We retain the general notations of paragraph 4. We shall also agree that, unless explicitly stated otherwise, all the modules considered are left modules, all the complexes considered are complexes of left modules.
1. Complexes of homomorphisms
Let (C,d) and (C′,d′) be two A-complexes. Consider the graded k-module HomgrA(C,C′) (II, p. 174, 175): for n∈Z, HomgrA(C,C′)n is the k-module of graded A-linear mappings of degree n from C into C′; in other words HomgrA(C,C′) is canonically identified with the a-module
Then (HomgrA(C,C′),D) is a complex of k-modules called the complex of homomorphisms of C into C′.
For example, Homgr_A (A, C') is canonically identified with C'. Note also that, for every n∈Z, one has HomgrA(C,C′)(n)=HomgrA(C,C′(n)).
The elements of Zn(HomgrA(C,C′)) are the graded homomorphisms f of (descending) degree n from C into C′ such that d′∘f=(−1)nf∘d, that is to say the morphisms of complexes from C into C′(n), or equivalently from C(p) into C′(p+n) for any fixed p. They are called the morphisms of complexes of degree (descending) nfrom C into C'; if f,g∈Zn(HomgrA(C,C′)) and s∈HomgrA(C,C′)n+1, then the condition g−f=Ds means that s is a homotopy linking the morphisms f and g from C into C′(n), so that Hn(HomgrA(C,C′))is the k-module of the homotopy classes of morphisms of descending degree n from C into C'.
Let α∈Hn(HomgrA(C,C′)) and p∈Z. Represent α by f∈Zn(HomgrA(C,C′)); then f is a morphism of complexes from C into C′(n), hence Hp(f) is a homomorphism from Hp(C) into Hp(C′(n))=Hp+n(C′); since Hp(f) depends only on the homotopy class α of f (X, p. 33, prop. 3), we deduce a canonical homomorphism of k-modules
Hn(HomgrA(C,C′))→HomA(Hp(C),Hp+n(C′)),
whence a graded k-linear mapping of degree 0, called canonical
λ(C,C′):H(HomgrA(C,C′))→Homgr˙A(H(C),H(C′)).
The homogeneous components of λ(C,C′) will often be denoted:
If C is zero on the right and C' zero on the left, then Homgr_A (C, C') is zero on the left, and the canonical k-linear mapping
λ0(C,C′):H0(HomgrA(C,C′))→HomA(H0(C),H0(C′))
is bijective.
We have exact sequences
0⟶H0(C′)iC′0d′0C′1C1d1C0pH0(C)⟶0.
On the other hand HomgrA0(C,C′) is identified with HomA(C0,C′0), Z0(HomgrA(C,C′)) then being identified with the set of f:C0→C′0 such that d′0∘f=0,f∘d1=0; B0(HomgrA(C,C′)) is zero; finally the mapping λ0 associates to the class of f modulo {0} the homomorphism φ:H0(C)→H0(C′) such that f=i∘φ∘p, whence the proposition.
Let u:C~→C and u′:C′→C~′ be morphisms of complexes; then the canonical homomorphism HomgrA(u,u′):HomgrA(C,C′)→HomgrA(C~,C~′), defined by f↦u′∘f∘u, is a morphism of complexes, as follows immediately from formula (1). Moreover, the following diagram is commutative
b) Let 0→C′uCvC′′→0 be a sequence of A-complexes which is split as an exact sequence of graded A-modules (this is the case for example if C′ is injective, or if C′′ is projective). Then for any complex E, the sequences
are exact for all p,q∈Z, and we apply II, p. 10, prop. 5 and II, p. 13, prop. 7. The proof of b) is analogous.
2. Complexes of homomorphisms and homotopies
Proposition 3
Let C,C~,C′,C~′ be four A-complexes, u:C~→C,v:C~→C,u′:C′→C~′ and v′:C′→C~′ four morphisms of complexes.
a) If u and u′ are homotopic to v and v′ respectively, then the two morphisms HomgrA(u,u′) and HomgrA(v,v′) from HomgrA(C,C′) to HomgrA(C~,C~′) are homotopic.
b) If u and u′ are homotopisms, HomgrA(u,u′) is a homotopism.
