In this paragraph, all the rings considered are commutative.
1. The complexes K(u), K.(u,C), K∗(u,C)
Let A be a ring, L an A-module, u:L→A a linear form, and Λ(L) the exterior algebra of the A-module L. For x∈Λ(L), denote by du(x) the interior product x⋅u (III, p. 161, example). According to loc. cit., p. 162, formula (60), one has
for e1,…,en in L. According to III, p. 164 and 165, the mapping du:Λ(L)→Λ(L) is an antidifferentiation of degree (−1) and of square zero. It is the unique antidifferentiation of the A-algebra Λ(L) which extends u:Λ1(L)→Λ0(L).
Definition 1
The complex (Λ(L),du) is denoted by KA(u) or K(u).
One should note that Kn(u)=Λn(L)=K−n(u). It is clear that K(u) is zero on the right and that H0(K(u))=Coker(u)=A/q where q is the ideal u(L) of A.
If the complex C is zero on the right (resp. on the left), then KA(u,C) (resp. KA∗(u,c)) is zero on the right (resp. on the left), and γ0 (resp. λ0) is bijective.
This follows from X, p. 62, prop. 1 and p. 82, prop. 1.
Proposition 1
Let x∈L; denote by Rx:y↦x∧y the left multiplication by x in the algebra Λ(L). Then du∘Rx+Rx∘du=u(x)⋅1Λ(L)=u(x)Λ(L).
Indeed (du∘Rx+Rx∘du)(y)=du(x∧y)+x∧du(y); since du is an antidifferentiation, du(x∧y)+x∧du(y)=du(x)∧y=u(x)⋅y.
Corollary 1
If u is surjective, K(u) is homotopic to zero (X, p. 34) as well as KA(u,C) and KA∗(u,C) for every complex C.
Indeed, there exists x∈L such that u(x)=1. Then K(u) is homotopic to zero by the proposition 1, and therefore also KA(u,C) (X, p. 64, prop. 3) and KA∗(u,C) (X, p. 83, prop. 3).
Corollary 2
Let C be a complex, Ann (C) its annihilator. Then q+Ann(C) annihilates HA(u,C) and HA∗(u,C).
For every λ∈q, the homothety λK(u) is homotopic to zero by the proposition, hence so also are 1C⊗λK(u) and Homgr (λK(u),1C) by X, p. 64, prop. 3 and X, p. 83, prop. 3; it follows that λ annihilates H(u,C) and H∗(u,C). If λ∈Ann(C), then 1K(u)⊗λC and Homgr (1K(u),λC) are zero.
Suppose L projective (resp. K(u) acyclic in degrees > 0). Then the complex Λ(L) is projective by III, p. 87, cor. 2 (resp. is a resolution of A/q); by X, p. 102 (resp. p. 100), one therefore has, for every A-module M, homomorphisms
(2) HrA(u,M)→TorrA(A/q,M), ExtAr(A/q,M)→HrA(u,M) resp. (3) TorrA(A/q,M)→HrA(u,M), Hr(u,M)→ExtAr(A/q,M).
If L is projective and K(u) acyclic in degrees > 0, the homomorphisms (2) and (3) above are bijective and reciprocal to one another (X, p. 102, prop. 1).
Proposition 2
Let (Li)i∈I be a family of A-modules, where the set I is finite and totally ordered. Let u be a linear form on ⨁i∈ILi, ui its restriction to Li.
The canonical isomorphism of A-algebras (III, p. 84)
g:i∈I⨂Λ(Li)→Λ(i∈I⨁Li)
is an isomorphism of the complex ⨂i∈IK(ui) (X, p. 63) onto the complex K(u).
In fact, by X, p. 64, remark 4, the differential D of the complex ⨂i∈IK(ui) is an antiderivation; the antiderivations du and g∘D∘g−1 of Λ(⨁Li) coincide on ⨁Li with the mapping x↦u(x).1 of ⨁Li into Λ(⨁Li), hence are equal (III, p. 128, cor.).
Let C and C′ be two complexes of A-modules. One has (X, p. 63 and p. 99) canonical isomorphisms of complexes
C⊗A(C′⊗AK(u))→(C⊗AC′)⊗AK(u)
HomgrA(C′,HomgrA(K(u),C))→HomgrA(C′⊗AK(u),C),
that is, isomorphisms
(4)
C⊗AKA(u,C′)→KA(u,C⊗AC′)
(5)
HomgrA(C′,KAi(u,C))→HomgrA(KA(u,C′),C)
Dans (4) and (5), take C′=K(u′), where u′:L′→A is a linear form on an A-module L′, and note that KA(u,K(u′)) which is equal by definition to K(u′)⊗AK(u) is identified, according to Prop. 2, with K(u′⊕u) where u′⊕u:L′⊕L→A is the linear form (x′,x)↦u′(x′)+u(x). We then obtain isomorphisms of complexes
(6)
KA(u′⊕u,C)→KA(u,KA(u′,C))
(7)
KAi(u′,KAi(u,C))→KAi(u′⊕u,C)
By passing to homology, we deduce isomorphisms of A-modules
HrA(u′⊕u,C)→HrA(u,KA(u′,C)),r∈Z,
HAr(u′,KAi(u,C))→HAr(u′⊕u,C),r∈Z.
