Abstract
The main results of the paper are two purity theorems for reductive group schemes over regular local rings containing a field. Using these two theorems a well-known Grothendieck-Serre conjecture on principal bundles is reduced to the simply-connected case. We point out that the mentioned reduction is one of the major steps in the proof of the conjecture that the author published in another work.
Bibliography: 25 titles.
Export citation and abstract BibTeX RIS
The author acknowledges the support of the Russian Foundation for Basic Research (grant no. 19-01-00513-а). |
§ 1. Main results
Recall ([3], Exp. XIX, Definition 2.7) that an -group scheme is called reductive (semi-simple, simple) if it is affine and smooth as an -scheme and if, moreover, for each ring homomorphism to an algebraically closed field , its scalar extension is a connected and reductive (semi-simple, simple, respectively) algebraic group over .
We stress that all the groups are connected. Observe also that the class of reductive group schemes contains the class of semi-simple group schemes which in turn contains the class of simple group schemes. This notion of a simple -group scheme coincides with the notion of a simple semi-simple -group scheme from Demazure-Grothendieck [3], Exp. XIX, Definition 2.7 and Exp. XXIV, § 5.3. Throughout this paper denotes an integral noetherian domain and denotes a reductive -group scheme, unless explicitly stated otherwise.
After the pioneering articles [2] and [21] on purity theorems for algebraic groups, various versions of purity theorems were proved in [1], [18], [25] and [13]. The most general result in the so-called constant case was given in [25], § 3.3. The papers [18] and [25] contain results for the nonconstant case and they are proved even if the base field is finite. However they only consider specific examples of algebraic scheme morphisms to a torus . The following theorem covers all the mentioned results of this shape and it looks like the final one.
Theorem 1.1. Let be a field. Let be the semi-local ring of finitely many closed points on a -smooth irreducible affine -variety . Let . Let
be a smooth -morphism of reductive -group schemes, with a torus . Suppose additionally that the kernel of is a reductive -group scheme. Then the sequence
is exact, where runs over the height primes of and each is the natural map (the projection to the factor group).
If the field is infinite, this is Theorem 1.0.1 (Theorem A) in [13].
Let , and be as in Theorem 1.1. Let be a semi-simple -group scheme. Let be a closed subgroup scheme of the centre of . It is known that is of multiplicative type. Let be the factor group and be the projection. It is known that is finite surjective and strictly flat. Thus the sequence of -group schemes
induces an exact sequence of group sheaves in the -topology. Thus for every -algebra the sequence (2) gives rise to a boundary operator
One can check that it is a group homomorphism (cf. [22], Ch. II, § 5.6, Corollary 2). Set
Theorem 1.2. Let , and be as in Theorem 1.1. Then the sequence
is exact, where runs over the height primes of and is the natural map (the projection to the factor group).
If the field is infinite, this is Theorem 1.0.2 in [13].
A well-known conjecture due to Serre and Grothendieck (see [22], Remarque, p. 31, [6], Remarque 3, pp. 26, 27, and [7], Remarque 1.11.a) asserts that, given a regular local ring and its field of fractions and given a reductive group scheme over , the map
induced by the inclusion of into , has a trivial kernel. A survey paper [14] on the topic was published in the proceedings of the 2018 ICM. The Grothendieck-Serre conjecture on principal bundles over a semi-local regular ring containing a field was proved in [17]. That proof is based heavily on Theorem 1.3 stated below and proved in the present paper. We derive Theorem 1.3 from Theorems 1.1 and 1.2.
Theorem 1.3. Let be a field. Let be the semi-local ring of finitely many closed points on a -smooth irreducible affine -variety . Let . Assume that for all semi-simple simply connected reductive -group schemes the pointed set map
induced by the inclusion of into its fraction field has a trivial kernel. Then for any reductive -group scheme the pointed set map
induced by the inclusion of into its fraction field has a trivial kernel.
If the base field is infinite, then this theorem is proved in [13], Theorem 1.0.3. Theorem 1.0.3 in [13] is used in [5] for the proof of the conjecture in the case of regular local rings containing an infinite field.
Remark 1.4. The proof of the latter theorem is subdivided into two steps. First, given a semi-simple -group scheme we prove that the Grothendieck-Serre conjecture holds for an -group scheme , provided it holds for its simply-connected cover and all inner forms of that simply-connected -group scheme .
