Exposé XII. Algebraic Geometry and Analytic Geometry
Mme M. Raynaud. [Translator note: according to unpublished notes of A. Grothendieck.]
Proceeding as in [XII.10], one associates to every scheme locally of finite type over the field of complex numbers an analytic space , whose underlying set is .
In nos. XII.2 and XII.3 of this exposé, we give a “dictionary” between the usual properties of and of , and between the properties of a morphism and of the associated morphism .
We then show that the comparison theorems between coherent sheaves on and , established in [XII.10, no. 12] for a projective variety, are still valid when is a proper scheme.
Finally, in no. XII.5 we prove the equivalence between the category of finite étale coverings of and the category of finite étale coverings of . As a bonus for the reader, we give a new proof of the Grauert-Remmert theorem [XII.6], using resolution of singularities [XII.8].
1. The Analytic Space Associated with a Scheme
Let be a scheme locally of finite type over . Let be the functor from the category of analytic spaces [XII.4, no. 9] to the category of sets which associates to an analytic space the set of morphisms of ringed spaces in -algebras . One has the following theorem:
Theorem-Definition.
The functor is representable by an analytic space and a morphism . One says that is the analytic space associated with .
If is the underlying set of , induces a bijection from to the set of points of with values in . Moreover, for each point of , the morphism
which is necessarily local, gives after passage to completions an isomorphism
In particular the morphism is flat.
Notice that the fact that induces a bijection from to follows from the universal property of . On the other hand, one has the following assertions:
a. If the theorem is true for a scheme , then it is also true for every subscheme of . Suppose first that is an open subscheme of . If is the canonical morphism, is an open subset
of , endowed with the analytic-space structure induced by that of . Since every morphism from an analytic space to factors through by the universal property of the latter, and hence through , which is the fiber product , is the analytic space associated with . Finally, the assertion concerning the is evident.
It remains only to consider the case where is a closed subscheme of . Let be the coherent -ideal defining . Then is a coherent sheaf of ideals on defining a closed analytic subspace of . As in the case of an open subscheme, one sees that is the analytic space associated with . Let be the canonical morphism. For every point of , the morphism is none other than the morphism
𝒪_Y,ψ(x)/I_ψ(x) → 𝒪_Y^an,x / I_ψ(x) · 𝒪_Y^an,x
induced by . Its completion
is an isomorphism, since is one; this proves a.
b. If one has two -schemes , , such that and exist, then also exists. Indeed, let and be the canonical morphisms, and let , be the two projections from . It follows formally from EGA I 1.8.1 that is the product of and in the category of ringed spaces in local rings. Consequently the morphisms and define a
morphism , and the pair represents the functor .
c. If denotes affine space of dimension 1, that is, the topological space endowed with the sheaf of holomorphic functions, the functor is representable by , the canonical morphism being the evident morphism. [Translator note: the corrected source adds in 2003 that denotes the algebraic affine line over .] Indeed, to give a morphism from an analytic space to is equivalent to giving an element of , which is also equivalent to giving a morphism from to . Plainly one has a bijection , and, for each point , the morphism is none other than the identity morphism of a ring of formal power series in one variable over .
It follows from b and c that the theorem is true for affine space , . Using a, one sees that it is also true for every affine scheme locally of finite type over . If is no longer assumed affine and if is a covering of by affine opens, it follows from the universal property and from a that the glue and thus define the analytic space associated with .
1.2.
Let be a morphism of -schemes locally of finite type. If and are the canonical morphisms, it follows from the universal property of that there exists a unique morphism such that the diagram
X^an → X
| |
f^an f
| |
Y^an → Y
is
commutative. We have therefore defined a functor from the category of -schemes locally of finite type to the category of analytic spaces.
The functor commutes with finite projective limits. Indeed it is enough to see that commutes with fiber products. But if , , are schemes locally of finite type over , it follows from the fact that is the fiber product of and over in the category of ringed spaces in local rings that satisfies the universal property characterizing .
1.3.
Let be a -scheme locally of finite type, let be the associated analytic space, and let be the canonical morphism. If is an -module, the inverse image is a sheaf of modules over . This defines a functor from the category of -modules to the category of modules on . This functor commutes with inductive limits (EGA 0 4.3.2). Since the sheaf is coherent [XII.4, no. 18, §2, th. 2], it sends coherent sheaves to coherent sheaves (EGA 0 5.3.11). Moreover:
Subproposition.