Finally note that the homomorphism deduced from the product in the algebra Λ(L)
m:KA(u)⊗AKA(u)→KA(u)
is a morphism of complexes (since du is an antiderivation). Supposing that L is free of rank n and composing with the morphism of complexes KA(u)→ΛnL(−n) which is the identity in degree n, we deduce a morphism of complexes
χ:KA(u)⊗AKA(u)→ΛnL(−n);
to this morphism corresponds canonically, according to X, p. 99, Prop. 12, a morphism of complexes
φ:KA(u)→HomgrA(KA(u),ΛnL(−n))
which is bijective (III, p. 87, formula (20)). For every complex C, we deduce a composed isomorphism
By passing to homology, we therefore have canonical isomorphisms
(8)
HrA(u,C)→HAn−r(u,C⊗AΛnL),r∈Z.
Remarks. — 1) \* The above remains valid when L is projective of rank n. \*
2) Since L is free of rank n, ΛnL is isomorphic to A, and we have non-canonical isomorphisms HrA(u,C)→HAn−r(u,C).
2. Functoriality
Let f:C→C′ be a morphism of complexes. We denote
KA(u,f):KA(u,C)→KA(u,C′),
KA(u,f):KA(u,C)→KA(u,C′),
the morphisms of complexes f⊗1K(u) and HomA(1K(u),f).
We denote by HA(u,f):HA(u,C)→HA(u,C′), HA(u,f):HA(u,C)→HA(u,C′), HrA(u,f):HrA(u,C)→HrA(u,C′), HrA(u,f):HrA(u,C)→HrA(u,C′) the morphisms induced in homology. The mapping f↦KA(u,f) is linear; if g:C′→C′′ is another morphism of complexes, we have KA(u,g∘f)=KA(u,g)∘KA(u,f); analogously for KA, HA, HA, HrA, HrA.
Let 0→C′fCgC′′→0 be an exact sequence of complexes.
a) Suppose L is flat; then Λ(L) is flat (X, p. 15, Cor.). The sequence
0→KA(u,C′)KA(u,f)KA(u,C)KA(u,g)KA(u,C′′)→0
is then exact, and gives rise (X, p. 30) to an exact sequence of homology
where φ is the canonical bijection (III, p. 85, Prop. 14);
2) for every complex of A-modules C, an isomorphism of complexes of A′-modules
KA(u,A′⊗AC)→A′⊗AKA(u,C),
whence homomorphisms of A′-modules
A′⊗AHnA(u,C)→HnA′(u′,A′⊗AC),
which are bijective when A′ is flat over A (X, p. 66, Cor. 2).
Let L′ be an A-module, u′:L′→A a linear form, f:L→L′ an A-homomorphism such that u′∘f=u. It follows from III, p. 161, formula (55), that the homomorphism Λ(f):Λ(L)→Λ(L′) satisfies du∘Λ(f)=Λ(f)∘du′, hence defines a morphism of complexesΛ(u):KA(u)→KA(u′). If C is an A-complex, it follows that there are morphisms of complexes
1C⊗Λ(u):KA(u,C)→KA(u′,C) and Homgr(Λ(u),1C):KA∗(u′,C)→KA∗(u,C).
If f is bijective, all these morphisms are isomorphisms.
3. Example 1: the complex S(L)⊗AΛ(L)
Let A be a ring, L an A-module, S(L) its symmetric algebra, S(L)⊗AL the S(L)-module deduced by extension of scalars, u:S(L)⊗AL→S(L) the linear form such that u(s⊗x)=sx for s∈S(L),x∈L. By the canonical isomorphism of S(L)-modules (III, p. 83, Prop. 8)
ΛS(L)(S(L)⊗AL)→S(L)⊗AΛ(L),
the differential of the complex KS(L)(u) is carried over into the mapping
d:S(L)⊗AΛ(L)→S(L)⊗AΛ(L)
such that, for x1,...,xp,y1,...,yq in L, one has
Note that d maps Sp(L)⊗Λq(L) into Sp+1(L)⊗Λq−1(L), hence that the complex of A-modulesS(L)⊗Λ(L) decomposes into the direct sum of the complexes described by the following diagrams:
If the A-module L is the direct sum of a finite family (Li)i∈I where I is totally ordered, the canonical bijection
i∈I⨂(S(Li)⊗AΛ(Li))→S(L)⊗AΛ(L)
is an isomorphism of complexes of A-modules (this follows from prop. 2 of X, p. 148 or from formula (9) above).
Proposition 3
If the A-module L is flat, the sequences (En) above are exact for n>0.
a) Let us first note that, if pL is the composite homomorphism
S(L)⊗Λ(L)αS0(L)⊗Λ0(L)βA,
where α is the tensor product of the canonical projections and β the canonical isomorphism, it remains to prove that H(pL) is bijective.
b) If L=0 or if L=A, the proposition is obvious.
c) Suppose L is free of finite rank; write it as the direct sum L1⊕...⊕Ln of free A-modules of rank 1. According to the remark preceding the proposition, the complex S(L)⊗Λ(L) is isomorphic to the tensor product of the nfree complexes S(Li)⊗Λ(Li) whose homology is free according to b). According to X, p. 79, cor. 4, the canonical homomorphism
γ:i=1⨁nH(S(Li)⊗Λ(Li))→H(S(L)⊗Λ(L))
is bijective. According to b)H(pLi) is bijective for every i. Since ⨂i=1nH(pLi)=H(pL)∘γ,
H(pL) is bijective.
d) In the general case, L is an inductive limit of a filtered inductive system (Li)i∈I of free modules of finite rank (X, p. 14, th. 1). Since the canonical bijective homomorphism
limS(Li)⊗Λ(Li)→S(L)⊗Λ(L)
is an isomorphism of complexes, the proposition follows from X, p. 28, prop. 1.
