In sections 4 to 8, we denote by k a commutative ring, and by A an associative unital k-algebra. The role of k is auxiliary; we mainly have in view the following three special cases:
a) we consider an arbitrary ring A, we set k=Z and we endow A with its natural structure of Z-algebra,
b) we consider an arbitrary ring A, we take for k the center of A,
c) we consider a commutative ring A and we take k=A.
1. Tensor product of two complexes
Let (C,d) be a complex of right A-modules and (C′,d′) a complex of left A-modules.
Let us endow the k-module C⊗AC′ with the grading such that
(C⊗AC′)n=p+q=n∑(Cp⊗Cq′)
and let D be the unique k-linear endomorphism of degree (−1) of C⊗AC′ such that
The complex of k-modules (C⊗AC′,D) is called the tensor product complex of the complexes (C,d) and (C′,d′).
Remark 1
When C′ is reduced to C0′=M, then (C⊗AC′)n=Cn⊗AM and D=d⊗1M; for example C⊗AAs is naturally identified with C. Similarly, when C is reduced to C0=P, then (C⊗AC′)n=P⊗ACn′ and D=1P⊗d.
Remark 2
For every integer r, we have (C⊗AC′)(r)=C(r)⊗AC′, but (C⊗AC′)(r) and C⊗AC′(r) do not in general have the same differential.
Let p,q be two integers, x∈Zp(C),x′∈Zq(C′); then the element x⊗x′ of Cp⊗Cq′ belongs to Zp+q(C⊗C′) by (1); moreover, if y∈Cp+1,y′∈Cq+1′, we have
(x+dy)⊗(x′+d′y′)=x⊗x′+D(y⊗x′+(−1)p(x+dy)⊗y′);
by passing to the quotients, we deduce a k-linear mapping, called canonical
γp,q(C,C′):Hp(C)⊗AHq(C′)→Hp+q(C⊗AC′);
if we endow H(C)⊗AH(C′) with the grading such that
(H(C)⊗H(C′))n=p+q=n∑Hp(C)⊗AHq(C′),
the γp,q define a graded k-linear mapping of degree 0
γ(C,C′):H(C)⊗AH(C′)→H(C⊗AC′).
Prop. 6 of II, p. 59, can be reformulated as follows:
Proposition 1
If the complexes C and C′ are zero on the right, C⊗AC′ is zero on the right and the canonical k-linear mapping*
γ0,0(C,C′):H0(C)⊗AH0(C′)→H0(C⊗AC′)
is bijective.
Let u:(C,d)→(C1,d1) be a morphism of complexes of right A-modules and u′:(C′,d′)→(C1′,d1′) a morphism of complexes of left A-modules; then u⊗u′:C⊗AC′→C1⊗AC1′ is a morphism of complexes of k-modules; indeed, it is graded of degree 0, and if we denote by D and D1 the differentials of C⊗C′ and C1⊗C1′, we have for p,q∈Z,x∈Cp,x′∈Cq′,
Let A∘ be the k-algebra opposite to A, C∘ (resp. C′∘) the complex C (resp. C′) considered as a complex of left A∘-modules (resp. right). Let us denote by
σ(C,C′):C⊗AC′→C′∘⊗A∘C∘
the unique graded k-linear mapping of degree 0 such that, for x∈Cp,x′∈Cq′,p,q∈Z, we have
σ(C,C′)(x⊗x′)=(−1)pqx′⊗x.
Proposition 2
The mapping σ(C,C′):C⊗AC′→C′∘⊗AC∘ is an isomorphism of complexes of k-modules, whose reciprocal isomorphism is σ(C′∘,C∘).
Since the mappings σ(C,C′) and σ(C′∘,C∘) are reciprocal to each other, it suffices to prove that σ(C,C′) is a morphism of complexes. Now, for
x∈Cp=Cp∘,x′∈Cq′=Cq′∘,p,q∈Z,
we have, denoting by D′ the differential of C′⊗A∘C,
Suppose for the end of this number that the ring A is commutative (cf. No. 9 for the general case).