The functor which associates to an -module its inverse image on is exact, faithful, and conservative.
Exactness follows from the fact that the morphism is flat (XIII.1.1). Let us prove that the functor is faithful. Taking exactness into account, it is enough to show that if is zero, then itself is zero. But for every point of one then has
F_φ(x) ⊗_𝒪_X,φ(x) 𝒪_X^an,x = 0.
Since the morphism is faithfully flat, one has for every closed point of ; and since is Jacobson (EGA IV 10.4.8), this implies that is zero.
The
fact that the functor is conservative is formal from exactness and faithfulness.
2. Comparison of Properties of a Scheme and of the Associated Analytic Space
Proposition.
Let be a -scheme locally of finite type, let be the associated analytic space, and let be an integer. Consider the property of being:
(i) nonempty
(i′) discrete
(ii) Cohen-Macaulay
(iii) (S_n)
(iv) regular
(v) (R_n)
(vi) normal
(vii) reduced
(viii) of dimension n.
Then has property if and only if has property .
Let be the canonical morphism. Assertion (i) follows from the fact that (XIII.1.1) and from the fact that is Jacobson (EGA IV 10.4.8). To say that , respectively , is discrete is equivalent to saying that , respectively by [XII.4, no. 19, §4, cor. 6]; hence (i′) follows from (viii).
Let be one of the properties (ii) through (vii). For to have property , it is necessary and sufficient that hold at every closed point of . Indeed, since is excellent (EGA IV 7.8.6 (iii)), the set of points
where satisfies is open, and if this open contains all closed points, it is equal to all of . Thus to say that , respectively , has property is equivalent to saying that for every point of , the local ring , respectively , has property . Since the fact that an excellent local ring has property can be detected after passage to the completion, the proposition follows from the isomorphisms
in cases (ii) through (vii). The same holds in case (viii), taking into account the relations
dim X = sup_x dim 𝒪_X,φ(x), dim X^an = sup_x dim 𝒪_X^an,x,
where . This completes the proof.
Proposition.
Let be a -scheme locally of finite type, let be the canonical morphism, and let be a locally constructible subset of . Then one has the relation
We may suppose that is a dense open subset of . Let be the reduced closed subscheme of whose underlying space is . The associated space is a closed analytic subspace of whose underlying space is . We must show that every point of belongs to . But at such a point , the germ of analytic space contains the subgerm , and this is defined by a non-nilpotent ideal of . It then follows from the Nullstellensatz [XII.4, no. 19, §4, cor. 3] that every open neighborhood of contains points of which do not belong to . This proves that .
Corollary.
Let
be a -scheme locally of finite type, let be the canonical morphism, and let be a locally constructible subset of . For to be an open subset, respectively a closed subset, respectively a dense subset, it is necessary and sufficient that have the corresponding property.
The corollary follows from XII.2.2 and from the fact that, since is a Jacobson scheme (EGA IV 10.4.8), two locally constructible subsets of that have the same trace on the very dense set are equal.
Proposition.
Let be a -scheme locally of finite type. For to be connected, respectively irreducible, it is necessary and sufficient that be connected, respectively irreducible.
Suppose is connected, respectively irreducible. The image of in is then connected, respectively irreducible. It follows that is connected, respectively irreducible, because closed subsets of and of correspond bijectively (EGA IV 10.1.2).
Conversely suppose is connected, respectively irreducible, and let us show that the same is true of . We may restrict to the case where is irreducible. Indeed, suppose is connected. Given a point of , the set of points for which there exists a finite sequence of irreducible closed subschemes of , with , , and for , is both open and closed, hence equal to all of . For a sequence as above, one also has for ; if the are known to be connected, then is connected as well.
From now on
suppose is irreducible. We may also suppose affine. Indeed, if is a covering of by affine opens, any two of these opens have nonempty intersection, and the same property is therefore true for the covering of . If the are known to be irreducible, then is irreducible as well.
We may further suppose that is normal. Indeed, let be the normalization of . Since the morphism is surjective, so is , which proves that if is irreducible, then is irreducible as well.
From now on suppose is affine normal. Since the local rings of are integral domains, saying that is irreducible is equivalent to saying that it is connected. Indeed, if is a closed analytic subset of , the set of points of at which is a closed analytic subset of [XII.4, no. 20 A, cor. 1] which is also open. If is connected, this proves that, whenever , is rare; hence is irreducible. We are thus reduced to showing that is connected.
