Difference Galois theory of linear differential equations
aa r X i v : . [ m a t h . A C ] A p r Difference Galois theory of linear differential equations
Lucia Di Vizio, Charlotte Hardouin and Michael WibmerSeptember 19, 2018
Abstract
We develop a Galois theory for linear differential equations equipped with the action ofan endomorphism. This theory is aimed at studying the difference algebraic relations amongthe solutions of a linear differential equation. The Galois groups here are linear differencealgebraic groups, i.e., matrix groups defined by algebraic difference equations.
Introduction
Linear differential equations with coefficients in a differential field (
K, δ ) and their behavior underthe action of an endomorphism σ of K are a frequent object of study. Let us start with someclassical examples. For instance, one can consider the field K = C ( α, x ) of rational functions inthe variables α, x and equip K with the derivation δ = ddx and the endomorphism σ : f ( α, x ) f ( α + 1 , x ). The Bessel function J α ( x ), which solves Bessel’s differential equation x δ ( y ) + xδ ( y ) + ( x − α ) y = 0satisfies the linear difference equation xJ α +2 ( x ) − α + 1) J α +1 ( x ) + xJ α ( x ) = 0 . Contiguity relations for hypergeometric series provide a large class of examples in a similar spirit.(See for instance [WW88, Chapter XIV].)Another occasion, where a linear differential equation comes naturally equipped with the actionof an endomorphism arises in the p -adic analysis of linear differential equations, when consideringFrobenius lifts. For example, let p be a prime number and let us consider the field C p with its norm | | , such that | p | = p − , and an element π ∈ C p verifying π p − = − p . Following [DGS94, ChapterII, §
6] the series θ ( x ) ∈ C p [[ x ]], defined by θ ( x ) = exp( π ( x p − x )), has a radius of convergencebigger than 1. Therefore it belongs to the field E † C p , consisting of all series P n ∈ Z a n x n with a n ∈ C p such that • ∃ ε > ∀ ρ ∈ ]1 , ε [ we have lim n →±∞ | a n | ρ n = 0 and • sup n | a n | is bounded.One can endow E † C p with an endomorphism σ : P n ∈ Z a n x n P n ∈ Z a n x pn . (For the sake ofsimplicity we assume here that σ is C p -linear.) The solution exp( πx ) of the equation δ ( y ) = πy ,where δ = ddx , does not belong to E † C p , since it has radius of convergence 1. Moreover, exp( πx ) isa solution of an order one linear difference equation with coefficients in E † C p , namely: σ ( y ) = θ ( x ) y. Lucia Di Vizio, Laboratoire de Math´ematiques UMR8100, UVSQ, 45 avenue des ´Etats-Unis 78035 Versaillescedex, France. e-mail: [email protected] .Charlotte Hardouin, Institut de Math´ematiques de Toulouse, 118 route de Narbonne, 31062 Toulouse Cedex 9,France. e-mail: [email protected] .Michael Wibmer, RWTH Aachen, Templergraben 64, 52062 Aachen, Germany. e-mail: [email protected] .Work supported by project ANR-2010-JCJC-0105 01 qDIFF. α α + 1 orthe Frobenius operator considered in the p -adic example above, remain out of reach. An approachto difference algebraic relations among solutions of linear difference equations, in the spirit of thisarticle can be found in [OW].The above described theories have been applied in various areas, for example in questions ofintegrability and isomonodromy ([MS12], [MS13], [Dre]) or in combinatorial problems ([CS12]).A prototypical application is to show that certain special functions are independent in a strongsense. For example, [HS08] provides a Galoisian proof of H¨olders theorem, stating that the Gammafunction, which satisfies the linear difference equation Γ( x + 1) = x Γ( x ), does not satisfy analgebraic differential equation over C ( x ). We refer to [DV12] for an overview of the Galoisianapproach to differential transcendence of special functions.The Galois theories in [CS06], [HS08] and [Lan08] study differential algebraic relations withrespect to a finite number of commuting derivations, i.e., partial differential equations are consid-ered. In this article we only study difference algebraic relations with respect to one endomorphism,i.e., only ordinary difference algebraic equations are considered. Certain aspects of our theory, e.g.the Galois correspondence, will generalize to the partial case in a straight forward manner. How-ever, other aspects, e.g. existence and uniqueness of Picard-Vessiot extensions appear to be morechallenging.Given the well-known analogy between difference and differential algebra, it might at first sightseem a rather straight forward matter to pass from differential algebraic relations to differencealgebraic relations. However, a closer look reveals quite the contrary. Indeed, some of our resultsactually differ from the statements anticipated by naive analogy. For example, a σ -Picard-Vessiotextension (a suitable minimal field containing all solutions and their transforms under σ ) for agiven linear differential equation is not unique, it is only unique up to powers of σ .There is a certain class of results in differential algebra whose difference analogs simply fail.For example, a difference ideal which is maximal in the set of all proper difference ideals of a givendifference ring, need not be prime. This jeopardizes the classical construction of a Picard-Vessiotextension by taking the quotient by some suitable ideal in the universal solution ring. So genuinelynew ideas are needed. Moreover, certain results, well-known in differential algebra, have not beenavailable in the literature for the difference case. For example, Kolchin’s theory of constrainedextensions ([Kol74]) plays a crucial role in the developments in [CS06] and [HS08]. A differenceanalog has been made available only recently by the third author in [Wib12a].2n the classical Picard-Vessiot theory of linear differential equations, it is usually assumed thatthe field k = { a ∈ K | δ ( a ) = 0 } of all constant elements in the base differential field ( K, δ ) isalgebraically closed. The reason for this is twofold:(i) It is needed to establish the existence and uniqueness of a Picard-Vessiot extensions L | K fora given differential equation δ ( y ) = Ay , A ∈ K n × n .(ii) Identifying a linear algebraic group G over k with its k -rational points G ( k ), we can makesense of the statement that the group Aut δ ( L | K ) of differential automorphism of L | K is alinear algebraic group over k , by identifying Aut δ ( L | K ) with a subgroup of Gl n ( k ) via thechoice of a fundamental solution matrix for δ ( y ) = Ay in L .If k is not algebraically closed the existence and uniqueness can not be guaranteed in all gener-ality but there are relative versions: After a finite algebraic extension of the constants, there alwaysexists a Picard-Vessiot extension. Similarly, two Picard-Vessiot extensions become isomorphic aftera finite base extension of the constants.Concerning (ii), Aut δ ( L | K ) can still be identified with the k -rational points G ( k ) of a linearalgebraic group G , but as k is not algebraically closed, G may contain significantly more informationthan G ( k ). The Galois correspondence no longer holds in the naive sense. For example, not everyelement of L , which is fixed by Aut δ ( L | K ), must lie in K . It is well known that this defect can beavoided by employing an appropriate functorial formalism ([Tak89, Appendix], [Dyc], [Del90]). Allthe expected results can be restored by considering not only the action of G ( k ) on the solutions of δ ( y ) = Ay , but also the action of G ( S ), where S is an arbitrary k -algebra.If we are interested not only in algebraic relations among the solutions of δ ( y ) = Ay , but indifferential algebraic relations or difference algebraic relations, then, instead of assuming that the δ -constants k are algebraically closed, it is natural to assume that k is a differentially closed ordifference closed field. While the algebraically closed field of complex numbers comes up in thetheory of linear differential equations rather naturally, differentially closed fields or difference closedfields are not the kind of objects a mathematician will encounter on a daily basis. We rather seethem as a tool to make certain things work: It is sometimes convenient to work inside these largefields so that you do not need to pass to extensions when performing certain constructions.However, for the applications of the theory, the assumption that the δ -constants are differen-tially or difference closed is somewhat of a hindrance: One has to first extend the field of constants,then apply the theory, and finally find some usually rather ad-hoc descent arguments to get backto the situation originally of interest.In this article we completely avoid the assumption that the δ -constants are difference closed.We provide relative versions of the existence and uniqueness theorem and we employ a functorialformalism for difference algebraic groups, allowing us to establish the Galois correspondence andrelated results in maximal generality. We think that, based on the approach of this article, it willbe a straight forward matter to also remove the assumption of differentially closed constants from[CS06] and [HS08].It is well-known that a functorial-schematic approach to algebraic groups has many benefits andintroduces genuinely new phenomena, which are not visible over the algebraic closure. For example,over a field k of characteristic p >
0, the center of Sl p is µ p , the group scheme of p -th roots of unity,whereas the center of Sl p ( k ) is trivial. A similar phenomenon occurs for difference algebraic groups,even in characteristic zero, since “difference nilpotence” is not restricted to positive characteristic.If G is a difference algebraic group over a difference field k , then, even if k is difference closed, G might contain significantly more information than G ( k ). For example, let G , G denote thedifference algebraic subgroups of Gl ,k given by G ( S ) = { g ∈ S × | g = 1 } ≤ Gl ( S )and G ( S ) = { g ∈ S × | g = 1 , σ ( g ) = g } ≤ Gl ( S ) . Here S is any difference algebra over k , i.e. a k -algebra, equipped with an endomorphism σ : S → S which extends σ : k → k . Then G ( k ) = { , − } = G ( k ), but of course G = G . Moreover, if the3 -Galois group of δ ( y ) = ay is G , then g (cid:18) σ ( y ) y (cid:19) = σ ( gy ) gy = σ ( y ) y for every g ∈ G ( S ) and every k - σ -algebra S . So, by the Galois correspondence, σ ( y ) y lies in thebase field. In other words, y satisfies a difference equation of the form σ ( y ) = by . This simpleand important difference algebraic relation, expressing the σ -integrability of δ ( y ) = ay , is notdetected by G ( k ). As illustrated in [DVHW], an equation δ ( y ) = ay with σ -Galois group G isnot σ -integrable. So we might loose a lot of valuable information about the differential equationif we replace its Galois group G by G ( k ). This is another reason why, in our opinion, a functorialapproach is indispensable. However, in Section 4.1, we explain in all detail what the outcome willbe if one really wants to insist to work with G ( k ) instead of G .The theory of difference algebraic groups is still in its infancy. In the context of groups definablein ACFA, certain groups defined by algebraic difference equations have played a quite crucial rolein some of the recent applications of model theory to number theory. (See the appendix forreferences.) However, there is no coverage of foundational results concerning difference algebraicgroups in the literature which fits our needs. We have therefore collected certain basic aspectsof the theory of difference algebraic groups in an appendix. This appendix will also be used in[OW], but in view of potential applications of difference algebraic groups beyond σ -Galois theory,we hope that this appendix will serve other purposes as well.Our main motivation to initialize this Galois theory was the creation of a versatile tool for thesystematic study of the difference algebraic relations among the solutions of a linear differentialequation. The application of our theory to this problem follows the usual paradigm: Since thedifference algebraic relations among the solutions of a linear differential equation are governed by adifference algebraic group, they must follow a rather restricted pattern. Using structure theoremsfor difference algebraic groups one is often able to elucidate this pattern. In Section 2, we showthat the Zariski closure of the Galois group (in the sense of our theory) of a linear differentialsystem δ ( y ) = Ay agrees with the Galois group G of δ ( y ) = Ay in the sense of classical Picard-Vessiot theory. Thus, if we can classify the Zariski dense difference algebraic subgroups of G , weobtain, via our Galois theory, a classification of the possible difference algebraic relations amongthe solutions of δ ( y ) = Ay . Such classifications are available for tori G nm (Lemma A.40), vectorgroups G na and the semidirect product G a ⋊ G m . Thanks to [CHP02], such a classification isalso available for almost simple algebraic groups. The applications of these structure theorems fordifference algebraic groups to the study of difference algebraic relations among the solutions of alinear differential equation will be presented in [DVHW]. Exemplarily, let us state here a resultwhich corresponds to the group G m and applies to the field C ( x ) of rational functions. Corollary 0.1 ([DVHW, Cor. 3.11]) . Let L be a field extension of C ( x ) eqipped with a derivation δ and an endomorphism σ such that δσ = σδ and { a ∈ L | δ ( a ) = 0 } = C . Assume that therestriction of δ to C ( x ) equals ddx and that σ ( f ( x )) = f ( x + 1) for f ∈ C ( x ) . If z ∈ L satsfies δ ( z ) = az with a ∈ C ( x ) × , then z is transformally dependent over C ( x ) , i.e., z, σ ( z ) , σ ( z ) , . . . arealgebraically dependent over C ( x ) , if and only if there exist P ∈ C [ x ] , f ∈ C ( x ) × and N ∈ Z × suchthat a = P + N δ ( f ) f . In Theorem 5.11 of [DVHW] we also show how the discrete integrability of a linear differentialequation can be characterized through our new Galois group. Combining this with the structure ofthe Zariksi dense difference algebraic subgroups of almost simple algebraic groups yields a methodto show that certain special function are independent in the sense of difference algebra. Forexample:
Corollary 0.2 ([DVHW, Cor. 6.10]) . Let A ( x ) and B ( x ) be two C -linearly independent solutionsof Airy’s equation δ ( y ) − xy = 0 . Then A ( x ) , B ( x ) and δ ( B )( x ) are transformally independent,i.e., the functions A ( x ) , B ( x ) , δ ( B )( x ) , A ( x +1) , B ( x +1) , δ ( B )( x +1) , A ( x +2) , . . . are algebraicallyindependent over C ( x ) .
4e now describe the content of the article in more detail. In Section 1 we introduce σ -Picard-Vessiot extensions and σ -Picard-Vessiot rings. These are the places where the solutions to our lineardifferential equations live. The basic questions of existence and uniqueness of a σ -Picard-Vessiotextension for a given linear differential equation have already been addressed by the third authorin [Wib12a]. So, concerning these questions, we largely only recall the results from [Wib12a].In Section 2, we introduce and study the σ -Galois group G of a σ -Picard-Vessiot extension L | K . It is a difference algebraic group over the difference field k of δ -constants of the base δσ -field K . Roughly speaking, the σ -Galois group consists of all automorphisms of the solutions whichrespect δ and σ . The choice of a fundamental solution matrix in L determines an embedding G ֒ → Gl n,k of difference algebraic groups. We also compute the σ -Galois group in some simple andclassical examples. Concerning applications, the most important result here is that the differencetranscendence degree of the σ -Picard-Vessiot extension agrees with the difference dimension ofthe σ -Galois group. As it is also illustrated in [DVHW], this allows one to reduce questions ofdifference transcendence to questions about difference algebraic groups.In Section 3, we establish the analogs of the first and second fundamental theorem of Ga-lois theory. Here we employ some ideas of the Hopf-algebraic approach to Picard-Vessiot theory([AMT09]).In Section 4, we present some refinements of our σ -Galois correspondence. We show how certainproperties of the σ -Galois group are reflected by properties of the σ -Picard-Vessiot extension. Mostnotably, this concerns the property of the σ -Galois group to be perfectly σ -reduced. Perfectly σ -reduced σ -schemes correspond to what goes under the name “difference variety” in the classicalliterature [Coh65] and [Lev08].In the first few sections of the appendix we have collected some basic results pertaining tothe geometry of difference equations. Then we present some foundational aspects of the theory ofdifference algebraic groups in a way that is suitable for the main text.We are grateful to Phyllis Cassidy, Zo´e Chatzidakis, Shaoshi Chen, Moshe Kamensky, AkiraMasuoka, Alexey Ovchinnikov and Michael Singer for helpful comments. We would also like toacknowledge the support of CIRM, where part of this work was conducted. σ -Picard-Vessiot extensions and σ -Picard-Vessiot rings In this section, we introduce the notions of σ -Picard-Vessiot extension and σ -Picard-Vessiot ringfor a linear differential equation. We establish some first properties of these minimal solution fields,respectively rings and show that the existence of a σ -Picard-Vessiot extension can be guaranteedunder very mild restrictions. We also show that a σ -Picard-Vessiot extension for a given differentialequation is essentially unique.Before really getting started, let us agree on some conventions: All rings are commutative withidentity and contain the field of rational numbers. In particular all fields are of characteristiczero. A differential ring (or δ -ring for short) is a ring R together with a derivation δ : R → R . Adifference ring (or σ -ring for short) is a ring R together with a ring endomorphism σ : R → R . Wedo not assume that σ is an automorphism or injective. A σ -ring with σ injective is called σ -reduced.If σ is an automorphism the σ -ring is called inversive. The δ -constants are R δ = { r ∈ R | δ ( r ) = 0 } and the σ -constants are R σ = { r ∈ R | σ ( r ) = r } .The basic algebraic concept that facilitates the study of difference algebraic relations amongsolutions of differential equations is the notion of a δσ -ring. A δσ -ring is a ring R , that is simul-taneously a δ and a σ -ring such that for some unit ℏ ∈ R δ δ ( σ ( r )) = ℏ σ ( δ ( r )) (1.1)for all r ∈ R . If ℏ = 1, then σ and δ commute. The element ℏ is understood to be part of the dataof a δσ -ring. So a morphism ψ : R → R ′ of δσ -rings is a morphism of rings such that ψσ = σ ′ ψ , ψδ = δ ′ ψ and ψ ( ℏ ) = ℏ ′ . The reason for not simply assuming ℏ = 1 is that this factor appears insome examples of interest. See Example 1.1 below. Note that condition (1.1) implies that R δ is a σ -ring. 5e refer the reader to Section A.1 in the appendix for an exposition of some basic notions indifference algebra. We largely use standard notations of difference and differential algebra as canbe found in [Coh65], [Lev08] and [Kol73]. For the convenience of the reader we recall the basicconventions: Algebraic attributes always refer to the underlying ring. For example a δσ -field isa δσ -ring whose underlying ring is a field. By a K - δσ -algebra R over a δσ -field K one meansa K -algebra R that has the structure of a δσ -ring such that K → R is a morphism of δσ -rings.Similarly for δ or σ instead of δσ . An extension of δσ -fields is an extension of fields such that theinclusion map is a morphism of δσ -rings. If R is a K - σ -algebra over a σ -field K and B a subsetof R then K { B } σ denotes the smallest K - σ -subalgebra of R that contains B . If R = K { B } σ forsome finite subset B of R we say that R is finitely σ -generated over K . If L | K is an extension of σ -fields and B ⊂ L then K h B i σ denotes the smallest σ -field extension of K inside L that contains B . Example 1.1.
Some basic examples of δσ -fields of interest for us are the following: • The field K = C ( x ) of rational functions in one variable x over the field of complex numbersbecomes a δσ -field by setting δ := ddx and σ ( f ( x )) := f ( x + 1) for f ∈ C ( x ). We have ℏ = 1and K δ = C .One can also take δ := x ddx and σ ( f ( x )) := f ( qx ) for some q ∈ C r { } . Again we have ℏ = 1and K δ = C .If we set δ := x ddx and σ ( f ( x )) := f ( x d ) for some integer d ≥
2. Then K becomes a δσ -fieldwith ℏ = d and K δ = C . • Let K = C ( α, x ) be the field of rational functions in two variables α and x . We set δ := ddx and σ ( f ( α, x )) := f ( α + 1 , x ). Then we have ℏ = 1 and K δ = C ( α ) with σ ( α ) = α + 1.Alternatively, one could also take σ defined by σ ( f ( α, x )) = f ( qα, x ). The field C ( α )(( x )) offormal Laurent series over C ( α ) naturally is a δσ -extension of K . • Let k be an ultrametric field of characteristic zero, complete with respect to a discretevaluation. Assume that the residue field of k is F q , a field of characteristic p > q elements. We denote by | | the p -adic norm of k , normalized so that | p | = p − . The ring E † k of all f = P n ∈ Z a n x n , with a n ∈ k , such that – there exists ε >
0, depending on f , such that for any 1 < ̺ < ε we havelim n →±∞ | a n | ̺ n = 0; – sup n | a n | is bounded;is actually a field with residue field F q (( x )). (Cf. [Ked10, Lemma 15.1.3, p. 263].)We consider the field K = E † k as a δσ -field with derivation δ := x ddx and endomorphism σ : K → K , a lifting of the Frobenius endomorphism of F q . Namely, we consider an endo-morphism σ of k such that | σ ( a ) − a p | <
1, for any a ∈ k , | a | ≤ σ to K by setting σ ( x ) = x p , so that σ ( P n ∈ Z a n x n ) = P n ∈ Z σ ( a n ) x pn . We have ℏ = p and K δ = k . Definition 1.2.
