Exposé IX. Descent of Étale Morphisms. Application to the Fundamental Group

1. Reminders on Étale Morphisms

We shall review here the properties of étale morphisms, developed in Exposé I, that we shall need, taking the opportunity to remove the superfluous noetherian hypotheses from the theory. The reader should note that even if one is interested only in noetherian schemes, descent techniques lead one to introduce non-noetherian schemes, such as , where is a noetherian local ring; and in order to apply the language of fibered categories, it is important to define notions such as étale morphism, etc., without introducing noetherian restrictions. A reader reluctant to verify or admit that the statements below are true without noetherian hypotheses may content himself with admitting them under the noetherian hypotheses of Exposé I, provided that the same noetherian hypotheses are introduced into the statements of the following numbers, and that Definition IX.1.1 below is used for the non-noetherian schemes that enter the arguments.

Definition.

Let be a morphism of preschemes, and let be a point of . We say that is étale at , or that is étale over at , if there exist an affine open neighborhood of , an affine open neighborhood of over , a noetherian affine scheme , an affine -scheme that is étale in the sense of Exposé I, a morphism , and a -isomorphism

V ≃ V₀ ×_U₀ U.

When is locally noetherian, this terminology agrees with that of loc. cit. Similarly, we shall say that is étale, or that is étale over , if is étale at every point of . With these definitions, the propositions below reduce without difficulty to the noetherian case, where they are proved in I.4, I.5, and I.7. For details, the reader may consult EGA IV. [Translator note: more precisely, EGA IV 17 and 18.]

Remarks.

If is étale at , then is “of finite presentation at ” (VIII.3.5), the local ring of in the fiber is a finite separable extension of , and is flat at . One can show that the converse is true; hence Definition IX.1.1 is the same as in the case where is locally noetherian, except that the condition “of finite type at ” must be replaced by “of finite presentation at .” Since this result has a delicate proof, we have not wanted to take it here as the definition of an étale morphism, as it does not lend itself directly to the proof of the properties that follow.

First note the trivial fact:

Proposition.

If is étale, then every morphism obtained from it by base change is also étale.

Thus one can say that étale morphisms form a fibered subcategory of the category of arrows in Sch (cf. VI VI.11 a). The object of the present exposé is the study of the exactness properties of this fibered category over Sch.

Proposition.

Let be a morphism of preschemes. It is an open immersion if and only if it is étale and radicial.

Cf. I.5.1. It follows that if is étale over , every section of over is an open immersion. Using IX.1.4 once more, one obtains:

Corollary.

Let be an étale -prescheme. Then there is a one-to-one correspondence between the set of sections of over and the set of open subsets of such that the morphism induced by the structural morphism is radicial and surjective.

If moreover is separated over , will be a subset of both open and closed, but this is immaterial. Making the evident base change, one can put IX.1.5 in the following apparently more general form:

Corollary.

Let and be two -preschemes, with étale over . Then the map associating to every -morphism from to the subset of underlying the graph of is a bijection from onto the set of open subsets of such that the morphism induced by is radicial and surjective.

Proposition.

Let be the subprescheme of defined by a quasi-coherent nil-ideal, that is, such that has the same underlying set as . Then the functor from the category of preschemes étale over to the category of preschemes étale over is an equivalence of categories.

The fact that this functor is fully faithful is an immediate consequence of IX.1.6. The fact that it is essentially surjective is contained in I.8.3. Notice that under the preceding equivalence, is of finite type, that is, quasi-finite over (respectively finite, that is, an étale covering of ), if and only if satisfies the analogous condition over ; the same remark applies to separatedness. These facts are immediate, and are also contained in IX.2.4 below.

Corollary.

Let be a complete noetherian local ring with residue field . Then the functor is an equivalence from the category of finite étale -algebras to the category of finite étale -algebras, that is, finite direct products of finite separable extensions of .

Proposition.

For to be an étale covering of , that is, finite and étale over , it is necessary and sufficient that be -isomorphic to the spectrum of an Algebra on , which is a locally free Module of finite type, and such that for every

, is a separable algebra over , hence in this case a direct product of finite separable extensions of .

Finally, the following result is less elementary in nature, being the conjunction of I.8.4 and the existence theorem for sheaves in algebraic geometry (EGA III 5; cf. also [IX.1], theorem 3).

Theorem.

Let be the spectrum of a complete noetherian local ring, let be a proper -scheme, and let be the fiber of at the closed point of , so that is a closed subscheme of . Then the restriction functor is an equivalence from the category of étale coverings of to the category of étale coverings of .

2. Submersive and Universally Submersive Morphisms

Definition.

A morphism of preschemes is called submersive if it is surjective and makes a quotient topological space of ; that is, a subset of whose inverse image is open is itself open. One says that is universally submersive if, for every morphism , the morphism deduced from by base change is submersive.

It is immediate that the composite of two submersive (respectively universally submersive) morphisms is submersive (respectively universally submersive), and that a base change of a universally submersive morphism is universally submersive. If fg is submersive (respectively universally submersive), then is so.