Second, given a reductive -group scheme we prove that the Grothendieck- Serre conjecture holds for , provided it holds for the derived -group scheme of and for all inner forms of .
The most basic obstacle in any attempt to prove Theorem 1.1 is as follows. Suppose the group , the torus and the morphism are defined over the field . Even under these additional assumptions we are not able to say that the functor is a presheaf with transfers in the sense of Voevodsky [24] or in any other weaker sense. If this were the case, then Theorem 1.1 could be derived in a standard way. Why are we not able to check that the functor is a presheaf with transfers in a weak sense? Because to check this is more or less the same as to check that the morphism from Theorem 1.1 satisfies the norm principle at least for finite separable field extensions. However, the latter is a big open question, which is not even addressed in the present paper.
Here is our approach. We use transfers for the functor , but we do not use at all the norm principle for the homomorphism . Given an element such that its image is an -unramified element, one can find a finite correspondence of the form
(see the diagram (10)) and use it in the 'constant group' case to write down a good candidate (see (17)) for a lift of the element to . In general, since the group-scheme does not come from the ground field, we need to equate two of its pull-backs and over . Here is the projection. We need to do the same with the torus . Due to these requirements our construction of the finite correspondence (6) and of a good candidate (see (18)) is quite involved.
It is especially involved in the case of a finite base field. Of course, we use Bertini-type results which are due to Poonen. Even this does not resolve all the difficulties. Nevertheless, there is a construction of the desired finite correspondence, as for the 'constant group' case, so for the general case. It is done below in Theorem 4.1.
The finite surjective morphism of the -schemes in Theorem 4.1 has the following property: for the corresponding fraction field extension the element
is such that its image is -unramified. The latter yields that the element is constant, that is, belongs to . Thus the evaluations of at and at coincide. Now a simple computation shows that the element is indeed a lift of the element . Details are given in § 7.
It seems plausible to expect a purity theorem in the following context. Let be a regular local ring. Let be a smooth morphism of reductive -group schemes with an -torus . Let be the covariant functor from the category of commutative -algebras to the category of abelian groups given by . Then should satisfy purity for .
The article is organized as follows. In § 2 a theorem about equating certain group scheme morphisms is proved. In §§ 3 and 4 geometric results from [15] are used to prove stronger versions of some results from [16]. In § 5 the construction of norm maps is recalled. In § 6 groups of unramified elements and specialization maps are defined. A homotopy invariance theorem for the group of unramified elements is recalled (see Theorem 6.5). It is proved that in certain cases the norm map takes a specific unramified element to an unramified one (see Lemma 6.7). In § 7 Theorems 1.1 and 1.2 are proved. Finally, in § 8 Theorem 1.3 is proved.
Throughout the paper is a base field.
§ 2. Equating group scheme morphisms
Let be a regular semi-local irreducible scheme such that the residue fields at all its closed points are finite over . Let and be two smooth -group scheme morphisms with -tori and . Suppose that and are reductive -group schemes which are forms of each other and suppose that and are forms of each other. Let be a connected nonempty closed subscheme of , and and be -group scheme isomorphisms. By [16], Theorem 4.1, there exists a finite étale morphism with an irreducible scheme and a section of over and -group scheme isomorphisms
such that and . For these , , , and the following result holds.
Theorem 2.1. Let be the above regular semi-local irreducible scheme. Let be the above connected nonempty closed subscheme of . Suppose that the above morphisms and , and are such that the diagram
commutes. Then as -group scheme morphisms.
Proof. Recall that can be naturally presented as a composition
Since , it remains to check that .
The equality holds by the assumption of the Theorem. It yields the equality . The equality now follows from [16], Proposition 4.7, since is irreducible.
This proves the theorem.
§ 3. Nice triples and group scheme morphisms
See [12], Definition 3.1, for the definition of a nice triple and see [12], Definition 3.2, for the definition of a morphism between nice triples. These definitions are reproduced in [15], Definitions 3.1 and 3.3. The notion of a special nice triple is given in [15], Definition 3.4. We need an extension of Theorem 3.9 in [15] and Theorem 3.9 in [16].