Let
be a compactification of , where is a normal projective -scheme and is a dominant open immersion. It then follows from [XII.10, no. 12, th. 1] that is connected. Since is obtained by removing from a rare closed analytic subset, it follows from XII.2.5 below that is also connected.
Lemma.
Let
be a connected normal analytic space, and let be a rare closed analytic subset. Then is connected.
When has codimension , the proposition follows from [XII.11, no. 3, prop. 4]. In the general case one may suppose, after removing from a closed analytic subset of codimension , that and , regarded as a reduced analytic subspace of , are regular. By the implicit function theorem, every point of has a neighborhood isomorphic to a ball in an affine space , such that is defined by the vanishing of a certain number of coordinate functions. This proves that is connected, and hence that is connected.
Corollary.
Let be a -scheme locally of finite type. The morphism
induced by the canonical morphism is bijective.
3. Comparison of Properties of Morphisms
Proposition.
Let be a morphism of -schemes locally of finite type, and let be the morphism deduced from on the associated analytic spaces. Let be the property of being:
(i) flat
(ii) net, that is, unramified
(iii) étale
(iv) smooth
(v) normal
(vi) reduced
(vii) injective
(viii) separated
(ix) an isomorphism
(x) a monomorphism
(xi) an open immersion.
Then has property if and only if has property .
Let and be the canonical morphisms. Let be a point of , and put . The morphisms and deduced from and give the same morphism after passage to completions (XII.1.1). By [XII.2, ch. 3, §5, prop. 4], respectively EGA IV 17.4.4, it is therefore equivalent to say that satisfies property (i), respectively (ii), and to say that satisfies (i), respectively (ii), at every closed point of . Since the set of points of where (i), respectively (ii), holds is open (EGA IV 11.1.1 and I 3.3), this proves (i) and (ii), hence also (iii).
Let be property (iv), respectively (v), respectively (vi). Taking XII.2.1 ((v), (vi), (vii)) into account, it is equivalent to say that the geometric fibers of at the various points of are regular, respectively normal, respectively reduced, and to say that the same is true of the geometric fibers of at the various closed points of . Cases (iv), respectively (v), respectively (vi), then follow from (i) and from the fact that the set of points of where the geometric fibers of are regular is open (EGA IV 12.1.7).
(vii). If is injective, so is . Conversely suppose is injective and let us show that is injective. We may suppose
of finite type. Since is injective, the fibers of at closed points of are radicial. Since the set of points of whose fiber is radicial is locally constructible (EGA IV 9.6.1), and since is a Jacobson scheme, all fibers of are radicial; hence is injective.
(viii). Let and be the diagonal immersions, and let be the canonical morphism. By XII.2.3, saying that is closed in is equivalent to saying that is closed in .
Since an open immersion is nothing other than an étale injective morphism (EGA IV 17.9.1 and [XII.4, no. 13, §1]), (xi) follows from (iii) and (vii). Since an isomorphism is the same thing as a surjective open immersion, (ix) follows from (xi) and XII.3.2 (i) below. Saying that is a monomorphism is equivalent to saying that the diagonal morphism is an isomorphism, so (x) follows from (ix).
Proposition.
Let and be two -schemes locally of finite type, let be a morphism of finite type, and let be the morphism deduced from on the associated analytic spaces. Let be the property of being:
(i) surjective
(ii) dominant
(iii) a closed immersion
(iv) an immersion
(v) proper
(vi) finite.
Then has property if and only if has property . [Translator note: the source footnote says that a morphism of analytic spaces is called proper if it is proper in the sense of [XII.1, ch. 1, §10, no. 1] and is separated.]
Let and be the canonical morphisms.
(i). If is surjective, then for every point of , is a nonempty closed subset of ; hence it contains at least one closed point, which proves that is surjective. Conversely, if is surjective, is a locally constructible subset of (EGA IV 1.8.4) containing all closed points of ; hence .
(ii) follows from XII.2.2.
(iii). If is a closed immersion, then so is by XII.1.1 a. Conversely, if is a closed immersion, then so is by XII.3.1 (x) and XII.3.2 (v), since this is equivalent to saying that is a proper monomorphism (EGA IV 8.11.5).