Let K be a δσ -field and A ∈ K n × n . A δσ -field extension L of K is called a σ -Picard-Vessiot extension for δ ( y ) = Ay (or A ) if (i) there exists Y ∈ Gl n ( L ) such that δ ( Y ) = AY and L = K h Y ij | ≤ i, j ≤ n i σ and (ii) L δ = K δ .A K - δσ -algebra R is called a σ -Picard-Vessiot ring for δ ( y ) = Ay if (i) there exists Y ∈ Gl n ( R ) such that δ ( Y ) = AY and R = K (cid:8) Y ij , Y ) (cid:9) σ and (ii) R is δ -simple, i.e., R has no non-trivial δ -ideals. δσ -field extension L of K is called a σ -Picard-Vessiot extension if it is a σ -Picard-Vessiotextension for some differential equation δ ( y ) = Ay with A ∈ K n × n ; similarly for σ -Picard-Vessiotrings. To simplify the notation we write K h Y i σ for K h Y ij | ≤ i, j ≤ n i σ and K (cid:8) Y, Y ) (cid:9) σ for K (cid:8) Y ij , Y ) (cid:9) σ . If R is a K - δσ -algebra, then a matrix Y ∈ Gl n ( R ) such that δ ( Y ) = AY is calleda fundamental solution matrix for δ ( y ) = Ay . Thus, a σ -Picard-Vessiot extension is a δσ -fieldextension of K without new δ -constants, σ -generated by the entries of a fundamental solutionmatrix.If Y, Y ′ ∈ Gl n ( R ) are two fundamental solution matrices for δ ( y ) = Ay in some K - δσ -algebrathen there exists C ∈ Gl n ( R δ ) such that Y ′ = Y C . This is simply because δ ( Y − Y ′ ) = δ ( Y − ) Y ′ + Y − δ ( Y ′ ) = − Y − δ ( Y ) Y − Y ′ + Y − AY ′ = − Y − AY Y − Y ′ + Y − AY ′ = 0 . Note that we obtain the usual definitions of Picard-Vessiot extension and Picard-Vessiot ringof a linear differential equation δ ( y ) = Ay if we require that σ is the identity (on K, L and R ) inDefinition 1.2.The keen reader might have noticed a slight deviation between our definition of a σ -Picard-Vessiot ring and the corresponding notion in [CS06] and [HS08]: We require a σ -Picard-Vessiotring to be δ -simple and not only to be δσ -simple. See [Wib12b, p. 167] for some comments on thisissue. σ -Picard-Vessiot rings and extensions Our first concern is to show that the notions of σ -Picard-Vessiot ring and σ -Picard-Vessiot extensionare essentially equivalent. σ -Picard-Vessiot extensions can be seen as σ -analogs of classical Picard-Vessiot extensions.There is, however, another relation between the classical Picard-Vessiot theory and our σ -Picard-Vessiot theory: As we will now explain, every σ -Picard-Vessiot extension is a limit of Picard-Vessiotextensions.Let K be a δσ -field. From δ ( σ ( a )) = ℏ ( σ ( δ ( a ))) for a ∈ K it follows that δ ( σ d ( a )) = ℏ d σ d ( δ ( a ))for d ≥ ℏ d := ℏ σ ( ℏ ) · · · σ d − ( ℏ ) . Given a differential equation δ ( y ) = Ay with A ∈ K n × n we can consider for every d ≥ σ -jets of order d of δ ( y ) = Ay , namely, the linear system δ ( y ) = A d y ,where A d := A · · · ℏ σ ( A ) 0 · · · · · · ℏ d − σ d − ( A ) 00 · · · ℏ d σ d ( A ) ∈ Gl n ( d +1) ( K ) . (1.2) Lemma 1.3.
Let K be a δσ -field and A ∈ K n × n . If L | K is a σ -Picard-Vessiot extension for δ ( y ) = Ay with fundamental solution matrix Y ∈ Gl n ( L ) , then R d := K h Y, Y ) , σ ( Y ) , σ ( Y )) , . . . , σ d ( Y ) , σ d ( Y )) i is a (classical) Picard-Vessiot ring for δ ( y ) = A d y for every d ≥ .Proof. For i = 0 , . . . , d we have δ ( σ i ( Y )) = ℏ i σ i ( δ ( Y )) = ℏ i σ i ( A ) σ i ( Y ). It follows that the fractionfield L d of R d is a δ -subfield of L and that Y d = Y · · · σ ( Y ) 0 · · · · · · σ d − ( Y ) 00 · · · σ d ( Y ) ∈ Gl n ( d +1) ( L d ) .
7s a fundamental solution matrix for δ ( y ) = A d y . Because L δd ⊂ L δ = K δ , we conclude that L d isa Picard-Vessiot extension for δ ( y ) = A d y . Therefore R d is a Picard-Vessiot ring for δ ( y ) = A d y .(See [vdPS03, Prop. 1.22, p. 16] for the case that K δ is algebraically closed or [Dyc, Cor. 2.6, p.6] for the general case.)In order to prove the next proposition, we need another simple lemma: Lemma 1.4.
Let R be a δ -simple δσ -ring. Then R is a σ -domain, i.e. , R is an integral domainand σ is injective on R . In particular, δ and σ naturally extend to the field of fractions L of R .Moreover, L δ = R δ .Proof. It is well known that a δ -simple differential ring is an integral domain. (See e.g. [vdPS03,Lemma 1.17, p. 13].) It follows from the commutativity relation (1.1) for δ and σ that the kernelof σ on R is a δ -ideal. So by δ -simplicity, σ must be injective on R .Let a ∈ L δ . Then, a = { r ∈ R | ar ∈ R } is a non-zero δ -ideal of R . Thus a ∈ R δ .As in the classical theory, σ -Picard-Vessiot rings and σ -Picard-Vessiot extensions are closelyrelated: Proposition 1.5.
Let K be a δσ -field and A ∈ K n × n . If L | K is a σ -Picard-Vessiot extensionfor δ ( y ) = Ay with fundamental solution matrix Y ∈ Gl n ( L ) , then R := K { Y, Y ) } σ is a σ -Picard-Vessiot ring for δ ( y ) = Ay . Conversely, if R is a σ -Picard-Vessiot ring for δ ( y ) = Ay with R δ = K δ , then the field of fractions of R is a σ -Picard-Vessiot extension for δ ( y ) = Ay .Proof. To prove the first claim we only have to show that R := K { Y, Y ) } σ is δ -simple. Supposethat a is a non-trivial δ -ideal of R . Then, for a suitable d ≥
0, the ideal a ∩ R d is a non-trivial δ -ideal of R d := K (cid:20) Y, Y ) , σ ( Y ) , σ ( Y )) , . . . , σ d ( Y ) , σ d ( Y )) (cid:21) . This contradicts the fact that R d is δ -simple (Lemma 1.3).The second claim is clear from Lemma 1.4.The next proposition states that the condition R δ = K δ in Proposition 1.5 is always satisfiedif K δ is σ -closed. But let us first recall what it means for a σ -field to be σ -closed: Definition 1.6. A σ -field k is called σ -closed if for every finitely σ -generated k - σ -algebra R whichis a σ -domain (i.e., R is an integral domain and σ : R → R is injective) there exists a morphism R → k of k - σ -algebras. Model theorists usually call σ -closed σ -fields “existentially closed” or a “model of ACFA”. (Seee.g. [Mac97] or [CH99].) The assumption that R is a σ -domain in the above definition is quitecrucial. If k is a σ -closed σ -field and R a finitely σ -generated k - σ -algebra, there need not exist a k - σ -morphism R → k . Indeed, there exists a k - σ -morphism R → k if and only if there exists a σ -prime σ -ideal q in R , i.e., a prime ideal q of R with σ − ( q ) = q . See Lemma A.7.As indicated in the introduction, in the classical Picard-Vessiot theory the assumption that theconstants are algebraically closed is widely-used to avoid certain technicalities and it seems thatmany authors consider this assumption as natural. So, by way of analogy, it would be naturalfor us to assume that K δ is σ -closed. However, none of the examples of δσ -fields relevant for us(Example 1.1) does us the favour to have σ -closed δ -constants. So we have been careful to avoidthis assumption. Proposition 1.7.
Let K be a δσ -field such that K δ is a σ -closed σ -field. Then R δ = K δ for every σ -Picard-Vessiot ring R over K .Proof. This is [Wib12a, Corollary. 2.18, p. 1393].The following simple lemma is a fundamental tool for the development of our σ -Galois theory.8 emma 1.8. Let K be a δσ -field and A ∈ K n × n . If R and R are σ -Picard-Vessiot rings for δ ( y ) = Ay with R δ = R δ = k := K δ , then the canonical map R ⊗ k ( R ⊗ K R ) δ −→ R ⊗ K R is an isomorphism of R - δσ -algebras. Moreover, ( R ⊗ K R ) δ is finitely σ -generated over k . Indeedif Y ∈ Gl n ( R ) and Y ∈ Gl n ( R ) are fundamental solution matrices for δ ( y ) = Ay , then ( R ⊗ K R ) δ = k { Z, Z ) } σ where Z := ( Y ⊗ − (1 ⊗ Y ) ∈ Gl n ( R ⊗ K R ) .Proof. Because Y ⊗ , ⊗ Y ∈ Gl n ( R ⊗ K R ) are fundamental solution matrices for δ ( y ) = Ay we have δ ( Z ) = 0. It follows from the commuting relation (1.1) that k { Z, Z } σ ⊂ ( R ⊗ K R ) δ .Since 1 ⊗ Y = ( Y ⊗ Z and R is σ -generated by Y and Y ) we see that R ⊗ K R = R · k { Z, Z } σ . It holds in general that the δ -constants of a δ -algebra over a δ -simple δ -ring R are linearly disjoint from R over R δ . (See e.g. [AM05, Corollary 3.2, p. 753].) Therefore R ⊗ K R = R ⊗ k k { Z, Z } σ . This also shows that ( R ⊗ K R ) δ = k { Z, Z } σ . In [Tak89] (see also [AMT09]) M. Takeuchi gave a very general definition of Picard-Vessiot exten-sions which does not require any finiteness assumptions. It simply postulates the validity of analgebraic reformulation of the torsor theorem.
Definition 1.9 (Definition 1.8 in [AMT09]) . An extension L | K of δ -fields is called Picard-Vessiotin the sense of Takeuchi if the following conditions are satisfied. (i) L δ = K δ . (ii) There exists a K - δ -subalgebra R of L such that L is the quotient field of R and the canonicalmap R ⊗ K δ ( R ⊗ K R ) δ → R ⊗ K R is surjective. (It is then automatically an isomorphism.) We call the δ -ring R a Picard-Vessiot ring in the sense of Takeuchi. As in Section A.4, we denote with ( − ) ♯ the forgetful functor that forgets σ . So if R is a δσ -ringthen R ♯ is a δ -ring. Remark 1.10.
Let L | K be a σ -Picard-Vessiot extension. Then L ♯ | K ♯ is a Picard-Vessiot extensionin the sense of Takeuchi. Proof.
This is clear from Lemma 1.8 (with R = R ).Once we have defined the σ -Galois group G of a σ -Picard-Vessiot extension L | K we will showthat also the Galois group of L ♯ | K ♯ in the sense of Takeuchi (an affine group scheme, in generalnot of finite type over k = K δ ) can be obtained from G by forgetting σ . See Remark 2.8.Since the Picard-Vessiot theory in [Tak89] (or [AMT09]) does not at all take into account σ ,it is clearly not an appropriate theory to discuss the questions of this article, e.g., the σ -algebraicrelations among the solutions of a linear differential equation. Nevertheless, it is sometimes veryconvenient to know that every σ -Picard-Vessiot extension can be seen as a Picard-Vessiot extensionin the sense of Takeuchi; it allows for some shortcuts in the proofs. This applies most notably toour proof of the σ -Galois correspondence, which can be interpreted as the restriction of Takeuchi’scorrespondence to the σ -stable objects on both sides.The fact that every σ -Picard-Vessiot extension can be seen as a Picard-Vessiot extension in thesense of Takeuchi is also used in the proof of the following lemma. Lemma 1.11.
Let L | K be a σ -Picard-Vessiot extension. Let A ∈ K n × n and Y ∈ Gl n ( L ) suchthat L | K is a σ -Picard-Vessiot extension for δ ( y ) = Ay with fundamental solution matrix Y . Let A ′ ∈ K n ′ × n ′ and Y ′ ∈ Gl n ′ ( L ) be another pair of matrices such that L | K is a σ -Picard-Vessiotextension for δ ( y ) = A ′ y with fundamental solution matrix Y ′ . Then the corresponding σ -Picard-Vessiot rings R = K { Y, Y ) } σ and R ′ = K { Y ′ , Y ′ ) } σ are equal. roof. It follows from Lemma 1.8 that R and R ′ are Picard-Vessiot rings in the sense of Takeuchi.Because a Picard-Vessiot ring in the sense of Takeuchi is unique (inside L) by [AMT09, Lemma1.11, p. 133], it follows that R = R ′ .By the above remark it makes sense to speak of the σ -Picard-Vessiot ring R of the σ -Galoisextension L | K without reference to a specific equation δ ( y ) = Ay and we shall henceforth adhereto this practice. σ -Picard-Vessiot extensions Before proceeding to develop the Galois theory of σ -Picard-Vessiot extensions, we shall be con-cerned with the fundamental questions of existence and uniqueness of σ -Picard-Vessiot extensions.These questions have already been addressed in [Wib12a] to illustrate the usefulness of constrainedextensions of σ -pseudo fields. So we largely only recall the results from [Wib12a] Proposition 1.12 (Existence of σ -Picard-Vessiot rings) . Let K be a δσ -field and A ∈ K n × n . Thenthere exists a σ -Picard-Vessiot R ring for δ ( y ) = Ay such that R δ is an algebraic field extensionof K δ .Proof. This is [Wib12a, Lemma 2.16, p. 1392]. Because of the importance of the result we recallthe contruction: For each of the systems δ ( y ) = A d y from Lemma 1.3 we are able to construct a(classical) Picard-Vessiot ring individually by taking the quotient of S d := K h X, X ) , σ ( X ) , σ (det( X )) , . . . , σ d ( X ) , σ d (det( X )) i by some δ -maximal δ -ideal m d of S d . Here X is an n × n -matrix of σ -indeterminates and the actionof δ on S d is determined by δ ( X ) = AX and the commutativity relation (1.1). The difficulty is tomake this construction compatible with σ : We need m d − ⊂ m d and σ ( m d − ) ⊂ m d . This difficultycan be resolved by a recourse to the prolongation lemma for difference kernels ([Coh65, Lemma 1,Chapter 6, p. 149]).We set m := S d ≥ m d and R := k { X, X ) } σ / m . So R is the union of the δ -simple rings R d := S d / m d . The δ -constants of a δ -simple δ -ring which is finitely generated as an algebra overa δ -field K are algebraic over K . (See [vdPS03, Lemma 1.17, p. 13] for the case K δ algebraicallyclosed or [Tak89, Theorem 4.4, p. 505] for the general case). Thus R δd is algebraic over K δ and itfollows that also R δ is algebraic over K δ .From Propositions 1.5 and 1.12 we immediately obtain the existence of σ -Picard-Vessiot exten-sions over δσ -fields with algebraically closed δ -constants: Corollary 1.13 (Existence of σ -Picard-Vessiot extensions) . Let K be a δσ -field and A ∈ K n × n .Assume that K δ is an algebraically closed field. Then there exists a σ -Picard-Vessiot extension for δ ( y ) = Ay over K . The standard assumption to guarantee the existence of (classical) Picard-Vessiot extensions is“algebraically closed constants”. Since we can get by with the same assumption the above resultsare more or less optimal. In all generality the existence of σ -Picard-Vessiot extensions can notbe guaranteed. Indeed, if L | K is a σ -Picard-Vessiot extension for δ ( y ) = Ay with fundamentalsolution matrix Y ∈ Gl n ( L ) then K ( Y ) ⊂ L is a (classical) Picard-Vessiot extension of K for δ ( y ) = Ay . Thus, if there is no Picard-Vessiot extension for δ ( y ) = Ay there can be no σ -Picard-Vessiot extension for δ ( y ) = Ay . A concrete example of a δ -field K and an equation δ ( y ) = Ay such that there exists no Picard-Vessiot extension for δ ( y ) = Ay over K has been provided bySeidenberg in [Sei56]. To obtain a concrete example of a δσ -field K and an equation δ ( y ) = Ay such that there exists no σ -Picard-Vessiot extension for δ ( y ) = Ay over K one simply has to add σ as the identity on K . In [Wib12a] it is assumed that δ and σ commute, i.e., ℏ = 1. However, the proofs in [Wib12a] generalize to theslightly more general setting of this article without difficulty. σ -Picard-Vessiot extension for a given linear differentialequation, even if the δ -constants are not algebraically closed. The next proposition gives anexample of such a “natural situation”. See also Examples 2.9 to 2.14. Proposition 1.14.
Let k be a σ -field and let K = k ( x ) denote the field of rational functions in onevariable x over k . Extend σ to K by setting σ ( x ) = x and consider the derivation δ = ddx . Thus K is a δσ -field with ℏ = 1 and K δ = k . Then for every A ∈ K n × n , there exists a σ -Picard-Vessiotextension L | K for δ ( y ) = Ay .Proof. Since we are in characteristic zero, there exists an a ∈ k σ which is a regular point for δ ( y ) = Ay . That is, no denominator appearing in the entries of A vanishes at a . We consider the field k (( x − a )) of formal Laurent series in x − a as a δσ -field by setting δ ( P b i ( x − a ) i ) = P ib i ( x − a ) i − and σ ( P b i ( x − a ) i ) = P σ ( b i )( x − a ) i . Then k (( x − a )) is naturally a δσ -field extension of K . Bychoice of a , there exists a fundamental solution matrix Y ∈ Gl n ( k (( x − a )) for δ ( y ) = Ay .Since k (( x − a )) δ = k it is clear that L := K h Y i σ ⊂ k (( x − a )) is a σ -Picard-Vessiot extensionfor δ ( y ) = Ay .Next we shall be concerned with the uniqueness of σ -Picard-Vessiot extensions and σ -Picard-Vessiot rings for a given linear differential equation δ ( y ) = Ay . To motivate our results we firstrecall the classical situation: Let K be a δ -field and A ∈ K n × n . If R and R are two Picard-Vessiot rings for δ ( y ) = Ay , then there exists a finite algebraic extension k ′ of k := K δ containing k := R δ and k := R δ and an isomorphism of K ⊗ k k ′ - δ -algebras R ⊗ k k ′ ≃ R ⊗ k k ′ . In particular, if k = K δ is an algebraically closed field, a Picard-Vessiot ring (and thus also aPicard-extension) for a given equation is unique up to K - δ -isomorphisms.To obtain a similar uniqueness result for σ -Picard-Vessiot extensions, one needs to understandthe σ -analog of finite algebraic extensions. This has been worked out in [Wib12a, Section 2.1],where it is shown that constrained extensions of σ -pseudo fields ([Wib12a, Definition 2.3, p. 1388])satisfy properties similar to algebraic extensions of fields. These constrained extensions can alsobe seen as σ -analogs of the constrained extensions of differential fields studied by E. Kolchin in[Kol74].To state the main uniqueness result we need the following definition which will also be relevantlater on in Section 4. Definition 1.15.