For this it is convenient to give two definitions in the following set up. Let be a field and be the semi-local ring of finitely many closed points on a -smooth irreducible affine -variety . Let . Let be a special nice triple over and be a reductive -group scheme, and let and . Let be a morphism between nice triples over (see [15], Definition 3.3). The latter means that is an étale morphism of -schemes such that , , , where . The following definition is from [16], Definition 4.1.
Definition 3.1. We say that the above morphism equates the reductive -group schemes and if there is an -group scheme isomorphism with .
Further, let be an -torus, and . Let be an -group scheme morphism which is smooth as a scheme morphism. Let , and let be the pull-back of to by means of .
Definition 3.2 (equating morphisms). We say that the morphism equates the reductive -group scheme morphisms and if there are -group scheme isomorphisms
with , and . Clearly, if the morphism equates morphisms and , then it equates with and with .
Remark 3.3. Let and be morphisms of nice triples over . If equates with , then equates and also.
Theorem 3.4. Let be as above in this section. Let be a special nice triple over . Let be a reductive -group scheme and and . Let be an -torus and and . Let be an -group scheme morphism which is smooth as a scheme morphism. Let and be the pull-back of to by means of .
Then there exists a morphism between nice triples over such that
(i) the morphism equates the reductive -group scheme morphisms and ;
(ii) the triple is a special nice triple over subject to the conditions and from Definition 3.7 in [15].
Proof of Theorem 3.4. Let be as in the theorem. Let be the special nice triple over as in the theorem. By the definition of a nice triple there exists a finite surjective morphism of -schemes. Construction 4.2 in [15] now gives us the data , where , and are closed subsets in , finite over . In particular, they are semi-local. If are all closed points of , then . Recall that .
Further, let , , and be as in the hypotheses of Theorem 3.4. Finally, let and be the identity isomorphisms. Recall that by the definition of a nice triple is -smooth and irreducible, hence is regular and irreducible. By Theorem 4.1 in [16] and Theorem 2.1 there exists a finite étale morphism , a section of over and isomorphisms and such that , and
where the scheme is irreducible. Consider now the diagram (6) from Construction 4.2 in [15]. We may and will now suppose that the neighbourhood of the points in that diagram is chosen so that there are -group scheme isomorphisms and with and . Clearly, and . It is less clear, but still true, that
Applying the second part of Construction 4.2 in [15] and Proposition 4.3 in [15] to the finite étale morphism and the section of over we get
- 1)a triple ;
- 2)the étale morphism of -schemes .
Further we get
- (i)the special nice triple over ;
- (ii)that the morphism is a morphism between the nice triples that equates the -group scheme morphisms and ;
- (iii)the equality .
To complete the proof of the theorem just apply Theorem 3.9 in [15] to the special nice triple and use Remark 3.3.
The theorem is proved.
§ 4. An extension of Theorem 6.1 from [16]
Let be an affine -smooth irreducible -variety, and let be closed points in . Let be the semi-local ring . Let and be the canonical embedding. Let be a reductive -group scheme and let be the pull-back of to . Let be an -torus and let be the pull-back of to . Let be an -group scheme morphism which is smooth as an -scheme morphism. Let . The following result is an extension of Theorem 6.1 in [16].
Theorem 4.1. Given a nonzero function vanishing at each point , there is a diagram of the form
with an irreducible affine scheme , a smooth morphism , a finite surjective -morphism , an essentially smooth morphism and a function , which enjoys the following properties:
(a) if is the closed subscheme of defined by the ideal , then the morphism is a closed embedding and the morphism is finite;
(a) and and , where is the zero section of the projection ;
(b) is étale in a neighbourhood of ;
(c) scheme theoretically for some closed subscheme and ;
(d) scheme theoretically for some closed subscheme and ;
(e) for one has ;
(f) there is a monic polynomial with , where the homomorphism 'bar' takes any to ;
(g) there are -group scheme isomorphisms and : with , and .
Remark 4.2. The triple is a nice triple over , since is a finite surjective -morphism.
The morphism is not equal to , since and the morphism is finite.
We stress that usually the scheme from diagram (10) does not coincide with the scheme in [16], Theorem 6.1.
Proof of Theorem 4.1. By [15], Proposition 3.6, one can shrink such that are still in and is affine, and then construct a special nice triple over and an essentially smooth morphism such that , and the set of closed points of is contained in the set of closed points of .
