Bourbaki

§ 10. Local cohomology, grothendieck duality

1. Local cohomology

In this No., we consider a local noetherian ring AA. Recall (VIII, § 3, No. 3, Lemma 2) that the ideals of definition of AA are the ideals of AA distinct from AA containing a power of mAm_A, or again the ideals a⊂mAa \subset m_A such that A/aA/a is of finite length. We denote by D\mathscr{D} the set of ideals of definition of AA, endowed with the order relation opposite to inclusion; it is filtering to the right.

Let MM be an AA-module. To every ideal of definition aa of AA let us associate the graded AA-module ExtA(A/a,M)\mathrm{Ext}_A(A/a, M); if aa and bb are ideals of definition with a⊂ba \subset b, let pab:A/a→A/bp_{ab} : A/a \to A/b denote the canonical mapping and consider the AA-linear mapping Ext(pab,1M):ExtA(A/b,M)⟶ExtA(A/a,M)\mathrm{Ext}(p_{ab}, 1_M) : \mathrm{Ext}_A(A/b, M) \longrightarrow \mathrm{Ext}_A(A/a, M). Thus one obtains an inductive system of graded AA-modules and graded AA-linear mappings of degree 0 relative to the ordered set D\mathscr{D}. The graded AA-module lim←⁡a∈DExtA(A/a,M)\varprojlim_{a \in \mathscr{D}} \mathrm{Ext}_A(A/a, M) is called the local cohomology module of MM, and is denoted by HA(M)H_A(M).

The ideals mAnm_A^n for n≥1n \geq 1 form a cofinal subset of D\mathscr{D}; hence there is a canonical isomorphism of graded modules lim←⁡nExtA(A/mAn,M)⟶HA(M)\varprojlim_n \mathrm{Ext}_A(A/m_A^n, M) \longrightarrow H_A(M). It follows that every element of HA(M)H_A(M) is annihilated by a power of mAm_A.

Remark 1

Let XX be the topological space Spec(A)\mathrm{Spec}(A), OX\mathscr{O}_X the structural sheaf of rings and M~\widetilde{M} the OX\mathscr{O}_X-module associated with MM. The graded AA-module HA(M)H_A(M) is identified with the module H{mA}(X,M~)H_{\{m_A\}}(X, \widetilde{M}) of cohomology with support in the closed point mAm_A of XX.

For every homomorphism f:M→Nf : M \to N of AA-modules, the mappings ExtA(1A/a,f):ExtA(A/a,M)⟶ExtA(A/a,N)\mathrm{Ext}_A(1_{A/a}, f) : \mathrm{Ext}_A(A/a, M) \longrightarrow \mathrm{Ext}_A(A/a, N) form an inductive system of graded linear mappings. By passing to the inductive limit, one obtains a graded homomorphism HA(f):HA(M)→HA(N)H_A(f) : H_A(M) \to H_A(N). For every sequence M→fN→gPM \xrightarrow{f} N \xrightarrow{g} P of AA-modules and homomorphisms, one has HA(g∘f)=HA(g)∘HA(f)H_A(g \circ f) = H_A(g) \circ H_A(f). Let

0→M→fN→gP→0 \begin{array}{ccccccccc} 0 & \to & M & \xrightarrow{f} & N & \xrightarrow{g} & P & \to & 0 \end{array}

be an exact sequence of AA-modules. By AA, X, p. 90, Prop. 8, the connecting homomorphisms of the extension modules ExtA(A/a,P)⟶ExtA(A/a,M)\mathrm{Ext}_A(A/a, P) \longrightarrow \mathrm{Ext}_A(A/a, M) form an inductive system of AA-linear mappings, graded of (ascending) degree +1. By passing to the inductive limit, one deduces an AA-homomorphism ∂(E):HA(P)→HA(M)\partial(\mathscr{E}) : H_A(P) \to H_A(M), graded of degree +1, which makes exact the sequence of homomorphisms

…⟶HAn−1(P)→∂n−1(E)HAn(M)→HAn(f)HAn(N)→HAn(g)HAn(P)→∂n(E)HAn+1(M)⟶… \ldots \longrightarrow H_A^{n-1}(P) \xrightarrow{\partial^{n-1}(\mathscr{E})} H_A^n(M) \xrightarrow{H_A^n(f)} H_A^n(N) \xrightarrow{H_A^n(g)} H_A^n(P) \xrightarrow{\partial^n(\mathscr{E})} H_A^{n+1}(M) \longrightarrow \ldots

Let M be an a-module. For every ideal α\alpha of A, the A-module HomA(A/α,M)\mathrm{Hom}_A(A/\alpha, M) is canonically identified with the submodule of M consisting of the elements annihilated by α\alpha. Thus HA0(M)H_A^0(M) is identified with the submodule of M consisting of the elements mm which are annihilated by a power of mA\mathfrak{m}_A, that is to say such that longA(Am)<+∞\mathrm{long}_A(Am) < +\infty. In particular one has HA0(M)=MH_A^0(M) = M when M is artinian.