Examples.

a) A surjective morphism that is open, or closed, is submersive. Hence a surjective universally closed or universally open morphism is universally submersive. For example, a proper surjective morphism is universally submersive. On the other hand, a faithfully flat and quasi-compact morphism is universally submersive (VIII.4.3). These will be the two most important cases for us.

One can apply to a submersive or universally submersive morphism the arguments of VIII.4.3. In particular one finds:

Proposition.

Suppose submersive. Then the following diagram of maps is exact:

Open(S) → Open(S′) ⇉ Open(S″),

where , and where denotes the set of open subsets of the prescheme .

Proposition.

Let be a universally submersive morphism, let be an -morphism, and let be the -morphism deduced from it by base change. For to be open (respectively closed), it is enough that be so. For to be universally open, respectively universally closed, respectively separated, it is necessary and sufficient that be so. If in addition is quasi-compact and is locally of finite type, then is proper if and only if is proper.

For this last point, note that if is proper, hence quasi-compact, then is quasi-compact (VIII.3.3), hence of finite type since it is locally of finite type. On the other hand it is separated and universally closed by what precedes; therefore it is proper.

Proposition.

Let be a prescheme of finite type over the spectrum of a complete noetherian local ring. Suppose that the fiber over the closed point of is finite, so that the local rings in at the points of this fiber are finite over . Let be the sum scheme of the spectra of the in question, regarded as a finite -scheme. Then is universally submersive if and only if the structural morphism is surjective.

Since there is a natural -morphism , and since a finite surjective morphism is universally submersive by IX.2.2, the stated condition is sufficient. Conversely, suppose is not surjective; we show that is not universally submersive. Let be a point of not in the image of . There then exists an -scheme , the spectrum of a discrete valuation ring, whose image in is {s,t}.

Notice that the image of in is open, because the morphism is a local isomorphism; moreover this image contains and does not meet . It follows that the inverse image of this open subset in is open and identical with the inverse image of the closed point of . This shows that is not submersive, and hence is not universally submersive.

Remark.

Using the criterion IV.6.3 for a constructible subset of a noetherian space to be open, one easily obtains the following valuative criterion for a morphism of finite type, with locally noetherian, to be universally submersive: it is necessary and sufficient that, for every -scheme that is the spectrum of a discrete valuation ring, putting , the inverse image in of the closed point of be non-open.

3. Descent of Morphisms of Étale Preschemes

Proposition.

Let be a surjective morphism of preschemes, let and be two preschemes over , and let , be their inverse images over . If is unramified over , then the canonical map

is injective.

Indeed, by IX.1.6, an -morphism is known once one knows the underlying set of its graph , which is a subset of . Since

is surjective (because is), this subset is known once one knows its inverse image in , which is nothing other than the underlying set of the graph of . This proves the assertion.

A subset of is the graph of an -morphism if and only if it is open and if the morphism induced by from to is radicial

and surjective; cf. IX.1. When the first property is verified, the second is verified if and only if the inverse image of in satisfies the same condition, by VIII.3.1. If one finally knows that is submersive, which will be the case in particular if is universally submersive, then is open if and only if is open.

Thus the set is then in one-to-one correspondence with the set of open subsets of such that the projection morphism is radicial and surjective (that is, corresponding to an -morphism ), and which are saturated for the equivalence relation defined by ; that is, whose two inverse images in , where , by the two projections, are equal. But these latter subsets are the graphs of the two -morphisms deduced from by base change along the two projections . We have therefore obtained:

Proposition.

Let be a universally submersive morphism of preschemes, let , let and be two -preschemes, let and be their inverse images over , and let and be their inverse images over . If is étale over , then the following canonical diagram of maps is exact:

Hom_S(X,Y) → Hom_{S′}(X′,Y′) ⇉ Hom_{S″}(X″,Y″).

Taking and étale over , one obtains the following statement, which moreover gives back IX.3.2, even if one restricts to ; indeed in IX.3.2 one can always reduce to that case by the base change .

Corollary.

A universally submersive morphism of preschemes is a descent morphism for the fibered category of preschemes étale over other preschemes.

I do not know, however, whether it is necessarily a morphism of effective descent for the fibered category in question, even under the additional hypotheses that be noetherian and of finite type, and even restricting to étale coverings. We shall nevertheless give useful criteria of effectivity in the next number.

Corollary.

Let be a universally submersive morphism whose fibers are “geometrically connected,” that is, for every extension , is connected. Then is connected if is. The functor from the category of preschemes étale over to the category of preschemes étale over defined by is fully faithful.

A subset of that is both open and closed is saturated for the set-theoretic equivalence relation defined by , since the fibers are connected; it is therefore the inverse image of a subset of , necessarily both open and closed because is submersive. Thus if is connected, is connected.

This also implies the following fact: the composite fg of two morphisms with universally connected fibers, with universally submersive, has universally connected fibers; if and over have universally connected fibers, the same is true of . In particular, under the conditions of IX.3.4, has universally connected fibers over .

Let and be étale over , and let be an -morphism from to . We prove that it is compatible with the descent data, which gives the desired conclusion by IX.3.3. Let and be the two -morphisms deduced from . The subprescheme of on which and coincide is an induced open subprescheme, fiberwise closed, as the inverse image of the diagonal prescheme of over . [Translator note: the source footnote observes that the fibers of over are separated.] It is therefore the inverse image of a subset of . Since it contains the diagonal in , it is all of . Hence , as required.

