1.10**.**
Let B be an r.c. algebra over A and let C be an A-algebra. Let h:B→C be a morphism in (Rings/A). Then h is a morphism in (Alg/A) if and only if h(Z(B))⊂Z(C).
Corollary 1.11**.**
Let B be an A-algebra. Then for a two-sided ideal I of A[Tλ;λ∈Λ], we have a bijection
[TABLE]
[TABLE]
Proposition 1.12**.**
Let B be an r.c. A-algebra and let C be an A-algebra.
(1) CB(A)=Z(B).
(2) The C-algebra B⊗AC is relatively commutative.
(3) The A-algebra B⊗AC is the push-out (i.e. co-fiber product) of B←A→C in the category (Alg).
(4) The category (Algr.c.) has push-outs. If B and C are r.c. A-algebras, B⊗AC is the push-out of B←A→C in (Algr.c).
Proof.
(1) Since B is generated by Z(B) and the image of A as a ring, an element of CB(A) commutes all elements of B.
(2) The ring B⊗AC is generated by Z(B)⊗1 and 1⊗C as a ring. Because B→B⊗AC is a morphism in (Alg) (1.5 (3)), Z(B)⊗1 is contained in the center of B⊗AC by 1.9.
(3) For a ring D and for a morphism B→D in (Alg), CB(A) is cent to Z(D) by (1) and by 1.9. By this, (3) follows from 1.5 (2).
(4) follows from (3).
∎
Proposition 1.13**.**
Assume that a ring A and a ring A′ are Morita equivalent. Then:
(1) The category of A-algebras and the category of A′-algebras are equivalent.
(2) The category of r.c. A-algebras and the category of r.c. A′-algebras are equivalent.
Proof.
There is a (A,A′)-bi-module P
which is finitely generated faithful and projective over these rings, and we have A′=EndA(P)∘ and A=End(A′)∘(P). Here ()∘ denotes the opposite ring. We have the functor from the category of A-algebras to the category of A′-algebras given by
B↦B′:=EndB(B⊗AP)∘. The converse functor is obtained by B′↦B=End(B′)∘(P⊗A′B′).
This functor induces also the equivalence in (2).
∎
1.14**.**
In the case A′ is the matrix algebra Mn(A) (n≥1), the equivalence in 1.13 is given by sending an A-algebra B to the Mn(A)-algebra B′=Mn(B)=A′⊗AB. The converse functor is given by B′↦CA′(B′).
Proposition 1.15**.**
Let A be a commutative ring and let A′ be an Azumaya algebra over A. Then the category of A-algebras and the category of A′-algebras are equivalent by the functor B↦B′=A′⊗AB from the former category to the latter category. The converse functor is given by B′↦B=CA′(B′). This equivalence induces an equivalence between the categories of r.c. algebras.
Proof.
Étale locally on Spec(A), A′ is isomorphic to Mn(A). Hence we are reduced to 1.14 by étale descent for the étale topology of Spec(A).
∎
2 Non-commutative schemes
2.1**.**
Recall that a prime ideal of a ring A (which may not be commutative) is a two-sided ideal p=A of A such that if I and J are two-sided ideals of A and if IJ⊂p, then we have either I⊂p or J⊂p.
Let Spec(A) be the set of all prime ideals of A. It has the Zariski topology (Jacobson topology). A closed subset is a set V(I)={p∈Spec(A)∣I⊂p} given for a two-sided ideal I of A. We have ∩I∈SV(I)=V(∑I∈SI) for a set S
of two-sided ideals of A and V(I)∪V(J)=V(IJ) for two-sided ideals I and J of A.
For f∈A, let D(f)={p∈Spec(A)∣f∈/p}. Then the topology of Spec(A) is the weakest topology for which D(f) is open for every f∈A. For p∈Spec(A), D(f) (f∈A) such that p∈D(f) form a base of neighborhoods of p.
2.2**.**
For a ring homomorphism A→B, the inverse image of a prime ideal of B need not be a prime ideal of A. A standard example is the ring homomorphism k×k→M2(k) for a commutative field k, the embedding as diagonal matrices. The ideal (0) of M2(k) is a prime ideal, but its inverse image is the ideal (0) of k×k which is not a prime ideal.
However, by Procesi [16], for a morphism h:A→B in (Alg), we have a map
Spec(B)→Spec(A);p↦h−1(p). This map is continuous.
2.3**.**
By a non-commutative space of prime ideals, we mean a triple (X,OX,(p(x))x∈X), where X is a topological space, OX is a sheaf of rings (which need not be commutative) on X, and p(x) for each x∈X is a prime ideal of the stalk OX,x satisfying the following conditions (i) and (ii).
(i) For each x∈X, the center Z(OX,x) of OX,x is a local ring, and Z(OX,x)∩p(x) is the maximal ideal of Z(OX,x).
(ii) If U is an open set of X and if f∈OX(U), the set {x∈U∣fx∈/p(x)} is open in U.
For non-commutative spaces of prime ideals X=(X,OX,(p(x))x∈X) and Y=(Y,OY,(p(y))y∈Y), a morphism X→Y means a pair of a continuous map f:X→Y and a homomorphism of sheaves of rings f−1(OY)→OX on X satisfying the following condition (iii).
(iii) For each x∈X with image y in Y, the ring homomorphism OY,y→OX,x is a morphism of (Alg), and p(y) is the inverse image of p(x).
2.4**.**
We consider the following
Condition (f) on a ring A:
As a Z(A)-algebra, A is finitely generated.
Proposition 2.5**.**
Let A be a ring satisfying (f) and let R be a flat commutative ring over Z(A). Then R is the center of R⊗Z(A)A.
Proof.
Take finitely many elements ai∈A (1≤i≤n)
which generate
A over Z(A) as a ring. We have an exact sequence 0→Z(A)→A→hAn, where h(x)=(aix−xai)i. Since R is flat over Z(A), the sequence 0→R→R⊗Z(A)A→h(R⊗Z(A)A)n is exact.∎
2.6**.**
For a ring A which satisfies (f) in 2.4, we have
a non-commutative space of prime ideals
(Spec(A),OA,(p(x))x∈Spec(A)), where OA and p(x) are as follows.
Define the sheaf OA of rings on Spec(A) as follows. Let Z(A) be the center of A, and let OZ(A) be the structure sheaf of Spec(Z(A)). Let OA be the inverse image of the sheaf OZ(A)⊗Z(A)A on Spec(Z(A)) under the canonical map Spec(A)→Spec(Z(A)) (2.2).
Let x∈X and let y be the image of x under Spec(A)→Spec(Z(A)) and let Z(A)y be the local ring of Z(A) at y. Then Z(A)y is the center of OA,x by 2.5. Let q be the prime ideal of A corresponding to x, and let p(x):=Z(A)y⊗Z(A)q⊂OA,x.
We denote this non-commutative space of prime ideals simply by Spec(A).
Remark 2.7**.**
We have the evident canonical ring homomorphism A→Γ(Spec(A),OA).
This map need not be bijective. See 2.12 (1).
Theorem 2.8**.**
Let A be a ring which is finitely generated as a Z(A)-algebra and let X be a non-commutative space of prime ideals. Consider the map
[TABLE]
(Mor denotes the set of morphisms of non-commutative spaces of prime ideals, and O(X):=Γ(X,OX)) which sends a morphism ψ on the left hand side to the composition A→O(Spec(A))→O(X) where the second arrow is induced by ψ. Then this map is injective and the image consists of all homomorphisms satisfying the following condition (i).
(i) For every x∈X, the composition A→O(X)→OX,x is a morphism in (Alg).
Proof.
Let H be the subset of Hom(A,O(X)) consisting of all elements which satisfy (i). Let Y=Spec(A), Z=Spec(Z(A)).
Let ψ:X→Y be a morphism and let h:A→O(X) be the associated ring homomorphism. We prove h∈H.
Let x∈X, and let y∈Y and z∈Z be the images of x.
Then the map A→OX,x is the composition A→OZ,z⊗Z(A)A=OY,y→OX,x. The first arrow of this composition is a morphism in (Alg). Since the second arrow is a morphism of (Alg), the composition is also a morphism in (Alg).
Let h∈H. We define a morphism ψ:X→Y of non-commutative spaces of prime ideals associated to h as follows. For x∈X, let ψ(x)∈Y be the inverse image of p(x) under A→OX,x. Then ψ is continuous by the condition (ii) in 2.3. For x∈X, since Z(A)→OX,x is a morphism in (Alg), the image of Z(A) in OX,x is contained in the center Z(OX,x) of OX,x by 1.9.
Let f∈Z(A) and let DX(f)={x∈X∣fx∈/p(x)inOX,x}.
Then for x∈DX(f), fx∈Z(OX,x) is not contained in the intersection of p(x) and Z(OX,x) which is the maximal ideal of the local ring Z(OX,x). Hence fx is invertible in Z(OX,x). By this, we have a homomorphism from the pullback of OZ on X to Z(OX). Hence we have a homomorphism from the inverse image of OZ⊗Z(A)A on X to OX. Thus we have a morphism of ringed spaces X→Y, which we still denote by ψ, and this is a morphism of non-commutative spaces of prime ideals.
It is easy to see that H→Mor(X,Y);h↦ψ and Mor(X,Y)→H;ψ↦h are the inverse maps of each other.
∎
2.9**.**
By a non-commutative scheme, we mean a non-commutative space of prime ideals X satisfying the following condition: For each x∈X, there is an open neighborhood U of x such that as a non-commutative space of prime ideals, U is isomorphic to an open subspace of Spec(A) (2.6) for some ring A which satisfies (f).
A morphism of non-commutative schemes is a morphism as non-commutative spaces of prime ideals.
2.10**.**
We have the contra-variant functor A↦Spec(A) from the full subcategory of (Alg) consisting of rings satisfying (f) to the category of non-commutative schemes.
Example.
Let k be a commutative field of characteristic 3 and let A be the group ring k[S3] of the symmetric group S3 of degree 3. We show:
(1) The canonical map A→O(X) is not bijective.
(2) The canonical map Hom(Alg)(A,A)→Mor(Spec(A),Spec(A)) is not bijective.
The group S3 is defined by generators α,β with relations α2=1, β3=1, αβα−1=β−1.
Let I be the two-sided ideal of A generated by β−1. Then I3=0 and A/I≅k×k where α∈A goes to (1,−1)∈k×k. The set Spec(A)
consists of two elements p1 and p2 where pi is the composition of A→A/I≅k×k and the i-th projection k×k→k. The center Z(A) is an Artinian local ring and we have Z(A)≅k[u,v]/(u2,v2,uv), where u corresponds to 1+β+β2 and v corresponds to α(1+β+β2).
As a non-commutative scheme, Spec(A) is the disjoint union of its two open subspaces U1={p1} and U2={p2} though A is not a direct product of two non-zero rings. The maps A→O(Ui) (i=1,2) are isomorphisms.
Hence O(Spec(A))=A×A, and A→O(Spec(A)) is the diagonal map A→A×A which is not an isomorphism. This shows (1).
By 2.8, we have Mor(Spec(A),Spec(A))=Hom(Alg)(A,Γ(Spec(A),OA))=Hom(Alg)(A,A×A)=Hom(Alg)(A,A)×Hom(Alg)(A,A). The diagonal map Hom(Alg)(A,A)→Hom(Alg)(A,A)×Hom(Alg)(A,A) is not bijective, for the set Hom(Alg)(A,A) has an element which is not the identity map (for example, the ring homomorphism x↦αxα−1). This shows (2).
Remark 2.12**.**
The author does not think that (1), (2) in 2.11 are weak points of our formulation of non-commutative geometry.
The fact Spec(A) is the disjoint union of U1 and U2 in 2.11 is important for the relation with zeta functions of non-commutative rings.
Étale cohomology theory was created by Grothendieck in his efforts to have a cohomology theory which explains properties of zeta functions of varieties over a finite field predicted in Weil conjectures. Our dream is to have an étale cohomology theory of non-commutative rings which can explain
zeta functions of non-commutative rings ([5]) over finite fields, and we hope that this paper is a starting point. In Appendix of this paper, T. Fukaya gives a partial result (9.23).
In 2.11, let k be the algebraic closure of F3 and let A0=F3[S3]. Then the zeta function ζA0(s) of A0 is (1−3−s)−2 where each (1−3−s)−1 corresponds to each point of Spec(A0). Because Spec(A) is the disjoint union of U1 and U2, the étale cohomology group
H0(Aet,Qℓ) is two-dimensional, and we have the good relation ζA0(s)=det(1−φ3−1u∣H0(Aet,Qℓ))−1 of the zeta function of A0 and the étale cohomology of Spec(A), where u=3−s and φ3∈Gal(kˉ/k) is the Frobenius operator which acts here as the identity map on the cohomology group.
If X is a ringed space such that O(X)=A, X must be connected because O(X) is not a product of two non-zero rings, and hence H0(X) should be
one-dimensional for any reasonable cohomology theory, and hence we would not have a presentation of ζA0(s) by using the cohomology of X. Such X is not nice for us in the relation to the zeta function.
3 Basics of non-commutative schemes, 1
We consider the non-commutative analogue 3.3 of the fact that a quasi-coherent sheaf B of commutative rings on a scheme S defines a scheme Spec(B) over S.
We also consider the “tensor products” and the fiber products in the category of non-commutative schemes.
In this section 3, X denotes a non-commutative scheme.
3.1**.**
For a ring A, by an A-algebra of finite presentation, we mean an A-algebra which is isomorphic over A to A⟨T1,…,Tn⟩/(f1,…,fm) for some n,m and f1,…,fm∈A⟨T1,…,Tn⟩.
By an OX-algebra locally of finite presentation (an l.f.p. OX-algebra for short), we mean a sheaf of rings over OX on X which is locally isomorphic to OX⟨T1,…,Tn⟩/(f1,…,fm) for some n,m and sections f1,…,fm of OX⟨T1,…,Tn⟩.
Lemma 3.2**.**
Let B be an l.f.p. OX-algebra.
Assume that we are given a ring A satisfying (f) and an isomorphism between X and an open subspace of Spec(A). Then locally on X, there are an A-algebra B of finite presentation and an isomorphism B≅OX⊗AB over OX.
Proof.
Let x∈X, and let z∈Spec(Z(A)) be the images of x. Take an open neighborhood U of x and an isomorphism B∣U≅OU⟨T1,…,Tn⟩/(f1,…,fm). Since OX,x≅OZ,z⊗Z(A)A, there is h∈Z(A) which is invertible at z such that the stalk fi,x comes from an element gi of A[1/h]⟨T1,…,Tn⟩ for 1≤i≤m. Replacing X by an open neighborhood of x, we may assume that fi coincides with the pullback of gi. Let B:=A[1/h]⟨T1,…,Tn⟩/(g1,…,gm) and U′=U∩D(h). Then B∣U′≅OU′⊗AB, and B is an A-algebra of finite presentation because A[1/h]=A[T]/(hT−1).
∎
Proposition 3.3**.**
Let B be an l.f.p. OX-algebra. Then there is a non-commutative scheme Spec(B) over X which represents the following contra-variant functor F from the category of non-commutative spaces of prime ideals over X to (Sets).
Let f:Y→X be a non-commutative scheme over X. Then F(Y) is the set of all homomorphisms f−1(B)→OY of sheaves of rings over f−1(OX) on Y such that for each y∈Y with image x in X, the induced map Bx→OY,y is a morphism in (Alg).
Proof.
Working locally on X, we may assume that we have A, B, and an isomorphism B≅OX⊗AB as in 3.2.
Then the inverse image of X in Spec(B), which is an open subspace of Spec(B), satisfies the condition of Spec(B) by 2.8 (which we apply by taking the present B and an open set U of Y as A and X there, respectively, to have the relation of Mor(U,Spec(B)) between Hom(Rings)(B,O(U))).
∎
3.4**.**
Let f:Y→X be a morphism of non-commutative schemes. We say f is locally of finite presentation (l.f.p. for short) if the following condition (i) is satisfied.
(i) Locally on X and of Y, there is an l.f.p. OX-algebra B such that Y is isomorphic over X to an open subspace of Spec(B).
Lemma 3.5**.**
(1) If f:Y→X and g:Z→Y are l.f.p., the composition f∘g:Z→X is also l.f.p.
(2) If Y→X and Z→X are l.f.p. and if f:Z→Y is a morphism over X, then f is l.f.p.
Proof.
(1) By the proof of 3.3, we may assume that there are a ring A satisfying (f), an A-algebra B of finite presentation, an isomorphism between X and an open subspace of Spec(A), and an isomorphism over X between Y and an open subspace of the inverse image of X in Spec(B). We may assume also that there are an l.f.p. OY-algebra C and an isomorphism over Y between Z and an open subspace of Spec(C). By 3.2 which we apply by taking (Y,B,C) as (X,A,B) of 3.2, we have that locally on Y and on Z, there is a B-algebra C of finite presentation such that C≅OY⊗BC over OY. Then Z is isomorphic over X to an open subspace of Spec(C).
(2) We may assume that X=Spec(A) for some ring A satisfying (f) and Y=Spec(B) for some A-algebra B of finite presentation. By 3.2, we may assume that there is a A-algebra C of finite presentation and that Z is an open subspace of Spec(C). Let z∈Z. Then there are an element h of the center of C such that z∈D(h), an open neighborhood U of z in D(h), and an A-homomorphism B→C[1/h] which induces the morphism U→Y. Since B and C are of finite presentation over A, C[1/h] is of finite presentation over B.
∎
Theorem 3.6**.**
Assume we are given morphisms f:Y→X and g:Z→X of non-commutative schemes. Assume at least one of f and g is l.f.p. Then there is a non-commutative scheme over X, which we denote by Y⊗XZ, which represents the following contra-variant functor F from the category of non-commutative spaces of prime ideals over X to (Sets).
For a non-commutative space of prime ideals W over X,
F(W) is the set of pairs (u,v) of morphisms u:W→Y and v:W→Z of non-commutative spaces of prime ideals over X satisfying the following condition:
For each w∈W with images y∈Y, z∈Z and x∈X, and for each a∈OY,y and b∈OZ,z which commute with the image of every element of OX,x, the images of a and b in OW,w commute.
If g is l.f.p., the canonical morphism Y⊗XZ→Y is l.f.p.
Proof.
We may assume that Z=Spec(B) for an l.f.p. OX-algebra B. Then Y⊗XZ is given as Spec(f∗B),
where f∗B=OY⊗f−1(OX)f−1(B).
∎
3.7**.**
Let A→B and A→C be morphisms in (Alg) and assume A, B and C satisfy (f). Assume at least one of A→B and A→C is of finite presentation. Then
Spec(B)⊗Spec(A)Spec(C)=Spec(B⊗AC).
3.8**.**
Let f:Y→X be a morphism of non-commutative schemes. We say f is r.c. (relatively commutative) if for every y∈Y and for x:=f(y), OY,y is an r.c. OX,x-algebra.
Clearly, if f and g:Z→Y are r.c., then f∘g is r.c.
Proposition 3.9**.**
Let S be a scheme, let B be a sheaf of Azumaya algebras over OS on S, and assume that X=Spec(B). Then the functor X⊗S gives an equivalence from the category of non-commutative schemes over S to the category of non-commutative schemes over X.
It gives an equivalence between the category of schemes over S and the category of r.c. non-commutative schemes over X.
Proof.
This follows from 1.15 and the fact that for an Azumaya algebra A over R, the sets of two-sided ideals are in bijection as is well known (it is deduced from the case A=Mn(R) by the étale descent for the étale topology of Spec(R)) and hence Spec(A)→Spec(R) is a homeomorphism.
∎
We consider fiber products in the category of non-commutative schemes.
Theorem 3.10**.**
Let f:Y→X and g:Z→X be morphisms of non-commutative schemes. Assume that at least one of f and g is l.f.p. and assume that at least one of f and g is r.c. Then Y⊗XZ is the fiber product of Y→X←Z in the category of non-commutative spaces of prime ideals. Write Y⊗XZ by Y×XZ in this case. If g is r.c., the canonical morphism Y×XZ→Y is r.c.
Proof.
This follows from 1.12 (3) and from the following 3.11.
∎
Lemma 3.11**.**
Let Y→X and Z→X be morphisms of non-commutative schemes, and assume at least one of these morphisms is l.f.p. Let W=Y⊗XZ, let w∈W, and let y, z, and x be the images of w in Y, Z, and X, respectively. Let S be the set of all elements of the center of OY,y⊗OX,xOZ,z whose images in OW,w are invertible. Then we have an isomorphism S−1(OY,y⊗OX,xOZ,z)→≅OW,w.
Proof.
We may assume that Z=Spec(B) for an l.f.p. OX-algebra B.
In the case Y=X, this follows from the construction of Spec(B) in the proof of 3.3. The general case follows from the case Y=X and the construction of Y⊗XZ in the proof of 3.6.
∎
Lemma 3.12**.**
Let Y→X be an l.f.p. morphism and assume that it is r.c. Assume that we are given a ring A satisfying (f) and an isomorphism between X and an open subspace of Spec(A). Then locally
on X and on Y, there exist an A-algebra B of the form A[T1,…,Tn]/(f1,…,fm) and an isomorphism over X between Y and an open subspace of the inverse image of X in Spec(B).
Proof.
By 3.2, we may assume that there are an A-algebra B of finite presentation and an isomorphism over X between Y and an open subspace of the inverse image of X in Spec(B). Take a finite presentation of B as a quotient of A⟨T1,…,Tr⟩ and let bi be the image of Ti in B. Let y∈Y and let s be the image of Y in S:=Spec(Z(B)). Since OY,y=OS,s⊗Z(B)B and this is relatively commutative over A, there are finitely many elements t1,…,tn of Z(B) and an element h of Z(B) which is invertible at s such that all bi are contained in the subring of B[1/h] generated by the image of A and t1,…,tn and 1/h. We have the surjective homomorphism A[T1,…,Tn,Tn+1]→B[1/h] over A which sends Ti to bi for 1≤i≤n and Tn+1 to 1/h.
Since B[1/h] is of finite presentation as an A-algebra, the kernel of this surjection is a finitely generated two-sided ideal. Furthermore, for U=D(h)⊂Y, U is isomorphic over X with an open subspace of the inverse image of X in Spec(B[1/h]).