Let C,C′,C′′ be three complexes of A-modules; the canonical homomorphism of A-modules (III, p. 64)
φ:(C⊗AC′)⊗AC′′→C⊗A(C′⊗AC′′)
is an isomorphism of complexes, as is immediately verified using the definitions.
More generally, let (C(i),d(i))i∈I be a family of complexes of A-modules, where the set I is finite and totally ordered; we shall identify I with the interval [1,r] of N to simplify the notation. Let us endow the A-module C=⨂i=1rC(i) with the grading such that
where xi∈(C(i))pi for i=1,…,n. Then (C,D) is a complex of A-modules called the tensor product complex of the family (Ci,di). For every strictly increasing sequence r0,...,rk of [0,r] such that r0=0,rk=r, the canonical associativity isomorphism
j=0⨂k−1i=rj+1⨂rj+1C(i)→i=1⨂rC(i)
is an isomorphism of complexes.
We define as above a graded homomorphism of degree 0
γ((C(i))):i∈I⨂H(C(i))→H(i∈I⨂C(i))
Remark 3
One can define the tensor product of a finite family of complexes without endowing the index set with a total order (X, p. 185, Exercise 3).
Remark 4
Suppose that each C(i) is endowed with a graded algebra structure compatible with its grading and such that the d(i) are antiderivations (III, p. 117). Let us then endow ⨂i∈IC(i) with the left graded tensor product algebra structure of the given structures (III, p. 49). Then D is an antiderivation. Indeed, using the associativity of the tensor product, we may assume that I={1,2}; let then p1,q1,p2,q2∈Z,x1∈(C(1))p1,y1∈(C(1))q1,x2∈(C(2))p2,y2∈(C(2))q2; we have
Let C,C1 be two complexes of right A-modules, C′,C1′ two complexes of left A-modules, and u:C→C1,v:C→C1,u′:C′→C1′,v′:C′→C1′ be morphisms of complexes.
a) If u and u′ are homotopic to v and v′ respectively, then the two morphisms u⊗u′ and v⊗v′ from C⊗AC′ to C1⊗AC1′ are homotopic.
b) If u and u′ are homotopy equivalences, u⊗u′ is a homotopy equivalence.
c) If C or C′ is homotopic to zero, C⊗AC′ is homotopic to zero.
Let us denote by the same letter d the differentials of the complexes C,C1,C′,C1′ and by D the differentials of the complexes C⊗AC′ and C1⊗AC1′.
If u (resp. u′) is homotopic to v (resp. v′), there exists a graded homomorphism of degree 1 s:C→C1 (resp. s′:C′→C1′) such that
(2)u−v=ds+sd(resp. u′−v′=ds′+s′d).
Let S:C⊗AC′→C1⊗AC1′ be the unique graded homomorphism of degree 1 such that, for x∈Cp,y∈Cq′,p,q∈Z, we have
Let us prove b). If u and u′ are homotopy equivalences, there exist homomorphisms of complexes α:C1→C and α′:C1′→C′ such that u∘α,α∘u′,α′∘u′ are homotopic respectively to IdC1,IdC,IdC1′,IdC′. Then (u⊗u′)∘(α⊗α′), which is equal to (u∘α)⊗(u′∘α′), is homotopic by a) to IdC1⊗IdC1′=IdC1⊗C1′, while (α⊗α′)∘(u⊗u′) is homotopic to IdC⊗C′, whence b). Finally, c) follows from b) applied to the case where C1 or C1′ is zero.
Corollary 1
Let C′ be a split complex of left A-modules such that H(C′) is flat. For every complex C of right A-modules, the canonical mapping
γ(C,C′):H(C)⊗AH(C′)⟶H(C⊗AC′)
is bijective.
