§21. Divisors

On the content of the present section, see the comments of the introduction to §20. For the global properties of divisors, the reader is referred to the section devoted to them in chap. V.

21.1. Divisors on a ringed space

(21.1.1). Let be a ringed space, the sheaf of germs of meromorphic functions on (20.1.3), the sheaf (of multiplicative groups) of germs of regular meromorphic functions on (20.1.8). It is clear that the sheaf (of multiplicative groups) of germs of invertible sections of is canonically identified with a subsheaf (of multiplicative groups) of .

Definition (21.1.2).

One calls sheaf of divisors on and denotes by the quotient sheaf (of commutative groups) ; the sections of this sheaf over are called divisors on ; they form a commutative group denoted . For every section of over (in other words, every regular meromorphic function on (20.1.8)), one calls divisor of and denotes by (or ) the divisor on image of by the canonical homomorphism .

The support of a divisor is the closed set of such that . One denotes it .

For every open of , one obviously has , , hence , and consequently the sheaf is equal to the presheaf .

When is affine, one writes instead of .

(21.1.3). We shall always denote the group of divisors on additively. For two regular meromorphic functions , on , one therefore has

  (21.1.3.1)             div(fg) = div(f) + div(g),

  (21.1.3.2)             div(f⁻¹) = −div(f).

By definition, for every regular meromorphic function on , one has the equivalence

  (21.1.3.3)             div(f) = 0  ⟺  f ∈ Γ(X, 𝒪_X^×),

whence, for two regular meromorphic functions , on ,

  (21.1.3.4)             div(f) = div(g)  ⟺  f = ug, u ∈ Γ(X, 𝒪_X^×).

(21.1.4). Let now be an invertible -Module, and let be a regular meromorphic section (20.1.8) of over . Every possesses an open neighbourhood such that is isomorphic to , hence isomorphic to ; by one of these isomorphisms, corresponds to a section , and since two isomorphisms of onto differ only by multiplication by an element of , the element of is independent of the chosen isomorphism; it is clear that these elements (for variable ) are the restrictions of a section of over , which one calls the divisor of and denotes (such a divisor is not necessarily of the form for a regular meromorphic function on ; see (21.2.9)). For the definition of coincides with that of (21.1.2). If , are two invertible -Modules, (resp. ) a regular meromorphic section of (resp. ) over , it is immediate that one has

  (21.1.4.1)             div(s ⊗ s') = div(s) + div(s')

  (21.1.4.2)             div(s^⊗ n) = n · div(s)        for every n ∈ ℤ

( being the regular meromorphic section of over defined in (20.1.10)), and, for two regular meromorphic sections , of over , one has the relation

  (21.1.4.3)             div(s) = div(s')  ⟺  s' = ts, with t ∈ Γ(X, 𝒪_X^×).

(21.1.5). The sheaf (20.1.3) whose sections over an open of are the regular elements of is a subsheaf of monoids of ; one can write

If one denotes by the sheaf whose sections over are the inverses in of the elements of , it is clear that one has , hence

Definition (21.1.6).

The subsheaf of sets of that is the canonical image of the subsheaf of is denoted ; its sections over are called positive divisors on , and their set is denoted .

Since is a sheaf of (multiplicative) monoids, one has

  (21.1.6.1)             Div^+(X) + Div^+(X) ⊂ Div^+(X)

and on the other hand, by virtue of (21.1.5.2) and (21.1.3.3),

These two relations show that is the set of positive elements for an order structure on the group , compatible with this group structure; one denotes this order relation ; in other words, one has

We shall always assume in what follows that is endowed with this order structure; it is clear that , hence , and one can therefore say that defines on a structure of sheaf of ordered groups. The stalk at a point of the sheaf is a submonoid of the group , the set of elements for an order structure compatible with the group structure; for a divisor on , it amounts to the same to say that or that for every .

By definition, for every regular meromorphic function on , one has the relation

  (21.1.6.4)             div(f) ≥ 0  ⟺  f ∈ Γ(X, 𝒮(𝒪_X));

in other words, signifies that is a regular section of , or also a section of invertible in .

More generally, given a divisor on , the relation is equivalent to the following: for every open of such that , where , there exists a regular element of such that .

(21.1.7). Let be an invertible -Module, a regular meromorphic section of over ; one has the relation

  (21.1.7.1)             div(s) ≥ 0  ⟺  s ∈ Γ(X, ℒ) ∩ Γ(X, (𝓜_X(ℒ))^×)

as follows at once from the definitions (21.1.4) and (21.1.6).

Proposition (21.1.8).

Let be a locally Noetherian prescheme, a divisor on . Suppose that for every such that one has (resp. ). Then one has (resp. ).

The question being local on , one may assume that , being a

regular meromorphic function on ; the relation is equivalent to , hence the hypothesis means that if , one has for every (since contains the maximal points of ). Consequently (5.10.5) the restriction homomorphism is bijective, which shows that there exists a section of over such that . But by definition of , this implies , hence and . The assertion relative to the relation follows at once by applying what precedes to , by virtue of (21.1.6.2).

Corollary (21.1.9).

Let be a locally Noetherian prescheme, a divisor on . Let be the support of . Then, for every maximal point of , one has .

Indeed set ; in view of (20.2.11) and (20.3.6), the sheaf is induced on X_1 by , hence one may restrict to the case where , in which case, since and is a maximal point of , one necessarily has . If one had , one would conclude, by virtue of (21.1.8), that , which contradicts the definition of .

Proposition (21.1.10).

Let be a Noetherian local ring; for , it is necessary and sufficient that (in other words, that the maximal ideal of be associated to (0, 16.4.6)).

Indeed, to say that means that in every regular element is invertible, or also that all the elements of are zero-divisors, which means that (Bourbaki, Alg. comm., chap. IV, §1, n° 1, cor. 3 of prop. 2).

21.2. Divisors and invertible fractional Ideals

(21.2.1). Let be a ringed space. One calls fractional Ideal on a sub--Module of the -Module of germs of meromorphic functions on . A fractional Ideal on which is an invertible -Module is called an invertible fractional Ideal.

Proposition (21.2.2).

For a fractional Ideal on to be invertible, it is necessary and sufficient that for every , there exist an open neighbourhood of and a section such that .

The condition is obviously sufficient, the map from to being obviously bijective for every open . To see that it is necessary, note that by hypothesis there exist an open neighbourhood of and an isomorphism of -Modules . If is the image of the section by this isomorphism, one may assume, restricting , that , where and , and the considered isomorphism then makes correspond, to every section (where is an open contained in ) the section ; to say that the map thus defined is bijective means that is a regular element of , hence .

One will note that for every open of such that with , the section is determined up to multiplication by an element of , since the

multiplication by these elements provides all the automorphisms of the -Module .

Corollary (21.2.3).

(i) Let be an invertible fractional Ideal; then the invertible -Module is canonically identified with the fractional Ideal (transporter of into ) defined in the following way: for every open of such that , where , one has .

(ii) If and are two invertible fractional Ideals, the canonical map is an isomorphism of -Modules.

Assertion (ii) follows at once from (21.2.2). On the other hand, the remark made at the end of (21.2.2) shows that there exists indeed one and only one fractional Ideal defined by the condition of the statement; the canonical isomorphism of onto is obtained by making correspond to every section (where is an open contained in and ) the homomorphism from to .

By virtue of (21.2.3, (i)), one will generally identify the -Modules and , considering therefore as a sub--Module of .

(21.2.4). It follows from (21.2.3) that the set Id.inv(X) of invertible fractional Ideals on is endowed with a structure of commutative group for the composition law , the neutral element of this group being . It is clear that for every open of , one has , hence the restriction map is a homomorphism of groups ; one thus defines a presheaf of commutative groups ; it is immediate that in fact this presheaf is a sheaf of commutative groups, which one denotes .

(21.2.5). For every regular meromorphic function , it follows from (21.2.2) that is an invertible fractional Ideal, and one obviously has ; in other words the map is a homomorphism from the commutative group into the commutative group Id.inv(X). Replacing by any open of and noting that the homomorphisms obtained are compatible with the operations of restriction, one obtains a canonical homomorphism of sheaves of commutative groups:

Note that if , one has ; one deduces at once that the homomorphism factors as

  (21.2.5.2)             𝓜_X^× → 𝓜_X^× / 𝒪_X^× = 𝒟iv_X → 𝐼𝑑.𝑖𝑛𝑣_X

where is a homomorphism from the sheaf of additive groups into the sheaf of multiplicative groups ; one therefore has for every open of a homomorphism of commutative groups, such that, for every section , one has

One concludes that , for every divisor , is the invertible fractional Ideal defined in the following way: for every open of such that , where , is the invertible fractional Ideal . One therefore has, by virtue of (21.1.6), for every regular meromorphic function , the relation

  (21.2.5.4)             f ∈ Γ(X, 𝐼(D))  ⟺  div(f) ≥ D.

Proposition (21.2.6).

The homomorphism is bijective.

One defines a homomorphism from Id.inv(X) into by making correspond to every invertible fractional Ideal on the divisor defined in the following way: for every open of such that , where (21.2.2), one takes ; by virtue of the remark following (21.2.2), this definition is independent of the generator chosen in , and determines indeed a divisor on . Moreover, this definition shows at once that the homomorphisms and are reciprocal to one another. Replacing by any open , one deduces the definition of the isomorphism reciprocal to , whence the proposition. One will set , so that one has, for every regular meromorphic function on ,

(21.2.7). One will often identify the sheaves and (resp. the groups and Id.inv(X)) by means of the preceding isomorphisms and (resp. and ). One will note that one has the relation

  (21.2.7.1)             D ≥ 0  ⟺  𝐼_X(D) ⊂ 𝒪_X       for D ∈ Div(X)

as follows at once from the definitions (21.1.6) and (21.1.5.1); in other words, the image is the set of Ideals of (also sometimes called integral Ideals) that are invertible -Modules: such an Ideal is again characterized by the fact that for every , there is an open neighbourhood of in such that , where is a regular element of . The set of these Ideals is therefore a submonoid of Id.inv(X), equal to the set of positive elements for an order relation compatible with the group structure of Id.inv(X), and it is immediate that this relation is none other than the relation opposite to inclusion; in other words, one has

  (21.2.7.2)             ℐ_1 ⊂ ℐ_2  ⟺  div(ℐ_1) ≥ div(ℐ_2)

for , in Id.inv(X).

(21.2.8). For every divisor on , one sets

is therefore an invertible fractional Ideal, defined in the following way: for every open of

such that , where , is the invertible fractional Ideal ; by virtue of (21.1.6), for every regular meromorphic function , one has the relation

  (21.2.8.2)             f ∈ Γ(X, 𝒪_X(D))  ⟺  div(f) ≥ −D.

Moreover, it is clear that one has canonical isomorphisms (21.2.3)

  (21.2.8.3)             𝒪_X(0) = 𝒪_X,   𝒪_X(D + D') = 𝒪_X(D) ⊗ 𝒪_X(D'),
                         𝒪_X(nD) = (𝒪_X(D))^⊗ n,   𝒪_X(−D) = (𝒪_X(D))⁻¹

for every integer , and for any two divisors , on .

(21.2.9). Let be an invertible fractional Ideal on . The canonical injection defines by tensorization a homomorphism of -Modules

  (21.2.9.1)             𝓜_X(ℐ) = ℐ ⊗_{𝒪_X} 𝓜_X → 𝓜_X ⊗_{𝒪_X} 𝓜_X = 𝓜_X

which is an isomorphism: indeed, if is an open of such that , where , the homomorphism (21.2.9.1) restricted to is none other than the isomorphism that to every section of over makes correspond the section of the same sheaf. In the isomorphism (21.2.9.1), to the regular meromorphic sections of over correspond the regular meromorphic functions on .

Consider in particular the case where , where is a divisor on ; one then has a canonical isomorphism

and one denotes by the regular meromorphic section of over which corresponds by this isomorphism to the section 1 of . If is an open of such that , where , one has in ; since one then has , one deduces (21.1.4) that one has

On the other hand, one deduces at once from the canonical isomorphisms (21.2.8.3) the formulas

  (21.2.9.4)             s_0 = 1,   s_{D + D'} = s_D ⊗ s_{D'},   s_{nD} = s_D^{⊗ n}   (n ∈ ℤ).

(21.2.10). Let us consider, between two pairs , , where and are two invertible -Modules, (resp. ) a regular meromorphic section of (resp. ) over , the relation: "there exists an isomorphism such that ", where is the isomorphism deduced from (one will note that the isomorphism verifying this condition is then uniquely determined). It is clear that this is an equivalence relation, and since there exists a set of invertible -Modules such that every invertible -Module is isomorphic to an element of this set , one can speak of the set of equivalence classes of pairs for the preceding relation. For every invertible -Module and every

regular meromorphic section of over , one will denote by the element of corresponding to the pair . It follows from that if , are two regular meromorphic sections of over , the relation is equivalent to the existence of a section such that .

It is immediate that if is equivalent to and to , the pairs and are equivalent; one therefore defines in a composition law by setting

  (21.2.10.1)            cl(ℒ, s) · cl(ℒ', s') = cl(ℒ ⊗ ℒ', s ⊗ s');

it is immediate that this is a commutative group law, whose neutral element is and where the inverse of is .

Proposition (21.2.11).

The maps

  (21.2.11.1)            D ↦ cl(𝒪_X(D), s_D),       cl(ℒ, s) ↦ div(s)

are reciprocal isomorphisms of onto and of onto respectively.

In view of (21.2.8.3), (21.2.9.4) and (21.2.10.1), it suffices to see that the composites of these two maps are the identity in and respectively. The first assertion is none other than (21.2.9.3). On the other hand, let , where is a regular meromorphic section of over , and let be an open of such that there exists an isomorphism of onto transforming into , so that , , and is the unit element of . There is therefore an isomorphism such that (notation of (21.2.10)) transforms into ; one sees at once that these isomorphisms are compatible with the operations of restriction, hence define an isomorphism such that . Q.E.D.

One can transport, by the first of the isomorphisms (21.2.11.1), the ordered group structure of to ; the elements of are therefore the classes such that , that is to say (21.1.7.1) such that

                         s ∈ Γ(X, ℒ) ∩ Γ(X, (𝓜_X(ℒ))^×).

(21.2.12). Let be a positive divisor on a prescheme ; the fractional Ideal is therefore an Ideal of which is an invertible -Module; let be the closed sub-prescheme of it defines. For every , there is by hypothesis an open neighbourhood of in and a regular section such that ; in other words, the canonical immersion is regular and of codimension 1 (19.1.4) at every point of . Conversely, if is a closed sub-prescheme of , regularly immersed in and of codimension 1 at every point of , there exists one and only one positive divisor such that , for every , there is a neighbourhood of in such that is defined by an Ideal of of the form , where is regular in .

One will note that one then has , for to say that means that (with the notations above) is not invertible, that is to say that .

21.3. Linear equivalence of divisors

(21.3.1). One says that a divisor on is principal if it is of the form , where is a regular meromorphic function on ; the regular meromorphic functions such that are then all those of the form tf, where (21.1.3.4). The set of principal divisors is a subgroup of , denoted , isomorphic to . Two divisors , are said to be linearly equivalent if is a principal divisor; the principal divisors are therefore the divisors linearly equivalent to 0.

(21.3.2). Recall that one can speak of the set of equivalence classes of invertible -Modules for the relation of isomorphy; one denotes this set by , and for every invertible -Module , one denotes by the equivalence class of -Modules isomorphic to ; moreover, is a commutative group for the multiplication defined by . It is clear that the map

  (21.3.2.1)             𝓁' : cl(ℒ, s) ↦ cl(ℒ)

is a homomorphism from the group (21.2.10) into the group . By composition, one therefore deduces a homomorphism

  (21.3.2.2)             𝓁 : Div(X) ⥲ D(X) → Pic(X)

(also denoted ) such that, for every divisor , one has

Note finally that, if is a morphism of ringed spaces, , two isomorphic invertible -Modules, the invertible -Modules and are isomorphic; since moreover, for any two invertible -Modules , , one has up to canonical isomorphism , one sees that the morphism canonically defines a homomorphism of commutative groups

  (21.3.2.4)             Pic(u) : Pic(X) → Pic(X').

Proposition (21.3.3).

(i) The kernel of the canonical homomorphism is the subgroup ; in other words, for and to be isomorphic, it is necessary and sufficient that and be linearly equivalent. One therefore has a canonical injective homomorphism

  (21.3.3.1)             Div(X) / Div.princ(X) → Pic(X)

deduced from .

(ii) For an invertible -Module to be such that is of the form , or also for to be isomorphic to an -Module of the form , it is necessary and sufficient that there exist a regular meromorphic section of .

The proposition follows at once from the definitions and from (21.2.10).

Proposition (21.3.4).

Let be a prescheme satisfying one of the two following hypotheses:

a) is locally Noetherian and is contained in an affine open of .

b) is reduced and the set of its irreducible components is locally finite.

Then the canonical homomorphism is surjective, and gives, by passage to the quotient, an isomorphism

                         Div(X) / Div.princ(X) ⥲ Pic(X).

It suffices to show that every invertible -Module admits a regular meromorphic section over (21.3.3). In the two cases, it suffices, thanks to (20.2.11, (ii)), to define a section of over a schematically dense open of , or also in case a), over an open containing (20.2.13, (iv)). Indeed, let be any open of such that is isomorphic to , so that there exists an isomorphism of onto , transforming into a section of over . Since is schematically dense in , it follows from (20.2.11) that there exists one and only one regular meromorphic function on such that , and this section therefore corresponds, by the isomorphism considered, to a regular meromorphic section of over such that . Moreover, if is a second open of such that is isomorphic to , the restrictions of and to are equal, for they correspond by isomorphism to two meromorphic functions which coincide in a schematically dense open , and one concludes again by (20.2.11); the are therefore the restrictions of a section of over .

This being so, in case b), one takes for each of the maximal points of an open containing , not meeting any irreducible component of distinct from and such that is isomorphic to ; one will take for the section such that is the section of corresponding by the preceding isomorphism to the unit section of .

In case a), one can take by hypothesis for an affine open (hence Noetherian); in other words, one may assume that , where is Noetherian, and , where is a projective -module of rank 1. If is the set of regular elements of , one has (20.2.12) and . But is the set of elements not belonging to any of the ideals associated to , hence is a semi-local ring whose maximal ideals come from the maximal elements of , and is a projective -module of rank 1, hence here free of rank 1 (Bourbaki, Alg. comm., chap. II, §5, n° 3, prop. 5); an element forming a basis of this -module is therefore (20.1.8) a regular meromorphic section of over .

Corollary (21.3.5).

If is a Noetherian prescheme such that there exists an ample invertible -Module (II, 4.5.3) (for example a quasi-projective prescheme over the spectrum of a Noetherian ring (II, 5.3.1 and 4.6.6)), the canonical homomorphism is surjective.

Indeed (II, 4.5.4), there then exists an affine open neighbourhood of the finite set .

Remark (21.3.6).

Recall that one has a canonical isomorphism defined in the following way. One starts from a 1-cocycle with values in corresponding to an open cover of , being an element of , and one associates with it the class of the invertible -Module obtained by gluing the along the isomorphisms defined by multiplication by . On the other hand, one deduces from the exact sequence of sheaves of commutative groups

the connecting homomorphism of the exact cohomology sequence

Let us show that the composite homomorphism

                                                ∂              π
                         Div(X) → H^1(X, 𝒪_X^×) ⥲ Pic(X)

is none other than the homomorphism defined in (21.3.2.2). Indeed, one must start from a divisor and an open cover of such that , where is a regular meromorphic function over ; is the cohomology class of the cocycle , where , denoting the restriction of to . It is clear that the image by of this cohomology class is the class of the invertible fractional Ideal such that for every , , which is none other by definition than (21.2.8).

21.4. Inverse images of divisors

(21.4.1). Let be a morphism of ringed spaces; we propose to give conditions allowing us to associate with a divisor on a divisor on , inverse image of by . Note first for this that for every section , the image of by the canonical homomorphism is again invertible, in other words belongs to . Consider then as given by the equivalence class of a pair , where is an invertible -Module and a regular meromorphic section of over (21.2.11). Form the invertible -Module ; to say that the inverse image of by exists (20.1.11) and is a regular meromorphic section of over amounts to saying that the inverse images by of and of exist, in other words that and ; the remark made above then shows that if is a pair equivalent to in the sense of (21.2.10), the inverse image exists and is a regular meromorphic section of over , and the pairs and are equivalent. One can therefore lay down the following definition:

Definition (21.4.2).

Given a morphism of ringed spaces, one says that the inverse image by of a divisor on exists if one has and (cf. (20.1.11)). One then calls inverse image of by , and denotes by , the divisor on which corresponds canonically (21.2.11) to the class of the pair .

It follows at once from this definition that if and have inverse images under , so do and (taking account of (21.2.9.4)) and that one has

. In other words, the set of divisors on whose inverse image by exists is a subgroup of , and the map is an increasing homomorphism from the ordered subgroup into the ordered group , making commutative the diagram

                         Div_f(X) ─𝓁_X─→ Pic(X)
                            │              │
  (21.4.2.1)              f^*│              │ Pic(f)
                            ↓              ↓
                         Div(X') ─𝓁_{X'}→ Pic(X')

(21.4.3). Definition (21.4.2) shows at once that, for to exist, it is necessary and sufficient that for every open of , the inverse image by (restriction of ) of exist. Now, if , where is a regular meromorphic function on , to say that the inverse image of by exists and is a regular meromorphic section of signifies (21.2.9) that the inverse image of by exists and is a regular meromorphic function on . One deduces at once a second description of and of : consider the subsheaf of groups of , denoted , formed of the germs of regular meromorphic functions on an open of whose inverse image by exists and is regular on the inverse image open (20.1.11). Then if , the canonical homomorphism (20.1.11) gives by restriction a homomorphism of sheaves of groups . Setting , one has , and the map corresponds to the homomorphism of sheaves of groups deduced from the preceding one by passage to the quotients.

(21.4.4). It follows at once from the preceding definitions that if is a second morphism of ringed spaces, a divisor on such that the inverse images and exist, then exists and is equal to .

Proposition (21.4.5).

Let be a morphism of ringed spaces. In each of the three following cases, the inverse image by of every divisor on is defined:

(i) is flat.

(ii) and are locally Noetherian preschemes and one has .

(iii) and are preschemes, the set of irreducible components of is locally finite, is reduced, and every irreducible component of dominates an irreducible component of .

It suffices to show that in the three cases one has . In case (i), this follows from (20.1.12). In case (iii), one may restrict to the case where and are affine; if is regular, it does not belong to any minimal prime ideal of (20.1.3.1), hence the hypothesis implies that its image in does not belong to any minimal prime ideal of , and is consequently a regular element of (20.1.3.1). In case (ii) the meromorphic functions on are identified with the pseudo-functions on (20.2.11), and the hypothesis, joined to (20.2.13, (iv)), ensures that the inverse image

by of every schematically dense open of is a schematically dense open of ; one therefore concludes by (20.3.12).

Corollary (21.4.6).

Let be a prescheme having one of the following properties:

(i) is locally Noetherian.

(ii) is reduced and the set of its irreducible components is locally finite.

Then, for every , one has a canonical isomorphism

This follows from (20.2.11), (20.3.7) and from the fact that identifies with the group of invertible elements of the ring .

(21.4.7). Let , be two preschemes, a morphism. If is a positive divisor on such that the inverse image is defined (21.4.2), then the closed sub-prescheme of is none other than the inverse image ; this follows at once from the definitions (21.4.2) and (21.2.12).

Proposition (21.4.8).

Let , be two preschemes; a faithfully flat morphism. Then, if a divisor on is such that (the existence of following from (21.4.5)), one has . In particular, the map from to is injective.

The question being local on , one may restrict to the case where , with , and being two regular sections of over . By hypothesis one has , hence, for every , if one sets , one has ; one concludes that by virtue of the hypothesis that is a faithfully flat -module and of Bourbaki, Alg. comm., chap. I, §3, n° 5, prop. 10, (ii); whence since is surjective, and consequently .

21.5. Direct images of divisors

(21.5.1). Let , be two preschemes, a morphism. We shall, in this n°, give sufficient conditions to be able to associate with every divisor on a divisor on , direct image of by . We shall restrict to the case where is a finite morphism (for more general conditions, see the chapter of this Treatise devoted to intersection theory).

Lemma (21.5.2).

Let be a ring, a free -module of finite rank. For an endomorphism of to be injective, it is necessary and sufficient that be a regular element of .

This is proved in Bourbaki, Alg., chap. III, 3rd ed., §8, n° 2, prop. 3.

(21.5.3). Suppose now that the morphism is finite, and moreover that verifies one of the two following properties:

I) is a finite locally free morphism, in other words (18.2.7) the quasi-coherent -Module of finite type is locally free.

II) is a reduced locally Noetherian prescheme, the -Module is locally free, and for every section ( open

in ), is a section of over (cf. (II, 6.5.1)). (One recalls that condition (II) is verified for every finite morphism when is a locally Noetherian normal prescheme (loc. cit.).)

One then knows (II, 6.5.5) that for every invertible -Module one defines the norm (which we shall also write ), which is an invertible -Module. Moreover, for every open of and every regular section , the norm (which we shall also write ) is a regular element of ; one is indeed at once reduced to the case where is affine, and then the conclusion follows from (21.5.2) under hypothesis (I); on the other hand, under hypothesis (II), the fact that is a flat -Module entails that the section of is also regular , and the conclusion follows again from (21.5.2) applied to the ring , taking into account the definition of the norm of a section (II, 6.5.3). This being so, let be a meromorphic section of over ; the morphism being affine, every point admits an open neighbourhood such that is of the form , where and is a regular section in ; the element (where is the section of defined in (II, 6.5.3)) is then a meromorphic section of over by virtue of what precedes, and it follows from the multiplicative properties of the norm (II, 6.5.3.1) that this section depends only on and not on its expression in the form ; for the same reason, when varies, the meromorphic sections of over thus defined are the restrictions of one and the same meromorphic section of over , which one calls the norm of and denotes (or simply ). The map thus defined

  (21.5.3.1)             N_{X'/X} : Γ(X', 𝓜_{X'}(ℒ')) → Γ(X, 𝓜_X(N_{X'/X}(ℒ')))

extends the norm defined in (II, 6.5.3); if is a regular meromorphic section of over , it follows at once from what precedes that is a regular meromorphic section of over , for (with the same notations) is regular if is. Finally, if , are two invertible -Modules, (resp. ) a meromorphic section of (resp. ) over , one has, by virtue of what precedes and of (II, 6.5.3.1),

  (21.5.3.2)             N(s_1' ⊗ s_2') = N(s_1') ⊗ N(s_2').

(21.5.4). Suppose still that verifies one of the hypotheses I), II) of (21.5.3). If , are two pairs each formed of an invertible -Module and a regular meromorphic section of that Module over , which are moreover equivalent in the sense of (21.2.10), then the pairs and are also equivalent, for an isomorphism of invertible -Modules has for norm an isomorphism of their norms (II, 6.5.3), and one has seen above that transforms sections of into those of . One can therefore lay down the following definition:

Definition (21.5.5).

Given a finite morphism of preschemes, verifying one of the conditions I), II) of (21.5.3), one calls direct image (or norm) of a divisor on by , and denotes (or ), the divisor on which corresponds canonically (21.2.11) to the class of the pair .

It follows at once from this definition, taking into account (21.2.9.4) and (21.5.3.2), that if , , are divisors on , one has

  (21.5.5.1)             f_*(D_1' + D_2') = f_*(D_1') + f_*(D_2')

and implies ; in other words, is an increasing homomorphism from the ordered group into the ordered group . Definition (21.5.5) moreover shows at once that for every open of , one has ( being the restriction of ), and the homomorphisms , for variable , therefore define a homomorphism of sheaves of ordered groups

Moreover, for every open of , every invertible -Module and every regular meromorphic section of over , one has, according to the preceding definitions and (21.1.4),

Proposition (21.5.6).

Let be a finite locally free morphism and suppose that is of constant rank . Then, for every divisor on , is defined and one has

The first assertion follows from the fact that is flat (21.4.5), and the second is an immediate consequence of the definitions and of (II, 6.5.3.2).

Proposition (21.5.7).

Let be a finite morphism verifying one of the hypotheses I), II) of (21.5.3), a finite locally free morphism of constant rank . Then verifies the same hypothesis as , and for every divisor D'' on X'', one has

In view of definition (21.5.5), it suffices to prove the following result:

Lemma (21.5.7.2).

Under the hypotheses of (21.5.7), one has a functorial isomorphism

in the category of invertible -Modules.

Indeed, taking into account the definition of the norm of a section of (II, 6.5.3) and definition (21.5.5), one will at once obtain (21.5.7.1). To prove (21.5.7.2), it suffices, in view of the definitions of (II, 6.5.2 and 6.5.3), to prove that for every section of over an open of , one has

The question is obviously local on , and one may therefore restrict to the case where is affine; one then has and , and one may suppose that A'' is a projective -module of rank . When is locally free, one may suppose that is a free -module of rank , and then A'' is a projective -module of rank mn, and by restricting to a suitable open, one may suppose that A'' is a free -module of rank mn; formula (21.5.7.4) then follows from the transitivity of the norm (Bourbaki, Alg., chap. VIII, §12, n° 2, prop. 7). When verifies hypothesis II), is Noetherian reduced, and if is its total ring of fractions, is a free -module of rank , hence is a projective -module of rank mn, and since is then a semi-local ring, this -module is free; the proposition follows again from the transitivity of norms.

Proposition (21.5.8).

Let be a finite morphism, a morphism; set , , . Suppose verified one of the following conditions:

(i) is locally free and is flat.

(ii) is locally free, and are locally Noetherian, and .

(iii) verifies hypothesis II) of (21.5.3), is locally Noetherian, and are reduced, and every irreducible component of (resp. ) dominates an irreducible component of (resp. ).

Then, for every divisor on , is defined, is defined, and one has

Indeed, in all the cases, it follows from (II, 6.5.8) that one has a functorial isomorphism

in the category of invertible -Modules; moreover (II, 6.5.4), if is a section of over , s'' the corresponding section of over ( open of ), is the section of over which corresponds to the section of over . Formula (21.5.8.1) will therefore follow from the definitions if one proves that and are defined, whatever the divisors on and on . As regards , this follows from the hypotheses made and from (21.4.5). As regards , in case (i) is flat, hence in all the cases is defined by virtue of (21.4.5).

21.6. 1-codimensional cycle associated with a divisor

(21.6.1). Let be a locally Noetherian prescheme, and let denote the set of irreducible closed parts of (which is in bijective correspondence with by the map ). In the product group , consider the subgroup of elements such that the set of such that (or, what amounts to the same,

the set of such that ) is locally finite. It is clear that is a subgroup of which contains the direct sum group (free group having for basis), and is equal to it when is Noetherian. The elements of are called cycles on and those of prime cycles (they do not in general form a basis of when is not Noetherian). One always considers as ordered by the order induced on this subgroup by the product order of , and one denotes the set of cycles .

For every cycle , equal to , one writes by abuse of language,

                         Z = ∑_{x ∈ X} n_x · ‾{x};

one calls multiplicity of at the point and one denotes by the integer (positive or negative). To say that means that for every . One calls support of and denotes by the union of the such that ; by definition of , this is a closed part of , as union of a locally finite family of closed parts. One calls dimension (resp. codimension) of and denotes by (resp. ) the dimension (resp. codimension in ) of .

(21.6.2). One says that a closed part of is purely of codimension (in ) if each of its irreducible components is of codimension in . One says that a cycle is purely of codimension , or (by abuse of language) is a -codimensional cycle, if its support is purely of codimension . One denotes by the set of such that , or, what amounts to the same (5.1.2.1), : the cycles purely of codimension form a subgroup of , isomorphic to the subgroup of formed of the such that the set of (or the set of ) for which is locally finite; this subgroup contains the free group and is identical to it when is Noetherian. One denotes by the set of elements of . It is clear that the ordered group contains the direct sum of the ordered subgroups , and is identical to this direct sum when is Noetherian; in this last case, one considers as graded by the for .

(21.6.3). Let be a cycle on , an open of ; one calls restriction of to and one denotes by the cycle on ; one has . It is clear that is a homomorphism of ordered groups from into (resp. from into ), and that if is an open contained in , one has ; the map (resp. ) is therefore a presheaf on of ordered commutative groups. In fact this presheaf is a sheaf, direct sum of the sheaves , where runs through (resp. ) and for each , is the canonical injection and is the simple sheaf associated with the constant sheaf on the space : this follows at once from the description of the sections of a direct sum of sheaves of commutative groups (G, II, 2.7). One denotes this sheaf (resp. ). One denotes by (resp. ) the subsheaf

of monoids (resp. ) of (resp. ). The sheaf is evidently the direct sum of the sheaves .

Note finally that the sheaves (hence also ) are flasque. Suppose indeed given a section of over an open , so that the set of such that is locally finite in ; this set is also locally finite in , for, for every and every Noetherian open neighbourhood of , is Noetherian, hence contains only a finite number of points of . One therefore defines a section of over by setting ; and since the closure of in is , one has indeed , whence our assertion.

(21.6.4). We propose to define a canonical homomorphism

of sheaves of commutative groups. It evidently suffices to define a homomorphism from into , which vanishes on (21.1.2); now, by definition, is the symmetrization of the sheaf of monoids , and a homomorphism is uniquely determined by the data of its restriction , which is a homomorphism of sheaf of monoids subject to the sole condition of being zero on ; it amounts to the same to say that to define , it suffices to define a homomorphism of sheaf of monoids

(21.6.5). Now, one has seen that to every positive divisor on corresponds the closed sub-prescheme of , defined by the Ideal , and which is regularly immersed and of codimension 1. At each of the maximal points of , one therefore has ((19.1.4) and (5.1.2)) , in other words ; moreover the set of these maximal points is locally finite in (3.1.6), and is an Artinian ring. At every point which is not a maximal point of , one necessarily has , hence . Set

  (21.6.5.1)             cyc(D) = ∑_{x ∈ X^{(1)}} long(𝒪_{Y(D),x}) · ‾{x}

which is therefore an element of .

Proposition (21.6.6).

The map is a homomorphism from the monoid into .

Everything reduces to seeing that for two positive divisors , , one has

                         cyc(D + D') = cyc(D) + cyc(D').

Now, one has (21.2.5) ; everything reduces to seeing that if , if one sets , and if and are two regular elements of , one has long(A / tt'A) = long(A / tA) + long(A / t'A); but since is regular, is isomorphic to , whence the proposition at once.

It follows at once from the definitions that for every open , one has

                         Y(D | U) = Y(D) ∩ U,

hence , and the homomorphisms therefore define a homomorphism of sheaves of monoids (21.6.4.2), whence the sought homomorphism (21.6.4.1) of sheaves of groups.

One has Supp(cyc(D)) ⊂ Supp(D) for every divisor and

  (21.6.6.2)             Supp(cyc(D)) = Supp(D)     when D ≥ 0;

one has indeed already seen the second relation (21.2.12); when is arbitrary, the relation entails the existence of an open neighbourhood of such that , whence , which proves our assertion.

(21.6.7). The homomorphism (21.6.4.1) gives, by passage to the groups of sections over , an increasing homomorphism of ordered groups , which we shall again denote ; the number for is also denoted or and is called multiplicity of the divisor at the point , or multiplicity of the prime cycle in ; this is a positive or negative integer, equal as one has seen to when is a positive divisor; one therefore has by definition

  (21.6.7.1)             cyc(D) = ∑_{x ∈ X^{(1)}} mult_x(D) · ‾{x},

and by virtue of the fact that is a homomorphism, one has

  (21.6.7.2)             mult_x(−D) = −mult_x(D),    mult_x(D + D') = mult_x(D) + mult_x(D')

for any two divisors , .

For every regular meromorphic function on , one sets in particular, for ,

and one says that (positive or negative integer) is the order of at the point . If (a regular element of this local ring by hypothesis), one therefore has

  (21.6.7.4)             ω_x(f) = long(𝒪_{X,x} / f 𝒪_{X,x});

for two regular meromorphic functions , on , one has

  (21.6.7.5)             ω_x(ff') = ω_x(f) + ω_x(f'),    ω_x(f⁻¹) = −ω_x(f)

for every . The 1-codimensional cycles

                         Z^+(f) = ∑_{x ∈ X^{(1)}, ω_x(f) > 0} ω_x(f) · ‾{x},
                         Z^−(f) = ∑_{x ∈ X^{(1)}, ω_x(f) < 0} (−ω_x(f)) · ‾{x}

are called respectively the cycle of zeros and the cycle of poles (or polar cycle) of , and one has obviously . The 1-codimensional cycles of the form are called principal (or linearly equivalent to 0) and form a subgroup of . The sections over of the image are

called locally principal cycles; such a cycle is therefore characterized by the fact that, for every , there is an open neighbourhood of in and a regular meromorphic function on such that . If , it amounts to the same to say that, for every , if one sets and , is principal; in other words there exists a regular meromorphic function on such that . This follows at once indeed from the preceding definition and from (20.3.7), which establishes a bijective correspondence between the regular meromorphic functions on and the germs of regular meromorphic functions on the open neighbourhoods of in when is locally Noetherian. One again expresses the fact that is principal by saying that is principal at the point . The set of points where is principal is evidently open, by virtue of what precedes.

One deduces from the canonical homomorphism a canonical homomorphism Div(X) / Div.princ(X) → 𝔷^1(X) / 𝔷_{princ}^1(X), by virtue of the definition of . One says that the group is the group of classes of 1-codimensional cycles on and one denotes it . Two elements of having the same image in are called linearly equivalent.

(21.6.8). Consider in particular the case where , where is an integrally closed Noetherian integral domain. Then is the set of prime ideals of height 1 of , and is therefore identified with the group of divisors (in the sense of N. Bourbaki) of the Krull ring (Bourbaki, Alg. comm., chap. VII, §1, n° 3, cor. of th. 2 and n° 6, th. 3).

Since on the other hand, the regular meromorphic functions on are then identified with the elements of the fraction field of , the map from into is identified with the map denoted in Bourbaki (loc. cit., §1, n° 1); is therefore identified with the group of principal divisors of in the sense of Bourbaki, and with the group of divisor classes of in the sense of Bourbaki (loc. cit., §1, n° 2 and n° 10).

Theorem (21.6.9).

Let be a locally Noetherian normal prescheme.

(i) The canonical homomorphism is injective and its image is formed of the locally principal cycles.

(ii) The following conditions are equivalent:

a) The homomorphism is bijective.

b) Every 1-codimensional cycle is locally principal.

c) For every , the local ring is factorial.

(One then says that is a locally factorial prescheme.)

(i) It suffices to prove that

since one has and . One is therefore reduced to proving that if is a divisor on such that for every , one has . Now, for every , the local ring is integral and integrally closed, and

of dimension 0 or , hence one has or (0, 16.5.1). On the other hand, at the points , the ring is a discrete valuation ring (II, 7.1.6), hence (0, 16.5.1); the only points of such that are therefore the points of , and the conclusion follows from (21.1.8).

(ii) The equivalence of a) and b) is clear by virtue of (i). According to the characterization of locally principal cycles given in (21.6.7), and the relation between 1-codimensional cycles on and divisors (in the sense of Bourbaki) of the ring when is an integrally closed Noetherian ring (21.6.8), condition b) is equivalent to saying that for every , every divisor of the ring is principal, in other words that the ring is factorial (Bourbaki, Alg. comm., chap. VII, §3, n° 1), whence the equivalence of b) and c).

(21.6.9.2). When is a factorial Noetherian ring, it is clear that is locally factorial (Bourbaki, Alg. comm., chap. VII, §3, n° 4, prop. 3). If, in this case, one writes an element of the fraction field of in the form , where and are two coprime elements of , whose divisors are well determined (loc. cit., §3, n° 3), these divisors are identified respectively with the cycle of zeros and the cycle of poles of (21.6.7).

Corollary (21.6.10).

Let be a locally Noetherian normal prescheme.

(i) There exists a canonical injective homomorphism

(ii) If is locally factorial, the homomorphism (21.6.10.1) is bijective, and conversely.

One has seen (21.6.7) that the image of by the homomorphism is ; one therefore deduces from (21.6.9) that the homomorphism Div(X) / Div.princ(X) → 𝔷^1(X) / 𝔷_{princ}^1(X) = Cl(X) deduced from cyc is injective, and that it is bijective if and only if is locally factorial. The conclusion therefore follows from the fact that, when is locally Noetherian and reduced, the canonical homomorphism Div(X) / Div.princ(X) → Pic(X) (21.3.3.1) is bijective (21.3.4, (ii)).

Corollary (21.6.11).

Let be a locally Noetherian and locally factorial prescheme. Then the sheaf is flasque, and for every open of , the canonical homomorphism is surjective.

Taking into account (21.6.9, (ii)), the first assertion is equivalent to saying that the sheaf is flasque, which has been seen above (21.6.3). For every open of , the canonical homomorphism is therefore surjective; since by virtue of (21.6.10), the homomorphism is canonically identified with , it is also surjective.

Proposition (21.6.12).

Let be a Noetherian reduced prescheme. Let be a decreasing filtered family of opens of verifying the following conditions:

1° If , one has for every .

2° For every , the ring is factorial.

Then there exist canonical isomorphisms

  (21.6.12.1)            lim Div(U_λ) ⥲ 𝔷^1(X),    lim Pic(U_λ) ⥲ Cl(X)

such that, for every open of , the diagrams

                         lim Div(U_λ) ─⥲─ 𝔷^1(X)
                              │             │
  (21.6.12.2)
                              ↓             ↓
                         lim Div(U_λ ∩ V) ─⥲ 𝔷^1(V)

                         lim Pic(U_λ) ─⥲─ Cl(X)
                              │             │
                              ↓             ↓
                         lim Pic(U_λ ∩ V) ─⥲ Cl(V)

are commutative.

Hypothesis 1° on implies that for every , one has (5.1.3.1); hence the restriction homomorphism is bijective and consequently one has a canonical isomorphism . The canonical homomorphisms define therefore, by passage to the inductive limit, the first of the canonical homomorphisms (21.6.12.1), and the second one is deduced from it by passage to the quotients. Moreover, it follows from condition 1° that the are dense in , hence schematically dense since is reduced (11.10.4), and consequently every meromorphic function on is entirely determined by its restriction to each ; one deduces at once from this that in the isomorphism , the image of is , hence one also has a canonical isomorphism . The second of the canonical homomorphisms (21.6.12.1) will therefore be an isomorphism if the first is, and it remains to prove this latter point, the commutativity of the diagrams (21.6.12.2) being trivial.

Let us show first that the homomorphism is injective. Set , and note that the form a fundamental system of neighbourhoods of . Indeed, for every open , is a closed subspace of , hence a Noetherian space every closed irreducible part of which admits a generic point; since the sets are closed in , form an increasing filtered family and have for union , our assertion follows from . This being so, it is a matter of seeing that if is such that in , then there exists such that . By virtue of what precedes, it will suffice to see that for every , one has ; indeed, there will then be a neighbourhood of in such that , and the union of the for contains some . Taking into account (21.4.6), one is therefore reduced to the case where ; but by hypothesis is factorial, hence integral and integrally closed, and is then normal; hence the conclusion follows from (21.6.9, (i)).

To prove that the homomorphism is bijective, it suffices to prove similarly that for every and every , there is a neighbourhood

of in and a divisor on such that . Indeed, one will then be able to cover the quasi-compact set by a finite number of , and by virtue of the first part of the demonstration, since the pairs are finite in number, there will exist an index such that in each of the , the restrictions of and coincide; since one may suppose moreover that is contained in the union of the , one sees that the restrictions are the restrictions of one and the same divisor which will be such that . One is therefore reduced again to the case where with ; but since is factorial, so are its localizations (Bourbaki, Alg. comm., chap. VII, §3, n° 4, prop. 3), and it suffices to apply (21.6.9, (ii)).

Corollary (21.6.13).

Let be an integrally closed Noetherian integral local ring of dimension . Set , and let be the closed point of , . For to be factorial, it is necessary and sufficient that be locally factorial and that .

Indeed, to say that is factorial is equivalent to saying that (Bourbaki, Alg. comm., chap. VII, §1, n° 4, cor. of th. 2 and §3, n° 2, th. 1); it therefore suffices to use the existence of the second isomorphism (21.6.12.1), taking the family restricted to the single open .

Corollary (21.6.14).

Let be a Noetherian local ring of dimension , , the closed point of , . The following conditions are equivalent:

a) is factorial.

b) and (conditions which we shall later (21.13) express by saying that the ring is parafactorial), and is locally factorial.

It is clear that if is factorial, it is normal, hence since (0, 16.5.1), and (21.6.13) shows that a) implies b). Conversely, if b) is verified, it suffices to see that is integrally closed to conclude by (21.6.13) that b) entails a). By virtue of Serre's criterion (5.8.6), it suffices to verify for the conditions (R_1) and (S_2). Now, being locally factorial verifies these conditions, and the hypothesis entails that verifies them also.

21.7. Interpretation of positive 1-codimensional cycles in terms of subpreschemes

(21.7.1). Let be a locally Noetherian prescheme, a positive 1-codimensional cycle (so that one has for every , and except on a locally finite set of points). Denote by the closed sub-prescheme of , closed image (I, 9.5.3 and 9.5.1) of the sum prescheme under the canonical morphism, and by (or ) the Ideal of defining .

Proposition (21.7.2).

Let be a locally Noetherian prescheme verifying condition (R_1) (5.8.2). For a closed sub-prescheme of to be of the form , where is a positive 1-codimensional cycle, it is necessary and sufficient that it verify the two following conditions:

(i) is purely of codimension 1.

(ii) verifies condition (S_1), in other words (5.7.5) has no immersed associated prime cycle.

One then has

The map is a bijection of onto the set of closed sub-preschemes of verifying conditions (i) and (ii).

The conditions are necessary (without assuming that verifies (R_1)). Indeed, the question being local on , one may suppose that , where is a Noetherian ring; one then has , where is by definition (I, 9.5.1) the kernel of the canonical homomorphism , where the are the prime ideals corresponding to the points for which , and where one has set . The inverse image of in is a -primary ideal, and one has . Moreover, since the are such that is of codimension 1, no point of can be adherent to another point of . Hence the are the minimal prime ideals of and the set of is equal to , which proves conditions (i) and (ii).

The conditions are sufficient. Denoting by the Ideal of defining , hypothesis (ii) implies that is identical to the set of maximal points of , and hypothesis (i) implies that is contained in ; hence (3.2.6) is identified with a sub--Module of , where for each , is an irredundant -Module such that . Now, since verifies (R_1), for each , is a discrete valuation ring, and consequently the primary ideals of this ring are the powers of the maximal ideal; supposing still that , one concludes that , where , the being the inverse images in of ideals , where the correspond to the points of . One sees therefore well that is of the form .

Corollary (21.7.3).

Let be a locally Noetherian prescheme. For every positive divisor on such that verifies (R_1) at the maximal points of the closed sub-prescheme (21.2.12), majorizes the closed sub-prescheme (21.7.1), and has the same underlying space; for these two sub-preschemes to be equal, it is necessary and sufficient that verify condition (S_1), or also that, for every distinct from a maximal point of , one have (condition always verified when is normal).

Indeed, the question being local, one may always suppose that , being a regular section of over ; at every maximal point of , necessarily belonging to , is by hypothesis a discrete valuation ring, hence for . One may suppose , and then, if , , it follows from what precedes and from (21.7.1) that and have the same non-immersed primary ideals; hence , since is the intersection

of these primary ideals (21.7.2). This proves that majorizes and that these two sub-preschemes are equal if and only if has no immersed associated prime cycle (in other words verifies (S_1)). Since for every , there is by hypothesis an open neighbourhood of in and a regular element such that is the restriction to of , to say that verifies (S_1) also means that at every non-maximal point , one has , that is to say (0, 16.4.6) . The assertion concerning the case where is normal is then immediate since verifies (S_2) and at the non-maximal points of one has (0, 16.3.4).

(21.7.3.1). One therefore sees that when is normal, is canonically identified with the set of closed sub-preschemes of verifying conditions (i) and (ii) of (21.7.2) and regularly immersed in .

Proposition (21.7.4).

Let be a locally Noetherian prescheme, reduced at each of its isolated points. The following conditions are equivalent:

a) The canonical homomorphism (21.6.4) is an isomorphism of sheaves of ordered groups.

a') Every prime 1-codimensional cycle on is the image by cyc of a positive divisor, and the canonical homomorphism is injective.

a") For every integral closed sub-prescheme of , of codimension 1, the canonical immersion is regular.

b) is normal and the homomorphism is bijective.

c) is locally factorial.

The equivalence of b) and c) has been proved in (21.6.9), as well as the fact that c) entails a). Moreover b) entails a") by virtue of (21.7.3.1), and a) trivially implies a'). It remains to see that a') or a") entails c).

Suppose then that condition a') is verified, and let us first show that is normal. Note first that if verifies a'), so does every local scheme . Consider then , so that is a Noetherian local ring of dimension 1. Applying condition a') to and to the prime 1-codimensional cycle formed of the closed point of , one sees that in the maximal ideal is generated by a single regular element of ; hence (0, 17.1.1) is a regular ring. Since the localized rings are also regular (0, 17.3.2), one sees that is regular at all its non-isolated maximal points; since it is also reduced (hence regular) at its isolated points by hypothesis, one concludes that verifies condition (R_1). Let us show moreover that also verifies (S_1), in other words that for every such that , one has (0, 16.4.6). Indeed, hypothesis a') applied to shows that there exists on this prescheme at least one divisor , in other words that one has , and it suffices to apply (21.1.10). One already deduces from these results that is reduced (5.8.5). Let us next prove that it verifies condition (S_2) (which will establish that is normal, by virtue of Serre's criterion (5.8.6)). Argue by contradiction by supposing that the set of points where does not verify (S_2) is non-empty, and let be a point for which is the smallest possible; since verifies (S_1), one has . In , the open verifies (S_2); let us show that X_1 verifies (S_2), so that one will have reached a contradiction. It suffices, by virtue of (5.10.5), to show that every section of over extends to a section of over X_1. Since X_1 is reduced and dense in X_1, is schematically dense in X_1, and consequently is the restriction to of a regular meromorphic section . Moreover, since , one has since ; since the homomorphism cyc is injective, one necessarily has , hence (21.6.4) is a section of over X_1, which establishes our assertion.

To show that a') implies c), remark then that condition a') implies that the canonical homomorphism is surjective; since is normal, it suffices to apply (21.6.9).

Let us now prove that a") entails c). It follows from (21.6.5) that a") entails that every prime 1-codimensional cycle on is the image by cyc of a positive divisor; the first part of the reasoning made above therefore proves again that verifies (R_1) and (S_1). It remains to see that is normal (the end of the reasoning being then the same), that is to say that if is such that , one has . Now, if is a generization of such that is of codimension 1 in , the reduced sub-prescheme of having for underlying

space is integral, hence regularly immersed in by hypothesis. This entails that there exists a regular element of such that the ideal is the prime ideal defining the prescheme Y_1, inverse image of in . Since is integral, this proves that and finishes the proof of (21.7.4).

Remark (21.7.5).

One cannot in (21.7.4) replace condition a') by the weaker condition that every prime 1-codimensional cycle of be the image by cyc of a positive divisor. This is shown by the example where is the affine scheme defined in (14.1.5) ("union of a plane and a line meeting it"), which is not normal; with the notations introduced in (14.1.5), the prime 1-codimensional cycles of are those of the plane X_1 and the closed points of the line X_2 distinct from the common point of X_1 and X_2; if , , are the images of T_1, T_2, T_3 in , one sees therefore that the prime 1-codimensional cycles of are defined by the principal prime ideals of generators ( irreducible in ) or , with in ; these elements being regular in , every 1-codimensional cycle is indeed the canonical image of a positive divisor.

Corollary (21.7.6).

Let be a locally Noetherian prescheme, reduced at each of its isolated points. The following conditions are equivalent:

a) The canonical homomorphism is an isomorphism of sheaves of ordered groups, and one has Div(X) = Div.princ(X).

a') The canonical homomorphism is injective, and every prime 1-codimensional cycle is the image by cyc of a positive principal divisor, in other words is of the form , where is a regular section of over .

a") For every integral closed sub-prescheme of , of codimension 1, there exists a regular section of over such that is defined by the Ideal .

b) is normal and every prime 1-codimensional cycle on is principal.

c) is locally factorial and .

d) is normal, and for every open of , one has .

The equivalence of a), a'), a"), b) and c) follows at once from (21.7.4) and from (21.6.11). Moreover, it follows at once from a") that every non-empty open of again verifies the same conditions; in other words these conditions imply d). It remains to see that d) entails b). Now, denote by the opens that are complements of finite unions of sets of the form , where ; it is clear that the form a decreasing filtered family, and that for every , one has , hence is a principal ring by virtue of the hypothesis. One can therefore apply to the family proposition (21.6.12), which shows that is isomorphic to , hence by virtue of the hypothesis, which proves b) by definition.

Remark (21.7.7).

When , where is a Noetherian integral ring, the equivalent conditions of (21.7.6) are equivalent to saying that is a factorial ring.

21.8. Divisors and normalization

Lemma (21.8.1).

Let be an integral morphism of preschemes. For every -Module locally free of constant rank and every , there exists an open neighbourhood of in such that, setting , is isomorphic to .

Since the question is local on , one may restrict to the case where is affine; one then has , where is an integral -algebra. Then is the inductive limit of its finite sub--algebras . Setting and denoting by the structure morphism, it follows from (8.5.2) and (8.5.5) that there exist an index and an -Module locally free of rank such that ; it will evidently suffice to prove the lemma for and , in other words, one may restrict to the case where the morphism is finite. Set and let be the ring of the affine scheme ; since is a local ring and is a finite -algebra, is a semi-local ring (Bourbaki, Alg. comm., chap. V, §2, n° 1, prop. 3); one concludes that the locally free -Module is isomorphic to (Bourbaki, Alg. comm., chap. II, §5, n° 3, prop. 5). Considering as the projective limit

of the preschemes induced by on the opens , where ranges over the filtered set of affine open neighbourhoods of in , following the method of (8.1.2, a), and applying (8.5.2.5), one obtains the desired conclusion.

Corollary (21.8.2).

Let be an integral morphism of preschemes. Then:

(i) .

(ii) The group is canonically isomorphic to .

(i) is the sheaf associated to the presheaf of commutative groups on (loc. cit.); the stalk of this sheaf at a point may therefore be identified with the commutative group , where ranges over the open neighbourhoods of , the transition homomorphisms for two opens being defined by (21.3.2.4). Now, for every , every open neighbourhood of in , and every element , it follows from (21.8.1) that there is an open neighbourhood of such that the image of in is zero. Hence the stalk of at is zero.

(ii) The Leray spectral sequence for the composite functor on sheaves of commutative groups on (T, 2.4) gives the exact sequence of low-degree terms (M, XV, 6)

                         0 → H¹(X, f_*(𝒪_{X'}^×)) → H¹(X', 𝒪_{X'}^×) → H⁰(X, R¹ f_*(𝒪_{X'}^×))

and the conclusion follows from (i) and from the isomorphism between and .

Proposition (21.8.3).

Let be an integral morphism of preschemes; suppose that, for every open of , the homomorphism sends regular elements to regular elements, which permits one canonically to extend the homomorphism to a homomorphism of sheaves of rings , whence homomorphisms of sheaves of multiplicative groups and , yielding by passage to quotients a homomorphism . One then has a commutative diagram

                         1 ——→ 𝒪_X^×  ——→  𝓜_X^×  ——→  𝒟iv_X  ——→  0
                                  │           │           │
  (21.8.3.1)                    θ^*│        θ'^*│        θ''^*│
                                  ↓           ↓           ↓
                         1 ——→ f_*(𝒪_{X'}^×) → f_*(𝓜_{X'}^×) → f_*(𝒟iv_{X'}) → 0

in which both rows are exact.

The only point to verify is the exactness of the second row of the diagram, which follows from applying the exact cohomology sequence for the functor to the exact sequence of sheaves of commutative groups on

and from (21.8.2).

Corollary (21.8.4).

If, in addition to the hypotheses of (21.8.3), the homomorphism is an isomorphism of sheaves of rings, then is injective, is surjective and is isomorphic to .

This is an immediate consequence of the snake-diagram lemma (Bourbaki, Alg. comm., chap. I, §2, n° 4, prop. 2) applied to the diagram (21.8.3.1).

Proposition (21.8.5).

Let , be two locally Noetherian preschemes, an integral morphism; assume that there exists a schematically dense open in such that is schematically dense in (cf. (20.3.5)), and that the morphism , the restriction of , is an isomorphism. Then:

(i) The homomorphism is defined and bijective, the homomorphism is injective, the homomorphism is surjective, is isomorphic to , and the support of the sheaf of multiplicative groups is contained in .

(ii) Assume in addition that the set is discrete and, to abbreviate, set . Then one has a commutative diagram

                                  j                            i_X    i_{X'}
  1 → ∏_{s ∈ S} 𝒪'^×_{X,s}/𝒪^×_{X,s} ─→ Div(X) ──→ Div(X') ─→ 0
                                  │           │
  (21.8.5.1)                      │           │
                                  ↓           ↓
  1 → (∏_{s ∈ S} 𝒪'^×_{X,s}/𝒪^×_{X,s}) / Im Γ(X', 𝒪_{X'}^×) ─→ Pic(X) ─→ Pic(X') ─→ 0
                                  j'

where the rows are exact and where the left vertical arrow is the canonical homomorphism.

(i) The hypothesis entails that one has a canonical isomorphism for the sheaves of germs of pseudo-functions (20.2.2), whence the assertion concerning , in view of the existence of the canonical isomorphisms and (20.2.11). The rest of assertion (i) follows from (21.8.4), except for what concerns the support of , which follows directly from the hypothesis on .

(ii) Applying the exact cohomology sequence to the two exact sequences of sheaves of commutative groups

                         1 → 𝒩 → 𝒟iv_X → f_*(𝒟iv_{X'}) → 0
  (21.8.5.2)
                         1 → 𝒪_X^× → f_*(𝒪_{X'}^×) → 𝒩 → 1

one obtains respectively the exact sequences

                         1 → Γ(X, 𝒩) ─→ Div(X) → Div(X') → H¹(X, 𝒩)
                                       j
  (21.8.5.3)
                         Γ(X', 𝒪_{X'}^×) → Γ(X, 𝒩) ─→ Pic(X) → Pic(X') → H¹(X, 𝒩)
                                                    ∂

taking account of the canonical isomorphism (21.8.2). Now, since the support of is contained in the discrete set , closed in , one has

(G, II, 4.9.2) and by definition of cohomology (G, II, 4.4). Similarly one has and , by virtue of the second exact sequence (21.8.5.2). It remains to make precise the injections and . One can describe a section of over by taking a covering of formed by and by opens such that is reduced to a single point and that for , then considering for each a section of over , necessarily such that and at the points . The germ comes from a section of over , which one may also consider as a section of over , hence a section of over (in virtue of the hypothesis); there corresponds canonically to this section a section , and since in the restriction of is identified with a section of over by virtue of the hypothesis, the restrictions of the to are all zero, so the are the restrictions of one and the same divisor , which is the image of the section under . Similarly, the image of under comes from the cocycle equal to the restriction of in for each , to in , whence the expression of j'(t) and the commutativity of the diagram (21.8.5.1).

Remarks (21.8.6).

(i) The conditions of application of (21.8.5, (i)) are in particular satisfied when is a reduced locally Noetherian prescheme having only finitely many irreducible components, its normalization, and is also locally Noetherian (II, 6.3.8); the -Module is then an extension of by the cokernel of , which one may regard as known. If moreover is a (reduced) algebraic curve over a field , one is in the conditions of application of (21.8.5, (ii)).

(ii) When the conditions of application of (21.8.5, (ii)) are satisfied and moreover the global canonical homomorphism is bijective, one sees that the kernels of the surjective homomorphisms and are isomorphic.

(iii) When the conditions of application of (21.8.5, (ii)) are satisfied, one sees that the homomorphism can be injective (in which case it is bijective) only if for every . For an such that the ring is local (which, taking into account that (II, 1.3), means that there exists only one point above ), this implies necessarily that the residue fields and are equal and that , hence finally is equivalent to the relation , or also (taking the hypothesis into account) to the fact that there is an open neighbourhood of in such that the morphism is an isomorphism. In the general case, the ring is semi-local (this is evident when the morphism is finite, and in the general case it follows from ); the canonical homomorphism defines, by passage to quotients, a homomorphism from the multiplicative group to the product of the multiplicative groups , the being the points (in

number > 1) of above . It is immediate that this homomorphism can be bijective only if and all the are equal to the field with 2 elements; moreover, if this condition is verified, it is necessary in addition that the multiplicative group have image , where is the radical of , or also that , which entails that is a direct sum of fields isomorphic to (which, when the morphism is finite, entails that it is unramified at the points (17.4.1)). If, for example, no residue field of is isomorphic to , the canonical homomorphism can be bijective only if is an isomorphism. In the case where the canonical homomorphism is bijective, one concludes from the preceding and from (ii) that in the previous considerations one may replace the homomorphism by the homomorphism .

21.9. Divisors on preschemes of dimension 1

(21.9.1). Let be a topological space, a point of , the canonical injection. If is a commutative group, one may regard it as a sheaf of commutative groups on the space reduced to a single point, and take its direct image , which is a sheaf of commutative groups on ; it follows at once from the definitions that for every open of , if , and reduces to 0 in the contrary case; hence, for one has , and for , .

If now is a sheaf of commutative groups on and a part of , for every open of one has a canonical homomorphism

  (21.9.1.1)             Γ(U, ℱ) → ∏_{x ∈ U ∩ Y} ℱ_x = ∏_{x ∈ Y} Γ(U, (i_x)_*(ℱ_x))

and since these homomorphisms commute with the restriction operators, they define a canonical homomorphism of sheaves

  (21.9.1.2)             j_Y : ℱ → ∏_{x ∈ Y} (i_x)_*(ℱ_x).

Lemma (21.9.2).

Let be a locally Noetherian topological space, X_0 the set of its closed points, a sheaf of commutative groups on . The following conditions are equivalent:

a) The canonical homomorphism is injective and has as image .

a') There exists a family of commutative groups such that is isomorphic to .

b) Every section of over an open of has a discrete support contained in X_0.

When this is so, for every , the group is determined up to unique isomorphism and is isomorphic to . Moreover, the sheaf is then flasque.

Since the points of X_0 are closed in , one has for every

; this remark proves the uniqueness assertion relative to the groups , and it is clear besides that a) implies a'). To see that a') implies b), one may restrict to the case where is Noetherian; then (G, II, 3.10) one has

                         Γ(U, ⨁_{x ∈ X_0} (i_x)_*(A(x))) = ⨁_{x ∈ X_0} Γ(U, (i_x)_*(A(x)))

so every section of over is a direct sum of finitely many sections () with . But since is closed in , the support of is reduced to , so the support of is contained in the finite set of the , which is evidently discrete since the points are closed. Let us finally show that b) entails a); for every Noetherian open , any support of a section of over , being discrete and quasi-compact, is finite. One thus deduces from hypothesis b) that for every Noetherian open , the image of the homomorphism (21.9.1.1) (for ) is contained in and that this homomorphism is injective, which establishes a).

Finally, to show that is flasque, consider a section of over an open of ; since the support of is a discrete and closed subspace of , one extends to a section of over by setting in .

Remarks (21.9.3).

(i) In condition b) of (21.9.2), it does not suffice to suppose that the support of every section of over an arbitrary open of is discrete; this is shown by the example where one takes for the spectrum of a discrete valuation ring, with a closed point and a generic point , and for the sheaf of commutative groups such that and .

(ii) Assume that the conditions of (21.9.2) are verified; let be a discrete part of X_0, and for every , let be an element of . Then there exists one and only one section of over such that for every and that the support of is contained in . Indeed, for every Noetherian open , is finite, and it suffices to take for the section of whose restriction to each Noetherian open is the sum of the for .

Proposition (21.9.4).

Let be a locally Noetherian prescheme of dimension , the set of points such that . One then has a canonical isomorphism

  (21.9.4.1)             𝒟iv_X ⥲ ⨁_{x ∈ X^{(1)}} (i_x)_*(Div(𝒪_{X,x}))

and is flasque.

Taking the isomorphism (21.4.6.1) into account, the homomorphism (21.9.4.1) is defined by (21.9.1.2): let us prove that it is a bijection. Since , the points of are the non-isolated closed points of . One is reduced to proving that: 1° verifies condition b) of (21.9.2); 2° for every isolated point , one has . The second point follows from the fact that is then an Artinian ring and so every regular element of is invertible. For the first, it suffices to note that for every open of and every divisor , the maximal points of the support of are such that

(21.1.9), so a fortiori belong to ; since , the set of these points equals , and is therefore discrete, the set of irreducible components of being locally finite.

Corollary (21.9.5).

Let be a locally Noetherian prescheme of dimension . For every discrete part and every family with , there exists one and only one divisor on such that the support of is contained in and for every .

This follows from (21.9.4) and (21.9.3, (ii)).

Corollary (21.9.6).

Let be a locally Noetherian prescheme of dimension ; every divisor on is of the form , where , D'' are two divisors with supports contained in that of .

In virtue of (21.9.5), one is at once reduced to the case where , a Noetherian local ring; it then suffices to observe that is the total ring of fractions of , and that a section of over is by definition expressible as a quotient of two regular elements of .

Corollary (21.9.7).

Let be a locally Noetherian prescheme of dimension , having no isolated point, and a dense open in . Then there exists a divisor on , with support contained in and meeting each of the irreducible components of .

One may assume that is the union of disjoint non-empty open sets , each contained in a single irreducible component of ; it then suffices to take in each a point closed in (such a point exists since the unique generic point of cannot be isolated by hypothesis), and to apply (21.9.5) to the discrete set of the .

The interest of this corollary is that it will allow one to prove that a separated algebraic curve over a field is quasi-projective, the divisor defined in (21.9.7) being then necessarily ample in virtue of the Riemann- Roch theorem for curves (chap. V).

(21.9.8). For locally Noetherian preschemes of dimension , proposition (21.9.4) reduces the determination of to the case where , a Noetherian local ring of dimension 1.

When is a regular local ring of dimension 1 (in other words, a discrete valuation ring), the group is canonically identified with (21.6.8); consequently, in virtue of (21.9.2):

Proposition (21.9.9).

Let be a regular locally Noetherian prescheme of dimension . Then the sheaf is canonically isomorphic to , where is the additive group considered as a sheaf of groups on the space .

(21.9.10). Assume only that the Noetherian local ring of dimension 1 is reduced; then, if is the normalization of (integral closure of in its total ring of fractions), is Noetherian in virtue of the Krull-Akizuki theorem (Bourbaki, Alg. comm., chap. VII, §2, n° 5, prop. 5), and one saw in (21.8.6) how

may be obtained as an extension of , the latter group being of the form .

Proposition (21.9.11).

Let be a Noetherian prescheme, X_0 a closed subprescheme of having the following properties:

.

2° For every locally closed part of such that is discrete, there exists a part of , closed in and open in , containing .

Under these conditions:

(i) Let Z_0 be the union of the sets as ranges over the part of formed by points such that is finite. Then, for every divisor D_0 on X_0 whose support does not meet Z_0, there exists a divisor on whose inverse image under the canonical injection exists (21.4) and equals D_0; if moreover , one may suppose .

(ii) Assume in addition that there exists in X_0 an affine open U_0 containing (a condition automatically satisfied when there exists an ample -Module (II, 4.5.4)). Then the canonical homomorphism (21.3.2.4) is surjective.

(i) Taking (21.9.6) into account, one may restrict to proving the assertion concerning divisors ; in virtue of (21.9.4), the support of D_0 is a finite discrete and closed set in X_0. One may suppose , that is, , otherwise there is nothing to prove. For every , there exists an element which is not a zero-divisor in this ring, belongs to its maximal ideal, and whose image in is . There exists an affine open neighbourhood of in and a section of over such that is the image in of the germ ; we shall see that by taking small enough, one can arrange for not to be a zero-divisor in , in other words (3.1.9), denoting by the closed part of formed by such that , one has . Now, if , the hypothesis entails that one has either , or is not isolated in . By replacing by a smaller open, one may suppose that the second case occurs for a ; would therefore contain an irreducible component of dimension 1 of containing ; but this would mean that the image of in would be in the nilradical of this ring, and consequently would belong to the nilradical of , which is absurd. One has therefore for every point , and since this set is finite, one may, by replacing by a smaller open neighbourhood of , suppose that .

By virtue of the choice of , one may define a divisor on by ; moreover, one saw above that is necessarily isolated in , so by replacing again by a smaller open neighbourhood of , one may suppose that is reduced to the point . But there exists then, in virtue of condition 2°, a part of , closed in and open in , such that . Replacing again by a smaller open neighbourhood of , one may therefore suppose

that , in other words that is closed in . One may then define a divisor on by the condition that and , which makes sense because in the restriction of is an invertible section, so the restrictions to this open of the two divisors considered on and are equal. It then suffices, to answer the question, to take , which makes sense since is finite.

(ii) Taking the commutative diagram (21.4.2.1) into account, the conclusion will follow from (i) if one proves that every invertible -Module is isomorphic to an -Module of the form (21.2.8), where D_0 is a divisor on X_0 whose support does not meet Z_0. Now, since U_0 is schematically dense in X_0 (20.2.13, (iv)), it suffices for this to prove that there exists a section such that at the points of (3.1.9). One may therefore restrict to the case where is affine; but then is ample (II, 5.1.4) and since the set is finite, the conclusion follows from (II, 4.5.4).

Corollary (21.9.12).

Let be a Henselian local ring (18.5.8), , the closed point of , a separated morphism of finite presentation, and suppose that the set is of dimension . Then, for every closed subscheme of having X_0 as underlying set and of finite presentation over , the canonical homomorphism (21.3.2.4) is surjective.

Let us first show that it suffices to prove the corollary when the Henselian local ring is moreover Noetherian. One knows in fact (18.6.15) that one can write , where the are Noetherian Henselian local rings, the homomorphisms being local; there exists in addition an index and a separated morphism of finite presentation such that and are deduced from and by the base change (8.10.5, (v)); with the usual notation (8.8.1), the morphisms will be separated for and one will have . Moreover, one may assume that, if is the unique closed point of , there is a closed subscheme of , having the same underlying set as , such that (8.6.3); one has, for , by transitivity of fibres and (4.1.4). It is a matter of seeing that if one has proved that the homomorphism is surjective for every , then it is the same for . One evidently has the commutative diagram of canonical homomorphisms

                         Pic(X_λ) ──→ Pic(X)
                              │           │
                              ↓           ↓
                         Pic(X'_{0λ}) ─→ Pic(X'_0)

For every invertible -Module , there exist a and an invertible -Module such that is deduced from it by the base change (8.5.2 and 8.5.5); it suffices to

consider an invertible -Module such that is the image of in , then to take the invertible -Module deduced from by the base change; in virtue of the commutativity of the preceding diagram, will be the image of .

Therefore assume Noetherian, hence so is , and verify that and satisfy the conditions of (21.9.11, (ii)). One has by hypothesis ; on the other hand, to verify condition 2° of (21.9.11), consider a subprescheme of having as underlying set; the morphism , restriction of , being quasi-finite at each of the points of (), one may apply (18.5.11, c) and one sees that is the sum of the open subpreschemes () and of a prescheme Y'' such that , and moreover the canonical injections are such that is a finite morphism. Since is separated, is also a finite morphism, so is closed in ; one therefore answers the question by taking .

It remains to verify the supplementary hypothesis of (21.9.11, (ii)). Now, being a separated curve over the field , is a quasi-projective -scheme (chap. V) 1, so there exists an ample -Module (II, 5.3.1), and this completes the proof.

Remark (21.9.13).

Under the conditions of (21.9.12), assuming moreover proper, the morphism is even projective: indeed, if is an ample -Module, there exists, in virtue of (21.9.12), an invertible -Module whose inverse image in is isomorphic to , hence is ample. Since every neighbourhood of in is necessarily all of , one then deduces from (9.6.4) that is an ample -Module, whence the conclusion (II, 5.3.1 and II, 5.5.3).

21.10. Inverse images and direct images of 1-codimensional cycles

In a later chapter, devoted to intersection theory, the notions of inverse image and direct image of cycles will be developed systematically. In the present number, we content ourselves with defining these notions in certain useful particular cases, and for 1-codimensional cycles, these definitions being chosen so that they are compatible with the corresponding definitions for divisors (21.4 and 21.5), taking account of the homomorphism cyc defined in (21.6).

(21.10.1). Let , be two locally Noetherian preschemes, a morphism, a part of ; assume that the image under of every maximal point of is a maximal point of , and that, for every (i.e. such that ), one has one of the three following conditions for the point :

(i) ;

(ii) and is a flat -module;

(iii) the ring is factorial and .

Under these conditions, we propose, for every 1-codimensional cycle on with support contained in , to define a 1-codimensional cycle , so

that is a homomorphism of ordered groups from the subgroup of formed by cycles with support contained in to the ordered group . For this, let

  (21.10.1.1)            Z = ∑_{x ∈ T ∩ X^{(1)}} n_x · {x}

where the family of such that is locally finite. For every , let us define an integer in the following way, setting :

1° if , take ;

2° if and is a flat -module, one knows (6.1.1) that ; in other words is an -module of finite length , and one takes ;

3° if is factorial and , one knows (21.6.9) that the canonical homomorphism cyc : Div(𝒪_{X,x}) → 𝔍^1(Spec(𝒪_{X,x})) is bijective, and on the other hand since and , consists solely of the maximal points of , so the hypothesis on implies that , and it follows from (21.4.5, (ii)) that the homomorphism is defined; finally, being the unique closed point of , is canonically isomorphic to . One has therefore a composite canonical homomorphism

                                    cyc⁻¹              f^*                cyc
  (21.10.1.2)            𝔍^1(Spec(𝒪_{X,x})) ─→ Div(𝒪_{X,x}) ─→ Div(𝒪_{X',x'}) ─→ 𝔍^1(Spec(𝒪_{X',x'})) ⥲ ℤ.

If is the cycle on , one takes for the image of under the homomorphism (21.10.1.2).

(21.10.2). We propose to show that:

A) When two of the conditions 1°, 2°, 3° of (21.10.1) are simultaneously satisfied, the corresponding values of coincide.

B) The set of such that is locally finite in .

To prove A), assume first that and that verifies one of conditions 2° or 3° of (21.10.1); then and if one is in case 2°, one has ; if one is in case 3°, one has since , so again . It remains to consider the case where one is at once in case 2° and in case 3°; then, since , is a discrete valuation ring; if is a uniformizer of this ring, the divisor corresponding to is in , and its image in is the divisor of , where is the (regular) element of image of . One may evidently restrict to the case where , and then the definition (21.6.5.1) shows that the image of under (21.10.1.1) is the number , which completes the proof of A).

Let us now prove B). Set , ; it suffices to prove that the relation implies that belongs to the set of maximal points of , the latter being locally finite in . It is immediate that one necessarily has ; if were not maximal in the closed set , there would exist a generization of in , distinct from , and since , would necessarily

be a maximal point of ; consequently would be a maximal point of by hypothesis; but this is absurd since and T_0 is purely of codimension 1 in .

(21.10.3). One can now set

                         f^*(Z) = ∑_{x' ∈ X'^{(1)}} n_{x'} · {x'},

the sum on the right-hand side making sense in virtue of what was proved in (21.10.2); one says that the 1-codimensional cycle is the inverse image of under . It is immediate that the map thus defined is a homomorphism of ordered groups. Moreover, if is an open of , an open of such that , and the restriction of , it follows at once from the definitions that one has

  (21.10.3.1)            f'^*(Z | U) = f^*(Z) | V.

Denote by the largest subsheaf of commutative groups of with support contained in ; it follows from relation (21.10.3.1) that the maps just defined determine a homomorphism of sheaves of ordered commutative groups on

where is the continuous map underlying the morphism .

Proposition (21.10.4).

Assume the conditions of (21.10.1) are verified. Then, for every divisor on such that and that is defined (21.4.2), one has

The question being local on , one may restrict to the case where is affine, , a regular non-invertible element of , and where the subprescheme (21.2.12) has a single maximal point , so that , where is the length of (21.6.5.1). One saw in the proof of (21.10.2, B) that the points such that are maximal points of . If the point is in case 3° of (21.10.1), the equality of the multiplicities at of the two members of (21.10.4.1) follows from the definition of by means of the homomorphism (21.10.1.1). Suppose on the contrary that is in case 2° of (21.10.1), and let ; since , one necessarily has . Let us remark now that , where is the image of in , and ; the multiplicity of at the point is therefore the length of the -module ; since , it follows from (4.7.1) that one has , so the multiplicities at of the two members of (21.10.4.1) are again equal, which completes the proof.

(21.10.5). Suppose now that is a morphism of locally Noetherian preschemes, sending every maximal point of to a maximal point of , and suppose in addition that for every rare closed part of , verifies the conditions

of (21.10.1); this means again that for every , either is a maximal point of , or verifies one of conditions (ii), (iii) of (21.10.1). If one takes into account that every 1-codimensional cycle has rare support in , one sees that is defined for every 1-codimensional cycle on ; in virtue of (21.10.3.1), one has thus defined a homomorphism of sheaves of ordered commutative groups

If, moreover, for every divisor on , is defined (21.4.5), the fact that the support of is rare in (21.6.6) entails that (21.10.4) is applicable, and one therefore has the formula (21.10.4.1) for every divisor on . In particular:

Proposition (21.10.6).

Let , be two locally Noetherian preschemes, a flat morphism. Then is defined for every 1-codimensional cycle on , is defined for every divisor on , and one has relation (21.10.4.1).

Indeed, if is such that is not maximal, it follows from (6.1.1) that one necessarily has , so one is in case (ii) of (21.10.1). One may therefore apply (21.10.5), taking account of (21.4.5) and (2.3.4).

Remark (21.10.7).

The existence of for every 1-codimensional cycle on already follows from the hypothesis that is flat at the points of of codimension in (i.e. such that ); indeed, for every maximal point , it follows from (6.1.1) that is a maximal point of since . Similarly, if , is maximal or belongs to by (6.1.1); one can therefore again apply (21.10.5).

Proposition (21.10.8).

Let , , X'' be three locally Noetherian preschemes, , two morphisms; assume that (resp. ) is flat at every point of codimension in (resp. X''). Then is flat at every point of codimension in X'' and for every 1-codimensional cycle on , one has .

The first assertion follows from (6.1.1) and (2.1.6). The second results from the fact that (resp. , resp. ) sends maximal points of (resp. X'', resp. X'') to maximal points of (resp. , resp. ), and that, if is such that and , one has

                  long(𝒪_{X'',x''} / 𝔪_x 𝒪_{X'',x''}) = long(𝒪_{X'',x''} / 𝔪_{x'} 𝒪_{X'',x''}) · long(𝒪_{X',x'} / 𝔪_x 𝒪_{X',x'}).

Indeed, setting , , , , , one has , and the formula

                         long_{A''}(A'' / 𝔪 A'') = long_{A'}(A' / 𝔪 A') · long_{A''}(A'' / 𝔪' A'')

follows from (4.7.1) and from the flatness hypothesis of A'' over .

(21.10.9). Let be a locally Noetherian prescheme, a commutative group, written additively and therefore considered as a -module; one will denote again by and the simple sheaves on associated to the constant presheaves equal respectively to and . The sheaf of commutative groups is called the sheaf of germs of 1-codimensional cycles with coefficients in . If , one says that the sections of

over are the 1-codimensional cycles with rational coefficients. Since the stalks of are torsion-free -modules (21.6.3), the canonical homomorphism is injective, so 1-codimensional cycles are identified with 1-codimensional cycles with rational coefficients.

(21.10.10). We are going to see that under certain conditions, one may broaden the definition of given in (21.10.3) for a 1-codimensional cycle on , but on condition of taking for a 1-codimensional cycle with rational coefficients on . The more general case in which we place ourselves is that where sends every maximal point of to a maximal point of , and where, at every point , one has one of conditions (i), (ii), (iii) of (21.10.1) or a fourth condition (setting ):

(iv) , , and moreover, if one sets , , and if denotes the total ring of fractions of , then is a finite -algebra and is a free -module.

Let us denote then by the rank of the free -module , by the degree of over , and set . For a 1-codimensional cycle with support contained in , given by (21.10.1.1), and every , one defines as equal to the number when one is in one of cases 1°, 2°, 3° of (21.10.1); but there remains here a fourth possibility:

4° if verifies condition (iv) above, take .

(21.10.11). It remains to prove that when condition 4° of (21.10.10) is verified simultaneously with one of conditions 1°, 2°, 3° of (21.10.1), one has . This is evident if since then . To study the two other cases, note that is closed for the -preadic topology, so the completion of for the latter topology is . If one is at once in case 2° and case 4°, is discrete for the -preadic topology, so isomorphic to . Since is a finite and flat -algebra , it is a free -module , and the rank of over is equal to the rank of over , hence also to that of over . On the other hand, this rank is also equal to the product of the length of (as -module, or as -module) by the rank , which proves the relation in this case.

Assume finally that one is at once in case 3° and case 4°. Then, since , is a discrete valuation ring, hence regular. On the other hand, is of dimension 1, and since , is a Cohen-Macaulay ring (0, 16.4.6); finally, since is an -module of finite type, is a -vector space of finite rank; since is contained in , it is also a -vector space of finite rank, hence an Artinian ring. Applying (6.1.5) then shows that is a flat -module, so one is also in case 2°, and one concludes by what was seen above.

This being so, in the case under consideration, one will set

                         f^*(Z) = ∑_{x' ∈ X'^{(1)}} c_{x'} · {x'}

which is therefore a 1-codimensional cycle on with rational coefficients. One has again defined in this way a homomorphism of ordered groups, satisfying (21.10.3.1), and consequently a homomorphism of sheaves of commutative groups

                         ψ^*(Γ_T(𝔍^1_X)) → 𝔍^1_{X'} ⊗_ℤ ℚ.

When verifies the preceding conditions for every closed part rare in , that is, when for every , either is a maximal point of , or verifies one of conditions (ii), (iii) or (iv), is then defined for every 1-codimensional cycle on , and one has defined a homomorphism of sheaves of ordered commutative groups

                         ψ^*(𝔍^1_X) → 𝔍^1_{X'} ⊗_ℤ ℚ

whence, by tensorization, a homomorphism of sheaves of ordered -vector spaces

  (21.10.11.1)           ψ^*(𝔍^1_X ⊗_ℤ ℚ) → 𝔍^1_{X'} ⊗_ℤ ℚ.

Remark (21.10.12).

When one is in the situation of (21.10.10), one may effectively, for 1-codimensional cycles on , have for 1-codimensional cycles with non-integral coefficients; in other words, the numbers may be non- integers. One has an example by taking the complete integral ring of (6.15.11, (ii)) and its integral closure : the closed point of verifies condition (iv) of (21.10.10) and one has .

Lemma (21.10.13).

Let be a Noetherian local ring of dimension 1, a regular element of belonging to the maximal ideal (which implies that ).

(i) For every -module of finite type, the kernel and cokernel of the homothety of ratio are of finite length. One sets .

(ii) If is an exact sequence of -modules of finite type, one has .

(iii) One has for every -module of finite type; for to hold, it is necessary and sufficient that be of finite length.

(iv) If is the total ring of fractions of and if is a free -module of rank , one has .

(v) If verifies the hypotheses of (iv) and moreover is -regular, one has .

(i) consists of the point and the minimal prime ideals ; since by hypothesis for every (Bourbaki, Alg. comm., chap. IV, §1, n° 1, cor. 3 of prop. 2), the image of in each of the is invertible, and the supports of the -modules of finite type and are therefore empty or reduced to ; one concludes (0, 16.1.10) that these modules are of finite length.

(ii) Since is regular, one has an exact sequence

                              t_A
                         0 → A ─→ A → A / tA → 0

and since for , the exact sequence of Tor gives on the one hand the exact sequence

                                                          t_M
                         0 → Tor_1^A(M, A / tA) → M ─→ M → M / tM → 0

and on the other, for ,

                         0 = Tor_i^A(M, A) → Tor_i^A(M, A / tA) → Tor_{i-1}^A(M, A) = 0

so one has and for ; the exact sequence of Tor gives an exact sequence

  0 → Tor_1^A(M', A / tA) → Tor_1^A(M, A / tA) → Tor_1^A(M'', A / tA) →
                              M' ⊗_A (A / tA) → M ⊗_A (A / tA) → M'' ⊗_A (A / tA) → 0

and this sequence is written, by the foregoing,

  0 → N_t(M') → N_t(M) → N_t(M'') → P_t(M') → P_t(M) → P_t(M'') → 0

which proves (ii).

To prove (iii), note that there is a composition series of whose quotients are isomorphic to or to one of the (Bourbaki, Alg. comm., chap. IV, §1, n° 4, th. 1), and for to be of finite length, it is necessary and sufficient that all these quotients be isomorphic to . Everything therefore reduces (in virtue of (ii)) to proving that and . Now, the image of in being 0, one has and , whence the first assertion; on the other hand, the image of in is regular, so and , which is not reduced to 0, whence the second assertion.

(iv) There is a basis of of the form , where is a regular element of and . Consider the homomorphism which sends the element () of the canonical basis of to , and let us show that and are of finite length; indeed, for every , the image of in is invertible, and since , the images of the in form a basis of this -module; so is bijective. As in (i), one concludes that the supports of and are contained in , so that these modules are of finite length. This being so, it follows from (ii) and (iii) that one has since is regular.

Finally, it is clear that (v) is at once deduced from (iv), since then .

This lemma allows one to generalize (21.10.4):

Proposition (21.10.13).

Assume that verifies the conditions of (21.10.10). Then, for every divisor on such that and that is defined, one has

Reasoning as in (21.10.4), everything reduces to seeing (with the same notation) that if is in case 4° of (21.10.10) and if is the length of , then the

length of is equal to , being the rational number defined in (21.10.10). Since is of finite length, it has the same length as its -preadic completion , which one may also write ; moreover, since is regular by hypothesis in , is also regular in by flatness , and when is considered as -module, one may also say that is -regular. Since is of dimension 1 and is an -module of finite type such that is a free -module of rank , one may apply (21.10.13, (v)) to and to , and the length of as -module is therefore . Since is a -vector space of rank , the length of as -module is therefore , which completes the proof.

(21.10.14). Let now , be two locally Noetherian preschemes, a morphism having the two following properties:

a) is finite;

b) the image under of every maximal point of is a maximal point of .

For every , the points all belong to , as follows from hypothesis b) and the inequality (0, 16.3.9.1), the fibre being discrete. Let then

                         Z' = ∑_{x' ∈ X'^{(1)}} n_{x'} · {x'}

be a 1-codimensional cycle on . For every , let us define an integer by the formula

                         n_x = ∑_{x' ∈ f⁻¹(x)} n_{x'} · [k(x') : k(x)]

which makes sense, the points of being finite in number and being a field of finite degree over (I, 6.4.4). Moreover, the set of such that is locally finite in , since it is contained in the image under of the set of such that , and the conclusion follows from the fact that the morphism is quasi-compact. One may thus define a 1-codimensional cycle on by setting

  (21.10.14.1)           f_*(Z') = ∑_{x ∈ X^{(1)}} n_x · {x}

and one says that is the direct image of under . It is clear that the map thus defined is a homomorphism of ordered groups. Moreover, if is an open of , the restriction of , it follows at once from the definitions that one has

  (21.10.14.2)           (f_U)_*(Z' | f⁻¹(U)) = f_*(Z') | U

so, denoting by the continuous map underlying the morphism , the maps just defined determine a homomorphism of sheaves of ordered commutative groups on

Proposition (21.10.15).

Let , , X'' be three locally Noetherian preschemes, , two morphisms verifying conditions a) and b) of (21.10.14). Then verifies the same conditions, and for every 1-codimensional cycle Z'' on X'', one has .

This follows at once from the definitions.

Proposition (21.10.16).

Let , , X_1 be three locally Noetherian preschemes, a morphism verifying conditions a) and b) of (21.10.14), a flat morphism. Set (so that is locally Noetherian) and note and the canonical projections. Then verifies conditions a) and b) of (21.10.14), and for every 1-codimensional cycle on , one has

It is clear that is finite, and it verifies condition b) of (21.10.14) by virtue of (2.3.7). To prove (21.10.16.1), one is at once reduced to the case where , and X_1 are spectra of Noetherian local rings of dimension 1, , and A_1, with a finite -module and A_1 a flat -module. Denoting by , and the closed points of , and X_1 respectively, it is a matter of seeing that one has

  (21.10.16.2)           ∑_{x'_1} λ_{x'_1}[k(x'_1) : k(x_1)] = λ_{x'}[k(x') : k(x)]

where ranges over the set of closed points of (i.e. the set of points lying above both and ) and where and . Since one has

                         [k(x'_1) : k(x_1)] · [k(x_1) : k(x)] = [k(x'_1) : k(x')] · [k(x') : k(x)]

the left-hand side of (21.10.16.2) is also written

                         ([k(x') : k(x)] / [k(x_1) : k(x)]) · long_{A'}(A'_1 / 𝔪_{x'} A'_1) = long_{A_1}(A'_1 / 𝔪_x A'_1)

where one has set . One therefore has , and since A_1 is a flat -module, one has by (4.7.1)

                         long_{A_1}(A'_1 / 𝔪_x A'_1) = long_A(A' / 𝔪_x A') · long_{A_1}(A_1 / 𝔪 A_1)
                                                    = [k(x') : k(x)] · λ_{x_1}

which completes the proof.

Proposition (21.10.17).

Let , be two locally Noetherian preschemes, a finite locally free morphism. Then verifies condition b) of (21.10.14), and for every divisor on , one has

Since is flat and finite, condition b) of (21.10.14) and the relation follow from (6.1.1). The definition (21.10.14.1) shows that one may reduce to the case where for an ; then , where is an

-algebra which is a free -module of finite rank; moreover, one may assume that , where is a regular element of ; one then has , where is a regular element of (21.5.2). One may restrict to the case where is not invertible in , which is equivalent to not being invertible in ; the ring is then of finite length and , and is a direct sum of Artinian local rings (), the residue field of being , where () are the points of above . If () is the image of in , is the direct sum of the ; since the product equals , one sees that the multiplicity at the point of the left-hand side of (21.10.17.1) is , so that the formula to prove reduces to

  (21.10.17.2)           long_A(A' / t' A') = long_A(A / tA).

This relation follows from the more general following lemma:

Lemma (21.10.17.3).

Let be a Noetherian local ring of dimension 1, a free -module of finite rank, an injective endomorphism of . Then one has

  (21.10.17.4)           long_A(Coker u) = long_A(A / (det u) A).

Let us distinguish several cases.

I) is a discrete valuation ring. Indeed, let be a uniformizer of , and remark that ; if is the rank of and () are the invariant factors of , is the direct sum of the -modules , so has length , and is the product of an invertible element and , whence the conclusion in this case (Bourbaki, Alg., chap. VII, §4, n° 5, cor. 1 of prop. 4).

II) is a complete integral ring (of dimension 1). One knows then (0, 19.8.8, (ii)) that there is a subring of which is a discrete valuation ring, such that is a local homomorphism making a -module of finite type; since this -module is evidently torsion-free, it is free (Bourbaki, Alg. comm., chap. VI, §3, n° 6, lemma 1). Denote by the set endowed with its (free) -module structure, by the endomorphism regarded as a -endomorphism. It follows from I) that one has

  (21.10.17.5)           long_B(Coker u') = long_B(B / (det u') B).

But one has (Bourbaki, Alg., chap. VIII, §12, n° 2, prop. 7)

                         det u' = N_{A/B}(det u),

hence, applying (21.10.17.4) to the homothety of the free -module , it comes

                         long_B(B / (det u') B) = long_B(A / (det u) A),

whence, substituting in (21.10.17.5) and dividing by , the length of the -module , one obtains (21.10.17.4) in the case envisaged.

III) is a complete ring. Note that one may suppose in addition that ; indeed, since is a regular element of , if one had , one would deduce , so would be invertible, an automorphism of , and the formula (21.10.17.4) becomes then trivial, both members being zero.

In what follows, for an endomorphism of a module over a ring , such that Ker v and Coker v be of finite length, one will set

                         χ(N, v) = long_R(Ker(v)) − long_R(Coker(v)).

One will note that the hypothesis on amounts to saying that the complex

                                            v
                         K^0 : 0 → N ──→ N → 0

has its cohomology modules of finite length and that . One deduces that if , , N'' are three -modules, if one has a commutative diagram

                                 r        s
                         0 → N' ─→ N ─→ N'' → 0
                              │     │     │
                              │v'   │v    │v''
                              ↓     ↓     ↓
                         0 → N' ─→ N ─→ N'' → 0
                                 r        s

whose rows are exact, and if two of the three numbers , and are defined, then it is the same of the third and one has

  (21.10.17.6)           χ(N, v) = χ(N', v') + χ(N'', v'').

This follows at once from the exact cohomology sequence.

Finally, if is an -module of finite length, one has .

With these notations, one has the following lemma:

Lemma (21.10.17.7).

Let be a Noetherian local ring of dimension 1 whose maximal ideal is such that ; let () be the minimal prime ideals of . Let be a free -module of finite type, an endomorphism of such that is defined; for each , set and let be the endomorphism of ; then, if is defined for each , one has

  (21.10.17.8)           χ(M, u) = ∑_{i=1}^n long(A_{𝔭_i}) · χ(M_i, u_i).

Since , one has a unique reduced primary decomposition , where the ideal is -primary for . If one sets , one then has an exact sequence

                         0 → M → ⨁_i M'_i → M'' → 0

of -modules, where M'' is of finite length: indeed, localizing the preceding exact sequence at each of the , one obtains , since and for (Bourbaki, Alg. comm., chap. IV, §2, n° 4, prop. 6), so and ;

the support of M'' being therefore reduced to , M'' is of finite length (0, 16.1.10). If one sets , and if u'' is the endomorphism of M'' deduced from by passage to quotients, one deduces from (21.10.17.6) that , and since M'' is of finite length, . To prove (21.10.17.8), one may therefore reduce to the case where . One will then denote by the unique minimal prime ideal, which is the nilradical of ; if , , , it is a matter of proving that if is defined, one has

  (21.10.17.9)           χ(M, u) = long A_𝔭 · χ(M_0, u_0).

Let () be the "-th symbolic power" of , inverse image in of the ideal of (), with and ; set

                         M_j = 𝔫_j M / 𝔫_{j+1} M       (0 ≤ j ≤ r − 1),

and denote by the endomorphism of deduced from by restriction to and passage to quotients. We shall first show that each of the numbers is defined and that one has

  (21.10.17.10)          χ(M, u) = ∑_j χ(M_j, v_j).

The first assertion will entail the second, by applying (21.10.17.6) to each of the exact sequences

                         0 → 𝔫_j M / 𝔫_{j+1} M → M / 𝔫_{j+1} M → M / 𝔫_j M → 0.

To prove the first assertion, one notes that if is the rank of the free -module , is isomorphic to , or also (since annihilates each of the quotients ), is an A_0-module isomorphic to . Denote by the rank of the A_0-module ; since the field of fractions K_0 of A_0 is the residue field of , is also the length of the -module . There is a system of generators of the A_0-module which contains a basis of ; so there is an A_0-homomorphism

whose localization at the ideal (0) of A_0 is an isomorphism, so that the supports of and of are reduced to the maximal ideal of A_0; and are therefore A_0-modules of finite length (0, 16.1.10). Since by hypothesis is defined, the same holds for , and in virtue of (21.10.17.6) and of the fact that and are of finite length, one sees that is defined and equal to ; relation (21.10.17.10) then gives

                         χ(M, u) = (∑_j l_j) χ(M_0, u_0),

and in virtue of a previous remark, is none other than the length of , which completes the proof of the lemma (21.10.17.7).

To apply this lemma when is a complete ring and , one observes that if is injective, the same holds for the (with the notation of the lemma): indeed, det u is then a regular element of , so does not belong to any of the , and its image in is consequently an element of this integral ring, which proves that is injective. Since is image of , it is also of finite length and is therefore defined for every ; one has accordingly the formula (21.10.17.8). On the other hand, since is a regular element of , it is contained in none of the ; the ideal (det u) A is therefore -primary and the quotient of finite length. Applying the same reasoning as above to the injective homothety of and its images in the , it comes

                         χ(A, t) = ∑_i long(A_{𝔭_i}) · χ(A / 𝔭_i, t_i).

But the rings are integral and complete, and applying the result of II) to each of them, one again obtains relation (21.10.17.4) for and .

IV) General case. Set , , ; one has , and by flatness, and ; since the formula (21.10.17.4) is true for and in virtue of III), it is also true for and in virtue of (4.7.1).

This completes the proof of (21.10.17).

Proposition (21.10.18).

Let , be two locally Noetherian preschemes, a finite morphism, sending every maximal point of to a maximal point of , and verifying for every one of conditions (ii), (iii), (iv) of (21.10.10). Assume in addition that there exists an integer such that, for every maximal point of , is a free -module of rank . Then, for every 1-codimensional cycle on , one has

("projection formula").

By virtue of the definitions, one is at once reduced to the case where is the spectrum of a Noetherian local ring of dimension 1, with closed point , and where ; set , where is a finite -algebra, and, for every minimal prime ideal of , is a free -module of rank . Let us show that one may further restrict to the case where is complete. Make in effect the base change , where , with , and set and ; the morphism is then finite, and since is flat, the maximal points of lie above those of ; above each of the , there is only a finite number of minimal ideals of , and is a free -module of rank . Finally, if verifies one of conditions (ii), (iii) or (iv) of (21.10.10) at each of the points of , verifies the corresponding condition at the unique point of above (denoting by the closed point of ); this is immediate for conditions (ii) and (iv); for condition (iii), it implies that

is a discrete valuation ring, so the same holds for , and the condition follows, by flatness, from the condition (3.3.1). The morphism therefore verifies the same conditions as ; if one proves the formula (21.10.18.1) for and , the same formula will be valid for and , in virtue of (21.10.16.1).

One may therefore assume that is complete; then so is , which is thus the direct sum of complete local rings; one may consequently restrict to the case where is a local ring, and it remains to verify (21.10.18.1) in each of the cases (ii), (iii), (iv) for the closed point of . In case (ii), being a flat -module of finite type, is a free -module , of rank in virtue of the hypothesis. Now, one has by definition (21.10.1 and 21.10.3) , where is the length of the -module , then ; but is the length of as -module, or also its rank as -vector space, hence equals .

In case (iii), is a discrete valuation ring, hence regular, and the hypothesis entails that is a Cohen-Macaulay ring (0, 16.4.6); since and is an Artinian ring, it follows from (6.1.5) that is a flat -module, and one is reduced to case (ii).

In case (iv), if is the total ring of fractions of , is by hypothesis a free -module of rank and by definition (21.10.10), one has , whence . Q.E.D.

One will note that the formula (21.10.18.1) is applicable in particular when the morphism is finite and flat and such that for every maximal point of , is a free -module of rank .

Corollary (21.10.19).

Under the hypotheses of (21.10.18), let be a divisor on such that is defined (21.4.5); then one has

This follows from (21.10.18) and (21.10.13).

21.11. Factoriality of regular local rings

Theorem (21.11.1) (Auslander-Buchsbaum).

A regular Noetherian local ring is factorial.

The proof that follows is due to I. Kaplansky.

Let be a regular Noetherian local ring of dimension ; we shall reason by induction on . For , is a field, and for , a discrete valuation ring, hence principal (and a fortiori factorial). Suppose then and the theorem proved for regular rings of dimension . Set , denote by the closed point of , and set . At every point , one has , and since is regular, the rings are also regular (0, 17.3.2), so the induction hypothesis entails that they are factorial. Moreover one has since is regular, hence Cohen-Macaulay (0, 17.1.3).

Using (21.6.14), one is reduced to proving that . Therefore consider an invertible -Module , and prove that it is isomorphic to . It follows from (I, 9.4.5) that there exists a coherent -Module such that . Since is regular, hence of finite cohomological dimension (0, 17.3.1), there exists a finite left resolution of :

                         0 ← ℱ ← 𝒪_X^{n_1} ← 𝒪_X^{n_2} ← … ← 𝒪_X^{n_h} ← 0

(0, 17.2.8 and 0, 17.2.2, (iii)). By restriction to , one therefore has a finite resolution

  (21.11.1.1)            0 ← ℒ ← 𝒪_U^{n_1} ← 𝒪_U^{n_2} ← … ← 𝒪_U^{n_h} ← 0.

The theorem will then follow from the general considerations that follow. On a ringed space , let be a locally free -Module of finite rank; one will denote by the invertible -Module which, in a neighbourhood of each point of , equals , denoting by the rank of in this neighbourhood (which may vary with the connected component of ). With this notation, one has the

Lemma (21.11.1.2).

Let be a ringed space in local rings, and

an exact sequence of -Modules locally free of finite rank; then the invertible -Module is isomorphic to .

Let us show how this lemma will allow one to conclude the proof of (21.11.1). It suffices to note for this that, for every integer , is isomorphic to , as is . Since on the other hand for every invertible -Module , the lemma (21.11.1.2), applied to the exact sequence (21.11.1.1), shows that is isomorphic to .

It remains to prove (21.11.1.2); one proceeds by induction on , the lemma being trivial for . For , is an -Module locally free of finite rank , and one has the two exact sequences

In virtue of the induction hypothesis, is isomorphic to ; it will therefore suffice to define a canonical isomorphism (Λ^{max} 𝒩) ⊗ (Λ^{max} ℰ_0) ⥲ Λ^{max} ℰ_1 to complete the proof. Now, there exists an open covering of such that in every , is the direct sum of and of an -Module locally free , whence a canonical isomorphism . Since one has a canonical isomorphism

                         r_α : (Λ^{max} 𝒩 | U_α) ⊗ (Λ^{max} ℳ_α) ⥲ (Λ^{max} ℰ_1) | U_α

one deduces from this, by means of , an isomorphism

                         u_α : (Λ^{max} 𝒩 | U_α) ⊗ (Λ^{max} ℰ_0 | U_α) ⥲ Λ^{max} ℰ_1 | U_α

and it is a matter of showing that and coincide in for any two indices , . Now, if and are the restrictions to of and respectively, one has , where is the "projection parallel to " such that for every section , with ; the identity of and follows at once from this fact and from the definition of the canonical isomorphism (Bourbaki, Alg., chap. III, 3rd ed.).

21.12. Van der Waerden's purity theorem for the ramification locus of a birational morphism

(21.12.1). Let and be two preschemes, a quasi-compact and quasi-separated morphism, so that is a quasi-coherent -Algebra (1.7.4). We call affine envelope of the -prescheme the -prescheme affine over

                       U^∘ = Aff(U/X) = Spec(f_*(𝒪_U)) = Spec(𝒜(U))                  (II, 1.1.1).

If is the structural morphism, one has therefore by definition

                       𝒜(U^∘) = f^∘_*(𝒪_{U^∘}) = f_*(𝒪_U) = 𝒜(U),

and to the identity isomorphism of there corresponds by (II, 1.2.7) a canonical -morphism

For to be an isomorphism, it is evidently necessary and sufficient that the morphism be affine.

For every -prescheme affine over , the map is a bijection

                       Hom_X(U^∘, V) ⥲ Hom_X(U, V)

functorial in : this results from the existence of the canonical bijections and (II, 1.2.7).

One can therefore say that represents the covariant functor in the category of -preschemes affine over . One deduces that, for fixed, is a covariant functor from the category of quasi-compact and quasi-separated -preschemes into the category of -preschemes affine over ; more precisely, if U_1, U_2 are two quasi-compact and quasi-separated -preschemes, to every -morphism there corresponds the unique -morphism rendering the diagram

                       U_1 ────────→ U_2
                        │             │
                  i_{U_1}│             │i_{U_2}
                        ↓             ↓
                       U_1^∘ ──g^∘──→ U_2^∘

commutative.

More generally, consider a commutative diagram

                       U ──u──→ U'
                       │        │
                      f│        │f'
                       ↓        ↓
                       X ──v──→ X'

where the morphisms , are quasi-compact and quasi-separated and the morphism affine. Setting , one has and is an affine morphism; one has consequently (relative to the morphism ), whence a unique -morphism rendering the diagram

                       U ──u──→ U'
                       │        │
                  i_U  │        │i_{U'}
                       ↓        ↓
                       U^∘ ─v^∘→ U'^∘

commutative.

Proposition (21.12.2).

Let be a quasi-compact and quasi-separated morphism, a flat morphism; set , . Then one has a canonical -isomorphism

  (21.12.2.1)                          Aff(U'/X') ⥲ Aff(U/X) ×_X X'.

Indeed, one has , and (II, 1.5.1); the isomorphism of the statement comes from the canonical isomorphism (2.3.1).

Corollary (21.12.3).

For every quasi-compact and quasi-separated morphism and every , one has, up to a canonical isomorphism

  (21.12.3.1)                          U^∘ ×_X Spec(𝒪_{X,x}) = (U ×_X Spec(𝒪_{X,x}))^∘.

It also follows from (21.12.2) that one has, up to a canonical isomorphism, for every open of

(21.12.5). We shall consider in particular the case where is an open immersion, so that is identified with a prescheme induced on an open of . Since the morphism is the identity, it follows from (21.12.4) that the morphism restriction of is an isomorphism, being therefore an open immersion permitting to be identified with a prescheme induced on an open of .

More generally, for an open of , the restriction of is an isomorphism of onto the prescheme induced on the open if and only if the open immersion is an affine morphism. It is clear (II, 1.2.1)

that the union of these opens is the largest of them, U_1, which contains ; by virtue of the foregoing, U_1 is also the largest open not meeting the set

("affineness defect" of the open relative to , which is empty if and only if is affine over ); in other words, the closed set is the closure of the set .

Note that for every flat morphism , if one sets , one has

as follows at once from (21.12.2.1) and (I, 3.4.8). In particular, for every open of , one has

  (21.12.5.2)                          Daf((U ∩ V)/V) = Daf(U/X) ∩ V

and for every ,

  (21.12.5.3)                          Daf((U ∩ Spec(𝒪_{X,x}))/Spec(𝒪_{X,x})) = Daf(U/X) ∩ Spec(𝒪_{X,x}).

We shall, when is locally Noetherian, give precise information on the nature of the set , showing for example that when is everywhere dense in , is not an arbitrary rare closed set:

Theorem (21.12.6).

Let be a locally Noetherian prescheme, a non-empty open of , the canonical injection. Then:

(i) The closure of is of codimension in .

(ii) If is of codimension , the morphism is surjective.

(i) Let us first show that for every point one necessarily has . Indeed, cannot evidently be a maximal point of , being contained in ; we have therefore to see that one cannot have . But this relation would entail, by (21.12.5.3), that , where . But the only opens of are itself and the subsets of the (finite) set of maximal points of . Now the open set reduced to a maximal point of is affine, by definition of preschemes; one concludes that all open sets in are affine, hence , contrary to the hypothesis made.

To prove (i) one must show more, namely that if is such that , there exists an open neighbourhood of in such that does not meet , that is to say such that the canonical injection is affine. But, with the same notations as above, one has just seen that the canonical injection is affine. One can evidently restrict to the case where is Noetherian; since the morphism is of finite presentation, the conclusion results from (8.10.5, (viii)) applied following the method of (8.1.2, a)).

(ii) We must prove that for every point , one has ; we shall first show that one may reduce to the case where , where is a complete Noetherian local ring, and the closed point of . For this, it suffices to make the base change , where ; if one sets , is the canonical injection , and since the morphism is flat, it follows from (21.12.2) that if one proves that belongs to , one deduces that . By virtue of (6.1.1), the reduction sought has indeed been effected.

Let then X_1 be a closed reduced sub-prescheme of whose underlying space is an irreducible component of , of maximal dimension among those which contain an irreducible component of , and set , ; one has therefore (0, 14.2.1), and the pair therefore verifies the same hypotheses as the pair (X_1 being the spectrum of a quotient ring of , hence local Noetherian and complete). One has moreover a commutative diagram

                       U_1 ──→ U
                        │      │
                       f_1│     │f
                        ↓      ↓
                       X_1 ──j→ X

where and are the canonical injections, being therefore an affine morphism; one deduces (21.12.1) the existence of a morphism rendering the diagram

                       U_1^∘ ──→ U^∘
                        │        │
                       f_1^∘│     │f^∘
                        ↓        ↓
                       X_1 ────→ X

commutative, and consequently, to prove that , it suffices to prove that . One can therefore replace by X_1, and one can consequently suppose that the ring is moreover integral. But is the quotient of a regular Noetherian ring by virtue of Cohen's theorem (0, 19.8.8, (i)), and since it is integral, one can apply (5.11.1) with the family reduced to the unique maximal point of ; since by hypothesis, one sees that is a coherent -Module, hence the morphism is finite; since this morphism is dominant ( being everywhere dense), it is surjective (II, 6.1.10), and consequently . Q.E.D.

Corollary (21.12.7).

Let be a locally Noetherian prescheme, a sub-prescheme induced on an open of , the canonical immersion; suppose that is an affine morphism. Then every irreducible component of is of codimension (and consequently of codimension 1 if is everywhere dense).

Suppose indeed that one of the irreducible components T_1 of is of codimension in . Replacing if needed by an open neighbourhood of the generic point of T_1, one can suppose irreducible and of codimension . But then the hypothesis

that is affine implies that is identified with and with , in other words that ; but this contradicts the conclusion of (21.12.6) which, under the dimension hypothesis made, implies that must be surjective.

(21.12.8). Let be a Noetherian local ring, , the unique closed point of , . Consider the following condition:

(W) For every open of contained in , containing no irreducible component of and such that the canonical immersion is affine, is itself an affine open.

This condition is entailed by the following:

(W̃) For every closed part of every irreducible component of which is of codimension 1 in , and such that for every irreducible component of , one has , the open is affine.

Indeed, if (W̃) is verified and if the open of verifies the hypothesis of (W), it follows from (21.12.7) that no irreducible component of can be of codimension ; since moreover contains none of the irreducible components of , condition (W̃) shows that is affine.

Note that condition (W̃) simplifies when is irreducible and is then equivalent to the following:

(W̃') For every irreducible closed part of , of codimension 1 in , is an affine open.

Indeed, it is clear that (W̃) entails (W̃') when is irreducible, and the converse is true by considering the irreducible components of and noting that the intersection of a finite number of affine opens is an affine open (I, 5.5.6).

Examples (21.12.9). If is a factorial Noetherian local ring, it verifies (W̃) (and a fortiori (W)), since every prime ideal of height 1 is principal (I, 1.3.6). But there are non-factorial Noetherian local rings verifying (W), for example those of dimension : indeed, one noted in the proof of (21.12.6, (i)) that all opens of are then affine. One can moreover prove by using local duality theory (chap. III, 3rd part) that every Noetherian local ring of dimension 2 verifies (W).

The interest of condition (W) lies in the following result:

Proposition (21.12.10).

Let , be two locally Noetherian preschemes; a morphism, a closed point of , , ; suppose that the morphism restriction of is an open immersion, and that the local ring verifies condition (W) (21.12.8). Then, for every irreducible component of , either is of codimension in , or its generic point is isolated in . If is locally of finite type over , the second alternative implies that is reduced to a single point.

The last assertion of the statement results from the fact that is then a Jacobson prescheme (10.4.7), and since the set of closed points of is then dense in , the generic point of can be isolated in only if is reduced to a single point.

Suppose that there exists in an irreducible component whose generic point is not isolated in , and such that . The question being local

on and since is non-isolated in , one can, by replacing by an open neighbourhood of in , suppose and affine, irreducible, not reduced to a point and such that . The image is an open of isomorphic to by hypothesis; let us show that on replacing if needed by an open neighbourhood of in , one can suppose that contains none of the irreducible components of whose closure contains . Indeed, one can first suppose that all maximal points of are generizations of , the set of these points being finite; hence the set of maximal points of is the set of images , by of the maximal points of (no maximal point of being able by hypothesis to be contained in , since ). Set . By hypothesis, one has and since is not isolated in , there exists in a point . Since , there exists in a point such that, if one sets , one has (10.5.9); this entails by hypothesis , hence is not a generization of . Replacing by , one sees that the image by of does not contain the images , hence contains none of the irreducible components of whose closure contains . That being the case, since and are affine, the morphism is affine, hence so is the restriction ; since is an open immersion, the canonical immersion is affine. Set , , . The foregoing proves that the canonical immersion is affine, hence U_1 is an affine open in Y_1 by virtue of (W), in other words the canonical immersion is affine. But since this immersion is of finite presentation (1.6.2), it follows from (8.10.5, (viii)) applied following the method of (8.1.2, a)) that on restricting if needed to an affine open neighbourhood of , one can suppose that the immersion is affine. One would conclude that the open of , isomorphic to , would be affine, which would contradict (21.12.7); the proposition is thus proved.

Corollary (21.12.11).

Let , be two locally Noetherian preschemes, being supposed irreducible; let be a morphism locally of finite type, and let be the largest open of such that the restriction is a local isomorphism. Suppose that for every point , the local ring verifies condition (W) (21.12.8). Then every irreducible component of is either of codimension , or such that its generic point is isolated in .

Set . The question depending only on the fibre and the local ring , one can by base change restrict to the case where and where is affine ((I, 3.6.5) and (I, 4.5.5)); replacing if needed by an open neighbourhood of in , one can suppose that is irreducible; moreover one can restrict to the case where is non-empty. The morphism can therefore be supposed separated and closed in ; since the restriction is a local isomorphism and the non-empty open of

is irreducible, it follows from (I, 8.2.8) that is an open immersion. The restriction is therefore also an open immersion, which shows that is either empty or reduced to a point rational over , hence closed in (I, 6.4.2). Replacing once again if needed by a smaller open neighbourhood of in , one can therefore restrict to the case where , in other words ; since moreover is the generic point of , one has , in other words . One is then exactly in the conditions of application of (21.12.10), whence the conclusion.

Theorem (21.12.12) (van der Waerden).

Let , be two locally Noetherian integral preschemes, a birational morphism locally of finite type. Suppose moreover that is normal and that for every , the ring verifies condition (W) of (21.12.8) (conditions fulfilled in particular when is locally factorial (21.12.9)). If is the largest open of such that the restriction is a local isomorphism, all irreducible components of are of codimension 1 in .

Note that since is birational, the open set is non-empty; it therefore suffices to prove that a maximal point of cannot be isolated in , where . But, by restricting to an open neighbourhood of , one can restrict to the case where ; then all the fibres () would be empty or reduced to a point, and it would follow from the Main theorem (8.12.10) that would be a local isomorphism, contrary to the hypothesis .

Corollary (21.12.13).

Suppose the hypotheses of (21.12.12) are verified and in addition suppose that is quasi-finite at each of the points of (recall that these are the points where ); then is a local isomorphism, and if moreover is separated, is an open immersion.

It suffices to prove the first assertion, the second being a consequence of the first and of (I, 8.2.8). Everything reduces to proving, with the notation of (21.12.12), that ; in the contrary case, a generic point of an irreducible component of would belong to by virtue of (21.12.12), and by hypothesis it would be isolated in with ; but we saw in the proof of (21.12.12) that this is not possible.

Remarks (21.12.14).

(i) The conclusion of (21.12.12) applies when is regular and integral and integral, since by virtue of the Auslander-Buchsbaum theorem (21.11.1), is then locally factorial. On the other hand, examples are known where and are normal algebraic schemes of dimension 3, over an algebraically closed field of arbitrary characteristic, and where the conclusion of (21.12.12) is no longer valid.

(ii) The set of (21.12.12) is also the set of points where is ramified. Indeed, if is unramified at a point , it is also unramified in an affine open neighbourhood of (17.3.7), hence is a separated, quasi-finite and birational morphism (17.4.3). Since is normal, one concludes from the Main theorem (III, 4.4.9) that is a local isomorphism at the point , hence . Conversely, is evidently unramified at every point where it is a local isomorphism. This justifies the title given to this section.

(iii) Without supposing that is locally factorial, but supposing on the other hand that is a complete intersection relative to (chap. V), we shall see in chap. V

that one has a more precise result than (21.12.12), by expressing as a 1-codimensional cycle of a section of an invertible -Module, canonically associated to . This will apply notably when and are both regular, or when is an -morphism, and being smooth preschemes over .

(iv) One may ask whether (21.12.10) admits a converse; in other words, for a given Noetherian local ring , if one sets , denoting the closed point of , and if, for every locally Noetherian prescheme and every morphism such that is an open immersion, every irreducible component of is of codimension in or is reduced to a point, is it then true that verifies condition (W) of (21.12.8)?

(v) Let be an integral regular prescheme, a locally Noetherian normal prescheme, a morphism locally of finite type; suppose moreover that, if is the generic point of , the fibre is étale over , and let be the complement of the largest open of such that is étale; is it then true that all irreducible components of are of codimension 1? This is indeed so when one supposes in addition that is locally quasi-finite (Zariski-Nagata "purity theorem"). One can show that the foregoing conjecture would be a consequence of the following one (and would even be equivalent to it if the conjecture of (iv) were verified): for a regular local ring contained in an integral and integrally closed local ring , such that is a finite -module, the open of points of at which is unramified over (or, what amounts to the same by (18.10.1), étale over ) is affine.

Proposition (21.12.15).

Let be a prescheme, a flat morphism locally of finite presentation, whose fibres are geometrically irreducible (4.5.2); a smooth morphism; one denotes the fibres of this morphism. Let be a proper -morphism such that, for every maximal point of , the morphism is an isomorphism. Then is an isomorphism.

Since is flat, every maximal point of is above a maximal point of (2.3.4), hence the union of the , when runs through the set of maximal points of , is dense in ; since the are isomorphisms, is dominant and hence surjective since it is proper. It therefore suffices to show that is an open immersion. Taking (17.9.5) into account, it suffices to prove that for every , is an open immersion. Since every is a generization of a maximal point , one can already, by making the base change , where is the reduced sub-prescheme of having as underlying space, reduce to the case where is a local and integral scheme of closed point : the fibres at the points and are indeed not changed and the properties of the morphisms , and are preserved by base change. Moreover, the question is local on ; replacing by an affine open neighbourhood in of a point of , and by , which is quasi-compact since is proper, one sees that one can suppose in addition that and are of finite presentation over . There then exists a Noetherian local sub-ring A_0 of such that is a local homomorphism, two morphisms of finite type , and an S_0-morphism

such that , and are deduced from , , by the base change ((8.9.1) and (5.13.3)); one can in addition suppose flat (11.2.7), smooth (17.7.9) and proper (8.10.5, (xii)). On the other hand, the projection on S_0 of the generic point of is the generic point of S_0 and it follows from (2.7.1, (viii)) that the morphism is an isomorphism; finally the closed point of is above the closed point of S_0, hence the fibre of is geometrically irreducible (4.5.6). One sees therefore that one can restrict to the case where is the spectrum of an integral Noetherian local ring, of closed point , weaken the hypothesis on the fibres by supposing only that and are geometrically irreducible, and prove that under these hypotheses is an open immersion. There is then a discrete valuation ring and a morphism which transforms the closed point of into and the generic point of into (II, 7.1.9); let , , be the morphisms deduced from , , by the base change , which again verify the hypotheses verified by , , and in the statement of (21.12.15); if one proves that is an open immersion, it will follow from (2.7.1, (x)) that the same will be true for . One can therefore finally restrict to the case where is the spectrum of a discrete valuation ring. Then, since is smooth, is regular (17.5.8), and since the question is local on , one can restrict to the case where is integral. Since is flat, the maximal points of are contained in (2.3.4) and since is an isomorphism, is irreducible and the local ring at its generic point is a field; moreover, by flatness (3.3.2), one sees that has no immersed prime cycles, hence is reduced (3.2.1) and consequently integral and is a birational and separated morphism. To prove that is an open immersion, one can therefore apply the criterion (21.12.13); to see that is quasi-finite at the points of , one remarks that the assertion is evident at those of these points which belong to ; the only point of belonging to is the generic point of , by virtue of (6.1.1). Now, it follows from (2.4.6) and (14.2.2) that the morphisms and are equidimensional; since and are isomorphic, the irreducible components of and of have all the same dimension. But by hypothesis is irreducible and one has seen that is surjective, hence so is , which entails that is also irreducible; if is the generic point of , one has therefore , and the relation then entails, by virtue of (5.6.6), that , in other words (and consequently also ) is indeed quasi-finite at the point , which completes the proof.

Corollary (21.12.16).

Let be a proper, flat morphism of finite presentation, a smooth, proper morphism of finite presentation, an -morphism. Suppose that the fibres of are geometrically irreducible. Let be the set of such that is an isomorphism. Then is open and closed in , and the morphism , restriction of , is an isomorphism.

The last assertion will result from the first and from (17.9.5). One already knows (9.6.1, (xi)) that is locally constructible in . By virtue of (1.10.1), it suffices to prove that is stable by specialization and by generization. To demonstrate these assertions one

can, by a base change , reduce to the case where is a local scheme of closed point and generic point , and one must show that, in order for to be an isomorphism, it is necessary and sufficient that be one. Now, the sufficiency of this condition results from (21.12.15). To show that it is necessary, one argues as in the proof of (21.12.15) (remarking, with the same notations, that the projection of the closed point of is the closed point of S_0) to reduce to the case where in addition is Noetherian; but then the conclusion results from (III, 4.6.7, (ii)).

Remarks (21.12.17).

(i) The conclusion of proposition (21.12.15) is no longer valid if one eliminates the hypothesis that the fibres are irreducible. Take indeed , where is a discrete valuation ring, , which is proper and smooth over (17.3.9). Denote again and the closed point and the generic point of ; let be a closed point of , for example a point rational over , and let be the -prescheme obtained by blowing up the point . Since the polynomial ring A[T] is regular (0, 17.3.7) and of dimension 2, one sees as in the proof of (15.1.1.6) that, if is the structure morphism, is isomorphic to , and on the other hand, the complement of in is isomorphic to , hence and Z_2, the closure in of the complement of Z_1, are the two irreducible components of . Moreover, is evidently proper and is flat, since is integral (II, 8.1.4) and for every affine open of , the homomorphism is injective (I, 1.2.7), hence is a torsion-free -module, and consequently flat since is a discrete valuation ring . It is clear that is an isomorphism, although is not.

(ii) The conclusion of (21.12.15) also fails if, in the hypotheses, one suppresses the hypothesis that is proper. It suffices, to see this, to define and as in (i), but to replace by the complement of Z_2 in the prescheme defined in (i), and by the restriction of ; the morphism , restriction of , is still flat, and this time is geometrically irreducible; is moreover isomorphic to the complement of a closed point in , hence is an affine scheme isomorphic to ; since its image in is reduced to the closed point , is not proper (III, 4.4.2) and is not an isomorphism, although is one.

(iii) It is possible that the statement of proposition (21.12.15) remains valid when one replaces the word "isomorphism" by "étale morphism" (cf. (21.12.14, (v))). The same will then still hold for (21.12.16).

21.13. Parafactorial couples. Parafactorial local rings

Definition (21.13.1).

Let be a ringed space, a closed part of , . One says that the couple is parafactorial if, for every open of , the restriction functor , from the category of -Modules invertible to that of -Modules invertible, is an equivalence of categories.

To say that the couple is parafactorial means therefore that, for every open of , the following conditions are verified:

1° the functor is fully faithful, in other words, for two invertible -Modules , , the restriction map

                       Hom_{𝒪_V}(ℒ, ℒ') → Hom_{𝒪_{U ∩ V}}(ℒ | (U ∩ V), ℒ' | (U ∩ V))

is bijective;

2° the functor is essentially surjective, in other words, for every

invertible -Module , there exists an invertible -Module such that is isomorphic to ; this is also expressed by saying that the canonical homomorphism (21.3.2.4)

is surjective.

Lemma (21.13.2).

Let be a morphism of ringed spaces; for every open of , one denotes the restriction of . The following conditions are equivalent:

a) For every open of , the functor from the category of -Modules locally free of finite rank into the category of -Modules locally free of finite rank is faithful (resp. fully faithful).

a') For every open of , the functor from the category of -Modules locally free of rank 1 into the category of -Modules locally free of rank 1 is faithful (resp. fully faithful).

b) For every open of , the canonical homomorphism is a monomorphism (resp. an isomorphism) for every -Module locally free of finite rank.

b') The canonical homomorphism is a monomorphism (resp. an isomorphism).

Suppose that the canonical homomorphism is bijective; then, for an -Module locally free of finite rank to be isomorphic to an -Module of the form , where is an -Module locally free of finite rank, it is necessary and sufficient that the two following conditions be fulfilled:

(i) is an -Module locally free;

(ii) The canonical homomorphism is an isomorphism.

When these two conditions are fulfilled, is isomorphic to .

To say that the functor is faithful (resp. fully faithful) means that for two -Modules , locally free of finite rank, the map is injective (resp. bijective); since this must hold on replacing by an open and , by , , and since , the condition means again that the canonical homomorphism of sheaves

  (21.13.2.1)             ℋom_{𝒪_V}(ℰ_1, ℰ_2) → (f_V)_*(f_V^*(ℋom_{𝒪_V}(ℰ_1, ℰ_2)))

is injective (resp. bijective). But since and are locally free of finite rank, , isomorphic to , is also locally free of finite rank; this already shows that b) entails a), and conversely b) is a particular case of a), taking into account the isomorphism . It is clear that a') is a particular case of a) and b') a particular case of b). Conversely, since b) is a local property, one can, to verify it, restrict to the case where , and this shows that b') entails b).

If the canonical homomorphism is bijective, and if conditions (i) and (ii) are fulfilled, it is clear that with , up to isomorphism.

Conversely, suppose that , with locally free; the question being local on , one can suppose that , whence , and conditions (i) and (ii) result from the hypothesis that is an isomorphism.

In this number, we shall apply the preceding lemma to a canonical injection , where is the ringed space induced by on an open of . With these notations, (21.13.2) translates into:

Corollary (21.13.3).

Let be a ringed space, a closed part of , , the canonical injection. The following conditions are equivalent:

a) For every open of , the restriction functor from the category of -Modules locally free of finite rank into the category of -Modules locally free of finite rank is faithful (resp. fully faithful).

a') For every open of , the restriction functor from the category of -Modules locally free of rank 1 into the category of -Modules locally free of rank 1 is faithful (resp. fully faithful).

b) For every open of , the canonical homomorphism is injective (resp. bijective) for every -Module locally free of finite rank.

b') The canonical homomorphism is injective (resp. bijective).

Suppose that the canonical homomorphism is bijective. Then, for an -Module locally free of finite rank to be of the form , where is an -Module locally free of finite rank, it is necessary and sufficient that be an -Module locally free, and in this case, one may take .

Lemma (21.13.4).

Let be a locally Noetherian prescheme, a closed part of , , the canonical injection. For the canonical homomorphism to be injective (resp. bijective), it is necessary and sufficient that (resp. ) (in other words, that (resp. ) for every ).

These assertions are particular cases of (5.10.2) and (5.10.5).

Proposition (21.13.5).

Let be a ringed space, a closed part of , , the canonical injection. For the couple to be parafactorial, it is necessary and sufficient that the canonical homomorphism be bijective, and that for every open of and every invertible -Module , be an invertible -Module (notation of (21.13.3)).

This is an immediate consequence of definition (21.13.1) and of (21.13.3).

Corollary (21.13.6).

Let be a ringed space, a closed part of .

(i) If the couple is parafactorial, the same is true of for every open of . Conversely, if is an open covering of such that each of the couples is parafactorial, then the couple is parafactorial.

(ii) If the couple is parafactorial, the same is true of (X, Y') for every closed part of .

(iii) Suppose that is a prescheme, and let be a prescheme, a faithfully flat and quasi-compact morphism; set . Suppose that the couple (X', Y') is parafactorial and that the open is retrocompact in . Then the couple is parafactorial.

(i) Since the fact that the canonical homomorphism is bijective is a local property on , everything reduces (by virtue of (21.13.5)) to seeing, denoting by the canonical injection , that one has the following property: if, for every and for every invertible -Module , is an invertible -Module, then is an invertible -Module. But the property of being an invertible -Module is local on , and is an open covering of ; since one has , this proves our assertion.

(ii) Set and let be the canonical injection. To say that the homomorphism is bijective amounts to saying that for every open of , the homomorphism is bijective; but the composite homomorphism

                       Γ(V, 𝒪_X) → Γ(V ∩ U', 𝒪_X) → Γ(V ∩ U, 𝒪_X)

is bijective by hypothesis (21.13.5), and on replacing by , one sees that is bijective, hence is bijective. Note next that if is the canonical injection and if is an invertible -Module, is an invertible -Module, hence is by hypothesis an invertible -Module. Since the couple is parafactorial by virtue of (i), one has , hence is an invertible -Module. It then suffices, to conclude, to replace in the preceding argument , and by , and , where is an open of .

(iii) Set and note that since one is dealing with preschemes, one can write ; let be the restriction of and let be the canonical injection, which one also writes . Let us first show that the canonical homomorphism is bijective; for this, note that by hypothesis the canonical homomorphism is bijective; but since the morphism is quasi-compact and separated by hypothesis, and the morphism flat, is canonically identified with by virtue of (2.3.1); since , one sees that the homomorphism is bijective; one concludes that is bijective since is faithfully flat (2.2.7). Consider next an invertible -Module , and let , which is an invertible -Module. By hypothesis is an invertible -Module, which is isomorphic to by (2.3.1) as above. But since is faithfully flat and quasi-compact, this entails that is an -Module locally free (2.5.2). To complete the proof, it suffices to replace by an open and by in the foregoing.

Definition (21.13.7).

One says that a local ring is parafactorial if, setting and denoting by the closed point of , the couple (X, {a}) is parafactorial.

Proposition (21.13.8).

The notations being those of (21.13.7), set . For to be parafactorial, it is necessary and sufficient that it verify the following conditions:

(i) The canonical homomorphism is bijective,

(ii) .

If moreover is Noetherian, condition (i) is equivalent to

(i bis) .

Indeed, the only open of containing is itself, and consequently every invertible -Module is isomorphic to , in other words ; the first assertion therefore results from definition (21.13.1) and from (21.13.3). The second assertion is a particular case of (21.13.4).

Examples (21.13.9).

(i) A parafactorial Noetherian local ring is necessarily of dimension by virtue of (21.13.8); in other words a Noetherian local ring of dimension is not parafactorial.

(ii) A factorial Noetherian local ring of dimension is parafactorial, as follows from (21.13.8) and (21.6.14).

(iii) If is a Noetherian local ring of dimension and parafactorial, it is not necessarily normal nor even reduced. Consider a regular local ring of dimension , and let (0, 18.2.3), isomorphic to ; one sees at once that one has prof(A) = prof(B) = dim(B) ⩾ 3. To show that is parafactorial, it suffices, with the notations of (21.13.8), to prove that . Let be the kernel of the augmentation , which is such that and which, as a -module, is isomorphic to . Since , and have the same underlying space; if , X_0 is the sub-scheme defined by , we shall denote U_0 the sub-prescheme induced by X_0 on the open . For every , set ; since , is invertible in and is the kernel of the canonical surjective homomorphism of multiplicative groups ; in other words, one has an exact sequence of commutative groups

                       0 → 𝒥 →^φ A^× → B^× → 1

(the last three groups being multiplicative, the first two additive). For the same reason, for every , one has the exact sequence

denoting by the image of in , since ; in other words, one has an exact sequence of sheaves of commutative groups on the topological space

whence by restriction to the open , an exact sequence

                       0 → ℐ | U → 𝒪_U^× → 𝒪_{U_0}^× → 1.

By the cohomology exact sequence, one deduces an exact sequence

  (21.13.9.1)             H^1(U, ℐ) → H^1(U, 𝒪_U^×) → H^1(U, 𝒪_{U_0}^×).

But since is a -module isomorphic to , one has , and it follows from chap. III, 3rd part (see also [41, III, Example III-1]) that one has

by reason of the relation . Moreover, since is factorial, one has ; one therefore draws indeed from the preceding exact sequence that .

(iv) There also exist Noetherian local rings of dimension 3 which are integral, integrally closed and parafactorial, but not factorial. Let indeed be a Noetherian local ring, complete, integral and integrally closed, of dimension and non-factorial (for example the completion of the localized ring of the integral algebra at the maximal ideal image of ). Then it will follow from what we shall see below (21.14.2) that the local ring of formal series in over is parafactorial, but it is not factorial, otherwise would be by virtue of (21.13.12) below.

(v) One can show that an absolute complete intersection local ring (19.3.1) of dimension is parafactorial (cf. [41, XI, 3.13 (i))]).

(vi) One has seen (Remark (ii)) that every factorial Noetherian local ring of dimension 2 is parafactorial. But there are Noetherian local rings of dimension 2 which are parafactorial and non-factorial. One can show that, for a Noetherian local ring of dimension 2 to be parafactorial and non-factorial, it is necessary and sufficient that it satisfy the three following conditions:

is a Cohen-Macaulay ring (in other words ).

is integral and if is its integral closure, is factorial and is a finite -algebra.

3° Let be the conductor of in (annihilator of the -module , or also the largest ideal of contained in ); set , ; then (which implies , in other words is not integrally closed), and the canonical map (21.4.5) is surjective.

One can show moreover that these conditions entail the following property:

4° The ring (and a fortiori ) is reduced, and the morphism is bijective.

If one sets , , , , so that (resp. ) is defined by the ideal of (resp. ), the structural morphism is an isomorphism of onto (Bourbaki, Alg. comm., chap. V, §1, n° 5, cor. 5 of prop. 16). One sees therefore by virtue of 4° that is a bijective morphism; in other words, is a unibranch prescheme (6.15.1) (and in particular is a Noetherian local ring); in general is not geometrically unibranch. The space , of dimension 1, is constituted by the closed point of and the maximal points () of , and the unique point of above is also a maximal point of ; since and are reduced, one deduces that one has for and ; this relation is also evidently verified for and the unique point of , hence for every and the unique point of . In particular, if the ring is of characteristic 0 (0, 21.1.1) one sees that is unramified over (but not étale in general).

We shall restrict ourselves here to demonstrating that conditions 1°, 2°, 3° above are sufficient for to be

parafactorial. Now, by virtue of condition 1° and of (21.13.8), it suffices to show that .

Since is factorial by virtue of 2°, one has , and one deduces from (21.8.5, (ii)) an exact sequence

                       1 → (⨁_{s ∈ S} (A'_{p'_s})^× / (A_{p_s})^×) / Im(Γ(U', 𝒪_{U'}^×)) → Pic(U) → Pic(U') → 0

and one must therefore show that the second term of this sequence is reduced to the neutral element, taking for the set of the () and noting that here is the -Algebra Ã'. Let , be the prime ideals of and corresponding to the points , , and remark that ; let us give for each an invertible element of with , ; since we have only to examine the quotient groups , one can suppose for all ; let be the canonical image of , so that is a non-zero element of . The set formed by the is an affine open of the form in , with , the maximal ideal of ; there exists therefore an invertible element whose are the canonical images, and since is regular in the integral ring , one can suppose of the form , where is invertible. Let a'' be an element of in the class b'', which is therefore necessarily invertible; is invertible in and for every , one has , whence ; but , hence is an invertible element of , and the classes of and in are the same, which completes the proof.

To have an explicit example of a parafactorial ring of dimension 2 obtained in this manner and non-factorial, consider the ring , whose integral closure identifies with (6.15.11). If one sets , the integral closure of is the ring (Bourbaki, Alg. comm., chap. V, §1, n° 4, prop. 14); one verifies at once that the conductor of in is the maximal ideal of , hence the conductor of in is , and one has , . It is then immediate that conditions 1°, 2° and 3° stated above are indeed verified, but is not even integrally closed.

One can vary this example, and the reader will see without difficulty that if is an algebraically closed field, the ring , localization of the ring at the maximal ideal generated by the images of , and , verifies also conditions 1°, 2° and 3° above.

It could be that these three conditions imply even that each of the rings is a discrete valuation ring, which would entail that is even a regular ring.

Proposition (21.13.10).

Let be a locally Noetherian prescheme, a closed part of . For the couple to be parafactorial it is necessary and sufficient that, for every , the local ring be parafactorial.

Set and let be the canonical injection.

By virtue of (21.13.4), to say that the canonical homomorphism is injective is equivalent to saying that each of the local rings for satisfies condition (i bis) of (21.13.8).

Let us first show that if the couple is parafactorial, each of the rings () is parafactorial. Taking the preceding remark into account, everything reduces to seeing that, if one sets and , every invertible -Module is isomorphic to . Now, when runs through the set of affine open neighbourhoods of in , it follows from (8.2.13) and (I, 2.4.2) that the prescheme is projective limit of the preschemes induced by on the opens which are quasi-compact since is locally Noetherian. Since the are separated, it follows from (8.5.2, (ii)) and (8.5.5) that there is an affine open neighbourhood of in and an invertible -Module such that is the sheaf induced by on . But the hypothesis entails, by virtue of (21.13.6, (i) and (ii)), that the couple is parafactorial; one concludes by definition (21.13.1) that there exists a -Module invertible inducing on U_V. Replacing if needed by a smaller open neighbourhood of , one can suppose that is isomorphic to , whence one concludes that is isomorphic to .

Conversely, let us prove that if all the () are parafactorial the couple is parafactorial. One is evidently reduced, taking the remark at the beginning into account, to showing that for every invertible -Module , is an invertible -Module (21.13.5). The question being local on , one can suppose Noetherian. The set of such that the restriction of to an open neighbourhood of in is invertible is evidently open in ; one must show that . For this, suppose the contrary and set . Let be a maximal point of ; since by definition, is by hypothesis a parafactorial ring. Replacing if needed by an open neighbourhood of in , one can suppose that the restriction of

to is invertible; hence, with the notations introduced above, the -Module induced by on is invertible. Since is parafactorial, is induced by an invertible -Module ; but since is the projective limit of the open neighbourhoods of in , one can again apply (8.5.2, (ii)) and (8.5.5), proving the existence of such a neighbourhood and of an invertible -Module inducing on , hence inducing on ; applying this time (8.5.2.5) and (8.5.2, (i)), one deduces that on replacing if needed by a smaller open neighbourhood of , one can suppose that the restrictions of and of to are equal. If one then sets and if one denotes by the invertible -Module equal to in , to in , one has and . One then concludes from the fact that the homomorphism is bijective and from (21.13.2, b)) that is isomorphic to , hence invertible. But this contradicts the definition of , and therefore concludes the proof of (21.13.10).

Corollary (21.13.11).

Let be a locally Noetherian prescheme, a closed part of , . Suppose that the couple is parafactorial and that is locally factorial; in other words (21.13.10), for every , the ring is: 1° parafactorial if ; 2° factorial if . Then is locally factorial (in other words, is in fact factorial for every ).

Suppose indeed the contrary, and let be a point of such that is not factorial and has minimal dimension among all points of having this property. Then, if one sets , , the hypothesis that is parafactorial entails and (21.13.8); moreover the choice of entails that is locally factorial. But then (21.6.14) proves that is factorial, contrary to hypothesis.

Proposition (21.13.12).

Let , be two Noetherian local rings, a local homomorphism making a flat -module. Then, if is factorial, so is .

One already knows that is integral and integrally closed (6.5.4), hence one can restrict to the case where , and reason by recurrence on . Let , the closed point of , , the structural morphism, , .

The local rings at points are factorial and of dimension (6.1.2), hence parafactorial (21.13.9, (ii)); one concludes (21.13.10) that the couple is parafactorial. By virtue of (21.13.6, (iii)), this shows that the ring is parafactorial, hence and (21.13.8). Finally, the local rings of being of dimension , the recurrence hypothesis proves that is locally factorial; the conclusion then results from (21.6.14).

(21.13.12.1). Note that the statement (21.13.12) where one replaces "factorial" by "parafactorial" is no longer exact, as shows the example where is a field and the factorial local ring . However, it follows from (21.13.6, (iii)) that, under the conditions of (21.13.12), if denotes the maximal ideal of and if is an ideal

of definition of (in other words, if ), then, when is parafactorial, so is . In particular, if the completion  of a Noetherian local ring is parafactorial, so is .

Remark (21.13.13).

A part of the foregoing definitions and results is attached to more general considerations on the cohomology of sheaves of commutative groups on a topological space, developed in chap. III, 3rd part, and we have given here an independent exposition only for the convenience of the reader. Indeed, if is a ringed space, one has a canonical equivalence between the category of invertible -Modules and that of principal homogeneous sheaves under the sheaf of groups (16.5.15). This equivalence is defined by associating to every invertible -Module the sheaf of germs of isomorphisms of onto ; it is immediate that is canonically equipped with a structure of principal homogeneous sheaf under . The verification of the fact that the functor is an equivalence of categories is immediate. That being the case, consider in a general way a topological space and a sheaf of groups on ; let be a closed part of , set , and let be an integer such that ; one will say that the couple is -pure for if, for every open of , the restriction functor , from the category of principal homogeneous sheaves under the sheaf of groups , into the analogous category under the sheaf of groups , is faithful (), resp. fully faithful (), resp. an equivalence of categories (). In the case where is a ringed space and , to say that is -pure means therefore for , that the homomorphism is injective; for , that this homomorphism is bijective; and finally, for , that the couple is parafactorial (21.13.3).

Returning to the general case, recall that the morphisms in the category of principal sheaves under are by definition the isomorphisms. To say that the couple is 0-pure means therefore that for every open of , the canonical homomorphism is injective; to say that the couple is 1-pure means that the canonical homomorphism is bijective (and then the canonical homomorphism is injective); finally, one can show that for the couple to be 2-pure, it is necessary and sufficient that the homomorphisms be bijective for and . When is a sheaf of commutative groups, introducing the cohomology sheaves defined in chap. III, 3rd part, to say that is -pure for means that for ; in this form, the notion generalizes immediately for every integer .

Proposition (21.13.14).

Let be a locally Noetherian normal prescheme, a filtering decreasing family of opens of . The following conditions are equivalent:

a) Every 1-codimensional cycle on such that there exists an index for which is locally principal (21.6.7), is locally principal.

b) For every such that and such that does not belong to , the ring is parafactorial.

b') For every closed part of such that and such that there exists for which , the couple is parafactorial.

Property a) can again be expressed in the following manner: if the closed set of points where is not principal is contained in one of the , then is locally principal. Since is normal, the condition entails that is a field or a discrete valuation ring, hence is principal; in other words, one has necessarily (5.1.3). Property a) is therefore equivalent to the following:

a') If there exists a closed set contained in one of the , such that and such that is locally principal, then is locally principal.

Note on the other hand that b) implies b') by virtue of (21.13.10); conversely, if b') is verified and if , then is contained in and one has , hence it follows again from (21.13.10) that is parafactorial, which proves the equivalence of b) and b'). One is thus reduced to proving the equivalence of a') and b'). Note that since is normal, one has, for every closed part of such that , (5.8.6); if one sets , the homomorphism is therefore bijective (21.13.4); since the conditions a') and b') are local, one sees that it suffices to show the equivalence of the following conditions, when is Noetherian and normal and is closed in and such that :

a'') For every 1-codimensional cycle on , the hypothesis that (where ) is locally principal entails that itself is locally principal.

b'') The canonical homomorphism is surjective.

Let us first prove that a'') entails b''); an element has for image in (21.6.11) the class of a 1-codimensional cycle Z_0 on , which is locally principal. The hypothesis entails that the restriction homomorphism is bijective, hence , where is a 1-codimensional cycle on ; since Z_0 is locally principal, so is by virtue of a''); it follows from (21.6.11) that the image of in is the image of a unique element , and it is clear then that is the image of .

Conversely, let us prove that b'') entails a''). Let then be a 1-codimensional cycle on such that is locally principal; the image of in is therefore the image of a unique element (21.6.11). By hypothesis there exists whose image in is ; the image of in is therefore the class of a 1-codimensional cycle on such that and are linearly equivalent. But since is schematically dense in , the image of by the isomorphism is , hence and are linearly equivalent, and since is locally principal, so is .

Corollary (21.13.15).

Let be a locally Noetherian and normal prescheme, a part of . The following conditions are equivalent:

a) Every 1-codimensional cycle on which is principal at the points of is locally principal.

b) For every which is not a generization of a point of (in other words, such that ) and which is such that , the ring is parafactorial.

It is clear that the set of points of generizations of points of is the intersection of the open neighbourhoods of . One can restrict to the case where is Noetherian (properties a) and b) being local on ); since every 1-codimensional cycle on which is principal at the points of is also so at the points of an open neighbourhood of , one sees that condition a) of (21.13.15) is equivalent to condition a) of (21.13.14) applied to the family of open neighbourhoods of . The corollary then results from the equivalence of a) and b) in (21.13.14).

Remark (21.13.16).

Under the general conditions of (21.13.14), suppose moreover that one has . Then condition a) of (21.13.14) is also equivalent to the following:

c) Every 1-codimensional cycle on whose support is contained in one of the sets is locally principal.

One can restrict to the case where is irreducible and all are non-empty. It is clear that a) entails c), for if the 1-codimensional cycle on has its support in , , hence is locally principal. Conversely c) entails a): let indeed be a 1-codimensional cycle on such that is locally principal; since is normal, , where is a divisor on (21.6.10, (i)), and the hypothesis implies (by virtue of (21.3.4)) that there is a set such that is equivalent to 0. If , where is a regular rational function on , can be considered as a regular rational function on . If , one sees therefore that has its support in ; by virtue of the hypothesis, Z'' is locally principal, whence the conclusion.

The condition will in particular be fulfilled if is the family of open neighbourhoods of a point , for every invertible -Module , there exists by definition a such that is isomorphic to . One sees therefore that, in the statement of (21.13.15), if , conditions a) and b) are also equivalent to the following:

c) Every 1-codimensional cycle on whose support does not contain is locally principal.

If moreover , one will conclude that this condition entails that every 1-codimensional cycle whose support does not contain is principal.

21.14. The Ramanujam-Samuel theorem

Theorem (21.14.1) (Ramanujam-Samuel).

Let be a Noetherian local ring of maximal ideal , such that its completion  is integral and integrally closed. Let be a Noetherian local ring such that , a local homomorphism making a formally smooth -algebra (for the preadic topologies (0, 19.3.1)) and such that the residue field of is finite over that of . Then every 1-codimensional cycle on which is principal at the point is a principal 1-codimensional cycle.

If is the residue field of , is a formally smooth -algebra (for

its preadic topology) (0, 19.3.5), hence regular, and in particular integral; in other words is indeed a prime ideal of , which justifies the statement. Everything evidently reduces to proving that every prime ideal of not contained in is principal.

Let Â, be the completions of and respectively, so that the maximal ideal of  is ; one knows (0, 19.3.6) that is a formally smooth Â-algebra for the adic topologies. Let be the residue field of , a finite extension of ; there exists a local homomorphism , where is a Noetherian local ring which is a finite and flat (hence free) Â-module and is such that is isomorphic to ; one deduces that is complete, and it then follows from (7.5.1), (7.5.3) and (6.5.4, (ii)) that is integral and integrally closed. Moreover, is a complete semi-local ring, direct composite of complete local rings one of which, D_0, has residue field (since is direct composite of local rings one of which is isomorphic to ). Since is formally smooth over , the same is true of D_0; consequently is a formally smooth -algebra, of residue field , which entails that it is -isomorphic to a formal series algebra (0, 19.6.4); one concludes, by (0, 19.7.1.5), that D_0 is -isomorphic to , and consequently integral and integrally closed (Bourbaki, Alg. comm., chap. V, §1, n° 4, prop. 14). Since the morphisms and are faithfully flat, one deduces that and are also integral and integrally closed (2.1.13). This proves that the ideal is divisorial (Bourbaki, Alg. comm., chap. VII, §1, n° 10, prop. 15) and not contained in , otherwise one would have by faithful flatness . One concludes that none of the prime ideals of height 1 in D_0 which contain can be contained in ; if one proves that these ideals are principal, it will follow that is principal, the divisors (in Bourbaki's sense) of and of a product of powers of the being equal (Bourbaki, Alg. comm., chap. VII, §1, n° 4, prop. 5). Since , one deduces by faithful flatness that is principal (2.5.2), using the fact that is a local ring.

One is thus reduced to proving the theorem in the particular case where is complete and . Let us first show that one can reduce to the case where . Indeed, proceed by recurrence on , and let be an element not belonging to ; it will suffice to show that there exists an -automorphism of such that does not belong to the ideal : indeed, if , is complete and has maximal ideal . Since one has and , it will follow from the recurrence hypothesis (taking into account that is integral and integrally closed) that the ideal is principal in , hence so is . Now, if is the image of ("reduced series" of ), one knows (Bourbaki, Alg. comm., chap. VII, §3, n° 7, lemma 3) that there is an -automorphism of such that , which evidently implies .

Suppose then and write in place of T_1, so that . It suffices to show that in the polynomial ring A[T], the ideal is principal;

in fact, let be an element of not belonging to ; it is a formal series whose reduced series is not zero; hence, by virtue of the preparation theorem (Bourbaki, Alg. comm., chap. VII, §3, n° 8, prop. 5), for every , there exist and a polynomial such that , and one has therefore ; on the other hand (loc. cit., prop. 6) there exist a non-constant distinguished polynomial and an invertible element such that , hence one also has , which proves that is generated by . Since is flat over A[T] , it follows from Bourbaki, Alg. comm., chap. VII, §1, n° 10, prop. 15, that is a prime ideal of height 1 in A[T]. Moreover, one has necessarily ; otherwise, would necessarily be of height > 1, and it would follow from (5.5.3) that one would have . But then, since , one would have contrary to the hypothesis on . If is the field of fractions of , is therefore a prime ideal distinct from 0 and from K[T] in K[T], hence of the form , where with and the , being irreducible in K[T]. But one has seen above that there exists in a non-constant distinguished polynomial . If is the class of in , is therefore a root of the polynomial F_0 in an extension of and consequently divides F_0 in K[T]; but since and F_0 are monic, this entails that the coefficients of are integral over (Bourbaki, Alg. comm., chap. V, §1, n° 3, prop. 11), hence belong to since is integrally closed. In other words, one has (Bourbaki, Alg. comm., chap. II, §1, n° 5, prop. 11); since every polynomial is divisible by in K[T] and is monic, the coefficients of belong to , hence . Q.E.D.

The statement (21.14.1) is equivalent to the following:

Corollary (21.14.2).

Under the hypotheses of (21.14.1) on , and , for every prime ideal of not contained in and such that , the ring is parafactorial. In particular, if (i.e. ((0, 19.7.1) and (6.1.1))), the ring is parafactorial.

Indeed, one has seen in the proof of (21.14.1) that the conditions of the statement entail that is integral and integrally closed; the equivalence of the statements (21.14.1) and (21.14.2) then results from (21.13.15) applied to and .

Proposition (21.14.3).

Let be a normal prescheme, a smooth morphism.

(i) If is locally Noetherian (hence also ), then, for every such that and such that is not a maximal point of its fibre , the ring is parafactorial. Every 1-codimensional cycle on such that contains no irreducible component of a fibre , is locally principal.

(ii) Let be a closed part of containing no irreducible component of a fibre , and such that for every maximal point of , is of codimension in . Then the couple is parafactorial.

Note that the conditions of (ii) are fulfilled if is a closed part of such that, for every , is of codimension in .

To prove (21.14.3), note first that under the hypotheses of (i), if one

sets , there exists an open neighbourhood of in such that and verify the conditions of (ii): indeed, it follows from the hypothesis and from (9.5.3) that one can take such that contains no irreducible component of an ; on the other hand, for every maximal point of such that , the points of are specializations of , hence at such a point one has and since ( being reduced to the point by virtue of (2.1.13)), is indeed of codimension in . Since the couple is then parafactorial by virtue of (ii), one concludes from (21.13.10) that the ring is parafactorial, that is to say the first assertion of (i). To prove the second, one can restrict to the case where , where ; since is an integrally closed local ring of dimension 1, it is a discrete valuation ring and is therefore principal at the point ; for every other point of , one therefore has and is not a maximal point of its fibre by hypothesis, hence is parafactorial. Apply then (21.13.15) taking for the set formed by and the maximal points of the fibres of ; it is clear that is principal at the points of since is the only point of belonging to ; since on the other hand condition b) of (21.13.15) is evidently fulfilled, one concludes that is locally principal.

One is therefore reduced to proving (ii). Set and let be the canonical injection.

Since the hypotheses and the conclusion are local on and on by virtue of (21.13.6, (i)), one can restrict to the case where and are affine, being therefore of finite presentation. Since the fibres of are regular, the hypothesis on entails that, for every

point , one has ; it follows from (19.9.8) that the canonical homomorphism is injective. Moreover, replacing by a larger closed part following the method of the proof of (19.9.8), and using (21.13.6, (ii)), one can suppose that is defined by an ideal of finite type of the ring of , hence the open is quasi-compact and quasi-separated and the closed set is constructible.

To prove (ii), it suffices, for every invertible -Module and every point given, to establish the existence of an open neighbourhood of in such that: 1° the canonical homomorphism is surjective; 2° there exists an invertible -Module such that ((21.13.5) and (21.13.3)).

Set ; we shall see that one can restrict to the case where . Let be a fundamental system of affine open neighbourhoods of in , and set , , , being the canonical injection; is then projective limit of the , (8.1.2, a)); suppose the proposition true for the morphism and for , and set , being the canonical injection. The projection being an affine morphism, one has (II, 1.5.2) and the canonical homomorphism is none other than , where is the canonical homomorphism; since by hypothesis is surjective, the same is true of for large enough (8.5.7). Let on the other hand be the invertible -Module restriction of , and note that the are affine, hence quasi-compact and quasi-separated; since by hypothesis there exists an invertible -Module such that , there exists a large enough and a quasi-coherent -Module of finite presentation such that is isomorphic to (8.5.2, (ii)); moreover (8.5.5) one can suppose that is invertible. Finally, since the are quasi-compact and quasi-separated and , there exists large enough such that and are isomorphic (8.5.2.5).

One is thus reduced to the case where , , where is a local ring; since is normal and smooth, is normal (17.5.7), hence is integral and integrally closed. Consider then as inductive limit of its sub--algebras of finite type; since the integral closure of such a sub--algebra is a subring of and is also a -algebra of finite type (7.8.3, (ii), (iii) and (vi)), is filtered inductive limit of the sub--algebras of finite type of which are integrally closed rings. By virtue of (1.8.4.2), there exists an index and an -algebra of finite type such that up to an -isomorphism. Set , ; if is the canonical projection, one can moreover suppose (since is constructible) that , where is a closed part of (8.3.11). Let be the morphism ; by virtue of the transitivity of fibres and of (4.2.6), contains no irreducible component of the fibres for any . Since is smooth, one can also suppose that is smooth (17.7.8); finally, the image of the generic point of in is the generic point of , and one has by virtue of (6.1.4). One sees therefore that , and verify all the hypotheses of (ii), and on setting for , the same is true for , and . Let us show that if one proves that the couple is parafactorial for every , the same is true for . Indeed, let , , , the canonical injections; the projection being an affine morphism, one has (II, 1.5.2), and consequently the canonical homomorphism is none other than , where is the canonical homomorphism; the latter being bijective by hypothesis, the same is true of . On the other hand, for every invertible -Module , there exists and an invertible -Module such that (denoting by the restriction of ) (8.5.2 and 8.5.5), being quasi-compact since is Noetherian. By hypothesis, there exists an invertible -Module such that is isomorphic to ; is then an invertible -Module such that is isomorphic to , which proves our assertion.

One is thus reduced to proving (ii) when the ring is a -algebra of finite type integrally closed; since the local rings of are then excellent integrally closed rings, their completions are also integrally closed (7.8.3, (ii), (iii) and (vii)). To prove that the couple is parafactorial, it suffices to show that for every , the ring is parafactorial (21.13.10). Let be a closed point of , which is a specialization of (5.1.11); if one sets , has a

integrally closed completion, and is a formally smooth -algebra for the preadic topologies (17.5.3); since contains no irreducible component of , one has (17.5.8). If , one has ; if on the contrary , one has by hypothesis ; hence in all cases . Moreover, the prime ideal of corresponding to the point is not contained in since contains no irreducible component of . Finally, since is closed in , is a finite extension of (I, 6.4.2). In all cases, one can apply to , and the result of (21.14.2), which completes the proof.

Remarks (21.14.4).

(i) It may be that the statements (21.14.1) and (21.14.2) remain valid without hypothesis on the residue field of . Stated with this generality, the result would be equivalent, by virtue of (21.13.15) and (21.13.12.1), to the following: Let be a Noetherian local ring complete, integral and integrally closed, a Noetherian local ring which is a formally smooth -algebra, such that ; then is parafactorial.

(ii) We shall see in chap. VI, by using a "finite descent" technique applied to the morphism , where is the normalization of , that the conclusion of (21.14.1) (or of (21.14.2)) remains valid on replacing the hypothesis that  is integrally closed by the hypothesis that  is reduced provided that . If one replaces this last condition by , one can even suppress the hypothesis that  is reduced. Similarly, the conclusion of (21.14.3) remains valid on replacing the hypothesis that is normal by the hypothesis that is reduced, provided that .

(iii) In chap. III, 3rd part, one proves that if is a smooth morphism, a locally constructible closed part of such that, for every , one has (with the notations of (21.14.3)) , then the couple is parafactorial ([41], XII, 4.8). This conclusion is no longer valid when one supposes only that for every and when one no longer supposes reduced. For example, let be a field, , algebra of dual numbers over , , ( indeterminates), so that , being the "zero section" of this bundle; if is the unique closed point of , one has and is the "zero section" of . To see that the couple is not parafactorial, it suffices to show that the ring , where is the ideal which defines , is not parafactorial (21.13.10). Let , and denote and U_0 the complements of the closed point in and ; arguing as in (21.13.9), one has the exact sequence, extension of (21.13.9.1)

                       Γ(U, 𝒪_U^×) → Γ(U_0, 𝒪_{U_0}^×) → H^1(U_0, 𝒪_{U_0}) → Pic(U) → Pic(U_0).

Now, one has , since (5.10.5). Moreover since B_0 is factorial, and [41, 3, Example III-1]; since the homomorphism is surjective, one concludes that , hence that is not parafactorial.

(iv) The result (21.14.3) was first proved by C. Seshadri [44] in the particular case where is a normal algebraic scheme over an algebraically closed field and , where is a -prescheme algebraic and smooth over . Seshadri's proof [44, p. 188-189] is global in nature and uses the theory of Picard schemes. It gives moreover (loc. cit.) results such as the following (for which one does not at present possess a proof by local means). Let , be two preschemes locally of finite type over a field , , a 1-codimensional cycle on (considered as -prescheme); suppose the following conditions are verified:

and are geometrically normal over (6.7.6);

2° For every maximal point of , the 1-codimensional cycle on the fibre , having the same multiplicity as at every point of , is locally principal (in other words, is the image of a divisor of , since is normal);

3° For every , is principal at the maximal points of the fibre .

Then is locally principal. In other words, being normal (6.14.1), for every which is not maximal in its fibre and which belongs to none of the "generic fibres" (which implies by virtue of (6.1.1)), the local ring is parafactorial, by virtue of (21.12.15).2

21.15. Relative divisors

(21.15.1). Let be a prescheme, a flat morphism locally of finite presentation. One has defined in (20.6.1) the sheaf of rings of germs of meromorphic functions on relative to , a subsheaf of ; it is clear that the canonical injection (20.1.4.1) sends onto a subsheaf of , with which one identifies it. Let be the sheaf (in multiplicative groups) of germs of invertible sections of ; it is therefore a subsheaf of and contains as a subsheaf.

Definition (21.15.2).

One calls sheaf of divisors on relative to , or sheaf of divisors on transversal to , and one denotes the quotient sheaf (of commutative groups) ; the sections of this sheaf over are called divisors on relative to , or divisors on transversal to ; they form a commutative group denoted .

It is clear that identifies canonically with a subsheaf of , and consequently with a subgroup of , which one again denotes additively.

For every open of , one has , hence , and the sheaf is therefore equal to the presheaf .

Since is a subsheaf of , the definitions, notations and formulas relating to the divisors of sections of over (21.1.3) apply without change to the sections of over .

(21.15.3). The structure of sheaf of ordered groups on (21.1.6) induces on the subsheaf a structure of sheaf of ordered groups, for which the sheaf of monoids of germs of positive sections is , which one denotes . One has ; one denotes this submonoid of by , and it is formed of elements for a structure of ordered group on . It follows from (21.1.5.1) that is the image in of the subsheaf of monoids

For every open of , is the set of sections of over such that be regular and that belong to , which means, with the notations of (20.6.1), that , so that the sheaf is none other than the sheaf denoted in (20.6.1). One can therefore write

up to a canonical isomorphism.

(21.15.3.3). Let and consider the closed sub-prescheme of defined by the Ideal of (21.2.12); by virtue of what precedes, for every , there is an open neighbourhood of in and a section such that ; since , the image of in belongs to the maximal ideal of , and moreover, by definition, for every , the image of in is a regular section. One deduces therefore from (11.3.8) and (19.2.4) that the canonical immersion is transversally regular relative to and of codimension 1 at the point . The converse being immediate, one sees that one can identify canonically the positive divisors on relative to with the closed sub-preschemes of such that the canonical injection be a transversally regular immersion relative to and of codimension 1. We shall ordinarily make this identification.

Proposition (21.15.4).

Let be a divisor on , the corresponding invertible fractional Ideal (21.2.5). For , it is necessary and sufficient that, for every , one have (or what amounts to the same, that one have ).

Indeed, to say that means that for every , there exists an open neighbourhood of and a section such that ; since is the invertible fractional Ideal , the proposition follows at once.

Proposition (21.15.5).

Let be a divisor on , the invertible fractional Ideal and the regular meromorphic section of defined canonically by (21.2.8 and 21.2.9). For , it is necessary and sufficient that .

Indeed, if is an open of such that , where , to say that means, by virtue of definitions (20.6.2), that , whence the proposition.

The interpretation of divisors on by means of the classes (21.2.11) therefore permits the interpretation of the elements of as the couples (up to isomorphism) such that be an invertible -Module and that be a meromorphic section of over , regular relative to (20.6.5, (iii)).

Proposition (21.15.6).

Let be a divisor on relative to , and suppose that for every such that , one has (resp. ). Then one has (resp. ).

As in (21.1.8), one can restrict to the case where , where is a regular meromorphic function relative to , and everything reduces to seeing that if at every point such that , is everywhere defined in . But this hypothesis means that, if , one has for every , and it suffices, to conclude, to apply (20.6.6).

(21.15.7). Let be a second -prescheme flat and locally of finite presentation over , and let be an -morphism. If the -morphism is flat, one knows (21.4.5) that the inverse image by of every divisor on is defined; if moreover , it follows from definition (21.15.2) and from (20.6.8) that one has then .

(21.15.8). Let be a second -prescheme flat and locally of finite presentation over , and let be a finite and flat -morphism. Note that is then necessarily of finite presentation (1.4.3, 1.4.6 and 1.6.3), hence is a flat and finite-presentation -Module, and consequently locally free (2.1.12); in other words is a locally free morphism (18.2.7); for every , the corresponding morphism is therefore also finite and locally free. One deduces then from (21.5.2) and (20.6.1) that for every open of and every section , the norm belongs to ; the reasoning of (21.5.3) then proves that for every invertible -Module and every meromorphic section of over , regular relative to , the norm is a meromorphic section of over , regular relative to . The interpretation of divisors relative to given in (21.15.5) and the definition of the direct image of a divisor (21.5.5) then prove that for every divisor , one has .

(21.15.9). Consider finally any morphism , and (under the hypotheses of (21.15.1)) set , which is flat and locally of finite presentation over ; if is the canonical projection, one has seen (20.6.9) that one has a canonical homomorphism , which evidently transforms every section of over an open , regular relative to , into a section of over , regular relative to (20.6.5, (iii)); one then concludes from the definition

(21.15.2) and from the right-exactness of the functor , that the foregoing homomorphism defines by passage to quotients a canonical homomorphism

which evidently transforms the elements of into elements of . One sets again (resp. ), and one sees at once that one has thus defined two contravariant functors

                       𝒟iv_{X/S} : Sch_{/S}^∘ → Ab,        𝒟iv_{X/S}^+ : Sch_{/S}^∘ → Set

from the category of -preschemes into that of commutative groups (resp. of sets). One will see later (chap. VI) important cases where the functor is representable .

For every divisor , the image of by the homomorphism (21.15.9.1) is none other than the inverse image (in the sense of (21.4.2)): the existence of this inverse image and the preceding assertion are indeed immediate consequences of (20.6.5, (iii)) and (20.6.9).

The elements of are often called, by abuse of language, "families of divisors on relative to , parametrized by the -prescheme "; this terminology is used especially when one is dealing with positive divisors.

1

The reader will verify that (21.9.12) is not used to prove this property in chap. V.

2

In fact, in the article cited above, Seshadri supposes that is algebraically closed, separated and "semi-complete" (i.e. such that is -isomorphic to ) and replaces hypothesis 3° by the stronger hypothesis that contains none of the fibres for . But since the statement is local on , one concludes at once that it suffices to make hypothesis 3°, and this proves that the conclusion (interpreted as above in terms of the parafactoriality property of the rings ) is local on and on , which allows one to eliminate completely the hypothesis that is "semi-complete" and that is algebraically closed, since (by passing first to the algebraic closure of ) one can suppose first affine, which allows one to embed it as an open of a projective normal scheme over , to which Seshadri's result applies. Note also that, thanks to this reduction, it suffices to do Seshadri's proof in the case where is projective (and not only "semi-complete"), a case in which the Picard scheme theory used by Seshadri is contained in the theory which will be developed in chap. VI of our Treatise.