Let k be a σ -field and R a k - σ -algebra. We say that R is σ -separable over k if σ is injective on R ⊗ k k ′ for every σ -field extension k ′ of k . Note that in characteristic p > σ ( a ) = a p . In this situation σ -separability isthe same thing as separability. The well-known characterizations of separability generalize in astraight forward manner. (Cf. [Hru04] or [Wib10].) For example, the well-known fact a reduced k -algebra over a perfect field is separable generalizes to “Every σ -reduced k - σ -algebra over aninversive σ -field k is σ -separable.” (Corollary A.14 (i).) Here a σ -ring R is called σ -reduced if σ : R → R is injective.Now we can state the general uniqueness theorem for σ -Picard-Vessiot rings ([Wib12a, Theorem2.19, p. 1393]). Theorem 1.16 (Uniqueness of σ -Picard-Vessiot rings) . Let K be a δσ -field such that K is σ -separable over k := K δ . Assume that R and R are two σ -Picard-Vessiot rings over K for thesame equation δ ( y ) = Ay , A ∈ K n × n . Then there exists a finitely σ -generated constrained σ -pseudofield extension k ′ of k containing k := R δ and k := R δ such that R ⊗ k k ′ and R ⊗ k k ′ areisomorphic as K ⊗ k k ′ - δσ -algebras. K is σ -separable over k := K δ ” is automatically satisfied if k is σ -closed because a σ -closed σ -field is inversive. (If a ∈ k and b ∈ k ∗ is an element in the inversiveclosure k ∗ ([Lev08, Def. 2.1.6, p. 109]) of k such that σ ( b ) = a then k { b } σ is a σ -domain.)Since the k ′ in Theorem 1.16 is a pseudo-field rather than a field, we obtain uniqueness over σ -closed δ -constants only up to powers of σ . Corollary 1.17.
Let K be a δσ -field such that K δ is a σ -closed σ -field. Let R and R be two σ -Picard-Vessiot rings for δ ( y ) = Ay with A ∈ K n × n . Then there exists an integer l ≥ such that R and R are isomorphic as K - δσ l -algebras.Proof. This is [Wib12a, Corollary 2.21, p. 1394]. To be precise, the corollary states that thereexists an integer l ≥ ψ : R → R of K - δ -algebras which commutes with σ l ,but maybe not with σ . Corollary 1.18.
Let K be a δσ -field such that K δ is a σ -closed σ -field. Let L and L be two σ -Picard-Vessiot extensions for δ ( y ) = Ay with A ∈ K n × n . Then there exists an integer l ≥ such that L | K and L | K are isomorphic as δσ l -field extensions of K .Proof. We know from Proposition 1.5 that L and L are the quotient fields of some σ -Picard-Vessiot rings R ⊂ L and R ⊂ L for δ ( y ) = Ay . The K - δσ l -isomorphism R → R which existsby Corollary 1.17 extends to an isomorphism of δσ l -field extensions of K .An example, illustrating that in general it is not possible to choose l = 1 in the above corollariescan be found in [Wib12a, Example 2.22, p. 1394]. In the remaining part of this subsection weprovide some information on when it is possible to choose l = 1. To formulate our results we needto recall the notion of compatibility of difference field extensions ([Lev08, Def. 5.1.1, p. 311]):Two extensions L | K and L | K of σ -fields are called compatible if there exists a σ -field extension M | K and K - σ -morphisms L → M and L → M . Proposition 1.19.
Let K be a δσ -field such that K δ is a σ -closed σ -field. Let L and L be two σ -Picard-Vessiot extensions of K for the same equation δ ( y ) = Ay , A ∈ K n × n . Then L and L are isomorphic (as δσ -field extensions of K ) if and only if L and L are compatible σ -fieldextensions of K .Proof. Of course the extensions L | K and L | K are compatible if they are isomorphic.Assume that L | K and L | K are compatible σ -field extensions. We have to show that L | K and L | K are isomorphic. Let R ⊂ L and R ⊂ L denote the corresponding σ -Picard-Vessiot ringsfor δ ( y ) = Ay and set k := K δ . From Lemma 1.8 we know that R ⊗ K R = R ⊗ k ( R ⊗ K R ) δ and that U := ( R ⊗ K R ) δ is a finitely σ -generated k - σ -algebra.Because L | K and L | K are compatible, there exists a σ -prime ideal in L ⊗ K L . (If M isa σ -field extension of K containing copies of L and L , then the kernel of L ⊗ K L → M is a σ -prime ideal.) Via the inclusion U ֒ → R ⊗ k U = R ⊗ K R ֒ → L ⊗ K L this σ -prime ideal of L ⊗ K L contracts to a σ -prime ideal of U . Because U is finitely σ -generatedover the σ -closed σ -field k , the existence of a σ -prime ideal in U is sufficient to guarantee theexistence of a morphism ψ : U → k of k - σ -algebras. This yields a morphism ϕ : R → R ⊗ K R = R ⊗ k U id · ψ −−−→ R of K - δσ -algebras. Because R is δ -simple ϕ is injective, and because R and R are σ -generated over K by a fundamental solution matrix for the same equation δ ( y ) = Ay we see that ϕ is surjective.So ϕ : R → R is an isomorphism. Of course ϕ extends to an isomorphism L ≃ L . Corollary 1.20.
Let K be a δσ -field such that K δ is a σ -closed σ -field. Let L and L be two σ -Picard-Vessiot extensions of K for the same equation. Assume that K is relatively algebraicallyclosed in L . Then L and L are isomorphic (as δσ -field extensions of K ). roof. In view of Proposition 1.19, it suffices to acknowledge that a σ -field extension L | K suchthat K is relatively algebraically closed in L is compatible with any other σ -field extension of K .This follows for example from [Lev08, Theorem 5.1.6, p. 313]. Remark 1.21.
The condition “ K is relatively algebraically closed in L ” in Corollary 1.20 canbe weakened to “The core of L | K is equal to K ”. This means that every finite σ -field extensionof K inside L is equal to K . (In general, the relative algebraic closure of K in L is of infinitedegree over K and may or may not contain finite σ -field extensions of K .) Proof.
This follows from the classical compatibility theorem [Lev08, Theorem 5.4.22, p. 342]. σ -Galois group of a linear differential equation In this section, we introduce the σ -Galois group of a linear differential equation δ ( y ) = Ay over a δσ -field K . More precisely, we will define the σ -Galois group of a fixed σ -Picard-Vessiot extensionfor δ ( y ) = Ay . It is a σ -algebraic group over K δ . We show that the Zariski closure of the σ -Galois group is the classical Galois group of δ ( y ) = Ay . We also explain the significance of thehigher order Zariski closures of the σ -Galois group and show that the σ -transcendence degree of a σ -Picard-Vessiot extension equals the σ -dimension of its σ -Galois group.For a brief introduction to σ -algebraic groups, we refer the reader to the appendix. Here weonly recall the definition. Definition 2.1.
Let k be a σ -field. A σ -algebraic group over k is a (covariant) functor G from thecategory of k - σ -algebras to the category of groups which is representable by a finitely σ -generated k - σ -algebra. I.e., there exists a finitely σ -generated k - σ -algebra k { G } such that G ≃ Alg σk ( k { G } , − ) . Here Alg σk stands for morphisms of k - σ -algebras. By the Yoneda lemma k { G } is unique up toisomorphisms. If R ⊂ S is an inclusion of δσ -rings, we denote by Aut δσ ( S | R ) the automorphismsof S over R in the category of δσ -rings, i.e., the automorphisms are required to be the identity on R and to commute with δ and σ . Definition 2.2.
Let L | K be a σ -Picard-Vessiot extension with σ -Picard-Vessiot ring R ⊂ L . Set k = K δ . We define σ - Gal( L | K ) to be the functor from the category of k - σ -algebras to the categoryof groups given by σ - Gal( L | K )( S ) := Aut δσ ( R ⊗ k S | K ⊗ k S ) for every k - σ -algebra S . The action of δ on S is trivial, i.e., δ ( r ⊗ s ) = δ ( r ) ⊗ s for r ∈ R and s ∈ S . We call σ - Gal( L | K ) the σ -Galois group of L | K . On morphisms σ - Gal( L | K ) is given by base extension: If ψ : S → S ′ is a morphism of k - σ -algebras, then ( σ - Gal( L | K ))( ψ ) : σ - Gal( L | K )( S ) → σ - Gal( L | K )( S ′ ) is the morphism of groupswhich associates to a K ⊗ k S - δσ -automorphims τ : R ⊗ k S → R ⊗ k S the K ⊗ k S ′ - δσ -automorphims R ⊗ k S ′ = ( R ⊗ k S ) ⊗ S S ′ τ ⊗ id −−−→ ( R ⊗ k S ) ⊗ S S ′ = R ⊗ k S ′ .Note that σ - Gal( L | K )( k ) = Aut δσ ( R | K ) = Aut δσ ( L | K ). To show that σ - Gal( L | K ) is a σ -algebraic group we shall need two simple lemmas. Lemma 2.3.
Let R be a δ -simple δ -ring, k := R δ and S a k -algebra, considered as a constant δ -algebra. Then ( R ⊗ k S ) δ = S and the assignments a R ⊗ k a and b S ∩ b define mutuallyinverse bijections between the set of ideals of S and the set of δ -ideals of R ⊗ k S . In particular,every δ -ideal b of R ⊗ k S is generated by b ∩ S as an ideal.Proof. The first claim follows immediately when choosing appropriate bases. The second claimfollows as in [Kov03, Prop. 5.6, p. 4484]. See also [Mau10, Lemma 10.7, p. 5443].
Lemma 2.4.
Let L | K be a σ -Picard-Vessiot extension with σ -Picard-Vessiot ring R ⊂ L and δ -constants k . If S is a k - σ -algebra, then every K ⊗ k S - δσ -endomorphism of R ⊗ k S is an auto-morphism. roof. Fix matrices A ∈ K n × n and Y ∈ Gl n ( L ) such that L | K is a σ -Picard-Vessiot extensionfor δ ( y ) = Ay with fundamental solution matrix Y . If τ : R ⊗ k S → R ⊗ k S is a K ⊗ k S - δσ -morphism then τ ( Y ) ∈ Gl n ( R ⊗ k S ) is a fundamental solution matrix for δ ( y ) = Ay . Since also Y (= Y ⊗ ∈ Gl n ( R ⊗ k S ) is a fundamental solution matrix, there exists a (unique) matrix[ τ ] Y ∈ Gl n (( R ⊗ k S ) δ ) = Gl n ( S ) such that τ ( Y ) = Y [ τ ] Y . Because R is σ -generated by Y , itfollows that τ is surjective. The kernel of τ is a δ -ideal of R ⊗ k S . It follows from Lemma 2.3 that τ is injective. Proposition 2.5.
Let L | K be a σ -Picard-Vessiot extension with σ -Picard-Vessiot ring R ⊂ L .Then σ - Gal( L | K ) is a σ -algebraic group over k = K δ . More precisely, σ - Gal( L | K ) is representedby the finitely σ -generated k - σ -algebra ( R ⊗ K R ) δ . The choice of matrices A ∈ K n × n and Y ∈ Gl n ( L ) such that L | K is a σ -Picard-Vessiot extension for δ ( y ) = Ay with fundamental solutionmatrix Y defines a σ -closed embedding σ - Gal( L | K ) ֒ → Gl n,k of σ -algebraic groups.Proof. Let S be a k - σ -algebra. From Lemma 2.4 and Lemma 1.8, we obtain the following chain ofidentifications: σ - Gal( L | K )( S ) = Alg δσK ⊗ k S ( R ⊗ k S, R ⊗ k S ) = Alg δσK ( R, R ⊗ k S ) == Alg δσR ( R ⊗ K R, R ⊗ k S ) = Alg δσR ( R ⊗ k ( R ⊗ K R ) δ , R ⊗ k S ) == Alg σk (( R ⊗ K R ) δ , S ) . (2.1)The last identity holds because every R - δσ -morphism R ⊗ k ( R ⊗ K R ) δ → R ⊗ k S restricts toa k - σ -morphism ( R ⊗ K R ) δ → S by taking δ -constants.We have already seen in Lemma 1.8 that ( R ⊗ K R ) δ = k { Z, Z ) } σ where Z = ( Y ⊗ − (1 ⊗ Y ) ∈ Gl n ( R ⊗ K R ). So ( R ⊗ K R ) δ is finitely σ -generated over k .As in the proof of Lemma 2.4, every automorphism τ ∈ Gal( L | K )( S ) is given by a matrix[ τ ] Y ∈ Gl n ( S ) = Gl n (1 ⊗ S ) such that τ ( Y ) = Y [ τ ] Y . Then τ is given in Alg δσR ( R ⊗ K R, R ⊗ k S )by τ (1 ⊗ Y ) = ( Y ⊗ τ ] Y . So, as an element of Alg δσR ( R ⊗ k ( R ⊗ K R ) δ , R ⊗ k S ) the morphism τ is given by τ ( Z ) = ( Y ⊗ − τ (1 ⊗ Y ) = [ τ ] Y . In summary, we see that, under the identification σ - Gal( L | K )( S ) = Alg σk (( R ⊗ K R ) δ , S ), an automorphism τ ∈ σ - Gal( L | K )( S ) corresponds tothe k - σ -morphism ( R ⊗ K R ) δ → S determined by Z [ τ ] Y . It follows that the identification σ - Gal( L | K )( S ) = Alg σk (( R ⊗ K R ) δ , S ) is functorial in S . So σ - Gal( L | K ) is represented by ( R ⊗ K R ) δ .Moreover, the assignment τ [ τ ] Y defines an injection σ - Gal( L | K )( S ) ֒ → Gl n,k ( S ). Thisyields an embedding of functors σ - Gal( L | K ) ֒ → Gl n,k . If we set S := ( R ⊗ K R ) δ = k { Z Z ) } σ the automorphism τ univ ∈ σ - Gal( L | K )( S ) corresponding to id ∈ Alg σk (( R ⊗ K R ) δ , S ) is given by τ univ ( Y ⊗
1) = ( Y ⊗ Z , i.e., [ τ univ ] Y = Z . Thus, the dual morphism of σ - Gal( L | K ) ֒ → Gl n,k isgiven by k { Gl n,k } = k { X, X ) } σ −→ ( R ⊗ K R ) δ = k { Z, Z ) } σ , X Z. This is clearly surjective. So σ - Gal( L | K ) ֒ → Gl n,k is a σ -closed embedding. (See DefinitionA.3.) Definition 2.6.
Let L | K be a σ -Picard-Vessiot extension with σ -Picard-Vessiot ring R , σ -field of δ -constants k , and σ -Galois group G := σ - Gal( L | K ) . We set k { G } = ( R ⊗ K R ) δ . By the above proposition k { G } is a finitely σ -generated k - σ -algebra representing G . From Lemma 1.8 we immediately obtain the algebraic recast of the torsor theorem:14 emma 2.7.
Let L | K be a σ -Picard-Vessiot extension with σ -Picard-Vessiot ring R , σ -field of δ -constants k , and σ -Galois group G . Then R ⊗ K R = R ⊗ k k { G } . We will next explain the relation between our σ -Galois group and the Galois group in the senseof Takeuchi. Let k be a σ -field and X = Alg σk ( k { X } , − ) a k - σ -scheme. (See Definition A.1.) The k -scheme obtained from X by forgetting σ is X ♯ = Alg k ( k { X } ♯ , − ). (See Section A.4.)Let L | K be a σ -Picard-Vessiot extension with σ -Picard-Vessiot ring R , σ -field of δ -constants k , and σ -Galois group G . Let T be a k -algebra. Replacing S by T and forgetting σ in theidentifications of equation (2.1) above, we see that G ♯ ( T ) = Aut δ ( R ⊗ k T | K ⊗ k T ) . Remark 2.8.
Let L | K be a σ -Picard-Vessiot extension with σ -Galois group G . We already notedin Remark 1.10 that L ♯ | K ♯ is a Picard-Vessiot extension in the sense of Takeuchi. The Galoisgroup of L ♯ | K ♯ (in the sense of [Tak89]) agrees with G ♯ , the group scheme obtained from G byforgetting σ . Proof.
Let R ⊂ L denote the σ -Picard-Vessiot ring of L | K . Initially the Galois group of L ♯ | K ♯ , ormore precisely its representing Hopf algebra D , is defined by some abstract algebraic manipulationswith corings. See [AMT09, Section 1]. As a k -algebra D = ( R ⊗ k R ) δ = k { G } ♯ . Moreover in [Tak89,Appendix, Theorem A.2, p. 508], it is shown that D represents the automorphism functor T Aut δ ( R ⊗ k T | K ⊗ k T ) . To get a better feeling for what is really going on, let us compute the σ -Galois group in somesimple examples, including the ones given in the introduction. Example 2.9.
Let K = C ( x ) be the δσ -field of rational functions in the variable x over C , where δ = ddx and σ ( f ( x )) = f ( x + 1). So k = K δ = C , with σ the identity map. Consider the equation δ ( y ) = 2 xy over K . The field M of meromorphic functions on C is naturally a δσ -field extensions of K (with δ = ddx and σ : f ( x ) f ( x + 1)). Since M δ = k , it is clear that L = K h e x i σ ⊂ M is a σ -Picard-Vessiot extension for δ ( y ) = 2 xy . The σ -Picard-Vessiot ring is R = K { e x , e − x } σ ⊂ L . Since x − x + 1) + ( x + 2) = 2, we have f σ ( f ) − σ ( f ) = e ∈ K, where we have set f = e x to simplify the notation. So R = K (cid:2) e x , e − x , e ( x +1) , e − ( x +1) (cid:3) . Let G = σ - Gal( L | K ) denote the σ -Galois group. We consider G as a σ -closed subgroup of Gl ,k viathe fundamental solution matrix Y = e x ∈ Gl ( L ). Let S be a k - σ -algebra. For g ∈ G ( S ) ⊂ S × we have e = g ( e ) = g ( f ) σ ( g ( f )) − σ ( g ( f )) = gσ ( g ) − σ ( g ) e . (Note that the above computation takes place in R ⊗ k S .) Therefore gσ ( g ) − σ ( g ) = 1. On theother side, the functions e x and e ( x +1) are algebraically independent over K . (This follows forexample from the Kolchin-Ostrowski theorem.) So for any g ∈ S × , satisfying gσ ( g ) − σ ( g ) = 1,we have a well-defined K ⊗ k S - δσ -automorphism of R ⊗ k S determined by e x ge x .In summary, G ≤ Gl ,k is given by G ( S ) = { g ∈ S × | gσ ( g ) − σ ( g ) = 1 } ≤ Gl ( S )for any k - σ -algebra S . 15 xample 2.10. As in the above example, let K = C ( x ) be the δσ -field of rational functions inthe variable x over C , where δ = ddx and σ ( f ( x )) = f ( x + 1). Consider the equation δ ( y ) = 12 x y over K .Fix an algebraic closure K of K . Then δ extends uniquely to K . We can also extend σ to K .The extension of σ to K is of course not unique but one can show that any two extensions areisomorphic. In particular, the σ -Galois group will be independent of this choice. The derivation δ and the endomorphisms σ also commute on K . I.e., K is a δσ -extension of K . Obviously √ x ∈ K is a fundamental solution matrix for δ ( y ) = x y . Since K δ = K δ = C =: k it is clear that L := K h√ x i σ = K ( √ x, √ x + 1 , . . . ) ⊂ K is a σ -Picard-Vessiot extension for δ ( y ) = x y . The σ -Picard-Vessiot ring R is equal to L . Let G = σ - Gal( L | K ) denote the σ -Galois group. Since a δσ -automorphism of R | K is determined byits action on √ x we see that G ( k ) = Aut δσ ( R | K ) = C , where C = { , − } is the group with two elements acting on R by √ x
7→ −√ x . On the otherhand, as the degree of K ( √ x, . . . , √ x + i ) over K is 2 i +1 for i ≥
0, we see that G ♯ ( k ♯ ) = Aut δ ( R | K ) = C ∞ = C × C × · · · . As a σ -closed subgroup of Gl ,k the σ -Galois group is given by G ( S ) = { g ∈ S × | g = 1 } ≤ Gl ( S )for every k - σ -algebra S . An elements g ∈ G ( S ) is acting on R ⊗ k S by √ x + i σ i ( g ) √ x + i for i ≥ Example 2.11.