By X, p. 35, def. 6, there exists a homotopy equivalence u′:C′→H(C′). By prop. 3, 1C⊗u′:C⊗AC′→C⊗AH(C′) is a homotopy equivalence; since
H(1C⊗u′)∘γ(C,C′)=γ(C,H(C′))∘(1C⊗H(u′)),
and H(1C⊗u′) and H(u′) are bijective, it suffices to prove that γ(C,H(C′)) is bijective, and we are reduced to the case where C′ is flat and has zero differential. In this case the canonical exact sequences
(I)0⟶Z(C)iC∂B(C)⟶0
(II)0⟶B(C)jZ(C)πH(C)⟶0
give exact sequences:
0⟶Z(C)⊗AC′i⊗1C⊗AC′δ⊗1B(C)⊗AC′⟶0
0⟶B(C)⊗AC′j⊗1Z(C)⊗AC′π⊗1H(C)⊗AC′⟶0.
Since d=i∘j∘δ, we have D=d⊗1C′=(i⊗1)∘(j⊗1)∘(δ⊗1), which shows that the canonical mappings Z(C)⊗AC′→Z(C⊗AC′) and B(C)⊗AC′→B(C⊗AC′) are bijective, hence also γ(C,C′) by passing to quotients.
Corollary 2
Let N be a flat left A-module. For every complex C of right A-modules, the canonical homomorphisms
γn(C,N):Hn(C)⊗AN⟶Hn(C⊗AN)
are bijective.
Corollary 3
Let C′ be a complex of left A-modules such that B(C′) and H(C′) are projective. For every complex C of right A-modules, the mapping γ(C,C′) is bijective.
Indeed, C′ is split (X, p. 35, example 4) and H(C′) is projective; therefore we can apply corollary 1.
Remark 1
Using the commutation isomorphisms, one deduces from corollaries 1, 2 and 3 the analogous statements obtained by exchanging the roles of the two arguments of the tensor products.
Remark 2
We shall see below (X, p. 79, Cor. 4) that the conclusion of Cor. 1 is also true when one assumes C′ and H(C′) flat and C′ bounded on the right.
3. Tensor product by a flat complex bounded on the right
Lemma 1
Let C be a complex of right A-modules and E a complex of left A-modules. We assume that H(C)=0 and that E is flat and bounded on the right. Then H(C⊗AE)=0.
For k∈Z, let T(k) be the subcomplex of C⊗AE such that
Tn(k)=p+q=nq≤k∑Cp⊗AEq;
then T(k−1)⊂T(k) and we have an exact sequence of complexes
0⟶T(k−1)ikT(k)πC⊗AEk(−k)⟶0
where ik is the canonical injection and where π projects the preceding direct sum onto its factor Cn−k⊗AEk=(C⊗AEk(−k))n. By Cor. 2 above, we have H(C⊗AEk(−k))=0, hence ik is a homomorphism. We have T(k)=0 for k sufficiently small, since E is bounded on the right, hence H(T(k))=0 for all k by induction on k. Finally, the canonical morphism ⟶limT(k)→C⊗AE is obviously an isomorphism, hence H(C⊗AE)=0 (X, p. 28, Prop. 1).
Lemma 2
If u:C→C′ is a morphism of complexes of right A-modules and E a complex of left A-modules, then the complexes Con(u)⊗AE and Con(u⊗1E) are isomorphic.
By definition, Con(u)⊗AE is the graded module (C′(−1)⊕C)⊗AE endowed with the differential D such that, for x∈Cp, y′∈C′(−1)p=Cp−1′, z∈Eq, we have
(4)
D(y′,x)⊗z=(−dy′,dx−u(y′))⊗z+(−1)p(y′,x)⊗dz,
while Con(u⊗1E) is the graded module (C′⊗AE)(−1)⊕(C⊗AE) endowed with the differential D1 such that, for x∈Cp, y′∈Cp−1′, z∈Eq, we have
Let u:C→C′ be a homology isomorphism of complexes of right A-modules and E a complex of left A-modules, flat and bounded on the right. Then
u⊗1E:C⊗AE→C′⊗AE
is a homology isomorphism of complexes of k-modules.