Set ; then . Thus the -group scheme in Theorem 3.4 and the -group scheme defined just above Theorem 4.1 are the same. Set , then . Thus the -group scheme in Theorem 3.4 and the -group scheme defined just above Theorem 4.1 are the same.
Following the notation from Theorem 3.4 write for and for . Let , and let : be the pull-back of to by means of .
By Theorem 3.4 there exists a morphism between nice triples over such that the triple is a special nice triple over subject to conditions and from Definition 3.7 in [15]. Additionally, the morphism equates the reductive -group scheme morphisms and . By Definition 3.2 the latter means that there are isomorphisms
of -group schemes such that , and
The triple is a special nice triple over subject to conditions and from Definition 3.7 in [15]. Thus by [15], Theorem 3.8, there is a finite surjective morphism of the -schemes satisfying conditions (a)–(f) from that theorem. Hence one has a diagram of the form
with the irreducible affine scheme , the smooth morphism , the finite surjective morphism , the essentially smooth morphism and the function .
Put , , , , and . With this new notation diagram (12) becomes a diagram of the form (10) enjoying properties (a)–(f) from Theorem 4.1. The equality (11) shows that the isomorphisms and are subject to the condition (g).
This proves Theorem 4.1.
To formulate a consequence of Theorem 4.1 (see Corollary 4.3), note that using items (b) and (c) of Theorem 4.1 one can find an element such that
- (1);
- (2);
- (3)is étale.
Here is the corollary. It is proved in [15], Corollary 7.2.
Corollary 4.3. The function in Theorem 4.1, the polynomial in item (f) of that theorem, the morphism and the function defined above enjoy the following properties:
(i) the morphism is étale;
(ii) the data satisfy the hypotheses of Proposition 2.6 in [1], that is, is a finitely generated -algebra, the element is not a zero-divisor in and ;
(iii) and ;
(iv) ;
(v) , and .
§ 5. Norms
In §§ 5 and 6 we prove results which will be used to prove Theorems 1.1, 1.2 and 1.3.
Let be field extensions and assume that is finite separable over . Let be a separable closure of and , , be the different embeddings of into . Let be a -smooth commutative algebraic group scheme defined over . One can define a norm map
by . In [13], following Suslin and Voevodsky (see [23], § 6), we generalized this construction to finite flat ring extensions. Let be a finite flat morphism of affine schemes. Suppose that its rank is constant, and equal to . Denote by the th symmetric power of over .
Let be a field. Let be the semi-local ring of finitely many closed points on a smooth affine irreducible -variety . Let be an affine smooth commutative -group scheme. Let be a finite flat morphism of affine -schemes (of constant degree) and be any -morphism. In [13] the norm of is defined as the composite map
Here we write '' for the group law on . The norm maps satisfy the following conditions.
(i) Base change: for any map of affine schemes, putting , we have a commutative diagram
(ii) Multiplicativity: if , then the following diagram commutes
(iii) Normalization: if and the map is the identity, then .
§ 6. Unramified elements
Let be a field and be the semi-local ring of finitely many closed points on a -smooth irreducible affine -variety . Let be the fraction field of , that is, . Let
be a smooth -morphism of reductive -group schemes, with a torus . Suppose additionally that the kernel of is a reductive -group scheme. We work in this section with the category of commutative Noetherian -algebras. For a commutative -algebra set
Let be an -algebra which is a domain and let be its fraction field. Define the subgroup of -unramified elements of as
where is the set of height prime ideals in . Obviously, the image of in is contained in . In most cases injects into and is simply the intersection of all . For an element we write for its image in . In this section we write for the functor (14).
Theorem 6.1 (see [9]). Let be an -algebra which is a discrete valuation ring with fraction field . Then the map is injective.
Lemma 6.2. Let be the above morphism of our reductive group schemes. Let . Then for an -algebra , where is a field, the boundary map is injective.
Proof. For an -rational point set . The action by left multiplication of on makes into a principal homogeneous -space and, moreover, coincides with the isomorphism class of . Now suppose that are such that . This means that and are isomorphic as principal homogeneous -spaces. We must check that for some one has .
Let be a separable closure of . Let be an isomorphism of principal homogeneous -spaces. For any and one has
Thus, for any and any one has
which means that the point is a -invariant point of . So . The following relation shows that the morphism coincides with right multiplication by . In fact, for any one has . Since is right multiplication by , one has , which proves the lemma.
