§2. Cohomological study of projective morphisms

2.1. Explicit calculations of certain cohomology groups

2.1.1.

Let be a prescheme and an invertible -module; consider the graded ring

  S = Γ_*(X, ℒ) = ⊕_{n ∈ ℤ} Γ(X, ℒ^{⊗n}).                                    (2.1.1.1)

Let be a finite family of homogeneous elements of , with ; set

  U_i = X_{f_i},   U = ⋃_i U_i,

and denote by the cover of . For every quasi-coherent -module , set

  H^•(𝔘, ℱ(*)) = ⊕_{n ∈ ℤ} H^•(𝔘, ℱ ⊗ ℒ^{⊗n})                                (2.1.1.2)
  H^•(U, ℱ(*)) = ⊕_{n ∈ ℤ} H^•(U, ℱ ⊗ ℒ^{⊗n}).                               (2.1.1.3)

The abelian groups (2.1.1.2) and (2.1.1.3) are bigraded, by setting

  (H^•(𝔘, ℱ(*)))^{m,n} = H^m(𝔘, ℱ ⊗ ℒ^{⊗n})

and an analogous definition for (2.1.1.3). For the second degree, it is clear that these groups are graded -modules, as follows for instance from the fact that and are functors.

2.1.2.

Consider now the graded -module

  M = Γ_*(ℱ) = H^0(X, ℱ(*)) = ⊕_{n ∈ ℤ} Γ(X, ℱ ⊗ ℒ^{⊗n}).                    (2.1.2.1)

If is a prescheme whose underlying space is Noetherian, or a quasi-compact scheme, it follows from (I, 9.3.1) that, setting as usual , we have, up to a canonical isomorphism,

  Γ(U_{i_0 … i_p}, ℱ(*)) = H^0(U_{i_0 … i_p}, ℱ(*)) = M_{f_{i_0} … f_{i_p}}.

One can also, with the notation of (1.2.2), identify with . This identification is an isomorphism of graded -modules, provided that one defines the degree of a homogeneous element as follows: is the canonical image of a homogeneous element of degree ; one then takes for the degree of the number . Taking the definition of the homomorphisms (1.2.2) into account, one sees at once that this definition does not depend on the "representative" of chosen. Denoting, as in (1.2.2), by the set of alternating maps from to (for every ), one defines in the same way as above a structure of graded -module on ; one has again as in (1.2.2)

  C^p(𝔘, ℱ(*)) = lim_→ C^p_a(M),                                              (2.1.2.2)

the isomorphism of the two sides preserving degrees. One then has, as in (1.2.2),

  C'^p(𝔘, ℱ(*)) = C^{p+1}_a(𝐟, M) = lim_→ K^{p+1}(𝐟^n, M),                    (2.1.2.3)

the isomorphism preserving degrees: the degree of an element of , the canonical image of a cochain whose values lie in a single homogeneous component of , is , and is independent of the choice of this cochain as a representative of the element considered.

Since the preceding isomorphisms are compatible with the coboundary operators, we conclude, as in (1.2.2), the following.

Proposition (2.1.3).

Let be a prescheme whose underlying space is Noetherian, or a quasi-compact scheme. There exists a canonical isomorphism, functorial in ,

  H^p(𝔘, ℱ(*)) ⥲ H^{p+1}(𝐟, M)                              for every p ≥ 1.   (2.1.3.1)

Moreover, one has an exact sequence functorial in

  0 → H^0(𝐟, M) → M → H^0(𝔘, ℱ(*)) → H^1(𝐟, M) → 0.                          (2.1.3.2)

Furthermore, all the homomorphisms introduced are of degree 0 for the graded -module structures ( being the ring (2.1.1.1)).

Corollary (2.1.4).

If is a quasi-compact scheme and the are affine, there exists a canonical isomorphism, functorial in , of degree 0,

  H^p(U, ℱ(*)) ⥲ H^{p+1}(𝐟, M)                              for p ≥ 1         (2.1.4.1)

and an exact sequence functorial in

  0 → H^0(𝐟, M) → M → H^0(U, ℱ(*)) → H^1(𝐟, M) → 0                           (2.1.4.2)

where all homomorphisms are of degree 0.

Proof. It suffices to apply (1.4.1) to the result of (2.1.3).

The "local" proposition analogous to (2.1.3) is the following.

Proposition (2.1.5).

Let be a graded ring with positive degrees, a homogeneous element of of degree , a graded -module. Let be the homogeneous prime spectrum of and set , , . There then exist canonical isomorphisms functorial in , of degree 0 for the graded -module structures,

  H^p(U, M̃(*)) ⥲ H^{p+1}(𝐟, M)                             for p ≥ 1         (2.1.5.1)

and an exact sequence functorial in

  0 → H^0(𝐟, M) → M → H^0(U, M̃(*)) → H^1(𝐟, M) → 0                          (2.1.5.2)

where all homomorphisms are of degree 0.

Proof. We have by definition (II, 2.5.2), hence . The rest of the argument is then the same as for the proof of (2.1.4), taking into account that is a scheme.

Remarks (2.1.6).