Indeed, by X, p. 38, Cor., u (resp. u⊗1E) is a homology isomorphism if and only if H(Con(u))=0 (resp. H(Con(u⊗1E))=0). The conclusion then follows from Lemmas 1 and 2.
Remark
Using the commutation isomorphisms, one deduces from the preceding statements the analogous statements obtained by exchanging the roles of the two arguments of the tensor products.
4. Definition and first properties of the torsion product
For every A-module E, we denote by pE:L(E)→E the canonical free resolution of E (X, p. 50).
Definition 1
Let M be a right A-module and N a left A-module. We call the torsion product of M and N the graded k-module
(4)
TorA(M,N)=H(L(M)⊗AL(N)).
The homogeneous components of TorA(M,N) are denoted
(5)
TornA(M,N)=Hn(L(M)⊗AL(N)).
Since L(M) and L(N) are zero on the right, we have
(6)
TornA(M,N)=0for n<0.
Remark 1
We will see below (X, p. 107, Proposition 6) finiteness properties of the modules TorA(M,N). For example, if A is commutative noetherian and if M and N are A-modules of finite type, every A-module TornA(M,N) is of finite type.
Let f:M→M′ be a homomorphism of right A-modules and g:N→N′ a homomorphism of left A-modules. We define TorA(f,g)=H(L(f)⊗AL(g)); this is a homomorphism of graded k-modules
TorA(f,g):TorA(M,N)→TorA(M′,N′)
whose homogeneous components are denoted
TornA(f,g):TornA(M,N)→TornA(M′,N′).
According to Proposition 1 of X, p. 62, the canonical homomorphism
γ0,0:H0(L(M))⊗AH0(L(N))→H0(L(M)⊗AL(N))
is bijective; using the isomorphisms M→H0(L(M)) and N→H0(L(N)), we deduce from it an isomorphism, called *canonical
(7)
γM,N:M⊗AN→Tor0A(M,N).
We will always identify Tor0A(M,N) with M⊗AN through this isomorphism. Hence the k-linear map Tor0A(f,g) is identified with f⊗g.
Remark 2
The complex morphism pM⊗pN:L(M)⊗AL(N)→M⊗AN induces on the degree 0 homology the isomorphism
γM,N−1:Tor0A(M,N)→M⊗AN
inverse to γM,N.
We have L(1M)=1L(M), L(1N)=1L(N), hence by passage to the homology:
(8)
TorA(1M,1N)=1TorA(M,N).
If f′:M′→M′′ (resp. g′:N′→N′′) is a homomorphism of right A-modules (resp. left A-modules), we have L(g′∘g)=L(g′)∘L(g) and L(f′∘f)=L(f′)∘L(f), hence
(9)
TorA(f′∘f,g′∘g)=TorA(f′,g′)∘TorA(f,g).
Consider the morphisms of k-complexes
L(M)⊗AN1⊗pNL(M)⊗AL(N)pM⊗1M⊗AL(N)
and the k-homomorphisms they induce in homology:
H(L(M)⊗AN)ψM(N)TorA(M,N)ψN(M)H(M⊗AL(N));
according to Proposition 4 of X, p. 67, 1⊗pN and pM⊗1 are homology morphisms. Therefore:
Corollary. — If M or N is flat, ToriA(M,N)=0 for i⩾0.
Suppose that N (resp. M) is flat. Then pM⊗1:L(M)⊗AN→M⊗AN (resp. 1⊗pN:M⊗AL(N)→M⊗AN) is a homology morphism (X, p. 67, Proposition 4), hence Hi(L(M)⊗AN) (resp. Hi(M⊗AL(N))) vanishes for i>0.
Remark 3. — If g:N→N′ is a homomorphism of left A-modules, then
Let I and J be two sets, (Mα)α∈I a family of right A-modules, (Nβ)β∈J a family of left A-modules. The homomorphism
α∈I,β∈J⨁TorA(Mα,Nβ)→TorAα∈I⨁Mα,β∈J⨁Nβ
deduced from the canonical homomorphismsMα→⨁MαandNβ→⨁Nβis bijective.
