Chapter 7

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

Last update: 4 June 2013

Operations on sheaves, quasi-coherent and coherent sheaves

Let (X,𝒪) be a ringed space. An 𝒪-Module (note the capital M) is a sheaf ℱ of abelian groups such that, for each open set U in X, the group ℱ(U) carries a structure of an 𝒪(U)-module, these structures being compatible with the restriction homomorphisms: explicitly, if U⊆V are open sets in X, then the restriction φ:ℱ(U)→ℱ(V) is compatible with the restriction ρ:𝒪(U)→𝒪(V), that is to say, if f∈ℱ(U) and a∈𝒪(U) then φ(af)=ρ(a)·φ(f). Then each stalk ℱx has a natural 𝒪X-module structure, defined as follows: if ax∈𝒪X, fx∈ℱx, say ax is the image of a∈𝒪(U), fx the image of f∈ℱ(U) for some sufficiently small open neighbourhood U of x; then ax·fx is the image of af in ℱx.

In particular, 𝒪 itself is an 𝒪-Module.

Most of the concepts of module theory have their counterparts for Modules: -

(i) An 𝒪-Module homomorphism φ:ℱ→𝒢 is a sheaf homomorphism (i.e. a family of homomorphisms φ(U):ℱ(U)→𝒢(U), commuting with the restrictions) such that each φ(U) is an 𝒪(𝔘)-module homomorphism. Then each φx:ℱx→𝒢x is an 𝒪X-module homomorphism.

(ii) Sub-Modules. A subsheaf ℱ′ of an 𝒪-Module ℱ is a sub-Module of ℱ if, for each open set U in X, ℱ′(U) is a sub-𝒪(U)-module of ℱ(U). Then each ℱx′ is a sub-𝒪X-module of ℱx, and the embedding ℱ′→ℱ is an 𝒪-Module homomorphism. In particular, a sub-Module of 𝒪 is called an Ideal (with a capital I).

(iii) Quotient Modules. Let ℱ be an 𝒪-Module, ℱ′ a sub-Module of ℱ. For each open set U in X, form ℱ(U)/ℱ′(U). U↦ℱ(U)/ℱ′(U), with the induced restriction homomorphisms, is a presheaf, but not necessarily a sheaf. So we form the sheaf associated with this presheaf: this is the quotient Module ℱ′′=ℱ/ℱ′. Since lim→ is exact, we have ℱx′′=ℱx/ℱx′.

(iv) Kernel. Let φ:ℱ→𝒢 be an 𝒪-Module homomorphism. For each open set U in X let ℱ′(U) be the kernel of φ(U):ℱ(U)→𝒢(U). Then U↦ℱ′(U) is a sheaf ℱ′, called the kernel of φ. Clearly ℱ′ is an 𝒪-Module. We have ℱx′=Ker(φx) for all x∈X.

(v) Image. For each open set U in X we can form Im(φ(U)), which is a submodule of 𝒢(U). U↦im(φ(U)) is a presheaf (not necessarily a sheaf). Let ℋ be the sheaf associated with this presheaf. Then ℋ is a subsheaf of 𝒢, called the image of 𝒪. Again by the exactness of lim→ we have ℋx=Im(φx). Also ℋ is isomorphic to the quotient ℱ/ℱ′, where ℱ′ is the kernel of φ.

(vi) Cokernel. The cokernel of φ is 𝒢/ℋ. We have the formulas

(Ker(φ))x≅ Ker(φx); (Im(φ))x≅ Im(φx); (Coker(φ))x≅ Coker(φx).

The class of 𝒪-Modules is an abelian category. Exact sequences are defined in the usual way.

Lemma (7.1). A sequence ℱ→φ𝒢→ψℋ is exact if and only if ℱx→φx𝒢x→ψxℋx is exact for all x∈X.

Proof.

ℱ→𝒢→ℋ is exact ⇔Im(φ)= Ker(ψ)⇔ (Im(φ))x= (Ker(ψ))x for all x∈X⇔Im(φx)=Ker(ψx) for all x∈X⇔ℱx→𝒢x→ℋx is exact for all x∈X.

□

Lemma (7.2). The "section functor" Γ(ℱ) (=Γ(X,ℱ)=ℱ(X)) is left exact: if 0→ℱ→𝒢→ℋ is exact, then 0→Γ(ℱ)→Γ(𝒢)→Γ(ℋ) is exact. This follows from (iii) above.