4. Descent of Étale Preschemes: Effectivity Criteria

Proposition.

Let be a faithfully flat and quasi-compact morphism. Then is a morphism of effective descent for the fibered category of preschemes étale, separated, and of finite type over other preschemes.

Indeed, it is a descent morphism for the fibered category in question, by IX.3.3 or by VIII.5.2. It remains to show that if is étale, separated, and of finite type over , and endowed with a descent datum relative to , then this datum is effective in the fibered category in question. But one sees easily that if is a prescheme over , then it is étale over if and only if it is étale over , by Definition IX.1.1 and VIII.3.6. Hence it is étale, separated, and of finite type over if and only if is so over ; cf. for example IX.2.4.

It is therefore enough to ensure effectivity of the descent datum on for the fibered category of arrows of Sch. But is quasi-affine over by VIII.6.2 and VIII.6.6. One can then conclude using VIII.7.9. Notice that if one restricts to preschemes étale and finite over others, the proof requires less, since one can invoke VIII.2.1 directly.

Corollary.

Let be a universally submersive morphism, let be an -prescheme étale, separated, and of finite type, endowed with a descent datum relative to , and let be faithfully flat and quasi-compact. Let and be deduced from and by base change, so that is universally submersive and is étale, separated, and of finite type over , endowed with a descent datum relative to . Then the descent datum on is effective if and only if the descent datum on is effective.

This follows from descent theory in categories [IX.D], taking IX.4.1 and IX.3.3 into account.

Similarly one proves:

Corollary.

Let be a universally submersive morphism, let be an -prescheme étale and endowed with a descent datum relative to , and let be an open covering of . Then the descent datum is effective if and only if, for every , the corresponding descent datum on , relative to the morphism , is effective.

This last result leads to a local effectivity criterion:

Proposition.

Let be a morphism of finite presentation (VIII.3.6) and universally submersive, let be a prescheme étale and of finite presentation over , endowed with a descent datum relative to , and let be a point of .

Then there exists an open neighborhood of such that the corresponding descent datum on relative to

is effective if and only if the corresponding descent datum on , relative to

,

is effective.

Necessity is trivial; let us prove sufficiency. We have an étale prescheme of finite type over , and an isomorphism

compatible with the descent data. By a standard and easy sorites on preschemes defined over an inductive limit of rings (here the rings , where is the ring of an affine open neighborhood of , and runs through the elements of not in the prime ideal corresponding to ), one can find an open neighborhood of , an étale prescheme of finite type X_U over , and an -isomorphism . Moreover, after taking small enough, one may suppose that the isomorphism (*) comes from an isomorphism

.

The latter might not be compatible with the descent data; however, after shrinking , it will be compatible with them. This completes the proof.

Corollary.

Under the conditions of IX.4.4, the descent datum on is effective if and only if, for every , the corresponding descent datum on relative to the morphism is effective. When is locally noetherian and is separated over , one may also replace by its completion in the preceding criterion.

The first assertion follows from IX.4.4 and IX.4.3, and the second is then a consequence of IX.4.2. Using IX.4.2 once more, and the fact that for every noetherian local ring one can find a complete noetherian local ring and a local homomorphism such that is flat over and is any prescribed extension of the residue field of , one obtains:

Corollary.

Under the conditions of IX.4.4, suppose in addition that is separated over and that is locally noetherian. Then the descent datum on is effective if and only if, for every prescheme over that is the spectrum of a complete local ring with algebraically closed residue field, the corresponding descent datum on , relative to , is effective.

Theorem.

Let be a finite, surjective morphism of finite presentation (the last hypothesis follows from the others if is locally noetherian). [Translator note: the source footnote says that in fact it is enough for to be integral, by a limit argument in the style of EGA IV 8.] Then is a morphism of effective descent for the fibered category of preschemes étale, separated, and of finite type over other preschemes.

We must show that if is étale, separated, and of finite type over , and endowed with a descent datum relative to , then this datum is effective. Using IX.4.3, one easily reduces to the case where is noetherian. By IX.4.5, one may then suppose is the spectrum of a noetherian local ring, hence in particular

.

We argue by induction on , the assertion being trivial for . Suppose therefore and the theorem proved in dimensions . By IX.4.6 we are reduced to the case where is the spectrum of a complete local ring; then is a finite union of spectra of complete local rings. Hence

,

where is finite over , and has no point above any of the closed points

of . Consider the morphisms

q₁,q₂: X″ ⇉ X′

corresponding to the descent datum, compatible with p₁,p₂: S″ ⇉ S′. One sees at once that

, ,

is the analogous canonical decomposition of over . This implies , and consequently and carry induced descent data.

Let be the open subset of complementary to its closed point. Then is the part of complementary to the set of closed points, and , which lies entirely over , is endowed with a descent datum relative to the morphism induced by . Since the latter is finite surjective and , this descent datum is effective by the induction hypothesis. Thus it remains only to prove that the descent datum on is effective; we may therefore suppose from now on that is étale and finite over . Notice that the induction argument is unnecessary if one restricts statement IX.4.7 to étale coverings.