Let K = C ( x ) denote the field of rational functions in the variable x over C . Weconsider K as δσ -field with derivation δ = x ddx and endomorphism σ , given by σ ( f ( x )) = f ( x d )for some integer d ≥
2. So ℏ = d and k := K δ = C . Let us consider the equation δ ( y ) = d y over K . It has the solution x d . The field L = C ( x d ) is naturally a δσ -field extension of K with σ ( x d ) = x . Since L δ = C , it is clear that L | K is a σ -Picard-Vessiot extension for x d . The σ -Picard-Vessiot ring R equals L . We consider the σ -Galois group G = σ - Gal( L | K ) as a σ -closedsubgroup of Gl ,k . For any k - σ -algebra S and every g ∈ G ( S ) ≤ Gl ( S ) we have σ ( g ( x d )) = σ ( g ) x .On the other hand, σ ( g ( x d )) = g ( σ ( x d )) = g ( x ) = x . Therefore σ ( g ) = 1. Since ( x d ) d ∈ K weeasily see that g d = 1 and that the σ -Galois group is given by G ( S ) = { g ∈ S × | g d = 1 , σ ( g ) = 1 } ≤ Gl ( S )for any k - σ -algebra S . Note that G is not σ -reduced, i.e., σ is not injective on the σ -coordinatering k { G } . Indeed, the image of x − k { G } = k { x, x − } σ / [ x d − , σ ( x ) −
1] is a non-zero elementin the kernel of σ . Example 2.12.
Let us consider Bessel’s differential equation x δ ( y ) + xδ ( y ) + ( x − α ) y = 0 . The matrix of the equivalent system is A = (cid:18) α x − − x (cid:19) .
16s demonstrated in Proposition 1.14, there exists a σ -Picard-Vessiot extension for δ ( y ) = Ay over C ( α, x ), where δ = ddx and σ ( f ( α, x )) = f ( α + 1 , x ). However, since the classical solutions J α ( x ) and Y α ( x ) are normalized by some factor meromorphic in α , it is more convenient to workwith meromorphic rather than rational functions in α . Let M denote the field of meromorphicfunctions on { α ∈ C | Re( α ) > } . We consider M as σ -field by σ ( f ( α )) = f ( α +1). Let K = M ( x )denote the δσ -field of rational functions in x over M with derivation δ = ddx and endomorphism σ : K → K , extended from M by σ ( x ) = x . Then k = K δ = M . As an ambient δσ -field,containing the Bessel function of the first kind J α ( x ) and the Bessel function of the second kind Y a ( x ) we can, for example, choose the field E = M (( x − x − M , where, as before, δ = ddx and σ ( P a i ( x − i ) = P σ ( a i )( x − i . For generalitieson Bessel functions we refer the reader to [Wat95]. The matrix Y = (cid:18) J α ( x ) Y α ( x ) δ ( J α ( x )) δ ( Y α ( x )) (cid:19) ∈ Gl n ( E )is a fundamental solution matrix for δ ( y ) = Ay . Since E δ = k , we see that L = K h Y i σ ⊂ E is a σ -Picard-Vessiot extension for δ ( y ) = Ay over K . The recurrence formulas2 αx Z α ( x ) = Z α − ( x ) + Z α +1 ( x )and 2 δ ( Z α ( x )) = Z α − ( x ) − Z α +1 ( x ) , satisfied by J α ( x ) and Y α ( x ) can be rewritten in matrix form as σ ( Y ) = BY , where B = (cid:18) αx − − α ( α +1) x + 1 α +1 x (cid:19) ∈ Gl n ( K ) . We consider the σ -Galois group G = σ - Gal( L | K ) as σ -closed subgroup of Gl ,k via the fundamentalsolution matrix Y . Let S be a k - σ -algebra and g ∈ G ( S ) ≤ Gl ( S ). We have g ( σ ( Y )) = g ( BY ) = Bg ( Y ) = BY g.
On the other hand, g ( σ ( Y )) = σ ( g ( Y )) = σ ( Y g ) =
BY σ ( g ) . So σ ( g ) = g . The functions J α ( x ) , Y α ( x ) , δ ( J α ( x )) , δ ( Y α ( x )) are not algebraically independent.Indeed, the Wronskian of J α ( x ) , Y α ( x ) equals πx . In particular, det( Y ) ∈ K . It follows thatdet( Y ) = g (det( Y )) = det( g ( Y )) = det( Y g ) = det( Y ) det( g ) . So det( g ) = 1.For a fixed α ∈ C , with α − / ∈ Z , the (classical) Galois group of Bessel’s equation over C ( x )is Sl , C ([Kol68, Appendix]). Roughly speaking, this means that det( Y ) ∈ C ( x ) is the “only”algebraic relation among J α ( x ) , Y α ( x ) , δ ( J α ( x )) , δ ( Y α ( x )) over C ( x ).This implies that det( Y ) ∈ K is the “only” algebraic relation among J α ( x ) , Y α ( x ) , δ ( J α ( x )) , δ ( Y α ( x ))over K . Because R = K { Y, Y ) } σ = K [ Y, Y ) ] we find that G ≤ Gl ,k is given by G ( S ) = { g ∈ Sl ( S ) | σ ( g ) = g } , for any k - σ -algebra S . Example 2.13.
Let q be a complex number of norm greater than 1. We consider the Jacobi Thetafunction θ q ( x ) = X n ∈ Z q − n ( n − / x n ℓ q ( x ) = δ ( θ q ( x )) θ q ( x ) , where δ = x ddx . Since | q | >
1, the formal series θ q naturally defines a meromorphic function on C ∗ := C r { } and satisfies the q -difference equation θ q ( qx ) = qxθ q ( x ) , so that ℓ q ( qx ) = ℓ q ( x ) + 1. This implies that σ ( δ ( ℓ q )) = δ ( ℓ q ), where σ is the q -differenceoperator f ( x ) f ( qx ). We want to give an interpretation of these classical formulas in thepresent framework.Inside the δσ -field M of meromorphic function on C ∗ , we consider the δσ -subfield K := M σ of q -elliptic functions and the differential equation δ ( y ) = δ ( ℓ q ) (2.2)with coefficients in K . Since the δσ -field L := K ( ℓ q ) is contained in M , we deduce that L δ = K δ = C =: k . This means that L | K is a σ -Picard-Vessiot extension for the system δ (cid:18) y y (cid:19) = (cid:18) δ ( ℓ q )0 0 (cid:19) (cid:18) y y (cid:19) associated with (2.2). The σ -Picard-Vessiot ring is R := K [ ℓ q ] and the σ -Galois group G := σ - Gal( L | K ) is naturally contained in the additive group G a,k . For any k - σ -algebra S and any g ∈ G ( S ) ≤ G a ( S ) we have g ( σ ( ℓ q )) = σ ( g ( ℓ q )) = σ ( ℓ q + g ) = ℓ q + 1 + σ ( g ) . On the other hand, g ( σ ( ℓ q )) = ℓ q + 1 + g and consequently σ ( g ) = g . Since ℓ q does not belong to K , and therefore is transcendental over K , we see that G ( S ) = { g ∈ S | σ ( g ) = g } ≤ G a ( S ) . Example 2.14.
We go back to the p -adic example of the Dwork exponential already presented inthe introduction. See also Example 1.1 for the notation. We assume that there exists π ∈ k suchthat π p − = − p . Our base δσ -field is K := E † k . We have already pointed out that exp( πx ) / ∈ K and that L := K (exp( πx )) is a δσ -field. In fact, since L δ = K δ = k , is is clear that L | K is a σ -Picard-Vessiot extension for δ ( y ) = πxy. The σ -Picard-Vessiot ring is R := K [exp( πx ) , exp( πx ) − ]. Let us consider the σ -Galois group G := σ - Gal( L | K ) as a σ -closed subgroup of Gl ,k . Notice that exp( πx ) p = exp( pπx ) ∈ K .This implies that g p = 1 for every g ∈ G ( S ) ≤ Gl ( S ) and any k - σ -algebra S . Moreover, since σ (exp( πx )) exp( πx ) − ∈ K , we also find that σ ( g ) = g . As exp( πx ) / ∈ K , it is now easy to see that G ( S ) = { g ∈ S × | g p = 1 , σ ( g ) = g } ≤ Gl ( S ) . The fact observed in Examples 2.9, 2.10 and 2.14, that the σ -algebraic relations satisfied by thesolution of a first order linear differential equations δ ( y ) = ay can be described by σ -monomials,is a general pattern which can be derived from the classification of the σ -closed subgroups of themultiplicative group (Lemma A.40). See [DVHW] for more details.We continue by describing the relation between the σ -Galois group and the classical Galoisgroup. Proposition 2.15.
Let L | K be a σ -Picard-Vessiot extension with σ -field of δ -constants k = K δ .Let A ∈ K n × n and Y ∈ Gl n ( L ) such that L | K is a σ -Picard-Vessiot extension for δ ( y ) = Ay with fundamental solution matrix Y . We consider the σ -Galois group G of L | K as a σ -closedsubgroup of Gl n,k via the embedding associated with the choice of A and Y . For d ≥ , set L d = K (cid:0) Y, σ ( Y ) , . . . , σ d ( Y ) (cid:1) ⊂ L .Then L d | K is a (classical) Picard-Vessiot extension for the linear system δ ( y ) = A d y , where A d ∈ Gl n ( d +1) ( K ) is defined in equation (1.2). he (classical) Galois group of L d | K is naturally isomorphic to G [ d ] , the d -th order Zariskiclosure of G inside Gl n,k . (See Definition A.11.) In particular, L = K ( Y ) is a (classical) Picard-Vessiot extension for δ ( y ) = Ay and the Zariski closure of G inside Gl n,k is the (classical) Galoisgroup of δ ( y ) = Ay .Proof. By Lemma 1.3, L d | K is a classical Picard-Vessiot extension for δ ( y ) = A d y and for d ≥ R d = K h Y, σ ( Y ) , . . . , σ d ( Y ) , Y ··· σ d ( Y )) i ⊂ L d is the Picard-Vessiot ring of L d | K . By Proposition 1.5, the K - δσ -algebra R = K { Y, Y ) } σ ⊂ L is the σ -Picard-Vessiot ring of L | K . We denote the Galois group of L d | K with G d . So G d ( T ) = Aut δ ( R d ⊗ k T | K ⊗ k T )for every k -algebra T . Since an automorphism τ ∈ G ♯ ( T ) = Aut δ ( R ⊗ k T | K ⊗ k T ) restrictsto an automorphism e τ ∈ G d ( T ) = Aut δ ( R d ⊗ k T | K ⊗ k T ), we obtain a morphism G ♯ → G d of group k -schemes. Because of the special shape of Y d , we see that G d is a closed subschemeof Gl n,k × · · · × Gl n,k = (Gl n,k ) d . (See Section A.5 for an explanation of this notation.) But,by definition, G [ d ] is the smallest closed subscheme of (Gl n,k ) d such that G ♯ → (Gl n,k ) d factorsthrough G [ d ] ֒ → (Gl n,k ) d . Thus G [ d ] ⊂ G d . The image of G ♯ → G d is a subfunctor of G d (notclosed in general) contained in G [ d ]. Since every element of L invariant under G ♯ must lie in K ,we see that every element of L d invariant under G [ d ] must lie in K . But then it follows from theGalois correspondence for L d | K that G [ d ] = G d .We will finish this subsection by showing that the σ -transcendence degree of a σ -Picard-Vessiotextension equals the σ -dimension of its σ -Galois group. But let’s first recall the definition of σ -transcendence degree. Definition 2.16 (Definition 4.1.7 in [Lev08]) . Let L | K be a σ -field extension. Elements a , . . . , a n ∈ L are called transformally (or σ -algebraically) independent over K if the elements a , . . . , a n , σ ( a ) , . . . , σ ( a n ) , . . . are algebraically independent over K . Otherwise, they are called transformally dependent over K .A σ -transcendence basis of L over K is a maximal transformally independent over K subset of L . Any two σ -transcendence bases of L | K have the same cardinality and so we can define the σ -transcendence degree of L | K , or σ - trdeg( L | K ) for short, as the cardinality of any σ -transcendencebasis of L over K . The definition of the σ -dimension σ - dim k ( G ) of a σ -algebraic group G over a σ -field k is givenin Section A.7 of the appendix. Proposition 2.17.
Let L | K be a σ -Picard-Vessiot extension with σ -Galois group G and constantfield k = K δ . Then σ - trdeg( L | K ) = σ - dim k ( G ) . Proof.
Let R ⊂ L denote the corresponding σ -Picard-Vessiot ring. Then R ⊗ K R = R ⊗ k k { G } .Therefore L ⊗ K R = L ⊗ k k { G } . It follows from Lemma A.27 that σ - dim k ( G ) = σ - dim L ( L ⊗ k k { G } ) = σ - dim L ( L ⊗ K R ) = σ - dim K ( R ) . By Lemma A.26 σ - dim K ( R ) = σ - trdeg( L | K ) . σ -Galois correspondence In this section, we will establish the σ -versions of the first and second fundamental theorem ofGalois theory. 19et L | K be a σ -Picard-Vessiot extension with σ -Picard-Vessiot ring R ⊂ L , k := K δ , the σ -field of δ -constants and G := σ - Gal( L | K ), the σ -Galois group of L | K . Let S be a k - σ -algebra, τ ∈ G ( S ) and a ∈ L . We follow [Dyc] and [Mau10] in giving meaning to the phrase “ a is invariantunder τ ”. By definition, τ is an automorphism of R ⊗ k S . The total quotientring Quot( R ⊗ k S )contains L . It might not be possible to extend σ to Quot( R ⊗ k S ), but in any case τ extends toan automorphism of rings τ : Quot( R ⊗ k S ) → Quot( R ⊗ k S ) and it makes sense to say that a isinvariant under τ . If we write a = r r with r , r ∈ R , r = 0 then a is invariant under τ if andonly if τ ( r ⊗ · r ⊗ r ⊗ · τ ( r ⊗ ∈ R ⊗ k S .If H is a subfunctor of G , we say that a ∈ L is invariant under H if a is invariant under everyelement of H ( S ) ⊂ G ( S ) for every k - σ -algebra S . The set of all elements in L , invariant under H is denoted with L H . Obviously L H is an intermediate δσ -field of L | K .If M is an intermediate δσ -field of L | K , then it is immediately clear from Definition 1.2 that L | M is a σ -Picard-Vessiot extension with σ -Picard-Vessiot ring M R , the ring compositum of M and R inside L . Let S be a k - σ -algebra. When we fix a fundamental solution matrix Y ∈ Gl n ( L ),then an M ⊗ k S - δσ -automorphism of M R ⊗ k S is given by a matrix [ τ ] Y ∈ Gl n ( S ). It follows that τ restricts to a K ⊗ k S - δσ -automorphism of R ⊗ k S . This yields an injection σ - Gal( L | M )( S ) ֒ → σ - Gal( L | K )( S ) whose image consists of precisely those τ ∈ G ( S ) that leave invariant everyelement of M . We will often identify σ - Gal( L | M ) with this subfunctor of σ - Gal( L | K ). Be-cause σ - Gal( L | M ) and σ - Gal( L | K ) can be realized as σ -closed subgroups of Gl n,k it follows that σ - Gal( L | M ) is then a σ -closed subgroup of σ - Gal( L | K ). Lemma 3.1.
Let L | K be a σ -Picard-Vessiot extension with σ -Picard-Vessiot ring R , σ -Galoisgroup G = σ - Gal( L | K ) and field of δ -constants k . Let a ∈ L and r , r ∈ R , r = 0 such that a = r r . If H ≤ G is a σ -closed subgroup of G , then a is invariant under H if and only if r ⊗ r − r ⊗ r lies in the kernel of R ⊗ K R = R ⊗ k k { G } → R ⊗ k k { H } . Moreover the invariantsof H can be computed as L H = { a ∈ L | ⊗ a − a ⊗ ∈ L ⊗ K L · I ( H ) } , (3.1) where I ( H ) ⊂ k { G } denotes the defining ideal of H . (See Definition A.3.)Conversely, if M is an intermediate δσ -field of L | K , the defining ideal of σ - Gal( L | M ) in k { G } can be computed as I ( σ - Gal( L | M )) = ker( L ⊗ K L → L ⊗ M L ) ∩ k { G } . (3.2) Proof.