(iv). It is clear that if is an immersion, then so is . Conversely suppose is an immersion, and let be the image of in , and the scheme-theoretic closure of . There is a factorization of
X --i→ T̄ --j→ Y,
where is a closed immersion and is the canonical morphism; from it one deduces the following factorization of :
X^an --i^an→ T̄^an --j^an→ Y^an.
Since
is a locally constructible subset of (EGA IV 1.8.4), one has, by XII.2.2, . It follows that is open in , hence that is open in . Consider the canonical factorization of
X --i₁→ i(X) --i₂→ T̄.
The morphism is a proper monomorphism, hence so is by XII.3.2 (v) and XII.3.1 (x). This proves that , and hence also , is an immersion.
(v). Suppose is proper and let us show that is proper. Since properness of is local on , we may suppose affine. By Chow’s lemma (EGA II 5.6.1), one can find a projective -scheme and a projective surjective morphism
The morphism is projective, hence proper; is surjective; and it follows from [XII.1, ch. 1, §10] that is proper.
Conversely suppose is proper and let us show that is proper. By XII.3.1 (viii), is separated. It remains to prove that is universally closed, and it is even enough to show that is closed. Indeed, for every -scheme locally of finite type, the morphism
f_(Y′) = h: X ×_Y Y′ → Y′
will also be closed since is proper. Let be a closed subset of . The set is locally constructible, and one has
Since
is proper, is a closed subset of , and therefore it follows from XII.2.2 that
This implies , that is, is closed; hence is proper.
(vi). Saying that a morphism is finite is equivalent to saying that it is proper with finite fibers (EGA III 4.4.2 and [XII.4, no. 19, §5]). Since the set of points where the fibers of are finite is locally constructible (EGA IV 9.7.9), the fibers of are finite if and only if the fibers of are finite. Thus (vi) follows from (v).
Remark.
a. Let be a morphism of -schemes locally of finite type. The fact that is a local isomorphism does not imply that is one. Indeed, if is étale, is étale and hence is a local isomorphism [XII.4, no. 13, §1], but this need not be true of .
b. The statement XII.3.2 is not true if is not assumed of finite type. For example, can be a closed immersion without being one. It is enough to take to be the sum of copies of , to be the affine line, and the morphism obtained by sending the points of to distinct points of forming a discrete subset.
4. Cohomological Comparison Theorems and Existence Theorems
The purpose of this number is to reprove the results of [XII.3, no. 2, ths. 5 and 6]. These generalize to the case of a proper scheme the theorems established in [XII.10, no. 12] when is projective, and extend them to the relative case. More general results, concerning relative proper schemes over an analytic space, are proved in [XII.7, ch. VIII, no. 3].
Recall that the Čech cohomology used in [XII.10, no. 12] coincides with the usual cohomology in the algebraic case as well as in the analytic case (EGA III 1.4.1 and [XII.5, II 5.10]).
4.1.
Let be a morphism of -schemes locally of finite type, and consider the commutative diagram
X^an --φ→ X
| |
f^an f
| |
Y^an --ψ→ Y.
If is an -module, then for every integer one has morphisms
Rᵖf_*F --i→ Rᵖf_*(φ_*F^an) --j→ Rᵖ(f·φ)_*F^an --k→ ψ_*(Rᵖf_*^an F^an),
where is deduced from the canonical morphism , and , are “edge homomorphisms” of Leray spectral sequences. With the composite there is associated a canonical morphism
Theorem.
Let
be a proper morphism of -schemes locally of finite type, and let be a coherent -module. Then, for every integer , the morphism (4.1.1)
is an isomorphism.
- The case where is projective. The proof is analogous to that of [XII.10, no. 13]. Let us recall it briefly. One reduces to the case where is a projective space of type over . Let and . One first proves that
f_*^an 𝒪_𝓟 = 𝒪_𝓨, Rᵖf_*^an(𝒪_𝓟) = 0 for p > 0.
To verify the preceding relations, one may reduce to the case where is a ball in an affine space . One considers the “standard covering” of by open subsets isomorphic to . Since these opens are Stein, one has, for every integer , isomorphisms
One can then express the sections of the structural sheaf on the opens and on their intersections in terms of Laurent series. An easy calculation proves that
H⁰(𝓟,𝒪_𝓟) ≃ H⁰(𝓨,𝒪_𝓨), Hᵖ(𝓟,𝒪_𝓟) = 0 for p > 0.
The proof is then completed by copying [XII.10, no. 12, lemma 5], with cohomology groups replaced by cohomology sheaves.