Remark 1
We shall see below (X, p. 158, example) another proof of part c) above.
Remark 2
If A is a Q-algebra, the conclusion of prop. 3 remains true without assumption on L (cf. X, p. 206, exercise 1).
Remark 3
Let G be a group and ρ:G→GL(L) a linear representation of G in a flat A-module L. Then the (En) are exact sequences of linear representations. Suppose L finitely generated projective, and denote by RA(G) the ring of representations of G in finitely generated projective A-modules. It follows from prop. 3 that one has in R_A(G) the relations
i=0∑n(−1)i[Si(L)][Λn−i(L)]=0,n>0.
If one considers the formal series
s(T)=i=0∑∞[Si(L)]Ti∈RA(G)[[T]],
λ(T)=i=0∑∞[Λi(L)]Ti∈RA(G)[[T]],
the relations (10) are written
s(T)λ(−T)=1∗.
4. Example 2 : the case of a free module
Let k be a ring, M a k-module, I a set and p an integer ⩾0. A mapping m:Ip→M is said to be alternating if it satisfies the following two conditions:
a) for every permutation σ∈Sp and every sequence (α1,...,αp)∈Ip, one has
m(ασ(1),...,ασ(p))=εσm(α1,...,αp),
b) for every sequence (α1,...,αp)∈Ip such that two of the indices α1,...,αp are equal, one has m(α1,...,αp)=0. (In the case where I is a k-module and m is multilinear, one recovers the notion introduced in III, p. 80.)
Suppose I finite and denote by CIp(M) the k-module of alternating mappings of Ip into M.
Let L0 be a k-module, (ei)i∈I a family of elements of L0; one defines two k-linear mappings
g:Homk(ΛpL0,M)→CIp(M)
h:CIp(M)→M⊗kΛpL0
as follows: if f∈Homk(ΛpL0,M), one sets
g(f)(α1,...,αp)=f(eα1∧...∧eαp);
let m∈CIp(M), define h(m)∈M⊗kΛpL0. For every element (α1,...,αp) of Ip, the element m(α1,...,αp)⊗(eα1∧...∧eαp) of ΛpL0⊗kM is zero if Card{α1,...,αp}<p and is independent of the order of the indices α1,...,αp if
Card{α1,...,αp}=p.
It depends only on the subset J=(α1,...,αp) of I; denote it by hJ(m); we have hJ(m)=0 if Card(J)<p; we then set:
h(m)=J∑hJ(m),
where J runs through the subsets of I having p elements.
Lemma 2
If the k-module L0 is free with basis (ei)i∈I, the k-linear mappings g and h are bijective.* This follows from III, p. 79, th. 1.
Let M be a k-module, and let x=(xi)i∈I be a family of k-endomorphisms of M, pairwise permutable. Consider the polynomial ring A=k[(Xi)i∈I] and endow M with the structure of an A-module such that P(Xi)m=P(xi)m for P∈A and m∈M. Let moreover L be the free A-module AI, (ei)i∈I its canonical basis and u:L→A the linear form which maps ei onto Xi for all i∈I. Consider the complexes of k-modules KA∗(u,M) and KA∗(u,M); one has canonical isomorphisms
KAp(u,M)=HomA(ΛAp(AI),M)→Homk(Λkp(kI),M),
M⊗kΛkp(kI)→M⊗AΛAp(AI)=KAp(u,M);
whence by composition with the isomorphisms g and hisomorphisms of k-modules
θp:KAp(u,M)→CIp(M),
θp:CIp(M)→KAp(u,M).
We denote by
∂p:CIp(M)→CIp+1(M),
∂p:CIp(M)→CIp−1(M),
the k-homomorphisms obtained by transporting the differentials of KA∗(u,M) and KA∗(u,M) by the isomorphisms θ. For example, one has: (12)
The complex formed by the CIp(M) and the ∂p (resp. the ∂p) is denoted by K∗(x,M) (resp. K.(x,M)) and is called the ascending (resp. descending) Koszul complex associated with the module M and the sequence of endomorphisms (x1,...,xn). One therefore has isomorphisms of complexes of k-modules
θ∗:KA∗(u,M)→K∗(x,M),
θ∗:K∗(x,M)→KA∗(u,M).
Remark
Conversely, let B be a k-algebra, L a free B-module with basis (ei)i∈I, and M a B-module. The datum of a linear form u:L→B is equivalent to that of a family x=(xi)i∈I of elements of B, by the relation xi=u(ei). The underlying complex of k-modules of KB∗(u,M) (resp. KB(u,M)) is then identified, by the isomorphism θ∗ (resp. θ∗) with the Koszul complex K∗(x,M) (resp. K∗(x,M)). For example KB(u) is identified with K∗(x,B).
The notations H∗(x,M), H∗(x,M), etc., are introduced as in No. 1 (X, p. 147), and all the results of Nos. 1 and 2 apply mutatis mutandis, the module AI being free. We have for example isomorphisms
H0(x,M)→M/(x)M
H0(x,M)→HomA(A/(x),M),
where (x) denotes the ideal ∑Axi of A. For example also, if K∗(x,A) is acyclic in degrees >0, we have isomorphisms
Hr(x,M)→TorrA(k,M),
ExtAr(k,M)→Hr(x,M).