c) If C or C′ is homotopic to zero, HomgrA(C,C′) 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 HomgrA(C,C′) and HomgrA(C1,C1′). If u (resp. u′) is homotopic to v (resp. v′), there exists a graded homomorphism of degree 1,
w:C1→C(resp. w′:C′→C1′)
such that
u−v=dw+wd(resp. u′−v′=dw′+w′d)(2)
Let W:HomgrA(C,C′)→HomgrA(C1,C1′) be the graded homomorphism of degree 1 such that, for f∈HomgrA(C,C′)n, n∈Z, one has
W(f)=w′fu+(−1)nv′fw.(3)
We then have
(DW+WD)(f)=D[w′fu+(−1)nv′fw]+W[df−(−1)nfd]
=dw′fu−(−1)n+1w′fud+(−1)ndv′fw+v′fwd
+w′dfu+(−1)n+1v′dfw−(−1)nw′fdu+v′fdw
=(dw′+w′d)fu+v′f(wd+dw)
=(u′−v′)fu+v′f(u−v)=u′fu−v′fv;
This is written as DW+WD=HomgrA(u,u′)−HomgrA(v,v′), whence a).
Let us prove b). If u and u′ are homotopy equivalences, let α:C→C and α′:C′→C′ be morphisms of complexes such that u∘α, α∘u, u′∘α′, α′∘u′ are homotopic respectively to IdC, IdC, IdC′, IdC′. Then Homgr(u,u′)∘Homgr(α,α′), which is equal to Homgr(α∘u,u′∘α′), is homotopic by a) to
HomgrA(IdC,IdC′)=IdHomgrA(C,C′);
analogously Homgr(α,α′)∘Homgr(u,u′) is homotopic to IdHomgr(C,C′), whence b).
Finally c) follows from b) (applied to the case where C or C′ is zero).
Corollary 1
If C is split and H(C) projective (resp. if C′ is split and Hn(C′) injective for each n), then the canonical homomorphism
λ(C,C′):H(HomgrA(C,C′))→HomgrA(H(C),H(C′))
is bijective.
Suppose for example C′ split and H(C′) injective for each n, the case where C is split and H(C) projective being proved in an analogous manner. By X, p. 35, def. 6, there exists a homotopy equivalence u′:C′→H(C′) ; by Proposition 3, HomgrA(1,u′) is a homotopy equivalence from HomgrA(C,C′) onto HomgrA(C,H(C′)) ; since
and that HomgrA(1,H(u′)) and H(HomgrA(1,u′)) are bijective, it is enough for us to prove that λ(C,H(C′)) is bijective, which brings us back to the case where C′ is injective and has zero differential.
The canonical exact sequences (X, p. 28)
(III)
0→B(C)iCδC/B(C)→0
(IV)
0→H(C)jC/B(C)δB(C)→0
give exact sequences (X, p. 83, Proposition 2, a))
Since dC=i∘δ∘p, the differential D of HomgrA(C,C′) is given by Dn=(−1)n+1Pn∘Δn∘ln; we then have
Z(HomgrA(C,C′))=Ker(P∘Δ∘I)=KerI=ImP,
B(HomgrA(C,C′))=Im(P∘Δ∘I)=P(ImΔ)=P(KerJ);
whence an isomorphism φ:H(HomgrA(C,C′))→HomgrA(H(C),C′) such that, if a∈HomgrA(C/B(C),C′), the image of the class of P(a) under φ is J(a); we immediately verify that φ coincides with the canonical homomorphism λ.
Corollary 2
Assume that B(C) and H(C) are projective (resp. Bn(C′) and Hn(C′) are injective for every n). Then λ(C,C′) is bijective.
Indeed, C (resp. C′) is then split, by X, p. 35, Example 4, and we apply Corollary 1.
Corollary 3
Let M be a projective A-module (resp. an injective A-module). For every complex C of A-modules and every integer n, the canonical homomorphism
Hn(HomgrA(M,C))→HomA(M,Hn(C))
(resp. Hn(HomgrA(C,M))→HomA(Hn(C),M)) is bijective.
Lemma 1
a) If C or C′ is bounded on the right, if C is projective and if H(C′)=0, then H(HomgrA(C,C′))=0. b) If C or C′ is bounded on the left, if C′ is injective and if H(C)=0, then H(HomgrA(C,C′))=0.
Let f∈Zn(HomgrA(C,C′)); f is then a morphism of complexes from C to C′(n); in case a) (resp. b)), fm vanishes for m sufficiently small (resp. sufficiently large). By X, p. 47, Proposition 1, f is then homotopic to zero, hence belongs to Bn(HomgrA(C,C′)), which gives the conclusion.