∎
3.13**.**
For a morphism f:Y→X of non-commutative schemes which is l.f.p. and r.c., we have the diagonal morphism Δf:Y→Y×XY which corresponds the pair (Y→Y,Y→Y) of the identity morphisms.
We consider immersions (open immersions, l.f.p. closed immersions, and l.f.p. immersions) of non-commutative schemes,
Lemma 3.14**.**
For a morphism f:Y→X of non-commutative schemes, the following conditions (i) and (ii) are equivalent.
(i) Locally on X, there are a ring A satisfying (f), a finitely generated two-sided ideal I of A, and an isomorphism between X and an open subspace of Spec(A), such that Y is isomorphic over X to the inverse image of X in Spec(A/I) regarded as an open subspace of Spec(A/I).
(ii) Y is isomorphic over X to Spec(B) over X for some l.f.p. OX-algebra B such that the map OX→B is surjective.
Proof.
Assume (i). Then locally on X, with A and I as in (i), we have Y=Spec(B), where B=OX/IOX. When we regard OY naturally as a sheaf on X via the homeomorphism from Y to the image of Y in X, which is a closed subset of X, B is identified with the image of OX→OY and hence is independent of the choice of (A,I). Hence B is glued on the whole X and we have Y=Spec(B) globally on X.
Assume (ii). Locally, take a ring A satisfying (f) and an isomorphism between X and an open subspace of Spec(A). Then locally on X, for some element h of Z(A) which is invertible on X and for some f1,…,fm∈A[1/h], we have B=OX/(f1,…,fn)=OX⊗A[1/h]A[1/h]/(f1,…,fm). Hence we have (i).
∎
3.15**.**
(1) A morphism of non-commutative schemes Y→X is called an open immersion if Y is isomorphic over X to an open subspace of X.
(2) A morphism of non-commutative schemes Y→X is called a closed immersion locally of finite presentation (l.f.p. closed immersion for short) if it satisfies the equivalent conditions in 3.14.
If f:Y→X is an l.f.p. closed immersion, f gives a homeomorphism from Y to a closed subset of X.
(3) A morphism of non-commutative schemes Y→X is called an l.f.p. immersion if it is a composite Y→aU→bX such that a is an l.f.p. closed immersion and b is an open immersion.
Lemma 3.16**.**
(1) If f:Y→X and g:Z→Y are open (resp. l.f.p. closed, resp. l.f.p.) immersions, the composition f∘g:Z→X is an open (resp. l.f.p. closed, resp. l.f.p.) immersion.
(2) If f:Z→X is an open (resp. l.f.p. closed, resp. l.f.p.) immersion and if Y→X is a morphism of non-commutative schemes, the induced morphism Y×XZ→Y is an open (resp. l.f.p. closed, resp. l.f.p.) immersion. (Note that an l.f.p. immersion is r.c. and hence we have Y×XZ.)
Proof.
(1) for open immersion is clear.
(1) for l.f.p. closed immersion. We may assume that X is an open subspace of Spec(A) for a ring A satisfying (f) and Y=X×Spec(A)Spec(A/I) for a finitely generated two-sided ideal I of A. We have Z=Spec(C) for some OY-algebra C locally of finite presentation such that OY→C is surjective. Working locally on X, we may assume that there are an element h of Z(A/I) which is invertible on Y and elements f1,…,fm∈A such that C=OY/(h−1f1,…,h−1fm). Let J be the ideal of A generated by I and f1,…,fm. Then we have Z=X×Spec(A)Spec(A/J).
(1) for l.f.p. immersion. Since the composition of open (resp. l.f.p. closed) immersions is an open (resp. l.f.p. closed) immersion as we have just proved, it is sufficient to prove that if f:Y→X is an l.f.p. closed immersion and if g:Z→Y is an open immersion,
then f∘g:Z→X is an l.f.p. immersion. There is an open set U of X such that Z=Y∩U. Then f∘g is the composition Z→U→X, where the first arrow is an l.f.p. closed immersion (because it is the restriction of Y→X to U) and the second arrow is an open immersion.
(2) is clear.
∎
3.17**.**
Let f:Y→X be an r.c. and l.f.p. morphism of non-commutative schemes. Consider the diagonal morphism Δf:X→Y×XY (3.13) which is l.f.p. by 3.5 (2).
We say f is separated (resp. unramified) if Δf is an l.f.p. closed immersion (resp. open immersion).
Lemma 3.18**.**
(1) If Y→X and Z→Y are separated (resp. unramified) (then these morphisms are assumed to be r.c. and l.f.p.), then the composition Z→X is separated (resp. unramified).
(2) If f:Z→X is separated (resp. unramified), then for every morphism Y→X of non-commutative schemes, the induced morphism Y×XZ→Y is separated (resp. unramified).
(3) Let f:Y→X and g:Z→Y be morphisms of non-commutative schemes. Assume that f is separated and f∘g is an l.f.p. immersion. Then g is an l.f.p. immersion.
Proof.
(1) The diagonal morphism Z→Z×XZ is the composition of Z→Z×YZ and Z×YZ→Z×XZ, in which the second arrow is the base change of Y→Y×XY.
(2) is evident.
(3) The morphism g is the composition Z→Y×XZ→Y. The first arrow is a base change of the diagonal Y→Y×XY and hence is an l.f.p. closed immersion. The second morphism is a base change of Z→X and hence an l.f.p. immersion.
∎
Lemma 3.19**.**
Let f:Y→X be an unramified morphism. Let y∈Y, x=f(y), A=OX,x, B=OY,y. Let m be a maximal two-sided ideal of A (which may not be p(x)) and assume that A/m is an Azumaya algebra over a commutative field k. Then for some n≥1, there are commutative fields k1,…,kn which are finite separable extensions of k such that B/Bm≅A/m⊗k(k1×⋯×kn) over A/m. (Note that the left ideal Bm of B is a two-sided ideal of B because B is an A-algebra.)
Proof.
Let S be the set of all elements of the center of B⊗AB whose images in B under B⊗AB→B;b⊗c↦bc are invertible. Then since the diagonal morphism Y→Y×XY is an open immersion, S−1(B⊗AB)→≅B by 3.11 applied to the case Y and Z are Y, y and z are y, and w=y∈Y⊂Y×XY.
By 1.15, B/Bm=A⊗kE for some commutative ring E over k. Hence we have
S−1(B/Bm⊗A/mB/Bm)=B/Bm, that is,
S−1(E⊗kE)→≅E, where S is the set of all elements of E⊗kE whose images are invertible in E. Hence E≅k1×⋯×kn for some finite separable extensions k1,…,kn of k.
∎
3.20**.**
As in [2] Chap. 2, Section 4.1, we say a topological space T is irreducible if every finite intersection of non-empty open subsets of X is non-empty.
If X is a non-commutative scheme, there is a bijection from X to the set of all irreducible closed subsets of X which sends x∈X to the closure of {x} in X. In fact this is reduced to the case X=Spec(A) for a ring A, and this case is well-known.
3.21**.**
Recall that a ring A is called a prime ring if the ideal {0} of A is a prime ideal.
We say that a non-commutative scheme X is prime if it is irreducible and locally on X, X is isomorphic to an open subspace of Spec(A) for some prime ring A.
Lemma 3.22**.**
Let X be a prime non-commutative scheme.
(1) Let η be the point of X whose closure is X. Then the prime ideal p(η) of OX,η is (0). For every non-empty open set U of X, the map OX(U)→OX,η=OX,η/p(η) is injective.
(2) Let A be a ring satisfying (f), and let I be a two-sided ideal of A. Let f:X→Spec(A) be a morphism whose image is contained in Spec(A/I)⊂Spec(A). Then f factors through a morphism X→Spec(A/I) in a unique way.
Proof.
(1) We may assume X=Spec(A) for a prime ring A satisfying (f). Note that Z(A) is an integral domain. The map OX(U)→∏x∈UOX,x is injective. For x∈U, the map OX(U)→OX,η is the composition OX(U)→OX,x→OX,η and the last arrow is Z(A)p⊗Z(A)A→Z(A)(0)⊗Z(A)A where p is the image of x in Spec(Z(A)). Since all non-zero elements of Z(A) are non-zero-divisors of A, this last arrow is injective.
(2) By 2.8, it is sufficient to prove that I is killed by the induced homomorphism A→O(X). By (1), it is sufficient to prove that I is killed by the map A→OX,η=OX,η/p(η). Let y∈Spec(A/I) be the image of η. Then the last map factors as A→OSpec(A),y/p(y)→OX,η/p(η) and the first arrow here kills I.
∎
Proposition 3.23**.**
Let X be a non-commutative scheme and let
C be an irreducible closed subset of X. Then:
(1) C has a unique structure of a prime non-commutative scheme with which C represents the functor on the category of prime non-commutative schemes
[TABLE]
The natural morphism C→X is underlain by the inclusion map from C to X.
(2) Assume that X is covered by open subspaces Spec(Aλ) such that for every λ, every two-sided ideal
of Aλ is finitely generated. Then with the structure of C in (1), the canonical morphism C→X is an l.f.p. closed immersion.
Proof.
(1) We can work locally and hence we may assume that X is an open subspace of Spec(A) for a ring A satisfying (f) and that C=X∩Spec(A/p) for a prime ideal p of A. Then C is endowed with the structure of a non-commutative scheme as an open subspace of Spec(A/p). Then C becomes a prime non-commutative scheme. The rest follows from 3.22 (2).
(2) is clear.
∎
Remark 3.24**.**
If C is a closed subset of a scheme S, C has a canonical structure of a scheme, the reduced closed subscheme structure. The author does not know whether a close subset C of a non-commutative scheme X has such a canonical structure, except the case of an irreducible closed subset discussed above. For example, consider the following X and C.
We have Spec(Z3[S3])={m,m′,p,p′,q}, where m=(α−1,β−1,3), m′=(α+1,β−1,3), p=(α−1,β−1), p′=(α+1,β−1), q=(β2+β+1) with α, β as in 2.11, m and m′ are closed points, the closure of p is {m,p}, the closure of p′ is {m′,p′}, and the closure of q is {m,m′,q}. Let X be the open subspace of Spec(A)
consisting of m,p,q, and let C be X itself. Then C can be regarded as the non-commutative scheme X but C has also the structure as an open subspace of Spec(Z3[S3]/(p∩q))={m,m′,p,q}. These two structures are different because the stalk OC,m for the former is Z3[S3] but OC,m for the latter is Z3[S3]/(p∩q). Since A and A/(p∩q) are semi-prime rings (that is, the intersection of all prime ideals is (0)) and semi-prime rings are non-commutative analogues of commutative reduced rings, these two structures on C are both like reduced scheme structure, and which one should be regarded as the best canonical structure is not clear to the author.
4 Basics of non-commutative schemes, 2
We discuss flat morphisms of non-commutative schemes. We then consider specially non-commutative schemes which are “finite over the center”. We discuss the openness of a flat morphism (4.16).
4.1**.**
Let h:A→B be a homomorphism of rings.
We say h is right (resp. leftt) flat if through h, B is flat as a right (resp. left) A-module. (That is, the functor B⊗A is exact for left (resp. right) A-modules.)
We say f is flat if it is right flat and left flat.
4.2**.**
Let f:X→Y be a morphism of non-commutative schemes.
We say f is flat (resp. right flat, resp. left flat) if OY,f(x)→OX,x is flat (resp. right flat, resp. left flat) for every x∈X.
In this Section 4, we consider the right flatness. But the left flatness is treated in the same way by considering opposite rings.
Lemma 4.3**.**
(1) If X→Y and Y→Z are right flat morphisms of non-commutative schemes, the composition X→Z is right flat.
(2) Let X→Y be a right flat morphism. Then for a morphism Y′→Y of non-commutative schemes, if at least one of X→Y and Y′→Y is l.f.p., the induced morphism X⊗YY′→Y′ is right flat.
Proof.
(1) is clear.
(2) Let x′ be a point of X′:=X⊗YY′, let x be the image of x′ in X, let y′ be the image of x′ in Y′, and let y be the image of y′ in Y. Let P=OX,x⊗OY,yOY′,y′ and Q=OX′,x′. Then by 3.11, there is a multiplicative subset S in the center of P such that the canonical map P→Q induces an isomorphism S−1P→≅Q. Since P is right flat over OY′,y′, Q is right flat over OY′,y′. ∎
Let A be a prime ring and assume that A is a subring of a ring B. Then there is a prime ideal p of B such that A∩p=(0).
Proof.
Consider the set S of all two-sided ideals I of B such that A∩I=(0). By Zorn’s lemma, S has a maximal element p. We prove that p is a prime ideal. Assume I and J are two-sided ideals of B such that p⊂I and p⊂J and IJ⊂p. Since (A∩I)(A∩J)=(0) and (0) of A is a prime ideal, we have either A∩I=(0) or A∩J=(0). Assume A∩I=(0).
By the property of p as a maximal element, we have I=p.
∎
Proposition 4.5**.**
Let A→B be a morphism of (Alg) and assume it is faithfully right flat (that is, it is right flat and B⊗AM=0 for every non-zero left A-module M).
Then the morphism f:Spec(B)→Spec(A) is surjective. If A and B satisfy (f) and g:Y→Spec(A) is a morphism of non-commutative schemes and if either f or g is l.f.p., the map Spec(B)⊗Spec(A)Y→Y is surjective.
Proof.
By 3.7 (applied to C=OY,y/p(y) for y∈Y), it is sufficient to prove that Spec(B)→Spec(A) is surjective.
Let p be a prime ideal of A. Note that since B is an A-algebra, the left ideal Bp is a two-sided ideal BpB of B.
We show that A/p→B/Bp is injective. Let M be the kernel of this map. By applying B⊗A to the exact sequence 0→M→A/p→B/Bp, we have an exact sequence 0→B⊗AM→B/Bp→B⊗AB/p. The map B/Bp→B⊗AB/Bp is injective because it has a section x⊗y↦xy. Hence B⊗AM=0, and hence M=0. By 4.4 applied to the injection A/p→B/Bp, there is a prime ideal q of B whose inverse in A is p.
∎
In the rest of this Section 4, we consider an algebra A over a commutative ring R which is finitely generated as an R-module, and non-commutative schemes which are locally an open subspace of Spec(A) for such A.
For a ring A, let max(A) be the set of all maximal two-sided ideals of A.
Lemma 4.6**.**
Let R be a commutative ring and let A be an R-algebra such that A is a finitely generated R-module.
(1) Let q∈Spec(A) and let p∈Spec(R) be the image of q. Then q∈max(A) if and only if p∈max(R).
(2) Let I be a two-sided ideal of A and let I′⊂R be the inverse image of I under R→A. Then the image of V(I)⊂Spec(A) under Spec(A)→Spec(R) is V(I′).
(3) The map Spec(A)→Spec(R) is a closed map.
(4) If R→A is injective, the map Spec(A)→Spec(R) is surjective.
Proof.
(1) First assume q∈max(A). We prove that R/p is a field. Let a be a non-zero element of R/p. Since the center of A/q is a field, a is invertible in A/q. Since the inverse b of a in A/q is integral over R/p and satisfies ∑i=0ncibi=0 for some ci∈R/p with cn=1, we have b=−∑i=0n−1cian−1−i∈R/p.
Next assume p∈max(R). Then A/pA is a finite-dimensional algebra over a commutative field R/p. Hence every prime ideal of A/pA is a maximal two-sided ideal. Hence q∈max(A).
(2)
It is clear that the map Spec(A)→Spec(R) sends V(I) to V(I′). Let p∈V(I′). Since R/I′⊂A/I, A/I⊗RRp⊃R/I′⊗RRp=(R/I′)p=0. Take a maximal two-sided ideal m of this non-zero ring A/I⊗RRp. By (1), the inverse image of m in (R/I′)p is the maximal ideal of (R/I′)p. Let q be the prime ideal of A defined to be the inverse image of m. Then q∈V(I) and q↦p under Spec(A)→Spec(R).
(3) follows from (2).
(4) follows from the case I=0 of (2).
∎
4.7**.**
Let A be a ring such that there is a commutative local ring R and A is an R-algebra and A is finitely generated as an R-module.
(1) Assume R is excellent and henselian. Then there is an idempotent for (A,m).
(2) Let e be an idempotent for (A,m). Let I be a two-sided ideal of A. Then I⊂m if and only if e∈/I.
(3) Let e be an idempotent for (A,m). Then D(e)={p∈Spec(A)∣p⊂m}. In particular, D(e) is the smallest neighborhood of m in Spec(A).
Proof.
Let mR be the maximal ideal of R.
(1) We first consider the case R is a complete Noetherian local ring. In this case, A=limnA/mRnA.
Let J=Ker(A→∏mA/m), where m ranges over all maximal two-sided ideals of A. Then Jn⊂mRA for some n≥1 and hence A→≅limnA/Jn. We have an element e1∈A/J whose image in A/m is 1 and whose image in A/m′ is [math] for every m∈max(A)∖{m}. We can lift an idempotent en∈A/Jn to an idempotent en+1∈A/Jn+1 by the standard method.
(If e~n denotes a lifting of en to A and if e~n2=e~n+x with x∈Jn, then for e~n+1:=e~n+(1−2e~n)x, we have e~n+12≡e~n+1modJn+1.)
We next reduce the proof of (1) to the case R is complete. Let R^=limnR/mRn be the completion of R. Then an idempotent e^ for (A⊗RR^m) exists. There is a finitely generated subring R′ of R^ over R such that e^ comes from an idempotent e′ of A′:=A⊗RR′ satisfying e′−1∈mA′ and
e′∈m′A′ for all m′∈max(A)∖{m}. By Artin’s approximation theorem generalized by Popescu [15], there is an R-homomorphism R′→R. Let e be the image of e′∈A′ under A′=A⊗RR′→A. Then e is an idempotent for (A,m).
(2) If I⊂m, then e∈/I clearly. Assume I is not contained in m. We prove e∈I. Let B=A/I. Then the image eˉ of e in B is contained in all maximal two-sided ideals of B. Since the kernel of B/mRB→∏m′B/m′, where m′ ranges over all maximal two-sided ideals of B, is a nil-ideal, we have eˉn∈mRB for some n≥1. Hence eˉ=eˉn∈mRB. Let P=Beˉ⊂B. Then we have a homomorphism
B→P;b↦beˉ and the composition P→B→P is the identity map. The induced homomorphisms P/mRP→B/mRB→P/mRP have the properties that the first map is the zero map and the composition is the identity map. Hence P/mRP=0. By Nakayama’s lemma, we have P=0. Hence eˉ=0.
(3) Straightforwards from (2).
∎
4.9**.**
We will often consider the following situation.
Let R be an excellent strict local ring, let S=Spec(R), let A be an R-algebra which is finitely generated as an R-module, and let m∈max(A). Denote m also by x. We denote the open subspace {p∈Spec(A)∣m⊃p} of Spec(A) (4.8 (3)) by U(x). This is the smallest neighborhood of x in Spec(A).
4.10**.**
The condition (F) on a non-commutative scheme X.
(F): Locally on X, there are an excellent commutative ring R and an R-algebra A which is finitely generated as an R-module, and an open immersion X→Spec(A).
The condition (FT) on a non-commutative scheme over T, where T is an excellent scheme.
(FT) Locally on X, there are a commutative ring R over T which is of finite type over T, an R-algebra A which is finitely generated as an R-module, and an open immersion X→Spec(A) over T.
Let X be a non-commutative scheme satisfying (F) and let x∈X. Then OX,x is a finitely generated module over the local excellent ring Z(OX,x), and κ(x):=OX,x/p(x) is a finite-dimensional simple algebra over the residue field of Z(OX,x).
Proposition 4.12**.**
Let f:X→Y be a morphism of non-commutative schemes. Assume X and Y satisfy (F).
Assume f is surjective. Let g:Y′→Y be a morphism of non-commutative schemes and assume that either f or g is l.f.p. Then the morphism X⊗YY′→Y′ is surjective.
Proof.
Let y′∈Y′. We prove that there is an element of X⊗YY′ whose image in Y′ is y′.
Let y the image of y′ in Y and let x be an element X whose image in Y is y. Let A=OY,y/p(y), B=OX,x/p(x), C=OY′,y′/p(y′). Replacing Y with Spec(A), X with Spec(B), and Y′ with Spec(C), it is sufficient to prove that Spec(B⊗AC)→Spec(C) is surjective. By 4.5, it is sufficient to prove that C→B⊗AC is faithfully right flat. Hence it is sufficient to prove that A→B is faithfully right flat. We have A≅Mn(D) for some n≥1 and for some finite-dimensional central division algebra D over a commutative field k. By 1.15, the A-algebra B is written as Mn(D⊗kB′) for some non-zero k-algebra B′ and hence is faithfully right flat over A.
∎
4.13**.**
Let X be a quasi-compact non-commutative scheme satisfying (F). Then X is an Noetherian space (that is, every open subset of X is quasi-compact). In fact, X is locally an open subspace of Spec(A) for some algebra A such that Z(A) is a Noetherian ring and A is finitely generated as a Z(A)-module. Since all two-sided ideals of A are finitely generated, Spec(A) is a Noetherian space.
Hence X is a finite union of irreducible closed subsets.
Proposition 4.14**.**
Let X→Y be an l.f.p. morphism of non-commutative schemes. Assume X and Y satisfy (F) and X and Y are quasi-compact. Then f(X) is a constructible subset of Y.
Proof.
By Noetherian induction, we may assume that X is a prime non-commutative scheme and that f(C) is constructible for every closed subset C=X of X. In fact, if f(X) is not constructible, since X is a Noetherian space, there is a minimal element in the set of all closed subsets C of X such that f(C) are not constructible. If this minimal element C is a finite union of closed subsets Ci, we have C=Ci for some i because otherwise, all f(Ci) would be constructible and hence f(C) would be constructible. Thus this minimal element C is irreducible. It is sufficient to consider the case X is this C.
We may assume that Y is the closure of the image of the generic point of X in Y, and we may assume that Y is a prime non-commutative scheme. If X is union of open subsets U and U′ such that all f(U) and f(U′) are constructible, then f(X)=f(U)∪f(U′) is constructible.