It suffices to prove that for every right module M (resp. every left module N), the canonical homomorphism
β∈J⨁TorA(M,Nβ)→TorA(M,β∈J⨁Nβ)
(resp. ⨁α∈ITorA(Mα,N)→TorA(⨁α∈IMα,N)) is bijective. Now this follows from the preceding, from Proposition 1 of X, p. 28, and from the canonical isomorphisms :
β⨁(L(M)⊗ANβ)→L(M)⊗A(β⨁Nβ),
α⨁(Mα⊗AL(N))→(α⨁Mα)⊗AL(N).
An analogous reasoning gives :
Proposition 8
Let I (resp. J) be a right filtered preordered set, ((Mα, uα′,α) (resp. ((Nβ, vβ′,β))) an inductive system of right A-modules (resp. left) relative to I (resp. J). The homomorphism of graded k-modules
deduced from the canonical A-homomorphismsMα→lim⟶MαandNβ→lim⟶Nβ, is bijective.
In particular, taking J=I and remarking that the (α,α),α∈I, form a cofinal subset of I×I, one obtains :
Corollary
Let I be a right filtered preordered set, (Mi,uji) (resp. (Ni,vji)) an inductive system of right A-modules (resp. left) relative to I. The homomorphism of graded k-modules
deduced from the canonical A-homomorphisms Mi→lim⟶Mi and Nj→lim⟶Nj is bijective.
5. Linking Homomorphisms and Exact Sequences
Let M be a right A-module, N a left A-module, A∘ the opposite ring of A, M∘ the A∘-left module underlying M, N∘ the A∘-right module underlying N. We have L(M∘)=L(M)∘ and L(N∘)=L(N)∘, whence a commutation isomorphism (X, p. 63, prop. 2)
σ(L(M),L(N)):L(M)⊗AL(N)→L(N∘)⊗A∘L(M∘).
Passing to homology, σ(L(M),L(N)) induces a graded isomorphism of degree 0 σM,N:TorA(M,N)→TorA∘(N∘,M∘) called the commutation isomorphism of torsion products.
Let us note that σN∘,M∘∘σM,N=IdTor(M,N) and that σM,N induces on the degree 0 terms the commutation homomorphism of the tensor product. On the other hand, if f:M→M′ and g:N→N′ are homomorphisms of A-modules, then
TorA∘(g,f)∘σM,N=σM′,N′∘TorA(f,g).
5. The linking homomorphisms and exact sequences
Let M be a right A-module. Let us recall that for every left A-module N, an isomorphism
ψM(N):TorA(M,N)→H(L(M)⊗AN)
was defined in the previous section (X, p. 69, prop. 5). Let
(E)
0→N′uNvN′′→0
be an exact sequence of left A-modules; then the sequence of k-complexes
(ME)
0⟶L(M)⊗AN′1⊗uL(M)⊗AN1⊗vL(M)⊗AN′′⟶0
is then exact (X, p. 66, Lemma 1); denote by
∂(ME):H(L(M)⊗AN′′)→H(L(M)⊗AN′)
the corresponding linking homomorphism (X, p. 29).
Definition 2
The linking homomorphism of torsion products, relative to the module M and the exact sequence E, is the composite homomorphism
It commutes according to (X, p. 69, Remark 3) and Definition 2. Furthermore, the bottom row is exact (X, p. 30, Thm. 1), and the different ψM are bijections (X, p. 69, prop. 5).
Corollary 1
If Tor1A(M,N′′)=0, the sequence
0⟶M⊗AN′1⊗uM⊗AN1⊗vM⊗AN′′⟶0
is exact.