(vii) Direct sum. Let (ℱi)i∈I be any family of 𝒪-Modules. Their direct sum ℱ=⨁i∈Iℱi is the sheaf U↦⨁i∈Iℱi(U). If each ℱi is equal to 𝒪, we write 𝒪(I) for the direct sum. In particular, if I is finite and has n elements, we write 𝒪n for the direct sum of n copies of 𝒪.

(viii) Tensor product. If ℱ,𝒢 are 𝒪-Modules, their tensor product ℱ⊗𝒪𝒢 is defined to be the sheaf associated with the presheaf U↦ℱ(U)⊗𝒪(U)𝒢(U). Since ⊗ commutes with lim→, we have (ℱ⊗𝒪𝒢)x≅ℱx⊗𝒪x𝒢x. This tensor product has all the usual properties: it is commutative, associative, distributive over ⨁, and is right exact in each variable (look at the stalks and use (7.1)). Also ℱ⊗𝒪𝒪≅ℱ.

(ix) Global Hom. Hom𝒪(ℱ,𝒢) is the group of all 𝒪-Module homomorphisms φ:ℱ→𝒢. It has a natural 𝒪(X)-module structure: if φ:ℱ→𝒢 and s∈𝒪(X), define sφ:ℱ→𝒢 by (sφ)(U)=s|U·φ(U).

(x) Sheaf ℋℴ𝓂. The presheaf U↦Hom𝒪|U is easily checked to be a sheaf, denoted by ℋℴ𝓂𝒪(ℱ,𝒢). Thus Γ(X,ℋℴ𝓂𝒪(ℱ,𝒢)) =Hom𝒪(ℱ,𝒢). ℋℴ𝓂𝒪(ℱ,𝒢) has a natural 𝒪-Module structure. Both Hom and ℋℴ𝓂 are left exact in each variable (contravariant in the 1st variable, covariant in the 2nd). We have ℋℴ𝓂𝒪(𝒪,𝒢)≅𝒢.

Let fx∈(ℋℴ𝓂𝒪(ℱ,𝒢))x. Then fx is represented by say f:ℱ|U→𝒢|U, which gives rise to a homomorphism ℱx→𝒢x, i.e. an element of Hom𝒪X(ℱx,𝒢x). Hence we have 𝒪x-module homomorphism

(ℋℴ𝓂𝒪(ℱ,𝒢))x →Hom𝒪x(ℱx,𝒢x)

which in general is neither injective nor surjective (but see (7.9)).

(xi) Direct image. Let Ψ= (ψ,θ): (X,𝒪X)→ (Y,𝒪Y) be a morphism of ringed spaces. If ℱ is an 𝒪X-Module (thus a sheaf on X), we define its direct image Ψ*ℱ, which is an 𝒪Y-Module (thus a sheaf on Y) as follows: Ψ*ℱ(V)=ℱ(ψ-1(V)) for each open set V in Y; ℱ(ψ-1(V)) is an 𝒪X(ψ-1(V))-module, hence an 𝒪Y(V)-module via the homomorphism θ(V)→𝒪Y(V)→𝒪X(ψ-1(V)).

Ψ* is a left-exact functor from 𝒪X-Modules to 𝒪Y-Modules. For the section functor Γ is left exact by (7.2). Hence if 0→ℱ′→ℱ→ℱ′′ is an exact sequence of 𝒪X-Modules, then 𝒪→ Γ(ψ-1(V),ℱ′)→ Γ(ψ-1(V),ℱ)→ Γ(ψ-1(V),ℱ′′) is exact for each open V⊆Y; hence 0→Ψ*ℱ′→ Ψ*ℱ→ Ψ*ℱ′′ is exact.

In particular, if Y is the ringed space consisting of a single point and the ring 𝒪(X), then Ψ*(ℱ)=ℱ(X)=Γ(X,ℱ). Thus Ψ*(ℱ)=ℱ(X)=Γ(X,ℱ). Thus Ψ* is a 'relativization' of the section functor Γ.