The K - δσ -morphism ψ : R → R ⊗ K R = R ⊗ k k { G } → R ⊗ k k { H } , where the first map is the inclusion into the second factor, extends to a K ⊗ k k { H } - δσ -morphism τ : R ⊗ k k { H } → R ⊗ k k { H } . It follows from Lemma 2.4 that τ is an automorphism. I.e., τ ∈ G ( S ),where S := k { H } . Now to say that a is invariant under τ precisely means that r ⊗ r − r ⊗ r lies in the kernel of R ⊗ K R = R ⊗ k k { G } → R ⊗ k k { H } .For the reverse direction, let S be any k - σ -algebra. According to the identifications made in theproof of Proposition 2.5, every τ ∈ H ( S ) ⊂ Aut( R ⊗ k S | K ⊗ k S ) is obtained from a k - σ -morphism k { H } → S by extending R ψ −→ R ⊗ k k { H } → R ⊗ k S to R ⊗ k S . This shows that a is invariantunder τ if r ⊗ r − r ⊗ r lies in the kernel of R ⊗ K R = R ⊗ k k { G } → R ⊗ k k { H } .In summary we see that a is invariant under H if and only if r ⊗ r − r ⊗ r lies in the idealof R ⊗ K R generated by I ( H ) = ker( k { G } → k { H } ). Working inside L ⊗ K L we can divide by r ⊗ r to obtain 1 ⊗ a − a ⊗ ∈ L ⊗ K L · I ( H ) if a is invariant under H . This is “ ⊂ ” of equation(3 . ⊃ ” it suffices to see that ( L ⊗ K L · a ) ∩ R ⊗ K R = a , where a = R ⊗ K R · I ( H ) isthe ideal of R ⊗ K R generated by I ( H ). Consider the inclusions of δ -rings k { G } ⊂ R ⊗ K R ⊂ L ⊗ K L. By Lemma 2.3 extension and contraction are mutually inverse bijections between the set of( δ -)ideals of k { G } and the set of δ -ideals of R ⊗ K R . Similarly, by [AMT09, Proposition 2.3,20. 135] extension and contraction are mutually inverse bijections between the set of ( δ -)ideals of k { G } and the set of δ -ideals of L ⊗ K L . This implies that extension and contraction also aremutually inverse bijections between the set of δ -ideals of R ⊗ K R and the set of δ -ideals of L ⊗ K L .In particular, ( L ⊗ K L · a ) ∩ R ⊗ K R = a .It remains to prove the identity (3.2). So let M be an intermediate δσ -field of L | K and set H := σ - Gal( L | M ). If S is a k - σ -algebra and τ ∈ G ( S ), then we denote with e τ : k { G } → S theelement of Alg σk ( k { G } , S ) corresponding to τ under G ( S ) ≃ Alg σk ( k { G } , S ). An element τ ∈ G ( S )leaves invariant an element a = r r ∈ L if and only if r ⊗ r − r ⊗ r lies in the kernel of R ⊗ K R = R ⊗ k k { G } id ⊗ e τ −−−→ R ⊗ k S . Thus τ ∈ G ( S ) leaves invariant every element of M , i.e., τ ∈ H ( S ), if and only if b lies in the kernel R ⊗ K R = R ⊗ k k { G } → R ⊗ k S , where b is theideal of R ⊗ K R generated by all elements of the form r ⊗ r − r ⊗ r with r , r ∈ R , r = 0and r r ∈ M . Note that b is precisely the kernel of R ⊗ K R → M R ⊗ M M R . In particular, b is a δσ -ideal. It follows from Lemma 2.3 that b = R ⊗ k ( b ∩ k { G } ). So τ ∈ G ( S ) lies in H ( S ) if andonly if b ∩ k { G } ⊂ ker e τ . This means that I ( H ) = b ∩ k { G } . So I ( H ) = b ∩ k { G } = ker( R ⊗ K R → M R ⊗ M M R ) ∩ k { G } = ker( L ⊗ K L → L ⊗ M L ) ∩ k { G } . Theorem 3.2 ( σ -Galois correspondence) . Let L | K be a σ -Picard-Vessiot extension with σ -Galoisgroup G = σ - Gal( L | K ) . Then there is an inclusion reversing bijection between the set of interme-diate δσ -fields M of L | K and the set of σ -closed subgroups H of G given by M σ - Gal( L | M ) and H L H . Proof.
We know from [AMT09, Theorem 2.6, p. 136] that the assignments M ker( L ⊗ K L → L ⊗ M L ) ∩ k { G } and I
7→ { a ∈ L | ⊗ a − a ⊗ ∈ L ⊗ K L · I } are inverse to each other, and yield a bijection between the set of all intermediate δ -fields M of L | K and all Hopf-ideals I of k { G } . The claim thus follows from Lemma 3.1. Theorem 3.3 (Second fundamental theorem of σ -Galois theory) . Let L | K be a σ -Picard-Vessiotextension with σ -Galois group G . Let K ⊂ M ⊂ L be an intermediate δσ -field and H ≤ G a σ -closed subgroup of G such that M and H correspond to each other in the σ -Galois correspondence.Then M is a σ -Picard-Vessiot extension of K if and only if H is normal in G . If this is thecase, the σ -Galois group of M | K is the quotient G/H . (See Definition A.41 for the definition andTheorem A.43 for the existence of the quotient
G/H .)Proof.
We first assume that M | K is σ -Picard-Vessiot. Let R ⊂ L denote the σ -Picard-Vessiot ringof L | K and R ′ ⊂ M the σ -Picard-Vessiot ring of M | K . First of all, we need to convince ourselvesthat R ′ ⊂ R : The ring compositum RR ′ inside L is a σ -Picard-Vessiot ring contained in L withquotient field L . Indeed, if R is a σ -Picard-Vessiot ring for δ ( y ) = Ay with fundamental solutionmatrix Y ∈ Gl n ( R ) and R ′ is a σ -Picard-Vessiot ring for δ ( y ) = A ′ y with fundamental solutionmatrix Y ′ ∈ Gl n ′ ( R ′ ), then RR ′ is a σ -Picard-Vessiot ring for δ ( y ) = (cid:18) A A ′ (cid:19) y with fundamental solution matrix (cid:18) Y Y ′ (cid:19) ∈ Gl n + n ′ ( RR ′ ) . Since R is the only σ -Picard-Vessiot ring inside L | K with quotient field L by Lemma 1.11, it followsthat RR ′ = R , i.e., R ′ ⊂ R .Set G ′ = σ - Gal( M | K ) and let S be a k - σ -algebra. Because R ′ ⊂ R and R ′ is a σ -Picard-Vessiot ring we see that every automorphism τ ∈ G ( S ) = Aut δσ ( R ⊗ k S | K ⊗ k S ) restricts to an21utomorphism τ ′ ∈ G ′ ( S ) = Aut δσ ( R ′ ⊗ k S | K ⊗ k S ). This defines a morphism φ : G → G ′ of group k - σ -schemes. Because the quotient field of R ′ is equal to M , it is clear that H = σ - Gal( L | M ) isthe kernel of φ : G → G ′ . In particular, H is normal in G .Thus, by Corollary A.44, to see that φ : G → G ′ is the quotient morphism of G modulo H itsuffices to see that φ ∗ is injective. To get an explicit description of φ ∗ : k { G ′ } → k { G } , one has totake S := k { G } and to chase id ∈ Alg δσk ( k { G } , S ) ≃ G ( S ) = G ( k { G } ) through the identifications ofthe proof of Proposition 2.5. One finds that φ ∗ is obtained from the inclusion R ′ ⊗ K R ′ ֒ → R ⊗ K R by taking δ -constants. I.e., φ ∗ : k { G ′ } = ( R ′ ⊗ K R ′ ) δ ֒ → ( R ⊗ K R ) δ = k { G } . So clearly φ ∗ isinjective and we conclude that G ′ = G/H .It remains to see that M | K is σ -Picard-Vessiot if H is normal in G . Let A ∈ K n × n and Y ∈ Gl n ( L ) such that L | K is a σ -Picard-Vessiot extension for δ ( y ) = Ay with fundamentalsolution matrix Y . As in Proposition 2.15, set L d = K (cid:0) Y, . . . , σ d ( Y ) (cid:1) for d ≥ G asa σ -closed subgroup of Gl n,k . The Galois group of L d | K equals G [ d ] and H [ d ] is a normal closedsubgroup scheme of G [ d ] by Lemma A.39. Now it follows from the (classical) second fundamentaltheorem of Galois theory (see [vdPS03, Corollary 1.40, p. 31] for the case of algebraically closedconstants or [AMT09, Theorem 2.11, p. 138] for the general case) applied to H [ d ] E G [ d ] that L H [ d ] d | K is Picard-Vessiot.An element a ∈ L d is invariant under H if and only if it is invariant under H [ d ]. (This followsfor example from Lemma 3.1.) Therefore L H [ d ] d = L H ∩ L d = M ∩ L d and M ∩ L d is a Picard-Vessiotextension of K .An intermediate σ -field of a finitely σ -generated σ -field extension is again finitely σ -generated([Lev08, Theorem 4.4.1, p. 292]). Thus we can find a , . . . , a m ∈ M such that M = K h a , . . . , a m i σ .Now choose d ′ ≥ a , . . . , a m ∈ L d ′ . Because M ∩ L d ′ | K is Picard-Vessiot, there existmatrices A ′ ∈ K n ′ × n ′ and Y ′ ∈ Gl n ′ ( M ∩ L d ′ ) such that M ∩ L d ′ | K is a Picard-Vessiot extensionfor δ ( y ) = A ′ y with fundamental solution matrix Y ′ .We have a , . . . , a m ∈ M ∩ L d ′ = K ( Y ′ ) and so M = K h a , . . . , a m i σ ⊂ K h Y ′ i σ . Thus, M = K h Y ′ i σ is a σ -Picard-Vessiot extension for δ ( y ) = A ′ y since M δ ⊂ L δ = k . Corollary 3.4.
Let L | K be a σ -Picard-Vessiot extension and K ⊂ M ⊂ L an intermediate δσ -field such that M | K is a σ -Picard-Vessiot extension. If R is the σ -Picard-Vessiot ring of L | K then R ∩ M is the σ -Picard-Vessiot ring of M | K .Proof. Let R ′ ⊂ M denote the σ -Picard-Vessiot ring of M | K . We have already seen at thebeginning of the proof of Theorem 3.3 that R ′ ⊂ R ∩ M . To see that R ′ = R ∩ M it suffices tonote that the corresponding statement is true if we forget σ . (See the proof of [AMT09, Theorem2.8, p. 137]). σ -separability Let R be a σ -ring. There are some natural conditions which we can impose on R : • R is σ -reduced, i.e., σ : R → R is injective. • R is perfectly σ -reduced, i.e., if f ∈ R and α , . . . , α n ∈ N such that σ α ( f ) · · · σ α n ( f ) = 0,then f = 0. • R is a σ -domain, i.e., R is an integral domain and σ -reduced.The importance of perfectly σ -reduced σ -rings stems from the fact that the finitely σ -generated,perfectly σ -reduced k - σ -algebras are precisely the σ -coordinate rings of the classical σ -varieties oversome σ -field k . See [Lev08, Section 2.6]. σ -domains correspond to irreducible σ -varieties.The main point of this section is to understand the implications on the σ -Picard-Vessiot exten-sion L | K if we impose one of the above conditions on the σ -coordinate ring k { G } of the σ -Galois22roup G = σ - Gal( L | K ). We also use this insight to explain what remains of the σ -Galois corre-spondence if one naively insists that the σ -Galois group of a σ -Picard-Vessiot extension L | K is theactual automorphism group Aut δσ ( L | K ) of L | K .To see the complete picture we need to study the above properties under extension of the base σ -field. Definition 4.1.
Let k be a σ -field and R a k - σ -algebra. We say that R is • σ -separable over k if R ⊗ k k ′ is σ -reduced; • perfectly σ -separable over k if R ⊗ k k ′ is perfectly σ -reduced; • σ -regular over k if R ⊗ k k ′ is a σ -domain;for every σ -field extension k ′ of k . Definition 4.2.
Let k be a σ -field and G a σ -algebraic group over k . We say that G is absolutely σ -reduced/perfectly σ -reduced/ σ -integral if k { G } is σ -separable/perfectly σ -separable/ σ -regular over k . The properties introduced above are studied in some more detail in Section A.6 of the appendix.Below we make use of these results. See the table after Definition A.18 for an overview of thenomenclature.
Proposition 4.3.
Let L | K be a σ -Picard-Vessiot extension with σ -field of δ -constants k and σ -Galois group G . Then (i) L is σ -separable over K if and only if G is absolutely σ -reduced. In particular, if k is inversive,then L is σ -separable over K if and only if G is σ -reduced. (ii) L is perfectly σ -separable over K if and only if G is absolutely perfectly σ -reduced. In partic-ular, if k is algebraically closed and inversive, then L is perfectly σ -separable over K if andonly if G is perfectly σ -reduced. (iii) L is σ -regular over K if and only if G is absolutely σ -integral. In particular, if k is alge-braically closed and inversive, then L | K is σ -regular if and only if G is σ -integral.Proof. We give a simultaneous proof of all the statements. Let R denote the σ -Picard-Vessiotring of L | K . Let L ∗ denote the inversive closure of L ([Lev08, Def. 2.1.6, p. 109]) and let L ′ denote an algebraic closure of L ∗ , equipped with an extension of σ . Then L ′ is an algebraicallyclosed and inversive σ -field. The fundamental isomorphism R ⊗ K R ≃ R ⊗ k k { G } extends to L ′ ⊗ K R ≃ L ′ ⊗ k k { G } .Assume that L is σ -separable/perfectly σ -separable/ σ -regular over k . Then also R is σ -separable/perfectly σ -separable/ σ -regular over k . Therefore L ′ ⊗ K R = L ′ ⊗ k k { G } is σ -reduced/perfectly σ -reduced/ σ -integral. It follows from Lemma A.13 that G is absolutely σ -reduced/perfectly σ -reduced/ σ -integral.Conversely, if G is absolutely σ -reduced/perfectly σ -reduced/ σ -integral, then R must be σ -separable/perfectly σ -separable/ σ -regular over K . It follows from Lemma A.16 that L is σ -separable/perfectly σ -separable/ σ -regular over K . The “in particular” statements are clear fromCorollary A.19. Corollary 4.4.
Let K be δσ -field with σ : K → K an automorphism and let L | K be a σ -Picard-Vessiot extension with σ -Galois group G . Then L is σ -separable over K and G is absolutely σ -reduced.Proof. By Corollary A.14 (i), since K is inversive, L | K is σ -separable.23 .1 A “naive” point of view: perfect σ -separability The purpose of this subsection is to explain what remains of the σ -Galois correspondence if onenaively insists that the σ -Galois group of a σ -Picard-Vessiot extension L | K is the actual automor-phism group Aut δσ ( L | K ) of L | K . Such an approach, closer to [CS06] and [HS08], is in principlepossible. It has the advantage of being notationally more convenient. For example, Aut δσ ( L | K )acts on L whereas σ - Gal( L | K ) is acting (functorially) only on R , the σ -Picard-Vessiot ring of L | K . One disadvantage is that one must assume that the δ -constants are σ -closed. In the caseof differential parameters, this is essentially the only disadvantage: If k is a δ -closed δ -field, thenthe δ -closed subgroups of Gl n ( k ) are the same as the δ -closed subgroups of Gl n,k . (Because incharacteristic zero every Hopf-algebra is reduced.) However, in our case, the case of a differenceparameter, there are more σ -closed subgroups of Gl n,k than σ -closed subgroups of Gl n ( k ) (even if k is σ -closed). See the introduction for an example. So only a certain part of the general σ -Galoiscorrespondence (Theorem 3.2) will remain if we replace σ - Gal( L | K ) with Aut δσ ( L | K ). Let usillustrate this phenomenon with an example. Example 4.5.
Let L | K be the σ -Picard-Vessiot extension from Example 2.10. Then Aut δσ ( L | K ) = C and so, in the “naive” sense, we only have two groups on the group side, namely C and thetrivial group. The trivial group corresponds to L and the intermediate δσ -field of L | K fixed by C is L C = K ( √ x + i p x + j | i, j ≥ ⊂ L. All the other intermediate δσ -fields of L | K are “lost”. For example, the intermediate δσ -field K ( √ x + 1 , √ x + 2 , . . . ) ⊂ L , which corresponds to the σ -closed subgroup H of σ - Gal( L | K ) given by H ( S ) = { g ∈ S × | g = 1 , σ ( g ) = 1 } for any k - σ -algebra S , does not appear in this correspondence.Let L | K be a σ -Picard-Vessiot extension with σ -Picard-Vessiot ring R , σ -field of δ -constants k and σ -Galois group G = σ - Gal( L | K ). As G ( k ) = Aut δσ ( R | K ) = Aut δσ ( L | K )the problem essentially boils down to “When can a σ -algebraic k - σ -scheme X be recovered fromits k -rational points X ( k )?” But this is well known, it will be possible if k is “big enough” and X is perfectly σ -reduced.By Proposition 4.3 (ii), the (absolutely) perfectly σ -reduced subgroups of G correspond tointermediate δσ -fields M of L | K such that L | M is perfectly σ -separable. We shall give a moreexplicit characterization of these intermediate δσ -fields. We first need a simple lemma. Lemma 4.6.
Let L | K be a σ -Picard-Vessiot extension and let K ′ denote the relative algebraicclosure of K inside L . Assume that k = K δ is algebraically closed. Then the field extension K ′ | K is Galois.Proof. As in Proposition 2.15, we can write L | K as a directed union of Picard-Vessiot extensions L d | K ( d ≥ k = ( L d ) δ is algebraically closed, we know that the relative algebraicclosure of K in L d is a Galois extension of K ([vdPS03, Proposition 1.34, p. 25]). Thus K ′ is thedirected union of Galois extension of K . So K ′ is itself Galois over K . Lemma 4.7.
Let L | K be a σ -Picard-Vessiot extension with σ -Galois group G . Assume that k = K δ is algebraically closed and inversive. Then G is perfectly σ -reduced if and only if L | K satisfies thefollowing properties: (i) L is σ -separable over K . (ii) The relative algebraic closure K ′ of K inside L is a finite field extension of K . (iii) Every (field) automorphism of K ′ | K commutes with σ : K ′ → K ′ . In the sense of scheme theory. roof. Assume that G is perfectly σ -reduced. As k is algebraically closed and inversive, it followsfrom Proposition 4.3 (ii) that L | K is perfectly σ -separable. A fortiori L | K is σ -separable. Because k is algebraically closed, we know from Lemma 4.6 that K ′ | K is Galois. We can thus apply LemmaA.23 to conclude that L | K satisfies conditions (ii) and (iii).The converse direction is similar: If L | K satisfies conditions (i), (ii) and (iii) then it followsfrom Lemma A.23 that L | K is perfectly σ -separable. So G is perfectly σ -reduced.It seems interesting to note that the δ -analogs of conditions (ii) and (iii) are automaticallysatisfied: If L | K is a finitely δ -generated extension of δ -fields then the relative algebraic closure K ′ of K inside L is finite and every automorphism of K ′ | K commutes with δ ([Kol73, Corollary2, Chapter II, Section 11, p. 113] and [Kol73, Lemma 1, Chapter II, Section 2, p. 90]).Let L | K be a σ -Picard-Vessiot extension for δ ( y ) = Ay with fundamental solution matrix Y ∈ Gl n ( L ). For the rest of this section, we will assume that k = K δ is σ -closed. Every automorphism τ ∈ Aut δσ ( L | K ) is given by a matrix [ τ ] Y ∈ Gl n ( k ) satisfying τ ( Y ) = Y [ τ ] Y . The mappingAut δσ ( L | K ) → Gl n ( k ) , τ [ τ ] Y is an injective morphism of groups. We shall henceforth identify Aut δσ ( L | K ) with the image ofthis embedding.The σ -closed subsets of Gl n ( k ) are defined as in Section A.3, i.e., as the solution sets of systemsof σ -polynomials in the matrix entries. Of course Aut δσ ( L | K ) is σ -closed in Gl n ( k ). Indeed,with the notation of the proof of Proposition 2.5, Aut δσ ( L | K ) is the solution set of the kernel of k { X, X ) } σ → k { Z, Z ) } σ , X Z . If e H is a σ -closed subgroup of Aut δσ ( L | K ) we set L e H := { a ∈ L | h ( a ) = a ∀ h ∈ e H } . Lemma 4.8.