The case where is proper. One uses EGA III 3.1.2 to reduce to the projective case. Let be the category of coherent -modules such that is an isomorphism for every . It is enough to prove that, for every exact sequence whose two terms are in , the third is also in ; that a direct factor of an object of is in ; and that, for every point of , one can find an object of such that .
The first condition follows by applying the five lemma to the following commutative diagram, whose rows are exact:
… → (Rᵖf_*F′)^an → (Rᵖf_*F)^an → (Rᵖf_*F″)^an → (Rᵖ⁺¹f_*F′)^an → …
↓ ↓ ↓ ↓
… → Rᵖf_*^an F′^an → Rᵖf_*^an F^an → Rᵖf_*^an F″^an → Rᵖ⁺¹f_*^an F′^an → …
The second condition is verified analogously.
To verify the third condition, one may restrict to the case where is an irreducible scheme with generic point . We could have supposed noetherian from the beginning. By Chow’s lemma (EGA II 5.6.1), one can find a projective -scheme and a projective surjective morphism . On the other hand, there exists an integer such that for all and such that the canonical morphism is surjective (EGA III 2.2.1). If one puts , the sheaf answers the question. Indeed ; moreover, the Leray spectral sequence
is degenerate, so one has an isomorphism
As in the algebraic case, one has a canonical isomorphism
and the diagram
is commutative. By 1, is an isomorphism; hence is also an isomorphism. This completes the proof.
Corollary.
Let be a proper -scheme, and let be a coherent -module. Then, for every integer , the canonical morphism
is an isomorphism.
Theorem.
Let be a proper -scheme. The functor which associates to every coherent -module its inverse image on is an equivalence of categories.
- The functor is fully faithful. Indeed, let and be two coherent -modules. The canonical morphism
Hom_𝒪_X(F,G) → Hom_𝒪_X^an(F^an,G^an)
identifies with the canonical morphism
H⁰(X,SheafHom_𝒪_X(F,G)) → H⁰(X^an,SheafHom_𝒪_X(F,G))
(EGA 0_I 6.7.6). Since SheafHom_𝒪_X(F,G) is coherent, it follows from XII.4.3 that this morphism is bijective.
- The functor is essentially surjective. When is projective, the assertion follows from [XII.10, no. 12, th. 3]. The general case reduces to the preceding one by using Chow’s lemma (EGA II 5.6.1). Indeed, let be a projective -scheme, let be a projective surjective morphism, and let be a dense open subset of such that induces an isomorphism . We reason by noetherian induction on ; hence we may suppose that for every coherent sheaf on for which one can find a closed subset of , distinct from , satisfying , there exists a coherent sheaf on such that .
Let be a coherent sheaf of modules over , and let and be the coherent sheaves defined by requiring the sequence
0 → 𝓚 → 𝓕 → f_*^an f^an*𝓕 → 𝓛 → 0
to be exact. Since is projective, there exists a coherent -module such that . From XII.4.2 one then deduces an isomorphism . Since and are zero, there exist coherent -modules and such that and . By 1, the morphism comes from a unique morphism . Let . The sheaf is then an extension of by , and it remains only to see that this extension comes
by inverse image from an extension of by . It is therefore enough to prove that the canonical morphism
(*) Ext^q_𝒪_X(I,K)^an ≃ Ext^q_𝒪_X^an(I^an,K^an), q = 1,
is bijective. [Translator note: the source prints “,” but the preceding sentence shows that the needed case is ; this is a mathematical correction rather than a change of argument.] Now one has isomorphisms
ExtSheaf^q_𝒪_X(I,K)^an ≃ ExtSheaf^q_𝒪_X^an(I^an,K^an)
for every integer (EGA 12.3.5), and a morphism of spectral sequences
Hᵖ(X,ExtSheaf^q_𝒪_X(I,K)) ⇒ Ext^(p+q)_𝒪_X(I,K)
↓ ↓
Hᵖ(X^an,ExtSheaf^q_𝒪_X^an(I^an,K^an)) ⇒ Ext^(p+q)_𝒪_X^an(I^an,K^an).
This morphism is an isomorphism because, by XII.4.3, it is so on the -terms; this proves the bijectivity of
(*).
Corollary.
The functor which associates to every proper -scheme is fully faithful.