By this, we may assume that Y is an open subspace of Spec(A) for some prime ring A such that Z(A) is Noetherian and A is finitely generated as a Z(A)-module,
X is an open subspace of Spec(B) for some prime ring B such that Z(B) is Noetherian and B is finitely generated as a Z(B)-module,
B is an A-algebra of finite presentation, A→B is injective, and the morphism Spec(B)→Spec(A) is compatible with f:X→Y. Note that then Z(A) and Z(B) are integral domains. There are a non-zero element a of Z(A) and a non-zero element b of Z(B) such that b/a∈Z(B) and such that A[1/a] is an Azumaya algebra over Z(A)[1/a], V:=Y×Spec(Z(A))Spec(Z(A)[1/a]) coincides with Spec(A[1/a]), B[1/b] is an Azumaya algebra over Z(B)[1/b], and U:=X×Spec(Z(B))Spec(Z(B)[1/b]) coincides with Spec(B[1/b]). Then via the isomorphisms of topological spaces V→≅Spec(Z(A)[1/a]) and U→≅Spec(Z(B)[1/b]), the constructibility of f(U) in V is reduced to the constructibility of the image of the morphism Spec(Z(B)[1/b])→Spec(Z(A)[1/a]) of finite presentation in the theory of schemes. Since f(X∖U) is constructible by our Noetherian induction, we have that f(X)=f(U)∩f(X∖U) is constructible.
∎
Proposition 4.15**.**
Let X→Y be a right flat morphism of non-commutative schemes. Assume both X and Y satisfy (F). Then f(X) is stable in Y under generalizations.
Proof.
Let x∈X with image y in Y. Let y′∈Y be a generalization of y (that is, y belongs to the closure of y′ in Y). We prove that there is an element x′ of X with image y′ in Y.
We may assume that Y is an open subspace of Spec(A), where A is an R-algebra over a commutative ring R such that A is finitely generated as an R-module. Replacing A by A/p where p is the prime ideal of A given by y′ and replacing Y by Y×Spec(A)Spec(A/p), we may assume A is a prime ring and y′ corresponds to the prime ideal (0) of A. By replacing R by the image of R in A, we may assume R is an integral domain.
We may assume that X is an open subscheme of Spec(B) where B is an algebra over an excellent commutative ring R′
and is finitely generated as an R′-module. By replacing R′ with the strict henselization at the image of
x in Spec(R′), we may assume that X=U(x) (4.9). We prove that there is a point of U(x) lying over the prime ideal
(0) of
A, that is, lying over the prime ideal (0) of R. If there is no such element,
U(x)×Spec(R)Spec(Q(R)) is empty, where Q(R) denotes
the field of fractions of R. Since every element in the intersection of all prime ideals of a ring is nilpotent ([13] Theorem 10.7),
the image of e in Q(R)⊗RA is nilpotent and hence is [math] because it is an idempotent. Hence there is a non-zero element a of
R such that ae=0. Since A→A;x↦ax is injective and the direct summand eB of B is right flat over A, eB↦eB;x↦ax
is injective. Since ae=0, we have e=0, a contradiction.
∎
Proposition 4.16**.**
Let X→Y be an l.f.p. right flat morphism of non-commutative schemes. Assume both X and Y satisfy (F). Then f is an open map.
In this section, we define the étale cohomology groups of non-commutative schemes imitating SGA 4 ([1]).
We prove a finiteness theorem (5.21) for non-commutative schemes satisfying (F). We also prove a proper base change theorem (5.29) for non-commutative schemes satisfying (F) and for relatively commutative morphisms between them.
5.1**.**
Let f:X→Y be a morphism of non-commutative schemes.
We say f is étale if it satisfies the following two conditions.
(ii) f is unramified (3.17, this tells that f is r.c. and l.f.p).
For schemes, this étaleness coincides with the usual étaleness.
Lemma 5.2**.**
(1) If X→Y and Y→Z are étale morphisms of non-commutative schemes, the composition X→Z is étale.
(2) Let X→Y be an étale morphism. Then for every morphism Y′→Y of non-commutative schemes, the induced morphism X×YY′→Y′ is étale.
(3) Let X→Z and Y→Z be étale morphisms of non-commutative schemes. Then every morphism X→Y over Z is étale.
Proof.
(1) (resp. (2)) follows from 4.3 (1) (resp. (2)) for flat morphisms and from 3.18 (1) (resp. (2)) for unramified morphisms.
(3) We first prove that the morphism X→Y is flat. This morphism is the composition X→X×ZY→Y. Here the first arrow is flat because it is the base change of the open immersion Y→Y×ZY by X×ZY→Y×ZY. The second arrow is flat because it is the base change of the flat morphism X→Z by Y→Z. The morphism X→X×YX is an open immersion because the composition X→X×YX→X×ZX is an open immersion and the morphism X×YX→X×ZX is the base change of the open immersion Y→Y×ZY by X×ZX→Y×ZY and hence is an open immersion.
∎
5.3**.**
For a non-commutative scheme X,
we define the étale site Xet as follows.
An object of Xet is a non-commutative scheme over X which is étale over X.
A morphism is a morphism of non-commutative schemes over X. (Note that a morphism in this site is étale by 5.2 (3).)
A morphism f:Y→Z in Xet is a covering if it is universally surjective (that is, for every morphism Z′→Z of non-commutative schemes, the map Y′:=Y×ZZ′→Z′ induced by f is surjective).
Remark 5.4**.**
By 4.12, if X satisfies (F) and f:Y→Z is a morphism in Xet, f is a covering if it is surjective. The author does not know any example of an r.c. l.f.p. morphism of non-commutative schemes which is surjective but not universally surjective.
5.5**.**
Functoriality. If X→Y is a morphism of non-commutative schemes, U↦X×YU sends an object of Yet to an object of Xet and a covering to a covering.
Hence we have a morphism of the associated topoi X~et→Y~et.
Proposition 5.6**.**
Let S be a scheme, let B be an Azumaya algebra over OS on S, and let X=Spec(B). Then the pullback functor X×S gives an equivalence of sites between Set and Xet.
For a non-commutative scheme X, the presheaves U↦O(U) and U↦Γ(U,Z(OU)) on the étale site Xet need not be a sheaf. Note that Γ(U,Z(OU)) is understood as the set of morphisms of non-commutative schemes U→Spec(Z[T]) (1.11, 2.8). Thus a representable functor is not necessarily a sheaf for the étale topology.
Let OXet be the sheafification of the presheaf U↦O(U) on Xet. See 6.20 for an example of X such that O(X)=Γ(X,OX) does not coincide with Γ(Xet,OXet) and Γ(X,Z(OX)) does not coincide with Γ(Xet,Z(OXet)).
See also 7.5 (1).
5.8**.**
Let X be a quasi-compact non-commutative scheme which satisfies (F).
Then X is a Noetherian space
(4.13).
By this fact and by the fact that every morphism between objects of Xet is an open map by 4.16, and by the fact every surjective morphism between objects of Xet is a covering (5.4), we have that the category of sheaves on Xet is a Noetherian topos in the sense of [1] vol. 2, Exp. VI, 2.11. Hence by loc.cit Theorem 5.1 and Theorem. 8.7.3, we have the following 5.9 and its generalization 5.11.
Proposition 5.9**.**
Let X be a quasi-compact non-commutative scheme satisfying (F), and let (Fλ)λ be a directed family of sheaves of abelian groups on Xet. Then for every m, we have an isomorphism
[TABLE]
5.10**.**
This is a preparation for 5.11. Let Λ be a directed ordered set and assume that we are given an inverse system (Xλ)λ∈Λ of non-commutative schemes. Let I be a finite set. Assume that for each λ∈Λ, we have an open covering (Uλ,i)i∈I of Xλ. Assume that for each i∈I, we have an inductive system (Aλ,i)λ∈Λ in the category (Alg). Assume that we have an open immersion Uλ,i→⊂Spec(Aλ,i) for each λ∈Λ and i∈I. We assume the following (i)–(iii).
(i) For λ,μ∈Λ such that λ≤μ and for each i∈I, the inverse image of Uλ,i⊂Xλ in Xμ is Uμ,i, the inverse image of Uλ,i⊂Spec(Aλ,i) in Spec(Aλ,i) in Spec(Aμ,i) is Uμ,i, and the morphism Uμ,i→Uλ,i induced by Spec(Aμ,i)→Spec(Aλ,i) coincides with the morphism induced by the transition morphism pμ,λ:Xμ→Xλ.
(ii) For each λ∈Λ, the center Z(Aλ,i) is an excellent ring and Aλ,i is finitely generated as a Z(Aλ,i)-module.
(iii) For each i∈I, if we write Ai=limλAλ,i, then Z(A) is an excellent ring and Ai is finitely generated as a Z(Ai)-module.
Let X=limλXλ as a topological space, endow X with the sheaf of rings limλpλ−1(OXλ) where pλ:X→Xλ is the projection, and define p(x):=limλp(xλ) where xλ is the image of x in Xλ. Then X=(X,OX,(p(x))x∈X) is a non-commutative scheme.
Assume that for each λ∈Λ, we are given a sheaf Fλ of abelian groups on Xλ,et and assume that for each λ,μ∈Λ such that λ≤μ, we have a homomorphism hλ,μ:pμ,λ−1(Fλ)→Fμ. We assume that hλ,λ is the identity homomorphism for λ∈Λ and that we have hλ,ν=hμ,ν∘pν,μ−1(hλ,μ) when λ≤μ≤ν.
Let F:=limλpλ−1(Fλ).
Lemma 5.11**.**
For each m, we have an isomorphism
[TABLE]
Proposition 5.12**.**
Let R, S, A, x, U(x) be as in 4.9. Let U→U(x) be an étale morphism such that there is u∈U whose image in U(x) is x. Then there is an open immersion U(x)→U over U(x).
Proof.
(In this proof here, we use the fact that f is right flat and also left flat.)
The integral closure R1 of the image of R in A is a finitely generated R-module and hence it is a finite product of local rings because R is henselian. Let R2 be the component of R1 in this direct product decomposition such that the image of x in Spec(R1) belongs to Spec(R2). By replacing R with R2 and A with A⊗R1R2, we may assume that R is the center of A and is a local ring. Then we have A=OU(x),x.
Since U→U(x) is r.c. and l.f.p.,
we may assume that
there are a finitely generated commutative ring R′ over R, an A⊗RR′-algebra B such that A⊗RR′→B is surjective, and an open immersion U→Spec(B) over A. Let V be the set of points of Spec(R′) at which the morphism Spec(R′)→Spec(R) is quasi-finite. Then V is open in Spec(R′) by [7] Cor. 13.1.4. Furthermore, the image t of u in Spec(R′) is contained in V by 3.19. By replacing Spec(R′) by an affine open neighborhood of t in V, we may assume that Spec(R′) is quasi-finite over S. Then since R is henselian, there is an open and closed neighborhood of t in Spec(R′) which is finite over S. By replacing Spec(R′) by this open an closed subset, we may assume that R′ is finitely generated as an R-module. Hence B is finitely generated as a left A-module. Let R1′ be the integral closure of the image of R′ in B. Since R′ is henselian, R1′ is a finite product of local rings. Let R2′ be the component of R1′ in this direct product decomposition such that the image of u in Spec(R1′) is contained in Spec(R2′). By replacing R′ with R2′ and replacing B with B⊗R1′R2′, we may assume that R′ is the center of B and is a local ring. We have B=OU,u.
By 3.19 and by the subjectivity of A⊗RR′→B, the maps A/m→B/Bm=B/mB are bijective for all m∈max(A). Hence the map A→B is surjective by Nakayama’s lemma. For the proof of 5.12, it is now sufficient to prove A→≅B. In fact, then we will have U→≅U(x) because U(x) is the smallest neighborhood of x in Spec(A).
Let I be the kernel of the surjection A→B. We prove I=0. Since B is left flat, by [14] Chap. 2, §4D Proposition 4.30, B is projective as a left A-module and hence A≅I⊕B as a left A-module.
Claim. If M is a subquotient of the left A-module I and N is a subquotient of the left A-module B, and if M≅N as a left A-module, then M=0 and N=0.
Proof of Claim. Since B is right flat over A, B⊗AI=0 and hence B⊗AM=0. On the other hand, B⊗AN=N. Hence N=0 and hence M=0.
Now by right multiplication, the opposite ring A∘ is isomorphic to the endomorphism ring of the left A-module A≅I⊕B. By Claim, this endomorphism ring is the direct product (as a ring) of the endomorphism ring of the left A-module I and the endomorphism ring of the left A-module B. Since the center of A is a local ring, the ring A can not be a product of two non-zero rings. Hence the endomorphism ring of the left A-module I is zero, that is, the right multiplication by 1 on I is the zero map. Hence I=0.
∎
5.13**.**
Let X be a non-commutative scheme satisfying (F). We consider stalks of a sheaf F on Xet.
Let x∈X.
Then κ(x):=OX,x/p(x) is a finite-dimensional central simple algebra over its center k which is a commutative field. Let κ(xˉ)=κ(x)⊗kkˉ, where kˉ is the separable closure of k, and let xˉ=Spec(κ(xˉ)). We have a canonical morphism ixˉ:xˉ→X of non-commutative schemes. The topos of sheaves on xˉet is equivalent to the category of sets via G↦G(xˉ). This is by the fact that κ(xˉ)≅Mn(kˉ) for some n≥1 and by 5.6.
For a sheave F on Xet, let Fxˉ=(ixˉ−1F)(xˉ). In the case k is separably closed, we denote Fxˉ also by Fx.
Lemma 5.14**.**
Assume X satisfies (F).
(1) For a morphism h:F→G of sheaves on Xet, h is an isomorphism if and only if the induced maps Fxˉ→Gxˉ are bijective for all x∈X.
(2) A sequence
F′→F→F′′ of sheaves of abelian groups on Xet is exact if and only if Fxˉ′→Fxˉ→Fxˉ′′ are exact for all x∈X.
Proof.
(1) Assume that Fxˉ→Gxˉ are bijective for all x∈X. We prove that h is injective (resp. surjective). Let U∈Xet. Let a1,a2∈F(U) and assume that there images in G(U) are the same (resp. let a∈G(U)). Let u∈U, and let x∈X be the image of u in X. Then the morphism xˉ→x factors as xˉ→u→x. Hence by the assumption, there are Uu∈Uet, and a morphism xˉ→Uu over U such that the image of xˉ in U is u and such that the images of a1,a2 in G(Uu) are the same (resp. there is b∈F(Uu) whose image in G(Uu) is the pullback of a). The morphism ∐u∈UUu→U is surjective and hence universally surjective by 4.12, and hence ∐u∈UUu→U is a covering in Xet. This proves that h is injective (resp. surjective).
(2) follows from (1).
∎
Proposition 5.15**.**
Let S, A, x, U(x) be as in 4.9 and let s be the closed point of S. Let F be a sheaf of abelian groups on U(x)et.
(1) We have RΓ(U(x)et,F)→≅Fx.
(2) Let ϵ:U(x)→S be the canonical morphism. Then (Rϵ∗F)s=(ϵ∗F)s=Fx.
Let i:Y→X be an l.f.p. closed immersion of non-commutative schemes. Assume X satisfies (F).
(1) The functor i∗ is exact for sheaves F of abelian groups on Yet. For x∈X, we have (i∗F)xˉ=Fxˉ if x∈Y, and (i∗F)xˉ=0 if x∈/Y.
(2) The functor i∗ gives an equivalence from the category of sheaves of abelian groups on Yet to the category of sheaves of abelian groups on Xet whose restrictions to U=X∖Y are zero. The converse functor is given by i−1. In particular, the former category depends only on the closed set Y of X, not on the non-commutative scheme structure of Y.
(2) If F is a sheaf of abelian groups on Yet, i−1i∗F→F is an isomorphism as is seen by checking the stalks. If G is a sheaf of abelian groups on Xet such that G∣U=0, G→i∗i−1G is an isomorphism as is seen by checking the stalks.
∎
Lemma 5.17**.**
Let j:U→X be an open immersion of non-commutative schemes. Assume X satisfies (F).
(1) The inverse image functor j−1 from the category of sheaves of abelian groups on Xet to the category of abelian groups on Uet has a left adjoint functor j!.
(2) j! is an exact functor for sheaves F of abelian groups on Uet. For x∈X, we have (j!F)xˉ=Fxˉ if x∈U, and (j!F)xˉ=0 if x∈/U.
(3) If i:Y→X is an l.f.p. closed immersion and if j:U→X is the complement,
we have an exact sequence 0→j!j−1F→F→i∗i−1F→0.
Proof.
For a sheaf of abelian groups F on Uet, define j!F to be the sheaf associated to the following presheaf G on Xet. For V∈Xet, G(V) is F(V) if the image of V in X is contained in U (then V→X factors uniquely as V→U→X and V is regarded as an object of Uet and hence F(V) is defined) and G(V)=0 otherwise. (1) and (2) follow from this definition. (3) is shown by checking the stalks by 5.14 (2).
∎
5.18**.**
Let f:Y→X be an l.f.p. immersion of non-commutative schemes. Assume X satisfies (F). We define f!.
For a sheaf F of abelian groups on Yet, by using a factorization
Y→iU→jX where i is an l.f.p. closed immersion and j is an open immersion, we define f!F:=j!i∗F. This is independent of the choice of the factorization. In fact, if Y→U′→Y is another factorization, we have the third factorization Y→kU∩U′→lX, and we have a canonical morphism l!k∗(F)→j!i∗(F).
We see that this is an isomorphism by comparing the stalks.
5.19**.**
Let X be a non-commutative scheme satisfying (F).
We say sheaf F on Xet is constructible if locally on X, there is a finite family of l.f.p. immersions Yi→X such that the set X is the disjoint union of these Yi and such that the pullback of F to each Yi is locally constant and finite.
By 5.17 (3), we see that a sheaf F of abelian groups on Xet is constructible if and only if locally on X, there is a finite filtration on F whose each graded quotient is isomorphic to f!G for some l.f.p. immersion f:Y→X and for some locally constant finite sheaf G of abelian groups on Yet.
(1) For a homomorphism h:F→G between constructible (resp. locally constant and finite) sheaves of abelian groups on Xet, the kernel and the cokernel of h are smooth (resp. constructible).
(2) For an exact sequence 0→F→G→H→0 of sheaves of abelian groups on Xet such that F and H are constructible (resp. locally constant and finite), G is also constructible (resp. locally constant and finite)
Proof.
These are proved as in the case of schemes treated in Lemma 2.1 and Proposition 2.6 in [1] vol. 3, Exp. IX. ∎
We consider a non-commutative version of the finiteness theorem of Gabber ([10] Exp. XIII, Exp. XXI; it is a generalization the finiteness theorem of Deligne in the chapter on finiteness in [4]).
Theorem 5.21**.**
Let X and Y be quasi-compact non-commutative schemes satisfying (F), and let f:X→Y be an l.f.p. morphism.
Let F be a constructible sheaf of abelian groups on Xet such that the orders of all stalks of F are invertible on Y. Then the sheaf Rmf∗F on Yet is constructible for every m and is zero for m≫0.
(2) Let F be a constructible sheaf of sets on Xet. Then the sheaf f∗F on Yet is constructible.
(3) Let F be a constructible sheaf of groups on Xet such that the orders of all stalks of F are invertible on Y. Then the sheaf R1f∗F on Yet is constructible.
To prove 5.21, we first consider the following situation.
Let A be a ring and let B be an A-algebra of finite presentation. Assume that A and B are prime rings, that they are finitely generated as modules over their centers, and that the map A→B is injective. Note that Z(A) and Z(B) are integral domains. We assume that Z(A) and Z(B) are excellent rings and Z(A) is regular.
Let Y be an open subspace of Spec(B), let X be an open subspace of Spec(B), and assume that X is contained in the inverse image of Y in Spec(B).
Let a be a non-zero element of Z(A), let b be a non-zero element of Z(B) such that b/a∈Z(B),
A[1/a] is an Azumaya algebra over Z(A)[1/a], B[1/b] is an Azumaya algebra over Z(B)[1/b], Y×Spec(Z(A))Spec(Z(A)[1/a])=Spec(A[1/a]), and X×Spec(Z(B)Spec(Z(B)[1/b])=Spec(B[1/b]).
Let f:X→Y, π:Y→Spec(Z(A)), j:U:=X×Spec(Z(B))Spec(Z(B)[1/b])=Spec(B[1/b])→X, g=π∘f∘j:U→Y, h:V:=Spec(Z(B)[1/b])→Spec(Z(A))
be the canonical morphisms. Note that j is an open immersion.
By 5.6, the topos of sheaves on Uet and the topos of sheaves on Vet are equivalent. We will identify a sheaf on Uet with the corresponding sheaf on Vet.
We use the following elementary lemma.
Lemma 5.23**.**
Let A be an Azumaya algebra over an integral domain R, and let e be a non-zero idempotent of A. Then D(e)={p∈Spec(A)∣e∈/p} coincides with the whole Spec(A).
Proof.
Let Q(R) be the field of fractions of R. Then Ae is a direct summand of the R-module A, and Q(R)⊗RAe=0. Hence Ae is a finitely generated projective R-module rank ≥1. Let p be a prime ideal of A. Then p=Aq for a prime ideal q of R. We have that 0=R/q⊗RAe⊂R/q⊗RA. This shows e∈/p.
∎
Lemma 5.24**.**
Let the situation be as in 5.22.
Then for a sheaf F of abelian groups (resp. sets, resp. groups) on Uet, the canonical morphism π−1Rmh∗F→Rmg∗F is an isomorphism for every m (resp. for m=0, resp. for m=1).
Proof.
Let y∈Y and let s be the image of y in S:=Spec(Z(A)). It is sufficient to prove that the map (Rmh∗F)sˉ→(Rmg∗F)yˉ is an isomorphism. Let Ssˉ be the strict henselization of S at sˉ. We consider the smallest neighborhood U(yˉ) of yˉ in Y×SSsˉ (4.9). Since R is regular, OS,sˉ is an integral domain. Hence by 5.23 applied to the Azumaya algebra
A⊗Z(A)OS,sˉ[1/a] over OS,sˉ[1/a], we have U(yˉ)×SSpec(Z(A)[1/a])=Y×SSpec(OS,sˉ[1/a]). From this, we have
U×SSsˉ=U×YU(yˉ). Hence we have
[TABLE]
where the first = follows from 5.15 (1).
On the other hand,
[TABLE]
By 5.6, we have
Hm((U×SSsˉ)et,F)≅Hm((V×SSsˉ)et,F).
∎
Lemma 5.25**.**
Let the situation be as in 5.22.