(i) Under the conditions of (2.1.5), the functors are exact in , by virtue of ; the same reasoning as in (1.2.5) then shows that if is an exact sequence of graded -modules (where the homomorphisms are of degree 0), one has commutative diagrams for every

  H^p(U, M̃''(*))    →    H^{p+1}(U, M̃'(*))

      ↓                          ↓

  H^{p+1}(𝐟, M'')   →    H^{p+2}(𝐟, M')                                      (2.1.6.1)

(ii) Proposition (2.1.5) will be especially interesting when is an -algebra generated by a finite number of elements of degree 1, being assumed Noetherian; for

then, every quasi-coherent -module is of the form (II, 2.7.7).

2.1.7.

We are going to apply (2.1.5) in the case , where is an arbitrary ring, the are indeterminates, with and . One is therefore essentially reduced to computing , where .

Lemma (2.1.8).

If , one has, with ,

  H^i(𝐓^n, S) = 0                                          if i ≠ r + 1      (2.1.8.1)

The -module thus has a basis over formed of the classes mod. of the monomials with , for all .

Proof. This is an immediate consequence of (1.1.3.5) and Proposition (1.1.4), whose hypotheses are trivially verified.

2.1.9.

Pass to the inductive limit on ; the relations (2.1.8.1) give for . For , the inductive system is formed of the , the homomorphism for being multiplication by . For , denote by the class of mod. ; one then has , and these elements thus have the same canonical image in the inductive limit ; by virtue of the definition of degree given in (2.1.2), the degree of is therefore equal to . It is clear that the for and form a basis of . One immediately deduces from (2.1.8):

Corollary (2.1.10).

With the notation of (2.1.8), one has

  H^i(𝐓, S) = 0                                            for i ≠ r + 1     (2.1.10.1)

and is a free -module with a basis formed of the elements such that for .

Remark (2.1.11).

Let be an arbitrary -module and set ; the reasoning of (2.1.8) shows that one has more generally

  H^i(𝐓^n, M) = 0                                          if i ≠ r + 1      (2.1.11.1)
  H^{r+1}(𝐓^n, M) = (S/(𝐓^n)) ⊗_A N,                                         (2.1.11.2)

since the latter formula follows directly from (1.1.3.5), and on the other hand it is clear that is identified with the tensor product , the ideal being a direct factor in the -module ; this allows one to apply (1.1.4) to , and one thus obtains (2.1.11.1).

Combining (2.1.10) and (2.1.5), one obtains:

Proposition (2.1.12).

Let be a ring, an integer > 0, and (II, 4.1.1). Then:

(i) One has for .

(ii) The canonical homomorphism (II, 2.6.2) is bijective.

(iii) is a free -module having a basis formed of elements , where for , being of degree , and the product being equal to .

Proof. Note that, in the exact sequence (2.1.5.2) applied to , one has and by (2.1.10.1), and that Proposition (2.1.5) applies to , since is the union of the (II, 2.3.14). It remains to identify the map of the exact sequence (2.1.5.1) with the canonical map ; but this follows from the canonical identification of and .

Corollary (2.1.13).

The only values of for which one can have are the following: and ; and .

Note that if , one effectively has for the pairs enumerated in (2.1.13); this follows from (2.1.12), since is then for all degrees .

In the applications which will be made in this chapter, we shall mostly use the less precise result:

Corollary (2.1.14).

The -modules are free of finite type; if , they are zero for .

Proposition (2.1.15).

Let be a prescheme, an -module locally free of rank , the projective bundle defined by , the structure morphism. The only values of and for which are and , and ; in addition, the canonical homomorphism (II, 3.3.2)

  α : 𝐒_{𝒪_Y}(ℰ) → 𝚪_*(𝒪_X) = R^0 f_*(𝒪_X(*)) = ⊕_{n ∈ ℤ} f_*(𝒪_X(n))

is an isomorphism.

Proof. The question being local on , we may suppose affine with ring and , where ; one is then immediately reduced to (2.1.12), taking (1.4.11) into account.

Remark (2.1.16).

We shall later complete the results of (2.1.15) by proving the following propositions: set , which is an invertible -module. Then:

(i) One has a canonical isomorphism

  ρ : R^r f_*(ω) ⥲ 𝒪_Y.                                                      (2.1.16.1)

(ii) The cup-product pairing (0, 12.2.2)

  R^r f_*(𝒪_X(n)) × R^0 f_*(ω(−n)) → R^r f_*(ω)                              (2.1.16.2)

composed with the isomorphism , defines an isomorphism of onto the dual of the locally free -module

  R^0 f_*(ω(−n)) = (⋀^{r+1} ℰ) ⊗_{𝒪_Y} (𝐒_{𝒪_Y}(ℰ))_{−n}.

2.2. The fundamental theorem of projective morphisms

Theorem (2.2.1) (Serre).

Let be a Noetherian prescheme, a proper morphism, an invertible -module ample for . For every -module , set for every . Then, for every coherent -module :

(i) The are coherent -modules.

(ii) There exists an integer such that for , one has for every .

(iii) There exists an integer such that, for , the canonical homomorphism is surjective.

Proof. Let us first note that if the theorem is true when one replaces by (), it is true in its initial form. Indeed, one may then write with and , and by hypothesis, for each there is an integer such that for , properties (ii) and (iii) hold for the -module ; taking for the largest of the , (ii) and (iii) will hold for . One may therefore suppose very ample relative to (II, 4.6.11); consequently, there is a dominant -open immersion , where , with a quasi-coherent graded -algebra with positive degrees, in which is of finite type and generates ; in addition, is isomorphic to (II, 4.4.7). But since is proper, so is (II, 5.4.4), so is an isomorphism . One may thus restrict to the case where and . Theorem (2.2.1) is then a consequence of the following.