We must show that, if and are two proper -schemes, the canonical map
is bijective. But to give a morphism from to , respectively from to , is equivalent to giving its graph, that is, a closed subscheme of , respectively a closed analytic subspace of , such that the restriction of the first projection to , respectively of to , is
an isomorphism. Since giving a closed subscheme of , respectively a closed analytic subspace of , is equivalent to giving a coherent sheaf of ideals on , respectively on , the corollary follows from XII.4.4.
Corollary.
Let be a proper -scheme. The functor which associates to every finite, respectively finite étale, scheme over is an equivalence from the category of finite, respectively finite étale, schemes over to the category of finite, respectively finite étale, analytic spaces over .
Indeed, to give a finite morphism , respectively , is equivalent to giving a coherent sheaf of algebras over , respectively over [XII.4, no. 19, §5, th. 2]. The corollary therefore follows from XII.4.4 in the non-respective case, and the respective case follows from it in view of XII.3.1 (iii).
5. Comparison Theorems for Étale Coverings
5.0.
Let us make precise the notion of a finite covering of an analytic space. If is an analytic space, an analytic space finite over is called a finite covering of if every irreducible component of dominates an irreducible component of .
Theorem (“Riemann existence theorem”).
Let be a -scheme locally of finite type, and let be the analytic space associated with . The functor which associates to every finite étale covering of is an equivalence
from the category of finite étale coverings of to the category of finite étale coverings of .
- The functor is fully faithful. Let and be two finite étale coverings of , and let us prove that the canonical map
is bijective. We may suppose connected. To give an -morphism from to is equivalent to giving a
connected component of such that the morphism induced by the first projection
is an isomorphism. Since the connected components of correspond bijectively to the connected
components of (XII.2.6), and since a morphism is an isomorphism if and
only if is one, this proves the bijectivity of (*).
- The functor is essentially surjective. Let be a finite étale covering of , and let us prove that there exists an étale covering of such that one has an isomorphism . In view of 1, the question is local on , so we may suppose affine.
a. Reduction to the case where is normal. We may suppose reduced. Indeed, suppose the theorem proved for . The functor which associates to a finite étale covering of the finite étale covering of is then an equivalence. Since it is obtained by composing with the functor which associates to a finite étale covering of its inverse image on , and since is fully faithful, this shows that is an equivalence of categories.
We
may suppose normal. Indeed, let be the normalization of , and let be the canonical morphism. Since is finite, is a morphism of effective descent for the category of étale coverings (IX.4.7). Supposing the theorem proved for , put . There exists an étale covering of and an isomorphism . It then follows from 1 that the natural descent datum on lifts to a descent datum on relative to . This proves the existence of an étale covering of such that one has an isomorphism
i: X′^an ×_{X^an} X̃^an ≃ 𝓧̃′,
whose inverse images by the two projections from are the same. By IX.3.2, whose proof is valid in the analytic case, the morphism is a morphism of descent for the category of étale coverings, and consequently comes from an isomorphism .
b. Reduction to the case where is regular. Let be the open subset of regular points of , and let and be the canonical morphisms. Since is normal, one has . Suppose that there exists an étale covering of such that , and let us show that then extends to an étale covering of such that . It is enough to see that extends to an étale covering of . Indeed, one will then have an isomorphism ; but if and are the coherent sheaves of algebras on defining respectively and , the fact that is normal and that implies that the canonical morphisms
𝓕 → i_*^an(𝓕|U^an), 𝓖 → i_*^an(𝓖|U^an)
are
isomorphisms [XII.11, no. 3, prop. 4]. It follows that and , and hence also and , are isomorphic.
Let be the canonical morphism. Since the problem of extending to is local on , it is enough to prove that, for every point of , the étale covering
U′_φ(y) = U′ ×_X Spec 𝒪_X,φ(y)
of
U_φ(y) = U ×_X Spec 𝒪_X,φ(y)
extends to . Let be the coherent -algebra defining . The canonical morphism
α: (i_*H)^an → i_*^an(H^an) = 𝓕
defines a morphism of sheaves of modules on :
whose restriction to
U_y = U_φ(y) ×_{Spec 𝒪_X,φ(y)} Spec 𝒪_X^an,y
is an isomorphism. But this proves that is trivial, hence that extends to .