Example 1

If M is injective, the A-module HAi(M)H_A^i(M) is zero for i>0i > 0 and injective for i=0i = 0 (§ 8, No. 2, lemma 1, c)).

Example 2

If HA0(M)=MH_A^0(M) = M (for example if M is artinian), HAi(M)H_A^i(M) is zero for i>0i > 0. Let in fact (I, e) be an injective envelope of M. The submodule HA0(I)H_A^0(I) of I is injective (example 1) and contains e(M)e(M), hence is equal to I. Put N=Coker eN = \mathrm{Coker}\,e and consider the exact sequence 0→M→eI→pN→00 \to M \xrightarrow{e} I \xrightarrow{p} N \to 0. Since I=HA0(I)I = H_A^0(I), one has N=HA0(N)N = H_A^0(N) and the homomorphism HA0(p)H_A^0(p) is surjective. Since HAi(I)H_A^i(I) is zero for i>0i > 0 (example 1), HA1(M)H_A^1(M) is zero and HAi(M)H_A^i(M) is isomorphic to HAi−1(N)H_A^{i-1}(N) for i>1i > 1; one concludes by reasoning by induction on the integer ii.

Example 3

Let Ω\Omega be a dualizing A-module. For i≠dim⁡(A)i \neq \dim(A), one has ExtAi(A/α,Ω)=0\mathrm{Ext}_A^i(A/\alpha, \Omega) = 0 for every ideal of definition α\alpha of A (§ 8, No. 5, example 3), whence HAi(Ω)=0H_A^i(\Omega) = 0; for i=dim⁡(A)i = \dim(A), the A-module HAi(Ω)H_A^i(\Omega), which is isomorphic to lim←⁡ExtAi(A/mAn,Ω)\varprojlim \mathrm{Ext}_A^i(A/\mathfrak{m}_A^n, \Omega), is a Matlis module (loc. cit., example 6).

Example 4

Let A be a noetherian local integral ring; let K denote its field of fractions, and suppose that A≠KA \neq K. It is an injective A-module (A, X, p. 18, example 1), so that the module HA(K)H_A(K) is zero (example 1). From the exact sequence 0→A→K→K/A→00 \to A \to K \to K/A \to 0, one obtains for every ii an isomorphism HAi(K/A)→HAi+1(A)H_A^i(K/A) \to H_A^{i+1}(A).

More generally, for every torsion-free A-module M and every integer ii, one deduces from the exact sequence

0→M→K⊗AM→(K/A)⊗AM→0 0 \to M \to K \otimes_A M \to (K/A) \otimes_A M \to 0

an isomorphism HAi((K/A)⊗AM)→HAi+1(M)H_A^i((K/A) \otimes_A M) \to H_A^{i+1}(M).

Example 5

Let us retain the hypotheses of the preceding example and suppose moreover that dim⁡(A)=1\dim(A) = 1. Let N be a torsion A-module; since every nonzero ideal of A distinct from A is an ideal of definition (VIII, § 1, No. 3, prop. 6, e)), one has HA0(N)=NH_A^0(N) = N, and consequently HAi(N)=0H_A^i(N) = 0 for i>0i > 0 (example 2).

Let M be an A-module; let T(M)T(M) denote its torsion submodule. Consider the long exact local cohomology sequence associated with the exact sequence

0→T(M)→M→M/T(M)→0; 0 \to T(M) \to M \to M/T(M) \to 0 ;

in view of the foregoing, one deduces from it canonical isomorphisms T(M)→HA0(M)T(M) \to H_A^0(M) and HA1(M)→HA1(M/T(M))H_A^1(M) \to H_A^1(M/T(M)). Since the canonical homomorphism (K/A)⊗AM→(K/A)⊗A(M/T(M))(K/A) \otimes_A M \to (K/A) \otimes_A (M/T(M)) is bijective, one finally obtains *canonical isomorphisms

HA0(M)→T(M),HA1(M)→(K/A)⊗AM. H_A^0(M) \to T(M) , \qquad H_A^1(M) \to (K/A) \otimes_A M .

Proposition 1

Let AA be a noetherian local ring and MM a finitely generated AA-module.

a) The AA-module HA(M)H_A(M) is artinian, and zero in degree >dim⁡(M)> \dim(M).

b) Let p=prof⁡A(M)p = \operatorname{prof}_A(M). We have HAi(M)=0H_A^i(M) = 0 for i<pi < p, and HAp(M)≠0H_A^p(M) \neq 0 if MM is nonzero.