Let L | K be a σ -Picard-Vessiot extension for δ ( y ) = Ay with fundamental solutionmatrix Y ∈ Gl n ( L ) and σ -Galois group G = σ - Gal( L | K ) . Assume that k = K δ is σ -closed. Theassignment H H ( k ) defines a bijection between the set of the σ -closed subgroups of G that areperfectly σ -reduced and the set of the σ -closed subgroups of Aut δσ ( L | K ) . Moreover, if H ≤ G isperfectly σ -reduced, then L H = L H ( k ) .Proof. The statement about the bijection follows from Lemma A.9.Let H be a perfectly σ -reduced σ -closed subgroup of G and let R denote the σ -Picard-Vessiotring of L | K . By definition L H ⊂ L H ( k ) . For h ∈ H ( k ) ⊂ G ( k ), let ev h : k { G } → k denotethe evaluation map. With the notation of the proof of Proposition 2.5 it is given by Z [ h ] Y .Then, for r ∈ R , h ( r ) is the image of 1 ⊗ r under R ⊗ K R = R ⊗ k k { G } id · ev h −−−−→ R . We also set m h = ker(ev h ).Let a ∈ L . We may write a = r r with r , r ∈ R , r = 0. We have h ( a ) = a if and only if r ⊗ r − r ⊗ r lies in the kernel of R ⊗ K R = R ⊗ k k { G } id · ev h −−−−→ R . Thus, a is invariant under H ( k ) if and only if r ⊗ r − r ⊗ r lies in R ⊗ k ( T h ∈ H ( k ) m h ).On the other side, by Lemma 3.1, a is invariant under H if and only if r ⊗ r − r ⊗ r lies in R ⊗ k I ( H ), where I ( H ) denotes the defining ideal of H in G . Thus L H = L H ( k ) because I ( H ) = T h ∈ H ( k ) m h by Lemma A.8.We now arrive at the reduced version of the σ -Galois correspondence which results if one wantsto avoid the use of schemes. Proposition 4.9.
Let L | K be a σ -Picard-Vessiot extension. Assume that k = K δ is a σ -closed σ -field. The assignments M Aut δσ ( L | M ) and H L H define mutually inverse bijections betweenthe set of all intermediate δσ -fields M of L | K such that L | M is perfectly σ -separable and the setof all σ -closed subgroups H of Aut δσ ( L | K ) .Moreover, for an intermediate δσ -field M of L | K , the extension L | M is perfectly σ -separableif and only if the following assertions are satisfied: L is σ -separable over M . (ii) The relative algebraic closure M ′ of M inside L is a finite field extension of M . (iii) Every automorphism of M ′ | M commutes with σ : M ′ → M ′ .Proof. The statement about the bijections follows from the general σ -Galois correspondence (The-orem 3.2) together with Proposition 4.3 (ii) and Lemma 4.8. The second statement follows fromLemma 4.7.Note that in the setting of Proposition 4.9 one has L Aut δσ ( L | K ) = K if and only if L | K isperfectly σ -separable. A Appendix: Difference algebraic groups
While differential algebraic groups, i.e., group objects in the category of differential varieties, area classical topic in differential algebra (see e.g. [Cas72], [Kol85], [Bui92]), their difference analoghas been neglected by the founding fathers of difference algebra. It appears that almost all resultspertaining to groups defined by algebraic difference equations are relatively recent, due to modeltheorists and motivated by number-theoretic applications. See [CH99], [CHP02], [Hru01], [Cha97],[SV99], [KP02], [KP07], [CH].Even though, both, the notion of a group definable in ACFA and our notion of a differencealgebraic group (Definition A.31), give precise meaning to the idea of a group defined by algebraicdifference equations, none of these notions encompasses the other. Our notion agrees with thenotion of a linear M -group in [Kam12], for a suitable choice of M . Also, a difference algebraicgroup in our sense, such that its coordinate ring is finitely generated as an algebra , is essentiallythe same thing as an affine algebraic σ -group in the sense of [KP07].The main purpose of this appendix is to provide a brief introduction to difference algebraicgroups, suitable for the applications in the main text. A more systematic and complete accountwill eventually be given by the third author. Standard references for difference algebra are [Coh65]and the more recent [Lev08]. Many ideas can also be found in [Hru04]. We consider most of theconstructions presented in this appendix as “well-known”. However, it is sometimes difficult to pindown suitable references. A.1 Some terminology from difference algebra
Throughout the text we use some basic notions from difference algebra. For the convenienceof the reader not well acquainted with difference algebra, we collect here some conventions andterminology.All rings are commutative with identity. A difference ring (or σ -ring for short) is a ring R together with a ring endomorphism σ : R → R . Algebraic attributes (e.g. Noetherian) areunderstood to apply to the underlying ring. Attributes that apply to the difference structure areusually prefixed with σ (e.g. finitely σ -generated). The expression σ is understood to be theidentity.A morphism of σ -rings is a morphism of rings that commutes with σ . Let R be a σ -ring. Byan R - σ -algebra , we mean a σ -ring S together with a morphism R → S of σ -rings. A morphismof R - σ -algebras is a morphism of R -algebras that is also a morphism of σ -rings. If S and S ′ are R - σ -algebras, we write Alg σR ( S, S ′ )for the set of R - σ -algebra morphisms from S to S ′ .An R - σ -subalgebra S ′ of an R - σ -algebra S is an R -subalgebra such that the inclusion morphism S ′ → S is a morphism of σ -rings. The tensor product S ⊗ R S of two R - σ -algebras S and S naturally carries the structure of an R - σ -algebra by virtue of σ ( s ⊗ s ) = σ ( s ) ⊗ σ ( s ). Even in σ -dimension zero this will rarely happen. k be a σ -field, i.e., a σ -ring whose underlying ring is a field. Let R be a k - σ -algebra and B a subset of R . The smallest k - σ -subalgebra of R that contains B is denoted with k { B } σ andcalled the k - σ -subalgebra σ -generated by B . As a k -algebra it is generated by B, σ ( B ) , . . . . If thereexists a finite subset B of R such that R = k { B } σ , we say that R is finitely σ -generated over k .The k - σ -algebra k { x } σ = k { x , . . . , x n } of σ -polynomials over k in the σ -variables x , . . . , x n isthe polynomial ring over k in the variables x , . . . , x n , σ ( x ) , . . . , σ ( x n ) , . . . , with an action of σ assuggested by the names of the variables.Let k be a σ -field. A σ -field extension k ′ of k is a σ -field containing k such that the inclusionmap is a morphism of σ -rings. We also say that k is a σ -subfield of k ′ . If B ⊂ k ′ , the smallest σ -field extension of k inside k ′ that contains B is denoted with k h B i σ . As a field extension of k itis generated by B, σ ( B ) , . . . . We say that k ′ is a finitely σ -generated σ -field extension of k if thereexists a finite subset B of k ′ such that k ′ = k h B i σ .Let R be a σ -ring. A σ -ideal a of R is an ideal a ⊂ R such that σ ( a ) ⊂ a . Then R/ a is naturallya σ -ring. Let B be a subset of R . We denote by [ B ] the σ -ideal generated by B in R . As an ideal itis generated by B, σ ( B ) , . . . . A σ -ideal a of R is called reflexive if σ − ( a ) = a , i.e., σ ( r ) ∈ a implies r ∈ a . A σ -ideal a of R is called perfect if σ α ( r ) · · · σ α n ( r ) ∈ a implies r ∈ a for all r ∈ R , n ≥ α , . . . , α n ≥
0. A σ -ideal q of R is called σ -prime if it is a prime ideal and reflexive. Notethat this property is stronger than being a prime σ -ideal. One can show that the perfect σ -idealsare precisely the intersections of σ -prime ideals.A σ -ring R is called inversive if σ : R → R is an automorphism. A σ -ring R is called σ -reduced if σ : R → R is injective. (Equivalently, the zero ideal is reflexive.) We say that R is perfectly σ -reduced if the zero ideal of R is perfect. If the zero ideal is σ -prime we say that R is a σ -domain .This is equivalent to saying that R is an integral domain with σ : R → R injective. A.2 σ -schemes Throughout the appendix k denotes an arbitrary σ -field. Because the main text deals with deriva-tions, we have made it a general assumption that all fields are of characteristic zero. However, thisappendix does not require the characteristic zero assumption. All products are understood to beproducts over k .It is widely recognized that a functorial approach to algebraic groups has many benefits([Wat79], [DG70], [Mil12]). Here we will adopt a similar point of view. Definition A.1.
Let k be a σ -field. A k - σ -scheme (or σ -scheme over k ) is a (covariant) functorfrom the category of of k - σ -algebras to the category of sets which is representable. Thus a functor X from the category of k - σ -algebras to the category of sets is a k - σ -scheme if and only if thereexists a k - σ -algebra k { X } and an isomorphism of functors X ≃ Alg σk ( k { X } , − ) . By the Yoneda lemma, the k - σ -algebra k { X } is uniquely determined up to unique k - σ -isomorphisms.We call it the σ -coordinate ring of X . A morphism of k - σ -schemes is a morphism of functors.If φ : X → Y is a morphism of k - σ -schemes, we denote the dual morphism of k - σ -algebras with φ ∗ : k { Y } → k { X } .A k - σ -scheme X is called σ -algebraic (over k ) if k { X } is finitely σ -generated over k . We saythat a k - σ -scheme X is σ -reduced/perfectly σ -reduced/ σ -integral if k { X } is σ -reduced/perfectly σ -reduced/a σ -domain. It would be somewhat more accurate to add the word “affine” into the above definition. How-ever, to avoid endless iterations of the word “affine” we make the following convention.
Convention:
All schemes and σ -schemes considered are affine. The above definition does not agree with the definition of a difference scheme given in [Hru04].The approach presented here is essentially equivalent to the approach in [MS11]. The classical In essence this is due to the fact that, starting with a difference ring R , one can not recover R from the globalsections on Spec σ ( R ). Indeed, if the worst comes to the worst, Spec σ ( R ), the set of σ -prime ideals of R , is empty. σ -reduced, σ -algebraic k - σ -schemes. Remark A.2.
By the Yoneda lemma, the category of k - σ -schemes is anti-equivalent to the cate-gory of k - σ -algebras. Definition A.3.
Let X be a k - σ -scheme. By a σ -closed σ -subscheme Y ⊂ X , we mean a subfunc-tor Y of X which is represented by k { X } / I ( Y ) for some σ -ideal I ( Y ) of k { X } . To be precise, the re-quirement is that there exists a σ -ideal I ( Y ) of k { X } and an isomorphism Y ≃ Alg σk ( k { X } / I ( Y ) , − ) such that Y ≃ (cid:15) (cid:15) (cid:31) (cid:127) / / X ≃ (cid:15) (cid:15) Alg σk ( k { X } / I ( Y ) , − ) (cid:31) (cid:127) / / Alg σk ( k { X } , − ) commutes. The ideal I ( Y ) of k { X } is uniquely determined by Y and vice versa. We call it the defining ideal of Y (in k { X } ).A morphism of k - σ -schemes φ : Y → X is called a σ -closed embedding if it induces an isomor-phism of Y with a σ -closed σ -subscheme of X . This is equivalent to saying that φ ∗ : k { X } → k { Y } is surjective. The reader displeased by the apparent foolery of the above definitions should indulge in thefollowing example. In principle we are only interested in the situation described in this example.
Example A.4.
Affine n -space over k (or difference affine n -space over k , if we want to be veryprecise) is the k - σ -scheme A nk such that A nk ( S ) = S n for every k - σ -algebra S . It is represented by k { x } σ = k { x , . . . , x n } σ – the σ -polynomial ring over k in the σ -variables x , . . . , x n .Let F ⊂ k { x } σ be a system of algebraic difference equations. For any k - σ -algebra S , weconsider the S -rational solutions V S ( F ) := { a ∈ S n | p ( a ) = 0 for all p ∈ F } of F in S n . The functor X defined by X ( S ) = V S ( F ) is a σ -closed σ -subscheme of A nk . It isrepresented by k { X } = k { x } σ / [ F ]. Here [ F ] denotes the difference ideal of k { x } σ generated by F .Note that the defining ideal I ( X ) of X in k { x } σ equals[ F ] = { p ∈ k { x } σ | p ( a ) = 0 for all a ∈ X ( S ) and all k - σ -algebras S } . Moreover, every σ -closed σ -subscheme of A nk is of the above described form.If X is a σ -algebraic σ -scheme over k , then choosing a σ -closed embedding of X into A nk isequivalent to specifying n generators of k { X } as k - σ -algebra. Lemma A.5.
Let φ : X → Y be a morphism of k - σ -schemes and Z ⊂ Y a σ -closed σ -subscheme.By setting φ − ( Z )( S ) = φ ( S ) − ( Z ( S )) for every k - σ -algebra S , we can naturally define a σ -closed σ -subscheme φ − ( Z ) of X (the inverse image of Z ). Indeed, φ − ( Z ) is the σ -closed σ -subschemeof X defined by the ideal of k { X } generated by φ ∗ ( I ( Z )) .Proof. Let a denote the ideal of k { X } generated by φ ∗ ( I ( Z )). Note that a is a σ -ideal. For ψ ∈ X ( S ) = Alg σk ( k { X } , S ) we have φ ( S )( ψ ) ∈ Z ( S ) ⇔ I ( Z ) ⊂ ker( φ ( S )( ψ )) = ker( ψ ◦ φ ∗ ) ⇔ φ ∗ ( I ( Z )) ⊂ ker ψ. Thus ψ ∈ φ − ( Z )( S ) if and only if a ⊂ ker ψ . This means that φ − ( Z ) is the σ -closed σ -subschemeof X defined by a . 28 .3 The semi-classical point of view The classical set-up for difference algebraic geometry, as it can be found in the standard textbooks[Coh65] and [Lev08], is in spirit close to the “Foundations of algebraic geometry” as laid down byAndr´e Weil. The story roughly runs as follows: Suppose we want to study σ -algebraic equationsover a fixed σ -field k . Usually k will not contain “enough” solutions, so one has to look for solutionsin σ -field extensions of k . One fixes a family of σ -overfields of k which is “large enough”, calledthe universal system of σ -overfields of k ([Lev08, Definition 2.6.1, p. 149]). A difference varietyover k is then the set of solutions in the universal system of σ -overfields of k of some set of σ -polynomials with coefficients in k . There is a one-to-one correspondence between the σ -varietiesdefined by σ -polynomials in the σ -variables x , . . . , x n and the perfect σ -ideals of the σ -polynomialring k { x , . . . , x n } σ ([Lev08, Theorem 2.6.4, p. 151]).It is a characteristic feature of difference algebra that one really needs to consider a family of σ -overfields of k , i.e., in general one can not find one big σ -overfield of k containing “enough”solutions. However, if we assume that k itself is “large enough”, we can discard the universalfamily and we arrive at a setting analogous to [Har77, Chapter I]. This is what we mean with thesemi-classical point of view, it is usually adopted by model theorists. See [Mac97],[CH99],[CHP02].We shall now outline very briefly the semi-classical set-up. The results below are used in Section4. We start by recalling the precise meaning of “large enough”: Definition A.6. A σ -field k is called σ -closed if for every finitely σ -generated k - σ -algebra R whichis a σ -domain, there exists a morphism R → k of k - σ -algebras. In other words, a σ -field k is σ -closed if and only if every system of algebraic difference equationsover k , which has a solution in a σ -field extension of k , already has a solution in k . The σ -closed σ -fields are also called models of ACFA. One has to exercise some caution: If R is a finitely σ -generated k - σ -algebra over a σ -closed σ -field k , there need not exist a morphism R → k . In fact,we have the following: Lemma A.7.
Let k be a σ -closed σ -field and R a finitely σ -generated k - σ -algebra. The followingstatements are equivalent: (i) There exists a morphism R → k of k - σ -algebras. (ii) There exists a σ -prime ideal in R .Proof. The implication (i) ⇒ (ii) is clear since the kernel of a k - σ -morphism R → k is a σ -primeideal. Conversely, if q is a σ -prime ideal of R , then R/ q is a σ -domain, and since k is σ -closed thereexists a k - σ -morphism R/ q → k which we can compose with R → R/ q to obtain a k - σ -morphism R → k .A maximal σ -ideal, i.e., a maximal element in the set of all proper σ -ideals ordered by inclusion,need not be prime. By a maximal σ -prime ideal , we mean a σ -prime ideal which is maximal inthe set of all σ -prime ideals ordered by inclusion. If R is a finitely σ -generated k - σ -algebra over a σ -closed σ -field k and q ⊂ R a maximal σ -prime ideal, then R/ q = k . So, in this case, a maximal σ -prime ideal is maximal as an ideal.The following lemma is surely well-known. For lack of a suitable reference we include a proof. Lemma A.8.
Let k be a σ -closed σ -field, R a finitely σ -generated k - σ -algebra and a a perfect σ -ideal of R . Then a is the intersection of all maximal σ -prime ideals of R containing a .Proof. Let f ∈ R such that f is contained in every maximal σ -prime ideal containing a . We haveto show that f ∈ a . Suppose f / ∈ a . Let g denote the image of f in R/ a . Since a is perfectand f / ∈ a , the multiplicatively closed subset S of R/ a generated by g, σ ( g ) , . . . does not containzero. So the localization R ′ := S − ( R/ a ) is not the zero ring. Note that R ′ is naturally a σ -ring.Moreover, R ′ = ( R/ a ) { g } σ is finitely σ -generated over k . It is easy to see that the zero ideal of R ′ is perfect. Because every perfect σ -ideal is the intersection of σ -prime ideals ([Coh65, p. 88]),29his implies that there exists a σ -prime ideal in R ′ . Since k is σ -closed we deduce the existence ofa k - σ -morphism R ′ → k from Lemma A.7. Composing with the canonical map R → R ′ this yieldsa k - σ -morphism ψ : R → k . By construction, the kernel of ψ is a maximal σ -prime ideal of R notcontaining f ; a contradiction.Let k be a σ -closed σ -field. A subset of k n is called σ -closed if it is of the form V k ( F ) := { a ∈ k n | p ( a ) = 0 ∀ p ∈ F } for some subset F of k { x , . . . , x n } σ . Equivalently, a subset of k n is σ -closed if it is of the form X ( k ) for some σ -closed σ -subscheme X of A nk . (Cf. Example A.4.) The σ -closed subsets of k n (forsome n ) are sometimes also called σ -varieties .If a is a perfect σ -ideal of k { x , . . . , x n } σ , then we can reinterpret Lemma A.8 as “Every σ -polynomial that vanishes on V k ( a ) must lie in a .” It follows that a V k ( a ) defines a bijectionbetween the σ -closed subsets of k n and the perfect σ -ideals of k { x , . . . , x n } σ . This in turn impliesthe following lemma. Lemma A.9.