c. The case where is affine regular. Let
be a compactification of , where is a projective -scheme and is a dominant open immersion. By the resolution of singularities theorem [XII.8], one can find a regular scheme and a projective morphism , such that induces an isomorphism and such that is the complement in of a divisor with normal crossings. Let
be the canonical immersion. We shall show that there exists a finite normal covering of , in the sense of XII.5.0, which extends the étale covering . By Proposition XII.5.3 below, such a covering is unique; the problem of extending is therefore local on near . But each point of has an open neighborhood isomorphic to a ball in an affine space , such that is defined by the vanishing of the first coordinate functions , with . The fundamental group of is isomorphic to , and every étale covering of is a quotient of a covering of the form
𝓤″ = 𝓤[T₁,…,T_p]/(T₁ⁿ¹ − z₁, …, T_pⁿᵖ − z_p),
where the are integers > 0, by a subgroup of the Galois group
of . But
extends to the regular covering
𝓥″ = 𝓥[T₁,…,T_p]/(T₁ⁿ¹ − z₁, …, T_pⁿᵖ − z_p)
of on which acts, and the quotient of by is the desired extension.
The proof is then completed by XII.4.6. The covering comes from a finite covering of ; the restriction of to is a covering of such that , and by XII.3.1 (iii), is an étale covering of .
Corollary.
Let
be a connected -scheme locally of finite type, let be the canonical morphism, and let be a point of . Let be the fundamental group of the topological space at , and let be the fundamental group of the scheme at (V.7). Then is canonically isomorphic to the completion of for the topology of subgroups of finite index.
Indeed, let be the category of finite étale coverings of , let be the functor from to Sets which associates to every finite étale covering of the set of points of above , and let be the profinite group associated with and as in V.4. Since every finite étale covering of is a quotient of the universal covering by a subgroup of finite index, is nothing other than the completion of for the topology of subgroups of finite index. The corollary therefore follows from XII.5.1 and V.6.10.
Proposition.
Let be a normal analytic space, and let be a closed analytic subset such that is dense in . Then the functor which associates to every normal finite covering of , in the sense of XII.5.0, its restriction to is fully faithful.
Let and be two finite normal coverings of . We must show that the canonical map
is bijective. Let , be two -morphisms from to whose restrictions to are the same, and let us prove that . The morphisms and
coincide on the dense open , hence on the underlying topological spaces. By [XII.4, no. 19, §4, cor. 5], this proves .
Let now be a -morphism from to , and let us show that it extends to all of . We may suppose regular. Indeed, since is normal, one can find an open subset of whose complement is an analytic subset of codimension , such that is regular. Let , and suppose the proposition proved for . Consider the commutative diagram
𝓥′ → 𝓧′
↓ ↓
𝓥 → 𝓧
and similarly 𝓥″ → 𝓧″ over 𝓥 → 𝓧.
With there is associated a morphism of -algebras
from which one deduces a morphism
Taking into account the isomorphisms and [XII.11, no. 3, prop. 4], one deduces a morphism of -algebras
hence the desired morphism .
We
now suppose regular. Let and . We regard as a reduced analytic subspace of . If is the singular closed subset of , then [XII.4, no. 20 D, th. 3]. Thus, by induction on the dimension of , one may suppose smooth. Since it is enough to extend to an open neighborhood of each point of , the implicit function theorem lets us suppose that is a ball in an affine space and that is the closed subset defined by the vanishing of the first coordinate functions , with .
To one associates a section of
p: 𝓧′ ×_𝓧 𝓧″ → 𝓧′
above . After restricting if necessary, one may suppose that is generated by elements of . Let be the images by of . Saying that extends to is equivalent to saying that extend to sections of . But, since is finite, each is a Laurent series in with coefficients that are convergent power series in , and satisfies integral-dependence relations. It follows that is bounded, hence is a convergent power series in , and therefore extends to .
One may ask whether the functor introduced in XII.5.3 is an equivalence of categories. This question has an answer by the theorem of Grauert-Remmert [XII.6], for which we give below a proof using resolution of singularities. One could also have used the Grauert-Remmert theorem to prove XII.5.1; that was what was done before [XII.8] was available.
Theorem (Grauert-Remmert theorem).
Let
be a normal analytic space, and let be a closed analytic subset such that is dense in . Let be a normal finite covering of . Suppose that there exists a rare closed analytic subset of such that the restriction of to is étale. Then there exists a normal finite covering of extending , and is unique up to isomorphism.
