Chapter 9

Arun Ram
Department of Mathematics and Statistics
University of Melbourne
Parkville, VIC 3010 Australia
aram@unimelb.edu.au

Last update: 11 June 2013

Cohomology of Affine schemes

Let A be any commutative ring with identity element, and let X=Spec(A). We recall that, for any f∈A, the 'basic open set' D(f) is the set of all x∈X such that f(x)≠0, i.e. such that f∉jx. Let M be any A-module; then we can form the module of fractions Mf, whose elements are all fractions of the form m/fn (m∈M n an integer ≥0). Mf is a module over the ring Af. We may then consider the presheaf D(f)→Mf, defined on the basis ℬ=D(f)f∈A of X. We denote by M∼ the presheaf on X which it determines. Since each Mf is an Af-module, M∼ is an A∼-Module.

Proposition (9.1). M∼ is a sheaf, and hence Γ(D(f),M∼)=Mf for all f∈A. In particular Γ(X,M∼)=M.

Proof.

Copy of the proof of (5.1) (ii).

□

Since formation of modules of fractions preserves exactness, it follows that the functor M↦M∼ (from A-modules to A∼-Modules) is exact. Moreover,

Corollary (9.2). If M,N are A-modules, then HomA(M,N)≅ HomA∼(M∼,N∼).

Proof.

φ:M→N gives rise to φf:Mf→Nf for each f∈A, hence to φ∼:M∼→N∼. Hence we have a homomorphism HomA(M,N)→ HomA∼(M∼,N∼). Conversely, given u:M∼→N∼ we have u(X):M∼(X)→N∼(X), i.e. by (9.1) a homomorphism M→N. Hence a map HomA∼(M∼,N∼)→ HomA(M,N). Verify that the two maps so defined are inverses of each other.

□

Theorem (9.3). Let ℱ be an A-Module. Then the following are equivalent:

(a) ℱ≅M∼ for some A-module M;
(b) there exists a finite open covering of X by basic open sets D(fi) such that ℱ|D(fi)≅M∼i for some Afi-module Mi and each index i;
(c) ℱ is quasi-coherent;
(d) ℱ satisfies the following two conditions:
(d1) for each g∈A and each s∈ℱ(D(g)) there exists an integer n≥0 such that gns can be extended to a global section of ℱ (i.e. an element of ℱ(X));
(d2) for each g∈A and each t∈ℱ(X) such that t|D(g)=0, there exists n≥0 such that gnt=0.

Proof according to the scheme
(a) ⟹ (b) ⟺ (c). ⟺ ⟸ (d)

(a) ⇒ (b). Take the covering of X consisting of the single set D(1)=X.

(b) ⇒ (c). Since quasi-coherence is a local property, it is enough to prove (a) ⇒ (c). We have an exact sequence A(I)→ A(J)→M→0, where A(I),A(J) are direct sums of copies of A; hence, since M↦M∼ is an exact functor, an exact functor, an exact sequence A∼(I)→ A∼(J)→ M∼→0. Hence M∼ is quasi-coherent.

(c) ⇒ (b). Each x∈X has a neighbourhood D(f) over which ℱ|D(f) is the cokernel of a homomorphism A∼(I)|D(f)→ A∼(J)|D(f), i.e. a homomorphism A∼f(I)→A∼f(J). Hence by (9.2) ℱ|D(f)=N∼ where N=coker(Af(I)→Af(J)). Since X is quasi-compact, (b) is proved.

(a) ⇒ (d). If g∈A and s∈ℱ(D(g))=Mg, then s=m/gn for some m∈M and some integer n≥0, hence sgn=m/1= image of m in Mg, i.e. sgn is the image of an element of M=ℱ(X). If g∈A, t∈M and t/1=0 in Mg, then tgn=0 for some integer n≥0, from the basic properties of modules of fractions.

(b) ⇒ (d). We have to show that if each ℱ|D(fi) satisfies (d), then so does ℱ. Take (d2) first. We have then g∈A, t∈ℱ(X) and t|D(g)=0. Then t|D(gfi)=0 (since D(gfi)=D(f)∩D(fi)); hence by (d2) applied to ℱ|D(fi) there exists an integer ni≥0 such that (fig)nit|D(fi)=0, i.e. (fig)nit=0 in Mi; now fi is a unit in Afi, hence gnit=0 in Mi. Let n be the largest of the ni, then we have gnt=0 in each ℱ|D(fi), hence gnt is the zero section of ℱ.