Let be the spectrum of the residue field of , let , and define , similarly from the fiber squares and cubes and of over . By IX.1.8, the morphisms , , etc. induce equivalences for the categories of étale coverings of and , of and , etc. From the sorites of descent theory in categories [IX.D], it follows that is a morphism of effective descent for the fibered category of étale coverings if and only if the same is true of . But this is indeed the case, for example as a special case of IX.4.1. This completes the proof.

Corollary.

The conclusion of IX.4.7 remains true if one assumes only that is universally submersive, of finite type, and quasi-finite, provided that is locally noetherian.

Indeed, by IX.4.6, one may suppose that is the spectrum of a complete noetherian local ring. Then by IX.2.5, there exists a finite surjective morphism and an -morphism . Since is a strict universal descent morphism for the fibered category under consideration by IX.4.7, and since is a universal descent morphism for it, IX.4.8 follows from the general sorites [IX.D].

Corollary.

Let be a morphism of finite type, surjective and universally open, with locally noetherian. Then is a morphism of effective descent for the fibered category of preschemes étale, separated, and of finite type over other preschemes.

Proceeding as in IX.4.7, one is reduced to the case where is the spectrum of a complete noetherian local ring . Let be a finite -algebra, with spectrum , such that is finite and surjective, hence a universal effective descent morphism for the fibered category under consideration by IX.4.7. It follows from the general theorems [IX.D] that is a morphism of effective descent for that fibered category if and only if the corresponding morphism is so. Since the latter satisfies the same hypotheses as , we are reduced to proving IX.4.9 for in place of .

Taking first for the direct product of the , for the minimal prime ideals of , we are reduced to the case where is integral. One then shows [Translator note: the source refers to EGA IV 14.3.13 and 14.5.4] that there exists an integral subscheme of , quasi-finite over and dominant over , passing through a point of the fiber of over the closed point of . This uses the fact that is universally open of finite type over the integral noetherian local and that is nonempty. Since is complete, is finite over , and since it dominates , the morphism is surjective. Replacing once more by , we are reduced to the case where has a section over , where the statement is trivial.

Theorem.

Let be a finite radicial surjective morphism of finite presentation. The last condition is superfluous if is locally noetherian. [Translator note: the source footnote says it even suffices that be integral, radicial, and surjective, by an easy reduction to the case in the text, in the style of EGA IV 8; cf. SGA 4 VIII 1.1.]

Then the inverse image functor induces an equivalence from the category of preschemes étale over to the category of preschemes étale over .

Since the diagonal morphisms from into and are surjective immersions, they induce, by IX.1.9, equivalences from the categories of preschemes étale over , respectively , with the category of preschemes étale over . It follows from the descent sorites [IX.D] that every étale over is endowed with one and only one descent datum relative to . Hence IX.3.3 implies that the inverse image functor by , from preschemes étale over to preschemes étale over , is fully faithful. It remains to show that it is essentially surjective, that is, that every étale over is isomorphic to the inverse image of an étale over . Since the question is plainly local on and on , one may suppose , , and affine. Then is separated and of finite type over , and one can apply the effectivity criterion IX.4.7.

Corollary.

The conclusion of IX.4.9 remains true if the hypothesis on is replaced by: is faithfully flat, quasi-compact, and radicial.

The proof is the same, invoking IX.4.1 instead of IX.4.7.

Notice that the proof of IX.4.7 is “elementary” in that it does not use the finiteness and comparison theorems for proper morphisms (EGA III 3, 4, 5). This is no longer true of the following result:

Theorem.

Let be a proper, surjective morphism of finite presentation (the last hypothesis follows from the first if is locally noetherian). Then is a morphism of effective descent for the fibered category of étale coverings of preschemes.

By IX.3.3 and IX.2.2, we are reduced to proving that for every étale covering over , endowed with a descent datum relative to , this descent datum is effective. Using IX.4.3, one is easily reduced

to the case where is noetherian; using IX.4.6, one may then suppose that is the spectrum of a complete noetherian local ring . Introduce and as usual, let be the spectrum of the residue field of , and let , , be deduced from , , by the base change , that is, the fibers of , , at the closed point of . By IX.1.10, the morphisms , , etc. induce equivalences from the category of étale coverings over the target scheme with the category of étale coverings over the source scheme. Consequently, is a strict descent morphism for the fibered category of étale coverings of preschemes if and only if is so; this is indeed the case by IX.4.1. This completes the proof of IX.4.12.

In this argument, from IX.1.10 one needed only the fact that the functor considered there is fully faithful, which does not use the existence theorem for coherent sheaves in algebraic geometry.

5. Translation in Terms of the Fundamental Group

Let

be a morphism of effective descent for the fibered category of étale coverings of preschemes, for example a proper, surjective morphism of finite presentation (IX.4.12), or a faithfully flat and quasi-compact morphism. Introducing as usual and , and denoting by , , , the categories of étale coverings of , , , respectively, one has a 2-exact diagram of categories

𝓒 → 𝓒′ ⇉ 𝓒″ ⇉⇉ 𝓒‴

corresponding to the diagram

.