Proposition 4
Let u:C′→C be a morphism of complexes, P a projective complex, E an injective complex.
a) If P is bounded on the right, or if C and C′ are bounded on the right, then
HomgrA(1,u):HomgrA(P,C′)→HomgrA(P,C)
is a morphism.
b) If E is bounded on the left, or if C and C′ are bounded on the left, then
HomgrA(u,1):HomgrA(C,E)→HomgrA(C′,E)
is a morphism.
Assume first that u is injective and set C′′=Cokeru. Since u is a morphism, C′′ is of zero homology. We have, on the other hand, exact sequences (Proposition 2)
==========
By Lemma 1, HomgrA(P,C′′) is of zero homology in case a, HomgrA(C′′,E) is of zero homology in case b, hence the conclusion.
In the general case, there exists (X, p. 38, cor. to Prop. 7) a complex C~′, which is right-bounded (resp. left-bounded) when C and C′ are, an injective morphism u~:C′→C~′ and a homotopy β:C~′→C such that u=β∘u~. Then u~ is a morphism (X, p. 34, cor. to Prop. 5); by what was shown above, HomgrA(lP,u~) (resp. HomgrA(u~,lE)) is a morphism in case a (resp. b). Moreover, HomgrA(lP,β) (resp. HomgrA(β,lE)) is a homotopy (Prop. 3); therefore, HomgrA(lP,u) (resp. HomgrA(u,lE)) is the composite of two morphisms, hence is a morphism.
3. Definition and First Properties of Extension Modules
For any A-module E, we denote by pE:L(E)→E (resp. eE:E→I(E)) the canonical free (resp. injective) resolution, cf. X, p. 50 (resp. p. 52).
Definition 1
Let M and N be two A-modules. The module of extensions of N by M is the graded k-module
(4)ExtA(M,N)=H(HomgrA(L(M),I(N))).
The homogeneous components of ExtA(M,N) are denoted
(5)ExtAn(M,N)=Hn(HomgrA(L(M),I(N))).
As L(M) (resp. I(N)) is zero to the right (resp. to the left), we have
(6)
ExtAn(M,N)=0pour n<0.
Remark 1
We will see below (X, p. 107, Prop. 6) some finiteness properties of the modules ExtAn(M,N). For example, if A is commutative and noetherian, and if M and N are A-modules of finite type, each A-module ExtAn(M,N) is of finite type.
Let f:M′→M and g:N→N’ be homomorphisms of A-modules. Set
ExtA(f,g)=H(HomgrA(L(f),I(g)));
this is a homomorphism of degree 0 of graded k-modules
ExtA(f,g):ExtA(M,N)→ExtA(M′,N′),
whose homogeneous components are denoted
ExtAn(f,g):ExtAn(M,N)→ExtAn(M′,N′).
According to Prop. 1 of [X], p. 82, the canonical homomorphism
According to Proposition 4 of X, p. 86, HomgrA(1,eN) and HomgrA(pM,1) are homologisms, hence:
Proposition 5
The k-homomorphisms*
φM(N):H(HomgrA(L(M),N))→ExtA(M,N)andφN(M):H(HomgrA(M,I(N)))→ExtA(M,N) are bijective.
Corollaire
If M is projective (resp. if N is injective), then ExtAi(M,N)=0 for i>0. Indeed, HomgrA(1,eN):HomA(M,N)→HomgrA(M,I(N)) (resp. HomgrA(pM,1):HomA(M,N)→HomgrA(L(M),N)) is then a homologism (X, p. 86, Prop. 4), yielding the conclusion.
Remarks. — 3) Let g:N→N′ be a homomorphism of A-modules. Then
Let f∈HomA(M′,M), g1,g2∈HomA(N,N′), λ1,λ2∈k. The morphisms HomgrA(L(f),λ1g1+λ2g2) and λ1HomgrA(L(f),g1)+λ2HomgrA(L(f),g2) from HomgrA(L(M),N) to HomgrA(L(M),N′) coincide, hence, by Proposition 5 and Remark 3,
Let λ∈k. If λ annihilates M or N, then it annihilates ExtA(M,N). Indeed, λ⋅1ExtA(M,N)=ExtA(λ1M,1N)=ExtA(1M,λ1N).
Proposition 7
Let I and J be two sets, (Mα)α∈I and (Nβ)β∈J two families of A-modules; the homomorphism
ExtAα∈I⨁Mα,β∈J∏Nβ⟶β∈J,α∈I∏ExtA(Mα,Nβ)
induced by the canonical homomorphisms Mα→⨁α∈IMα and ∏β∈JNβ→Nβ is bijective.