To prove (d1): take g∈A and s∈ℱ(D(g)). By applying (d1) to ℱ|D(fi), there exists an integer ni≥0 and an element si′∈ℱ(D(fi)) which extends (fig)nis|D(fig). Since fi is a unit in Afi, there exists si′=finisi, and si extends gnis|D(fig); and we may take all the ni to be equal, say ni=n. By construction, si-sj restricted to D(fifjg) is zero; now since ℱ|D(fi)=M∼i, it follows that each ℱ|D(fi)∩D(fj) satisfies (a) and therefore (d), hence by (d2) applied to ℱ|D(fi)∩D(fj) there exist integers mij≥0 such that (fifjg)mij(si-sj) restricted to D(fi)∩D(fj)=D(fifj) is zero; but fifj is a unit in ℱ(D(fifj)), hence gm(si-sj) restricted to D(fi)∩D(fj) is zero, where m=max(mij). Hence the gmsi∈Γ(D(fi),ℱ) are all of them restrictions of a global section s′ of ℱ. This section s′ is an extension of gm+ns, hence (d1) is proved.

(d) ⇒ (a). Let M=ℱ(X)=Γ(X,ℱ). We shall define a homomorphism u:M∼→ℱ and show that it is an isomorphism. For this we must define uf:Mf→ℱ(D(f)) for each f∈A, satisfying the usual compatibility conditions. Start with the restriction homomorphism ℱ(X)→ℱ(D(f)), i.e. M→ℱ(D(f)). Since f is a unit in Af, this homomorphism factorizes through Mf: M⟶Mf⟶uf ℱ(D(f)). This defines uf. We shall show that (d1) implies uf surjective, and (d2) implies uf injective.

Let s be any element of ℱ(D(f)). Then by (d1) fns lifts to a global section of ℱ, for some integer n≥0, i.e. fns is in the image of M, hence is in uf(Mf). Hence, as f is a unit in Af, we have s∈uf(Mf) and thus uf is surjective.

If z/fn∈Mf is such that uf(z/fn)=0, then uf(z/1)=0 and therefore the restriction of z (∈ℱ(X)) to D(f) is zero; hence by (d2) there exists an integer m≥0 such that zfm=0; hence z/fn=0 in Mf, hence uf is injective.

□

Corollary (9.4). Γ is exact on quasi-coherent Modules over an affine scheme.

Proof.

Let ℱ→𝒢→ℋ be an exact sequence of quasi-coherent A∼-Modules. By (9.2) and (9.3) this sequence is of the form M∼→u∼N→v∼P∼ (M=ℱ(X), etc.). If Q=Im(u), R=Ker(v) then Q∼=R∼ (since the functor M↦M∼ is exact), hence Q=Q∼(X)=R∼(X)=R. Hence the sequence M→N→P is exact, i.e. the sequence Γ(X,ℱ)→ Γ(X,𝒢)→ Γ(X,ℋ) is exact.

□

Theorem (9.5). Let A be a Noetherian ring, ℱ an A∼-Module. Then the following are equivalent:

(i) ℱ is coherent;
(ii) ℱ is of finite type and quasi-coherent;
(iii) ℱ≅M∼ for some finitely-generated A-module M.

Proof.

(i) ⇒ (ii) is always true (from the definitions).

(ii) ⇒ (iii): By (9.3) we have ℱ≅M∼ for some A-module M. Since ℱ is of finite type and X is quasi-compact, there exists a finite covering of X by basic open sets D(fi), and exact sequences A∼pi→ℱ→0 (over D(fi)), i.e. exact sequences A∼fipi→ M∼fi→0; hence, by (9.4), exact sequences Afipi→ Mfi→0. Thus each Mfi is a finitely-generated Afi-module, generated say by tij/1 (1≤j≤pi,tij∈M). Let N be the submodule of M generated by all the tij. If z∈M, then z/1∈Mfi is of the form ∑j(tij/1)· (aij/fimj), hence zfim∈N for all indices i and some integer m>0. Since the D(fi) cover X, the fim generate the unit ideal, i.e. we have an equation of the form ∑igifim=1, where gi∈A. Hence z=∑izfimgi∈N, consequently M=N and therefore M is finitely generated.