Then for a constructible sheaf F of abelian groups on Uet, the sheaf Rmg∗F is constructible for every m and is zero for m≫0. For a constructible sheaf F of sets (resp. groups) on Uet, f∗F (resp. R1f∗F) is constructible.
Proof.
By the finite theorem of Gabber ([10] Exp. XIII and Exp XXI), for a constructible sheaf F of abelian groups F on Vet, Rmh∗F is constructible for every m and is zero for m≫0, and for a constructible sheaf F of sets (resp. groups) on Vet, h∗F (resp. R1h∗F) is constructible. This lemma follows from it by 5.24.
∎
For a sheaf F on Xet and for a closed subset C of X, there is a unique quotient sheaf F of F such that Fxˉ→≅(FC)xˉ if x∈C and (FC)xˉ is a one point set if x∈X∖C. The uniqueness is clear. The existence of FC is seen as follows. In the case C is irreducible, C has the canonical structure of a prime non-commutative scheme and we have the l.f.p. closed immersion i:C→X. We have FC:=i∗i−1F. In general,
let Ck (1≤k≤r) be all the irreducible components of C. Then we obtain FC as the image of F→∏k=1rFCk. If F is a sheaf of groups (resp. abelian groups), FC has the unique structure of a sheaf of groups (resp. abelian groups) which is compatible with that of F.
We say that F has supports in C if F=FC, that is, if the restriction of F to X∖C is the constant sheaf associated to a one point set.
We prove (1) of 5.21.
For closed subsets C and C′ of X, we have an exact sequence
(1) 0→FC∪C′→FC⊕FC′→FC∩C′→0.
Thus if 5.21 (1) is true for F with supports in C and for F with supports in C′, then it is true for F with supports in C∪C′.
Note that X is a Noetherian space. Hence by Noetherian induction, we may assume that X is irreducible and 5.21 (1) is true if F has supports in some closed subset C=X of X. Hence we assume that X is a prime non-commutative scheme. Let η be the generic point of X and let Y′ be the closure of f(η) in Y with the prime non-commutative scheme structure. We can replace Y by Y′ and hence we assume that Y is also a prime non-commutative scheme.
For open immersions U→X and U′→X such that X is the union of U and U′, we have an exact sequence (let U′′=U∩U′)
where fU denoted the composition U→X→fY and fU′, fU′′ are defined similarly, FU denotes the restriction of F to U, and FU′, FU′′ are defined similarly. By this, we are reduced to the situation of 5.22. (The regularity of Z(A) in 5.22 is attained since the set of regular points in an excellent scheme is open.)
Let C=X∖U and let j:U→X be the inclusion morphism.
We have a distinguished triangle
(3) F→Rj∗(FU)⊕FC→(Rj∗(FU))C→.
Note that Rj∗(FU) is constructible by the case X=Y of 5.25, and hence (Rj∗(FU))C is constructible.
By Noetherian induction, Rf∗(FC) and Rf∗((Rj∗(FU))C) are constructible. On the other hand, Rf∗Rj∗(FU)=Rg∗(FU) is constructible by 5.25. Hence Rf∗F is constructible.
The proof of 5.21 (2) is similar to the above proof of 5.21 (1). We just replace the above exact sequences (1) and (2) and the distinguished triangle (3), by the following facts (4), (5), and (6), respectively.
(4) The sheaf FC∪C′ is the fiber product of FC→FC∩C′←FC′.
(5) The sheaf F is the fiber product of fU,∗(FU)→fU′′,∗(FU′′)←fU′,∗(FU′).
(6) The sheaf F is the fiber product of FC→(j∗(FU))C←j∗(FU).
We prove (3) of 5.21.
Because we have proved (2) of 5.21, we have
Claim. Let F and F′ be constructible sheaves of groups on Xet and let F→F′ be an injective homomorphism. If R1f∗(F′) is constructible, then R1f∗F is constructible.
In the case of schemes, this is Lemma 3.1.1 of Exp. XXI of [10]. The same proof works.
For a quasi-compact non-commutative scheme X satisfying (F), let SX (resp. SX,C for a closed subset C of X) be the statement that 5.21 (3) is true for every Y and every f:X→Y and every F (resp. every F with supports in C) as in the hypothesis of 5.21.
For closed subsets Ci (1≤i≤n) of X, if SX,Ci is true for 1≤i≤n, then SX,C for C=∪i=1nCi is true. This is seen by the above Claim applied to the injective homomorphism FC→∏i=1nFCi.
By this, for the proof of SX, we may assume that X is a prime non-commutative scheme and that SX,C is true for every closed subset C=X of X. For open subsets Ui (1≤i≤n) of X such that X=∪i=1nUi, if SUi are true for 1≤i≤n, then SX is true. In fact, for each i, if we denote by ji the inclusion morphism Ui→X, R1(f∘ji)∗(FUi) and R1ji,∗(FUi) are constructible. Hence f∗R1ji,∗(FUi) is constructible by (2) of 5.21 and hence
the kernel R1f∗ji,∗(FUi) of R1(f∘ji)∗(FUi)→f∗R1ji,∗(FUi) is constructible. By Claim applied to the injective homomorphism F→∏i=1nji,∗(FUi), R1f∗F is constructible. By this, we are reduced to the situation of 5.22.
Now we consider the injective homomorphism F→j∗(FU)×FC. The sheaf R1f∗j∗(FU) is constructible because it is the kernel of
R1g∗(FU)→f∗R1j∗(FU) and R1g∗(FU) (resp. f∗R1j∗(FU)) is constructible by 5.24 (resp. by the case Y=X of 5.25 and by (2) of 5.21). The sheaf R1f∗(FC) is constructible by the Noetherian induction. Hence R1f∗F is constructible by Claim. This completes the proof of 5.21.
The following 5.27 is a non-commutative version of the proper base change theorem in the usual étale cohomology theory. This will be further generalized in 5.29.
Theorem 5.27**.**
Let Y be a non-commutative scheme satisfying (F).
Let T→S be a proper morphism of schemes, let Y→S be a morphism of non-commutative schemes, and let X=Y×ST. Then we have the following proper base change theorem for f:X→Y. Let Y′ be a non-commutative scheme satisfying (F) and let h:Y′→Y be a morphism, and consider the following commutative diagram with X′=X×YY′ (note that f is relatively commutative by our assumption and hence the fiber product is defined).
[TABLE]
Then for any sheaf of torsion abelian groups F on Xet, the canonical morphism h−1Rf∗F→Rf∗′g−1F is an isomorphism.
Proof.
We may assume that S=Spec(R) for an excellent commutative ring R and X=U(x) for some R-algebra A
which is finitely generated as an R-module and for some closed point x of Spec(A).
It is sufficient to prove that (h−1Rf∗F)yˉ′→(Rf∗′g−1F)yˉ′ is an isomorphism
for every y′∈Y′. Let y be the image of y′ in Y. Let X(y)=X×Yy, X′(yˉ′)=X′×Y′yˉ′ and consider the cartesian squares
[TABLE]
It is sufficient to prove 5.27 in the cases of these squares (that is, in the cases the cartesian square in 5.27 is replaced by these cartesian squares).
In the second square, by localizing Y′, we may assume that y′=yˉ′.
It is sufficient to consider the first and the third squares. Note that X(y)=T(s)×sy, X′(y′)=T(s)×sy′, where T(s)=T×Ss.
Now we consider the first square. Let the notation be ϵX:X→T, ϵY:Y→S, ϵX(y):X(y)→T(s), ϵy:y→s, fy:X(y)→y, p:T→S, ps:T(s)→s.
Now 5.27 for the first square is proved by
[TABLE]
[TABLE]
Here, the second and the fifth = are the evident ones. The first and the last (sixth) = are by 5.15 (2). In the third =, we use the usual proper base change theorem of schemes for the proper morphism p:T→S. The fourth = is by
Claim.(ϵ∗XF)∣T(s)=ϵ∗X(y)(F∣X(y)).
We prove Claim.
Let t∈T(s) and let x the unique point of X(y) lying over t. Then if we denote the the strict localization of T at t by Ttˉ, the morphism X×TTtˉ→Ttˉ satisfies the present condition of X→S and X×TTtˉ=U(x). Hence we have
(ϵ∗XF)∣T(s))tˉ=Fxˉ and
(ϵ∗X(y)(F∣X(y)))tˉ=(F∣X(y))xˉ=Fxˉ by 5.15 (2).
This proves Claim and hence 5.27 for the first square.
Now we consider the third square. Note that s=Spec(k), y=Spec(Mn(k)), y′=Spec(Mmn(k′)) for some separably closed commutative fields k and k′ and for some m,n≥1. Since X(y)=T(s)×sy, a sheaf of torsion abelian groups on X(y)et comes form a sheaf of torsion abelian groups on T(s)et by 5.6.
Let s′=Spec(k′). Then X′(y′)=(T(s)×ss′)×s′y′. Hence by 5.6, 5.27 for the third square follows from the usual proper base change theorem for the proper morphism T(s)→s and the base change by s′→s.
∎
5.28**.**
Let X and Y be non-commutative schemes satisfying (F) and let f:X→Y be a morphism satisfying the following condition 5.28.1.
5.28.1. Locally on Y, there are schemes S and T, a proper morphism T→S, a morphism Y→S, and an l.f.p. immersion j:X→Y×ST such that f is the composition X→jY×ST→Y.
We define Rf!F for a sheaf F of torsion abelian groups on Xet. Assume we have T→S, Y→S, and X→jY×ST as above. We define
[TABLE]
where p:Y×ST→Y. This is independent of the choices. In fact, for other T′→S′, Y→S′, and X→j′Y×S′T′, we have the third choices
T′′=T×T′→S′′=S×S′ (here × is the fiber product over Spec(Z)), Y→S×S′, and X→j′′Y×S′′T′′, and we have
a commutative diagram
[TABLE]
in which T′′→T×S′ is proper. We show that j′′ is an l.f.p. immersion and that j!=Rπ∗∘j!′′ for sheaves of torsion abelian groups. Then we will have Rp∗∘j!=Rp∗∘Rπ∗∘j!′′=Rp∗′′∘j!′′ where p′′:Y×S′′T′′→Y, and we have similarly Rp∗′∘j!′=Rp∗′′∘j!′′ and hence consequently Rp∗∘j!=Rp∗′∘j!′.
This j!=Rπ∗∘j!′′ is reduced to the following. Assume we are given non-commutative schemes X and Y satisfying (F), an l.f.p. immersion j:X→Y, a proper morphism of schemes T→S, a morphism Y→S, and a morphism j′:X→Y×ST over Y. Then j′ is an l.f.p. immersion and j!=Rp∗∘j!′ for sheaves of torsion abelian groups, where p:Y×ST→Y.
The fact j′ is an l.f.p. immersion follows from 3.18 (3). We have a canonical morphism j!F→Rp∗∘j!′F. We prove (j!F)yˉ→(Rp∗∘j!′F)yˉ is an isomorphism for every y∈Y. By 5.27, this is reduced to the case Y=yˉ. In this case, X is either yˉ or ∅. In the former case, both j!F and Rp∗∘j!F are F. The latter case is clear.
Thus Rf!F is well defined for sheaves of torsion abelian groups on Xet.
If X, Y, Z are non-commutative schemes satisfying (F) and if f:X→Y and g:Y→Z are morphisms satisfying the following condition 5.28.2 for a fixed scheme S, we have
[TABLE]
for sheaves of torsion abelian groups.
5.28.2. There are proper schemes T and T′ over S and l.f.p. immersions j:X→Y×ST and j′:Y→Z×ST′ such that f is the composition X→jY×ST→Y and g is the composition Y→j′Z×ST′→Z.
Note that then g∘f is the composition X→Z×S(T×ST′)→Z in which the first arrow is an l.f.p. immersion and the T×ST′→S is proper, and hence R(g∘f)! is defined.
Theorem 5.29**.**
Let X and Y be non-commutative schemes satisfying (F), and let f:X→Y be a morphism satisfying the condition 5.28.1.
(1) If Y′ is a non-commutative scheme satisfying (F) and if Y′→Y is a morphism, Rmf! and Rmf!′ with f′:X×YY′→Y′ commute with pullbacks of sheaves of torsion abelian groups.
(2) Rmf! sends constructible sheaves of abelian groups on Xet to constructible sheaves of abelian groups on Yet. If m≫0, Rmf!F=0 for all constructible sheaves F of abelian groups on Xet.
Remark. For a constructible sheaf F of abelian groups on Xet such that the orders of all stalks of F are invertible on Y, the constructibility of Rmf!F in (2) follows from 5.21.
Proof.
(1) follows from 5.27. ((1) for the case f is an l.f.p. closed immersion and the case f is an open immersion are clear by checking stalks.)
(2) By Noetherian induction, we may assume that Y is a prime non-commutative scheme and (Rf!F)C is constructible for every closed subset C=Y of Y and every constructible sheaf of abelian groups F on Xet.
In fact, if Rf!F is not constructible for some constructible F, there is a minimal element in the set of all closed subsets C of Y satisfying the following condition: (Rf!F)C′ is not constructible for some closed subset C′ of C and for some constructible sheaf F of abelian groups on Xet . If C=C1∪C2 for closed subsets Ci, we would have C=C1 or C=C2 because otherwise, for every closed subset C′ of C, the exact sequence
(1) in the proof of 5.21, which we apply by replacing X, F, C and C′ there with Y, Rmf!(F), C1∩C′ and C2∩C′ here, would show that (Rf!F)C′ is constructible. Hence this minimal C is irreducible.
We have (Rf!F)C=Rf!′(FX′) for this minimal C which we endow with the prime non-commutative scheme structure, where f′:X′=X×YC→Y. We may assume that Y is this minimal C.
By Noetherian induction by using the exact sequence (1) in the proof of 4.6, we may assume that X is a prime non-commutative scheme and Rf!(FC) is constructible for every closed subset C=X of X.
We may assume that Y is the closure of the image of the generic point of X in Y and that Y has the prime non-commutative scheme structure.
If X=U∪U′ for open subsets of X, we have an exact sequence 0→j!′′(FU′′)→j!(FU)⊕j!FU′→F→0, where U′′=U∩U′ and j, j′, j′′ are inclusion morphisms from U, U′, U′′ to X, respectively.
By this, we may assume that Y is an open subspace of Spec(A) for some prime ring A such that Z(A) is an excellent ring and A is finitely generated as a Z(A)-module,
X is an open subspace of Spec(B) for some prime ring B such that Z(B) is an excellent ring and B is finitely generated as a Z(B)-module,
B is an A-algebra of finite presentation, A→B is injective, and the morphism Spec(B)→Spec(A) is compatible
with f:X→Y. There are a non-zero element a of Z(A) and a non-zero element b of Z(B) such that b/a∈Z(B) and such that A[1/a] is
an Azumaya algebra over Z(A)[1/a], V:=Y×Spec(Z(A))Spec(Z(A)[1/a]) coincides with Spec(A[1/a]), B[1/b] is an Azumaya algebra over Z(B)[1/b], and U:=X×Spec(Z(B))Spec(Z(B)[1/b]) coincides with Spec(B[1/b]).
Consider the exact sequence 0→j!(FU)→F→FC→0 on Xet, where C=X∖U and j is the inclusion morphism U→X. By our Noetherian induction on X, Rf!(FC) is constructible. Let D=Y∖V. By our Noetherian induction on Y, (Rf!j!(FU))D is constructible. It remains to prove that the pullback of
Rf!j!(FU) on Vet is constructible, that is, Rg!(FU) is constructible, where g:U→V. Let h:Spec(Z(B)[1/b])→Spec(Z(A)[1/a]) and
let G be the sheaf on (Spec(Z(B)[1/b])et corresponding to FU via 5.6. Then
Rg!(FU) corresponds to Rh!G via 5.6 and the constructibility of the former is reduced to the constructibility of the latter in the scheme theory ([1], vol. 3, Exp. XIV, Theorem 1.1).
∎
5.30**.**
Let X be a non-commutative scheme satisfying (F) and let ℓ be a prime number which is invertible on X. We consider constructible Λ-sheaves on X for Λ=OE,E,Qˉℓ for a finite extension E of Qℓ (OE denotes the integer ring of E) and for an algebraic closure Qˉℓ of Qℓ, following [1], [3], [9].
An OE-sheaf on X is a family (Fn)n≥1 of sheaves Fn of OE/ℓnOE -modules on Xet endowed with an isomorphism Fn+1⊗OE/ℓn+1OEOE/ℓnOE≅Fn for each n. We say a Λ-sheaf is smooth if Fn are locally constant and finite. We say a Λ-sheaf is constructible if locally on X, there is a finite family of l.f.p. immersions Yi→X such that the set X is the disjoint union of these Yi and such that the pullback of F to each Yi is smooth.
The category of constructible E-sheaves on X is defined as the quotient category of the category of constructible OE-sheaves by the full subcategory consisting of objects which are killed by some powers of ℓn.
The category of constructible Qˉℓ-sheaves on X is defined to be the inductive limit of categories of constructible E-sheaves, where E ranges over all finite extensions of Qℓ in Qˉℓ.
In the case X satisfies (FZ), we consider mixed sheaves following [3]. Fix an isomorphism Qˉℓ≅C of commutative fields. A mixed Qˉℓ-sheaf on X is a constructible Qˉℓ-sheaf F satisfying the following condition. There is a finite family F(i) (0≤i≤m+1) of constructible Qˉℓ-subsheaves F(i) of F such that F(0)⊃F(1)⊃⋯⊃F(m+1)=0 and such that for 0≤i≤m, there is an integer w(i) satisfying the following condition (*). For each closed point x of X, since OX,x/p(x)≅Mr(Fq) for some r≥1 and for some finite field Fq, the stalk (F(i)/F(i+1))xˉ, which is a finite-dimensional Qˉℓ-vector space, has a linear action of Gal(Fˉq/Fq) and in particular, the action of the Frobenius φx∈Gal(Fˉq/Fq), where φx(a)=aq for a∈Fˉq.
(*) For each x∈X, all the eigen values of φx−1 on (F(i)/F(i+1))xˉ regarded as elements of C× have absolute value qw(i)/2.
Theorem 5.31**.**
Let X, Y and f:X→Y be as in 5.21, let ℓ be a prime number which is invertible on Y, and let F be a constructible Qˉℓ-sheaf on Xet. Then Rmf∗F is a constructible Qˉℓ-sheaf on Y for every m and is [math] for m≫0.
Proof.
This follows from 5.21 as in the case of schemes treated in [9]. ∎
Theorem 5.32**.**
Let X, Y and f:X→Y be as in 5.29. Let ℓ be a prime number which is invertible on Y.
(1) Rmf! sends constructible Qˉℓ sheaves on X to constructible Qˉℓ-sheaves on Y. If m≫0, Rmf!F=0 for all constructible Qˉℓ-sheaves F on X.
(2) For a non-commutative scheme Y′ satisfying (F) and for a morphism Y′→Y, Rmf! and Rmf!′ with f′:X×YY′→Y′ commute with pullbacks of constructible Qˉℓ-sheaves.
(3) If Y satisfies (FZ), Rmf! sends mixed Qˉℓ sheaves on X to mixed Qˉℓ sheaves on Y.
Proof.
(1) and (2) are deduced from (2) and (1) of 5.29, respectively, as in the case of schemes treated in [9]. (Actually, (1) follows from 5.31.)
(3) By the method of the proof of (2) of 5.29, we are reduced to the case X and Y are schemes, and to the theory of mixed sheaves of Deligne in [3].
∎
6 Betti cohomology
For a scheme S over C locally of finite type, the set S(C) of C-valued points has the topology called the classical topology.
Recall that as a set, S(C) is identified with the set of all closed points of S.
In this Section 6, we consider a non-commutative version of this classical topology. For a non-commutative scheme X over C satisfying the finiteness condition (FC) (4.10), we endow the set Xcl of all closed points of X with a topology which we call the classical topology. (The notation cl in Xcl stands for classical.) We will call the cohomology of Xcl the Betti cohomology of X.
We will prove a comparison theorem (6.11) between the étale cohomology of X and the Betti cohomology of X.
Lemma 6.1**.**
Assume X satisfies (FC*), and let x∈X. Then x∈Xcl if and only if the residue field of the local ring Z(OX,x) is C.*
For X satisfying (FC), the classical topology on Xcl is defined in 6.4 (1). The style of this definition follows the following 6.2 and 6.3 concerning the classical topologies of S(C) for schemes S locally of finite type over C.
6.2**.**
Recall that there is a unique way to endow the set S(C) for every scheme S over C locally of finite type with a topology (the classical topology) satisfying the following (i) and (ii).
(i) If U is an open set of S, U(C) is open in S(C) and the classical topology of U(C) is the restriction of the classical topology of S(C).
(ii) If S is Spec(R) for a
finitely generated commutative ring R over C, the classical topology of S(C) is the topology of simple convergence of HomC(R,C)=S(C), and it is also understood as in 6.3 below.
Lemma 6.3**.**
Let R be a commutative finitely generated algebra over C, let S=Spec(R), and let Rcont be the ring of all C-valued continuous functions on S(C) for the classical topology. Let ι:S(C)→max(Rcont) be the map which sends s∈S(C) to the kernel of the evaluation Rcont→C at s. Then ι is injective and the classical topology of S(C) coincides with the pullback of the Zariski topology (Jacobson topology) of max(Rcont) via ι.
Proof.
This is well known.
We are reduced to the case R=C[T1,…,Tn]. In this case, for α=(α1,…,αn)∈S(C)=Cn, for hr∈Rcont (r>0) defined by hr(z):=∏i=1nmax(r−∣zi−αi∣,0) (z∈Cn), the sets {z∈Cn∣hr(z)=0}={z∈Cn∣∣zi−αi∣<1(1≤i≤n)} for r>0 form a base of neighborhoods of α.
∎
Proposition 6.4**.**
(1) There is a unique way to endow Xcl with a topology for every non-commutative scheme X over C satisfying the condition (FC*), satisfying the following conditions (i) and (ii).*
(i) If U is an open subspace of X, Ucl is open in Xcl and the topology of Ucl is the restriction of the topology of Xcl.
(ii) Let R be a finitely generated commutative ring over C, let A be an R-algebra which is finitely generated as an R-module, and let X=Spec(A).