Suppose that the preschemes , , , are disjoint sums of connected preschemes; this is the case in particular if they are locally connected, hence a fortiori if they are locally noetherian (for example if is of finite type over a locally noetherian ). Then the categories , , ... in the preceding diagram are multigaloisian categories (V.9), each described by a collection of totally disconnected compact topological groups, namely the fundamental groups of the connected components of , , , .

For simplicity we suppose connected, and we shall give a calculation procedure for its fundamental group in terms of the fibered category formed from , , , made explicit using the fundamental groups expressing these categories. The reader should note that the sketched procedure is in fact valid in the general setting of multigaloisian categories, which need not come from given preschemes , , , . It is the analogue of the well-known procedure for calculating the fundamental group of a topological space , a locally finite union of closed subspaces (or an arbitrary union of open subspaces ), from the fundamental groups of the components of the and of the components of . Of course, the analogous situation for preschemes fits exactly into the general framework of descent by introducing the prescheme that is the sum of the and the canonical morphism .

Put

, , ,

where denotes the functor “set of connected components.” Since the fiber products of over form a simplicial object of Sch, applying gives a simplicial set whose components in dimensions 0, 1, 2 are , , . We shall use the simplicial maps

, , , .

Objects of will be denoted with a prime, such as ; objects of and will be denoted with double and triple primes. The fact that is connected is expressed by , where is the simplicial set defined by ; equivalently, the equivalence relation in generated by the pair of maps is transitive.

Choose once and for all an element in , and for each in choose an element such that and , thereby displaying the connectedness of . [Translator note: the source footnote warns that such an element need not exist in all cases; the stated theorem must then be modified, though the corollaries remain valid.] For every choose a geometric point underlined in the connected component of ; this point enters in fact through the corresponding fiber functor on the multigaloisian category . The automorphism group of this functor, that is, the fundamental group of at that geometric point, will be denoted . Choose similarly geometric points underlined and underlined , giving fiber functors and and fundamental groups and . Thus

, , .

For every , lies in the same connected component as underlined , so there is an isomorphism of fiber functors

that is, a “path class” from to underlined . The same observation applies to and to the . Choose all these path classes:

, ,

for and . They give in particular group homomorphisms

, .

Finally, recall that the split fibered-category structure with fibers , , also contains isomorphisms of functors

, , ,

deduced from isomorphisms of the two sides respectively with (), where are the three projections from to . Making these data explicit, one finds for every a well-determined element

,

where , , are the three maps defined by , subject to the compatibility conditions such as

,

with , , and the two analogous conditions involving and .

The data just described allow one to reconstruct, up to equivalence of fibered categories, the fibered category with fibers , , . Hence in principle they must allow one to reconstruct up to equivalence, and therefore its fundamental group up to isomorphism. In fact, we shall determine the fundamental group at the geometric point of , that is, the automorphism group of .

An object of is essentially the same as the data of finite sets , for , on which the act continuously.

A gluing datum on such an object is then the giving, for every , of a bijection

compatible with the actions of , acting on the two sides through the homomorphisms . Taking first the of the form , one sees that such data define bijections

,

which allow all the to be identified with the same set , on which all the groups will then act. With this understood, the bijections correspond to bijections

,

subject first to the commutation relations with :

a) for and ,

and also to the relations

b) for ,

which express the way in which we identified the with one another. Making explicit the condition that such a gluing datum is in fact a descent datum gives the relations

c) for .

This gives an equivalence between the category of objects of endowed with a descent datum

and the category of finite sets on which the groups act continuously, endowed in addition with bijections satisfying relations a), b), c). Let be the group generated by the groups and the new generators , subject to relations a), b), c), and let be the inverse limit of the quotients of by subgroups of finite index whose inverse images in the groups are open subgroups. One also says that is the group of galoisian type generated by the and the , subject to relations a), b), c). One checks at once that the category under consideration is also equivalent to the category of finite sets on which the topological group acts continuously. This proves:

Theorem.

Let be a morphism of preschemes that is a morphism of effective descent for the fibered category of étale coverings of preschemes (cf. IX.4.9 and IX.4.12). Suppose connected, and suppose , its fiber square , and its fiber cube are sums of connected preschemes (for example, this holds if is of finite type over a locally noetherian connected ). Choose as above: a geometric point in every connected component of , , ; certain path classes; an ; and for every an whose two images in are and . Here , , denote the sets of connected components of , , respectively. Then the fundamental group of at the geometric point image of is canonically isomorphic to the group of galoisian type generated by the , for , and generators , for , subject to relations a), b), c) above, involving the elements of the groups and the elements , for and , introduced above.

Corollary.

Suppose and have only finitely many connected components, and that the fundamental groups of the connected components of are topologically finitely generated. Then the fundamental group of is topologically finitely generated.

Thus we shall prove later that the fundamental group of a normal projective scheme over an algebraically closed field is topologically finitely generated. Using Chow’s lemma and normalization of algebraic schemes, it will follow that the same result is true for every proper scheme over an algebraically closed field.

Corollary.

Suppose , , have only finitely many connected components, that the fundamental groups of the connected components of are topologically finitely presented, and that the fundamental groups of the connected components of are topologically finitely generated. Then the fundamental group of is topologically finitely presented.