Corollary 2
Let 0→C′uCvC′′→0 be an exact sequence of complexes of left A-modules, and let E be a complex of right A-modules. If C′′ or E is flat, the sequence
0⟶E⊗AC′1⊗uE⊗AC1⊗vE⊗AC′′⟶0
is exact.
Indeed, Tor1A(E,C′′)=0 according to X, p. 69, Corollary to prop. 5.
Example
Let a be an ideal of A. The exact sequence
0→a→As→A/a→0
of left a-modules, gives rise to an exact sequence of torsion products, in which the terms ToriA (M, A) are zero for i>0. From this one deduces isomorphisms
Tori+1A(M,A/α)→ToriA(M,α),i>0
and an exact sequence
0→Tor1A(M,Aα)→M⊗Aα→M⊗AA→M⊗Aα→0:
it follows that Tor1A(M,A/α) is identified with the kernel of the canonical homomorphism M⊗Aα→M.
For example, taking for M a module of the form Ad/b, where b is a right ideal of A, one obtains an isomorphism of Tor1A(A/b,A/α) onto (α∩b)/ba.
Proposition 9
Let f : M → M1 be a homomorphism of right a-modules and
We shall see later other commutation relations (cf. X, p. 131, cor. 1*).*
6. Flat modules and torsion products
Theorem 2
Let E be a right A-module. The following conditions are equi-* *valent :
(i) E is flat ;
(ii) for every left A-module F, and every integern>0, one has
TornA(E,F)=0;
(iii) For every cyclic left A-module F of finite presentation, we have
Tor1A(E,F)=0;
(iv) For every left ideal a of A of finite type, the canonical mapping E⊗Aa→E is injective; (v) For every exact sequence of right A-modules of the form
0→GνHwE→0,
and every left A-module F, the sequence
0⟶G⊗AFν⊗1H⊗AFw⊗1E⊗AF⟶0
is exact. (i) ⇒ (ii): This is the corollary to Prop. 5 of X, p. 69. (ii) ⇒ (iii): This is trivial. (iii) ⇔ (iv): Every cyclic left A-module of finite presentation is isomorphic to a quotient A/a, where a is a left ideal of finite type, so (iii) is equivalent to (iv) according to X, p. 72, Example. (iii) ⇒ (i): According to X, p. 8, Prop. 3, X, p. 72, Th. 1, E is flat as soon as Tor1A(E,F)=0 for every left A-module F. If (iii) is satisfied, this is the case when F is cyclic and of finite presentation. According to X, p. 11, Prop. 7, every A-module (resp. every cyclic A-module) is a filtered inductive limit of modules of finite presentation (resp. of cyclic modules of finite presentation); hence, according to X, p. 70, Prop. 8, it suffices to prove that if Tor1A(E,F)=0 when F is cyclic, then this is also the case when F is of finite type. We therefore proceed by induction on the cardinal of a system of generators (f1,…,fn) of F; the exact sequence
0→Af1→F→F/Af1→0
gives rise to an exact sequence
Tor1A(E,Af1)→Tor1A(E,F)→Tor1A(E,F/Af1),
so that Tor1A(E,F)=0 since Tor1A(E,Af1)=0 and Tor1A(E,F/Af1)=0 by the induction hypothesis. (i) ⇒ (v): This is Corollary 2 to Th. 1 (X, p. 72). (v) ⇒ (iii): The exact sequence (X, p. 50)
0⟶Z0(E)iEL0(E)pEE⟶0
gives rise, for every left A-module F, to an exact sequence
For simplicity, set B=B(C), Z=Z(C), H=H(C), and H′=H(C′).
Since B is flat, from (I) we deduce an exact sequence (X, p. 72, Corollary 2)
(12)
0⟶Z⊗AC′j⊗1C⊗AC′δ⊗1(B⊗AC′)(−1)⟶0.
Lemma 3
The connecting homomorphism H(B⊗AC′)→H(Z⊗AC′) associated with the exact sequence (12) is equal to H(i⊗1).