Let S=Spec(R), let Rcont be the ring of all C-valued continuous functions on S(C), and consider the injection ι:max(A)→max(Rcont⊗RA) which sends m∈max(A) to the kernel of
Rcont⊗RA→C⊗RA=A/m′A→A/m, where m′∈max(R) is the inverse image of m under R→A, the first arrow is defined by the evaluation Rcont→C at m′, and the map R→C in C⊗R is the evaluation at m′.
Then the topology of Xcl coincides with the pullback of the Zariski topology (Jacobson topology) of max(Rcont⊗RA) via ι.
We will call this topology of Xcl the classical topology.
(2) The classical topology of Xcl is the weakest topology such that the following sets DU,A(f) are open. Here A is a C-algebra such that Z(A) is a finitely generated C-algebra and A is finitely generated as a Z(A)-module, U is an open set of X endowed with a morphism U→Spec(A) over C, f is an element of Z(A)cont⊗Z(A)A (Z(A)cont is as in (1) (ii)), and DU,A(f)⊂Ucl denotes the inverse image of
D(f):={m∈max(Z(A)cont⊗Z(A)A)∣f∈/m} under Ucl→max(A)→ιmax(Z(A)cont⊗Z(A)A).
The following (3) (resp. (4)) is similar to (2) (resp. (1) (ii)) but uses analytic functions instead of continuous functions.
(3) The classical topology of Xcl is the weakest topology such that the following sets DU,A,V(f) are open. Here A and U are as in (2), V is an open set of S(C), where S:=Spec(Z(A)), f is an element of AV:=OSan(V)⊗Z(A)A,
where OSan is the structure sheaf of the complex analytic space associated to S, and DU,A,V(f)⊂Ucl×S(C)V denotes the inverse image of D(f):={m∈max(AV)∣f∈/m} under the map Ucl×S(C)V→max(AV) which sends (u,v)∈Ucl×S(C)V with image s in S(C) to the kernel of AV→C⊗O(S)O(U)→OU,u/p(u). Here the first arrow is induced by the evaluation OSan(V)→C at v and O(S)→C is the evaluation at s.
(4) Let R be a finitely generated commutative ring over C, let A be an R-algebra which is finitely generated as an R-module, and assume that X is an open subspace of Spec(A). Then the classical topology of Xcl is the weakest topology such that the following sets DV(f) are open.
Here V is an open set of S(C), where S:=Spec(R), f is an element of AV:=OSan(V)⊗RA,
and DV(f)⊂Ucl×S(C)V denotes the inverse image of D(f)⊂max(AV).
(5) Let x∈Xcl. Take an open neighborhood U of x, a finitely generated commutative ring R over C, an R-algebra A which is finitely generated as an R-module, and an open immersion U→Spec(A) over C.
Then a base of neighborhoods of x in Xcl (for the classical topology) is given as follows.
Denote Spec(R) by S, the canonical map Ucl→S(C) by π, and let s=π(x).
Take an open neighborhood W of s in S(C) and an element e∈OSan(W)⊗RA such that the image of e in OS,san⊗RA is an idempotent for (OS,san⊗RA,m), where m is the maximal two-sided ideal corresponding to x (4.7, such e exists by 4.8 (1)).
Let D(e) be the open subset of π−1(W)⊂Xcl consisting of all points y such that the image of e in OX,y/p(y) is not zero.
Then when W′ ranges over all open neighborhoods of s in W, D(e)∩π−1(W′) form a base of neighborhoods of x in Xcl.
Proof.
Let T2 (resp. T3) be the topology of Xcl defined as in the above (2) (resp. (3)) in 6.4. Assuming we are given R and A and an open immersion X→Spec(A) as in (ii) of (1) (resp. as in (4)), let T1 (resp. T4) be the topology of Xcl defined as in (ii) of (1) (resp. as in (4)). Let x∈X. For the proof of 6.4, it is sufficient to prove that for each k=1,2,3,4 and for the topology Tk, D(e)∩π−1(W′) in (5) form a base of neighborhoods of x in Xcl.
We first prove that D(e)∩π−1(W′) is a neighborhood of x for Tk. The cases k=3,4 are clear. The case k=2 follows from the case k=1. In the case k=1, take an open set V and a closed set C of S(C)
for the classical topology such that s∈V⊂C⊂W. Write e=∑i=1nbi⊗ai with bi∈OSan(W) and ai∈A, and let ci be a C-valued continuous function on S(C)
which coincides with bi on V and with [math] on S(C)∖C. Let e′=∑i=1nci⊗ai∈Rcont⊗RA. Then e and e′ coincide on π−1(V). This shows that for an open neighborhood W′ of s in V, D(e)∩π−1(W′) is a neighborhood of x for T1.
We next prove that each neighborhood of x in Xcl for Tk contains D(e)∩π−1(W′) for some W′. For a C-algebra A such that Z(A) is a finitely generated C-algebra and A is finitely generated as a Z(A)-module and for a morphism X→Spec(A) over C, locally on X, we have a C-algebra B such that Z(B) is a finitely generated C-algebra and B is finitely generated a Z(B)-module and
an open immersion X⊂Spec(B) over C with B, and a morphism A→B in (Alg/C) such that X→Spec(A) factors as X→Spec(B)→Spec(A). From this, the case T2 is reduced to the case T1, and the case T3 is reduced to the case T4. We consider T1 (resp. T4).
Let f be an element as in the explanation of T1 (resp. T4). It is sufficient to prove that if x∈/DX,A,S(f) (resp. DV(f)), DX,A,S(f) (resp. DV(f)) contains D(e)∩π−1(W′) for some open neighborhood W′ of s in W. Let OScont be the sheaf of C-valued continuous functions on S(C). For (∗)=an,cont, let fx be the image of f in OS,s(∗)⊗RA. By 4.8 (2) applied to I=(fx), e∈(fx) in this ring. Hence for some open neighborhood
W′ of s in W (resp. W∩V), we have e∈(f) in OS(∗)(W)⊗RA. Hence D(e)∩π−1(W′)⊂DX,A,S(f) (resp. DV(f)).
∎
Lemma 6.5**.**
For a morphism Y→X of non-commutative schemes over C satisfying (FC*), the induced map Ycl→Xcl is continuous.*
Let S be a scheme over C locally of finite type, and let X=Spec(B) for an Azumaya algebra B over OS. Then the canonical map Xcl→Scl
is homeomorphism.
Proof.
We may assume that S=Spec(R) for a finitely generated commutative C-algebra R and X=Spec(A) for an Azumaya algebra A over R. Then max(Rcont⊗RA)→max(Rcont) is a homeomorphism for the Zariski topology because Rcont⊗RA is an Azumaya algebra over Rcont.
∎
6.7**.**
For a scheme S over C of finite type, we have the sheaf OSan on S(C). This is generalized to a non-commutative version as follows.
For a non-commutative scheme X over C satisfying (FC), we have a sheaf OXan of rings on Xcl defined as follows. We will show that we have a non-commutative spaces of prime ideal Xan=(Xcl,OXan,(p(x))x).
Let X be a non-commutative scheme over C satisfying (FC). For an open set U of Xcl, let F(U):=∏x∈UO^X,x where O^X,x denotes the completion of OX,x by the ideal generated by the maximal ideal of Z(OX,x). Then OXan is the subsheaf of F characterized as follows. If A is a C-algebra such that Z(A) is a finitely generated C-algebra and A is a finitely generated Z(A)-module, and if U is an open set of X and if U→Spec(A) is an open immersion over C, the restriction of OXan to Ucl coincides with the image of the canonical injection f−1(OSan⊗Z(A)A)→F, where S=Spec(Z(A)) and f is the canonical map Ucl→S(C).
For x∈Xcl, the stalk OX,xan is identified with OS,san⊗Z(OX,x)OX,x where s is the image of x in S(C) and its center is identified with OS,san.
We have the non-commutative space of prime ideals Xan=(Xcl,OXan,(p(x))x), where p(x) for x∈Xcl is the kernel of
the composite map Z(OX,xan)⊗Z(OX,x)OX,x→C⊗Z(OX,x)OX,x→OX,x/p(x), in which the first arrow is given by the evaluation Z(OX,xan)→C at x.
We have a functor X↦Xan from the category of non-commutative schemes over C satisfying (FC) to the category of non-commutative spaces of prime ideals.
Theorem 6.8**.**
Let X, Y, Z be non-commutative schemes over C satisfying (FC*), and assume that we are given morphisms Y→X and Z→X over C. Then
the canonical map*
[TABLE]
is a homeomorphism.
Proof.
First we consider the case X=Spec(C). Write Y⊗Spec(C)Z by Y⊗CZ. We prove that the canonical map (Y⊗CZ)cl→Ycl×Zcl is a homeomorphism. We are reduced to the case where Y=Spec(B), Z=Spec(C) for C-algebras. (B and C are C-algebras whose centers are finitely generated C-algebras and which are finitely generated as modules over the centers.) Then this canonical map is bijective because we have the converse map Ycl×Zcl→(Y⊗CZ)cl=(B⊗CC)cl which sends a pair (p,q) for p∈Ycl and q∈Zcl to the kernel of B⊗CC→B/p⊗CC/q≅Mm(C)⊗CMn(C)≅Mmn(C).
We show that this bijection is a homeomorphism. Note that B⊗CC is a finitely generated module over Z(B)⊗CZ(C). Let S=Spec(Z(B)), T=Spec(Z(C)).
Let R be the image of Z(B)⊗CZ(C)→A⊗CB and let P=Spec(R). Then the composition (Y⊗CZ)cl→P(C)→⊂S(C)×T(C) is identified with the product of the surjections Ycl→S(C) and Zcl→T(C), and hence the closed immersion P(C)→S(C)×T(C) is bijective and hence is a homeomorphism.
Let (p,q)∈Ycl×Zcl and let x be the corresponding point of (Y⊗CZ)cl. Let e be as in (5) of 6.4 for (Y,p) and let e′ be one for (Z,q). Then e⊗e′ is as in (5) of 6.4 for (Y⊗CZ,x). Let π:Ycl×Zcl→S(C)×T(C) be the projection, and let (s,t)=π(p,q). Then by (5) of 6.4, when U ranges over sufficiently small neighborhoods of s in S(C) and U′ ranges over sufficiently small neighborhoods of t in T(C), (D(e)×D(e′))∩π−1(U×U′) form a base of neighborhoods of (p,q) in Ycl×Zcl. Denote the projection (Y⊗CZ)cl→P(C) also by π and let y=π(x). By (5) of 6.4, when U ranges over sufficiently small neighborhoods of y in P(C), D(e⊗e′)∩π−1(U) form a base of neighborhoods of x in (Y⊗CZ)cl. We have D(e⊗e′)=D(e)×D(e′). This shows the coincidence of the topology of (Y⊗CZ)cl and that of Ycl×Zcl.
Now consider a general X. We may assume that X=Spec(A), and Y=Spec(B), Z=Spec(C) for A-algebras B, C, respectively. We first show that the map (Y⊗XZ)cl→Ycl×XclZcl is bijective. This is a map induced from the bijection (Y⊗CZ)cl→Ycl×Zcl. Hence it is enough to prove the surjectivity of the map in problem. Assume p∈Ycl and q∈Zcl have the same image r in Xcl. Then A/r≅Mr(C) for some r. Since the category of C-algebras and the category of Mr(C)-algebras are equivalent by E↦Mr(E) (1.15), the A/r-algebra B/p corresponds to a C-algebra E and B/p≅Mr(E). Since B/p≅Mm(C) for some m, E must be isomorphic to Mb(C) for some b and hence B/p≅Mb(Mr(C)) for some b. Similarly, C/q≅Mc(Mr(C)) for some c. Hence B/p⊗A/rC/q≅Mbc(Mr(C))≅Mbcr(C). Let s∈(Y⊗XZ)cl be the kernel of B⊗AC→Mbcr(C). Then the map in problem sends s to (p,q).
Since the topology of (Y⊗XZ)cl is the restriction of the topology of (Y⊗CZ)cl and the topology of Ycl×XclZcl is the restriction of the topology of Ycl×Zcl, this bijection (Y⊗XZ)cl→Ycl×XclZcl is a homeomorphism.
∎
Proposition 6.9**.**
Let X,Y be non-commutative schemes over C satisfying (FC*), and let f:X→Y be an étale morphism over C. Then the induced map fcl:Xcl→Ycl is
a local homeomorphism.*
Proof.
Let x∈Xcl, y:=fcl(x)∈Ycl. By 5.12, there are morphisms of non-commutative schemes U→⊂V′→V→⊂Y, where V→Y and U→V′ are open immersions and V′→V is a morphism of the type explained below, an open immersion U→X×YU over U, and a point u of U whose image under U→X×YU→X is x. Here V′→V is such that there are a scheme S over C of finite type, a scheme S′ over S which is étale over S, a morphism V→S over C, and V′=V×SS. By 6.8, (V×SS′)cl=Vcl×S(C)S′(C) and (X×YU)=Xcl×YclUcl. The maps Vcl→Ycl, Ucl→(V′)cl, Ucl→(X×YU)cl are open immersions of topological spaces. The map S′(C)→S(C) is a local homeomorphism. Hence the maps Ucl→Xcl and Ucl→Ycl are local homeomorphisms. This shows that there are an open neighborhood of x in Xcl and an open neighborhood of y in Ycl such that fcl induces a homeomorphism between these open neighborhoods.
∎
6.10**.**
By 6.9, we have a morphism from the topos X~cl of sheaves on Xcl to the topos X~et of sheaves on Xet.
The following is an analogue of Artin’s comparison theorem ([1] vol. 3, Exp. XVI, 4).
Theorem 6.11**.**
Let X and Y be quasi-compact non-commutative schemes over C satisfying (FC*), and let f:X→Y be an l.f.p. morphism over C.*
Let F be a constructible sheaf of abelian groups (resp. sets, resp. groups) on Xet. Then we have an isomorphism (Rmf∗F)cl→≅Rmfcl,∗(Fcl) for every m (resp. for m=0, resp. for m=1). Here ()cl denotes the pullback to Ycl, Xcl, respectively.
Let the situation be as in 5.22 and assume that the center of Z(A) is a finitely generated C-algebra. Via the homeomorphism Ucl→Vcl(6.6), identify a sheaf on Ucl with the corresponding sheaf on Vcl.
Then for a sheaf F of abelian groups (resp. sets, resp. groups) on Ucl, the canonical morphism πcl−1Rmhcl,∗F→Rmgcl,∗F is an isomorphism for every m (resp. for m=0, resp. for m=1).
Proof.
Let S=Spec(Z(A)). Let y∈Ycl and let s=πcl(y)∈S(C) (πcl is the map Ycl→S(C) induced by π:Y→S). It is sufficient to prove that the map (Rmhcl,∗F)s→(Rmgcl,∗F)y is an isomorphism.
Recall that by 6.4 (5), a base of neighborhoods of y in Ycl is given as follows. Take an open neighborhood W of s in S(C) and an idempotent e∈OSan(W)⊗Z(A)A such that the image of e in OS,san⊗Z(A)A is an idempotent for (OS,san⊗Z(A)A,m) where m is the maximal two-sided ideal corresponding to y (4.7).
Let D(e) be the open subset of πcl−1(W)⊂Ycl consisting of all points y′ such that the image of e in OY,y′/p(y′) is not zero.
Then when W′ ranges over all open neighborhoods of s in W, D(e)∩πcl−1(W′) form a base of neighborhoods of y in Ycl. Let Vcl(W′)⊂Vcl be the inverse image of W′ under Vcl→S(C) and let Ucl(W′)⊂Ucl be the inverse image of W′ under Ucl→S(C), so Ucl(W′) is the inverse image of πcl−1(W′) in Ucl under Ucl→Ycl. By 5.23, the map Ucl→Ycl factors through D(e)∩πcl−1(W′). From this, we have
Let the situation be as in 6.12, Then
for a constructible sheaf F of abelian groups (resp. sets, resp. groups) on Uet, we have an isomorphism (Rmg∗F)cl→≅Rmgcl,∗(Fcl) for every m (resp. for m=0, resp. for m=1). .
Proof.
By the comparison theorem of Artin, (Rmh∗F)cl→≅Rmhcl.∗Fcl is an isomorphism. This comparison for Rmg∗F follows from it by 5.24 and 6.12.
∎
The following 6.14 and 6.15 are preparations for the proof of 6.11 (3).
6.14**.**
We give basic things about torsors, We work on a site S.
(1) Let G, H be sheaves of groups, let p,q:G→H be homomorphisms, and let F⊂G be the equalizer of (p,q). Assume that the morphism ν:G→H;g↦p(g)q(g)−1 is surjective (equivalently, we have an isomorphism of sheaves of sets G/F→≅H;g↦p(g)q(g)−1). Then we have the following (1.1) and (1.2).
(1.1) The category of F-torsors is equivalent to the categories of pairs (T,ι) where T a G-torsor and ι is an isomorphism of H-torsors p∗(T)→≅q∗(T). Here p∗(T) (resp. q∗(T)) denotes the H-torsor H×GT where G→H is p (resp. q). The equivalence is given by T↦(T′,ι) where T′=G×FT and ι is the evident isomorphism.
The converse functor sends (T,ι) to the equalizer of the two maps T→q∗(T), one is the evident map and the other is the composition T→p∗(T)→ιq∗(T).
(1.2) The canonical map from H1(S,F) to the equalizer of the two maps H1(S,G)→H1(S,H) induced by p and q, respectively, is surjective.
This follows from (1.1).
(2) Let F be a sheaf of group and let E an F-torsor. Then we have a sheaf F′ of groups which we call the twist of F by T defined as follows. It is the sheaf of morphisms h:E→F such that h(fx)=fh(x)f−1 for f∈F, x∈E. The group structure of F′ is as follows. For h1,h2∈F′, (h1h2)(x)=h1(x)h2(x) (x∈T).
The category of F-torsors and the category of F′-torsors are equivalent by T↦T′ where T′ is the sheaf of morphisms t:E→T such that t(fx)=ft(x) for f∈F, x∈E. The F′-torsor structure on T′ is as follows. For h∈F′ and t∈T′, (ht)(x)=h(x)t(x).
This functor sends the F-torsor E to the the trivial F′-torsor F′.
The converse functor sends an F′-torsor T′ to the F-torsor T of sheaves of morphisms s:E→T′ which satisfy the following condition. For x∈E, f∈F, h∈F′, s(fx)=hs(x) if h(x)=f−1. The action of F on T is given by (fs)(x)=s(f−1x) (f∈F, x∈E).
Lemma 6.15**.**
Let f:X→Y be as in the hypothesis of 6.11. Let G,H be sheaves of groups on Xet and let p,q:G→H be homomorphisms such that ν:G→H;g↦p(g)q(g)−1 is surjective.
Let F⊂G the equalizer of (p,q). (That is, we are in the situation of 6.14 with S=Xet). Then if (Rmf∗G)cl→≅Rmfcl,∗Gcl and (Rmf∗H)cl→≅Rmfcl,∗Hcl for m=0,1, we have (R1f∗F)cl.→≅R1fcl,∗Fcl.
Proof.
We apply 6.14 to the sites S=(X×YY′)et for Y′∈Yet and for the pullbacks of F, G, H on S.
We prove that the map
(1) (R1f∗F)y→(R1fcl,∗Fcl)y
is bijective for y∈Ycl. We use the commutative diagram of exact sequences of pointed sets.
[TABLE]
The vertical arrows are assumed to be isomorphisms except the third arrow. Note that for an element E of (R1f∗F)y, we have the twists of F, G, H by E and the images of E in (R1f∗G)y and (R1f∗H)y (6.14 (2)), defined on some étale neighborhood of y, and
we have the corresponding diagram as above for these twists.
We prove the injectivity of (1). Let a,b∈(R1f∗F)y and assume that they have the same image in (R1Fcl,∗Fcl)y. By twisting F by b, we may assume b is the trivial element of (R1f∗F)y. Then since the image of a under (R1f∗F)y→(R1f∗G)y→≅(R1fcl,∗Gcl)y is the trivial element, the image of a in (R1f∗G)y is the trivial element. The rest is the diagram chasing using the above diagram.
We prove the subjectivity of (1). Let a∈(R1fcl,∗Fcl)y. The image a2 of a in (R1fcl,∗Gcl)y comes from an element a3 of (R1f∗G)y.
Since p(a2)=q(a2) in (R1fcl,∗Hcl)y, we have p(a3)=q(a3) in (R1f∗H)y.
By (1.2) in 6.14, a3 comes from an element a4 of (R1f∗F)y. By twisting F by a4, we may assume that a4 is the trivial element of (R1f∗F)y. Then a3 and hence a2 are trivial elements. Thus we are reduced to the case a2 is the trivial element. The rest is the diagram chasing using the above diagram.
∎
6.16**.**
We prove 6.11.
The proof is similar to the proof of 5.21 given in 5.26.
We first prove the part of 6.11 for sheaves of abelian groups.
By using the exact sequence (1) in 5.26 and by the arguments as in 5.26 using this exact sequence, we are reduced to the following case: X and Y are prime non-commutative schemes, the image of X in Y is dense, and 6.11 for sheaves F of abelian groups on Xet is known to be true if F has supports in some closed subset C=X of X.
By using the exact sequence (2) in 5.26 and the similar exact sequence
⋯→Rmfcl,∗Fcl→RmfUcl,∗(FU)cl)⊕…, we are reduced to the situation 6.12. Assume we are in the situation 6.12, and consider the distinguished triangle (3) in 5.26. We have an isomorphism
(Rf∗Rj∗(FU))cl→≅Rfcl,∗(Rj∗(FU))cl because
(Rf∗Rj∗(FU))cl≅R(f∘j)cl,∗(FU,cl)≅Rfcl,∗Rj∗,cl(FU,cl)≅Rf∗,cl(Rj∗(FU))cl, where the first isomorphism is by 6.13 and the third isomorphism is by the case X=Y of 6.13. Furthermore, Rj∗(FU) is constructible (5.21) and hence (Rj∗(FU))C is constructible. Hence by the distinguished triangle (3) in 5.26 and by Noetherian induction, we obtain the comparison theorem for F.
The proof of the part of 6.11 for sheaves of sets is similar to that for sheaves of abelian groups by using (4), (5), (6) in 5.26.
The proof of the part of 6.11 for sheaves of groups is also similar to that for sheaves of abelian groups, but we replace (1), (2), (3) in 5.26 by the following (1), (2), (3), respectively.