Quasi-coherent and coherent sheaves

If ℱ is an 𝒪-Module, a homomorphism u:𝒪→ℱ gives rise to s=u(X)(I)∈ℱ(X), i.e. to a global section of ℱ. Conversely, given s∈ℱ(X) we may reconstruct u: if U is open in X and t∈𝒪(U), then u(t)=t·(s|U). Hence we have a one-one correspondence between 𝒪-Module homomorphisms 𝒪→ℱ and global sections of ℱ, hence between 𝒪-Module homomorphisms u:𝒪(I)→ℱ and families (si)i∈I of global sections of ℱ, where I is any index set. u is an epimorphism if and only if each ℱx is generated (as an 𝒪x-module) by the (si)x (for u is an epimorphism if and only if each ux:(𝒪x)(I)→ℱx is an epimorphism, by (7.1)).

ℱ is said to be quasi-coherent if each x∈X has an open neighbourhood U such that ℱ|U is the cokernel of a homomorphism 𝒪(I)|U→𝒪(J)|U, where the index sets I,J are of arbitrary cardinal (and depend on U). Clearly 𝒪 itself is quasi-coherent as an 𝒪-Module.

Thus ℱ is quasi-coherent if and only if ℱ is locally generated by its sections and if the 'sheaf of relations' is locally generated by its sections.

An 𝒪-module ℱ is of finite type if each x∈X has an open neighbourhood U such that ℱ|U is generated by a finite set of sections of ℱ over U, i.e. if there exists an epimorphism 𝒪p|U→ℱ|U for some integer p>0. If ℱ,G are of finite type, then so are ℱ⊕𝒢 and ℱ⊗𝒪𝒢 (the latter because ⊗ is right exact). If 𝒢 is of finite type and 𝒢 is a homomorphic image of ℱ, then 𝒢 is of finite type.

ℱ is said to be coherent if

(i) ℱ is of finite type;
(ii) for each open set U in X and each homomorphism φ:𝒪n|U→ℱ|U (n a positive integer), Ker(φ) is of finite type.

Clearly a coherent sheaf is quasi-coherent. All these properties (quasi-coherence, finite type, coherence) are local with respect to the base-space X.

We shall use the following notation. If U is an open set in X, the phrase 'f:ℱ→𝒢 (over U)' shall mean f:ℱ|U→𝒢|U. Similarly for diagrams of sheaves and homomorphisms.

Lemma (7.3). If ℱ is a subsheaf of 𝒢, and 𝒢 is of finite type and 𝒢 is coherent, then ℱ is coherent.

Proof.

Let i:ℱ→𝒢 be the embedding. If we have f:𝒪n→ℱ (over U), then i∘f:𝒪n→𝒢 (over U); but 𝒢 is coherent, hence ker f=ker i∘f is of finite type.

□

Lemma (7.4). Let 𝒢,H be 𝒪-Modules. If we have a diagram

𝒢 ⟶h ℋ ⟶ 0 ↗f 𝒪p (over a neighbourhood of x∈X)

with the row exact, then there exists an 𝒪-Module homomorphism g:𝒪p→𝒢 (over a (smaller) neighbourhood of x), such that h∘g=f.

Proof.

The map f defines p sections si (1≤i≤p) belonging to ℋ(U) (U some open neighbourhood of x). Explicitly, f(U) maps 𝒪(U)p into ℋ(U), and si is the image of the ith generator ei of 𝒪(U)p. Since h is an epimorphism, there exist gi,x∈𝒢x such that hx(gi,x)=(si)x (1≤i≤p). Each gi,x is represented by say gi′∈ℋ(Ui); h(gi′) agrees with si at x, hence in some open neighbourhood of x, say Vi (⊂Ui∩U). Let V=V1∩…∩Vp, then the gi=gi′|V define g:𝒪p→𝒢 (over V), and we have (over V) h∘g(ei)=h(gi)=si|V=f(ei), hence h∘g=f.

□

Theorem (7.5). If 0→ℱ→f𝒢→gℋ→0 is an exact sequence of 𝒪-Modules on X, and if any two of ℱ,𝒢,ℋ are coherent, then so is the third.

Proof.

(1) 𝒢,ℋ coherent. By (7.3) it is enough to show that ℱ is of finite type. Let x∈X. Since 𝒢 is of finite type we have an epimorphism u:𝒪p→𝒢 (over some neighbourhood of x). Since ℋ is coherent, the kernel of g∘u is of finite type, hence we have an exact sequence

𝒪q⟶v𝒪p ⟶guℋ⟶0 (over some neighbourhood U of x).