Let us prove a) by induction on dim⁡(M)\dim(M). The case where dim⁡(M)≤0\dim(M) \leq 0 follows from Example 2 above. Suppose dim⁡(M)>0\dim(M) > 0 and first take MM of the form A/pA/\mathfrak{p}, where p\mathfrak{p} is a prime ideal of AA distinct from mA\mathfrak{m}_A. Let xx be an element of mA−p\mathfrak{m}_A - \mathfrak{p}; we have an exact sequence 0→M→xMM⟶M/xM→00 \to M \xrightarrow{xM} M \longrightarrow M/xM \to 0, with dim⁡(M/xM)=dim⁡(M)−1\dim(M/xM) = \dim(M) - 1. Hence we deduce an exact sequence of local cohomology

HAi−1(M/xM)⟶HAi(M)→xHAi(M) H_A^{i-1}(M/xM) \longrightarrow H_A^i(M) \xrightarrow{x} H_A^i(M)

Every element of HAi(M)H_A^i(M) is annihilated by a power of mA\mathfrak{m}_A; to prove that this module is artinian, it is enough therefore to prove that the socle of HAi(M)H_A^i(M) is of finite dimension over κA\kappa_A (§ 8, No. 3, Lemma 3). By the induction hypothesis, the kernel NN of the homothety of ratio xx in HAi(M)H_A^i(M) is artinian; since xx belongs to mA\mathfrak{m}_A, the socle of HAi(M)H_A^i(M) identifies with that of NN, and is therefore of finite dimension. If i>dim⁡(M)i > \dim(M), we have HAi−1(M/xM)=0H_A^{i-1}(M/xM) = 0 by the induction hypothesis, so that the homothety of ratio xx is injective in HAi(M)H_A^i(M); since every element of HAi(M)H_A^i(M) is annihilated by a power of xx, we deduce that HAi(M)=0H_A^i(M) = 0, whence a) in the case considered.

Let us pass to the general case. The AA-module MM admits a composition series (Mj)0≤j≤n(M_j)_{0 \leq j \leq n} such that each quotient Mj/Mj+1M_j/M_{j+1} is isomorphic to A/pjA/\mathfrak{p}_j, where pj\mathfrak{p}_j is a prime ideal of AA (IV, § 1, No. 4, Theorem 1). Let us prove by induction on nn that MM satisfies a). The case n=0n = 0 is trivial. The exact sequence 0→M1→M→A/p0→00 \to M_1 \to M \to A/\mathfrak{p}_0 \to 0 yields an exact sequence of local cohomology

HAi(M1)⟶HAi(M)⟶HAi(A/p0) H_A^i(M_1) \longrightarrow H_A^i(M) \longrightarrow H_A^i(A/\mathfrak{p}_0)

the AA-module HAi(M1)H_A^i(M_1) is artinian by the induction hypothesis, and the same is true of HAi(A/p0)H_A^i(A/\mathfrak{p}_0) by the cases already treated; consequently HAi(M)H_A^i(M) is artinian. If i>dim⁡(M)i > \dim(M), the modules M1M_1 and A/p0A/\mathfrak{p}_0 are of dimension <i< i; the modules HAi(M1)H_A^i(M_1) and HAi(A/p0)H_A^i(A/\mathfrak{p}_0) are therefore zero by the induction hypothesis and the cases already treated, which implies that HAi(M)=0H_A^i(M) = 0.

Suppose MM nonzero, and let us prove b) by induction on the integer p=prof⁡(M)p = \operatorname{prof}(M). The case p=0p = 0 follows from the definition of depth. Suppose p>0p > 0 and choose an element xx of mA\mathfrak{m}_A such that the homothety xMx_M is injective. As above we obtain a local cohomology exact sequence

HAi−1(M/xM)⟶HAi(M)→xHAi(M) H_A^{i-1}(M/xM) \longrightarrow H_A^i(M) \xrightarrow{x} H_A^i(M)

We have prof⁡(M/xM)=prof⁡(M)−1\operatorname{prof}(M/xM) = \operatorname{prof}(M) - 1 (§ 1, No. 4, prop. 7), whence HAi−1(M/xM)=0H_A^{i-1}(M/xM) = 0 for i<pi < p by the induction hypothesis, which implies as above that HAi(M)=0H_A^i(M) = 0. In particular HAp−1(M)H_A^{p-1}(M) is zero, so that the homomorphism HAp−1(M/xM)⟶HAp(M)H_A^{p-1}(M/xM) \longrightarrow H_A^p(M) is injective; thus HAp(M)H_A^p(M) is nonzero by the induction hypothesis.

It can be shown that the module HAdim⁡(M)(M)H_A^{\dim(M)}(M) is nonzero when MM is nonzero (exerc. 4; cf. No. 3, cor. of th. 2).

Corollary

Let MM be a nonzero Macaulay AA-module of finite type. The AA-module HAi(M)H_A^i(M) is zero for i≠dim⁡(M)i \neq \dim(M) and nonzero for i=dim⁡(M)i = \dim(M).