(1) Let C and C′ be closed subsets of X and let F be a sheaf of groups on Xet. Let G:=FC×FC′, let H=FC∩C′, and let p,q:G→H be the two projections. Then the morphism G→H;g↦p(g)q(g)−1 is surjective and FC∪C′ coincides with the equalizer of (p,q).
(2) Let Ui (i=1,2) be open sets of X such that X=U1∪U2 and let U3=U1∩U2. Let F be a sheaf of groups on Xet, let G=j1,∗(FU1)×j2,∗(FU2), H=j3,∗(FU3), where ji is the inclusion morphism Ui→X, and let p,q:G→H be the two canonical homoorphisms.
Then the morphism G→H;g↦p(g)q(g)−1 is surjective and F coincides with the equalizer of (p,q).
(3) Let the situation be as in 6.12. Let F be a sheaf of groups on Xet, let G=j∗(FU)×FC, H=(j∗(FU))C, and let p,q:F→G be the two canonical homomorphisms. Then the morphism G→H;g↦p(g)q(g)−1 is surjective and F coincides with the equalizer of (p,q).
For a quasi-compact non-commutative scheme X over C satisfying (FC), let SX (resp. SX,C for a closed subset C of X) be the statement that 6.11 for sheaves of groups with m=1 is true for every Y and every f:X→Y and every F (resp. every F with supports in C) as in the hypothesis in 6.11.
By (1) and 6.15, if
SX,C and SX,C′ are true, then SX,C∪C′ is true. By this, the proof of SX is reduced to the following case: X and Y are prime non-commutative schemes, the image of X in Y is dense, and SX,C is true if F has supports in some closed subset C=X of X.
We prove that in (2), if SUi are true for i=1,2,3, then SX is true. By (2) and by 6.15, it is sufficient to prove that the comparison theorem is true for the sheaf of groups Fi=ji,∗(FUi) on Xet for each i. Write FUi by E for simplicity. Then
(R1f∗Fi)cl is the kernel of (R1(f∘ji)∗E)cl→(f∗R1ji,∗E)cl,
R1fcl,∗Fi,cl=R1fcl,∗ji,cl,∗Ecl (we use 6.11 for sheaves of sets in this identification) is the kernel of R1(f∘ji)cl,∗Ecl→fcl,∗R1ji,cl,∗Ecl. We have (R1(f∘ji)∗E)cl≅R(f∘ji)cl,∗Ecl by SUi, (f∗R1ji,∗E)cl≅fcl,∗(R1ji,∗E)cl≅fcl,∗R1ji,cl,∗Ecl by 6.11 for sheaves of sets and by the case Y=X of SUi.
By this, we are reduced to the situation of 6.12. Assume we are in the situation 6.12, and consider
the above (3). Note that j∗(FU) is constructible (5.21) and hence (j∗(FU))C is constructible. By Noetherian induction, the comparison theorem is true for FC and (j∗(FU))C. We show that the comparison theorem is also true for j∗(FU). In fact, R1f∗j∗(FU) is the kernel of R1(f∘j)∗(FU)→f∗R1j∗(FU) and the above proof for the comparison theorem for ji,∗(FUi) works by replacing SUi with 6.13. By (3) and 6.15, we have the comparison theorem for F. This completes the proof of 6.11.
From the case Y=Spec(C) of 6.11 for sheaves of sets (with m=0) and for sheaves of groups (with m=1), we obtain
Proposition 6.17**.**
Let X be a non-commutative scheme over C satisfying (FC*). Let F a constructible sheaf of groups on Xet. Then the category of F-torsors on Xet and the category of Fcl-torsors on Xcl are equivalent via T↦Tcl.*
We give some examples of the comparison of the étale topology and the classical topology.
First, we compare an algebraic theory 6.18 and an analytic theory 6.19.
6.18**.**
Example. Let R be a (commutative) Dedekind domain with field of fractions F, let A be an R-algebra which is finitely generated and torsion free as an R-module, and assume that A⊗RF is a central simple algebra over F. Let a:Spec(A)→Spec(R) be the canonical morphism, and let n≥1.
Then for the étale topology, we have:
(1)
R0a∗(Z/nZ)=Z/nZ and R1a∗(Z/nZ)=0.
(2) H1(Ret,Z/nZ)→≅H1(Aet,Z/nZ).
(3) Assume n is invertible in R. Then R2a∗(Z/nZ)≅⊕sis,∗(Z/nZ(−1))♯(Σ(s))−1 where s ranges over all points of Spec(R) of codimension one, is:s→Spec(R) is the inclusion morphism, and Σ(s) is the inverse image of s in Spec(A). We have Rma∗(Z/nZ)=0 for m≥3.
Proof.
(2) follows from (1) by the exact sequence 0→H1(Ret,R0a∗(Z/nZ))→H1(Aet,Z/nZ)→H0(Ret,R1a∗(Z/nZ)).
To prove (1) and (3), we may assume that R is a strict local discrete valuation ring. Let s be the closed point of Spec(R). Then Σ(s) is the set of all maximal two-sided ideals of A. For x∈Σ(s), let U(x)⊂X be the smallest neighborhood (4.9) of x. We have an exact sequence 0→Hx0(U(x))→H0(U(x))→H0(F)→Hx1(U(x))→H1(U(x))→H1(F)→… of étale cohomology groups, where the coefficients of the cohomology are Z/nZ. From this and from Hm(U(x))=0 for m≥1 and H0(U(x))=H0(F)=Z/nZ, we have Hxm(U(x))=0 for m=0,1, and Hxm(U(x))≅Hm−1(F) for m≥2. If n is invertible in R, then H1(F,Z/nZ)=Z/n(−1), Hm(F,Z/nZ)=0 for m≥2, and hence Hxm(U(x)) is Z/nZ(−1) if m≥2 and is [math] if m≥3. We have Hxm(X)≅Hxm(U(x)), and an exact sequence
0→⊕x∈Σ(s)Hx0(X)→H0(X)→H0(F)→⊕x∈Σ(s)Hx1(X)→H1(X)→H1(F)→…. (1) and (3) follow from this.
∎
6.19**.**
Let G be the group defined by generators α, β and the relations α2=1, αβα−1=β−1. Let R be a commutative ring in which 2 is invertible, and let A be the group ring R[G]. Then
Z(A) is the polynomial ring R[T] with T=β+β−1. We have an injective ring homomorphism
[TABLE]
over R which is injective and whose image consists of all matrices ((aij)1≤i,j≤2) such that a21≡0modT2−4. Hence A[1/(T2−4)]≅M2(R[T][1/(T2−4)]).
These facts about general R will be used in 9.24, but now let R=C, and let X=Spec(A)=Spec(C[G]), S=Spec(Z(A))=Spec(C[T]).
For n∈Z, we have
[TABLE]
The case n=0 of this is the consequence of 6.18 and the comparison theorem 6.11. In fact, for a:X→S and for Rma∗ for the étale topology, by 6.18, Rma∗(Z/nZ)=0 for m=0,2, R0a∗(Z/nZ)=Z/nZ,
and R2a∗(Z/nZ) is the sheaf which is zero outside {(T−2),(T+2)} and whose stalks at (T+2) and at (T−2) are Z/nZ.
We give an analytic proof of the above results on Hm(Xcl,Z/nZ) including the case n=0.
Note that Xcl→C=S(C) is a homeomorphism outside {2,−2}⊂C and that the inverse image of 2∈C (resp. −2∈C) in Xcl is a two point set.
Let E⊂C×R be the union of {(z,0)∣∣z−2∣≥1,∣z+2∣≥1} and two spheres
{(z,t)∣∣z−2∣2+t2=1} and {(z,t)∣∣z+2∣2+t2=1}.
We have a continuous map π:E→Xcl defined as follows. If z=2 and z=−2, π(z,t) is the unique point lying over z∈C=S(C) (that is, the maximal two-sided ideal of A generated by T−z). If z=2 (resp. z=−2), π(z,t) (here t=±1) is the maximal two-sided ideal of A generated by (β−1,α−t) (resp. (β+1,α−t) of A. Then the pullback by π induces isomorphisms Hm(Xcl,Z/nZ)→≅Hm(E,Z/nZ) for all m. The isomorphism H2(Xcl,Z/nZ)→≅(Z/nZ)2 is given by the pullback to the H2 of the two spheres inside E.
Let G be the group defined by generators α,β and the relation αβα−1=β−1 and let A=C[G]/(β−1)2, X=Spec(A).
Then Γ(X,OX)=A, and Γ(X,Z(OX))=Z(A)=C[t±1] with t=α2. Hence Γ(X,Z(OX)) does not contain a square root of t. We show that Γ(Xet,Z(OXet)) contains a square root ot t. After that, we will understand in an analytic way that a square root of t exists in Γ(Xcl,Z(OXan)) where OXan is as in 6.7, and then explain that the existence of a square root of t in Γ(Xet,Z(OXet)) can be deduced also from it.
Let A′=C[t±1/2]⊗C[t±1]A and let Y=Spec(A′). Let H={1,σ} be the Galois group of C[t±1/2] over C[t±1]. Then the canonical morphism Y→X is a covering in the étale site of X, H acts on Y over X, and H×Y→≅Y×XY;(h,a)↦(a,ha). To prove that a square root of t exists in
Γ(Xet,Z(OXet)), it is sufficient to prove that there is a square root of t in Γ(Y,Z(OY)) which is invariant under the action of H.
We have Γ(Y,OY)=A′×A′ on which σ acts as (a,b)↦(σ(b),σ(a)) (a,b∈A′). Then (t1/2,−t1/2)∈A′×A′ is H-invariant and is a square root of t.
We next show that Γ(Xcl,Z(OXan)) contains a square root of t. We have a homeomorphism Xcl≅C× in which u∈C× corresponds to the maximal two-sided ideal (α−u,β−1) of A. Let S=Spec(Z(A)). Then we have a homeomorphism S(C)≅C× in which v∈C× corresponds to the maximal ideal (t−v), and the canonical map f:Xcl→S(C) is identified with C×→C×;z↦z2. We have OXan=f−1(OSan⊗Z(A)A) and Z(OXan)=f−1(OSan). There is an element of Γ(Xcl,f−1(Z(OSan))) which has value u at the point u∈C×=Xcl. This element is a square root of t.
Now in general, let X be a non-commutative scheme over C satisfying (FC). Consider the exact sequence of sheaves
0→Z/2Z→Z(OXet)×→2Z(OXet)×→0 on Xet and
0→Z/2Z→Z(OXan)×→2Z(OXan)×→0 on Xcl. These exact sequences induce a commutative diagram
[TABLE]
Since H1(Xet,Z/2Z)→≅H1(Xcl,Z/2Z) (6.11), this diagram shows that for h∈Γ(X,Z(OX))×,
a square root of h exists in Γ(Xet,Z(OXet)) if and only if a square root of f exists in
Γ(Xcl,Z(OXan)).
7 Fundamental groups
7.1**.**
Let X be a non-commutative scheme. We say Xet is connected if the topos X~et of sheaves on Xet is connected. This means that the following equivalent conditions (i)–(iii) are satisfied. (i) The canonical map Σ→Γ(Xet,Σ) is bijective for every set Σ. (ii) The canonical map {0,1}→Γ(Xet,{0,1}) is bijective. (iii) X is not empty, and the constant sheaf on Xet associated to a one-point set is not a disjoint union of two non-empty subsheaves.
These conditions (i)–(iii) are equivalent to the following condition (iv). (iv) X is not empty, and there is no idempotent in Γ(Xet,Z(OXet)) other than [math], 1. In fact, since Z(OY,y) are local rings for all non-commutative schemes Y and for all y∈Y, the sheaf of idempotents in Z(OXet) on Xet is identified with the constant sheaf associated to the set {0,1}. Hence (iv) is equivalent to (ii).
If we replace Γ(Xet,Z(OXet)) in the above condition (iv) by Γ(X,Z(OX)), then the modified condition becomes equivalent to the condition X is connected. Hence if Xet is connected, then X is connected. Note that Γ(X,Z(OX)) and Γ(Xet,Z(OXet)) may be different (6.20). However,
the author does not know any example of a non-commutative scheme X such that X is connected but Xet is not connected.
Proposition 7.2**.**
Assume X satisfies (F). Then X is connected if and only if Xet is connected.
Proof.
Let e be the constant sheaf on Xet associated to a one-point set, and assume e=e1∐e2 for subsheaves ei of e. Then there are Ui∈Xet (i=1,2) such that U1∐U2→X is a covering and the canonical element of Γ(Ui,et,e) comes from Γ(Ui,et,ei) for i=1,2.
Let Vi be the image of Ui in X. Then Vi are open by 4.16, and V1∩V2 is empty. If X is connected, we have either V1 is empty or V2 is empty and hence have either U1 is empty or U2 is empty. Hence either e1=e or e2=e.
∎
7.3**.**
Assume Xet is connected.
Let k be a separably closed field, let n≥1, and let a:Spec(Mn(k))→X be a morphism (we call this morphism a base point). (We assume here that such a exists.) By 5.6, the topos of sheaves on Spec(Mn(k))et is equivalent to the category of sets, and hence the pullback by a is a point of the topos X~et of sheaves on Xet which we call the point a.
We define the fundamental group π1(X,a) as the pro-finite fundamental group of the connected topos X~et with the point a ([11] 8.49). More precisely:
Let C be the category of sheaves on Xet which are locally constant and finite. Then
with the pullback functor by a, C is a Galois category in the sense of [8] Section V 5. The pro-finite group π1(X,a) is the Galois group of this Galois category. If X=Spec(A), we denote π1(X,a) also as π1(A,a).
If (X′,a′) is another pair as (X,a), a commutative diagram
[TABLE]
induces a homomorphism π1(X′,a′)→π1(X,a).
If we only assume that a base point a→X exists but do not fix it, we have π1(X) as an object of the category of pro-finite groups in which morphisms are continuous homomorphisms considered modulo conjugacy. If we have a morphism X′→X (we assume (X′)et is connected and has a base point which is not fixed), we have a morphism π1(X′)→π1(X) in this category.
Theorem 7.4**.**
Let X be a non-commutative scheme over C satisfying (FC).
(1) The map from the set of all connected components of Xcl to the set of all connected components of X is bijective.
(2) The category of locally constant finite sheaves on Xet and that on Xcl are equivalent via the functor F↦Fcl.
(3) Assume X is connected and let a∈Xcl. Then π1(X,a) is canoniacally isomorphic to the pro-finite completion of π1(Xcl,a).
Proof.
We prove (1). Let Σ={0,1}. By 7.2, Γ(X,Σ)→Γ(Xet,Σ) is bijective. By 6.11 (2), Γ(Xet,Σ))→Γ(Xcl,Σ) is bijective. (1) follows from these.
We prove (3). Assume X is connected. For a finite group G, the set of all continuous homomorphisms (resp. all homomorphisms) from π1(X,a) (resp. π1(Xcl,a)) to G is identified with the set of all isomorphisms of pairs (T,ι) where T is a G-torsor on Xet (resp. Xcl) and ι is a trivialization of the G-torsor Ta. Hence (2) follows from the equivalence 6.17 of the categories of G-torsors.
(2) follows from (1) and (3).
∎
7.5**.**
For a scheme S, we have an equivalence between the category of finite étale schemes over S and the category of locally constant finite sheaves on Set, which sends an object T of the former category to the sheaf MorS(,T′) of the latter category.
We show an example of a non-commutative scheme X satisfying (F) such that:
(1) Let S and T be as above, and let X→S be a morphism and let Y=X×ST∈Xet. Then it can happen that the presheaf MorX(,Y) on Xet is not a sheaf.
(2) Let F be a locally constant finite sheaf on Xet. It can happen that F is not isomorphic to the sheaf associated to the presheaf MorX(,Y) for any object Y of Xet.
Let A, X=Spec(A) and S=Spec(C[t±1]) be as in 6.20.
For (1), let T=Spec(C[t±1/2]) and let Y=X×ST. Let F be the presheaf MorX(,Y) and let F~ be the associated sheaf. Then F(X)=MorX(X,Y) is empty because Γ(X,Z(OX)) does not contain a square root of t, but F~(X) is not empty because Γ(Xet,Z(OXet)) has a square root of t.
For (2), fix a square root t1/2 of t in Γ(Xet,Z(OXet)). Let G be the sheaf of square roots of this t1/2 in Z(OXet) on Xet. Then G is isomorphic locally on Xet to the constant sheaf associated to a set of order 2. However, there is no Y∈Xet such that G is the sheaf associated to the presheaf MorX(,Y). Assume Y exists. Since A/(β−1)≅C[α±1] and α is a square root of t in this ring, Y⊗AA/(β−1) must be isomorphic to Spec(C[α±1][T]/(T2−α)). Take any point y of Y and let x be the image of y in X. Let A′=C(t)⊗C[t±1]OX,x and let B′=C(t)⊗C[t±1]OY,y.
Since Y→X is flat, A→B and A′→B′ are flat, and B′ is a free A′-module of rank 2. Let Z be the center of B′. Since B′ is r.c. over A′, Z⊗C(t)A′→B′ is surjective. Note B′/(β−1)=C(α1/2).
Claim. The ring Z is a local ring. Let Z1 be the residue field of Z. Then dimC(t)(Z1)≤2.
We prove Claim. If Z is not a local ring, it is a product of two non-zero rings, and hence B′ is also a product of two non-zero rings, but this contradicts the fact
B′ divided by some two-sided nilpotent ideal is C(α1/2).
There is a subring Z0 of Z over C(t) such that Z0→Z1 is an isomorphism. We show 1, α, β−β−1 are linearly independent over Z0. Assume a+bα+c(β−β−1)=0 with a,b,c∈Z0. Then by applying α(−)α−1 to this, we have a+bα−c(β−β−1)=0. Hence c=0 and a+bα=0. If b=0, α=−a/b∈Z0 is central, a contradiction, and hence b=0 and hence a=0. This linear independence shows dimC(t)(Z0)⋅3≤dimC(t)B′=8, and hence dimC(t)(Z0)≤8/3<3. Hence dimC(t)(Z0)≤2. This completes the proof of Claim.
We have a surjection Z⊗C(t)C(α)→B′/(β−1)=C(α1/2). This induces a surjection Z1⊗C(t)C(α)→C(α1/2). This is impossible because α is a square root of t and dimC(t)(Z1)≤2.
8 Chow groups, relation to class field theory
We fix a quasi-compact excellent scheme T.
We define the Chow groups CHi(X) for non-commutative schemes X over T satisfying (FT) (4.10), and show that CHi is a covariant functor for X which are proper (8.8) over T. In the case T=Spec(Z), we relate CH0 to the unramified class field theory like in [12].
8.1**.**
For i≥0, let Ti be the set of points t of T such that the closure of t in T is i-dimensional.
8.2**.**
Let X be a non-commutative scheme over T satisfying (FT). For x∈X, let κ(x)=OX,x/p(x). Then κ(x) is a finite-dimensional simple algebra over a commutative field (4.11). If t denotes the image of x in T, the center of κ(x) is a finitely generated field over the residue field κ(t) of t. This is because locally on X, there are an affine scheme S=Spec(R) of finite type over T and an R-algebra A such that R is the center of A and such that A is finitely generated as an R-module, and an open immersion X→Spec(A) over T.
8.3**.**
Let X be as in 8.2.
Let Xi be the set of all points x of X such that the image t of x in T belongs to Tj and the center of κ(x) is of transcendence degree k over κ(t) for some j,k≥0 such that i=j+k.
If S, R, A and X→Spec(A) are as in 8.2, then for x∈X with image s in S, x∈Xi if and only if s∈Si. Here Si is defined by taking X=S, and not defined as the set of points of S whose closures in S are of dimension i. (For example, if T=Spec(Zp) and S=Spec(Qp), the unique point of S belongs to S1 though its closure on S is of dimension [math]. But see 8.9 concerning this point. )
Lemma 8.4**.**
Let X be as in 8.2. Let x′∈Xi, x∈Xj and assume that x belongs to the closure X′ of x′ in X. Endow X′ with the structure of a prime non-commutative scheme (3.23). Then the center Z(OX′,x) is an excellent local ring of dimension i−j. In particular, we have i≥j. We have i>j if x′=x.
Proof.
By replacing X by X′, we may assume that X is the closure of x′. Working locally on X, we may assume that there are S, R, A, and X→Spec(A) as in 8.2. By replacing R by Z(A), we may assume R=Z(A). Let s′ be the image of x′ in S and let s be the image of x in S. Then OS,s=Z(OX,x). Since s′∈Si and s∈Sj, OS,s is of dimension i−j. If i=j, we have s′=s and x and x′ correspond to maximal two-sided ideals of OX,x by 4.6 (1). Hence x=x′.
∎
where K1 and K0 are the algebraic K-groups of finitely generated projective left modules and ∂ is defined as below.
Note that for a ring A, K0(A) is generated by the classes [P] and K1(A) is generated by the classes [P,α], where P is a finitely generated projective left A-modue and α is automorphism of P. We have a homomorphism A×→K1(A) which sends a∈A× to (A,ra) where ra(b)=ba (b∈A). For x∈X, we have
Z→≅K0(κ(x)) by sending 1∈Z to the class [E] of a simple κ(x)-module (any left κ(x)-module is projective, and the class [E] is unique) and we have an isomorphism κ(x)×/[κ(x)×,κ(x)×]→≅K1(κ(x)).
For x∈Xi and u∈Xi+1, the component ∂u,x:K1(κ(u))→K0(κ(x)) of ∂ is as follows.
If x is not in the closure of u in X, ∂u,x=0. Assume that x is contained in the closure of u. Let C the closure of u in X endowed with the prime non-commutative scheme structure (3.23). Let M be the category of finitely generated left OC,x-modules. Since OC,x is a left (and also right) Noetherian ring, M is an abelian category. Let Mfin⊂M be the subcategory of M consisting of OC,x-modules of finite length. By 8.4, Z(OC,x) is a Noetherian local ring of dimension 1 and OC,x is a finitely generated module over Z(OC,x). Hence the quotient category M/Mfin is equivalent to the category of finitely generated κ(u)-modules. By the localization theory in K-theory (Quillen [17]), we have an exact sequence
[TABLE]
and isomorphisms
[TABLE]
where m ranges over all maximal two-sided ideals of OC,x.
The map ∂u,x is defined to be the composition K1(κ(u))→∂K0(Mfin)≅⊕mZ→Z, where the last arrow is to take the p(x)-component.