Let , and be as above in this section. Let be a field containing and be a discrete valuation vanishing on . Let be the valuation ring of . Clearly, . Let and be the completions of and with respect to . Let be the inclusion. By Theorem 6.1 the map is injective. We will identify with its image under this map. Set
The inclusion induces a map which is injective by Theorem 6.1. So both groups and are subgroups of . The following lemma shows that coincides with the subgroup of consisting of all elements unramified at .
Proof. We only have to check the inclusion . Let be an element. It determines elements and , which coincide when regarded as elements of . We denote this common element of by . Let and let be the boundary map.
Let , and . Clearly, and both coincide with when regarded as elements of . Thus one can glue and to get a which maps to under the map induced by the inclusion and maps to under the map induced by the inclusion .
Now we show that has the form for some . In fact, observe that the image of in is trivial. By [9], Theorem 4.2, and [8], Theorem 1.1, the map
has a trivial kernel. Therefore, the image of in is trivial. Thus there exists an element with .
We now prove that coincides with in . Since and are both subgroups of , it suffices to show that coincides with the element in . By Lemma 6.2 the map
is injective. Thus it suffices to check that in . This is indeed the case because and , and coincides with when regarded over . We have proved that coincides with in . Thus the inclusion is proved, whence the lemma.
For a regular domain with fraction field and each height prime in we construct a specialization map , where is the residue field of at the prime .
Definition 6.4. Let and be the maps induced by the canonical -algebra homomorphism . Define a homomorphism by , where is a lift of to . Theorem 6.1 shows that the map is well defined. It is called the specialization map. The map is called the evaluation map at the prime .
Obviously for one has .
The following two results are Theorem 5.0.16 in [13] and Corollary 5.0.17 in [13], respectively.
Theorem 6.5 (homotopy invariance). Let be the rational function field in one variable over the field as above in this section. Define by formula (15). Then
Corollary 6.6. Let be the specialization maps at and at (at the primes and ). Then .
Let , and be as above in this section.
Lemma 6.7. Let be a finite extension of -smooth algebras which are domains and each has dimension one. Let and let be such that the induced map is an isomorphism. Suppose for an ideal co-prime to the principal ideal .
Let and be the fields of fractions of and , respectively. Let be such that is -unramified. Then, for the class is -unramified.
Proof. The only primes at which could be ramified are those which divide . Let be one of them. Check that is unramified at .
To do this we consider all primes in lying over . Let be the unique prime dividing and lying over . Then
with . If and are the fields of fractions of and , then
and . We will write for and for . Let
Clearly, for one has and with and . Now coincides with the product
Thus . Let be the inclusion and be the induced map. Clearly, in . Now Lemma 6.3 shows that the element belongs to . Hence is -unramified.
The lemma is proved.
§ 7. Proof of Theorems 1.1 and 1.2
Let be a field. Let be the semi-local ring of finitely many closed points on a -smooth irreducible affine -variety . Let . Let
be a smooth -morphism of reductive -group schemes, with a torus . Suppose additionally that the kernel of is a reductive -group scheme. For each -scheme we write for in this section.
Definition 7.1. Let be an irreducible regular affine -scheme. Let . Its class is called unramified at a height prime ideal of if is in the image of the group . Let be a multiplicative system. The class is called -unramified if it is unramified at any height prime ideal of . In particular, the class is called -unramified if it is unramified at any height prime ideal of .
The following lemma is obvious.
Lemma 7.2. Let be a smooth morphism, where is an irreducible affine scheme. Then this morphism induces an obvious map
which takes -unramified elements to -unramified elements. If is a multiplicative system, then the homomorphism takes -unramified elements to -unramified elements.
Proof of Theorem 1.1. The -algebra is of the form , where is a -smooth irreducible affine variety. In general, , and do not come from . However, clearing denominators, we may assume that , and are defined over , is reductive over , is a torus and is an -group scheme morphism which is smooth. Let be such that its class in is -unramified. Then there is a nonzero function such that the element is defined over , that is, there is an element such that the image of in coincides with . Shrinking we may assume further that vanishes at each . Shrinking further, we may and will assume also that is -unramified. One can choose these shrinkings of so that the resulting scheme remains an affine neighbourhood of the points . By Theorem 4.1 there is a diagram of the form (10) together with a scheme , morphisms , , , and a function , which enjoys properties (a)–(g) from that theorem. From now on and until the end of this proof we use the notation from Theorem 4.1. However, we write for , for and for .