Remark 2

For every ideal of definition α\alpha of AA, the AA-module ExtA(A/α,M)\mathrm{Ext}_A(A/\alpha, M) is annihilated by α\alpha, and A/αA/\alpha identifies with A^/αA^\widehat{A}/\alpha \widehat{A}; consequently, the graded AA-module ExtA(A/α,M)\mathrm{Ext}_A(A/\alpha, M) identifies with A^⊗AExtA(A/α,M)\widehat{A} \otimes_A \mathrm{Ext}_A(A/\alpha, M), hence also with ExtA^(A^/αA^,A^⊗AM)\mathrm{Ext}_{\widehat{A}}(\widehat{A}/\alpha \widehat{A}, \widehat{A} \otimes_A M) (A, X, p. 111, prop. 10). The set of ideals αA^\alpha \widehat{A}, for α∈D\alpha \in \mathscr{D}, contains the powers of mA^m_{\widehat{A}}, hence is cofinal in the set of ideals of definition of A^\widehat{A}; we therefore deduce from the foregoing a canonical isomorphism of graded AA-modules

HA(M)⟶HA^(A^⊗AM) H_A(M) \longrightarrow H_{\widehat{A}}(\widehat{A} \otimes_A M)

If the AA-module MM is of finite type, the AA-module A^⊗AM\widehat{A} \otimes_A M identifies with the completion M^\widehat{M} of MM (III, § 3, No. 4, th. 3), and we have an isomorphism HA(M)→HA^(M^)H_A(M) \to H_{\widehat{A}}(\widehat{M}), graded of degree 0.

2. Local cohomology on a Macaulay ring

In this number, it is assumed that AA is a local Macaulay ring; we put dim⁡(A)=d\dim(A) = d.

The ideals generated by a sequence of elements of mAm_A completely secant for AA and of length d=dim⁡(A)d = \dim(A) form a cofinal subset Dcs\mathscr{D}_{cs} in the set D\mathscr{D} of ideals of definition of AA. In fact, let (x1,…,xd)(x_1, \ldots, x_d) be a sequence of elements of mAm_A completely secant for AA (§ 2, Proposition 3); for every integer nn, the sequence (x1n,…,xdn)(x_1^n, \ldots, x_d^n) is completely secant for AA (A, X, p. 158, Proposition 6, c)), and generates an ideal of definition (VIII, § 3, Corollary to Proposition 3 and Theorem 1) contained in mAnm_A^n.

Let α∈Dcs\alpha \in \mathscr{D}_{cs}, and let π:L→A/α\pi : L \to A/\alpha be a finite free resolution, zero in degree >d> d (for example the Koszul complex associated with a completely secant sequence for AA generating α\alpha). Consider the dual L∗=HomgrA(L,A)L^* = \mathrm{Homgr}_A(L, A) of LL; since the depth of AA is equal to dd, one has ExtAi(A/α,A)=0\mathrm{Ext}_A^i(A/\alpha, A) = 0 for i<di < d (§ 1, Corollary 2 to Proposition 2). Since L∗L^* is of length ≤d\leq d, it follows that Hi(L∗)H^i(L^*) is zero for i≠di \neq d and that Hd(L∗)H^d(L^*) is identified with ExtAd(A/α,A)\mathrm{Ext}_A^d(A/\alpha, A) (A, X, p. 100, Theorem 1). Hence there is a homologism

π∗:L∗(−d)⟶ExtAd(A/α,A) \pi^* : L^*(-d) \longrightarrow \mathrm{Ext}_A^d(A/\alpha, A)

which defines a finite free resolution of ExtAd(A/α,A)\mathrm{Ext}_A^d(A/\alpha, A).

Let MM be an AA-module; consider the canonical isomorphisms (loc. cit.)

φ(L,M):H(HomgrA(L,M))→ExtA(A/α,M) \varphi(L, M) : H(\mathrm{Homgr}_A(L, M)) \to \mathrm{Ext}_A(A/\alpha, M)
ψ(M,L∗(−d)):TorA(M,ExtAd(A/α,A))→H(M⊗AL∗)(−d) \psi(M, L^*(-d)) : \mathrm{Tor}^A(M, \mathrm{Ext}_A^d(A/\alpha, A)) \to H(M \otimes_A L^*)(-d)

Since the complex L is finite free, the canonical morphism of complexes M⊗AL∗→HomgrA(L,M)M \otimes_A L^* \to \mathrm{Homgr}_A(L, M) is an isomorphism; it follows that there is an isomorphism of graded A-modules H(M⊗AL∗)→H(HomgrA(L,M))H(M \otimes_A L^*) \to H(\mathrm{Homgr}_A(L, M)). By composition of the preceding isomorphisms, one obtains an isomorphism of graded A-modules, called canonical,

τ(L,M):TorA(M,ExtAd(A/α,A))(d)⟶ExtA(A/α,M) \tau(L, M) : \mathrm{Tor}^A(M, \mathrm{Ext}_A^d(A/\alpha, A))(d) \longrightarrow \mathrm{Ext}_A(A/\alpha, M)

which induces for each integer ii an isomorphism

τi(L,M):Tord−iA(M,ExtAd(A/α,A))⟶ExtAi(A/α,M) \tau^i(L, M) : \mathrm{Tor}_{d-i}^A(M, \mathrm{Ext}_A^d(A/\alpha, A)) \longrightarrow \mathrm{Ext}_A^i(A/\alpha, M)

For M=AM = A, τd(L,A)\tau^d(L, A) is the canonical isomorphism of A⊗AExtAd(A/α,A)A \otimes_A \mathrm{Ext}_A^d(A/\alpha, A) onto ExtAd(A/α,A)\mathrm{Ext}_A^d(A/\alpha, A).