Uniqueness follows from XII.5.3. The problem of extending is therefore local on . We may suppose regular and étale over . Indeed, the set of regular points of is a dense open subset of whose complement is an analytic subset [XII.4, no. 20 D, th. 2], and it is enough to replace by the open subset .
Let be a point of and let us show that one can extend to a neighborhood of . After restricting to an open neighborhood of , it follows from the resolution of singularities theorem [XII.8] that one can find a regular analytic space and a projective morphism inducing by restriction to an isomorphism , such that is the complement in of a divisor with normal crossings. Let us show that extends to a normal finite covering of . Since the question is local on , one may suppose that is a ball in an affine space and that is defined by the vanishing of the first coordinate functions , with . [Translator note: the corrected source fixes to in this list.] The étale covering of is a quotient of a covering of the form
𝓤₂ = 𝓤₁[T₁,…,T_p]/(T₁ⁿ¹ − z₁, …, T_pⁿᵖ − z_p)
by a subgroup of the Galois group of . The covering extends
to the covering
𝓧₂ = 𝓧₁[T₁,…,T_p]/(T₁ⁿ¹ − z₁, …, T_pⁿᵖ − z_p)
of on which acts, and extends to .
Let denote the normal finite covering of extending , and let
be the coherent 𝒪_𝓧₁-algebra defining . By the finiteness theorem of
Grauert-Remmert [XII.4, no. 15, th. 1.1], is a coherent -algebra. It
therefore corresponds to a finite covering of , which is normal since is,
and is the desired extension of .
Remark.
In the statement XII.5.4, one cannot remove the hypothesis on the locus of points where the morphism is not étale. For example, let be the unit disk in the complex plane, let be the complement of the origin in , and let
where is the coordinate function on . Then is a normal finite covering of which does not extend to . Indeed, suppose extended to a finite covering of . The locus of points of where the morphism is not étale would then be a closed analytic subset containing all points such that , which is absurd.
One can, however, remove the hypothesis on the singular locus of the morphism when . Indeed, one may suppose regular. The locus of points of where is not étale is a divisor of , and it follows from the Remmert-Stein theorem [XII.9, th. 3] that it is
the trace on of a divisor of . In this case, if is a coherent -algebra such that , and
if is the canonical morphism, it is enough to take ; indeed one knows that is
coherent [XII.11, no. 1, th. 1]. [Translator note: the corrected source fixes the adjective “coherent” to agree with
i_*𝓐.]
Remark (M. Raynaud, added in 2003).
There exist nontrivial finitely presented groups which have no subgroups of finite index distinct from , for example G. Higman’s group; cf. J.-P. Serre, Arbres et amalgames, Astérisque no. 46, prop. 6, ch. I, §1. Consequently, if such a group is realized as the topological fundamental group of a scheme over , say smooth and projective, the topological space underlying is not simply connected, but is algebraically simply connected. At present no such is known. Let us nevertheless mention that D. Toledo constructed smooth projective schemes over whose topological fundamental group is not separated for the topology of subgroups of finite index; the natural morphism is not injective. [D. Toledo, “Projective varieties with non-residually finite fundamental group,” Publ. Math. IHÉS 77 (1993), pp. 103-119.]
Bibliography
- N. Bourbaki, Topologie Générale, Hermann, Paris, 1960.
- N. Bourbaki, Algèbre Commutative, Hermann, Paris, 1961.
- H. Cartan, Séminaire E.N.S., Paris, 1956-57.
- H. Cartan, Séminaire E.N.S., Paris, 1960-61.
- R. Godement, Théorie des Faisceaux, Hermann, Paris, 1958.
- H. Grauert and R. Remmert, “Komplexe Räume,” Math. Ann. 136 (1958), pp. 245-318.
- M. Hakim, Schémas relatifs, thesis, Paris, 1967.
- H. Hironaka, “Resolution of singularities of an algebraic variety over a field of characteristic zero,” Ann. of Math. 39 (1964), pp. 109-236.
- R. Remmert and K. Stein, “Ueber die wesentlichen Singularitäten analytischer Mengen,” Math. Ann. 126 (1953), pp. 263-306.
- J.-P. Serre, “Géométrie algébrique et géométrie analytique,” Ann. Inst. Fourier (Grenoble) 6 (1956), pp. 1-42.
- J.-P. Serre, “Prolongement de faisceaux analytiques cohérents,” Ann. Inst. Fourier (Grenoble) 16 (1966), pp. 363-374.