The above map K1(κ(u))→∂K0(Mfin) is described as [P,α]↦[M2/α(M1)]−[M2/M1] where Mi are finitely generated left OC,x-submodules of P such that κ(u)⊗OC,xMi=P and such that M2⊃M1 and M2⊃α(M1).
Lemma 8.6**.**
Let X and Y be non-commutative schemes over T satisfying (FT*), let f:X→Y be a morphism over T, and let x∈Xi. Then y:=f(x) belongs to Yj for some j≤i.
If j=i, κ(x) is a finitely generated projective left (resp. right) κ(y)-module.*
Proof.
Working locally on X and Y, we may assume that there are affine schemes S=Spec(R) and S′=Spec(R′) over T of finite type, a morphism S′→S over T, an R-algebra A which is finitely generated as an R-module, an R′-algebra A′ which is finitely generated as an R′-module, and open immersions Y→Spec(A) and X→Spec(A′) via which the morphisms S′→S and X→Y are compatible. Let s be the image of y in S and let s′ be the image of x in S′. Then s∈Sj and s′∈Si′, and the image of s′ under S′→S is s. Hence we have that i≥j and that if i=j, κ(s′) is a finite extension of κ(s). Since κ(y) is a finitely generated κ(s)-module and κ(x) is a finitely generated κ(s′)-module, the last fact shows that if i=j, κ(x) a finitely generated as a κ(y)-module. Every finitely generated projective κ(y)-module is projective.
∎
8.7**.**
For non-commutative schemes X and Y over T satisfying (FT) and for a morphism f:X→Y over T, and for i,m≥0, we have a homomorphism
[TABLE]
defined as follows. For x∈Xi and y∈Yi, the (x,y)-component Km(κ(x))→Km(κ(y)) of this map is [math] unless y=f(x). In the case y=κ(x), κ(x) is a finitely generated projective left κ(y)-module by 8.6, and hence any finitely generated projective left κ(x)-module is regarded as a finitely generated projective left κ(y)-module. This defines an exact functor from the exact category of finitely generated projective left κ(x)-modules to that for κ(y), and defines Km(κ(x))→Km(κ(y)).
8.8**.**
Let X be a non-commutative scheme over T satisfying (FT).
We say X is proper over T if the following conditions (i) and (ii) are satisfied.
(i) X is l.f.p. over T and quasi-compact.
(ii) Let C be a one-dimensional Noetherian integral scheme over T with function field K, let L be a finite-dimensional simple K-algebra, and assume that a morphism Spec(L)→X over T is given. Then the closure of the image of Spec(L)→C×TX which is endowed with the canonical structure of a prime non-commutative scheme (3.23) is isomorphic to Spec(B) over C for some coherent OC-algebra B.
If X is a scheme, this definition of properness coincides with the usual one. In fact, if X is proper over T in the usual sense, the above closure in C×TX is proper and quasi-finite over C, and hence finite over C. If the above condition of properness is satisfied, X is proper over T in the usual sense by the dvr criterion of properness for Noetherian schemes.
We denote by PT the category of non-commutative schemes over T which satisfy (FT) and which are proper over T.
For example, if S is a proper scheme over T and X=Spec(B) for a coherent OS-algebra on S, then X∈PT. If X is proper over T and Y→⊂X is an l.f.p. closed immersion, then Y is proper over T.
Lemma 8.9**.**
Let X be a non-commutative scheme over T satisfying (FT*). Assume that either one of the following (i) and (ii) is satisfied.*
(i) X is proper over T.
(ii) T is covered by open subschemes which are Spec of Jacobson rings. (Recall that a commutative ring is called a Jacobson ring if every prime ideal p is the intersection of all maximal ideals which contain p.)
Let x∈Xi. Then i is the largest integer such that there are points xj of X for 0≤j≤i with the following properties: xd=x, and for 0≤j<d, xj belongs to the closure of xj+1 in X and xj=xj+1.
Proof.
We may replace X by the closure of x with the prime non-commutative scheme structure. It is sufficient to prove that Xi−1 is not empty if i>0.
Since X is a Noetherian space, there is a closed point y of X. We have y∈Xj for some j≤i. We may assume that either j≤i−2 or j=i.
There is are open neighborhood U of y in X, an affine scheme S=Spec(R) over T of finite type, an R-algebra A such that R is the center of A, A is finitely generated as an R-module, and A is a prime ring, and an open immersion U→⊂Spec(A) over T.
First assume j≤i−2.
There is a non-zero element a∈R such that A[1/a] is an Azumaya algebra over R[1/a] and such that U⊗SSpec(R[1/a])=Spec(A[1/a]). Since dim(R)≥2, there is a prime ideal p of height one which does not contain a. Take a maximal two-sided ideal m of the non-zero ring A⊗RRp. Then by 4.6 (1), the inverse image of m in Rp is the unique maximal ideal of Rp. Let q be the inverse image of m in A. Since A⊗RRp=A[1/a]⊗RRp, q gives a point z of X which lies over p. We have z∈Xi−1 by 8.4.
Now assume i=j. Then x is a closed point of X. We prove x∈X0.
We first assume that the condition (ii) is satisfied. Since the image s of x in S is a closed point of S by 4.6 (1), the condition (ii) shows s∈S0 and hence x∈X0 by 8.4.
We assume the condition (i) is satisfied.
Let t be the image of x in T. If t∈Tk with k>0, let t′ be an element of Tk−1 which belongs to the closure T′ of t in T, endow T′ with a reduced scheme structure, and let C=Spec(OT′,t′). If t∈T0 and the center Z(κ(x)) of κ(x) is of transcendence degree >0 over κ(t), take a curve C over κ(t) whose function field is contained in Z(κ(x)). In both cases, let D be the closure of the image of x in C×TX. Since X is proper over T, D is isomorphic to Spec(B) for some coherent OC-algebra B on C. Take any closed point w of D and let z be the image of w in X. Then z belongs to the closure of x in X and z=x, a contradiction.
∎
We show that X↦CHi(X) is a covariant functor from PT to the category of abelian groups.
Lemma 8.10**.**
For a morphism X→Y of PT, the following diagram is commutative, where the vertical arrows are given by 8.7.
[TABLE]
Hence we have the induced homomorphism CHi(X)→CHi(Y).
Proof.
Let y∈Yi, u∈Xi+1, and consider the composite map
(1) K1(κ(u))→∂⊕xK0(κ(x))→K0(κ(y)),
where x ranges over all elements of X which lie over y and belong to the closure of u. Let v be the image of u in Y.
It is sufficient to prove the following Claim 1 and Claim 2.
Claim 1. Assume v∈Yi+1 and y belongs to the closure of v in Y. Then the composite map (1) coincides with the composite map K1(κ(u))→K1(κ(v))→∂K0(κ(y)).
Claim 2. Assume v=y. Then the composite map (1) is the zero map.
We prove Claim 1.
We may replace X (reso, Y) by the closure of u (resp. v) in X (resp. Y) with the prime non-commutative structure.
We may assume that there are an affine scheme S=Spec(R) over T of finite type, an R-algebra A such that A is finitely generated as an R-module and such that R is the center of A, an open neighborhood V of y in Y, and an open immersion V→Spec(A) over T. Let s be the image of y in S. Then OS,s is the center of OY,y. Let C=Spec(OS,s), Y′=Y×SC=Spec(OY,y), X′=X×SC=X×YSpec(OY,y).
Let D⊂C×TY and E⊂C×TX be the closures of the images of v and of u, respectively, endowed with the prime non-commutative scheme structures (3.23). By the properness of Y and of X over T, we have D=Spec(P) and E=Spec(Q) for some OS,s-algebras P and Q which are finitely generated as OS,s-modules. We have P⊂κ(v), Q⊂κ(u), and P⊂Q in κ(u).
Since the image of v in Y′ is dense in Y′ and the image of u is dense in X′, the morphisms Y′→C×TY and X′→C×TX induce morphisms Y′→D and X′→E, respectively.
Claim 3. We have P=OY,y in κ(v).
We prove Claim 3. Let z be the image of y∈Y′ in P. Then OD,z⊂κ(v). The morphisms Y′→D→Y induce homomorphisms OY,y→OD,z⊂OY,y whose composition is the identity morphism. Hence OD,z=OY,y in κ(v). We prove that P→OD,z is an isomorphism. The image of the center of P in OY,y is contained in the center of OY,y (1.9) which is OS,s, and since OS,s is contained in P, we have that the center of P is OS,s. This proves OD,z=P and hence Claim 3.
Claim 4. The morphism X′→E×DY′ is an isomorphism.
We prove Claim 4. The morphisms D→Y and E→X induce a morphism E×DY′→X×YY′=X′. The composition X′→E×DY′→X′ is the identity morphism. The morphism Mor(,E×DY′)→Mor(,X′) of functors on the category of non-commutative schemes is injective because
Mor(,E)⊂Mor(,C)×Mor(,X) and hence Mor(,E)⊂Mor(,Y′)×Mor(,X). This proves Claim 4.
We complete the proof of Claim 1. By Claim 4, the set of points of X lying over y is identified with the set of two-sided maximal ideals of Q. By the localization theory in K-theory, the commutative diagram of categories
[TABLE]
induces a commutative diagram
[TABLE]
Let n be the maximal two-sided ideal of P corresponding to y, and let Σ be the set of all maximal two-sided ideals of Q lying over n.
Then the two arrows in the composition (1) are identical with the two arrows in the part K1(κ(u))→⊕m∈ΣK0(Q/m)→K0(P/n) of this diagram, respectively, and the map K1(κ(v))→∂K0(κ(y)) is identical with the part K1(κ(v))→∂K0(P/n) of this diagram. Hence the commutativity of this diagram proves Claim 1.
We prove Claim 2. Let k be the center of κ(y).
We replace T by t=Spec(k), Y by y, and X by X×Yy.
Let C be the proper regular curve over T=Spec(k) whose function field is the center of κ(v). Let X′⊂C×TX be the closure of the image of u in C×TX endowed with the prime non-commutative scheme structure (3.23). Since X is proper over T, X′=Spec(B) for some coherent OC-algebra B on C, and X′ also belongs to PT.
Claim 2 for (X,Y,u,y) is reduced to Claim 2 for (X′,Y,u,y) and Claim 1 for (X′,X,u,x) for x∈X0. Claim 1 is already proved. Claim 2 for (X′,Y,u,y) is reduced to Claim 2 for (X′,T,u,t), and this is reduced to Claim 1 for (X′,C,u,w) where w is the generic point of C and Claim 2 for (C,T,w,t). This Claim 2 for (C,T,w,t) is just the usual theorem for C that the degree of a principal divisor is [math].
∎
8.11**.**
Now we take T=Spec(Z). Let X∈PZ.
We will relate CH0(X) with abelian coverings of X following the higher dimensional unramified class field theory in [12].
Let π1ab(X):=Hom(H1(Xet,Q/Z),Q/Z). In the case X is connected (then Xet is connected by 7.2), π1ab(X) is identified with the abelianization of the fundamental group of X in Section 7.
Define a quotient group π~1ab(X) of π1ab(X) as follows.
Let H~1(Xet,Q/Z) be the subgroup of H1(Xet,Q/Z) consisting of all elements which are killed by H1(Xet,Q/Z)→H1(Spec(Mr(R))et,Q/Z)=H1(Spec(R)et,Q/Z)=Z/2Z for every integer r and for every morphism Spec(Mr(R))→X.
Let π~1ab(X)=Hom(H~1(Xet,Q/Z),Q/Z). Thus π1ab(X)→π~1ab(X) is surjective and its kernel is killed by 2.
8.12**.**
Let X∈PZ. For x∈X0, we have κ(x)≅Mr(x)(Fq) for some r(x)≥1 and for a finite field Fq of q elements.
The inclusion morphism x=Spec(κ(x))→X induces π1ab(x)→π1ab(X). Let φx∈π1ab(X) (called the Frobenius at x) be the image of the canonical generator of π1(x)≅π1(Fq)=Gal(Fˉq/Fq), the q-th power map Fˉq→Fˉq;x↦xq.
The image of φx in π~1ab(X) is also denoted by φx.
Theorem 8.13**.**
Let X∈PZ.
(1) There is a unique homomorphism
[TABLE]
which sends the class of 1∈Z at x∈X0 to φxr(x). Here r(x) and φx are as in 8.12.
(2) For a morphism X→Y in PZ, we have a commutative diagram
[TABLE]
The proof is given below.
Lemma 8.14**.**
For a morphism X→Y in PZ, we have a commutative diagram
[TABLE]
Proof.
Let x∈X0, y=f(x)∈Y0. Then κ(y)=Mr(Fpa) and κ(x)=Mr(Ms(Fpab))=Mrs(Fpab) for some r,s,a,b≥1. The map K0(κ(x))→K0(κ(y)) sends the generator [Fpabrs]∈K0(κ(x)) to bs times the generator [Fpar]∈K0(κ(y)). On the other hand, the homomorphism π1ab(X)→π1ab(Y) sends φx to φyb and hence sends φxrs to φyrsb=(φyr)bs.
∎
8.15**.**
We prove 8.13.
By 8.14, for the proof of 8.13, it is sufficient to prove that the composite map K1(κ(u))→⊕xZ→π~1ab(X) is the zero map for X∈PZ and for u∈X1, where x ranges over all points in X0 which belong to the closure of u in X. This is reduced to the classical reciprocity law of class field theory as follows.
Let F be the center of κ(u). If F is of positive characteristic, let
C be the proper smooth curve over a finite field with function field F. If F is of characteristic [math], let C=Spec(OF), where OF is the integer ring of the number field F. Let X′ be closure of the image of
u→C×Spec(Z)X endowed with the structure of a prime non-commutative scheme. Then X′∈PZ, u∈X1′, and by 8.10 and 8.14, the above map K1(κ(u))→π~1ab(X) is the composition K1(κ(u))→π~1ab(X′)→π~1ab(X). Hence we may assume X=X′.
Then π1ab(X)→≅π1ab(C) by 6.18 (2). We have a commutative diagram
[TABLE]
Hence it is sufficient to prove that the composition K1(κ(u))→K1(F)→π1ab(C)≅π1ab(X)→π~1ab(X) is the zero map.
Let Σ1 be the set of real places of F such that Fv⊗Fκ(u)≅Mr(R) for some r. Then π~1ab(X) is regarded as the quotient of π1ab(C) by the images of Gal(Fˉv/Fv) for v∈Σ1. Let Σ2 be the complement of Σ1 in the set of real places of F. By the reciprocity law of the class field theory of F, elements of K1(F)=F× which are totally positive at Σ2 are killed in π~1ab(X). All elements of the image of K1(κ(u))→F× are totally positive at Σ2 because Fv⊗Fκ(u)≅Mr(H) for some r if v∈Σ2.
9 Appendix: Zeta and L-functions of non-commutative schemes. By Takako Fukaya and Kazuya Kato
In the paper [5], the first author of this Appendix defined the Hasse zeta function of a finitely generated non-commutative ring over Z. In this Appendix, we generalize
it to the zeta function of a con-commutative scheme. L-functions for non-commutative rings were not considered in [5]. We discuss here L-functions
for non-commutative schemes using Sections 5, 7, 8 of this paper.
We also give a
partial result on the relation of zeta functions of non-commutative schemes over a finite field and étale cohomology (Thm. 9.23).
We obtain it by extending the definition of Rf! slightly to f which need not be r.c.
9.1**.**
Let X be a quasi-compact non-commutative scheme locally of finite presentation over Z.
Let X0 be the set of all points of X such that OX,x/p(x) is finite as a set. For x∈X0, OX,x/p(x)≅Mr(Fq) for some r≥1 and for some finite field Fq of q-elements. Let N(x)=q for these x and q.
We define the zeta function ζ(X,s) of X as
[TABLE]
If X is a scheme, this coincides with the Hasse zeta function of X. If X=Spec(A) for a (not necessarily commutative) ring A, this coincides with the zeta function ζA(s) defined in [5].
9.2**.**
It can happen that ζ(X,s) diverges for all s∈C. For example, this happens in the case X=Spec(Fp⟨T1,T2⟩)
with Fp⟨T1,T2⟩ the non-commutative polynomial ring in two variables over Fp. If R is a finitely generated commutative ring over Z and A is an R-algebra, ζA(s) absolutely converges when Re(s)≫0 for example, if A is finitely generated as an R-module (see 9.9), or if A is the group ring R[G] for a group G which has a finitely generated nilpotent subgroup of finite index ([6]).
Remark 9.3**.**
After the author of [5] studied zeta functions of non-commutative rings, she wondered whether we can have L-functions for non-commutative rings. For A=Z and B=Z[i], we have the well-known decomposition ζB(s)=ζA(s)L(s,χ), where L(s,χ) is the unique Diriclet character of conductor 4.
For a homomorphism A→B of rings which need not be commutative such that B is, say, free of rank 2 as a left A-module, the hope was to decompose ζB(s) similarly into the product of ζA(s) and some L-function of A. The issue as in the following examples (1) and (2) soon appeared.
(1) A=Fp×Fp embedded in B=M2(Fp) as the diagonal.
(2) A=Fp2 embedded in B=M2(Fp) as a subring.
In these (1) and (2), ζB(s)=(1−p−s)−1. In the case (1), ζA(s)=(1−p−s)−2 , and in the case (2), ζA(s)=(1−p−2s)−1=(1−p−s)−1(1+p−s)−1. Thus in both cases, it is impossible to decompose ζB(s) as the product of ζA(s) and an L-function of A.
Now we see that B is not an A-algebra in these examples, and we do not have a morphism Spec(B)→Spec(A) of non-commutative schemes.
The following Claim shows that a bad thing like in the above (2) does not happen for a morphism of non-commutative schemes.
Claim Let f:X→Y be a morphism between quasi-compact non-commutative schemes locally of finite presentation over Z. Let x∈X0 and y=f(x)∈Y. Then y∈Y0 and N(x)=N(y)n for some integer n≥1.
Proof of Claim. The fact y∈Y0 is clear. Assume OY,y/p(y)≅Mr(Fq). Since OX,x/p(x) is a OY,y/p(y)-algebra, we have by 1.15 that OX,x/p(x)≅Mr(B) for an Fq-algebra B. Since OX,x/p(x) is simple, B must be simple and hence B≅Ms(Fqn) for some s,n≥1. Hence OX,x/p(x)≅Mrs(Fqn).
Our hope is that for a morphism X→Y as above,
ζ(X,s) would be expressed by using L-functions of Y, and more generally, L-functions of X would be expressed by using L-functions of Y. We study this question.
9.4**.**
Assume Xet is connected (7.1). (If X satisfies (F), this is the same as X is connected, by 7.2). Let x∈X0. Then we have the Frobenius
[TABLE]
defined modulo conjugacy as follows. Assume OX,x/p(x)≅Mr(Fq). The morphism Spec(OX,x/p(x))→X induces a homomorphism π1(Fq)=π1(Mr(Fq))→π1(X) of pro-finite groups modulo conjugacy (7.3). We define φx as the image of the canonical generator Fˉq→Fˉq;x↦xq of π1(Fq)=Gal(Fˉq/Fq).
If X→Y is a morphism between quasi-compact non-commutative schemes locally of finite presentation over Z which sends x∈X0 to y∈Y0, π1(X)→π1(Y) sends
φx to φyf where f is the integer such that N(x)=N(y)f.
9.5**.**
The following questions arise, but the authors do not know the answers.
Question. Is there a Chebotarev density theorem for non-commutative schemes about the distribution of φx?
Fix an isomorphism of commutative fields Qˉℓ≅C.
Question. Assume X is over a finite field k, X is irreducible, ℓ is not the characteristic of k, and let ρ:π1(X)→GLn(Qˉℓ) be a continuous irreducible representation. Then is ρ pure? That is, is there an integer w such that the complex eigenvalues of ρ(φx) are N(x)w/2 for all x∈X0?
For a finite-dimensional continuous representation ρ of π1(X) over Qˉℓ, we define the L-function L(X,ρ,s) of ρ as
[TABLE]
If ρ is the trivial one-dimensional representation, we have L(X,ρ,s)=ζ(X,s).
If ζ(X,s) absolutely converges when Re(s)≫0 and if ρ has bounded weights (this means that there are a,b∈R such that for every x∈X0, all eigen values α∈C of ρ(φx) satisfy N(x)a≤∣α∣≤N(x)b), then L(X,ρ,s) absolutely converges in C when Re(s)≫0.
9.6**.**
In the following 9.7, let X→Y be a morphism of quasi-compact non-commutative schemes which are l.p.f over Z. Assume that there is a covering Y′→Y in Yet such that X×YY′ is isomorphic over Y′ to a disjoint union of m copies of Y′ for an integer m≥1. Assume that Xet and Yet are connected (7.1) and that X0 is non-empty, so π1(X) and π1(Y) are defined.
Then the homomorphism π1(X)→π1(Y) is injective, the image of it is an open subgroup of π1(Y) of index m, and the sheaf on Yet associated the presheaf MorY(,X) corresponds to the π1(Y)-set π1(Y)/π1(X).
Let ℓ be a prime number which is invertible on Y, and let ρ:π1(X)→GLn(Qˉℓ) be a continuous representation.
Fixing an isomorphism Qˉℓ≅C of commutative fields, we assume that ρ has bounded weights and ζ(X,s) absolutely converges when Re(s)≫0, so, L(X,ρ,s) absolutely converges in C when Re(s)≫0.
Let ρ~:π1(Y)→GLmn(Qˉℓ) be the induced representation of ρ.
Proposition 9.7**.**
Let the notation be as in 9.6. Then L(Y,ρ~,s) absolutely converges when Re(s)≫0, and we have
[TABLE]
Proof.
For y∈Y0, let X0(y) be the inverse image of y in X0. For y∈Y0 and for x∈X0(y), let ρx be the composition π1(x)→π1(X)→GLn(Qˉℓ) and let
ρ~y be the composition π1(y)→π1(X)→GLmn(Qˉℓ). We have L(X,ρ,s)=∏y∈Y0,x∈X0(y)L(x,ρx,s) and L(Y,ρ~,s)=∏y∈Y0L(y,ρ~y,s), and hence it is sufficient to prove L(y,ρ~y,s)=∏x∈X0(y)L(x,ρx,s) for each y∈Y0.