Let k be a σ -closed σ -field. The assignment X X ( k ) defines a bijection betweenthe perfectly σ -reduced σ -closed σ -subschemes of A nk and the σ -closed subsets of k n . Let X ⊂ k n and Y ⊂ k m be σ -closed. If one defines a morphism f : X → Y to be a mappinggiven by σ -polynomials then one finds easily that the category of σ -varieties is equivalent to thecategory of perfectly σ -reduced, σ -algebraic k - σ -schemes. A.4 The k - σ -scheme associated with a k -scheme There is a natural way to associate a k - σ -scheme to a scheme over k which formalizes the fact thatsolutions of a system of algebraic equations can be interpreted as solutions of difference equations,i.e., algebraic equations are difference equations. Cf. [Hru04, Section 3.2, p. 23] and [Kam12,Section 3.2, p. 25]. We first treat the algebraic point of view.Let R be a k -algebra and d ≥
0. We set σ d R = R ⊗ k k , where the tensor product is formedby using σ d : k → k on the right hand side. We consider σ d R as k -algebra via the right factor. Soif R = k [ x ] / a then σ d R equals k [ x ] / a ′ where a ′ ⊂ k [ x ] = k [ x , . . . , x n ] denotes the ideal generatedby the polynomials obtained from polynomials from a by applying σ d to the coefficients. Thus if a ∈ k n is a k -rational point of R then σ d ( a ) ∈ k n is a k -rational point of σ d R .There is a natural map ψ d from σ d R = R ⊗ k k to σ d +1 R = R ⊗ k k given by ψ d ( r ⊗ λ ) = r ⊗ σ ( λ ).We set R d = R ⊗ k σ R ⊗ k · · · ⊗ k σ d R. We have natural inclusions R d ֒ → R d +1 of k -algebras and ring morphisms σ d : R d → R d +1 definedby σ d ( r ⊗ · · · ⊗ r d ) = 1 ⊗ ψ ( r ) ⊗ · · · ⊗ ψ d ( r d ). The σ d ’s are not morphisms of k -algebras butmake the diagram R d σ d / / R d +1 k O O σ / / k O O commutative. Now we can define [ σ ] k R as the limit (i.e., the union) of the R d ’s ( d ≥ σ d ’s yields a morphism σ : [ σ ] k R → [ σ ] k R , turning [ σ ] k R into a k - σ -algebra. Theinclusion R = R ֒ → [ σ ] k R is characterized by the following universal property. Lemma A.10.
Let R be a k -algebra. There exists a k - σ -algebra [ σ ] k R together with a morphism ψ : R → [ σ ] k R of k -algebras satisfying the following universal property: For every k - σ -algebra S It is not hard to see that this actually defines a topology on k n . nd every morphism ψ ′ : R → S of k -algebras there exists a unique morphism ϕ : [ σ ] k R → S of k - σ -algebras making R ψ / / ψ ′ (cid:30) (cid:30) ❃❃❃❃❃❃❃❃ [ σ ] k R ϕ } } S commutative.Proof. This follows immediately from the universal property of the tensor product and the limit.Note that if R = k [ x ], the polynomial ring in the variables x = ( x , . . . , x n ), then [ σ ] k R = k { x } σ , the σ -polynomial ring in the σ -variables x = ( x , . . . , x n ), and the inclusion map R → [ σ ] k R is simply saying that a polynomial is a σ -polynomial. Moreover, R d ⊂ k { x } σ is the k -subalgebraof all σ -polynomials of order at most d .More generally, if R = k [ x ] / a then [ σ ] k R = k { x } σ / [ a ]. In particular, if R is finitely generatedas a k -algebra, then [ σ ] k R is finitely generated as a k - σ -algebra.Alternatively R d can be described as the k -subalgebra of [ σ ] k R generated by all elements ofthe form σ i ( r ) for i ≤ d and r ∈ R .If ϕ : R → R ′ is a morphism of k -algebras, then so is R → R ′ → [ σ ] k R ′ , and from the universalproperty we obtain a morphism [ σ ] k ( ϕ ) : [ σ ] k R → [ σ ] k R ′ . We thus obtain a functor [ σ ] k from thecategory of k -algebras to the category of k - σ -algebras.If S is a k - σ -algebra, we denote by S ♯ the underlying k -algebra of S . I.e., ( − ) ♯ is the forgetfulfunctor from k - σ -algebras to k -algebras that forgets σ . For every k -algebra R and every k - σ -algebra S , we have Alg σk ([ σ ] k R, S ) ≃ Alg k ( R, S ♯ ) . In other words, [ σ ] k is left adjoint to ( − ) ♯ .We now return to schemes. If V = Spec( k [ V ]) is a scheme over k , we can define a functor [ σ ] k V from the category of k - σ -algebras to the category of sets by setting([ σ ] k V )( S ) = V ( S ♯ )for every k - σ -algebra S . Then [ σ ] k V is a k - σ -scheme. Indeed, as([ σ ] k V )( S ) = V ( S ♯ ) = Alg k ( k [ V ] , S ♯ ) = Alg σk ([ σ ] k k [ V ] , S )for every k - σ -algebra S , we find that [ σ ] k V is represented by [ σ ] k k [ V ], i.e., k { [ σ ] k V } = [ σ ] k k [ V ].If confusion is unlikely, we shall sometimes denote the k - σ -scheme [ σ ] k V associated with V withthe same letter V . For example, we shall write k { V } instead of k { [ σ ] k V } , A nk instead of [ σ ] k A nk ,as in Example A.4 or Gl n,k instead of [ σ ] k Gl n,k . Note that if V is algebraic over k , then [ σ ] k V is σ -algebraic over k .From a k - σ -scheme X , one can obtain a scheme X ♯ over k by forgetting the σ -structure, i.e., X ♯ = Spec( k { X } ♯ ) or X ♯ = Alg k ( k { X } ♯ , − ). This defines a forgetful functor ( − ) ♯ from thecategory of k - σ -schemes to the category of k -schemes. If V is a scheme over k and X a k - σ -schemethen Hom( X, [ σ ] k V ) ≃ Hom( X ♯ , V ) . So, on schemes, [ σ ] k is right adjoint to ( − ) ♯ . 31 .5 Zariski closures We next introduce the Zariski closures of a σ -closed σ -subscheme of a scheme. Cf. [Hru04, Section4.3].Let V be a k -scheme. For d ≥
0, we set σ d V = V × Spec( k ) Spec( k ), where the morphismon the right hand side is induced from σ d : k → k . Note that if V descents to k σ , i.e., V = V ′ × Spec( k σ ) Spec( k ) for some scheme V ′ over k σ , then σ d V = V . (This, for example, is the casefor V = A nk or V = Gl n,k .)We also set V d = V × σ V × · · · × σ d V. Of course this notation is compatible with the notation from the previous section: If V =Spec( k [ V ]), then σ d V = Spec( σ d k [ V ]) and V d = Spec( k [ V ] d ). Since k { V } = [ σ ] k k [ V ] is theunion of the k -subalgebras k [ V ] d of k { V } we can see [ σ ] k V as the projective limit of the V d ’s.By a σ -closed σ -subscheme X of V , we mean a σ -closed σ -subscheme of [ σ ] k V . By definition, X is given by a σ -ideal I ( X ) of k { V } .We define X [ d ] to be the closed subscheme of V d defined by the ideal I ( X ) ∩ k [ V ] d ⊂ k [ V ] d .Clearly I ( X [ d ]) = I ( X ) ∩ k [ V ] d is the largest ideal of k [ V ] d such that k [ V ] d → k { V } → k { X } = k { V } / I ( X ) factors through k [ V ] d → k [ V ] d / I ( X [ d ]). The geometric significance of this is subsumedin the following definition. Definition A.11.
Let V be a scheme over k and X a σ -closed σ -subscheme of V . For d ≥ , thesmallest closed subscheme X [ d ] of V d such that X ♯ → V d factors through X [ d ] ֒ → V d is called the d -th order Zariski closure of X inside V . The -th order Zariski closure is also called the Zariskiclosure . If the Zariski closure of X inside V is equal to V = V we say that X is Zariski dense in V . The above definition can be subsumed by saying that X [ d ] is the scheme-theoretic image of X ♯ → V d . (Cf. [Har77, Exercise II.3.11 (d), p. 92].)We have natural dominant projections X [ d + 1] → X [ d ] and X ♯ = lim ←− X [ d ]. Example A.12.
Let V = A nk and X the σ -closed σ -subscheme of V defined by a σ -ideal I ( X ) of k { A nk } = k { x } = k { x , . . . , x n } . (Cf. Example A.4.) Then k { X } = k { x } / I ( X ) = k { x } and X [ d ] = Spec( k [ x, σ ( x ) , . . . , σ d ( x )])for every d ≥ k n where k is a σ -closed σ -field. A.6 Some properties related to base extension
In this section, we study the σ -analogs of separable and regular algebras, which, in classical al-gebraic geometry, correspond to absolutely reduced and absolutely integral schemes. Cf. [Hru04,Lemma 3.26, p. 25]. This is used in Section 4.Let k be a σ -field. Recall (Definition 4.1) that a k - σ -algebra R is called σ -separable/perfectly σ -separable/ σ -regular if R ⊗ k k ′ is σ -reduced/perfectly σ -reduced/a σ -domain for every σ -fieldextension k ′ of k . Lemma A.13.
Let k be a σ -field and R a k - σ -algebra. (i) Let k ′ be an inversive σ -field extension of k . Then R is σ -separable over k if and only if R ⊗ k k ′ is σ -reduced. (ii) Let k ′ be an inversive algebraically closed σ -field extension of k . Then R is perfectly σ -separable/ σ -regular over k if and only if R ⊗ k k ′ is perfectly σ -reduced/a σ -domain. This is called the d -th order weak Zariski closure in [Hru04, Section 4.3, p. 32]. roof. Point (i) follows from [Wib10, Proposition 1.5.2, p. 17], cf. [Hru04, Lemma 3.26 (2), p.25]. The case of σ -regularity in (ii) follows from (i) and the fact that R is regular over k if R ⊗ k k ′ is an integral domain. (See e.g. [Bou90, Corollary 1, Chapter 5, §
17, No. 5, A.V.143].)It remains to see that R is perfectly σ -separable if R ⊗ k k ′ is perfectly σ -reduced. So let k ′′ be a σ -field extension of k . We have to show that R ⊗ k k ′′ is perfectly σ -reduced. Because k ′ is algebraically closed there exists a σ -field extension k ′′′ of k containing k ′ and k ′′ (cf. [Lev08,Theorem 5.1.6, p. 313]). If R ⊗ k k ′′′ is perfectly σ -reduced then also R ⊗ k k ′′ ⊂ R ⊗ k k ′′′ is perfectly σ -reduced. Therefore we can assume that k ′ ⊂ k ′′ . As R ⊗ k k ′′ = ( R ⊗ k k ′ ) ⊗ k ′ k ′′ we can reduceto showing that every perfectly σ -reduced k - σ -algebra over an algebraically closed inversive σ -field k is perfectly σ -separable over k . In other words, we may assume that k = k ′ .Let q be a σ -prime ideal of R . Then R/ q is a σ -domain and it follows from the case of σ -regularity proved above that ( R/ q ) ⊗ k k ′′ is a σ -domain. Consequently q ⊗ k ′′ is a σ -prime ideal of R ⊗ k k ′′ . Because R is perfectly σ -reduced, the intersection of all σ -prime ideals of R is the zeroideal ([Lev08, Proposition 2.3.4, p. 122] or [Coh65, End of Section 6, Chapter 3, p. 88]). It followsthat \ q ( q ⊗ k ′′ ) = ( \ q q ) ⊗ k ′′ = (0) ⊂ R ⊗ k k ′′ , where the intersection is taken over all σ -prime ideals q of R . Thus the zero ideal of R ⊗ k k ′′ isthe intersection of σ -prime ideals. This shows that R ⊗ k k ′′ is perfectly σ -reduced. Corollary A.14.
Let R be a k - σ -algebra. (i) If k is inversive, then R is σ -separable over k if and only if R is σ -reduced. (ii) If k is inversive and algebraically closed, then R is perfectly σ -separable/ σ -regular over k ifand only if R is perfectly σ -reduced/a σ -domain.Proof. This is clear from Lemma A.13.
Lemma A.15.
Let R be a σ -ring and S a multiplicatively closed σ -stable subset of R consistingof non-zero divisors. If R is σ -reduced/perfectly σ -reduced/a σ -domain, then so is S − R .Proof. This is a straight forward verification.
Lemma A.16.
Let k be a σ -field and R a k - σ -domain. If R is σ -separable/perfectly σ -separable/ σ -regular over k , then also the quotientfield of R is σ -separable/perfectly σ -separable/ σ -regular over k .Proof. Let k ′ be a σ -field extension of k and S = R r { } the multiplicatively closed subset ofnon-zero divisors of R . Because R is a σ -domain, S is stable under σ and so the quotientfield L of R is naturally a σ -ring. Since R ⊗ k k ′ is σ -reduced/perfectly σ -reduced/a σ -domain it followsfrom Lemma A.15 that also S − ( R ⊗ k k ′ ) is σ -reduced/perfectly σ -reduced/a σ -domain. But L ⊗ k k ′ = S − ( R ⊗ k k ′ ).The property of a k - σ -scheme to be σ -reduced/perfectly σ -reduced/ σ -integral is not stableunder base extension. So we need to supplement these definitions. To speak meaningfully aboutbase extensions of k - σ -schemes we record that: Remark A.17.
The category of k - σ -schemes has products. Indeed, if X and Y are k - σ -schemesthen X × Y is represented by k { X } ⊗ k k { Y } . Proof.
This follows from Remark A.2 and the fact that the tensor product is the coproduct in thecategory of k - σ -algebras.Let X be a k - σ -scheme and k ′ a σ -field extension of k . Let Y denote the k - σ -scheme representedby the k - σ -algebra k ′ . We say that X k ′ := X × Y is obtained from X via the base extension k ′ | k .33 efinition A.18. Let X be a k - σ -scheme. We say that X is absolutely σ -reduced/perfectly σ -reduced/ σ -integral if X k ′ is σ -reduced/perfectly σ -reduced/ σ -integral for every σ -field extension k ′ of k . Thus a k - σ -scheme X is absolutely σ -reduced/perfectly σ -reduced/ σ -integral if and only if k { X } is σ -separable/perfectly σ -separable/ σ -regular over k . Corollary A.14 is reinterpreted as: Corollary A.19.
Let X be a k - σ -scheme. (i) If k is inversive, then X is absolutely σ -reduced if and only if X is σ -reduced. (ii) If k is inversive and algebraically closed, then X is absolutely perfectly σ -reduced/ σ -integralif and only if X is perfectly σ -reduced/ σ -integral. A.6.1 More on perfect σ -separabilityLemma A.20. Let L | K be an extension of σ -fields and let K ′ denote the relative algebraic closureof K inside L . Assume that L | K is σ -separable and separable (as field extension). Then L | K isperfectly σ -separable if and only if K ′ | K is perfectly σ -separable.Proof. Clearly K ′ | K is perfectly σ -separable if L | K is perfectly σ -separable.Assume that K ′ | K is perfectly σ -separable. Let M be a σ -field extension of K . We have toshow that L ⊗ K M = L ⊗ K ′ ( K ′ ⊗ K M ) is perfectly σ -reduced. Because L | K is separable also L | K ′ is separable and since K ′ is relatively algebraically closed in L we see that the field extension L | K ′ is regular. This implies that L ⊗ q is a prime ideal of L ⊗ K ′ ( K ′ ⊗ K M ) for every prime ideal q of K ′ ⊗ K M . As K ′ ⊗ K M is perfectly σ -reduced, the zero ideal of K ′ ⊗ K M is the intersectionof σ -prime ideals. It follows that the zero ideal of L ⊗ K M = L ⊗ K ′ ( K ′ ⊗ K M ) is the intersectionof the prime σ -ideals L ⊗ q , where q runs through the σ -prime ideals of K ′ ⊗ K M . In particular,the zero ideal of L ⊗ K M is the intersection of prime σ -ideals.Since L | K is σ -separable, σ is injective on L ⊗ K M . It is now easy to see that L ⊗ K M isperfectly σ -reduced: Indeed, let a ∈ L ⊗ K M and α , . . . , α n ≥ σ α ( a ) · · · σ α n ( a ) = 0.If q ′ is a prime σ -ideal of L ⊗ K M then σ α ( a ) ∈ q ′ , where α denotes the maximum of the α i .Therefore σ α ( a ) = 0 and it follows that a = 0. Lemma A.21.
Let L | K be a finitely σ -generated perfectly σ -separable extension of σ -fields suchthat the underlying field extension is algebraic. Then L | K is finite.Proof. Let K denote an algebraic closure of K containing L . We can extend σ : L → L to σ : K → K . By assumption there is an L -tuple a such that L = K h a i σ . Because L is algebraic over K , wehave L = K { a } σ . By assumption L ⊗ K K is perfectly σ -reduced. Because L ⊗ K K is a finitely σ -generated K - σ -algebra, it follows from the σ -basis theorem (See [Lev08, Theorem 2.5.5, p. 143and Theorem 2.5.11, p. 147].) that the zero ideal of L ⊗ K K is the intersection of finitely many σ -prime ideals. In particular L ⊗ K K has only finitely many minimal prime ideals. This is onlypossible if L | K is finite. Lemma A.22.
Let L | K be an extension of σ -fields such that the underlying extension of fields isfinite and Galois. Then L | K is perfectly σ -separable if and only if every (field) automorphism of L | K commutes with σ .Proof. Because L | K is Galois, there is a bijection between the prime ideals of L ⊗ K L and theautomorphisms of L | K : If q is a prime ideal of L ⊗ K L then k ( q ) = ( L ⊗ K L ) / q and the inclusions τ and τ into the first and second factor, respectively, are isomorphisms. So τ := τ q := τ − τ isan automorphism of L | K . Conversely, if τ is an automorphism of L | K , then the kernel q = q τ of L ⊗ K L → L, a ⊗ b aτ ( b ) is a prime ideal of L ⊗ K L . The relation between τ and q is determinedby 1 ⊗ a − τ ( a ) ⊗ ∈ q for every a ∈ L . 34et q , . . . , q m denote the prime ideals of L ⊗ K L . They are maximal and minimal and q ∩· · · ∩ q m = (0). We have a mapping q σ − ( q ) from the prime ideals of L ⊗ K L into the primeideals of L ⊗ K L .A σ -ideal a of a σ -ring which is the finite intersection of prime ideals, is perfect if and only ifthe prime ideals minimal above a are σ -prime ideals. (See [Lev08, Proposition 2.3.4, p. 122] or[Coh65, End of Section 6, Chapter 3, p. 88].) It follows that L ⊗ K L is perfectly σ -reduced if andonly if q σ − ( q ) is the identity.Let q be a prime ideal of L ⊗ K L and a ∈ L . Then 1 ⊗ a − τ σ − ( q ) ( a ) ⊗ ∈ σ − ( q ) and so1 ⊗ σ ( a ) − σ ( τ σ − ( q ) ( a )) ⊗ ∈ q . Therefore τ q ( σ ( a )) = σ ( τ σ − ( q ) ( a )) for every a ∈ L . In otherwords, τ q σ = στ σ − ( q ) . If q σ − ( q ) is the identity, then σ commutes with every automorphismof L | K . Conversely, if τ q σ = στ q for every prime ideal q , then στ q = τ q σ = στ σ − ( q ) implies τ q = τ σ − ( q ) and so q = σ − ( q ).In summary, we see that L ⊗ K L is perfectly σ -reduced if and only if σ commutes with everyautomorphism of L | K . If L | K is perfectly σ -separable then L ⊗ K L is perfectly σ -reduced and so σ must commute with every automorphism of L | K .It remains to see that L | K is perfectly σ -separable if σ commutes with every automorphismof L | K . Let M be an inversive algebraically closed σ -field extension of K containing L . Byassumption, we have L ⊗ K L = L ⊕ · · · ⊕ L with σ given by σ ( a ⊕ · · · ⊕ a m ) = σ ( a ) ⊕ · · · ⊕ σ ( a m ).Thus M ⊗ K L = M ⊗ L ( L ⊗ K L ) = M ⊕ · · · ⊕ M is perfectly σ -reduced. It follows from Lemma A.13 that L is perfectly σ -separable over K . Lemma A.23.