Hence a commutative diagram with exact rows:

𝒪q ⟶v 𝒪p ⟶gu ℋ ⟶ 0 w ⤏AAAA ww ↓u ↓id 0 ⟶ ℱ ⟶f 𝒢 ⟶g ℋ ⟶ 0 (over U).

We wish to define w:𝒪q→ℱ such that fw=uv, and show that w is an epimorphism (over U). Since guv=0, Im(uv)⊆Ker(g)=Im(f), so we can define w to be f-1uv. To show that w is an epimorphism, let y∈U, consider the corresponding diagram of stalks over y, and verify that wy is an epimorphism by diagram-chasing. Hence by (7.1) w is an epimorphism and therefore ℱ is of finite type.

(2) ℱ,𝒢 coherent. 𝒢 is of finite type, hence so is ℋ. Let x∈X and let u:𝒪p→ℋ be a homomorphism (over an open neighbourhood of x). By (7.4) we can lift u to v:𝒪p→𝒢 (over a smaller open neighbourhood of x), so that gv=u. ℱ is of finite type, hence we have say e:𝒪q→ℱ (over some open neighbourhood of x). Hence we have the following diagram:

0 ⟶ ℱ ⟶f ℱ ⟶g ℋ ⟶ 0 e↑ t↑ ↓u 0 ⟶ 𝒪q ⇄rh 𝒪p+q ⇄sk 𝒪p ⟶ 0 (over a neighourhood of  x).

in which the rows are exact and the bottom row is split: rh=1, ks=1, hr+sk=1. Define t=fer+vk:𝒪p+q→𝒢, then the diagram is commutative. Since 𝒢 is coherent, the kernel of t is of finite type and we can therefore enlarge the diagram:

0 ⟶ ℱ ⟶f ℱ ⟶g ℋ ⟶ 0 e↑ t↑ ↓u 0 ⟶ 𝒪q ⟶h 𝒪p+q ⟶k 𝒪p ⟶ 0 k↑ ↑kw 𝒪n ⟶id 𝒪n (over a neighourhood of  x).

Verify that the right-hand column is exact, e.g. by considering the corresponding diagram of stalks over a point y∈U. Hence ℋ is coherent.

(3) ℱ,ℋ coherent. Since ℋ and ℋ are of finite type we have

0 ⟶ ℱ ⟶f ℱ ⟶g ℋ ⟶ 0 u↓ v ⤏AAAA vv ↓w 0 ⟶ 𝒪q ⟶ 𝒪p+q ⟶ 𝒪p ⟶ 0 (over a neighourhood of  x).

with u,w epimorphisms; hence as in (2) we can define v:𝒪p+q→𝒢. Since u,w are epi, so is v (by the 5 lemma). Hence 𝒢 is of finite type.

Now let u:𝒪r→𝒢 be a homomorphism (over some open neighbourhood of x); we have to show that Ker(u) is of finite type. Since ℋ is coherent we have an exact sequence of the form 𝒪s⟶v𝒪r⟶guℋ (over some open neighbourhood u of x), hence a diagram

0 ⟶ ℱ ⟶f ℱ ⟶g ℋ ⟶ 0 w ⤏AAAA ww ↑u ↑id 0 ⟶ 𝒪s ⟶v 𝒪r ⟶gu ℋ ⟶ 0 (over U).

here we have guv=0, hence Im(uv)⊆Ker(g)=Im(f), so we can define w:𝒪s→ℱ (over U) so that uv=fw. Now ℱ is coherent, hence the kernel of w is of finite type, hence we can enlarge the diagram:

0 ⟶ ℱ ⟶f ℱ ⟶g ℋ ⟶ 0 w↑ ↑u ↑id 0 ⟶ 𝒪s ⟶v 𝒪r ⟶gu ℋ p↑ ↑vp 𝒪t ⟶id 𝒪t (over a neighourhood of  x).

Here the first column (as well as the top row) is exact, and we verify (e.g. by diagram-chasing in the stalks) that the second column is exact. Hence Ker(u) is of finite type and therefore Q is coherent.