We have y≅Spec(Mr(Fy)) for a finite field Fx of order N(y) and for some r≥1. For each x∈X0(y), x≅Spec(Ms(Fx)) for a finite field Fx of order N(x) and for some s≥1, the homomorphism π1(x)=Gal(Fˉx/Fx)→π1(y)=Gal(Fˉy/Fy) is injective, and the π1(y)-set π1(Y)/π1(X) is identified with the disjoint union of the π1(y)-sets π1(y)/π1(x) for x∈X0(y). From this, we see that ρ~y is the direct sum of the representations ρ~x of π1(y) induced from ρx for x∈X0(y). Hence L(y,ρ~y)=∏x∈X0(y)L(y,ρ~x,s)=∏x∈X0(y)L(x,ρx,s).
∎
9.8**.**
Fix an isomorphism of commutative fields Qˉℓ≅C. Let X be a quasi-compact non-commutative scheme satisfying (FZ), let ℓ be a prime number which is invertible on X, and let F be a mixed Qˉℓ-sheaf on X.
We define the L-function of F by
[TABLE]
This absolutely converges in C when Re(s)≫0. If X is connected and F is smooth, it is the L-function of the corresponding representation of π1(X) considered in
9.5.
For an exact sequence 0→F′→F→F′′→0 of mixed Qˉℓ-sheaves, we have L(X,F,s)=L(X,F′,s)L(X,F′′,s).
Proposition 9.9**.**
Let X be a quasi-compact non-commutative scheme satisfying (FZ*), let ℓ be a prime number which is invertible on X, and let F be a mixed Qˉℓ-sheaf on X. Then L(X,F,s)=∏i=1nL(Si,Fi,s)m(i) for some n≥0, for some schemes Si of finite type over Z on which ℓ is invertible, for some mixed Qˉℓ-sheaves on Si, and for some m(i)∈Z. If X is over a finite field k, we can take as Si schemes of finite type over k and hence L(X,F,s) is a rational function in q−s.*
By this and by Noetherian induction, we may assume that X is a prime non-commutative scheme and the statement of 9.9 is true if F has supports in a closed subset C=X of X.
For open subsets U,U′ of X such that X=U∪U′, we have
L(X,F,s)=L(U,F,s)L(U′,F,s)L(U∩U′,s)−1.
By this, we may assume that there are a prime ring A such that Z(A) is a finitely generated Z-algebra and A is finitely generated as a Z(A)-module and an open immersion X→⊂Spec(A). There is a non-zero element a of Z(A) such that A[1/a] is an Azumaya algebra over Z(A)[1/a] and U:=XSpec(Z(A)Spec(Z(A)[1/a]) coincides with Spec(A[1/a]). Let V=Spec(Z(A)[1/a]) and let G be the mixed Qˉℓ-sheaf on V corresponding to F via 5.6. Let Y=X×Spec(Z(A)Spec(Z(A)/aZ(A)). Then
L(X,F,s)=L(U,F,s)L(X,FY,s)=L(V,G,s)L(X,FY,s),
and 9.9 is true for L(X,FY,s) by Noetherian induction.
∎
Proposition 9.10**.**
Let X and Y be quasi-compact non-commutative schemes satisfying (FZ*) and let X→Y be a morphism
satisfying the condition 5.28.1. Let ℓ be a prime number which is invertible on Y and let F be a mixed Qˉℓ-sheaf on X. Then we have*
[TABLE]
Proof.
We are reduced to the case X=Y×ST with T→S a proper morphism of schemes. By 5.29, we have L(Y,Rf!F,s)=∏y∈Y0L(y,Rf!y(F∣X(y)),s) where X(y)=X×Yy, fy:X(y)→y, and F∣X(y) is the inverse image of F on X(y). Hence we are reduced to the case Y=y=Spec(Mr(Fq)) for r≥1 and for a finite field Fq. The morphism Y=Spec(Mr(Fq))→S factors as Y→Spec(Fq)→S. Hence by 5.6, we are reduced to the case Y=Spec(Fq)=S and X is a proper scheme over Fq, and hence we are reduced to the classical formula of Grothendieck on L-functions of Qˉℓ-sheaves.
∎
The L-function of an abelian character of the Galois group of a number field is understood by class field theory.
The following 9.12 is a version of it.
Lemma 9.11**.**
Let X be a quasi-compact non-commutative scheme satisfying (FZ*). Then r(x) for x∈X0 are bounded.*
Proof.
We may assume that X=Spec(A) for a finitely generated commutative ring R over Z and for an R-algebra A which is finitely generated as an R-module.
Then for some n≥1, A is generated by n elements as an R-module. Let m∈max(A) and let m′ be the image of m under max(A)→max(R). Then
A/m≅Mr(x)(k) is generated by n elements as a module over k′=R/m′, where k and k′ are finite fields such that k⊃k′. Hence r(x)2≤n.
∎
Proposition 9.12**.**
Let X be an object of PZ (8.8) and let π~1ab(X) as in 8.11.
Let ρ be a continuous homomorphism π~1ab(X)→Qˉℓ× of bounded weights. Let m=l.c.m of r(x) for x∈X0 (9.11).
Assume that ρ=(ρ′)m for some
continous homomorphism ρ′:π~1ab(X)→Qˉℓ×.
Then there is a homomorphism
χ:CH0(X)→Qˉℓ×
such that
[TABLE]
Proof.
Let χ be the composite map CH0(X)→π~1ab(X)→ρ′Qˉℓ× (8.13). Then ρ(φx)−1=ρ′(φxm)−1=χ(x)−m/r(x).
∎
Now we develop a theory of Rf! which is different form that in Section 5 and which works for some non-r.c. morphisms f.
Proposition 9.13**.**
Let S be an excellent scheme, let X be a non-commutative scheme, let f:X→S be a morphism, and assume that for some coherent OS-algebra B on S, X is isomorphic over S to an open subspace of Spec(B). Then the following conditions (i)–(iii) are equivalent.
(i) For every morphism S′→S from an excellent scheme S′, the map X×SS′→S′ is a homeomorphism.
(ii) For every étale morphism S′→S, the map X×SS′→S′ is a homeomorphism.
(iii) For every s∈S, if OS,sˉ denote the strict henselization of the local ring OS,s, there are an OS,sˉ-algebra A which is finitely generated as an OS,sˉ-module and a closed point xˉ of Spec(A) such that X×SSpec(OS,sˉ) is isomorphic to U(xˉ) (4.9) over Spec(OS,sˉ).
Proof.
The implication (i) ⇒ (ii) is clear.
(ii) ⇒ (iii). Let S′=Spec(OS,sˉ). By the limit argument, we see that the map X′:=X×SS′→S′ is a homeomorphism. Let xˉ be the unique point of X′ lying over the closed point sˉ of S′. Let A be the stalk Bsˉ of B at sˉ. Then X′ is an open set of Spec(A) containing xˉ. Since S′ is the smallest neighborhood of sˉ in S′, X must be the smallest neighborhood U(xˉ) of x in Spec(A).
(iii) ⇒ (i). For any s∈S, there is only one point of X×SSpec(OS,sˉ) lying over the closed point sˉ of Spec(OS,sˉ). This shows that for every morphism S′→S from an excellent scheme S′, the map X′:=X×SS′→S′ is bijective. We prove that this is a homeomorphism.
Let x′,y′∈X′, let s′,t′∈S′ be the images of x′,y′ in S′, respectively, and assume t′ converges to s′. Since X′ is locally a Noetherian space,
it is sufficient to prove that y′ converges to x′. Let s be the image of s′ in S, let OS,sˉ be the strict henselization of OS,s, and let OS′,sˉ′ be the strict henselization of OS′,s′. We have a homomorphism OS,sˉ→OS′,sˉ′ over OS,s which is compatible with S′→S. Since X×SSpec(OS,sˉ) is D(e) in Spec(Bsˉ) for an idempotent e for (Bsˉ,m) where m is a maximal ideal of Bsˉ, X′×S′Spec(OS′,s′) is D(e) in Spec(Bsˉ⊗OS,sˉOS′,sˉ′), and hence by 4.8 (3), all points of X′×S′Spec(OS′,s′) converges to the unique closed point xˉ′ of Spec(Bsˉ⊗OS,sˉOS′,sˉ′). Since Spec(OS′,sˉ′)→Spec(OS′,s′) is surjective, there is a point u of Spec(OS′,sˉ′) whose image in Spec(OS′,s′) is t′. Let z be the point of X′×Spec(OS′,s′)Spec(OS′,sˉ′) whose image in Spec(OS′,sˉ′) is u. Then the image of z in X′ is y′ because its image in S′ is t′ and X′→S′ is bijective. Since z converges to xˉ′, the image y′ of z converges to the image x′ of xˉ′.
∎
9.14**.**
We will denote the equivalent conditions in 9.13 by (H1).
Proposition 9.15**.**
Let f:X→S be a morphism as in 9.13 satisfying the condition (H1).
(1) The category of sheaves on Xet and the category of sheaves on Set are equivalence by f∗ and its inverse f−1.
(2) Let x∈X and s=f(x)∈S, and let F be a sheaf on Xet. Then we have (f∗F)sˉ→≅Fxˉ.
(3) Over C: Xcl→Scl is a homeomorphism.
Proof.
(2) follows from the condition (iii) in 9.13 by 5.15 (2). (1) follows from (2) by 5.14.
(3) follows from the condition (iii) in 9.13 by (5) of 6.4.
∎
9.16**.**
The condition (H2) for a morphism f from a non-commutative scheme X to an excellent scheme S:
(H2) Étale locally on S, there are an open covering (Ui)i∈I of X, an open covering (Vi)i∈I of S, and a morphism fi:Ui→Vi which is compatible with f and which satisfies (H1) for each i∈I.
If f satisfies (H2), X×SS′→S′ satisfies (H2) for every excellent scheme S′ and a morphism S′→S.
Proposition 9.17**.**
Let f:X→S be a morphism as in 9.16 satisfying (H2).
(1) The inverse image functor f−1 from the category of sheaves of abelian groups on Set to the category of sheaves of abelian groups on Xet has a left adjoint functor f!.
(2) For s∈S, we have an isomorphism ⊕xˉFxˉ→≅(f!F)sˉ, where xˉ ranges over all points of X×Ssˉ.
Here the map in (2) is defined as follows. By (1) of 9.17, there is a canonical map F→f−1f!F corresponding to the identity map f!F→f!F. Hence for each xˉ, an element of Fxˉ gives an element of (f−1f!F)xˉ=(f!F)sˉ. This gives the canonical map in (2).
Proof.
Work étale locally on S, let Ui and Vi be as in 9.16 and let Vij be the image of Uij:=Ui∩Uj. Then Uij→Vij also satisfies (H1). Denote the morphisms Ui→X, Uij→X, Vi→Y, Vij→Y, Ui→Vi, and Uij→Vij by ai, aij, bi, bij, fi, and fij, respectively. For a sheaf F of abelian groups on Xet, define f!F to be the cokernel of ⊕i,jbij,!fij,∗aij−1F→⊕kbk,!fk,∗ak−1F. Here the map from the (i,j)-component to the k-component is the inclusion map if k=i=j, the minus of the
inclusion map if k=j=i, and is the zero map otherwise. Since fi and fij satisfy (H1),
[TABLE]
[TABLE]
where the second = follows from 9.15 (1). This proves (1).
(2) follows from the definition of f!F by 9.15 (2).
∎
This f! is compatible with base changes by a morphism S′→S of excellent schemes.
Proposition 9.18**.**
Let S be an excellent scheme of dimension 1 such that the strict henselization OS,sˉ of the local ring OS,s is an integral domain for every s∈S. Let X=Spec(B) for a coherent OS-algebra B on S such that OS→≅Z(B). Then the morphism X→S satisfies (H2).
Proof.
Working étale locally on S, we may assume that for every x∈X with the image s in S, the center of OX,x/p(x) is a purely inseparable extension of the residue field κ(s) of s. Then for some dense open set V of S, B∣V
is an Azumaya algebra over OV. Let U be the inverse image of V in X. For each x∈X, let Ux:=U∪{x} and let Vx=V∪{s} where s is the image of x in S. Then (Ux)x∈X is an open covering of X, (Vx)x∈X is an open covering of S, and for any étale morphism S′→Vx from an excellent scheme S′, Ux×VxS′→S′ is a homeomorphism, that is, Ux→Vx satisfies (H1).
∎
9.19**.**
The condition (H3) on a morphism f:X→S from a non-commutative scheme X to an excellent scheme S.
(H3) There is a factorization X→f1Y→f2T→f3S of f such that
T is a scheme and f3 is compactifiable, that is, via f3, T is isomorphic to an open subscheme of a proper scheme over S,
f2 satisfies (H2), and
f1 is an l.f.p. immersion.
Such decomposition f=f3∘f2∘f1 will be called a good factorization of f.
For f satisfying (H3), for a sheaf of torsion abelian groups F on Xet. we define Rf!F on Set relative to a good factorization of f as above by
[TABLE]
The authors can not prove that this Rf!F is independent of the choice of a good factorization of f. The issue is that for another good factorization X→Y′→T′→S of f, we do not have the diagonal embedding X→Y×SY′ because Y×SY′ is not defined (though Y⊗SY′ is defined) and hence it is hard to compare two factorizations. The comparison with the classical topology in the case over C in 9.22 below suggests that Rf!F may be independent of the choice of a good factorization.
We give three remarks.
(1) Assume f:X→S satisfies (H3) and let g:X′→X be a morphism satisfying 5.28.1 which is expressed as X′→X×ST′→X where T′ is a proper scheme over S and the first arrow is an l.f.p. immersion. Then f′:=f∘g:X′→S satisfies (H3). In fact, we have a good factorization f′=f3′∘f2′∘f1′ of f′ where f1′ is the composition X′→X×ST′→Y×ST′, f2′ is the morphism Y×ST′→T×ST′, and f3′ is the morphism T×ST′→S, which are induced by f1,f2,f3, respectively. We have Rf!′=Rf!Rg! where Rf! and Rf!′ are defined with respect to the good factorizations f=f3∘f2∘f1 and f′=f3′∘f2′∘f1′. respectively.
(2) If X→S satisfies (H3), then f′=X×SS′→S′ satisfies (H3) for every excellent scheme S′ with a morphism S′→S. If we define Rf! and Rf!′ by using a good factorization of f and the induced good factorization of f′, then they commute with the pullbacks of torsion sheaves of abelian groups.
(3) In the case S=Spec(k) for a separably closed commutative field k, if f satisfies (H3), Rmf!F (defined by fixing a good factorization of f) is denoted by Hcm(Xet,F). This will be used in 9.23 to have cohomological expressions of L-functions.
Proposition 9.20**.**
Let f:X→S be a morphism from a non-commutative scheme to an excellent scheme satisfying (H3). Fix a good factorization of f to define Rf!. Let ℓ be a prime umber which is invertible on S.
(1) Rmf! sends constructible sheaves of abelian groups on Xet to constructible sheaves of abelian groups on Set. If m≫0, Rmf!F=0 for all sheaves of torsion abelian groups on Xet. It sends constructible Qˉℓ-sheaves on X to constructible Qˉℓ-sheaves on S.
(2) If S is of finite type over Z, Rmf! sends mixed Qˉℓ-sheaves on X to mixed Qˉℓ-sheaves. For a mixed Qˉℓ-sheaf on X, we have L(X,F,s)=L(S,Rf!F,s).
Proof.
This is reduced to the cases f=f1, f=f2, and f=f3. The case f=f1 is easy and the case f=f3 is the usual theory for schemes. In the case f=f2, (1) is clear and (2) follows from 9.17 (2). (The statement in (1) about Qˉℓ-sheaves follows also from 5.31.)
∎
9.21**.**
In the following, when we say locally compact, the Hausdorff condition is already included.
To compare the above Rf! with the classical topology over C, we slightly generalize Rf! in Verdier [18] for locally compact spaces to non-Hausdorff spaces.
If X is a non-commutative scheme over C satisfying (H3), then by 9.15 (3), each point of Xcl has a locally compact open neighborhood. However, Xcl itself need not be Hausdorff.
Let X,Y be topological spaces and let f:X→Y be a continuous map. Assume Y is locally compact and X has a finite covering by locally compact open sets (but X itself need not be Hausdorff). We define Rf!F for a sheaf F of abelian groups on X. Write
X=∪i∈IUi with I a totally ordered finite set and with Ui a locally compact open set of X. Let
F→I∙ be an injective resolution of F.
Then Rf!F is defined as the simple complex associated to the double complex (Gs,t)s,t defined as follows. Gs,t=0 if s>0 or if t<0. Assume s≤0 and t≥0. For a subset J of I, let UJ=∩i∈JUi, fJ:UJ→Y the induced map. Then Gs,t=⊕J(fJ)!(It∣Uj) where J ranges over all subsets of I of order 1−s and (fJ)! is the part of compact supports of (fJ)∗. The differential Gs,t→Gs,t+1 is induced by d:It→It+1, and the differential Gs,t→Gs+1,t is defined in the following way. For a subset J of I of order 1−s and a subset K of order −s, the part (fJ)!(It)→(fK)!(It) is [math] if K is not contained in J and (−1)r times the canonical map if J={a1,…,a1−s} (a1<⋯<a1−s) and K={a1,…,ar−1,ar+1,…,a1−s}.
If X is locally compact, this Rf!F coincides with that considered in [18].
In the case Y is a one point set, the unique stalk of Rmf!F will be denoted by Hcm(X,F).
Proposition 9.22**.**
Let f:X→S be as in 9.19 satisfying (H3), and assume that S is of finite type over C and separated. Let F be either a constructible sheaf of abelian groups on Xet or a constructible Qˉℓ-sheaf on X, and let Fcl be the pullback of F on Xcl. Fix a good factorization of f. Then we have
By the classical comparison theorem of Artin ([1] vol. 3, Exp. XVI, 4), we are reduced to the case f satisfies (H2), and this case follows from 9.15 (3). ∎
Theorem 9.23**.**
Let X be a non-commutative scheme over a finite field k=Fq such that the morphism X→Spec(k)
satisfying (H3). Fix a good factorization of this morphism.
Let ℓ be a prime number which is different from the characteristic of k and let F be a mixed Qˉℓ-sheaf on X. Then
[TABLE]
Here φq denotes the action of the generator kˉ→kˉ;x↦xq of Gal(kˉ/k).
Example. Let the group G=⟨α,β∣α2=1,αβα−1=β−1⟩ be as
in 6.19.
Then we have
[TABLE]
for every finite field Fq of characteristic =2. On the other hand,
[TABLE]
Thus like in Weil conjectures, the zeta function of a space Spec(Fq[G]) over a finite field and the cohomology of a space Spec(C[G])cl over C
have similar shapes, and this is explained as follows.
Let R be a commutative ring in which 2 is invertible. The center of R[G] is R[T] where T=β+β−1.
The morphism π:Spec(R[G])→Spec(R[T]) satisfies (H2). In fact, for 1≤i≤4, let
Ui=Spec(R[G])−V(Ii) where
[TABLE]
and let Vi be the open set D(T−2) (resp. D(T+2)) of Spec(R[T]) for i=1,2 (resp. i=3,4).
Then Spec(R[G])=∪i=14Ui, and Ui→Vi satisfies (H1).
By using this, for Λ=Z/nZ with an integer n which is invertible in R, we have an exact sequence 0→F→π!(Λ)→Λ→0 on Spec(R[T]), where F restricted to D(T−2)∩D(T+2) in Spec(R[T]) is [math] and the pullback of F to V(T−2) and that to V(T+2) in Spec(R[T]) are isomorphic to Λ. (The fact 2 is invertible in R is used for the fact V(T−2) and V(T+2) in Spec(R[T]) are disjoint.) From this we see that f:Spec(R[G])→Spec(R) satisfies (for the étale topology)
R0f!Λ=Λ2, R2f!Λ=Λ(−1), Rmf!Λ=0 for m=0,2. Via 9.23 and 9.22, this relates the zeta function of Spec(Fq[G]) and the compact support cohomology of Spec(C[G]).
9.25**.**
For a group G which has a finitely generated nilpotent subgroup of finite index, it is proved in [6] that the zeta function ζA(s) of the group ring A=Fq[G]
converges when Re(s)≫0. We expect that this zeta function is expressed by some compact support étale cohomology theory, but the method of this paper is limited to the case A is a finitely generated module over its center. For example, if G is the Heizenberg group (the group of unipotent upper triangular (3,3)-matrices with entries in Z), A is not a finitely generated module over its center, and we ask whether its zeta function ζA(s)=(1−q−s)(1−qs−1)−3(1−qs−2)3(1−qs−3)−1 can be explained by some compact support etale cohomology theory.
Bibliography18
The reference list from the paper itself. Each links out to its DOI / PubMed record.
1[1] M. Artin, A. Grothendieck, J.-L. Verdier , Séminaire de Géométrie Algébrique du Bois Marie – 1963–64 – Théorie des topos et cohomologie étale des schémas (SGA 4) , vol. 1. Lecture Notes in Math. 269 (1972), vol. 2. Lecture Notes in Math. 270 (1972), vol. 3. Lecture Notes in Math. 305 (1972), Springer.
2[2] N. Bourbaki , Elements of mathematics, Commutative algebra , Translated from the French. Hermann, Paris; Addison-Wesley Publishing Co., Reading, Mass. (1972).
3[3] P. Deligne , La conjecture de Weil, II , IHES. Publ. Math. 43 (1980), 157–252.
4[4] P. Deligne , Séminaire de Géométrie Algébrique du Bois Marie - Cohomologie étale (SGA 4 1 2 1 2 \frac{1}{2} ) , Lecture notes in Math. 569 , Springer (1977).
5[5] T. Fukaya , Hasse zeta functions of non-commutative rings , J. of Algebra, 208 (1998), 304–342.
6[6] T. Fukaya , On Hasse zeta functions of group algebras of almost nilpotent groups , Kodai Math. J. 22 (1999), 140–151.
7[7] A. Grothendieck, J. Dieudonné , Éléments de géométrie algébrique: IV. Étude locale des schémas et des morphismes de schémas, Troisième partie , Publ. Math. IHES. 28 (1966) 5–255.
8[8] A. Grothendieck , Séminaire de Géométrie Algébrique du Bois Marie – 1960–61 – Rev e ^ ^ 𝑒 \hat{e} tements étales et groupe fondamental (SGA 1) , Lecture Notes in Math. 224 , Springer (1971).