Assume first that is 'constant', that is, there are a reductive group , a torus over the field and an algebraic -group morphism and -group scheme isomorphisms
such that .
The morphism in Theorem 4.1 is finite surjective and the schemes and are regular and irreducible. Thus by a theorem of Grothendieck (see [4], Theorem 18.17) the morphism is flat and finite. Set , where is the restriction of to and set
be the generic points of and , respectively. Then .
Since , this claim completes the proof of Theorem 1.1 in the constant case. To prove the claim consider the scheme and its closed and open subschemes as -schemes via the morphism . Set . Taking the base change of , and via the morphism we get a morphism of the -schemes . Recall that the class is -unramified. By Lemma 7.2 the class is -unramified. Hence its image in is -unramified too.
Items (ii) and (v) of Corollary 4.3 and Lemma 6.7 show that for the element the class is -unramified. By Theorem 6.5 the class is constant, that is, it comes from the field . By Corollary 6.6 its specializations at the -points and of the affine line coincide: .
Properties (c)–(e) and the equality from Theorem 4.1 show that . Thus there is a Zariski open neighbourhood of the -points and in such that . Hence for one has the equality in . Thus
(see the remark at the end of Definition 6.4). By properties (i)–(iii) of the norm maps (see § 5) one has
and
By the base change property of the norm maps one has the equality
Hence in . Finally, the composite map
coincides with the canonical map . Hence , which proves Claim 7.3. This proves Theorem 1.1 in the constant case.
In the general case there are two functors on the category of -schemes. Namely, and . If is a scheme morphism, then , where is regarded as an -scheme via the morphism . On the other hand , where is regarded as a -scheme via the morphism . The -group scheme isomorphisms and from Theorem 4.1 induce a group isomorphism
which respects -scheme morphisms. Moreover, if the scheme is regarded as an -scheme via the morphism , then the isomorphism is the identity. And similarly for any -scheme regarded as an -scheme via the morphism , the isomorphism is the identity. Set , where is as above in this proof. Let be a unique element such that . Set
We leave it to the reader to prove the following claim.
be as above in this proof. Then
Since , this claim completes the proof of Theorem 1.1.
Proof of Theorem 1.2. Just repeat literally the proof of Theorem 10.0.30 in [13], replacing the reference to Theorem 1.0.1 in [13] with one to Theorem 1.1.
§ 8. Proof of Theorem 1.3
Proof of the semi-simple case of Theorem 1.3. Let and be the same as in Theorem 1.3 and assume additionally that is semi-simple. We need to prove that
Let be the corresponding simply-connected semi-simple -group scheme.
Claim 8.1. Under the hypotheses of Theorem 1.3, for all semi-simple reductive -group schemes the map is injective.
In fact, let be two elements such that their images and in are equal. Let and be the corresponding principal -bundles over and be the inner form of the -group scheme corresponding to . The -scheme is a principal -bundle over , which is trivial over . Since is simply-connected semi-simple reductive over , the -scheme has an -point by the hypotheses of Theorem 1.3. Whence the claim.
To finish the proof of the semi-simple case of Theorem 1.3 it remains to repeat literally the arguments from the proof of the semi-simple case of Theorem 1.0.3 in [13], § 11. This proves the semi-simple case of Theorem 1.3.
In turn, the semi-simple case of Theorem 1.3 has the following consequence, which is proved analogously to the proof of Claim 8.1.
Claim 8.2. Under the hypotheses of Theorem 1.3, for all semi-simple -group schemes the map is injective.
The end of the proof of Theorem 1.3. Just repeat literally the arguments from the corresponding part of the proof of Theorem 1.0.3 in [13], § 11.
Acknowledgements
The author thanks A. Suslin for his interest in the topic of the present article. He also thanks A. Stavrova for drawing his attention to Poonen's works [19] and [20] on Bertini-type theorems for varieties over finite fields. He thanks D. Orlov for useful comments concerning the weighted projective spaces tacitly involved in the construction of elementary fibrations. He thanks M. Ojanguren for many inspiring ideas arising from joint works with him [10] and [11].