It suffices to prove that for every A-module M (resp. N), the homomorphisms
Ext(M,β∏Nβ)→β∏Ext(M,Nβ)
(resp. Ext(⨁Mα,N)→∏αExt(Mα,N)) are bijective. This follows from what has already been shown, from Proposition 1 of X, p. 28, and the canonical isomorphisms HomgrA(L(M),∏Nβ)→∏HomgrA(L(M),Nβ) and HomgrA(⨁Mα,I(N))→∏HomgrA(Mα,I(N)).
Remarque 5
Let P∘ and Q∘ be two right A-modules. We define ExtA(P,Q) by
Consider indeed the diagram on page X.91. It is commutative by the remark 3 of X, p. 88 and def. 2 ; moreover the bottom row is exact (X, p. 30, th. 1), and the vertical arrows are bijective (X, p. 88, prop. 5).
It is a graded k-homomorphism of ascending degree 1, whose homogeneous components are denoted δn(F,N):ExtAn(M′,N)→ExtAn+1(M′′,N). One then proves as above the following statements:
Theorem 2
The right-unbounded sequence of homomorphisms of k-modules
5. Projective modules, injective modules and extension modules
Proposition 10
Let M be an A-module. The following conditions are equivalent:
(i) M is projective. (ii) ExtAi(M,N)=0 for every A-module N and for every integer i>0. (iii) ExtA1(M,N)=0 for every A-module N. (iv) There exists an exact sequence
0→KνPιM→0,
where P is projective, and where ExtA1(M,K)=0.
(i) ⇒ (ii) : this is the corollary of prop. 5 of X, p. 88. (ii) ⇒ (iii) : this is trivial. (iii) ⇒ (iv) : this is clear since M is quotient of a free module P. (iv) ⇒ (i) : since ExtA1(M,K)=0, the canonical mapping
HomA(M,P)→HomA(M,M)
is surjective (X, p. 90, corollary); hence there exists an A-linear section of ν and M is isomorphic to a direct factor of P, and therefore is projective.
Proposition 11
Let N be an A-module. The following conditions are equivalent:
(i) N is injective. (ii) ExtAi(M,N)=0 for every A-module M and every integer i>0.
(iii) ExtA1(M,N)=0 for every A-module M.
(iv) There exists an exact sequence
0→NuIvC→0,
where I is injective and where ExtA1(C,N)=0;
(v) ExtA1(M,N)=0 for every cyclic A-module M.
(i) ⇒ (ii) : this is the corollary of Prop. 5 of X, p. 88.
(ii) ⇒ (iii) ⇒ (v) : this is trivial.
(iii) ⇒ (iv) : this is clear since N is a submodule of an injective module (X, p. 19, Cor. 3).
(iv) ⇒ (i) : since ExtA1(C,N)=0, the canonical homomorphism
HomA(I,N)→HomA(N,N)
is surjective (X, p. 93, Corollary); there therefore exists an A-linear retraction of u and N is isomorphic to a direct factor of I, hence is injective (X, p. 16, Prop. 9).
(v) ⇒ (i) : if α is an ideal of A, we have ExtA1(A/α,N)=0; the canonical mapping HomA(A,N)→HomA(α,N) is therefore surjective and N is injective (X, p. 16, Prop. 10).
6. Universal coefficient formula
In this number, we consider two complexes of A-modules (C,d) and (C′,d′). Let us consider the canonical exact sequences:
One can prove an analogous statement, assuming Bn(C′) and C′/Bn(C′) are injective for each n.
Set for simplicity B=B(C), Z=Z(C), H=H(C) and H′=H(C′). Since B is projective, we derive from (1) an exact sequence
(12) 0→HomgrA(B,C′)(1)Homgr(δ,1)HomgrA(C,C′)
Homgr(i,1)HomgrA(Z,C′)→0.
Lemma 2
The connecting homomorphism
Hn(HomgrA(Z,C′))→Hn(HomgrA(B,C′))
associated to (12) is equal to (−1)n+1H(Homgr(i,1)).
Proof. Indeed, let a∈Zn(HomgrA(Z,C′)); this is a morphism of complexes of ascending degree n from Z to C′, whose values are therefore in Z(C′). Since the exact sequence (1) is split (B being projective), a extends to an element b of HomgrAn(C,Z(C′)). By definition, the image of the class of a by the connecting homomorphism in question is the class in Hn(HomgrA(B,C′)) of the homomorphism u from B(C) to C′ such that for x∈C, we have
This diagram is commutative by construction of the homomorphisms λ. Moreover, since the complexes B and Z are split and projective, λB and λZ are bijective (λ, p. 84, cor. 1). One deduces from this, on the one hand that λCn is surjective, with kernel equal to ImHn(Homgr(δ,1)), and on the other hand that φn−1∘λBn−1 is surjective, with kernel equal to KerHn(Homgr(δ,1)). The theorem follows immediately from this.