One may express IX.4.9 (restricted to étale coverings) by saying that a finite radicial surjective morphism of noetherian preschemes induces an isomorphism of fundamental groups. Figuratively, one can therefore say that the fundamental group is a topological invariant for preschemes. More generally, with the help of IX.5.1, one can make explicit the effect on the fundamental group of operations on preschemes, such as “pinching” a prescheme along a finite set of points, which have a simple topological meaning. For example:

Corollary.

Let be a finite morphism of finite presentation, and let be a discrete subset of . For every , let be the “geometric number of points” in the fiber , which can also be made explicit as the separable degree of over , the sum of the separable degrees of its residue extensions. Suppose that for one has . For every , let be an algebraically closed extension of , let be the set of geometric points of with values in (a set with elements), let be the complement of one chosen point of , and let be the union of the . Suppose connected. Then the fundamental group of is isomorphic to the group of galoisian type generated by the fundamental group of and generators for , subject to no additional relation.

The details of the proof are left to the reader. The statement obtained is only the translation, into the language of group theory, of the fact that one has an equivalence between the category of étale coverings of and the category of étale coverings of endowed, for every , with a transitive system of bijections between the fibers of at the points of with values in . In this intrinsic form, of course, it is no longer necessary to suppose connected.

Example.

One proves easily that the rational curve over an algebraically closed field is simply connected. [Translator note: the source refers to Expos\acute{e} XI.1.1.] Hence the fundamental group of a complete rational curve having exactly one double point, with analytic branches, is the free group of galoisian type on generators. For example, in the case of an ordinary double point, one finds the fundamental group , as announced in I.11 a). On the other hand, the existence of a cusp (which is a “geometrically unibranch” point) has no influence on the fundamental group.

Corollary.

Let be a universally submersive morphism of preschemes, with geometrically connected fibers, being connected. Then is connected, and, choosing a geometric point in and denoting by its image in , the homomorphism

is surjective. If is a morphism of effective descent for the fibered category of étale coverings of preschemes (cf. IX.4.12), introducing the geometric point of and the two homomorphisms

induced by the two projections, is isomorphic to the cokernel of this pair of morphisms in the category of groups of galoisian type, that is, to the quotient of by the closed normal subgroup generated by the elements

, .

Indeed, by IX.3.4 the functor from étale coverings of to étale coverings of is fully faithful; this is equivalent to saying that the homomorphism on fundamental groups is an epimorphism (V.6.9). The last assertion is an immediate consequence of the description IX.5.1.

Remark.

It is not known at present whether the fundamental group of a proper scheme over an algebraically closed field is topologically finitely presented. [Translator note: the source footnote says this seems very unlikely for smooth curves of genus in characteristic ; for the largest prime-to- quotient, however, known techniques seem to give an affirmative answer, even without properness, and it refers to work in preparation by J.-P. Murre.] Using IX.5.3, a well-known technique of hyperplane sections, and desingularization of normal surfaces, one reduces to the case of a smooth surface over . This at least allows one to show, by transcendental methods, that the answer is affirmative in characteristic 0, without having to assume triangulability of singular algebraic varieties. In characteristic , the main difficulty seems to lie in the case of curves, for which one only knows that the fundamental group is a quotient of the one appearing in the classical case (cf. the next exposé), but the kernel by which one divides is very poorly known.

Remark.

One could make explicit other special cases besides IX.5.4 and IX.5.6 in which IX.5.1 takes a particularly simple form. An interesting case is that in which is the quotient of by a finite group of automorphisms. Then the category of étale coverings of is equivalent to the category of étale coverings of on which acts compatibly with its action on , in such a way that for every and every (where denotes the inertia group of in ), acts trivially on the fiber .

If is connected, this statement is interpreted as follows. Let be the category of étale coverings of on which acts compatibly with its action on , without necessarily satisfying the preceding condition on inertia groups of points of . One sees easily that this is a galoisian category (V.5), and that for every geometric point of , the fiber functor on is a fundamental functor. Let be the automorphism group of this functor, with its usual topology. Then there is a canonical exact sequence

.

Moreover, for every geometric point of , one has an isomorphism , defined up to inner automorphism coming from . Since maps evidently into the first group, one obtains a homomorphism

,

defined up to inner automorphism coming from , whose composite with is the canonical immersion . With this understood, the fundamental group is canonically isomorphic to the quotient group of by the closed normal subgroup generated by the images of the homomorphisms . In particular, the image of in is a normal subgroup, and the corresponding quotient is isomorphic to a quotient of .

One can reduce the number of “relations” introduced as follows. For every , , introduce the subprescheme where the automorphisms and coincide; choose a geometric point in each connected component of , then one of the corresponding homomorphisms , giving lifts of in . It is enough to take the quotient of by the closed normal subgroup generated by the .

When is fixed by , acts naturally on , and identifies with the corresponding semidirect product. Identifying with a subgroup of , one sees that among the relations introduced above, taking gives “” for in the inertia group. Therefore if has a geometric point fixed by (that is, a point whose inertia group is ), then is a quotient of the quotient group of galoisian type of obtained by making the actions of on trivial; it is even isomorphic to this latter group if, for every , the inertia locus is connected, hence passes through the locality of .