Finally, suppose that I is (finite and) totally ordered, for example I={1,...,n}; identify Λn(AI) with A thanks to the basis element e1∧...∧en and translate the isomorphism KpA(u,M)→KAn−p(u,M) of X, p. 149. It becomes, by transport by (θp) and (θn−p), the isomorphism
C1p(M)→C1n−p(M)
which associates with m∈C1p(M) the element m~ of C1n−p(M) such that
m(α1,...,αp)=m~(β1,...,βn−p)
if (α1,...,αp,β1,...,βn−p) is an even permutation of {1,...,n}. Let us also remark that when I={1,...,n}, we can identify C1p(M) with the set of families m(α1,...,αp) of elements of M where α1<α2<...<αp; formula (12) remains valid, as does relation (13).
Example
Take M=k[T1,...,Tn]; the Koszul complex K(∂/∂T,M) associated with the sequence of endomorphisms (∂/∂T1,...,∂/∂Tn) is identified with the de Rham complex of k[x1,...,xn] over k (X, p. 44): with m∈C{1,...,n}p(M), we associate the differential form
If L=A, put u(1)=x∈A. The complex K(u) is then of length 1, we have K0(u)=K1(u)=A and d1(a)=xa, hence for every A-module M, K0(u,M), K1(u,M), K0(u,M) and K1(u,M) are identified with M, the differentials
d1:K1(u,M)→K0(u,M)andd0:K0(u,M)→K1(u,M)
being m↦xm. We therefore have isomorphisms
H0(x,M)→A/xA⊗AM←H1(x,M),
H1(x,M)→HomA(A/xA,M)←H0(x,M).
Lemma 3
Let K be a complex such that Ki=0 for i=0,1, and let C be a complex and p an integer.
a) If K is flat, there is for every p∈N an exact sequence
0→H0(K⊗AHp(C))→Hp(K⊗AC)→H1(K⊗AHp−1(C))→0.
b) If K is projective, we have for every p∈N an exact sequence
0→H1(HomgrA(K,Hp−1(C)))→Hp(HomgrA(K,C))
→H0(HomgrA(K,Hp(C)))→0.
Let us prove a), the proof of b) being analogous. For every i, denote by K(i) the complex Ki(−i). We have an exact sequence of complexes, split as a sequence of A-modules
0→K(0)αKβK(1)→0;
the sequence
0⟶K(0)⊗ACα⊗1K⊗ACβ⊗1K(1)⊗AC⟶0
is exact, and K being flat, the homomorphisms
γ0,p(K(0),C):K(0)⊗AHp(C)→Hp(K(0)⊗AC)
γ1,p−1(K(1),C):K(1)⊗AHp−1(C)→Hp(K(1)⊗AC)
are bijective (X, p. 66, cor. 2). Let us calculate the connecting homomorphism ∂(α⊗1,β⊗1); by definition, it maps the class of the cycle ∑ai⊗bi onto that of ∑dai⊗bi, which means that
∂(α⊗1,β⊗1)∘γ(K(1),C)=γ(K(0),C)∘(dK⊗1).
The exact homology sequence associated with (14) therefore has the form
K1⊗Hp(C)dk⊗1K0⊗Hp(C)⟶Hp(K⊗C)
⟶K1⊗Hp−1(C)dk⊗1K0⊗Hp−1(C),
whence a).
Applying Lemma 3, a) to the complex K(u), and using the commutation isomorphisms, we obtain:
Proposition 4
For every complex C, there are exact sequences .
0⟶A/xA⊗AHp(C)⟶Hp(x,C)⟶HomA(A/xA,Hp−1(C))⟶0.
Corollary 1
For Hp(x,C)=0 to hold, it is necessary and sufficient that the homothety with ratio x in Hp(C) be surjective and that the homothety with ratio x in Hp−1(C) be injective.
Corollary 2
Let x=(x1,…,xn) be a sequence of elements of A, M an A-module, x′ the sequence (x1,…,xn−1). We have exact sequences
For Hi(x,M) to be zero for all i>0, it is necessary and sufficient that the homothety with ratio xn in Hi(x′,M) be bijective for i>0, and that the homothety with ratio xn in M/(x′)M be injective.
6. Completely transverse families
Let A be a ring, M an A-module, x=(xi)i∈I a family of elements of A.
Definition 2
The family x is said to be completely transverse for M if one has Hi(x,M)=0 for i>0.
If I is finite, it amounts to the same thing (X, p. 150) to say that one has Hi(x,M)=0 for i<Card(I).
The following proposition makes it possible to give examples of completely transverse families.
Proposition 5
Let x=(x1,…,xn) be a sequence of elements of A. If for i=1,…,n, the homothety of ratio xi in the A-module M/(x1M+⋯+xi−1M) is injective, the sequence x is completely secant for M.
A sequence satisfying the conditions of the proposition is said to be regular for M, or M-regular. We shall note that this property depends in general on the order of the xi; for example the sequence (1,0) is always M-regular, whereas the sequence (0,1) is so only if M is zero. On the other hand, the fact that a sequence is completely secant does not depend on the order of the terms.
Let us prove the proposition by induction on n, the case n=0 being immediate. Put x′=(x1,...,xn−1); if the sequence x is M-regular, the sequence x′ is M-regular and multiplication by xn in M/(x′)M is injective; by the induction hypothesis, we have Hi(x′,M)=0 for i>0; it then follows from cor. 3 of X, p. 157, that Hi(x,M)=0 for i>0.