Proposition (2.2.2).

Let be a Noetherian ring, an -algebra graded with positive degrees, in which S_1 is an -module having a system of generators, and which generates the algebra . Let . For every coherent -module :

(i) The -modules are of finite type.

(ii) One has for .

(iii) There exists an integer such that for , one has for every .

(iv) There exists an integer such that for , is generated by its sections over .

Proof. Let us first show how (2.2.2) entails (2.2.1): in (2.2.1) (reduced to the particular case considered above), is quasi-compact, so can be covered by a finite number of affine open sets, with Noetherian rings, such that the restriction of to each of these open sets is generated by a finite number of sections of over . Assuming (2.2.2) proved, it then suffices to take for in parts (ii) and (iii) of (2.2.1) the largest of the analogous integers corresponding to the (taking (1.4.11) and (II, 3.4.7) into account).

To prove (2.2.2), note that identifies with a closed subscheme of (II, 3.6.2); in addition, if is the canonical injection, is a coherent -module, and one has (II, 3.4.5 and 3.5.2). Taking (G, II, cor. of th. 4.9.1) into account, one is therefore reduced to proving (2.2.2) in the particular case where and . As is covered by the affine opens

in number , (ii) results from (1.4.12). Note on the other hand that (iv) has already been proved (II, 2.7.9).

We shall prove (i) and (iii) simultaneously. Note that these assertions are true for (2.1.13); they are therefore also true when is a direct sum of a finite number of -modules of the form . On the other hand, (i) and (iii) are trivially true for by virtue of (ii). We shall proceed by descending induction on . We know that is isomorphic to a quotient of a direct sum of a finite number of sheaves (II, 2.7.10); in other words, one has an exact sequence , where is coherent and satisfies (i) and (iii). Since is an exact functor in , one also has the exact sequence

for every . One deduces the exact cohomology sequence

  H^{q−1}(X, ℰ(n)) → H^{q−1}(X, ℱ(n)) → H^q(X, ℛ(n)).

Since is a direct sum of the (II, 2.5.14), is of finite type, and so is by the induction hypothesis; since is Noetherian, one concludes that is of finite type for every , and in particular for . On the other hand, by the induction hypothesis, there exists an integer such that for one has ; furthermore, one may also suppose chosen so that for , since satisfies (iii); one concludes that for , which completes the proof.

Corollary (2.2.3).

Under the hypotheses of (2.2.1), let be an exact sequence of coherent -modules. There then exists an integer such that for , the sequence

is exact.

Proof. Let , , be the kernel, image, and cokernel of ; is the kernel and the image of , let be the cokernel of this homomorphism; all these -modules are coherent . Since is an exact functor in , it suffices to show that for large enough, each of the sequences

is exact; consequently, one may assume that is exact. One then has the exact cohomology sequence

  0 → f_*(ℱ(n)) → f_*(𝒢(n)) → f_*(ℋ(n)) → R^1 f_*(ℱ(n)) → ⋯

and the conclusion follows from (2.2.1, (ii)).

Corollary (2.2.4).

Let be a Noetherian prescheme, a morphism of finite type, an invertible -module ample for ; for every -module , set (for ). Let be an exact sequence of coherent -modules,

such that the supports of and are proper over (II, 5.4.10). There then exists an integer such that, for , the sequence

is exact.

Proof. The same reasoning as at the beginning of (2.2.1) shows that if the corollary is true for (), it is also true for ; one may therefore restrict to the case where is very ample for (II, 4.6.11), and consequently one may identify with an open set in a -scheme , where is a quasi-coherent graded -algebra with positive degrees, in which is of finite type and generates , so that , where is the canonical immersion (II, 4.4.7). This being so, as is closed in and contained in , it is proper over ; the supports of , , are thus closed in (II, 5.4.10). The sheaves , , are therefore coherent -modules, and the sequence is exact; in addition, if is the structure morphism, one has , and it is clear that and similarly for and ; the conclusion thus follows from (2.2.3) applied to , , .

Remarks (2.2.5).

(i) Assertion (i) of (2.2.1) is still true when one supposes only that is locally Noetherian; indeed, the property is evidently local on ; on the other hand, the hypotheses of (2.2.1) imply that for every open , the restriction of to is a projective morphism (II, 5.5.5, (iii)) and is ample for this morphism (II, 4.6.4).

(ii) Assertion (iii) of (2.2.1) is still valid, as we have seen, when one supposes only that is a quasi-compact scheme or a prescheme whose underlying space is Noetherian, and a quasi-compact morphism (II, 4.6.8). But it should be noted that even when one supposes that is the spectrum of a field and that is quasi-projective, assertion (ii) of (2.2.1) is no longer necessarily verified. For example, let and let be the union of the affine opens of (); since the immersion is quasi-compact, the structure morphism is quasi-affine (II, 5.1.10), so is very ample for (II, 5.1.6). But the ring identifies with the intersection of the rings of fractions for (I, 8.2.1.1), that is, with . Consequently, it follows from formulas (1.4.3.1) and (1.1.3.5) that one has for every .

2.3. Application to graded sheaves of algebras and of modules

Theorem (2.3.1).