□

Corollary (7.6). ℱ and 𝒢 are coherent if and only if ℱ⊕𝒢 is coherent.

Proof.

If ℱ,𝒢 are coherent, the exact sequence 0→ℱ→ℱ⊕𝒢→𝒢→0 shows that ℱ⊕𝒢 is coherent. If ℱ⊕𝒢 is coherent then ℱ is of finite type because it is a homomorphic image of ℱ⊕𝒢; it is also a subsheaf of ℱ⊕𝒢, hence coherent by (7.3).

□

Corollary (7.7). If φ:ℱ→𝒢 is a homomorphism of coherent 𝒪-Modules, then the kernel, image and cokernel of φ are all coherent.

Proof.

Im(φ) is a homomorphic image of ℱ, hence is of finite type; it is also a sub-Module of 𝒢, hence coherent by (7.3). Now apply (7.5) to the exact sequences

0⟶Ker(φ)⟶ℱ ⟶Im(φ)⟶0 0⟶Im(φ)⟶𝒢⟶ Coker(φ)⟶0.

□

Corollary (7.8). If ℱ1⟶φℱ2 ⟶ℱ3⟶ℱ4 ⟶ψℱ5 is an exact sequence in which all but ℱ3 are coherent, then ℱ3 is coherent.

Proof.

From (7.7) and the exact sequence 0⟶Coker(φ)⟶ℱ3⟶Ker(ψ)⟶0.

□

Proposition (7.9). If ℱ,𝒢 are coherent 𝒪-Modules, then ℱ⊗𝒪𝒢 and ℋℴ𝓂𝒪(ℱ,𝒢) are coherent.

Proof.

Consider ℱ⊗𝒪𝒢. Let x∈X; since ℱ is coherent there is an exact sequence.

(✶) 𝒪q⟶𝒪p⟶ℱ⟶0 (over some open neighourbood U of  x);

hence, as tensoring with 𝒢 is right exact and 𝒪⊗𝒢≅𝒢, an exact sequence

𝒢q⟶𝒢p⟶ℱ⊗𝒢 ⟶0(over U);

since 𝒢 is coherent, so are 𝒢p,𝒢q by (7.6), hence ℱ⊗𝒢 is coherent by (7.7) and the fact that coherence is a local property.

For ℋℴ𝓂𝒪(ℱ,𝒢), operate on (✶) ℋℴ𝓂𝒪( ,𝒢). The argument is similar.

□

Proposition (7.10). If ℱ,𝒢 are 𝒪-Modules and ℱ is coherent, the mapping

(ℋℴ𝓂𝒪(ℱ,𝒢))x→ Hom𝒪x(ℱx,𝒢x)

is an isomorphism.

Proof.

From (✶) we have 𝒪xq→ 𝒪xp→ ℱx→0 exact, hence by the left exactness of ℋℴ𝓂𝒪 and Hom𝒪x we have exact sequences

0 ⟶ (ℋℴ𝓂𝒪(ℱ,𝒢))x→ ⟶ (ℋℴ𝓂𝒪(𝒪p,𝒢))x→ ⟶ (ℋℴ𝓂𝒪(𝒪q,𝒢))x→ ↓ ↓ ↓ 0 ⟶ Hom𝒪x(ℱx,𝒢x) ⟶ Hom𝒪x(𝒪xp,𝒢x) ⟶ Hom𝒪x(𝒪xq,𝒢x).

Since ℋℴ𝓂𝒪(𝒪p,𝒢)=𝒢p, the second and third vertical arrows are isomorphisms, hence so is the first.

□

If 𝒪 itself is coherent as an 𝒪-Module, we shall say that 𝒪 is a coherent sheaf of rings.

Proposition (7.11). Let 𝒪 be a coherent sheaf of rings and let 𝒢 be an 𝒪-Module. Then ℱ is coherent if and only if it is locally finitely presented, i.e. for each x∈X there is an exact sequence 𝒪q→𝒪p→ℱ→0 over some neighbourhood of x.

Proof.

If ℱ is coherent it is locally finitely presented (whether 𝒪 is coherent or not). Conversely, if 𝒪 is coherent, so are 𝒪p and 𝒪q by (7.6), hence so is ℱ by (7.7) (since coherence is a local property).

□

Notes and References

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

page history