This last assertion is contained in the second description above of the relations to be introduced in .

This result applies in particular if one takes to be the cartesian power of a connected prescheme over an algebraically closed field, to be the symmetric group acting in the usual way, and to be the -th symmetric power of . Taking to be a geometric point localized on the diagonal, one is under the preceding conditions, since all inertia loci contain the diagonal. Using the fact, proved in the next exposé, that if is proper and connected over , the fundamental group of identifies with , one obtains the following amusing result: If is proper and connected over an algebraically closed , the fundamental group of its -th symmetric power, , is isomorphic to the abelianization of the fundamental group of . I do not know whether the analogous fact in algebraic topology is known; it should be provable by the same descent method. Taking for example to be the rational curve , one obtains yet another proof that is simply connected, using the fact that is. Taking now to be a nonsingular curve over , and , so that is fibered over the Jacobian with projective-space fibers, and hence, as will be seen using the results of the next two exposés, has the same fundamental group as , one recovers without dévissage the well-known fact that the fundamental group of the Jacobian of is isomorphic to the abelianization of the fundamental group of .

6. A Fundamental Exact Sequence. Descent by Morphisms with Relatively Connected Fibers

Theorem.

Let be the spectrum of an artinian ring with residue field , let be an algebraic closure of , let be an -prescheme, , , let ā be a geometric point of , let be its image in , and let be its image in . Suppose is quasi-compact and geometrically connected over . (If is proper over , this means that is a local artinian ring with residue field radicial over .) Then the sequence of canonical homomorphisms

is exact, and one has

= the Galois group of over .

Since fundamental groups do not change after killing nilpotents, one may suppose , which already makes the last isomorphism evident. Let be the separable closure of in , and consider and the image of ā in . One has a canonical sequence

,

where . There is also a canonical homomorphism from this sequence to the corresponding sequence for , coming from the evident diagram. This homomorphism of sequences is an isomorphism by IX.4.11. We are therefore reduced to proving that the second sequence is exact, that is, we may suppose perfect.

Let be the finite Galois subextensions of in , put , and let be the image of ā in . The reader may verify that the natural homomorphism

is an isomorphism. This simply means that an étale covering of comes from an étale covering of some , and that the latter is essentially unique after passing to an with .

On the other hand, let be the Galois group of over , that is, the opposite group of the group of -automorphisms of . Since the functor from étale coverings of to étale coverings of is fully faithful by IX.3.4, it follows that is also isomorphic to the opposite of the group of -automorphisms of the connected principal covering of . Hence by V.6.13 one has an exact sequence

.

Passing to the inverse limit over in these exact sequences gives an exact sequence, since we are in the category of groups of galoisian type, and this is precisely the sequence considered in IX.6.1.

The geometric translation of right exactness in IX.6.1 is the following:

Corollary.

With the preceding notation, let be an étale covering of , and let be the corresponding étale covering of . The following conditions are equivalent:

  1. There exists an étale over and an -isomorphism . The is then determined up to unique isomorphism by IX.3.4.
  2. is completely decomposed over .

If is connected, these conditions are also equivalent to:

2 bis. has a section over .

This last supplement is essential: the equivalence of 1 and 2 means only that is the quotient group of by the closed normal subgroup generated by the image of , and not by this image itself. Under the preceding conditions, we shall say that is a geometrically trivial covering of .

Remark.

In IX.6.1 one cannot replace by an arbitrary algebraically closed extension of , even if is already assumed algebraically closed. In other words, it is not generally true that if is a connected algebraic scheme over an algebraically closed field , its fundamental group is unchanged after replacing by an algebraically closed extension. This already fails, for instance, in characteristic for the affine line over , because of “higher ramification” phenomena at the point at infinity, which imply a “continuous” structure for the fundamental group. We shall see in the next exposé, however, that such phenomena cannot occur if is proper over . We shall also show by transcendental methods that the same is true if has characteristic zero.

Corollary.

Suppose that is localized at an that is rational over (or more generally has residue field radicial over ). Then the exact sequence IX.6.1 is split.

One may suppose . If is rational over , it corresponds to a section of over , sending to and defining a homomorphism that is the required splitting. If is radicial over , one reduces to the preceding case by the base extension .

Theorem.

Let be a proper and surjective morphism of finite presentation, with geometrically connected fibers; let be a prescheme of finite presentation and proper over ; let be a point of ; let be the fiber of at ; and let be a connected component of the fiber of at . There exists an open neighborhood of in , an -scheme étale over , and an -isomorphism if and only if is étale over at the points of and is a geometrically trivial covering of .

Necessity is trivial, so it remains to prove sufficiency. One reduces easily to the case where is noetherian. Consider the Stein factorization of , where is the spectrum of the Algebra on . Since the fibers of over are geometrically connected and is surjective, the morphism is finite, surjective, and radicial; hence by IX.4.10 every étale over comes by inverse image from an étale over . This reduces us to proving IX.6.5 with replaced by , that is, to the case where .

Consider then the Stein factorization of the proper morphism , where is the spectrum of the Algebra . The morphisms and define a canonical morphism

,

and our assertion is contained in the following:

Corollary.