Let be a Noetherian prescheme, a quasi-coherent graded -algebra of finite type with positive degrees, , the structure morphism, a quasi-coherent graded -module satisfying condition (TF). Then there exists an integer such that, for , the canonical homomorphism (II, 8.14.5.1)

  α_n : ℳ_n → q_*(𝒫roj(ℳ(n))) = q_*((𝒫roj(ℳ))_n)

is bijective. In other words, the canonical homomorphism

is a (TN)-isomorphism.

Proof. One may restrict to the case where is an -module of finite type (II, 3.4.2).

As is quasi-compact, there exists an integer such that is generated by the quasi-coherent -module , the latter being of finite type (II, 3.1.10), hence coherent since is Noetherian. Note now that is a direct sum of the for , and that each of the is a quasi-coherent -module of finite type, as follows from (II, 2.1.6, (iii)), the question being local on . Now, it obviously suffices to prove that each of the canonical homomorphisms is a (TN)-isomorphism. Taking (II, 8.14.13) into account (and notably the diagram (8.14.13.4)), one sees that one is reduced to proving the theorem when is generated by and is a coherent -module. Since is Noetherian, the same reasoning as at the beginning of (2.2.2) shows that one may restrict to the case where , , , being a Noetherian ring, S_1 an -module of finite type and a graded -module of finite type. Let us show that it then suffices to prove the theorem when . Indeed, in the general case, one has an exact sequence , where and are direct sums of graded modules of the form . If the result is true for , it is also true for , hence for and . Consider then the commutative diagram

The second line is exact by virtue of (2.2.3) as soon as is large enough; as the same holds for the first, and as the two left vertical arrows are isomorphisms, so is the third.

This being so, to prove the theorem when , suppose first that ( indeterminates); in this case, our assertion is none other than (2.1.11, (ii)). In the general case, identifies with a quotient of a ring by a graded ideal, hence with a closed subscheme of (II, 2.9.2). If is the canonical injection , is none other than the -module where is considered as a graded -module; this follows immediately from (II, 2.8.7). As is an -module satisfying (TF), the canonical homomorphism is bijective for large enough, by virtue of what precedes; this completes the proof, since .

Corollary (2.3.2).

Under the hypotheses of (2.3.1), let , and let be a quasi-coherent graded -module of finite type. Then satisfies condition (TF).

Proof. We have seen in the proof of (2.3.1) that , which is isomorphic to (II, 3.1.8), is of finite type over (II, 3.4.1). It then follows from (II, 8.14.9) that is isomorphic to an -graded module of the form , where is a quasi-coherent graded -module of finite type. By virtue of (2.3.1), is (TN)-isomorphic to , and consequently satisfies (TF).

Scholium (2.3.3).

Let be a Noetherian prescheme, an -algebra graded satisfying the conditions of (2.3.1) and . Let be the abelian category of quasi-coherent graded -modules satisfying (TF), the subcategory of formed of -modules satisfying (TN); finally, let be the category of quasi-coherent graded -modules of finite type (which amounts to saying, since is periodic (II, 8.14.4 and 8.14.12), that the are coherent -modules). Then the functors in and in define, by virtue of (II, 8.14.8 and 8.14.10) and (2.3.2), an equivalence (T, I, 1.2) of the quotient category (T, I, 1.11) with the category . When is generated by , one may replace by the category of coherent -modules (II, 8.14.12).

Proposition (2.3.4).

Let be a Noetherian prescheme.

(i) Let be a quasi-coherent graded -algebra of finite type with positive degrees. Let and . Then is a periodic graded -algebra (II, 8.14.12) whose homogeneous components are coherent -modules, and if is a period of , is an invertible -module -ample. In addition, the canonical homomorphism is a (TN)-isomorphism.

(ii) Conversely, let be a projective morphism, and let be a graded -algebra whose homogeneous components () are coherent -modules, and which admits a period such that is an invertible -module ample for . Then is a quasi-coherent graded -algebra with positive degrees of finite type, and there exists a -isomorphism such that is isomorphic (as a graded -algebra) to .

Proof. (i) All the assertions have essentially already been proved, the last being none other than a particular case of (2.3.2). The fact that is periodic has been seen in (II, 8.14.14), and the fact that there is a period such that is invertible and -ample is none other than (II, 4.6.18). Finally, for , is a -module of finite type (II, 8.14.14), so each of the is a quasi-coherent -module of finite type by virtue of (II, 2.1.6, (ii)), the question being local; as is coherent, so are the .

(ii) Up to replacing the period by one of its multiples, one may suppose that is an -module very ample relative to (II, 4.6.11). We have in addition by hypothesis, so ; we know (II, 3.1.8 and 3.2.9) that there is a -isomorphism of onto such that

. One will therefore establish the existence of a -isomorphism if one proves the following.

Proposition (2.3.4.1).

Let be a Noetherian prescheme, a projective morphism, an invertible -module very ample for . Then is a quasi-coherent graded -algebra of finite type, such that for large enough, and there exists a -isomorphism such that .

Proof. As is a projective morphism, it follows from (II, 5.4.4 and 4.4.7) that there exists a -isomorphism , where is a quasi-coherent -algebra such that is an -module of finite type and generates , and one has . One then has , where is the structure morphism, and it follows from (2.3.1) that for large enough, the canonical homomorphism is bijective; as , one has a fortiori as soon as is large enough. In addition, as the canonical homomorphism of graded -algebras is a (TN)-isomorphism, Φ = Proj(α) : Proj(𝒮) → Proj(𝒯) is an isomorphism (II, 3.6.1) and one has (II, 3.5.2); but since the -graded modules and are (TN)-isomorphic, one has for every (II, 3.4.2); to complete the proof of (2.3.4.1), it remains to show that is an -algebra of finite type; now the are coherent -modules by virtue of (2.2.1) and, since for , is generated by , which is coherent, whence our assertion (I, 9.6.2).

Let us return to the proof of (2.3.4), whose notation we resume. We have proved the existence of a -isomorphism such that for every ; we shall denote by q'' the structure morphism . Note now that is a direct sum of the -graded modules ; each of the latter is a quasi-coherent -module of finite type, by virtue of the periodicity of and the hypothesis that the are -modules of finite type (II, 8.14.12). Set , so the are quasi-coherent graded -modules of finite type; consequently (II, 8.14.8), the canonical homomorphism is an isomorphism of -modules. But one has and for , this last -module is by definition equal to . In other words, the canonical injection is a (TN)-isomorphism, hence (II, 3.4.2) one has , and consequently . It remains to note that up to a canonical isomorphism (II, 8.14.13.1) in order to have proved the isomorphism of and . Finally, by virtue of (2.3.2), each of the satisfies condition (TF), so the same holds for each of the ; in addition, since the are coherent, the same holds for the by (2.2.1), and one concludes at once that the are -modules of finite type. As we have seen in (2.3.4.1) that is an -algebra of finite type, one concludes that is also an -algebra of finite type.

Proposition (2.3.5).

Let be an integral Noetherian prescheme, an integral prescheme, a birational projective morphism. There then exists a coherent fractional ideal sheaf (II, 8.1.2) such that is -isomorphic to the prescheme obtained by blowing up (II, 8.1.3). In addition, there exists an open set of such that the restriction of to is an isomorphism of onto (cf. (I, 6.5.5)), and such that is invertible.

Proof. As there exists an invertible -module very ample for (II, 4.4.2 and 5.3.2), one may apply (2.3.4.1), and one sees that identifies with , where . We know in addition that the are torsion-free -modules (I, 7.4.5), so the same holds for the -module , and consequently identifies canonically with a sub--module of (I, 7.4.1); the latter is a simple sheaf (I, 7.3.6), which is known when one knows its restriction to a nonempty open set, for instance to a nonempty open such that is isomorphic to . Since by hypothesis the are then isomorphic to , one sees that is an -module isomorphic to , where is an indeterminate, and is (TN)-isomorphic to the sub--algebra generated by the canonical image of in (2.3.4.1); but if one identifies with , the image of identifies with , where is a coherent sub--module (2.2.1) of , whose restriction to is isomorphic to , and which consequently is such that is invertible. One then sees that is (TN)-isomorphic to , which completes the proof.

Corollary (2.3.6).

Under the hypotheses of (2.3.5), suppose in addition that for every coherent sub--module of , there exists an invertible -module such that ; then, in the statement of (2.3.5), one may suppose that is an ideal of . This additional condition is always satisfied if there exists an ample -module.

Proof. Indeed, one has

  ℒ ⊗ ℋom(𝒥, 𝒪_Y) = ℋom(ℒ^{−1}, ℋom(𝒥, 𝒪_Y)) = ℋom(𝒥 ⊗ ℒ^{−1}, 𝒪_Y);

the hypothesis thus signifies that there is a nonzero homomorphism of into . As, for every , identifies with a sub--module of the field of fractions of (I, 7.1.5), is necessarily injective, so is an isomorphism of onto an ideal of . But since and are -isomorphic (II, 3.1.8), this proves the first assertion of the corollary. To prove the second, note that is coherent and , since there exists an open set of such that is invertible. If is an ample -module, there exists an integer such that is generated by its sections over (II, 4.5.5); a fortiori, one has , whence the conclusion.

Corollary (2.3.7).

Let and be two integral schemes, projective over a field , and let be a birational -morphism. Then is -isomorphic to a -scheme obtained by blowing up a closed subscheme (not necessarily reduced) of .

Proof. Indeed, is projective (II, 5.5.5, (v)), and as is projective over , the additional condition of (2.3.6) is satisfied; it then suffices to consider the closed subscheme of defined by the coherent ideal of cor. (2.3.6).

Remark (2.3.8).

In Chap. IV, in studying the notion of divisor, we shall see that if, in the statement of (2.3.5), one supposes that the rings () are factorial (which is the case for instance if is non-singular), then may be deduced from by blowing up a closed sub-prescheme of whose underlying space is contained in .

2.4. A generalization of the fundamental theorem

Theorem (2.4.1).

Let be a Noetherian prescheme, a quasi-coherent -algebra of finite type. Let be a projective morphism, , a quasi-coherent -module of finite type. Then:

(i) For every , is an -module of finite type.

(ii) Let in addition be an invertible -module ample for , and set for every . There exists an integer such that, for , one has

for every , and the canonical homomorphism is surjective.

Proof. Set , , so that (II, 1.5.5); let , be the structure morphisms, which are affine by definition, and ; one therefore has a commutative diagram

        g'
   X ←──── X'
   |       |
 f |  ↘ h  | f'
   ↓       ↓
   Y ←──── Y'
        g

and the morphism is projective (II, 5.5.5, (iii)); set .

(i) Let be the -module associated to the quasi-coherent -module , when is considered as an affine -scheme (II, 1.4.3), so that one has ; as is an -module of finite type, is an -module of finite type (II, 1.4.5); as is of finite type, since and are (II, 1.3.7 and I, 6.3.4, (ii)), is Noetherian (I, 6.3.7) and is consequently coherent. This being so, as is affine, the canonical homomorphism is bijective (1.3.4). In addition, this homomorphism is a homomorphism of -modules; indeed, from the canonical homomorphism

  g^*(𝒮) ⊗_{𝒪_{X'}} g'^*(ℳ) → g'^*(ℳ)                                       (2.4.1.2)

which defines the -module structure of (recalling that ), one canonically deduces a homomorphism

  f_*(g^*(𝒮)) ⊗ R^p f_*(g'^*(ℳ)) → R^p f_*(g'^*(ℳ))

(0, 12.2.2), and since (2.4.1.2) itself comes (by application of ) from the homomorphism defining the -module structure of , the diagram

  f_*(g'_*(𝒪_{X'})) ⊗ R^p f_*(g'_*(ℳ̃))    →    R^p f_*(g'_*(ℳ̃))

           ↓                                            ↓

  h_*(𝒪_{X'}) ⊗ R^p h_*(ℳ̃)                →    R^p h_*(ℳ̃)

is commutative (0, 12.2.6); composing the horizontal arrows with the homomorphism coming from the canonical homomorphism , one obtains our assertion. On the other hand, since is affine and is separated and quasi-compact, the canonical homomorphism is bijective (1.4.14), and one shows as above that it is an isomorphism of -modules (this time using the commutativity of (0, 12.2.6.2)). Now, since is projective and is coherent, is a coherent -module by virtue of (2.2.1); one concludes that is an -module of finite type (II, 1.4.5).

(ii) Let , which is an invertible -module; for every , one has up to an isomorphism; one may apply to the reasoning made in (i) for , which proves that is isomorphic to . Now is ample for (II, 4.6.13, (iii)), so it follows from (2.2.1) that there exists an integer such that for every and every , which proves (2.4.1.1). Finally, it follows again from (2.2.1) that one may suppose chosen so that for , the canonical homomorphism is surjective; as is an exact functor (II, 1.4.4), the corresponding homomorphism

  g'_*(f'^*(f'_*(ℳ̃ ⊗ ℒ'^{⊗n}))) → g'_*(ℳ̃ ⊗ ℒ'^{⊗n}) = ℳ(n)

is surjective. Now, one has (II, 1.5.2) and since , one finally sees that one has

  g'_*(f'^*(f'_*(ℳ̃ ⊗ ℒ'^{⊗n}))) = f^*(f_*(g'_*(ℳ̃ ⊗ ℒ'^{⊗n}))) = f^*(f_*(ℳ(n))),

which completes the proof.

2.4.2.

We shall in particular have to apply (2.4.1) when is an -graded algebra with positive degrees, an -graded module. Then (with the

same hypotheses of finiteness on and ) one concludes from (2.4.1) that is an -module of finite type for every , and (under the additional hypotheses of (2.4.1, (ii))) that there exists such that for , one has for every and every , and that the canonical homomorphism is surjective for every .

2.5. Euler–Poincaré characteristic and the Hilbert polynomial

2.5.1.

Let be an Artinian ring, an -scheme projective over . For every coherent -module , the () are -modules of finite type (2.2.1), hence here of finite length since is Artinian. One knows in addition (2.2.1) that except for a finite number of values of ; the integer

  χ_A(ℱ) = Σ_{i=0}^∞ (−1)^i long(H^i(X, ℱ))                                  (2.5.1.1)

is thus defined for every coherent -module . When is an Artinian local ring, one says that is the Euler–Poincaré characteristic of (with respect to the ring ). For , one says that is the arithmetic genus of (with respect to ).

Proposition (2.5.2).

Let be an exact sequence of coherent -modules; one then has

Proof. As the cohomology modules of , , are zero except for a finite number of them, there is an integer such that the exact cohomology sequence is written

  0 → H^0(X, ℱ') → H^0(X, ℱ) → H^0(X, ℱ'') → H^1(X, ℱ') → ⋯
       ⋯ → H^r(X, ℱ') → H^r(X, ℱ) → H^r(X, ℱ'') → 0.

Now, we know that in an exact sequence of -modules of finite length, with 0 at both ends, the alternating sum of the lengths is zero (0, 11.10.1); applying this result, one immediately finds the formula (2.5.2.1).

One notes that the result of (2.5.2) applies whenever one knows that is a quasi-compact -scheme and that the -modules are of finite type for every coherent -module (1.4.12).

Theorem (2.5.3).

Let be an Artinian local ring, a scheme projective over , an invertible -module very ample relative to , a coherent -module; set for every .

(i) There exists a unique polynomial such that for every (one says that is the Hilbert polynomial of with respect to ).

(ii) For large enough, one has .

(iii) The leading coefficient of is .

Let us add that in Chap. IV, in the paragraph devoted to the notion of dimension, we shall in addition show that the degree of is equal to the dimension of the support of .

Proof. As one has for every as soon as is large enough (2.2.1),

one has for large enough, whence (ii); this implies for large enough, and (iii) thus follows from (i); as moreover the assertion of uniqueness in (i) is immediate, it remains to prove the existence of the polynomial .

Let us first show that one may suppose , where is the maximal ideal of . Indeed, there exists an integer such that , and thus admits a finite filtration

  ℱ(n) ⊃ 𝔪 ℱ(n) ⊃ ⋯ ⊃ 𝔪^{s−1} ℱ(n) ⊃ 0.

By induction, one deduces from (2.5.2.1) that

  χ_A(ℱ(n)) = Σ_{k=1}^s χ_A(𝔪^{k−1} ℱ(n) / 𝔪^k ℱ(n));

since , where , this proves our assertion.

So suppose ; if is the closed subscheme of , inverse image by the structure morphism of the unique closed point of , and the canonical injection, one has , where is a coherent -module; is a scheme projective over , where . If , is very ample relative to (II, 4.4.10), and one has , where . One concludes that (G, II, 4.9.1), and one is thus reduced to the case where is a field.

Note now that can be considered as a closed subscheme of for a suitable (II, 5.5.4, (ii)); if is the canonical injection, one sees as above that one has , so that one may restrict to the case where with , being a field.

This being so, one has , where is a graded -module of finite type (II, 2.7.8); there exists consequently a finite resolution of by graded free -modules of finite type

  0 → L_q → L_{q−1} → ⋯ → L_1 → M → 0

by virtue of Hilbert's syzygy theorem (M, VIII, 6.5); as is an exact functor in (II, 2.5.4), one also has an exact sequence

  0 → L̃_q → L̃_{q−1} → ⋯ → L̃_1 → M̃ → 0

and consequently, for every , the sequence

  0 → L̃_q(n) → L̃_{q−1}(n) → ⋯ → L̃_1(n) → M̃(n) → 0

is exact; applying by induction on Proposition (2.5.1), it comes

  χ_A(M̃(n)) = Σ_{i=1}^q (−1)^{i+1} χ_A(L̃_i(n))

and to prove (i), one is therefore reduced to the case where is free and graded of finite type, hence to the case where for an . As we then have (II, 2.5.15), one finally sees that the theorem will follow from the following.

Lemma (2.5.3.1).

Let be a field, an integer > 0, and ; one then has for every .

Proof. Indeed, for , one has , which is the number of monomials in the of total degree , that is, (2.1.12). For , one has similarly ; if , the dimension of over is the number of sequences of integers such that (2.1.12), or equivalently the number of sequences of integers () such that ; this is therefore the number . Finally, for , one has and on the other hand for every (2.1.12), which proves the lemma.

Corollary (2.5.4).

Let be an Artinian local ring, a graded -algebra of finite type generated by S_1, a graded -module of finite type, . One then has for large enough.

Proof. This follows from the fact that and are isomorphic for large enough (2.3.1).

2.6. Application: ampleness criteria

Proposition (2.6.1).

Let be a Noetherian prescheme, a proper morphism, an invertible -module. The following conditions are equivalent:

a) is ample for .

b) For every coherent -module , there exists an integer such that for , one has for every .

c) For every coherent ideal sheaf of , there exists an integer such that for one has .

Proof. We have seen that a) entails b) (2.2.1, (ii)). It is trivial that b) entails c), and it remains to prove that c) implies a). One may restrict to the case where is affine (II, 4.6.4), and prove in this case that is ample; it will suffice to show that as runs over the set of sections of the () over , those of the which are affine form a cover of (II, 4.5.2). For this, let us show that for every closed point of and every affine open neighbourhood of , there exist an and an such that ; will necessarily be affine (I, 1.3.6) and the union of these will be an open set of containing all the closed points of , and consequently itself, since is Noetherian . Let (resp. ) be the quasi-coherent ideal sheaf of defining the closed reduced subprescheme of having for underlying space (resp. ) (I, 5.2.1); it is clear that and are coherent (I, 6.1.1), that , and that is a coherent -module with support and such that . As is locally free, the sequence is exact for every , and by hypothesis there exists large enough such that ; the exact cohomology sequence

therefore proves that the homomorphism is surjective. A section of over such that is therefore the image of a section (since by virtue of , is a sub--module of ); one has by definition and for , which completes the proof.

Proposition (2.6.2).

Let be a Noetherian prescheme, a morphism of finite type, a finite surjective morphism, an invertible -module and . Suppose the following condition is satisfied: there exists a subset of , proper over (II, 5.4.10), such that for every , either is normal at the point , or is a free -module. Under these conditions, for to be ample for , it is necessary and sufficient that be ample for .

Proof.

(2.6.2.1)

Since is affine, the condition is necessary (II, 5.1.12). To see that it is sufficient, one may suppose affine (II, 4.6.4). Let us further show that one may restrict to the case where is reduced. Indeed, let be the canonical injection, and set , , so that one has the commutative diagram

         j'
    X' ←───── X'_1
    |          |
  g |          | g_1
    ↓          ↓
    X  ←───── X_1
         j

                                                                              (2.6.2.2)

The morphism is then of finite type (I, 6.3.4) and is a finite morphism (II, 6.1.5, (iii)); if is ample for , is ample for since is a closed immersion (II, 5.1.12 and I, 4.3.2). If one sets , Z_1 is proper over (II, 5.4.10); on the other hand, if is normal at a point , the same is evidently true of ; finally, if is a free -module, it follows at once from (II, 1.5.2) that is a free -module. Finally, since is Noetherian (I, 6.3.7), if is ample, is ample (II, 4.5.14), and as , this completes the reduction announced. We therefore suppose henceforth affine and reduced.