Indeed, let a∈H(B⊗AC′); since B is flat, a belongs to the image of B⊗AZ(C′), hence can be written as ∑λdaλ⊗bλ, where aλ∈C, bλ∈C′, and dbλ=0. The image of the class of a under the connecting homomorphism is, by definition, the class of D(∑aλ⊗bλ)=∑daλ⊗bλ=(i⊗1)(a), which proves the lemma.
The associated homology exact sequence to (12) is therefore
H(B⊗AC′)H(i⊗1)H(Z⊗AC′)H(j⊗1)H(C⊗AC′)
H(δ⊗1)H(B⊗AC′)(−1)H(i⊗1)H(Z⊗AC′)(−1).
On the other hand, since Z is flat, from (II) we obtain a sequence of graded k-modules
which is commutative by definition of the homomorphisms γ. But since the complexes B and Z are split and flat, γB and γZ are bijective (X, p. 65, Corollary 1). From this, on the one hand, we deduce that γC is injective and has image equal to KerH(δ⊗1), and on the other hand, that γB∘∂(II,H′) is injective with image equal to ImH(δ⊗1). The theorem follows immediately from this.
Corollary 1
If B(C) and Z(C) are flat, then for every left A-module N and every integer n, there is an exact sequence
Suppose B(C) and B(C′) are projective and Z(C) is flat. Then the sequences of k-modules (11) and (13) are exact and split.* This follows from the theorem and the following lemma:
Lemma 4
If B(C) and B(C′) are projective, then the canonical homomorphism*
γ(C,C′):H(C)⊗AH(C′)→H(C⊗AC′)
has a k-linear retraction. Indeed by X, p. 65, remark b), there exist homologisms φ:C→H(C) and φ′:C′→H(C′) such that H(φ)=1H(C) and H(φ′)=1H(C′). In the commutative diagram
H(φ)⊗H(φ′) and γ(H(C),H(C′)) are the identity, whence the assertion.
Corollary 3 (universal coefficient formula). — Suppose the ring A is principal. If the complexes C and C′ are free, the sequences of A-modules (11) are exact and split; if the complex C is free, the sequences of A-modules (13) are exact and split for every A-module N. Indeed, B(C),Z(C) and B(C′) are submodules of the free modules C,C,C′, so are free (VII, § 3, cor. 2 to th. 1), and we apply cor. 2.
Corollary 4 (« Künneth formula »)
Suppose C bounded on the right, C and H(C) flat; then the canonical homomorphism
γ(C,C′):H(C)⊗AH(C′)→H(C⊗AC′)
is bijective.
By the theorem, it suffices to proving that B(C) and Z(C) are flat. Now one has exact sequences
whence, by X, p. 75, cor. 1, implications (B_{n-1}(C) is flat) ⇒ (Z_n(C) is flat) ⇒ (B_n(C) is flat) ; we conclude by remarking that B_n(C) = 0 for n small enough.
Corollary 5
Let u : C \to C' be a homologism of complexes of right a-modules, flat and bounded on the right. For every complex E of left a-modules, the morphism u⊗1E:C⊗AE→C′⊗AE is a homologism.
Indeed, Con (u) is a flat complex, bounded on the right and of zero homology; then one has H(Con (u) \otimes_A E) = 0 by cor. 4, hence H(Con (u \otimes 1_E)) = 0 (X, p. 67, lemma 2), and u⊗1E is a homologism.
8. Bounded and flat complexes over a noetherian ring
Proposition 11
Suppose A left noetherian, and let C be a bounded and flat complex of left a-modules such that H(C) is a finitely generated a-module. Let a and b be two integers such that a \leq b and H_n(C) = 0 for n < a, C_n = 0 for n > b. There exists a complex P of left a-modules such that P_n is projective and finitely generated for each n, and P_n = 0 for n \notin [a, b], and a homologism u : P \to C. Moreover, for every complex E of right a-modules, the homomorphism
H(1E⊗u):H(E⊗AP)→H(E⊗AC)is bijective.