Let be a proper morphism of locally noetherian preschemes such that , and let be a prescheme proper over . Consider the Stein factorization for and the canonical morphism . Let be a point of , and let be a point of above , corresponding to a connected component of the fiber of at . The morphism is an isomorphism above an open neighborhood of étale over if and only if is étale over at the points of and is a geometrically trivial covering of the fiber .

Necessity is again trivial; it remains to prove sufficiency. The conclusion also says that a) the morphism deduced from by base change is an isomorphism, and b) is étale over at , that is, is étale over . In this form, the conclusion is invariant under the base change . Since the hypotheses are likewise stable under this base change, one may suppose is the spectrum of a complete noetherian local ring. One may also plainly suppose connected, which here implies and .

Since the set of points of at which is étale over is open and contains the fiber , and since is proper over , it follows that is étale over . Since it induces on an étale covering isomorphic to , where is étale over , IX.1.10 implies that it is isomorphic to a covering of the form , with étale over . Here again it is enough to use full faithfulness of the functor in IX.1.10, which follows from the fact that a formal isomorphism of coherent sheaves on comes from an isomorphism of those sheaves.

Thus, if is defined by the finite -algebra , identifies with the spectrum of the Algebra over . Since , it follows at once that is defined by , hence the canonical homomorphism is precisely the isomorphism under consideration. This completes the proof.

Corollary.

Under the conditions of IX.6.5, there exists a prescheme étale over and an -isomorphism if and only if is étale over and for every , the fiber is a geometrically trivial covering of .

Indeed, if this holds, is the union of open subsets that are isomorphic to inverse images of étale over . One then sees easily that these glue to an étale over , and that one obtains an isomorphism . For example, one may say that the carry descent data relative to , which necessarily glue to a descent datum on all of relative to ; since this datum is effective on the , it follows easily (by a sorites omitted in no. IX.4) that it is effective. One can also state IX.6.7 as follows:

Corollary.

Let be a proper surjective morphism of finite presentation, with geometrically connected fibers. Then is a morphism of effective descent for the fibered category of preschemes finite étale over other preschemes. The functor induces an equivalence from the category of preschemes finite étale over to the category of preschemes

finite étale over that induce on each fiber a geometrically trivial covering.

Remark.

Let be a proper and surjective morphism, with locally noetherian. Then factors as a morphism satisfying the hypothesis of IX.6.8 followed by a finite surjective morphism covered by IX.4.7. Thus is a composite of two morphisms that are universal effective descent morphisms for the fibered category of preschemes finite étale over other preschemes. It follows that itself is a universal effective descent morphism for the fibered category in question. This recovers IX.4.12 by a different method.

Remark.

The conclusion of IX.6.7 does not remain valid if the hypothesis that is proper is replaced by: is of finite type over and admits a section over (so is universally submersive and a descent morphism for the fibered category of preschemes étale over other preschemes), even when is the spectrum of a discrete valuation ring and is an étale covering of . To see this, start with a proper over whose generic fiber is a nonsingular rational curve and whose special fiber consists of two intersecting lines. For example, if is a uniformizer of the valuation ring , take the closed subscheme of defined by the homogeneous equation . Let be the complement of the singular point of in the union . The fibers of are and , hence geometrically simply connected, meaning that every étale covering of such a fiber is geometrically trivial.

However, proceeding as in no. IX.4, one easily constructs étale coverings of that do not come from étale coverings of , by gluing trivial coverings of and of . It is possible, on the other hand, that the conclusion of IX.6.7 remains true if the properness hypothesis is replaced by the hypothesis that be universally open of finite presentation over . [Translator note: the source adds that this is now proved, with g only universally open and surjective; cf. SGA 4 XV 1.15.] This is at least true if the fibers of over are geometrically irreducible, and not merely geometrically connected. We only point out that in this question one can reduce to the case where is the spectrum of a complete discrete valuation ring with algebraically closed residue field.

The interpretation of IX.6.7 in terms of the fundamental group is the following:

Corollary.

Let be a proper surjective morphism of finite presentation, with geometrically connected fibers. Suppose , hence , connected. Let be a geometric point of , its image in , and for every choose an algebraic closure of , a geometric point of with values in this extension, and a path class from to . This gives a homomorphism

,

where . Then the homomorphism is surjective, and its kernel is the closed normal subgroup of generated by the images of the .

Remark.

Under the conditions of IX.6.7, assuming noetherian, one sees easily that the set of points such that the corresponding fiber is geometrically trivial over is constructible; if is proper over , it is even open, as one sees from IX.6.6. Thus, if is a Jacobson prescheme (for example of finite type over a field), or if is proper over ,

it is enough, in order to verify the conditions of IX.6.7, to restrict to the closed points of . Likewise, in IX.6.11 it is then enough to take the for the closed points of .

Bibliography

[IX.D] J. Giraud, Méthode de la descente, Mémoire no. 2 de la Société Mathématique de France, 1964.

[IX.1] A. Grothendieck, Géométrie algébrique et Géométrie formelle, Séminaire Bourbaki, vol. 11, 1959, no. 182.