Corollary 1
Assume B(C) and Z(C) projective and B^n(C') injective for each n. Then the exact sequences (11) are split. This follows from the theorem and the following lemma:
Lemma 3
If B(C) is projective and B_n(C') injective for each n, the canonical homomorphism λ(C,C′):H(HomgrA(C,C′))→HomgrA(H(C),H(C′)) has a linear section k.
Indeed, by X, p. 35, remarks a) and b), there exist homologisms
φ:C→H(C)andφ′:H(C′)→C′
such that H(φ)=1H(C) and H(φ′)=1H(C′). In the commutative diagram
Assume A principal and C free. One has for every A-module N and every integer n a split exact sequence (13). Indeed B(C) and Z(C) are free as submodules of the free module C (VII. § 3, cor. 2 to th. 1).
Corollary 4
If C is bounded on the right and if C and H(C) are projective, then
λ(C,C′):H(HomgrA(C,C′))→HomgrA(H(C),H(C′))
is bijective. By the theorem, it suffices to prove that B(C) and Z(C) are projective. Now one has exact sequences
therefore (Bn−1(C) is projective)⇒(Zn(C) is projective)⇒(Bn(C) is projective). One concludes by remarking that Bn(C)=0 for n sufficiently small.
7. Generalization to complexes of multimodules ; the canonical isomorphisms
Let B,B′ be two rings, C a complex of (A,B)-bimodules, C′ a complex of (A,B′)-bimodules ; then (HomgrA(C,C′),D) is a complex of (B,B′)-bimodules and the canonical homomorphism λ:H(HomgrA(C,C′))→HomgrA(H(C),H(C′)) is a homomorphism of (B,B′)-bimodules.
If M is an (A,B)-bimodule and N an (A,B′)-bimodule, then HomgrA(L(M),I(N)) is a complex of (B,B′)-bimodules, so that ExtA(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 HomA(M,N) (II, p. 35).
If λ∈B,λ′∈B′ and if we denote by λM,λN′,λE,λE′ the homotheties x↦xλ,y↦yλ′,z↦λz,z↦zλ′ of M,N,ExtA(M,N),ExtA(M,N) respectively, then
λE=ExtA(λM,1),λE′=Ext(1,λN′),
which gives another description of the bimodule structure of ExtA(M,N).
We leave to the reader the task of generalizing Nos. 4 and 6 to the case of complexes of multimodules.
Let C,C′,C′′ be complexes of A-modules. The composition of the mappings defines a graded homomorphism of degree zero :
(14)
HomgrA(C′,C′′)⊗kHomgrA(C,C′)→HomgrA(C,C′′).
Let B, B', E be rings, C, C', C'' complexes of (B', A)-bimodules, (A, E)-bimodules, (B, E)-bimodules respectively. By restriction of the canonical isomorphism of II, p. 73, one obtains a bijective homomorphism of (B, B')-bimodules :
Finally, let B be a ring, C a complex of right B-modules, C' a complex of right A-modules, C'' a complex of (B, A)-bimodules. One deduces from the canonical homomorphisms (II. p. 75)
Cp⊗BHomA(Cq′,Cr′′)→HomA(Cq′,Cp⊗BCr′′)
a graded homomorphism of degree zero :
(16) C⊗BHomgrA(C′,C′′)→HomgrA(C′,C⊗BC′′).
This homomorphism is bijective when C is a finitely generated projective module (II, p. 75, prop. 2).
Proposition 12
The homomorphisms (14), (15), (16) are morphisms of complexes.
Let us prove it for example for the homomorphism (14). Denote
κ:HomgrA(C′,C′′)⊗kHomgrA(C,C′)→HomgrA(C,C′)
this homomorphism. Let f∈HomgrA(C′,C′′)p and g∈HomgrA(C,C′)q; then by definition one has κ(f⊗g)=f∘g. Moreover :
(18) HomgrA(C′,C′′)⊗kC′→C′′ which maps f⊗x to f(x) is a morphism of complexes of left A-modules ; there is associated to it a canonical homomorphism (X, p. 80) of graded A-modules γ:H(HomgrA(C′,C′′))⊗kH(C′)→H(C′′), which corresponds to the canonical homomorphism of k-modules