(iii) ⇒ (i). Suppose ℱ=M∼ where M is a finitely generated A-module. Then we have an exact sequence of the form Ap→M→0 for some integer p≥0, hence A∼→M∼→0; thus M∼ is of finite type. It remains to show that if A∼p→M∼ over some open set (which we may take to be D(f) for some f∈A), then the kernel is of finite type. We have a homomorphism A∼fp→M∼f, hence a homomorphism Afp→Mf by (9.2); now Af is Noetherian (since A is), hence the kernel is finitely generated. This completes the proof.

Remark. The Noetherian assumption intervenes only in the proof of (iii) ⇒ (i).

□

Corollary (9.6). If A is Noetherian, A∼ is a coherent sheaf of rings.

Proposition (9.7). Let ℱ be a quasi-coherent A∼-Module, and let 𝓊 be a covering of X=Spec(A) by basic open sets D(fi). Then Hp(𝓊,ℱ)=0 for all p>0 (and of course H0(𝓊,ℱ)= ℱ(X)).

Proof.

By (9.3) we have ℱ=M∼, where M=ℱ(X) is an A-module. Consider the Čech resolution of M∼ (Chapter 8):

0⟶M∼⟶C0 (𝓊,M∼)⟶ C′(𝓊,M∼) ⟶…

whose sections over X form the Čech complex 0→M→C0(𝓊,M∼) →C1(𝓊,M∼) →…. Recall that Cq(𝓊,M∼) is the sheaf associated with the pre sheaf

D(g)⟼∏σM∼ (Uσ∩D(g));

now if σ=(i0,…,iq) we have Uσ∩D(g)=D(fi0)∩…∩D(fiq)∩D(g)=D(fi0…fiqg)=D(fσg) say; hence Cq(𝓊,M∼) is the sheaf associated with the presheaf D(g)↦ ∏σMfσg= (∏σMfσ)g, so that Cq(𝓊,M∼)=(∏σMfσ)∼. Hence, by (9.3), the sheaf Cq(𝓊,M∼) is quasi-coherent; now Γ is exact on quasi-coherent sheaves (9.4), hence the Čech complex is exact, i.e. Hp(𝓊,ℱ)=0 for all p>0.

□

Theorem (9.8). If X is an affine scheme and ℱ a quasi-coherent sheaf on X, then Hp(X,ℱ)=0 for all p>0.

Proof.

Since finite basic open coverings are cofinal in the class of all open coverings of X, it follows from (9.7) that Hˇp(X,ℱ)=0 for all p>0. Hence for any basic open set U=D(f) we have Hˇq(U,ℱ|U)=0 for all q>0 (since U is an affine open set and ℱ|U is quasi-coherent). Hence by Cartan's criterion (8.3) we have Hˇp(X,ℱ)= Hp(X,ℱ) for all p>0. Hence Hp(X,ℱ)=0 for all p>0.

□

Remark. There is another proof, due to Chevalley, of (9.8) avoiding the use of (8.3) (which we didn't prove). Let 0→ℱ→ℐ0→ℐ1 →… be an injective resolution of a quasi-coherent sheaf ℱ on X. Then we have short exact sequences

(Ep):0⟶ 𝒢p⟶ℐp⟶ 𝒢p+1⟶0

where 𝒢0=ℱ and 𝒢p=Im(ℐp-1→ℐp) for p>0.

Lemma (9.9). Let f∈A and let 𝓊 be any finite covering of D(f) by basic open sets. Then Hq(𝓊,𝒢p|D(f))=0 for all p≥0 and all q>0.

Proof.

Proof by induction on p. True for p=0 by (9.7). Let p≥0 and assume (9.9) true for this value of p (and all q>0). Then H1(𝓊,𝒢p|D(f))=0 for any finite covering of D(f) by basic open sets. Since such open coverings of D(f) are cofinal in the class of all open coverings of D(f) it follows that Hˇ1(D(f),𝒢p|D(f))=0 and therefore that H1(D(f),𝒢p|D(f))=0 (since H1=Hˇ1 always). Hence, from the exact cohomology sequence of (Ep), we have an exact sequence

0⟶𝒢p(D(f))⟶ ℐp(D(f))⟶ 𝒢p+1(D(f)) ⟶0.