Let bb be an ideal of Dcs\mathscr{D}_{cs} contained in α\alpha. Let ρ:R→A/b\rho : R \to A/b be a finite free resolution of length ≤d\leq d and let pab:A/b→A/αp_{ab} : A/b \to A/\alpha be the canonical surjection. By A, X, p. 49, Proposition 3, there exists a morphism of complexes PLR:R→LP_{LR} : R \to L such that π∘PLR=pab∘ρ\pi \circ P_{LR} = p_{ab} \circ \rho. By Proposition 2 of A, X, p. 103, one has a commutative diagram

TorA(M,ExtAd(A/α,A))(d)→Tor(1M,Extd(pab,1A))TorA(M,ExtAd(A/b,A))(d)↓ψ(L∗(−d),M)↓ψ(R∗(−d),M)H(M⊗AL∗)→H(1M⊗tPLR)H(M⊗AR∗)↓↓H(HomgrA(L,M))→H(Homgr(PLR,M))H(HomgrA(R,M))↓φ(L,M)↓φ(R,M)ExtA(A/α,M)→Ext(pab,1M)ExtA(A/b,M) \begin{array}{ccc} \mathrm{Tor}^A(M, \mathrm{Ext}_A^d(A/\alpha, A))(d) & \xrightarrow{\mathrm{Tor}(1_M, \mathrm{Ext}^d(p_{ab}, 1_A))} & \mathrm{Tor}^A(M, \mathrm{Ext}_A^d(A/b, A))(d) \\ \downarrow \psi(L^*(-d), M) & & \downarrow \psi(R^*(-d), M) \\ H(M \otimes_A L^*) & \xrightarrow{H(1_M \otimes^t P_{LR})} & H(M \otimes_A R^*) \\ \downarrow & & \downarrow \\ H(\mathrm{Homgr}_A(L, M)) & \xrightarrow{H(\mathrm{Homgr}(P_{LR}, M))} & H(\mathrm{Homgr}_A(R, M)) \\ \downarrow \varphi(L, M) & & \downarrow \varphi(R, M) \\ \mathrm{Ext}_A(A/\alpha, M) & \xrightarrow{\mathrm{Ext}(p_{ab}, 1_M)} & \mathrm{Ext}_A(A/b, M) \end{array}

It follows first, by taking α=b\alpha = b, that the isomorphism τ(L,M)\tau(L, M) does not depend on the choice of the resolution L of A/αA/\alpha; let us denote it by τα(M)\tau_\alpha(M). It then follows that the τα(M)\tau_\alpha(M) for α∈Dcs\alpha \in \mathscr{D}_{cs} form an inductive system of isomorphisms. Passing to the inductive limit, and taking account of A, X, p. 70, prop. 8, one obtains for each integer ii an isomorphism of a-modules

τi(M):Tord−iA(M,HAd(A))⟶HAi(M) \tau^i(M) : \mathrm{Tor}_{d-i}^A(M, H_A^d(A)) \longrightarrow H_A^i(M)

For M=AM = A, τd(A)\tau^d(A) is the canonical isomorphism of A⊗AHAd(A)A \otimes_A H_A^d(A) onto HAd(A)H_A^d(A).

Remark 1

Let f:M→Nf : M \to N be a homomorphism of a-modules. Using A, X, p. 103, prop. 2, one proves that the following diagrams are commutative:

Tord−iA(M,HAd(A))→τi(M)HAi(M)↓Tord−i(f,1)↓Hi(f)Tord−iA(N,HAd(A))→τi(N)HAi(N) \begin{array}{ccc} \mathrm{Tor}_{d-i}^A(M, H_A^d(A)) & \xrightarrow{\tau^i(M)} & H_A^i(M) \\ \downarrow \mathrm{Tor}_{d-i}(f,1) & & \downarrow H^i(f) \\ \mathrm{Tor}_{d-i}^A(N, H_A^d(A)) & \xrightarrow{\tau^i(N)} & H_A^i(N) \end{array}

Remark 2

Let

(E)0→M→N→P→0 (\mathscr{E}) \quad 0 \to M \to N \to P \to 0
be an exact sequence of a-modules. Using A, X, p. 104, prop. 3 and p. 106, prop. 4, one proves that the following diagrams are commutative:

Tord−iA(P,HAd(A))→τi(P)HAi(P)↓∂d−i(E,HAd(A))↓∂i(E)Tord−i−1A(M,HAd(A))→τi+1(M)HAi+1(M) \begin{array}{ccc} \mathrm{Tor}_{d-i}^A(P, H_A^d(A)) & \xrightarrow{\tau^i(P)} & H_A^i(P) \\ \downarrow \partial_{d-i}(\mathscr{E}, H_A^d(A)) & & \downarrow \partial^i(\mathscr{E}) \\ \mathrm{Tor}_{d-i-1}^A(M, H_A^d(A)) & \xrightarrow{\tau^{i+1}(M)} & H_A^{i+1}(M) \end{array}

Remark 3

Consider the isomorphism

τd(M):M⊗AHAd(A)⟶HAd(M) \tau^d(M) : M \otimes_A H_A^d(A) \longrightarrow H_A^d(M)
For x∈Mx \in M, let us denote by fxf_x the mapping a↦axa \mapsto ax of A into M. For every u∈HAd(A)u \in H_A^d(A), one has
τd(M)(x⊗u)=HAd(fx)(u) \tau^d(M)(x \otimes u) = H_A^d(f_x)(u)
This follows in fact from remark 1 applied to the homomorphism fx:A→Mf_x : A \to M.

3. Grothendieck duality over a Macaulay ring

We still suppose that A is a local Macaulay ring, of dimension d. Let Ω\Omega be a dualizing A-module. The A-module HAd(Ω)H_A^d(\Omega) is a Matlis module (§ 8, No. 5, example 6); in accordance with the notation of § 8, we shall put D(M)=HomA(M,HAd(Ω))D(M) = \mathrm{Hom}_A(M, H_A^d(\Omega)) for every A-module M.

Consider the isomorphism τd(Ω):Ω⊗HAd(A)→HAd(Ω)\tau^d(\Omega) : \Omega \otimes H_A^d(A) \to H_A^d(\Omega) (No. 2). From it one deduces a homomorphism ω:HAd(A)→HomA(Ω,HAd(Ω))\omega : H_A^d(A) \to \mathrm{Hom}_A(\Omega, H_A^d(\Omega)) which associates to an element uu of HAd(A)H_A^d(A) the homomorphism x↦HAd(fx)(u)x \mapsto H_A^d(f_x)(u) (remark 3 above).

Proposition 2

Let A be a local Macaulay ring, of dimension d*, and let Ω\Omega be a dualizing A-module. The homomorphism ω:HAd(A)→D(Ω)\omega : H_A^d(A) \to D(\Omega) is bijective.

The homomorphism ω\omega is the limit of the inductive system of mappings (ωα)α∈Dcs(\omega_\alpha)_{\alpha \in \mathscr{D}_{cs}}, where

ωα:ExtAd(A/α,A)⟶HomA(Ω,ExtAd(A/α,Ω)) \omega_\alpha : \mathrm{Ext}_A^d(A/\alpha, A) \longrightarrow \mathrm{Hom}_A(\Omega, \mathrm{Ext}_A^d(A/\alpha, \Omega))

associates to an element uu of ExtAd(A/α,A)\mathrm{Ext}_A^d(A/\alpha, A) the mapping x↦fx∘ux \mapsto f_x \circ u (A, X, p. 114). It is therefore enough to prove that each of the mappings ωα\omega_\alpha is bijective.

Let α\alpha be an ideal of Dcs\mathscr{D}_{cs}, generated by a completely secant sequence x=(x1,…,xd)x = (x_1, \ldots, x_d) for AA. The Koszul complex K∙(x,A)K^\bullet(x, A) furnishes a projective resolution of A/αA/\alpha; for every AA-module MM, the AA-module Hd(HomgrA(K∙(x,A),M))H^d(\mathrm{Homgr}_A(K^\bullet(x, A), M)) identifies canonically with M/αMM/\alpha M (A, X, p. 155). From this one deduces an isomorphism (A, X, p. 100)

φM:M/αM⟶ExtAd(A/α,M). \varphi_M : M/\alpha M \longrightarrow \mathrm{Ext}_A^d(A/\alpha, M) .
Let x∈Ωx \in \Omega. Taking account of loc. cit., p. 103, prop. 2, one has a commutative diagram