By X, p. 53, prop. 7, there exists a complex (L, d) such that L_n is free and finitely generated for each n, and zero when n < a, and a homologism f : L \to C. Let P be the quotient complex L/L', where L'_n = 0 for n < b, L'_n = L_n for n > b, L'_b = B_b(L). Since C_n = 0 for n > b, f(L') = 0, so f factors through a morphism of complexes u : P \to C.
Since f is a homologism, one has H(Con(f))=0, whence an exact sequence
⋯⟶Lb+1db+1Lb⟶Lb−1⊕Cb⟶Lb−2⊕Cb−1⟶⋯
One then has an exact sequence
0⟶Pb⟶Lb−1⊕Cb⟶Lb−2⊕Cb−1⟶⋯.
This shows on the one hand that the cone of u is of zero homology, so that u is a homologism, on the other hand that the module Pb is flat (X, p. 76, cor. 2) ; since Pb is finitely generated as a quotient of Lb+1, it is projective (X, p. 13, cor.). The pair (P,u) then satisfies the required condition. The last assertion follows from X, p. 79, cor. 5.
\* Example. — Let A be a commutative noetherian ring, X a proper and flat A-scheme, F a coherent CX-module, flat over A. There exists a bounded complex P consisting of finitely generated projective A-modules such that for every A-module M, H(X,F⊗AM) identifies naturally with H(P⊗AM). Indeed, let U be a covering of X by a finite number of affine open sets, C(U,F) the associated Čech complex. One shows that Hi(C(U,F)) is isomorphic to the A-module Hi(X,F), and that the latter is finitely generated; moreover, for every A-module M, the complex C(U,F)⊗AM is isomorphic to C(U,F⊗AM). Applying Prop. 11 to the complex C(U,F) (which is bounded), one obtains a complex P that answers the question.
For every point y of Spec (A), let κ(y) denote the residue field of A at y, Xy=X⊗Aκ(y) the fibre of X over y, Fy=F⊗Aκ(y), and set hp(y)=dimκ(y)Hp(Xy,Fy) for p≥0.
One easily deduces the following results from the existence of the complex P :
(i) the function hp is upper semi-continuous on Spec (A) ;
(ii) the function p≥0∑(−1)php is locally constant on Spec (A). \*
9. Generalization to multimodule complexes
Let B and B′ be two rings, C a complex of (B, A)-bimodules, C′ a complex of (A, B′)-bimodules (X, p. 43); then (C⊗AC′,D) (X, p. 61) is a complex of (B, B′)-bimodules and the canonical homomorphism
γ:H(C)⊗AH(C′)⟶H(C⊗AC′)
is compatible with the (B, B′)-bimodule structures of the two members.
If B″ is a third ring, and C″ a complex of (B′, B″)-bimodules, the canonical homomorphism (II, p. 64, Prop. 8)
(C⊗AC′)⊗B′C′′⟶C⊗A(C′⊗B′C′′)
is an isomorphism of complexes of (B, B″)-bimodules.
More generally, we leave it to the reader to develop the theory of tensor products of finite, totally ordered families of multimodule complexes on the model of No. 1 (X, p. 63) and of II, pp. 65–72 (associativity, commutativity, … isomorphisms).
Let B and B' be two rings, M a (B, A)-bimodule, N an (A, B')-bimodule : then L(M)⊗AL(N) is a complex of (B, B')-bimodules, so that TorA(M,N) is endowed with a natural structure of graded (B, B')-bimodule; on the term of degree 0, this structure coincides with that of M⊗AN.
If λ∈B, λ′∈B′, and if we denote by λM,λN,λT,λT′ the homotheties x↦λx, y↦yλ′, z↦λz, z↦zλ′ of M, N, TorA(M,N), TorA(M,N) respectively, then
λT=TorA(λM,1N),λT′=TorA(1M,λN′),
which gives another description of the bimodule structure of TorA(M,N). We leave it to the reader to generalize Nos. 5 and 7 to the case of multimodule complexes.