The hypotheses of (II, 6.6.11) then being verified, there exists a reduced -prescheme X_2 and a -morphism finite and birational such that the restriction of to is an isomorphism onto and that is ample. Replacing by X_2, one sees that one is reduced to proving the proposition supposing in addition that has the properties just enumerated for . We shall again denote by a sub-prescheme of having for underlying space, which is proper over (II, 5.4.10).

(2.6.2.3)

Let now X_1 be a closed sub-prescheme of , the canonical injection, its inverse image, the canonical injection, so that one has the commutative diagram (2.6.2.2); set , , so that is ample for (II, 5.1.12). If one sets , the closed sub-prescheme Z_1 of X_1 is proper over (II, 5.4.2, (ii)). In other words, the hypotheses of (2.6.2) are verified for X_1, , , and Z_1.

This will allow us to prove (2.6.2) by Noetherian induction in the case where the restriction of to is an isomorphism onto (which is sufficient for our purpose, as one has seen in (2.6.2.2)): it will suffice to establish that if, for every closed sub-prescheme X_1 of whose underlying space is , the conclusion of (2.6.2) is true for the sheaf , then it is also true for the sheaf .

(2.6.2.4)

Let then , , so that is a sub--algebra of , which is a coherent -module; in addition, the restriction is equal to . Let be the conductor of over , that is, the largest sub--module quasi-coherent of such that (or equivalently the annihilator of the -module ), which entails . It is clear that at every point admitting a neighbourhood such that is an isomorphism of onto , and in particular at every point of and in a neighbourhood of every generic point of an irreducible component of . Consider then the closed sub-prescheme of defined by ; it is still proper over , since the sub-space Z_1 is closed in (II, 5.4.10). Moreover, the definition of shows that ; one thus sees that one can always reduce to the case where , and as we have seen that is a nonempty open set of , one may always suppose that the space is distinct from .

(2.6.2.5)

Consider as equal to (since is affine) and let , a coherent ideal sheaf of such that (II, 1.4.1); the closed sub-prescheme of is defined by and equal to (II, 1.4.10); as is a finite morphism (II, 6.1.5, (iii)), is proper over (II, 6.1.11 and 5.4.2, (ii)).

This being so, we must prove that for every and every open neighbourhood of , there exists a section of an () over such that (II, 4.5.2); we distinguish two cases:

1° One has ; one may evidently then suppose that one also has , so the open set does not meet . As is ample by hypothesis, there exist an and a section of over such that (II, 4.5.2). In addition, one may suppose that is generated by its sections over (II, 4.5.5), so, since , there is a section s'' of these such that ; multiplying it by (which amounts to replacing by 2n), one sees that one may also suppose that . This being so, it follows from that one has a canonical isomorphism

  Γ(X, 𝒦 ⊗ ℒ^{⊗n}) ⥲ Γ(X', 𝒦' ⊗ ℒ'^{⊗n}).

The section of corresponding to s'' under this isomorphism evidently has the desired properties.

2° One has . Let be the coherent ideal sheaf of defining the closed reduced sub-prescheme of having for underlying space , and consider in the coherent ideals

and , so that one has the diagram of inclusions

   𝒥 ℬ    →    ℬ
   ↑           ↑
   𝒥     →    𝒜
   ↑           ↑
   𝒥 𝒦 ℬ = 𝒥 𝒦  →   𝒦                                                       (2.6.2.6)

Let be the coherent ideal sheaf of , so that , , and consequently (II, 1.4.4). As for every open set not meeting , one sees that the support of is contained in . As is proper over , one may apply (2.2.4) and one sees that for large enough, the canonical map

  Γ(X', 𝒥' ⊗ ℒ'^{⊗n}) → Γ(X', (𝒥' / 𝒥' 𝒦') ⊗ ℒ'^{⊗n})

is surjective.

But by virtue of , one concludes that the canonical map

  Γ(X, 𝒥 ℬ ⊗ ℒ^{⊗n}) → Γ(X, (𝒥 ℬ / 𝒥 𝒦 ℬ) ⊗ ℒ^{⊗n})

is surjective.

This being so, let be the canonical injection, the canonical injection, so that one has the commutative diagram

         i'
    X' ←───── Z'
    |          |
  g |          | h
    ↓          ↓
    X  ←───── Z
         i

Let , ; as is ample, is ample (II, 5.1.12), and on the other hand ; one concludes therefore from the hypothesis of Noetherian induction (since ) that is ample. Consequently is generated by its sections over for large enough (II, 4.5.5). As , one deduces again from that there exists a section of over (for large enough) such that , since one has by definition of and by hypothesis. The diagram (2.6.2.6) shows that is also a section of over , hence is the canonical image of a section of over . But by definition, the canonical image of mod is in , hence by virtue of (2.6.2.6), this implies that is a section of over , and a fortiori a section of . One has seen above that , so , and by definition of , in which is the support of ; thus , which completes the proof.

Remark (2.6.3).

When is proper over , one can prove (2.6.2) more simply, by reasoning as in Chevalley's theorem (II, 6.7.1), using (2.6.1) and Lemma (II, 6.7.1.1).