A/α→φAExtAd(A/α,A)↓fx↓Ext(1A/α,fx)Ω/αΩ→φΩExtAd(A/α,Ω) \begin{array}{ccc} A/\alpha & \xrightarrow{\varphi_A} & \mathrm{Ext}_A^d(A/\alpha, A) \\ \downarrow f_x & & \downarrow \mathrm{Ext}(1_{A/\alpha}, f_x) \\ \Omega/\alpha \Omega & \xrightarrow{\varphi_\Omega} & \mathrm{Ext}_A^d(A/\alpha, \Omega) \end{array}

where fxf_x is the homomorphism deduced from fxf_x by passing to quotients. It follows that if for every AA-module MM one identifies ExtAd(A/α,M)\mathrm{Ext}_A^d(A/\alpha, M) with M/αMM/\alpha M by means of φM\varphi_M, the homomorphism ωα\omega_\alpha is identified with the AA-linear mapping of A/αA/\alpha into HomA(Ω,Ω/αΩ)\mathrm{Hom}_A(\Omega, \Omega/\alpha \Omega) which sends 1 to the canonical surjection, that is again with the canonical mapping A/α⟶EndA/α(Ω/αΩ)A/\alpha \longrightarrow \mathrm{End}_{A/\alpha}(\Omega/\alpha \Omega). But since the A/αA/\alpha-module Ω/αΩ\Omega/\alpha \Omega is dualizing (§ 9, No. 2, prop. 4), this mapping is bijective (§ 9, No. 4, prop. 6), which proves the proposition.

Let us identify the Matlis bidual D(D(Ω))D(D(\Omega)) with Ω^\widehat{\Omega} by the isomorphism α^Ω\widehat{\alpha}_\Omega (§ 8, No. 3, th. 2, b)).

Corollary

The homomorphism* D(ω):Ω^→D(HAd(A))D(\omega) : \widehat{\Omega} \to D(H^d_A(A)) is an isomorphism.

Let MM be an AA-module, and ii an integer. Consider the canonical homomorphisms (§ 8, No. 7)

ρd−i(M,Ω):Tord−iA(M,D(Ω))⟶D(ExtAd−i(M,Ω)) \rho_{d-i}(M, \Omega) : \mathrm{Tor}^A_{d-i}(M, D(\Omega)) \longrightarrow D(\mathrm{Ext}_A^{d-i}(M, \Omega))
θd−i(M,HAd(A)):ExtAd−i(M,D(HAd(A)))⟶D(Tord−iA(M,HAd(A))). \theta^{d-i}(M, H^d_A(A)) : \mathrm{Ext}_A^{d-i}(M, D(H^d_A(A))) \longrightarrow D(\mathrm{Tor}^A_{d-i}(M, H^d_A(A))) .

With the aid of the isomorphisms ω:HAd(A)→D(Ω)\omega : H^d_A(A) \to D(\Omega), D(ω):Ω^→D(HAd(A))D(\omega) : \widehat{\Omega} \to D(H^d_A(A)) (cor. 1 of prop. 2) and τi(M):Tord−iA(M,HAd(A))→HAi(M)\tau^i(M) : \mathrm{Tor}^A_{d-i}(M, H^d_A(A)) \to H^i_A(M) (No. 2), one deduces from them *canonical homomorphisms of AA-modules

γi(M):HAi(M)⟶D(ExtAd−i(M,Ω)) \gamma^i(M) : H^i_A(M) \longrightarrow D(\mathrm{Ext}_A^{d-i}(M, \Omega))
δi(M):ExtAd−i(M,Ω^)⟶D(HAi(M)). \delta^i(M) : \mathrm{Ext}_A^{d-i}(M, \widehat{\Omega}) \longrightarrow D(H^i_A(M)) .

Theorem 1 (Grothendieck Duality)

Let AA be a local Macaulay ring, of dimension dd, and let Ω\Omega be a dualizing AA-module.

a) The AA-module HAd(Ω)H_A^d(\Omega) is a Matlis module; for every AA-module PP, let D(P)D(P) denote the Matlis dual HomA(P,HAd(Ω))\mathrm{Hom}_A(P, H_A^d(\Omega)).

b) For every finitely generated AA-module MM and every integer ii, the canonical homomorphism

γi(M):HAi(M)⟶D(ExtAd−i(M,Ω)) \gamma^i(M) : H_A^i(M) \longrightarrow D(\mathrm{Ext}_A^{d-i}(M, \Omega))
is an isomorphism of artinian AA-modules.

c) For every AA-module MM and every integer ii, the canonical homomorphism

δi(M):ExtAd−i(M,Ω^)⟶D(HAi(M)) \delta^i(M) : \mathrm{Ext}_A^{d-i}(M, \widehat{\Omega}) \longrightarrow D(H_A^i(M))
is an isomorphism of A^\widehat{A}-modules.

This follows from prop. 6 of § 8, No. 7.

Corollary

Let AA be a local Macaulay ring, and MM a nonzero finitely generated AA-module, of dimension ee. The AA-module HAe(M)H_A^e(M) is nonzero.

By Remark 2 of No. 1, one may suppose that the local ring AA is complete. In this case AA has a dualizing module Ω\Omega (§9§ 9, No. 3, Corollary 3 of Proposition 6); if HAe(M)H_A^e(M) is zero, the same is true of its Matlis dual ExtAd−e(M,Ω)\mathrm{Ext}_A^{d-e}(M, \Omega), which contradicts Proposition 3, b) of §9§ 9, No. 1.