Example
Let k be a ring; take A=k[X1,...,Xn] and x=(X1,...,Xn). The sequence x is A-regular and the proposition gives again the acyclicity in degrees >0 of the complex Sk(kn)⊗kΛk(kn) (cf. X, p. 152, prop. 3). Likewise in the ring of formal series A^=k[[X1,...,Xn]], the sequence (X1,...,Xn) is A^-regular, hence completely secant for A^.
Proposition 6
a) If ∑i∈IxiA=A, the family (xi)i∈I is completely secant for M. b) Let x=(x1,...,xn) be a sequence of elements of A. Let (aij)∈GLn(A); put
yi=j∑aijxj.
If the sequence x is completely secant for M, the same is true of the sequence (y1,...,yn). c) Let k1,...,kn be integers ⩾1; in order that the sequence (x1k1,...,xnkn) be completely secant for M, it is necessary and sufficient that the sequence x be completely secant for M. Assertion a) follows from cor. 1, p. 148. Let us prove b). Let f:An→An be the isomorphism defined by the matrix t(aij); it follows from X, p. 151, that 1M⊗Λ(f) is an isomorphism of the complex K.(y,M) onto the complex K.(x,M), whence b). To prove c), it is obviously enough to prove that if k is an integer ⩾1, the sequence (x1,...,xn−1,xnk) is completely secant for M if and only if the same is true of the sequence x. Let x′=(x1,...,xn−1); by cor. 3, p. 157, the first condition (resp. the second) means that the homothety of ratio xnk (resp. xn) is bijective in Hi(x′,M) for i⩾1 and injective in M/(x′)M. These two conditions are clearly equivalent, whence c).
Remark 1
The assertion analogous to c) for regular sequences is true (X, p. 207, exercise 5).
Proposition 7
a) Let N be a flat A-module. If the family x is completely secant for M, it is so for M⊗AN.
b) Let 0→M′→M→M′′→0 be an exact sequence of A-modules. If the family x is completely transverse for M′ and for M′′, it is so for M.
The complex K.(x,M⊗AN) is isomorphic by definition to K.(x,M)⊗AN; since N is flat, one deduces from this an isomorphism H.(x,M)⊗AN→H.(x,M⊗AN) (X, p. 66, cor. 2), whence a).
The complex K.(x) being flat, one has an exact sequence of complexes
0→K.(x,M′)→K.(x,M)→K.(x,M′′)→0;
assertion b) follows from the associated exact homology sequence.
Remark 2
The analogous assertions to a) and b) for regular sequences are immediate.
Remark 3
If the family x is completely transverse for A, the complex K.(x,A) defines a free resolution of the A-module A/x, with x=∑i∈IxiA; therefore for every integer i⩾0 and for every A-module M there are isomorphisms
ExtAn−i(A/x,M)→Hn−i(x,M)→Hi(x,M)→ToriA(A/x,M).
Remark 4
One says that the sequence x=(x1,...,xn) is M-coregular if (denoting by (xi)M the homothety of ratio xi in M) the homothety of ratio xi in the module
Ker(x1)M∩...∩Ker(xi−1)M
is surjective for i=1,...,n. One then hasHi(x,M)=0 for i<n: this is proved in the same way as prop. 5.
Take for example A=k[D1,...,Dn], where k is a Q-algebra, and M=k[T1,...,Tn], endowed with the A-module structure such that DiP=∂P/∂Ti for every P∈M (1⩽i⩽n). One verifies immediately that the sequence (D1,...,Dn) is M-coregular; taking into account the example p. 155, one deduces that the de Rham complex ofk[T1,...,Tn]overkis acyclic in degrees>0.
7. A criterion for completely transverse sequences
Let A be a ring, M an A-module, x an ideal of A. The x-adictopology on M is defined to be the topology compatible with the group structure of M for which the set of submodules xrM (r⩾0) is a fundamental system of neighbourhoods of zero (TG, III, p. 5, example). This topology is separated if and only if
r⩾0⋂xr=0.
Suppose now that the ideal x is generated by a sequence x=(x1,...,xn) of elements of A. Consider the graded A-module ⨁r≥0xrM and the graded A-homomorphism of degree 0
aMx:A[X1,...,Xn]⊗AM→r≥0⨁xrM
such that aMx(P⊗m)=P(x1,...,xn)m if P is a homogeneous polynomial in X1,...,Xn and m∈M. Let d be the ideal of A[X1,...,Xn] generated by the elements (xiXj−xjXi) for 1≤i<j≤n. We have P(x1,...,xn)=0 if P∈d, so that aMx gives by passing to the quotient an A-homomorphism graded of degree 0
αMx:(A[X1,...,Xn]/d)⊗AM→r≥0⨁xrM.
By tensor product with A/x, we deduce from αMx an A-homomorphism graded of degree 0
βMx:(A/x)[X1,...,Xn]⊗AM→r≥0⨁(xrM/xr+1M).
The homomorphisms aMx,αMx and βMx are surjective.
Theorem 1
Consider the following conditions: (i) The sequence x is M-regular (X, p. 158). (ii) The sequence x is completely secant for M (X, p. 157, def. 2). (iii) One has H1(x,M)=0. (iv) The homomorphism αMx:(A[X1,...,Xn]/d)⊗AM→⨁r≥0xrM is bijective. (v) The homomorphism βMx:(A/x)[X1,...,Xn]⊗AM→⨁r≥0(xrM/xr+1M) is bijective.