Since this sequence is exact for. every f∈A, it follows that the sequence of Čech complexes

(✶) 0⟶C• (𝓊,ℱp|D(f)) ⟶C• (𝓊,ℐp|D(f)) ⟶C(𝓊,𝒢p+1|D(f)) ⟶0

is exact. Now ℐp is injective, hence its restriction to the open set D(f) is injective and therefore the complex C•(𝓊,ℐp|D(f)) is acyclic; consequently, from the cohomology exact sequence of (✶), we get

Hq(𝓊,𝒢p+1|D(f)) ≅Hq+1 (𝓊,𝒢p|D(f)) (q>0)

and the term on the right is zero by the inductive hypothesis.

□

Taking f=1, q=1 in (9.9), we have H1(𝓊,𝒢p)=0 for all p≥0, hence Hˇ1(X,𝒢p)=0, hence H1(X,𝒢p)=0. But from the exact sequences (Ep) we get (since each ℐp is injective)

Hp(X,ℱ)= Hp(X,𝒢0) ≅Hp-1 (X,𝒢1)≅…≅ H1(X,𝒢p-1) =0(p>0).

Theorem (9.10). If (X,𝒪X) is a scheme and ℱ is a quasi-coherent 𝒪X-Module, then Hq(X,ℱ)≅Hq(𝓊,ℱ) for any covering 𝓊 of X by affine open sets.

Proof.

Let 𝓊=(Ui)i∈I be an affine open covering of X. Since, X is a scheme, each Uσ=Ui0∩…∩Uiq is affine and hence by (9.8) Hp(Uσ,ℱ|Uσ)=0 for all σ and all p>0. Hence by the comparison theorem (8.2) we have Hp(X,ℱ)≅ Hp(𝓊,ℱ) for all p≥0.

□

Corollary (9.11). Hp(X,ℱ)≅ Hˇp(X,ℱ) under the hypotheses of (9.10).

There is a converse of (9.8):

Theorem (9.12). (Serre's criterion.) Let X be either a quasi-compact scheme or a prescheme whose underlying space is Noetherian. If H1(X,ℱ)=0 for every quasi-coherent 𝒪X-Module ℱ (or even only for every quasi-coherent Ideal ℱ of 𝒪X), then X is an affine scheme.

For the proof we refer to (E.G.A., II, 5.2.1). (9.8) and (9.12) show that the vanishing of the Hp(X,ℱ) for p>0 and ℱ quasi-coherent characterizes affine schemes.

Let X be a projective algebraic variety over an algebraically closed field k, and let ℱ be a coherent 𝒪X-Module, where 𝒪X is the sheaf of local rings on X. Serre proved that

(i) Hq(X,ℱ)=0 for q>dim X;
(ii) Hq(X,ℱ) is a finite-dimensional k-vector space for 0≤q≤dim X.

The proof of (i) is easy: by (9.10) (or rather its counterpart for algebraic varieties) it is enough to find a covering of X by d+1 affine open sets, where d=dim X, and this can be achieved by intersecting X by suitably chosen hyperplanes in the projective space P in which X is embedded. (ii) is proved by reducing to the case where X=P and calculating the Hq(P,ℱ) quite explicitly.

Grothendieck subsequently generalized this theorem, firstly to the case where X is complete (but not necessarily projective) and then to a statement about proper morphisms. If f:X→Y is a morphism of algebraic varieties, then f✶ (Chapter 7) is a left-exact functor from 𝒪X-Modules to 𝒪Y-Modules, hence has right derived functors Rpf✶(p≥0). Explicitly, if ℱ is an 𝒪X-Module, Rpf✶(ℱ) is the sheaf on Y associated to the presheaf U↦Hp(f-1(U),ℱ) (U open in Y). Then:

If X,Y are algebraic varieties over k,f:X→Y a proper morphism, ℱ a coherent 𝒪X-Module, then the 'higher direct images' Rpf✶(ℱ) are coherent 𝒪Y-Modules. (The statement for a complete variety X is obtained by taking Y to consist of a single point.)

Finally, this theorem generalizes to the case of a proper morphism of preschemes:

Let X,Y be preschemes, Y locally Noetherian (this means that Y can be covered by affine open sets each of which is the scheme of a Noetherian ring). If f:X→Y is a proper morphism and ℱ a coherent 𝒪X-Module, then the Rpf✶(ℱ) are coherent 𝒪Y-Modules (E.G.A., III, 3.2.1).

Notes and References

This is a typed excerpt of the book "Algebraic Geometry: Introduction to Schemes - I.G. Macdonald".

page history