Remark 1

When the AA-module MM is finitely generated, the A^\widehat{A}-module ExtAd−i(M,Ω^)\mathrm{Ext}_A^{d-i}(M, \widehat{\Omega}) is identified with A^⊗AExtAd−i(M,Ω)\widehat{A} \otimes_A \mathrm{Ext}_A^{d-i}(M, \Omega) (AA, X, p. 108, Proposition 7, c)), and δi(M)\delta^i(M) can also be obtained by composing D(γi(M))D(\gamma^i(M)) with the biduality isomorphism.

Remark 2

Let u:M→M′u : M \to M' be a homomorphism of AA-modules. By Remark 1 of No. 2 and that of §8§ 8, No. 7, the following diagrams are commutative:

HAi(M)→γi(M)D(ExtAd−i(M,Ω))↓HAi(u)↓D(Ext(u,1))HAi(M′)→γi(M′)D(ExtAd−i(M′,Ω)) \begin{array}{ccc} H_A^i(M) & \xrightarrow{\gamma^i(M)} & D(\mathrm{Ext}_A^{d-i}(M, \Omega)) \\ \downarrow H_A^i(u) & & \downarrow D(\mathrm{Ext}(u,1)) \\ H_A^i(M') & \xrightarrow{\gamma^i(M')} & D(\mathrm{Ext}_A^{d-i}(M', \Omega)) \end{array}
ExtAd−i(M′,Ω^)→δi(M′)D(HAi(M′))↓Ext(u,1)↓D(HAi(u))ExtAd−i(M,Ω^)→δi(M)D(HAi(M)) \begin{array}{ccc} \mathrm{Ext}_A^{d-i}(M', \widehat{\Omega}) & \xrightarrow{\delta^i(M')} & D(H_A^i(M')) \\ \downarrow \mathrm{Ext}(u,1) & & \downarrow D(H_A^i(u)) \\ \mathrm{Ext}_A^{d-i}(M, \widehat{\Omega}) & \xrightarrow{\delta^i(M)} & D(H_A^i(M)) \end{array}

Remark 3

Let

(E)(\mathscr{E})
0→M′→M→M′′→00 \to M' \to M \to M'' \to 0

be an exact sequence of AA-modules. By Remark 2 of No. 2 and that of § 8, No. 7, the following diagrams are commutative:

HAi−1(M′′)→γi−1(M′′)D(ExtAd−i+1(M′′,Ω))↓↓(−1)d−i+1D(δdi(E,Ω))HAi(M′)→γi(M′)D(ExtAd−i(M′,Ω)) \begin{array}{ccc} H_A^{i-1}(M'') & \xrightarrow{\gamma^{i-1}(M'')} & D(\mathrm{Ext}_A^{d-i+1}(M'', \Omega)) \\ \downarrow & & \downarrow (-1)^{d-i+1}D(\delta^d i(\mathscr{E}, \Omega)) \\ H_A^i(M') & \xrightarrow{\gamma^i(M')} & D(\mathrm{Ext}_A^{d-i}(M', \Omega)) \end{array}
ExtAd−i(M′,Ω^)→δi(M′)D(HAi(M′))↓↓(−1)d−i+1D(∂i−1(E))ExtAd−i+1(M′′,Ω^)→δi−1(M′′)D(HAi−1(M′′)) \begin{array}{ccc} \mathrm{Ext}_A^{d-i}(M', \widehat{\Omega}) & \xrightarrow{\delta^i(M')} & D(H_A^i(M')) \\ \downarrow & & \downarrow (-1)^{d-i+1}D(\partial^{i-1}(\mathscr{E})) \\ \mathrm{Ext}_A^{d-i+1}(M'', \widehat{\Omega}) & \xrightarrow{\delta^{i-1}(M'')} & D(H_A^{i-1}(M'')) \end{array}

Example

Let AA be an integral noetherian local ring of dimension 1; let KK denote its field of fractions. Let Ω\Omega be a dualizing AA-module, and let MM be a finitely generated AA-module. The AA-modules HA0(M)H_A^0(M) and HA1(M)H_A^1(M) are canonically identified with T(M)T(M) and (K/A)⊗AM(K/A) \otimes_A M (No. 1, Example 5). With these identifications, the duality isomorphisms

γ0(M):T(M)⟶D(ExtA1(M,Ω)),γ1(M):(K/A)⊗AM⟶D(HomA(M,Ω))\gamma^0(M) : T(M) \longrightarrow D(\mathrm{Ext}_A^1(M, \Omega)) \quad , \quad \gamma^1(M) : (K/A) \otimes_A M \longrightarrow D(\mathrm{Hom}_A(M, \Omega))

(Theorem 1) are none other than the isomorphisms defined in § 9, No. 6.

Exercises

Exercises

See the exercises for § 10.

Machine translation of the English, by gpt-5.4, against glossary version 34. Not checked by a person.