Then one has the implications (i) ⇒ (ii) ⇒ (iii) ⇔ (iv) ⇒ (v). If for 1≤i≤n the A-module M/(x1M+⋯+xi−1M) is separated for the x-adic topology, the conditions (i) to (v) are equivalent.
The theorem will be proved in Nos 8 and 9.
*COROLLARY 1. — If A is noetherian, if the A-module M is of finite type and if the xi belong to the radical of A, the conditions (i) to (v) of the theorem are equivalent. Indeed, on each of the modules M/(x1M+⋯+xi−1M) the x-adic topology is separated (AC III, § 3, No. 3, prop. 6).
Corollary 2
Suppose that A is a graded ring with positive degrees, M a bounded below graded A-module, and the xi homogeneous elements of degree > 0 of A. Then the conditions (i) to (v) of the theorem are equivalent. The x-adic topology is indeed separated for every bounded below graded A-module N, since if Nn=0 for n<n0 one has xaN⊂∑j≥n0+aNj for all a≥0.
Corollary 3
Suppose that the modules M/(x1M+⋯+xi−1M) are separated for the x-adic topology (1≤i≤n) ; let p be an integer between 1 and n. In order that the sequence x be completely secant for M, it is necessary and sufficient that the sequence (x1,…,xp) be completely secant for M and that the sequence (xp+1,…,xn) be completely secant for M/(x1M+⋯+xpM).
Indeed the corollary is evident if in the statement one replaces “completely secant sequences” by “regular sequences”; but the two notions coincide here by the theorem.
Remark
Let x=(x1,…,xn) be a sequence of elements of A such that H1(x,A)=0; then the kernel of the surjective homomorphism u:An→x such that
u(∑aiei)=∑aixi
is generated by the elements Xjei−Xiej; consequently, the A-algebra A[X1,…,Xn]/d is isomorphic to the symmetric algebra SA(x) (III, p. 69, prop. 4). It therefore follows from Theorem 1 that the algebra homomorphism
SA(x)⟶n⨁xn
induced by the canonical injection of x into ⨁nxn is an isomorphism. The same holds for the homomorphism
SA(x/x2)⟶n⨁xn/xn+1.
8. Proof of Theorem 1: first part
The implication (i) ⇒ (ii) has already been proved (X, p. 157, prop. 5). The implication (ii) ⇒ (iii) is obvious; the same is true of (iv) ⇒ (v), since βMx is identified with αMx⊗1A/x.
To show that (iv) implies (iii), consider the homomorphism (αMx)1 induced on the components of degree 1. Let E denote the graded A-module A[X1,…,Xn]; the A-module E1 is free with basis X1,…,Xn and d1 is the sub-A-module of E1 generated by the elements (xiXj−xjXi) for 1≤i<j≤n. Consequently ((E/d)⊗AM)1 is identified with K1(x,M)/B1(K∙(x,M)), the homomorphism (αMx)1 being identified with the mapping of K1(x,M)/B1(K∙(x,M)) into B0(K∙(x,M)) induced by d1. Thus the vanishing of H1(x,M) is equivalent to the injectivity of (αMx)1, whence the implication (iv) ⇒ (iii).
To prove that (iii) implies (iv), we shall use the following lemma:
Lemma 4
Let A0 be the ring Z[T1,…,Tn], and let u:A0[X1,…,Xn]→A0[U] be the homomorphism of A0-algebras such that u(Xi)=TiU. The kernel of u is the ideal d0 of A0[X1,…,Xn] generated by the elements (TiXj−TjXi) for 1≤i<j≤n. If t is the ideal of A0 generated by (T1,…,Tn), u induces an isomorphism
uˉ:A0[X1,…,Xn]/d0⟶r≥0⨁tr.
It is obviously enough to prove the first assertion. For every sequence of natural integers α=(α1,…,αn) and every integer k∈[0,n], let Pα,k denote the monomial
T1α1⋯TkαkXk+1αk+1⋯Xnαn;
let N be the sub-Z-module of A0[X1,...,Xn] generated by the Pα,k for α∈Nn and 0≤k≤n. We shall prove that the restriction of u to N is injective and that A0[X1,...,Xn]=d0+N; since we have d0⊂Keru, this will entail the lemma.
Observe that N is generated by the set S formed by P0,0=1 and by those of the Pα,k for which αk=0. To prove the injectivity of the restriction of u to N, it suffices to show that two distinct elements of S have as their images by u distinct monomials in A0[U]. Now we have u(Pα,k)=TαUi≥k∑αi, so that the equality u(Pα,k)=u(Pα′,k′) entails α=α′ and ∑i≥kαi=∑i≥k′αi. Suppose that Pα,k and Pα′,k′ belong to S. If α=0, then we have k=k′=0; if α=0, we obtain k=k′ since αk and αk′ are different from zero, whence the result.
Show that every monomial TαXβ∈A0[X1,...,Xn] is congruent modulo d0 to a Pη,k. For every sequence λ∈Nn, we shall denote by i(λ) (resp. j(λ)) the smallest (resp. the largest) integer k∈[1,n] such that λk=0. Among the monomials TγXδ which are congruent to TαXβ modulo d0, choose one for which the rational integer j(γ)−i(δ) is minimal; let us show that we then have j(γ)−i(δ)<0. Suppose that we have j(γ)≥i(δ); put j=j(γ),i=i(δ), and let ε=inf(γj,δi). Since (TjεXiε−TiεXjε) is divisible by (TjXi−TiXj), and therefore belongs to d0, we see that TγXδ is congruent modulo d0 to Tγ′Xδ′, where γi′=γi+ε,γj′=γj−ε,γk′=γk for k=i,j, and δi′=δi−ε,δj′=δj+ε,δk′=δk for k=i,j. Since γj′ or δi′ is zero, we have j(γ′)−i(δ′)<j(γ)−i(δ) which contradicts the minimal character of j(γ)−i(δ).