Let L | K be a finitely σ -generated σ -separable extension of σ -fields and let K ′ denote the relative algebraic closure of K inside L . Assume that L | K is separable and that K ′ is Galois over K . Then L is perfectly σ -separable over K if and only if K ′ | K is finite and everyautomorphism of K ′ | K commutes with σ .Proof. Assume that L | K is perfectly σ -separable. Then K ′ | K is also perfectly σ -separable. Anintermediate σ -field of a finitely σ -generated σ -field extension is finitely σ -generated ([Lev08, The-orem 4.4.1, p.292]). Therefore K ′ | K is finitely σ -generated. It follows from Lemma A.21 that K ′ | K is finite and from Lemma A.22 that every automorphism of K ′ | K commutes with σ .The reverse direction follows from Lemma A.22 and Lemma A.20. A.7 Difference dimension
Let k be a σ -field. We would like to define a notion of dimension for a k - σ -scheme X which is σ -algebraic over k , i.e., k { X } is finitely σ -generated over k . If X is σ -integral, this is classical:The σ -dimension of X (or k { X } ) is the σ -transcendence degree of the “function field” Quot( k { X } )over k . (See [Lev08, Section 7.1, p. 394].)As in [Hru04, Section 4] one can generalize this definition by consideringsup { σ - trdeg( k ( q ) | k ) | q is a σ -prime ideal of k { X }} . This definition has some drawbacks: Firstly, it does not quite make sense if k { X } has no σ -prime ideals. Secondly, it is not stable under extension of the base σ -field. Our aim here is tointroduce a notion of σ -dimension which agrees with the classical definition if k { X } is σ -integraland which is stable under extension of the base σ -field. It is well-known that the σ -dimensioncan be computed as the leading coefficient of an appropriate dimension polynomial ([Lev08, Def.4.2.21, p. 273]). Here we follow this idea.Let a = ( a , . . . , a m ) be a σ -generating set for k { X } over k . The basic idea is to define the σ -dimension of k { X } over k as the “growth rate” of the sequence d i := dim( k [ a, σ ( a ) , . . . , σ i ( a )]) , i = 0 , , . . . where dim denotes the usual Krull-dimension. There are two difficulties: First we need to makeprecise what we mean by “growth rate”, and then we need to show that the definition is independentof the choice of generators a . 35 roposition A.24. Let k be a σ -field, R a k - σ -algebra and a = ( a , . . . , a m ) a σ -generating setfor R over k . Then lim sup i →∞ (cid:0) dim( k [ a, . . . , σ i ( a )]) / ( i + 1) (cid:1) is independent of the choice of a .Proof. Let a ′ = ( a ′ , . . . , a ′ m ′ ) be another σ -generating of R over k . Then all the components of a ′ lie in k [ a, . . . , σ j ( a )] for some j ∈ N . It follows that k [ a ′ , . . . , σ i ( a ′ )] ⊂ k [ a, . . . , σ j + i ( a )] . We abbreviate d i := dim( k [ a, . . . , σ i ( a )]) and d ′ i := dim( k [ a ′ , . . . , σ i ( a ′ )]). Because of the aboveinclusion d ′ i ≤ d j + i for i ∈ N . Since k [ a, . . . , σ j + i ( a )] can be generated by mj elements over k [ a, . . . , σ i ( a )], we have d j + i ≤ d i + mj . In summary, d ′ i i +1 ≤ d j + i i +1 ≤ d i i +1 + mji +1 . Because lim i →∞ mji +1 = 0 we obtain lim sup i →∞ d ′ i i +1 ≤ lim sup i →∞ d i i +1 . By symmetry the above values are actually equal.Because we want the difference dimension to be an integer we make the following definition.
Definition A.25.
Let k be a σ -field and R a finitely σ -generated k - σ -algebra. We define the σ -dimension of R over k as σ - dim k ( R ) = (cid:22) lim sup i →∞ (cid:0) dim( k [ a, . . . , σ i ( a )]) / ( i + 1) (cid:1)(cid:23) , where ⌊ x ⌋ denotes the largest integer not greater than x and a = ( a , . . . , a m ) is a σ -generating setof R over k . (By Proposition A.24 this definition does not depend on the choice of a .) If X is a k - σ -scheme such that k { X } is finitely σ -generated over k , then we set σ - dim k ( X ) = σ - dim k ( k { X } ) . Lemma A.26.
Let k be a σ -field and R a σ -domain which is finitely σ -generated over k . Then σ - dim k ( R ) = σ - trdeg(Quot( R ) | k ) . Proof.
Assume that a = ( a , . . . , a m ) σ -generates R over k . Set d = σ - trdeg(Quot( R ) | k ). Thereexists an integer e such thatdim( k [ a, . . . , σ i ( a )]) = d ( i + 1) + e for i ≫ . See [Lev08, Def. 4.2.21, p. 273]. Thereforelim sup i →∞ (cid:0) dim( k [ a, . . . , σ i ( a )]) / ( i + 1) (cid:1) = d. Lemma A.27.
Let k be a σ -field and R a finitely σ -generated k - σ -algebra. If k ′ is a σ -fieldextension of k then σ - dim k ′ ( R ⊗ k k ′ ) = σ - dim k ( R ) . Proof.
Assume that a = ( a , . . . , a m ) σ -generates R over k . Then a also σ -generates R ⊗ k k ′ over k ′ . The claim now follows from the fact thatdim( k [ a, . . . , σ i ( a )]) = dim( k [ a, . . . , σ i ( a )] ⊗ k k ′ ) = dim( k ′ [ a, . . . , σ i ( a )]) . emma A.28. Let k be a σ -field and R a finitely generated k -algebra. Then σ - dim k ([ σ ] k R ) = dim( R ) . Proof.
Assume that a = ( a , . . . , a m ) generates R over k . Then a σ -generates [ σ ] k R over k .Moreover, with the notation of Section A.4, for every i ≥ k [ a, . . . , σ i ( a )] = R i . Thereforedim( k [ a, . . . , σ i ( a )]) = dim R i = ( i + 1) dim( R )and σ - dim k ([ σ ] k R ) = lim i →∞ dim( k [ a, . . . , σ i ( a )]) / ( i + 1) = dim( R ).The geometric interpretation of the above lemma is: Corollary A.29.
Let k be a σ -field and V a scheme of finite type over k . Then σ - dim k ([ σ ] k V ) = dim( V ) . Remark A.30.
In the situation of Definition A.25, one can show that if R is a k - σ -Hopf algebra(i.e., the σ -coordinate ring of a σ -algebraic group, see Section A.8 below), thenlim i →∞ (cid:0) dim( k [ a, . . . , σ i ( a )]) / ( i + 1) (cid:1) exists and is an integer. So the floor function and the limes superior are not needed. Since we shallnot require this fact we omit the details. A.8 Group k - σ -schemes We already noted in Remark A.17 that the category of k - σ -schemes has products: If X and Y are k - σ -schemes then X × Y is represented by k { X } ⊗ k k { Y } . There also is a terminal object:the functor sending every k - σ -algebra to a one element set. It is represented by k . Therefore thefollowing definition makes sense. Definition A.31. A group k - σ -scheme is a group object in the category of k - σ -schemes. In otherwords, a group k - σ -scheme is a k - σ -scheme G such that G ( S ) is equipped with a group structurewhich is functorial in S . A morphism of group k - σ -schemes φ : G → H is a morphism of k - σ -schemes such that φ ( S ) : G ( S ) → H ( S ) is a morphism of groups for every k - σ -algebra S .A σ -algebraic group over k is a group k - σ -scheme that is σ -algebraic (over k ). Let G be a group scheme over k . Then [ σ ] k G is a group k - σ -scheme. (This is clear from([ σ ] k G )( S ) = G ( S ♯ ).) By a σ -closed subgroup of G , we mean a σ -closed subgroup of [ σ ] k G . As inSection A.4, we write k { G } instead of k { [ σ ] k G } for the σ -coordinate ring of G . So, for example,the σ -coordinate ring of the general linear group Gl n,k (over k ) is k { Gl n,k } = k { X, X ) } σ . Here X = ( x ij ) ≤ i,j ≤ n is an n × n matrix of σ -indeterminates over k and k { X, X ) } σ is obtainedfrom the σ -polynomial ring k { x ij | ≤ i, j ≤ n } σ by localizing at the multiplicatively closed subsetgenerated by det( X ) , σ (det( X )) , . . . . Example A.32.
Let G be defined by G ( S ) = { g ∈ Gl n ( S ) | gσ ( g ) T = σ ( g ) T g = I } for every k - σ -algebra S , where I denotes the identity matrix of size n . Then G is a σ -closedsubgroup of Gl n,k . Example A.33.
A homogeneous, linear σ -polynomial p = a n σ n ( x ) + · · · + a σ ( x ) + a x ∈ k { x } σ defines a σ -closed subgroup G of the additive group G a,k by G ( S ) = { g ∈ S | p ( g ) = 0 } ≤ G a ( S ) , for any k - σ -algebra S . 37 xample A.34. Given m , m , . . . , m n ∈ Z , we can define a σ -closed subgroup G of the multi-plicative group G m,k by G ( S ) = { g ∈ S × | g m σ ( g ) m · · · σ n ( g ) m n = 1 } ≤ G m ( S ) , for any k - σ -algebra S . Definition A.35. A k - σ -Hopf algebra is a k - σ -algebra equipped with the structure of a Hopf algebraover k such that the Hopf algebra structure maps are morphisms of difference rings. A morphism of k - σ -Hopf algebras is a morphism of Hopf algebras over k which is also a morphism of k - σ -algebras. Remark A.36.
The category of group k - σ -schemes is anti-equivalent to the category of k - σ -Hopfalgebras. Proof.
This is all tautology, cf. [Wat79, Section 1.4].
Definition A.37.
Let G be a group k - σ -scheme. By a σ -closed subgroup H of G (In symbols: H ≤ G .) we mean a σ -closed σ -subscheme H of G such that H ( S ) is a subgroup of G ( S ) for every k - σ -algebra S . We call H normal if H ( S ) is a normal subgroup of G ( S ) for every k - σ -algebra S .In symbols: H E G . Remark A.38. σ -Closed subgroups correspond to σ -Hopf ideals, i.e., Hopf ideals which are dif-ference ideals. Normal σ -closed subgroups correspond to normal σ -Hopf-ideals, i.e., σ -Hopf-idealswhich are normal Hopf ideals.If φ : G → H is a morphism of group k - σ -schemes, then we can define a functor ker( φ ) from k - σ -algebras to groups by setting ker( φ )( S ) = ker( φ ( S )) for every k - σ -algebra S . It follows fromLemma A.5 that ker( φ ) = φ − (1 H ) is a σ -closed σ -subscheme of G . Here 1 H ⊂ H denotes the σ -closed σ -subscheme of the identity element. Obviously ker( φ ) is a σ -closed normal subgroup of G . The question, whether every σ -closed normal subgroup of G is the kernel of some morphism G → H will be answered in Section A.9 below.Let G be a group scheme over k and let d ≥
0. Then σ d G is a group scheme over k and also G d = G × σ G × · · · × σ d G is a group scheme over k . Moreover the natural projection ([ σ ] k G ) ♯ → G d is a morphism of group schemes over k . Lemma A.39.
Let G be a group scheme over k , H a σ -closed subgroup of G and d ≥ . (i) Then H [ d ] is a closed subgroup scheme of G d and H ♯ → H [ d ] is a morphism of group schemesover k . (ii) If N is a normal σ -closed subgroup of H , then N [ d ] is a normal closed subgroup scheme of H [ d ] .Proof. For (i) we note that the intersection of the Hopf ideal I ( H ) ⊂ k { G } with k [ G ] d yields aHopf ideal I ( H [ d ]) = k [ G ] d ∩ I ( H ) ⊂ k [ G ] d because k [ G ] d is a k -sub-Hopf algebra of k { G } .For (ii) let I ( H ) ⊂ I ( N ) ⊂ k { G } denote the ideals of H and N respectively. By assumption, I ( N ), the image of I ( N ) in k { H } = k { G } / I ( H ) is a normal Hopf ideal. The ideal I ( N [ d ]) ⊂ k [ H [ d ]] = k [ G ] d / I ( H [ d ]) of N [ d ] in H [ d ] is obtained from I ( N ) by intersecting with k [ H [ d ]] via k [ G ] d / I ( H [ d ]) ֒ → k { G } / I ( H ). Thus I ( N [ d ]) is a normal Hopf ideal.To illustrate the use of Zariski closures, let us describe the σ -closed subgroups of tori. Let k be a σ -field. As usual, we denote by G m the multiplicative group scheme over k . We think of G nm as a σ -algebraic group over k , i.e., G nm ( S ) = ( S × ) n for any k - σ -algebra S . The σ -coordinate ring of G nm is k { G nm } = k { x , . . . , x n , x , . . . , x n } σ . By a multiplicative function ψ ∈ k { x , . . . , x n , x , . . . , x n } σ we mean an element which is of the form ψ ( x ) = x α σ ( x α ) · · · σ l ( x α l )for some α i ∈ Z n and l ∈ N . Here x β := x β · · · x β n n for β ∈ Z n .38 emma A.40. Let k be a σ -field and let G be a σ -closed subgroup of G nm . Then there exists a set Ψ of multiplicative functions such that G ( S ) = { g ∈ G nm ( S ) | ψ ( g ) = 1 for ψ ∈ Ψ } for any k - σ -algebra S .Proof. For d ∈ N , let G [ d ] be the d -th order Zariski closure of G inside G nm . Then G [ d ] is analgebraic subgroup of ( G nm ) d = G n ( d +1) m (Lemma A.39). The claim now follows from the fact thatan algebraic subgroup of G n ( d +1) m is the intersection of kernels of characters of G n ( d +1) m . A.9 Quotients
In the category of groups, the quotient
G/N of a group G by a normal subgroup N is characterizedby the following universal property: Every morphism of groups φ : G → H such that N ⊂ ker( φ )factors uniquely through G → G/N . Replacing groups with group k - σ -schemes, we arrive at thefollowing definition. Definition A.41.
Let G be group k - σ -scheme and N E G a normal σ -closed subgroup. By a quotient of G modulo N , we mean a morphism π : G → G/N of group k - σ -schemes with N ⊂ ker( π ) satisfying the following universal property: For every morphism φ : G → H of group k - σ -schemeswith N ⊂ ker( φ ) there exists a unique morphism ψ : G/N → H of group k - σ -schemes making G π / / φ (cid:31) (cid:31) ❄❄❄❄❄❄❄❄ G/N ψ } } H commutative. As usual, if the quotient exists it is unique up to unique isomorphisms. In all generality, theexistence of quotients of group schemes is a somewhat delicate issue. Since we are only interestedin the affine case and normal closed subgroups everything can be done on the ring side and noheavy geometric machinery is necessary. We will follow the purely Hopf-algebraic approach of M.Takeuchi presented in [Tak72].Below we will use some standard notations from the theory of Hopf algebras: If R is a Hopfalgebra over a field k then ∆ : R → R ⊗ R denotes the comultiplication and ε : R → k denotes thecounit. The kernel of ε is denoted with R + , i.e., R + is the vanishing ideal of the unit element. Proposition A.42 (M. Takeuchi) . Let k be a field, R a Hopf algebra over k with comultiplication ∆ and a ⊂ R a normal Hopf ideal. Set R ( a ) = { r ∈ R | ∆( r ) − r ⊗ ∈ R ⊗ k a } . Then R ( a ) is a sub-Hopf algebra of R with RR ( a ) + = a , i.e., the ideal of R generated by R ( a ) + is equal to a . Moreover, R ( a ) is the only sub-Hopf algebra of R with this property and the inclusion map ι : R ( a ) ֒ → R satisfies the following universal property: Every morphism ψ : R ′ → R of k -Hopfalgebras such that ψ ( R ′ + ) ⊂ a factors uniquely through ι .Proof. The statement that R ( a ) is a sub-Hopf algebra of R is Lemma 4.4 in [Tak72]. That RR ( a ) + = a is proved in [Tak72, Theorem 4.3]. By [Tak72, Corollary 3.10] the map T RT + from sub-Hopf algebras of R to normal Hopf-ideals of R is injective. Thus R ( a ) is the only sub-Hopf algebra of R with RR ( a ) + = a . If ψ : R ′ → R is a morphism of k -Hopf algebras such that ψ ( R ′ + ) ⊂ a then ψ ( R ′ ) is a sub-Hopf algebra of R and ψ ( R ′ ) + = ψ ( R ′ + ) ⊂ a . By [Tak72, Lemma4.7] the Hopf-algebra R ( a ) is the greatest sub-Hopf algebra of R such that R ( a ) + ⊂ a . Therefore ψ ( R ′ ) ⊂ R ( a ), i.e., ψ factors through ι . Theorem A.43.
Let G be a group k - σ -scheme and N E G a normal σ -closed subgroup. Then thequotient π : G → G/N exists and satisfies N = ker( π ) . roof. We know from Proposition A.42 that k { G } ( I ( N )) = { r ∈ k { G }| ∆( r ) − r ⊗ ∈ k { G }⊗ k I ( N ) } is a sub-Hopf algebra of k { G } . It is also a k - σ -subalgebra. Let H denote the group k - σ -schemewith σ -coordinate ring k { H } = k { G } ( I ( N )). We will show that the morphism φ : G → H of group k - σ -schemes corresponding to the inclusion k { H } ⊂ k { G } of k - σ -Hopf algebras is the quotient of G modulo N .So let ϕ : G → H ′ be a morphism of group k - σ -schemes with N ⊂ ker( ϕ ). Since ker( ϕ ) = ϕ − (1 H ′ ) is the σ -closed σ -subscheme of G defined by the ideal of k { G } generated by ϕ ∗ ( k { H ′ } + )(Lemma A.5) we conceive that the algebraic meaning of N ⊂ ker( ϕ ) is ϕ ∗ ( k { H ′ } + ) ⊂ I ( N ). Itfollows from Proposition A.42 that the morphism ϕ ∗ : k { H ′ } → k { G } factors uniquely through k { H } ֒ → k { G } , i.e., ϕ ∗ ( k { H ′ } ) ⊂ k { H } . The induced morphism k { H ′ } → k { H } of k - σ -Hopfalgebras gives rise to the desired morphism H → H ′ making G φ / / ϕ ❆❆❆❆❆❆❆ H ~ ~ H ′ commutative. Thus φ : G → H is the quotient of G modulo N . The algebraic meaning of N =ker( φ ) is k { G } k { H } + = I ( N ). This identity holds by Proposition A.42. Corollary A.44.
Let G be a group k - σ -scheme, N E G a normal σ -closed subgroup and φ : G → H a morphism of group k - σ -schemes such that N = ker( φ ) and φ ∗ : k { H } → k { G } is injective. Then φ is the quotient of G modulo N , i.e., H = G/N .Proof.
Identifying k { H } with the image of φ ∗ , we may assume that k { H } ⊂ k { G } . The as-sumption N = ker( φ ) translates to I ( N ) = k { G } k { H } + . But from Proposition A.42 we know that k { G } ( I ( N )) is the only sub-Hopf algebra with this property. Thus k { H } = k { G } ( I ( N )) = k { G/N } and H = G/N as desired.
References [AM05] Katsutoshi Amano and Akira Masuoka. Picard-Vessiot extensions of Artinian simplemodule algebras.
J. Algebra , 285(2):743–767, 2005.[AMT09] Katsutoshi Amano, