Exposé XI. Application to the Picard group
This Exposé is modeled on the preceding one, but this time the result of no. 1 is weaker.
Throughout this Exposé, will denote a locally noetherian prescheme, a quasi-coherent ideal of O_X (so that
is a closed part of ), a variable open neighborhood of in , and the formal completion
of along . For every ringed space , we denote by the category of invertible O_Z-Modules — in
other words, locally free of rank 1 — and by the group of isomorphism classes of invertible
Modules on .
1. Comparison of Pic(X̂) and Pic(Y)
For every , set and . The sequence of sheaves of abelian groups on
0 ⟶ P_n ──u──→ O*_{X_{n+1}} ──v──→ O*_{X_n} ⟶ 1
is exact. Let us be precise: the group structure on is the additive structure, for every , and is the homomorphism deduced from the injection . We see that is surjective by remarking that, for every , is a local ring, the quotient of by a nilpotent ideal; the rest is equally trivial. From (1.1) we deduce an exact cohomology sequence:
(∗) H¹(Y, P_n) ──u¹──→ H¹(Y, O*_{X_{n+1}}) ──v¹──→ H¹(Y, O*_{X_n}) ──d──→ H²(Y, P_n).
On the other hand, for every , one knows how to identify with ; moreover, if is an invertible -Module corresponding to a cohomology class , the cohomology class corresponding to the inverse image of on is equal to .
Whence the following proposition:
Proposition.
Retain the notations introduced above. Let . The map :
- is injective for , if for ;
- is an isomorphism for , if for and .
Of course, the exact sequence (∗) contains more information than the proposition above. The reader will have noticed that we have said nothing about the functor . Given two invertible -Modules , , the sheaf is also invertible. If we indicate reduction modulo by a subscript , we find an exact sequence:
0 ⟶ H_0 ⊗ P_n ⟶ Hom(E_{n+1}, F_{n+1}) ⟶ Hom(E_n, F_n) ⟶ 0.
Whence an exact cohomology sequence that we shall not write down and whose interpretation is evident; one may use this remark to study the functor .
2. Comparison of Pic(X) and Pic(X̂)
The reader will find in Exposé X, no. 2, the proof of what follows:
Proposition.
Suppose that holds; then for every open neighborhood of in , the functor
is fully faithful, so that the map
is injective. If holds, then the map (2.3) is an isomorphism:
lim→_U Pic(U) ⟶ Pic(X̂).
Corollary.
Suppose that holds and that for every integer ; then for every open , the maps
Pic(X) ⟶ Pic(U) ⟶ Pic(Y_n)
are injective for . If holds and if, moreover, for every integer and , then the map
lim→_U Pic(U) ⟶ Pic(Y_n)
is an isomorphism for .
3. Comparison of P(X) and P(U)
A definition:
Definition. 1
Let be a prescheme and let be a closed part of . Set . We say that is parafactorial at the points of if, for every open set of , the functor is an equivalence of categories. We also say that the pair is parafactorial.
Recall that denotes the category of Modules locally free of rank 1 on .
Definition.
A noetherian local ring is said to be parafactorial if the pair is parafactorial.
One proves the following proposition, which shows that the notion is "pointwise":
Proposition.
Suppose is locally noetherian. In order that the pair be parafactorial, it is necessary and sufficient that, for every , the local ring be so.
Note that in "parafactorial" there is "fully faithful". One proves, as in Lemma 3.5 of Exposé X, the:
Lemma.
If is a locally noetherian prescheme and if is a closed part of , the following conditions are equivalent:
- for every open set of , the functor is fully faithful;
- the homomorphism is an isomorphism;
- for every , one has .
Thus "parafactorial" means that the conditions of 3.4 are satisfied and that, for every open set of , the homomorphism is surjective. In particular, if is the spectrum of a noetherian local ring, we find:
Proposition.
Let be a noetherian local ring; in order that it be parafactorial, it is necessary and sufficient that and , where we have set and is the unique closed point of .
Note that a local ring of dimension is never parafactorial, since its depth is . Hence "factorial" does not imply "parafactorial"; however, the converse holds for noetherian local rings of dimension , as we shall see below.
Lemma.
Let be a locally noetherian prescheme and let be a closed part of . Let be a faithfully flat and quasi-compact morphism. Set . If is parafactorial, then so is .
We first remark that, if denotes the canonical immersion of into , the formation of
the direct image by of a quasi-coherent O_U-Module commutes with the base change , since the latter is flat. It
is therefore equivalent to assume the equivalent conditions of Lemma 3.5 for or for , since
is a morphism of descent for the category of quasi-coherent sheaves. It remains to prove that, for every open set of
, is surjective. We make the base change , which
changes nothing (sic), and we are reduced to the case . We then remark that, if is an invertible
O_U-Module and if admits a locally free prolongation, this prolongation is isomorphic to , because of
what has just been seen. It remains to prove that is invertible. Using once more the fact that the direct
image by commutes with flat base change, and that "locally free of rank 1" is a property that descends by faithfully
flat and quasi-compact morphism, we are done.
Corollary.
Let be a noetherian local ring; if  is parafactorial, so is .
Do not believe that, if is parafactorial, so is Â.2
Before stating the principal theorem of this section, let us make the connection with the theory of divisors and the notion of factorial ring.3
Let be a noetherian and normal prescheme. Let be the free abelian group generated by the such that . The local ring of such a point is a discrete valuation ring. We shall write for the corresponding normalized valuation. Let be the ring of rational functions on and let
be the map that to every associates the codimension-one cycle:
(f) = Σ_{x ∈ X, dim O_{X,x} = 1} v_x(f) · x.
The image of is denoted , and its elements are called principal divisors.4 We set
Let be the subgroup of whose elements are the locally principal divisors. One knows that
and consequently is identified with a subgroup of .
Note that if is a dense open of , then is an isomorphism, and that if , i.e. if every such that belongs to , the homomorphism , and consequently , is also an isomorphism. Finally, if every is factorial — i.e. is so — then , and so .
Proposition.
Let be a noetherian and normal prescheme. Let be a family of open sets of such that:
- the form a filter base;5
- if one sets , then for every ;
- if for every , then is factorial.
Then one has an isomorphism:
lim→_{i ∈ I} Pic(U_i) ──≅──→ Cl(X).
Note that b) implies that every such that belongs to for every . Hence the are dense, and moreover the homomorphism is an isomorphism, as is . So . To prove what is desired, it therefore suffices to show that every belongs to for a suitable . It suffices to do this for irreducible positive "divisors". Let then be such that . It suffices to prove that there exists such that is locally principal at the points of . Let be the largest ideal of definition of the closed set . The set of points in whose neighborhood is free is an open set . Now by c). If we set , then with ; now is closed, so admits a finite number of generic points, so is contained in the union of finitely many , hence in some for a , because the form a filter base. Thus . QED.
Corollary.
Let be a noetherian and normal prescheme and let be a closed part of codimension . Suppose that, for every , is factorial; then
Pic(X − Y) ⟶ Cl(X − Y) ⟶ Cl(X)
are isomorphisms.
Corollary.
Let be a noetherian, normal local ring. Set and . In order that be factorial, it is necessary and sufficient that and that implies factorial.
Indeed, in order that be factorial, it is necessary and sufficient that .6
Corollary.
Let be a noetherian local ring of dimension . Set and let . Set . The following conditions are equivalent:
- is factorial;
- a) for every , is factorial, and b) is parafactorial, i.e. and .
Before proving this corollary, let us state the:
Serre's criterion of normality. 7
Let be a noetherian local ring. In order that be normal, it is necessary and sufficient that
- for every prime ideal of such that , be normal;
- for every prime ideal of such that , one have .
Let us prove 3.10.
(i) ⇒ (ii). Knowing that a localization of a factorial ring is factorial, we have (ii) a). Moreover is normal, so , since (3.11 (ii)). Finally is parafactorial; indeed (cf. 3.9).
(ii) ⇒ (i). We first prove that is normal by applying Serre's criterion. Since , condition (i) of the criterion is among the hypotheses. Moreover, for every , is factorial, hence normal, hence of depth , at least if . Finally by (ii) b). It remains to apply 3.9.
Let us summarize the preceding:
Proposition.
Let be a locally noetherian prescheme and let be a quasi-coherent ideal of . Set . Let . Suppose that:
- holds (Exposé X);
- if or
2and if ; - for every open neighborhood of in and every , the ring is parafactorial.
Then, for every and every open neighborhood of , the homomorphisms
Pic(X) ⟶ Pic(U) ⟶ Pic(X_n)
(with the canonical commutative triangle) are isomorphisms.
One knows some parafactorial rings:
Theorem.
- (Auslander–Buchsbaum)8 A regular noetherian local ring is factorial (hence parafactorial if its dimension is ).
- A noetherian local ring of dimension that is a complete intersection is parafactorial.
Corollary (Samuel conjecture)9.
A noetherian local ring that is a complete intersection and that is factorial in codimension (i.e. implies that is factorial) is factorial.
Proof of the corollary. We argue by induction on the dimension of . If , then is factorial by hypothesis. If , by the induction hypothesis, and remarking that a localization of a complete intersection is also a complete intersection, all localizations of other than itself are factorial. By Theorem 3.13 (ii), is parafactorial, hence factorial by 3.10.
Proof of 3.13 (i) (following Kaplansky).10
Let be a regular noetherian local ring; set . If or 1, the result is known. Suppose
, and argue by induction on : suppose and the theorem proved for rings of dimension
. Set and , where . The localizations of other than
are regular and of dimension , hence factorial. Moreover . It therefore suffices to
prove that (Cor. 3.10). Let then be an invertible O_X-Module; one knows that one can
prolong it to a coherent -Module . There exists a resolution of by free -Modules:
since the cohomological dimension of is finite. By restriction to one obtains a finite free resolution. It therefore suffices to prove the following lemma:
Lemma.
Let be a ringed space and let be a locally free O_X-Module that admits a finite resolution by free
modules of finite type. Then .
Recall that one defines as the maximal exterior power of . In the case envisaged, since is invertible, so the lemma allows us to conclude. Let us prove this lemma. Let
0 ⟵ L_0 ⟵ L_1 ⟵ L_2 ⟵ ⋯ ⟵ L_n ⟵ 0
be the announced exact sequence, where . Since everything is locally free, one has:
⨂_{0 ⩽ i ⩽ n} (det(L_i))^{(−1)^i} ≃ O_X;
now all the for are free, so their determinants are free as well, hence so is the determinant of . QED.
It remains to prove (ii) of the theorem. Beforehand, let us prove a lemma that will permit us to proceed by induction:
Lemma.
Let be a noetherian local ring that is a quotient of a regular ring. Let be an -regular element. Suppose that is complete for the -adic topology. Set , , , , , . Suppose that:
- for every closed in , one has ,
- ,
then the map is injective. In particular, if is parafactorial, then so is .
One knows that a) implies thanks to X 2.1. If we prove that for every , we shall know thanks to (2.2) that is injective. If, moreover, is parafactorial, we shall know that (3.5), hence ; now since is -regular, so will be parafactorial by 3.5.11
Let be the -Module associated with the ideal tA. In no. 1, we set
for every . Now is -regular, so . It therefore
remains to prove that . Now is an open subset of , so we have an exact sequence
(I (27)):
H¹(Y′, O_{Y′}) ⟶ H¹(Y, O_Y) ⟶ H²_x(Y′, O_{Y′}),
whose right-hand term is zero by virtue of hypothesis b), and whose left-hand term is zero because is affine. QED.
Lemma.
Retaining the hypotheses of 3.16, suppose moreover that:
- (c) for every closed in , ,
- (d) (stronger than b),
- (e) for every closed in with , the ring is parafactorial.
Then the map is an isomorphism; in particular, in order that be parafactorial, it is necessary and sufficient that be so.
One knows (X 2.1) that a) and c) imply . Moreover, by the reasoning just made, d) implies that for every and or . Furthermore, for every open neighborhood of in , the complement of in consists of a finite number of closed points. Thanks to e) and Theorem 3.12, we deduce that is an isomorphism. On the other hand, ; by criterion 3.5, we deduce that is parafactorial if and only if is so.
Let us now prove 3.13 (ii). Let be a regular noetherian local ring. Let be an -sequence. Set and suppose . We must prove that is parafactorial. We argue by induction on . If , then is regular, hence factorial by 3.13 (i), hence parafactorial by 3.10. Suppose and the theorem proved for . Set , so . We may suppose complete by 3.7. By the induction hypothesis, is parafactorial. Let us prove that we may apply Lemma 3.17. We have supposed complete, hence so is , and therefore is complete for the -adic topology. If , and if is closed in , then is a complete intersection of dimension , with . By the induction hypothesis, is parafactorial, and moreover of depth . This gives a), c), and e). Moreover , whence d). QED.
Theorem.
Let be a locally noetherian prescheme and let be a coherent sheaf of ideals on . Set . Let be an integer. Suppose that:
- holds (cf. examples X 2.1 and X 2.2);
- for every , one has for and ;
- for every open and every , the ring is regular of dimension or a complete intersection of dimension .
Then, for every open and every integer , the homomorphisms
Pic(X) ⟶ Pic(U) ⟶ Pic(Y_p)
are isomorphisms, where denotes the prescheme .
It suffices to combine 3.12 and 3.13.
For a more detailed study of the notion of parafactoriality, and the proof of 3.3, cf. EGA IV 21.13, 21.14.
N.D.E. For a precise study of the link between the factoriality of and that of its completion, see (Heitmann R., "Characterization of completions of unique factorization domains", Trans. Amer. Math. Soc. 337 (1993), no. 1, pp. 379–387).
For the generalities that follow, cf. also EGA IV 21.
In conformity with the terminology of EGA IV 21, we now prefer to reserve the name "divisors" for "locally principal divisors" or "Cartier divisors".
N.D.E. i.e. a decreasing filtered family.
N.D.E. See Bourbaki, Algèbre commutative VII.1.4, cor. to th. 2, and VII.3.2, th. 1.
Cf. EGA IV 5.8.6.
N.D.E. To be compared with the following purity result, due to Gabber. Let be the spectrum of a regular local ring of dimension 3, an element of nonzero differential, i.e. , and the complement of . Then a vector bundle on is free (for a simple proof, see Swan R.G., "A simple proof of Gabber's theorem on projective modules over a localized local ring", Proc. Amer. Math. Soc. 103 (1988), no. 4, pp. 1025–1030). The rank-1 case is a particular case of Theorem 3.13. For purity results concerning vector bundles of arbitrary rank, in either the analytic or the algebraic setting, see (Gabber O., "On purity theorems for vector bundles", Internat. Math. Res. Notices (2002), no. 15, pp. 783–788).
N.D.E. For a proof in the same vein, but more elementary, see Call F. & Lyubeznik G., "A simple proof of Grothendieck's theorem on the parafactoriality of local rings", in Commutative algebra: syzygies, multiplicities, and birational algebra (South Hadley, MA, 1992), Contemp. Math., vol. 159, American Mathematical Society, Providence, RI, 1994, pp. 15–18.
It is the proof reproduced in EGA IV 21.11.1.