Consequently we have j(γ)<i(δ), whence TγXδ∈N, which completes the proof of the lemma.
Let us prove that (iii) implies (iv). Consider the ring A0=Z[T1,...,Tn] and the ideal t of A0 generated by T1,...,Tn. Endow M with the structure of A0-module for which Tim=xim for m∈M,1≤i≤n. According to X, p. 155, Hi(x,M) is identified canonically with Hi(T,M).
Since the sequence T is regular for A0, it follows from the implication (i) ⇒ (ii) that the complex K.(T,A0) defines a free resolution of the A0-module A0/t. Let us remark that the latter is identified with Z, endowed with the structure of A0-module such that TiZ=0 for 1≤i≤n. Thus the condition (iii) is equivalent to Tor1A0(M,Z)=0.
Let us show that (iii) implies Tor1A0(M,A0/tr)=0 for all r≥1. This results from what precedes for r=1. In the general case, consider the exact sequence
0→tr/tr+1→A0/tr+1→A0/tr→0,
The A0-module tr/tr+1 is isomorphic to a finite product of copies of Z; we therefore have Tor1A0(M,tr/tr+1)=0. It follows, by induction on r, that
Tor1A0(M,A0/tr)=0for all r,
The exact sequence (16) therefore provides, by tensor product with M, an exact sequence
0→(tr/tr+1)⊗A0M→M/xr+1M→M/xrM→0
whence an isomorphism of (t′/t′r+1)⊗A0M onto x′M/x′r+1M. By considering the exact sequence 0→t′r+1→t′→t′/t′r+1→0, one then sees by induction on r that the mapping mr:t′⊗A0M→x′M, induced by the operation of A0 in M, is an isomorphism.
To prove (iv), it remains to observe that the diagram
The kernel of pr⊗1M is identified with x′M/x′r+1M; moreover t′/t′r+1 is annihilated by the Ti and is identified with the homogeneous component of degree r of A0, so that the homomorphism (t′/t′r+1)⊗A0M→x′M/x′r+1M deduced from ir⊗1M is identified with the homogeneous component of degree r of the homomorphism βMx. It follows therefore from the exact sequence (18) that condition (v) is equivalent to
(v') : the homomorphismTor1A0(pr,1M) : Tor1A0(A0/t′r+1,M)→Tor1A0(A0/t′,M)is surjective for allr≥1.
It remains for us to prove the implication (v) ⇒ (i) when the modules
M/(x1M+⋯+xi−1M)
are separated for the x-adic topology (1≤i≤n). Let M denote the A-module M/x1M. By definition, the sequence x is regular for M if and only if (x1)M is injective and the sequence x′=(x2,...,xn) is regular for M. Reasoning by induction on n, it therefore suffices to prove that, if M is separated for the x-adic topology and if βMx is bijective, then (x1)M is injective and βMx′ is bijective. Now the bijectivity of βMx implies in particular that the homothety with ratio x1 in ⨁rx′M/x′r+1M is injective, hence that Ker(x1)M⊂⋂ixiM and consequently that (x1)M is injective if the x-adic topology on M is separated.
We are thus reduced to proving that if (x1)M is injective and if M satisfies condition (v'), then M satisfies condition (v') relative to the sequence x′.
By assumption, we have an exact sequence
0⟶M(x1)MM⟶M⟶0;
let A0=A0/T1A0,t=tA0. Let q:L→M be a free resolution of the A0-module M; since the homothety with ratio T1 is injective in A0, it is injective in L, and we have an exact sequence of complexes
0⟶L(x1)LL⟶L⟶0
with L=L/x1L, and a commutative diagram
00⟶⟶L(x1)LL⟶L⟶0↓q↓q↓qM(x1)MM⟶M⟶0.
Since q is a homomorphism of complexes inducing isomorphisms in homology, q:L→M is a free resolution of the A0-module M (cor. 1, p. 30). For every A0-module P, there is a canonical isomorphism
P⊗A0L→P⊗A0L,
whence by passing to homology an isomorphism
φP:Tor1A0(P,M)→Tor1A0(P,M).
If u:P→P′ is a A-homomorphism, then
φP′∘Tor1A0(u,1M)=Tor1A0(u,1M)∘φP.
That being so, suppose that condition (v') is satisfied for M, and let us prove that it is true for M. Let r be an integer ⩾1; we have a commutative diagram with exact rows
Now multiplication by x1 defines an injection of M/xrM into M/xr+1M: this follows immediately, by induction on r, from the exact sequence
0→xr−1M/xrM→M/xrM→M/xr−1M→0
and from the injectivity of the homothety with ratio x1 in ⨁r≥0(xrM/xr+1M). Condition (v) therefore implies that the homomorphism Tor1A0(pr,1M) is surjective for every r≥1. By composition with the isomorphisms (φA0/tr+1)−1 and φA0/tr, one deduces that the homomorphism
is surjective for r≥1, whence condition (v') for M. This completes the proof of the theorem.
10. Class of extensions associated with a regular sequence
Let A be a ring, M an A-module, x=(x1,...,xn) a sequence of elements of A. Let Mi denote the A-module M/(x1M+⋯+xi−1M) for i=0,...,n+1, so that M0 and M1 are identified with M and that Mn+1=M/(x)M. Let
xi:Mi−1→Mi,i=1,...,n,
the A-homomorphism composed of the homothety of Mi−1 with ratio xi and of the canonical projection of Mi−1 onto Mi. Finally, let us denote by p:Mn→M/(x)M the canonical projection. The diagram
(19)
0⟶Mx1M1x2M2⟶⋯xnMnpM/(x)M⟶0
is an exact sequence if and only if the sequence x is M-regular. Henceforth suppose that the sequence x is regular for M. The element θx∈ExtAn(M/(x)M,M) associated with the exact sequence (19) is also said to be associated with the M-regular sequence x.
Let i be an integer, 1≤i≤n. Note that the sequence (19) can be decomposed into the two exact sequences
(20)
0⟶Mx1M1x2M2⟶⋯xiMi⟶M/(x1M+⋯+xiM)⟶0
(21)
0⟶M/(x1M+⋯+xiM)xi+1Mi+1⟶⋯⟶MnpM/(x)M⟶0
which are none other than the exact sequences associated with the sequence (x1,...,xi), which is regular for M, and with the sequence (xi+1,...,xn), which is regular for M/(x1M+⋯+xiM). Denoting by
the extension classes associated with (20) and (21), one has, by X, p. 118, prop. 3,
(22)
θ(x1,…,xn)=θ(x1,…,xi)∘θ(xi+1,…,xn).
Moreover, by prop. 5 (X, p. 157), the Koszul complex K.(x,M) is acyclic in degrees =n, whence an exact sequence
(23)
0⟶M∂0K1(x,M)∂1K2(x,M)∂2⋯⟶Kn(x,M)qM/(x)M⟶0,
where K0(x,M) has been identified with M and where q maps each element m∈Kn(x,M) to the class in M/(x)M of m(1,2,…,n)∈M.
Proposition 8
Suppose the sequence x is regular for M. The element of ExtAn(M/(x)M,M) associated with the exact sequence (23) is (−1)n(n+1)/2θx.* For i=0,1,…,n, define an A-linear mapping
ai:Ki(x,M)→Mi=M/(x1M+⋯+xi−1M)
as follows: if m∈Ki(x,M), ai(m) is the class in Mi of the element m(1,2,…,i) of M. It is clear that a0 is the identity mapping of M and that p∘an=q. Moreover ai+1∘∂i(m) is the image in Mi+1 of the element
k=1∑i+1(−1)k+1xkm(1,2,…,k−1,k+1,…,i+1).
Since xk annihilates Mi+1 for k=1,…,i, ai+1∘∂i(m) is also the image of
(−1)ixi+1m(1,2,…,i),
hence
ai+1∘∂i=(−1)ixi+1∘ai.
By X, p. 120, cor. 1 and 2, the element of ExtAn(M/(x)M,M) associated with (23) is equal to ∏i=1n(−1)i⋅θx, whence the assertion.
Corollary
Suppose moreover that the modules M/(x1M+⋯+xi−1M) are separated for the (x)-adic topology, and let (aij)∈GLn(A). Put*
yi=j∑aijxjandy=(y1,…,yn).
Then the sequence y is regular for M, and one has θy=det(aij)−1θx. Indeed, the sequence y is regular for M by Prop. 6 (X, p. 158) and Theorem 1 (X, p. 160); the last assertion follows from Prop. 8, and from Prop. 4 of X, p. 119.
Proposition 9
Suppose the sequence x regular for M. If N is an A-module such that (x)N=0, one has ExtAi(N,M)=0 for i<n, and the mapping α↦θx∘α* from Hom_A (N, M/(x) M) into Ext^n_A (N, M) (which is also the iterated connecting homomorphism associated with (19), cf. X, p. 127, Corollary 3) is bijective.
It remains to prove that the homomorphism ψi:α↦θx∘α from ExtAi−n(N,M/(x)M) into ExtAi(N,M) is bijective for i≤n. Let us reason by induction on n, the assertion being trivial for n=0. Put M1=M/x1M, x′=(x2,...,xn), so that x′ is M1-regular. By the induction hypothesis, the homomorphism
ψi−1:α↦θx′∘α
from ExtAi−n(N,M/(x)M) into ExtAi−1(N,M1) is bijective for i<n. On the other hand, consider the exact sequence
0⟶M(x1)MM⟶M1⟶0;
the connecting homomorphism ExtAi(N,M1)→ExtAi+1(N,M) associated with it is φi:β↦θx1∘β (X, p. 125, Prop. 5) ;
since one has Exti(1N,(x1)M)=Exti((x1)N,1M)=0, one deduces from this exact sequences
0⟶ExtAi(N,M)⟶ExtAi(N,M1)φiExtAi+1(N,M)⟶0.
Since ExtAi(N,M1)=0 for i<n−1, it follows that ExtAi+1(N,M)=0 for i<n−1, that is to say for i+1<n; it follows that φi is bijective for i<n. Since ψi(α)=θx∘α=θx1∘θx′∘α=φi−1∘ψi(α) for α∈ExtAi−n(N,M/(x)M), ψi is indeed bijective for i≤n.