§6. Integral morphisms and finite morphisms

6.1. Preschemes integral over another

Definition.

Let be an -prescheme and its structure morphism. We say that is integral over , or that is an integral morphism, if there exists a cover of by affine opens such that, for every , the induced prescheme is an affine scheme whose ring is an integral algebra over the ring of . We say that is finite over , or that is a finite morphism, if is integral and of finite type over .

When is affine of ring , we shall also say "integral (resp. finite) over " instead of "integral (resp. finite) over ".

(6.1.2)

It is clear that if is integral over , then it is affine over . For an affine prescheme over to be integral (resp. finite) over , it is necessary and sufficient that the associated quasi-coherent -algebra be such that there exists a cover of by affine opens with the property that, for every , is an integral (resp. integral and of finite type) algebra over . A quasi-coherent -algebra with this property is said to be integral (resp. finite) over . To give an integral (resp. finite) prescheme over is therefore the same (1.3.1) as to give a quasi-coherent -algebra that is integral (resp. finite) over . Note that a quasi-coherent -algebra is finite if and only if it is an -module of finite type (I, 1.3.9); this is equivalent to saying that is an integral -algebra of finite type, since an algebra that is integral and of finite type over a ring is an -module of finite type.

Proposition.

Let be a locally Noetherian prescheme. For an affine prescheme over to be finite over , it is necessary and sufficient that the -algebra be coherent.

Proof. Taking the preceding remark into account, this comes down to noting that, if is locally Noetherian, the quasi-coherent -modules of finite type are exactly the coherent -modules (I, 1.5.1).

Proposition.

Let be an integral (resp. finite) prescheme over and its structure morphism. Then, for every affine open of ring , is an affine scheme whose ring is an integral (resp. finite) algebra over .

Proof. We first prove the following lemma.

Lemma.

Let be a ring, an -module, and a finite system of elements of such that the () cover . If, for every , is an -module of finite type, then is an -module of finite type.

Proof. We may assume that each admits a finite system of generators with , with the same for all indices . We show that the form a system of generators of . Let be the sub--module of generated by these elements, and let be an element of . By hypothesis, for each , there exist and an integer (independent of ) such that, in , ; this implies that there exists an integer such that, for every , . Now, since the cover , the ideal of generated by the is equal to ; in other words, there exist elements such that , and consequently , whence the lemma.

This being so, we already know (1.3.2) that is affine. If is the homomorphism corresponding to , there exists a finite cover of by opens () such that, if , then is an integral (resp. integral and finite) algebra over . Indeed, there is a cover of by affine opens such that, if , is an integral (resp. finite) algebra over . Every belongs to some , so there exists such that ; if is the image of in , then . Let and let be the image of in ; we have

and, since is integral (resp. finite) over , is integral (resp. finite) over . It now suffices to use the fact that is quasi-compact to obtain the desired cover.

If we first suppose that the are integral and finite over the , then since , viewed as an -module, also coincides with , Lemma (6.1.4.1) shows that in this case is an -module of finite type.

Now suppose only that each is integral over ; let , and let be the sub--algebra of generated by . For every , is the algebra over generated by in ; it follows from the hypothesis that each is an -module of finite type, hence (6.1.4.1) is an -module of finite type, which proves that is integral over .

Proposition.

  • (i) A closed immersion is finite (and a fortiori integral).
  • (ii) The composition of two finite (resp. integral) morphisms is finite (resp. integral).
  • (iii) If is a finite (resp. integral) -morphism, then is finite (resp. integral) for every base extension .
  • (iv) If and are two finite (resp. integral) -morphisms, then so is .
  • (v) If and are morphisms such that is finite (resp. integral) and is separated, then is finite (resp. integral).
  • (vi) If is a finite (resp. integral) morphism, then so is .

Proof. By (I, 5.5.12), it suffices to prove (i), (ii), and (iii). To prove that a closed immersion is finite, we may restrict to the case , and everything then comes down to noting that a quotient ring is a monogeneous -module. To prove that the composition of two finite (resp. integral) morphisms , is finite (resp. integral), we may again assume that (and hence and (1.3.4)) is affine, and the claim then amounts to: if is a finite (resp. integral) -algebra and a finite (resp. integral) -algebra, then is a finite (resp. integral) -algebra, which is immediate. Finally, to prove (iii), we may restrict to the case , since is identified with (I, 3.3.11); we may further assume that and ; then is affine of ring (1.3.4), is affine of ring , and it suffices to remark that if is a finite (resp. integral) -algebra, then is a finite (resp. integral) -algebra.

We also note that if and are two -preschemes that are finite (resp. integral) over , their sum is a finite (resp. integral) prescheme over , since this comes down to seeing that if and are finite (resp. integral) -algebras, then so is .

Corollary.

If is an integral (resp. finite) prescheme over , then, for every open , is integral (resp. finite) over .

Proof. This is a particular case of (6.1.5, (iii)).

Corollary.

Let be a finite morphism. Then, for every , the fibre is a finite algebraic scheme over , and a fortiori its underlying space is discrete and finite.

Proof. Indeed, as a -prescheme, is identified with (I, 3.6.1), which is finite over (6.1.5, (iii)); it is thus an affine scheme whose ring is an algebra of finite rank over (6.1.4). The corollary then follows from (I, 6.4.4).

Corollary.

Let , be two integral preschemes and a dominant morphism. If is integral (resp. finite), then the field of rational functions on is algebraic (resp. algebraic of finite degree) over the field of rational functions on .

Proof. Let be the generic point of ; the -prescheme is integral (resp. finite) over (6.1.5, (iii)) and contains, by hypothesis, the generic point of ; the local ring of in , being equal to (I, 3.6.5), is a local ring of an integral (resp. finite) algebra over (6.1.4), whence the corollary.

Remark.

The hypothesis that is separated is essential for the validity of (6.1.5, (v)): indeed, if is not separated over , then the identity 1_Y is the composite morphism , but is not an integral morphism, as follows from (6.1.10):

Proposition.

Every integral morphism is universally closed.

Proof. Let be an integral morphism; by (6.1.5, (iii)), it suffices to show that is closed. Let be a closed part of ; there exists a subprescheme of whose

underlying space is (I, 5.2.1), and it therefore follows from (6.1.5, (i) and (ii)) that we may restrict to proving that is closed in . By (6.1.5, (vi)), we may assume that and are reduced; furthermore, if is the closed reduced subprescheme of whose underlying space is (I, 5.2.1), we know that factors as , where is the injection morphism (I, 5.2.2), and since is separated (I, 5.5.1, (i)), it follows from (6.1.5, (v)) that is an integral morphism. We may therefore assume that is dense in . Finally, since the question is local on , we may restrict to the case . Then , where is an -algebra that is integral over (6.1.4); moreover, is reduced (I, 5.1.4) and the hypothesis that is dense in implies that the homomorphism corresponding to is injective (I, 1.2.7). To say that under these conditions means that every prime ideal of is the trace on of a prime ideal of , which is none other than the first theorem of Cohen–Seidenberg ([13], t. I, p. 257, th. 3).

Corollary.

Every finite morphism is projective.

Proof. Since is affine, is a very ample -module relative to (5.1.2); furthermore, is a quasi-coherent -module of finite type (6.1.2); finally, is separated, of finite type, and universally closed (6.1.10), so we are in the conditions of application of criterion (5.5.4, (i)).

Proposition.

Let be a finite morphism, and let (which is a quasi-coherent -algebra and an -module of finite type). Let be a quasi-coherent -module; for to be locally free of rank , it is necessary and sufficient that be a locally free -module of rank .

Proof. It is clear that if is isomorphic to ( open in ), then is isomorphic to (1.4.2). Conversely, suppose that is locally free of rank , and let us show that is locally isomorphic to as a -module. Let be a point of ; as runs over a fundamental system of affine neighbourhoods of , runs over a fundamental system of affine neighbourhoods (1.2.5) of the finite set , since is closed (6.1.10). The proposition then follows from the following lemma:

Lemma.

Let be a prescheme, a locally free -module of rank , and a finite part of contained in an affine open . Then there exists a neighbourhood of such that is isomorphic to .

Proof. We may evidently assume affine; for every , there exists in the closure at least one closed point (0, 2.1.3); if is the set of the , every neighbourhood of is a neighbourhood of , and we may therefore assume that is closed in and discrete. Consider the closed reduced subprescheme of having as its underlying space (I, 5.2.1) and let be the canonical injection; is locally free of rank on the discrete scheme , and is therefore isomorphic to ; in other words, there exist sections () of over such that the homomorphism defined by these sections is bijective. But is affine, is defined by an ideal of , and we have , where is an -module; the are elements

of and are therefore images of elements . For every there is then a neighbourhood of such that the restrictions of the to define an isomorphism (0, 5.5.4); the neighbourhood given by the union of the therefore meets the question.

Proposition.

Let be an integral morphism of preschemes, a normal locally integral prescheme, and a rational map from to such that is an everywhere defined rational map (I, 7.2.1). Then is everywhere defined.

Proof. If and are two morphisms (from dense opens of to ) in the class , it is clear that and are equivalent morphisms, which justifies the notation for their equivalence class. We recall also that, if we further suppose to be locally Noetherian, then the hypothesis that is normal already implies that is locally integral (I, 6.1.13).

To prove (6.1.13), let us first note that the question is local on , and so we may assume that there exists in the class a morphism . Consider the inverse image , and note that the morphism is integral (6.1.5, (iii)). In view of the correspondence between rational maps from to and rational -sections of (I, 7.1.2), we see that we are reduced to proving the particular case of (6.1.13) in which ; in other words:

Corollary.

Let be a normal locally integral prescheme, an integral morphism, and a rational -section of . Then is everywhere defined.

Proof. Since the question is local on , we may assume integral, and then is identified with a morphism from an open of to (I, 7.2.2) which is a -section of . Since is separated, is a closed immersion of into (I, 5.4.6); let be the closed subprescheme of associated to (I, 4.2.1), which is isomorphic to , hence integral; let be the reduced subprescheme of whose underlying space is the closure of in (I, 5.2.1); then is a subprescheme induced on an open of (I, 5.2.3), and, being irreducible, so is , which is therefore integral. The morphism can then be considered as a rational -section of ; since the restriction of to is an integral morphism (6.1.5, (i) and (ii)), we are finally reduced to proving (6.1.14) in the particular case ; in other words:

Corollary.

Let be a normal integral prescheme, an integral prescheme, and an integral morphism. If there exists a rational -section of , then is an isomorphism.

Proof. Since the question is local on , we may assume affine of integral ring , and then is affine of ring integral over (6.1.4) and integral; furthermore, the argument of (6.1.14) shows that there exists a dense open in isomorphic to a dense open of , hence and have the same field of fractions. On the other hand, by (I, 8.2.1.1) and the hypothesis that the are integrally closed, the ring is integrally closed, hence , which completes the proof of (6.1.13).

6.2. Quasi-finite morphisms

Proposition.

Let be a morphism locally of finite type, a point of . The following conditions are equivalent:

  • a) The point is isolated in its fibre .
  • b) The ring is a quasi-finite -module (0, 7.4.1).

Proof. Since the question is evidently local on and on , we may assume and affine, with a -algebra of finite type (I, 6.3.3). Moreover, we may replace by without changing the fibre or the local ring (I, 3.6.5); we may therefore assume that is a local ring (equal to ); if is the maximal ideal of , then is an affine scheme of ring , of finite type over (I, 6.4.11). This being so, if a) is satisfied, we may further assume that is reduced to the point ; then is of finite rank over (I, 6.4.4), that is, is a quasi-finite -module. Conversely, if this is so, is an Artinian affine scheme, hence discrete (I, 6.4.4); consequently is isolated in its fibre, which shows that b) implies a).

Corollary.

Let be a morphism of finite type. The following conditions are equivalent:

  • a) Every point is isolated in its fibre (in other words, the subspace is discrete).
  • b) For every , the prescheme is a finite -prescheme.
  • c) For every , the ring is a quasi-finite -module.

Proof. The equivalence of a) and c) follows from (6.2.1). On the other hand, since is a -algebraic prescheme (I, 6.4.11), the equivalence of a) and b) follows from (I, 6.4.4).

Definition.

When is a morphism of finite type satisfying the equivalent conditions of (6.2.2), we say that is quasi-finite, or that is quasi-finite over .

It is clear that every finite morphism is quasi-finite (6.1.8).

Proposition.

  • (i) An immersion which is closed, or such that is Noetherian, is a quasi-finite morphism.
  • (ii) If and are quasi-finite morphisms, then is quasi-finite.
  • (iii) If and are -preschemes and a quasi-finite -morphism, then is quasi-finite for every base extension .
  • (iv) If and are two quasi-finite -morphisms, then
      f ×_S g : X ×_S X' → Y ×_S Y'
    
    is quasi-finite.
  • (v) Let and be morphisms such that is quasi-finite; if, further, is separated, or is Noetherian, or is locally Noetherian, then is quasi-finite.
  • (vi) If is quasi-finite, so is .

Proof. If is an immersion, then every fibre is reduced to a point, and assertion (i) therefore follows from (I, 6.3.4 (i) and 6.3.5). To prove (ii), we remark first that is of finite type (I, 6.3.4 (ii)); furthermore, if and , then is isolated in , so there exists an open neighbourhood of in meeting no point of other than ; therefore is an open neighbourhood of meeting

none of the , where is in ; since is isolated in , it is isolated in . To prove (iii), we may restrict to the case (I, 3.3.11); we remark again first that is of finite type (I, 6.3.4, (iii)); on the other hand, if and if we set and , then is identified with (I, 3.6.5); since is of finite rank over by hypothesis, is of finite rank over , hence discrete. The assertions (iv), (v), (vi) follow from (i), (ii), (iii) by the general method (I, 5.5.12), except when the hypotheses in (v) are other than " separated"; to treat this case, we remark first that if is isolated in , it is a fortiori isolated in ; the fact that is of finite type then follows from (I, 6.3.6).

Proposition.

Let be a complete local Noetherian ring, an -scheme locally of finite type, a point of over the closed point of , and suppose that is isolated in its fibre (where is the structure morphism ). Then is an -module of finite type, and is -isomorphic to the sum (I, 3.1) of (which is a finite -scheme) and an -scheme X''.

Proof. It follows from (6.2.1) that is a quasi-finite -module. Since is Noetherian (I, 6.3.7) and the homomorphism is local, the hypothesis that is complete implies that is an -module of finite type (0, 7.4.3). Let be the local scheme of at the point (I, 2.4.1), and let be the canonical morphism. Since the composite is finite (6.1.1) and is separated, is finite (6.1.5, (v)), hence is closed in (6.1.10); on the other hand, since is of finite type, is a local isomorphism at the closed point of by the definition of and (I, 6.5.4); but since is the only open neighbourhood of , this means that is an open immersion, hence is also open in , which completes the proof.

Corollary.

Let be a complete local Noetherian ring, , a separated quasi-finite morphism. Then is -isomorphic to a sum , where is a finite -scheme and X'' a quasi-finite -scheme such that, if is the closed point of , .

Proof. Indeed, the fibre is finite and discrete by hypothesis, and the corollary therefore follows, by induction on the number of points of this fibre, from (6.2.5).

Remark.

In Chapter V, we will see that if is locally Noetherian, then a separated quasi-finite morphism is necessarily quasi-affine.

6.3. Integral closure of a prescheme

Proposition.

Let be a ringed space, a (commutative) -algebra, a section of over . The following properties are equivalent:

  • a) The sub--module of generated by the for (0, 5.1.1) is of finite type.
  • b) There exists a sub--algebra of that is an -module of finite type, such that .
  • c) For every , is integral over the fibre .

Proof. Since the sub--module of generated by the is an -algebra, it is clear that a) implies b). On the other hand, b) implies that for every , the -module is of finite type, which implies that every element of the algebra , and in particular , is integral over . Finally, if for some we have a relation of the form

  f_x^n + (a₁)_x f_x^{n−1} + … + (a_n)_x = 0

where the () are sections of over an open neighbourhood of , then the section is zero over a neighbourhood of , whence it follows immediately that all the () are linear combinations with coefficients in of the for ; we conclude that c) implies a).

When the equivalent conditions of (6.3.1) are satisfied, we say that the section is integral over .

Corollary.

Under the hypotheses of (6.3.1), there exists a (unique) sub--module of such that, for every , is the set of germs that are integral over . For every open , the sections of over are the sections that are integral over .

Proof. The existence of is immediate, by taking to be the set of such that is integral over for every . The second assertion follows immediately from (6.3.1).

It is clear that is a sub--algebra of ; we say that it is the integral closure of in .

(6.3.3)

Let , be two ringed spaces and

  g = (ψ, θ) : X → Y

a morphism. Let (resp. ) be an -algebra (resp. a -algebra) and let

be a -morphism (0, 4.4.1). Then, if (resp. ) is the integral closure of (resp. ) in (resp. ), the restriction of to is a -morphism

Indeed, if is the canonical injection , it suffices to show that

  v = u^♯ ∘ g*(j) : g*(ℬ') → 𝒞

sends into . Now, an element of is integral over by the definition of , hence so is its image under , which establishes our assertion.

Proposition.

Let be a prescheme, a quasi-coherent -algebra. Then the integral closure of in is a quasi-coherent -algebra, and for every affine open of , is the integral closure of in .

Proof. We may restrict to the case where is affine and , where is a

-algebra; let be the integral closure of in . Everything comes down to seeing that for every , an element of integral over necessarily belongs to , which follows from the fact that, for a commutative ring , the operations of integral closure in a -algebra and of passage to a ring of fractions (with respect to a multiplicative part of ) commute ([13], t. I, p. 261 and 257).

The -scheme is then called the integral closure of relative to (or relative to ); it is clear that is integral over (6.1.2).

We immediately deduce from (6.3.4) that if is the structure morphism, then, for every open of , is the integral closure of the prescheme induced by on , relative to .

(6.3.5)

Let , be two preschemes, a morphism, (resp. ) a quasi-coherent -algebra (resp. quasi-coherent -algebra), and let be an -morphism. We have seen (6.3.3) that one deduces from it an -morphism , where (resp. ) is the integral closure of (resp. ) in (resp. ). Consequently, if (resp. ) is the integral closure of (resp. ) relative to (resp. ), one canonically deduces from a morphism (1.5.6) making the diagram

  X' ─────f'────→ Y'
  │                │                                                       (6.3.5.1)
  ↓                ↓
  X ──────f─────→ Y

commute.

(6.3.6)

Suppose that has only a finite number of irreducible components (), with generic points , and consider in particular the integral closure of relative to a quasi-coherent -algebra (quasi-coherent as an -algebra or as an -algebra, which amounts to the same thing). We know (I, 7.3.5) that is a direct sum of quasi-coherent -algebras , with the support of contained in , and the sheaf induced by on being a simple sheaf whose fibre is an algebra over . It is then clear (6.3.4) that the integral closure of in is the direct sum of the integral closures of in each of the , and consequently the integral closure of relative to is an -scheme that is the sum of the ().

Suppose further that the -algebra is reduced, or, equivalently, that each of the algebras is reduced, and hence may be considered as an algebra over the field (equal to the field of rational functions of the reduced subprescheme of having as its underlying space); then (1.3.8) each of the is a reduced -scheme and is also the integral closure of . Suppose moreover that each of the algebras is a direct sum of a finite number of fields (); if is the sub-algebra of corresponding to , it is clear that is the direct sum of the integral closures of in each of the . Consequently, is then the sum of the -schemes (). Furthermore, under these hypotheses and with this notation:

Proposition.

Each of the is an integral normal -prescheme, and its field of rational functions is canonically identified with the algebraic closure of in .

Proof. By the above, we may assume integral, so and , so that the unique algebra A_1 is a field ; let be the generic point of , and let be the structure morphism. For every non-empty affine open of , is identified with the spectrum of the integral closure in the field of the integral ring (6.3.4); since the ring is integral and integrally closed, so are the local rings of the points of its spectrum, hence is by definition an integral and normal scheme ((0, 4.1.4) and (I, 5.1.4)). Moreover, since (0) is the only prime ideal of above the prime ideal (0) of B_U ([13], t. I, p. 259), reduces to a single point , and is the field of fractions of , which is none other than the algebraic closure of in . Finally, is irreducible, because as runs over the non-empty affine opens of , the constitute an open cover of by irreducible opens; moreover, the intersection of two such opens contains , hence is non-empty, and we conclude by (0, 2.1.4).

Corollary.

Let be a reduced prescheme having only a finite number of irreducible components (), and let be the generic point of . The integral closure of relative to is the sum of -schemes which are integral and normal. If is the structure morphism, then reduces to the generic point of and we have ; in other words, is birational.

In this case we say that is the normalisation of the reduced prescheme ; note that , being birational and integral, is surjective (6.1.10). In order to have , it is necessary and sufficient that be normal. When is an integral prescheme, it follows from (6.3.8) that its normalisation is integral.

(6.3.9)

Let , be two integral preschemes, a dominant morphism, , the fields of rational functions of and ; there canonically corresponds to an injection , and when we identify (resp. ) with the simple sheaf (resp. ), this injection is an -morphism. Let (resp. ) be an extension of (resp. ) and suppose given a monomorphism such that the diagram

  K₁ ────→ L₁
  ↑         ↑
  K ─────→ L

commutes; if (resp. ) is considered as a simple sheaf on (resp. ), hence as an -algebra (resp. an -algebra), this means that is an -morphism. With this, if (resp. ) is the integral closure of (resp. ) relative to (resp. ), then (resp. ) is an integral normal prescheme (6.3.6) whose field of rational functions is canonically identified with the algebraic closure (resp. )

of (resp. ) in (resp. ), and there exists a canonical (necessarily dominant) morphism making the diagram (6.3.5.1) commute.

The most important case is that in which one takes , so is then an extension of contained in , and one supposes integral and normal, hence . What precedes then shows that when is normal, and is the integral closure of relative to a field , every dominant morphism factors as

  f : X →^{f'} Y' → Y

where is dominant; moreover, when the monomorphism is given, is necessarily unique, as one sees by reducing to the case where and are affine. We thus see that, for the data of , , and a -monomorphism , the integral closure of relative to is the solution of a universal problem.

Remark.

Let us take up the hypotheses of (6.3.6), supposing further that each of the algebras is of finite rank over (which implies that is a direct sum of a finite number of fields); we can assert in certain cases that the structure morphism is not only integral, but even finite. Let us restrict to the case where is reduced; since the question is local on , we may further assume that is affine of ring , and that has only a finite number of minimal ideals () such that the are integral; then will be finite over if the integral closure of each in every extension of finite degree of its field of fractions is a -module of finite type (6.3.4). We know that this condition is always satisfied when is an algebra of finite type over a field ([13], t. I, p. 267, th. 9), or over ([9, I, p. 93, th. 3]), or over a complete local Noetherian ring ([25], p. 298). We conclude that will be a finite morphism whenever is a scheme of finite type over a field, over , or over a complete local Noetherian ring.

6.4. Determinant of an endomorphism of an -module

(6.4.1)

Let be a (commutative) ring, a free -module of rank , an endomorphism of ; we recall that to define the characteristic polynomial of , one considers the endomorphism of the free A[T]-module of rank , ( an indeterminate), and one sets

  P(u, T) = det(T · I − (u ⊗ 1))                                            (6.4.1.1)

( the identity automorphism of ). We have

  P(u, T) = T^n − σ₁(u) T^{n−1} + … + (−1)^n σ_n(u)                         (6.4.1.2)

where is an element of , equal to a homogeneous polynomial of degree (with integer coefficients) in the entries of the matrix of relative to an arbitrary basis of ; we say that the are the elementary symmetric functions of , and in particular we have and . Recall that, by the Hamilton–Cayley theorem, we have

  P(u, u) = u^n − σ₁(u) u^{n−1} + … + (−1)^n σ_n(u) = 0                     (6.4.1.3)

which can also be written as

  (det u) · I_E = u · Q(u)                                                  (6.4.1.4)

(I_E the identity automorphism of ), with

  Q(u) = (−1)^{n+1} (u^{n−1} − σ₁(u) u^{n−2} + … + (−1)^{n−1} σ_{n−1}(u)).  (6.4.1.5)

Let be a ring homomorphism, making an -algebra; consider the -module , which is free of rank , and the extension of to an endomorphism of ; it is immediate that one has for every index .

(6.4.2)

Now suppose that is an integral ring, with field of fractions , and let be an -module of finite type (but not necessarily free). Let be the rank of , i.e. the rank of the free -module ; to every endomorphism of canonically corresponds the endomorphism of ; by abuse of language, we still call characteristic polynomial of , and denote by , the polynomial , whose coefficients (which belong to ) we also denote by and call the elementary symmetric functions of ; in particular by definition. With this notation, formulas (6.4.1.3) to (6.4.1.5) make sense and remain valid, provided one interprets as the homomorphism obtained by composing the endomorphism of with the canonical homomorphism .

If is the torsion submodule of and if , then , hence, by passage to quotients, gives an endomorphism of ; moreover, is identified with and with , hence for .

When is torsion-free, is canonically identified with a sub--module of , and the relation is equivalent to . When is a free -module, the two definitions of given in (6.4.1) and (6.4.2) coincide by the above, which justifies the notation adopted.

Note also that when is a torsion module — in other words, — the exterior algebra of reduces to and the determinant of the unique endomorphism of is equal to 1.

Proposition.

Let be an integral ring, an -module of finite type, an endomorphism of ; then the elementary symmetric functions of (and in particular det u) are elements of integral over .

Proof. Let be the integral closure of ; since A'[T] is an integrally closed ring ([24], p. 99), it is the integral closure of A[T] in its field of fractions . Replacing by and by A[T], we see that we are reduced to proving that det u is integral over . If is the rank of , we have and , so we may assume . But then the map is a homomorphism from the -module to ; since is of finite type, is isomorphic to a sub--module of the -module of finite type (if admits a system of generators), so the elements det u belong to a sub--module of of finite type, and are consequently integral over .

Corollary.

Under the hypotheses of (6.4.3), if we further suppose normal, then the (in particular Tr u and det u) belong to .

Proposition.

Let be an integral ring, an -module of finite type of rank , an endomorphism of such that the belong to . For to be an automorphism of , it is necessary that det u be invertible in ; this condition is sufficient when is torsion-free.

Proof. The condition is sufficient, since the hypothesis and the formulas (6.4.1.4) and (6.4.1.5) (valid in , and not only in , since is torsion-free) prove that is the inverse of .

The condition is necessary, since if is invertible, it follows from (6.4.3) that belongs to the integral closure of in its field of fractions , and is evidently the inverse of det u in . Our assertion then follows from:

Lemma.

Let be a subring of a ring such that is integral over . If an element is invertible in , it is also invertible in .

Proof. In the contrary case, would belong to a maximal ideal of , and it follows from the first theorem of Cohen–Seidenberg ([13], t. I, p. 257, th. 3) that there would be a maximal ideal of such that ; we would then have , which is absurd.

Corollary.

Let be an integral and integrally closed ring, a torsion-free -module of finite type, an endomorphism of . For to be an automorphism of , it is necessary and sufficient that det u be invertible in .

Proof. This follows from (6.4.4) and (6.4.5).

Remark.

We will need below a generalisation of the preceding results. Consider a reduced Noetherian ring ; let () be its minimal ideals, the field of fractions of the integral ring , the total ring of fractions of , the direct sum of the fields . Let be an -module of finite type, and suppose that is a free -module of rank (which here is no longer a consequence of the other hypotheses); it amounts to the same to say that all the -vector spaces have the same dimension ; if then is an endomorphism of , we again set and , and in particular ; the are thus elements of . It is immediate that is the direct sum of the and that the latter are stable under ; the restriction of to is in fact the extension of to this -vector space; we deduce that is the element of whose components in the are the . Since the integral closure of in is the direct sum of the integral closures of in the , the are integral over .

Lemma.

The sub--algebra of generated by all the elements (), as runs over , is an -module of finite type.

Proof. It suffices to prove that the sub-A[T]-algebra of K[T] generated by the is an A[T]-module of finite type, since if the () form a system of generators of this A[T]-module, then the coefficients of the are integral over , hence generate an -algebra which is an -module of finite type ([13], t. I, p. 255, th. 1). We may therefore

replace by A[T] (which is Noetherian) and by , which is such that is a free K[T]-module of rank . Returning to the initial notation, we therefore see that it suffices to prove that the -module generated by the elements det u, as runs over , is of finite type; a fortiori (since every submodule of an -module of finite type is of finite type) it suffices to prove that as runs over the set of endomorphisms of , the -module generated by the det v is of finite type; in other words, we are again reduced to the case . But then the proposition follows from the fact that is an -module of finite type, and that is a homomorphism from this -module to .

Let be the kernel of the canonical homomorphism and let ; we see as above that is identified with , that , and that if is the endomorphism of deduced from by passage to quotients, then is identified with , and for every . If we have , then formulas (6.4.1.3) and (6.4.1.5) make sense and remain valid when is identified with a submodule of and the with homomorphisms ; consequently, Proposition (6.4.5) also extends to this case, with the same proof.

(6.4.8)

Let be a ringed space, a locally free -module (of finite rank), an endomorphism of . There exists by hypothesis a base of the topology of such that for every , is isomorphic to (for some that may vary with ). Let be an endomorphism of ; for every , is thus an endomorphism of the -module , which is free by hypothesis; the determinant is therefore defined and belongs to . Moreover, if form a basis of , their restrictions to any open form a basis of over , so is the restriction of to . There exists therefore a unique section of over , which we denote by det u and call the determinant of , such that the restriction of det u to every is . It is clear that for every we have ; for two endomorphisms , of , we have

  det(v ∘ u) = (det v)(det u)                                               (6.4.8.1)

as well as

and, if the rank of is constant (which will be the case (0, 5.4.1) if is connected) and equal to ,

  det(s · u) = s^n det u                                                    (6.4.8.3)

for every (note that if , but for ). Moreover, for to be an automorphism of , it is necessary and sufficient that det u be invertible in .

If the rank of is constant, one can define in the same way the elementary symmetric functions , which are elements of ; the relations (6.4.1.3) to (6.4.1.5) still hold.

We have thus defined a homomorphism of multiplicative monoids; if we note that by definition, we see that we can replace in this definition by an arbitrary open of , which immediately defines a homomorphism of sheaves of multiplicative monoids. When has constant rank, one similarly defines homomorphisms of sheaves of sets; for , the homomorphism is a homomorphism of -modules.

Let be a second ringed space, and let be a morphism of ringed spaces; if is a locally free -module, then is a locally free -module (of the same rank as if the latter is of constant rank) (0, 5.4.5). For every endomorphism of , is an endomorphism of , and it follows immediately from the definitions that is the section of over which canonically corresponds to . We can also say that the homomorphism is the composite

  f*(𝓗𝓸𝓶_ℬ(ℱ, ℱ)) →^{γ^♯} 𝓗𝓸𝓶_𝒜(f*(ℱ), f*(ℱ)) →^{det} 𝒜

(0, 4.4.6). We have analogous results for the .

(6.4.9)

Suppose now that is a locally integral prescheme, so that the sheaf of rational functions on is a locally simple sheaf of fields (I, 7.3.4), quasi-coherent as an -module. If is a quasi-coherent -module of finite type, then is a locally free -module (I, 7.3.6); for every endomorphism of , is then an endomorphism of , and is a section of over , which we still call determinant of and still denote by det u. It follows from (6.4.3) that det u is a section of the integral closure of in (6.3.2); if, further, is normal, then det u is a section of over , and if we further suppose that is torsion-free, for to be an automorphism of , it is necessary and sufficient that det u be invertible, by virtue of (6.4.6). The formulas (6.4.8.1) to (6.4.8.3) still hold; from the homomorphism , applied to the modules of sections of , one deduces a homomorphism of sheaves , which takes its values in when is normal. We have analogous definitions and results for the other elementary symmetric functions when is of constant rank; if moreover is normal, the are sections of over .

Finally, let , be two integral preschemes, and let be a dominant morphism. We know that there then exists a canonical homomorphism (I, 7.3.8), whence we obtain, for every quasi-coherent -module of finite type, a canonical homomorphism . If is an

endomorphism of , then is an endomorphism of , and one has a commutative diagram

  f*(ℱ ⊗_{𝒪_Y} ℛ(Y)) ─────f*(v⊗1)────→ f*(ℱ ⊗_{𝒪_Y} ℛ(Y))
        │                                       │
       θ↓                                      θ↓
  f*(ℱ) ⊗_{𝒪_X} ℛ(X) ───f*(v)⊗1────→ f*(ℱ) ⊗_{𝒪_X} ℛ(X)

One then concludes easily that is the canonical image, under the homomorphism , of the section det v of ; indeed, one reduces immediately to the case where and are affine, and being integral rings with fields of fractions and respectively, the homomorphism being injective and extending therefore to a monomorphism ; if , where is a -module of finite type, the rank over of is equal to that of over , and is the image in of for every endomorphism of , whence the conclusion.

(6.4.10)

Suppose finally that is a locally Noetherian reduced prescheme, so that the sheaf of rational functions on is again a quasi-coherent -module (I, 7.3.4); let furthermore be a coherent -module such that is locally free (of finite rank). By virtue of (6.4.7), if is of constant rank, we can, for every endomorphism of , define the , which are sections of over . When is no longer assumed of constant rank, we can still define the homomorphism .

6.5. Norm of an invertible sheaf

(6.5.1)

Let be a ringed space and let be a (commutative) -algebra. The -module is canonically identified with a sub--module of , a section of over an open of being identified with multiplication by this section. If and satisfy one of the conditions enumerated in (6.4.8), (6.4.9), or (6.4.10), we can therefore define (and in certain cases the ), which are sections of or of over , which we shall call the norm of (resp. the elementary symmetric functions of ) and denote by . We shall suppose that one of the following conditions is satisfied:

  • (I) is a locally free -module (of finite rank).
  • (II) is a locally Noetherian reduced prescheme, is a coherent -module such that is a locally free -module, and, for every section over an open , is a section of over .

Note that this last condition is automatically satisfied when the locally Noetherian prescheme is normal (6.4.9).

On the other hand, the hypothesis that is locally free can also be expressed as follows: denote by the closed reduced subpreschemes of having as their underlying spaces the irreducible components of (I, 5.2.1), which are therefore locally Noetherian integral preschemes. Every belongs to a finite number of the subspaces ; on the other hand, is a locally free -module of constant rank (I, 7.3.6); to say that is a locally free -module means that, for every , the ranks corresponding to the indices such that are equal. The question is in fact local, and one reduces to the case , where is a reduced Noetherian ring, and , where is a -algebra that is a -module of finite type; if () are the minimal prime ideals of , the total ring of fractions of is the direct sum of the fields of fractions of the integral rings , and is the direct sum of the , whence the conclusion.

It is clear that under hypotheses (I) or (II), one thus defines a homomorphism of sheaves of multiplicative monoids , also denoted by if no confusion may arise, and called the norm homomorphism. For two sections , of over the same open , one therefore has

for the corresponding sections of over ;

finally, for every section of over ,

if the rank of is constant and equal to (for , this formula gives if , and if ).

Under hypothesis (I), for to be invertible it is necessary and sufficient that be so. Under hypothesis (II), this condition is necessary; it is sufficient (by (6.4.7)) when one supposes that is injective and that the following more restrictive hypothesis is satisfied:

  • (II bis) is a locally Noetherian reduced prescheme, is a coherent -module such that is a locally free -module, and, for every section over an open such that is of constant rank over , the () are sections of over .

(Again note that this condition is satisfied when is normal.)

(6.5.2)

Suppose that one of the hypotheses (I), (II) of (6.5.1) is satisfied, and let be an invertible -module. We shall associate to it canonically (up to unique isomorphism) an invertible -module in the following way. Denote by (resp. ) the subsheaf of (resp. ) such that (resp. ) is the set of invertible

elements of (resp. ) for every open ; these are sheaves of multiplicative groups, and , restricted to , is a homomorphism of sheaves of groups (6.5.1). Let be the set of pairs having the following property: is an open of and is an isomorphism of -modules. By hypothesis, the form a cover of ; for any two indices , , one sets , which is an automorphism of , canonically identified with a section of over , and is a 1-cocycle of the cover with values in (0, 5.4.7). The fact that is a homomorphism then implies that is a 1-cocycle of with values in , and therefore corresponds (up to unique isomorphism) to an invertible -module (0, 5.4.7) which we shall denote by and call the norm of the invertible -module .

Let be a subset of such that the corresponding still form a cover of , and let be this cover; the restriction of the cocycle to defines a 1-cocycle of with values in , the restriction of the 1-cocycle of ; it is clear that there is a canonical isomorphism from the invertible -module defined by this 1-cocycle of onto , which allows the invertible -module to be defined by any sub-cover of . This possibility immediately shows that, if , are two invertible -modules, then, by virtue of (6.5.1.1) and (6.5.1.2),

  N(ℒ'₁ ⊗_ℬ ℒ'₂) = N(ℒ'₁) ⊗_𝒜 N(ℒ'₂)                                       (6.5.2.1)

and

as well as

up to canonical isomorphisms. Moreover, it follows from (6.5.1.3) that if is an invertible -module and if is of constant rank over in case (I) (resp. of constant rank over in case (II)), one has, up to canonical isomorphism,

  N_{ℬ/𝒜}(ℒ ⊗_𝒜 ℬ) = ℒ^{⊗n}.                                               (6.5.2.4)

(6.5.3)

We now show that is a covariant functor in the category of invertible -modules. Indeed, let be a homomorphism of invertible -modules, and let be an open cover of such that for every one has -isomorphisms and ; there is then for each an endomorphism of such that , and one can evidently identify with a section of over (0, 5.1.1). Hence, for every pair of indices, the restrictions to of and coincide. We deduce that for the 1-cocycles and with values in corresponding to and , the relation

  ω_{λμ}^{(2)} h'_μ = h'_λ ω_{λμ}^{(1)}.

If we set , we therefore have the analogous relations

  N(ω_{λμ}^{(2)}) h_μ = h_λ N(ω_{λμ}^{(1)})

and consequently the define a homomorphism which we shall denote by or . Under hypothesis (I), for to be an isomorphism, it is necessary and sufficient (since this is a local question) that be one. Under hypothesis (II), this condition is again necessary; it is sufficient if hypothesis (II bis) is satisfied and is injective.

Taking in particular , the homomorphisms are identified (0, 5.1.1) with the sections of over , whence a canonical map

  N_{ℬ/𝒜} : Γ(X, ℒ') → Γ(X, N_{ℬ/𝒜}(ℒ')).

It also follows from (6.5.1.1) that if , , one has

  N(f'₁ ⊗ f'₂) = N(f'₁) ⊗ N(f'₂).                                          (6.5.3.1)

For every invertible -module and every section , one has, taking (6.5.2.4) into account,

when is of constant rank under hypothesis (I) (resp. when is of constant rank under hypothesis (II)). Finally, for the homomorphism corresponding to a section of over to be an isomorphism, it is necessary and sufficient that be a basis of for every ; under hypothesis (I), this condition is therefore equivalent to saying that is a basis of for every ; under hypothesis (II), this condition is again necessary, and it is sufficient when satisfies (II bis) and is injective.

(6.5.4)

Let , be two ringed spaces,

a morphism, an -algebra, the inverse image -algebra. One supposes that one of the following hypotheses is satisfied:

  • satisfies hypothesis (I) of (6.5.1).
  • and satisfy hypothesis (II) of (6.5.1), is a locally Noetherian reduced prescheme, and, if one denotes by and the closed reduced subpreschemes of and respectively having as underlying spaces the irreducible components of these spaces, the restriction of to each is a dominant morphism from into some .

Under these conditions, satisfies hypothesis (I) or hypothesis (II) of (6.5.1) respectively; the first assertion is immediate. To establish the second, it suffices to see that for every , the ranks of the for the such that are the same. But if the restriction of to is a dominant morphism into , the rank of is equal to that of (as one sees at once by reducing to the affine case as in (6.4.9)), whence our assertion by virtue of hypothesis (II) and (6.5.1).

This being so, it follows from (6.4.8), (6.4.9), and (6.4.10) that if is a section of over an open , the corresponding section of over , then is the section of over which corresponds to the section of over .

If is an invertible -module, one deduces from the preceding that if (which is an invertible -module), one has up to canonical isomorphism.

(6.5.5)

Suppose henceforth that is a prescheme. The data of a quasi-coherent -algebra that is an -module of finite type is then equivalent, as we know, to that of a finite morphism such that , defined up to -isomorphism (6.1.2 and 1.3.1). Furthermore, to give a quasi-coherent -module is equivalent to giving a quasi-coherent -module such that (1.4.3), and for to be invertible it is necessary and sufficient that be so (6.1.12). To translate the preceding results in terms of the finite morphism , it will be necessary to suppose either that is a locally free -module (of finite type), or that and satisfy hypothesis (II). For every invertible -module , one then sets

and one calls this the norm (relative to ) of . Similarly, if is a homomorphism of invertible -modules, one sets

In particular, for , one thus obtains a canonical map

  N_{X'/X} : Γ(X', ℒ') → Γ(X, N_{X'/X}(ℒ')).                                (6.5.5.3)

We leave to the reader most of these translations, and confine ourselves to spelling out the following:

Proposition.

Let be a finite morphism, and suppose that either is a locally free -module or that and satisfy (II bis) (which will be the case in particular when is locally Noetherian and normal). For a homomorphism of invertible -modules to be an isomorphism, it is necessary and sufficient, under the first hypothesis, that be one; under the second hypothesis, this condition is again necessary, and it is sufficient when the homomorphism is injective.

Proof. Note that here we use the fact that, for to be an isomorphism, it is necessary and sufficient that be one (1.4.2).

Corollary.

Let be a finite morphism, and suppose that either is a locally free -module or that and satisfy (II bis) and is injective. Let be an invertible -module, a

section of over , and the corresponding section of over (6.5.5.1). Then , and is the largest open subset of such that .

Proof. Indeed, is closed in (6.1.10); it therefore suffices to prove the last assertion. Now the relation is equivalent to the fact that the homomorphism defined by is an isomorphism. By (6.5.6), this is equivalent to saying that the homomorphism defined by is an isomorphism, that is, to the relation .

Proposition.

Let be a finite morphism, a morphism; let , , , so that one has the commutative diagram

  X' ←─f'── Y'
  │          │
  g↓         ↓g'
  X ←──f──── Y

Suppose that either is locally free, or that and satisfy (II), that is a locally Noetherian reduced prescheme, and that the restriction of to every irreducible component of is a dominant morphism into an irreducible component of . Then, for every invertible -module , one has

up to canonical isomorphism.

Proof. Note that one has by virtue of (1.5.2), and in particular ; if is locally free, then so is . The conclusion then follows from the definitions and (6.5.4).

Remark.

We shall later generalise the notion of norm developed above, and relate it to the notion of direct image of a divisor.

6.6. Application: criteria for ampleness

Proposition.

Let be a prescheme, a quasi-compact morphism, a finite and surjective morphism. Suppose that either is a locally free -module, or that and satisfy (II bis). Then, for every invertible -module that is ample for , is ample for .

Proof. We may assume affine (4.6.4), and then by virtue of (4.6.6) the statement is equivalent to:

Corollary.

Let be a quasi-compact prescheme, a finite surjective morphism such that either is a locally free -module or that and satisfy (II bis). Then, for every ample -module , is ample.

Proof. Under the second hypothesis, we may further assume that the canonical homomorphism is injective. Indeed, if not, let

be the kernel of this homomorphism, which is a coherent ideal of (I, 6.1.1), and set ; we then have a commutative diagram

  X'' ──h──→ X'
   \         /
   g'\      / g
       ↘  ↙
         X

where is a closed immersion (1.4.10). Furthermore, we know that the support of is a closed set (0, 5.2.2) rare in (I, 7.4.6), whence we conclude that for the generic point of an irreducible component of , there is an affine open neighbourhood of such that . Since is by hypothesis surjective, we thus conclude that ; is therefore dominant, and being a finite morphism, it is surjective (6.1.10); by definition, , so and satisfy (II bis), and moreover is injective. Finally, is an ample -module (4.6.13, (i bis)), and we have . Indeed, to define these two invertible -modules one can use a single affine open cover of such that the restrictions of and to are respectively isomorphic to and . One sees at once that to every isomorphism there canonically corresponds an isomorphism

so that, if and are the 1-cocycles corresponding to the systems of isomorphisms and (6.5.2), is the canonical image in of . By virtue of the definition of , one therefore concludes that

(with ), whence the asserted equality.

Suppose therefore that the homomorphism is injective when we are under hypothesis (II bis). It suffices to prove that, as runs over the sections of (for ) over , the form a base of the topology of (4.5.2). Now let and let be an arbitrary neighbourhood of ; since is finite (6.1.7) and is ample, there exists an integer and a section of over such that is a neighbourhood of contained in (4.5.4). Since

it suffices to take ; indeed, then (6.5.7), so .

Corollary.

Under the hypotheses of (6.6.1), for an invertible -module to be ample for , it is necessary and sufficient that be ample for .

Proof. The condition is necessary, since is affine (5.1.12). To prove that the condition is sufficient, we may assume affine (4.6.4), hence and quasi-compact and ample (4.6.6), and it remains to show that is ample. Now, the set of points

in a neighbourhood of which (resp. ) has a given rank in the first (resp. second) hypothesis is both open and closed in , so is the prescheme sum of a finite number of these opens, and we may therefore assume that it is equal to one of them (4.6.17). But then , hence is ample by virtue of (6.6.2), and so is (4.5.6).

Corollary.

Suppose that the hypotheses of (6.6.1) are satisfied and suppose further that is of finite type. Then, for to be quasi-projective, it is necessary and sufficient that be so. If, moreover, we suppose that is a quasi-compact scheme, or a prescheme whose underlying space is Noetherian, then, for to be projective, it is necessary and sufficient that be so.

Proof. The hypothesis implies that is of finite type. Taking into account the definition of quasi-projective morphisms (5.3.1), the first assertion follows from (6.6.1) and (6.6.3). Taking this result and (5.5.3, (ii)) into account, it remains to prove that when is assumed quasi-projective, for to be proper, it is necessary and sufficient that be proper. Now is then separated (5.3.1) and of finite type; since is surjective, our assertion follows from (5.4.2, (ii)) and (5.4.3, (ii)).

In particular:

Corollary.

Let be a prescheme of finite type over a field , a finite-degree extension of . For to be projective (resp. quasi-projective) over , it is necessary and sufficient that be projective (resp. quasi-projective) over .

Proof. The condition is indeed necessary (5.3.4, (iii) and 5.5.5, (iii)). Conversely, suppose it is satisfied, and let be the canonical projection. It is clear that is a finite morphism (6.1.5, (iii)) and surjective (I, 3.5.2, (ii)). Moreover, is a locally free -module, being isomorphic to (1.5.2). It then follows from the hypothesis and from (6.1.11) and (5.5.5, (ii)) that is projective (resp. quasi-projective) over ; we then deduce from (6.6.4) that is projective (resp. quasi-projective) over .

In Chapter V, we will show that the statement of (6.6.5) remains valid when is an arbitrary extension of .

The end of this section is devoted to the proof of the criterion (6.6.11), which is a rather technical refinement of (6.6.1); it may be omitted on a first reading.

Lemma.

Let be a reduced Noetherian prescheme, a coherent -module such that is a locally free -module of constant rank . Then there exists a reduced Noetherian prescheme and a finite birational morphism having the following property: the morphisms of sheaves of sets () (cf. (6.4.10)) send into the coherent -algebra .

Proof. Consider an affine open of of ring ; let , and let C_U be the sub-algebra of generated by the as runs over ; we have seen (6.4.7.1) that this -algebra is of finite rank. Moreover, it is clear that the formation of the algebras C_U commutes with the restriction operations from an affine open to an affine open . We have thus defined a finite sub--algebra of such that for every affine open of . We

shall take , and for the structure morphism, which is therefore finite (6.1.2); since is reduced, is a reduced Noetherian prescheme (1.3.8). Finally, the total ring of fractions of C_U is by definition, and since C_U is contained in the integral closure of in , there is a bijective correspondence between minimal prime ideals of and minimal prime ideals of C_U ([13], t. I, p. 259), which proves that is birational and completes the proof.

Corollary.

Under the hypotheses of (6.6.6), let be an open of such that, for every , either is normal at the point , or is a free -module. Then we may suppose defined so that the restriction of to is an isomorphism of onto .

Proof. Indeed, either hypothesis implies that if is an affine open, then, with the notation of (6.6.6), for every (6.4.3), so , and the conclusion follows from the definition of given in (6.6.6).

(6.6.8)

Let be a reduced Noetherian prescheme, a finite surjective morphism, so that is a coherent -algebra, and suppose moreover that is a locally free -module of constant rank . We can then apply the lemma (6.6.6) taking , whence, with the notation of (6.6.6), a homomorphism of sheaves of multiplicative monoids , and composing this homomorphism with the canonical homomorphism (6.5.1), one therefore obtains a homomorphism of sheaves of multiplicative monoids:

  N' : ℬ = g_*(𝒪_{X'}) → h_*(𝒪_Z) = 𝒞.                                     (6.6.8.1)

This being so, for every invertible -module , is an invertible -module (6.1.12), and the method of (6.5.2) allows us to associate to functorially an invertible -module, which we shall denote by .

Lemma.

Let be a reduced Noetherian prescheme, a finite surjective morphism such that is a locally free -module of constant rank. Then there exists a reduced Noetherian prescheme and a finite birational morphism having the following property: for every ample -module , the invertible -module such that (notation of (6.6.8)) is ample.

Proof. Suppose first that the homomorphism is injective. Define and as in (6.6.6) (with ). Let ; we have to show that there exists an integer and a section of over such that is an affine open containing (4.5.2). Let , and let be an affine open of containing ; then is an affine open neighbourhood of , and it suffices to find such that , since will then necessarily be affine (5.5.8). There exists by hypothesis an integer and a section of over such that

by virtue of (4.5.4). By definition, is also a section of over , and there corresponds to it, as in (6.5.2), a section of over .

We shall see that if is the section considered as a section of over , then answers the question. Set

  V = X − g(X' − X'_{s'})                                                  (6.6.9.2)

which is an open of containing and contained in , by virtue of (6.6.9.1) and (6.1.10). We shall show that

which will complete the proof. It amounts to saying that the set of such that is invertible contains and is contained in . For this, consider first an affine open contained in ; is an affine open in , and by virtue of (6.6.9.2), is invertible for every ; by virtue of the hypotheses on and , one can apply the results of (6.4.7) and one sees that if , then is invertible; in other words, . On the other hand, it also follows from (6.4.7) that, conversely, if is invertible, so is , which implies by (6.6.9.1), and therefore , whence in this case.

One passes from this to the general case by the same reasoning as in (6.6.2), replacing by X'' such that is injective, and by an ample -module such that . The lemma (6.6.9) is therefore proven in all cases (with a suitable choice of ).

Corollary.

Suppose that the hypotheses of (6.6.9) are satisfied; for every invertible -module such that is ample, is ample.

Proof. Indeed, if we set , then (0, 5.4.10), so

  N'(g_*(ℒ')) = (ℒ ⊗_{𝒪_X} 𝒞)^{⊗n}

(by the same reasoning as for (6.5.2.4)). We conclude that , and since is ample, so is (4.5.6).

Proposition.

Let be an affine scheme, a reduced Noetherian prescheme, a quasi-compact morphism, a finite surjective morphism. Let be an open part of such that, for every , either is normal at the point , or there exists an open neighbourhood of such that is a locally free -module. Then there exists a reduced -prescheme and a finite and birational -morphism such that the restriction of to is an isomorphism and which possesses the following property: for every invertible -module such that is ample relative to , is ample relative to .

Proof. Since is affine, is ample, and it is then a matter of proving that for a suitable choice of , is ample (4.6.6). We shall see that we can replace by a finite surjective morphism such that is ample and is a locally free -module of constant rank; we will then be reduced to the conditions of (6.6.10) and the proposition will be proven.

For this, let ; denote by () the closed reduced subpreschemes of having as their underlying spaces the irreducible components of (I, 5.2.1); they are integral by hypothesis. Let be the closed

subprescheme of , the morphism restricted to , which is finite (6.1.5, (iii)) and surjective; let be the rank of the -algebra . Since is a constant presheaf (I, 7.3.5), the rank is also the rank of the -algebra for every open of meeting only the single irreducible component . If is an open set in such that is isomorphic to , it follows from the preceding remark that the numbers are equal to for all the indices such that . Let then be the open of formed by the points in a neighbourhood of which is a locally free -module, and let () be its connected components, which are open in and finite in number (since is Noetherian); denote by the closed subprescheme of , closure of the subprescheme induced on the open (I, 9.5.11). By the above, for all indices such that , the ranks are all equal to a single integer ; note moreover that a single cannot meet two of distinct indices. Let be the indices such that . Consider the product of all the , set , and let X'' be the prescheme defined as follows. For each (), one considers preschemes isomorphic to , and for each , preschemes isomorphic to ; X'' is the sum of all these preschemes. One defines a morphism reducing to the canonical injection on each of the summands of X''; it is clear that g'' is a finite dominant morphism, hence surjective (since a finite morphism is closed (6.1.10)); set , which is a finite surjective morphism ; one has , so is an ample -module (5.1.12). It is clear then that for this new prescheme X'', the ranks defined like the for are all equal to ; taking (I, 7.3.3) into account, one immediately concludes that for every affine open of , is an -module isomorphic to .

Corollary.

If, in the statement of (6.6.11), we have , then for an invertible -module to be ample relative to , it is necessary and sufficient that be ample relative to .

Remark.

In Chapter III, we will see that, if is Noetherian, of finite type, and if the restriction of to the closed reduced subprescheme of having as its underlying space is proper, then the conclusion of (6.6.12) is still valid. But we shall give in Chapter V examples of algebraic schemes over a field (the structure morphism not being proper) whose normalisation is quasi-affine, but which is not quasi-affine (so that is not ample, even though is (5.1.12) and the morphism is finite and surjective (6.3.10)). We shall see in the next section that this circumstance cannot occur when we replace "quasi-affine" by "affine".

6.7. Chevalley's theorem

We shall establish (with the help of Serre's criterion (5.2.1)) the following theorem, which was proven by C. Chevalley by other methods, in the case of algebraic schemes.

Theorem.

Let be an affine scheme, a Noetherian prescheme, and a finite surjective morphism. Then is an affine scheme.

Proof. It is clear that f_red : X_red → Y_red is finite (6.1.5, (vi)); since is an affine scheme, and saying that is affine is equivalent to saying that is affine (since is Noetherian (I, 6.1.7)), we see that we may suppose reduced. For every closed part of , there is then a single reduced subprescheme of having as its underlying space (I, 5.1.2); its inverse image , canonically isomorphic to (I, 4.4.1), is affine as a closed subprescheme of , and the restriction of to , which is identified with , is a finite surjective morphism (6.1.5, (iii)). By virtue of the principle of Noetherian induction (0, 2.2.2), we may therefore (taking (I, 6.1.7) into account) reduce to proving the theorem under the hypothesis that for every closed part , every closed subprescheme of having as its underlying space is affine. We thus conclude that, for every coherent -module whose (closed) support is distinct from , one has . Indeed, there exists a closed subprescheme of having as its underlying space and such that, if is the canonical injection, one has (I, 9.3.5); consequently (5.2.3) by (I, 5.1.9.2).

Suppose first that is not irreducible, and let be an irreducible component of and ; we again denote by the closed reduced subprescheme of having as its underlying space, and by the canonical injection . Let be a coherent -module, and consider the canonical homomorphism

  ρ : ℱ → ℱ' = j_*(j*(ℱ))

(0, 4.4.3); is a coherent -module by virtue of (0, 5.3.10) and (0, 5.3.12), since , denoting by the sheaf of ideals of defining the subprescheme . Consequently and are also coherent -modules (0, 5.3.4); now by definition the fibre of at the generic point of is equal to , since is interior to and therefore , since is reduced. We conclude that is not contained in the (closed) support of ; moreover, the support of (and a fortiori that of ) is contained in ; in other words, the supports of and are distinct from . We deduce that , and the exact sequence of cohomology applied to the exact sequence shows that . We conclude by Serre's criterion (5.2.1).

So suppose irreducible, and hence integral. We may also suppose integral: indeed, if we denote by the closed reduced subpreschemes of having as their underlying spaces the irreducible components of (I, 5.2.1) and by the restriction of to , then at least one of the is dominant, and since it is a finite morphism (6.1.5), it is surjective (6.1.10); since is also an affine scheme, we see that we may replace by in the statement.

Lemma.

Let , be two integral Noetherian preschemes, (resp. ) the generic point of (resp. ), a finite surjective morphism. Let be an invertible -module such that there exists an affine open neighbourhood of and a section for which

. Then there exist two integers , , a homomorphism , and an open neighbourhood of such that the restriction is an isomorphism of onto .

Proof. Let be the (integral) ring of , its field of fractions: since is finite, is affine (1.3.2); let be its (integral) ring of fractions ; by hypothesis is a -module of finite type (6.1.4), so is an extension of finite rank of . The fibre is identified with (I, 3.6.5); let () be elements of forming a basis of over . There exists such that the sections of over extend to sections () of over (I, 9.3.1). The are also, by definition, sections of over and therefore define a homomorphism (0, 5.1.1); we shall see that answers the question. One has , where is a -module of finite type, so if is the injection corresponding to the morphism given by the restriction of , then is a -module of finite type; consequently

(I, 1.6.3) is coherent, and since is an arbitrary affine open of , is coherent; moreover, , where is a -homomorphism , and is the homomorphism ; but the latter is by definition an isomorphism, since the form a basis of over , being by hypothesis a generator of . We conclude that the supports of the -modules Ker u and Coker u do not contain ; since these -modules are coherent (0, 5.3.3), their supports are closed (0, 5.2.2), whence the lemma.

This being so, the hypotheses of the lemma (6.7.1.1) are fulfilled in the case we are considering, by taking , since is affine (I, 1.1.10); we shall set and . By virtue of Serre's criterion (5.2.1.1), it suffices to prove that for every coherent -module , one has ; it even suffices to prove this when , which entails that is torsion-free since is integral; in fact we shall show that for every coherent torsion-free -module . Now the homomorphism defines a homomorphism

  v : 𝒢 = 𝓗𝓸𝓶_𝒜(ℬ, ℱ) → 𝓗𝓸𝓶_𝒜(𝒜^m, ℱ) = ℱ^m.

We first show that is injective: by hypothesis has a support not meeting , hence is a torsion -module (I, 7.4.6); the exact sequence

gives, by left exactness of the functor , the exact sequence

  0 → 𝓗𝓸𝓶_𝒜(𝒯, ℱ) → 𝒢 →^v ℱ^m.

But since is torsion-free, one has (0, 5.2.6), whence our assertion. We therefore have the exact sequence

  0 → 𝒢 → ℱ^m → Coker v → 0

where and Coker v are coherent -modules ((0, 5.3.4) and (0, 5.3.5)); by virtue of the exact sequence of cohomology, it will suffice to show that in order to deduce , and therefore . Now the restriction is an isomorphism, hence the support of Coker v is distinct from , whence by the hypothesis at the start. On the other hand, is a coherent -module (I, 9.6.4); since is affine over , there exists a quasi-coherent -module such that is isomorphic to (1.4.3); since is affine, one has (I, 5.1.9.2), hence also by (5.2.3), which completes the proof of theorem (6.7.1).

Corollary.

Let be a Noetherian prescheme, a finite cover of the space consisting of closed sets. For to be affine, it is necessary and sufficient that, for each , there exist a closed subprescheme of having as its underlying space and which is affine.

Proof. Indeed, if this is so, let be the scheme sum of the ; it is clear that is affine, and one defines a surjective morphism by taking as the restriction of to the canonical injection. Everything comes down to verifying that is finite, by virtue of (6.7.1), and this has been seen in (6.1.5).

Corollary.

For a Noetherian prescheme to be affine, it is necessary and sufficient that the closed reduced subpreschemes having as their underlying spaces the irreducible components of be affine.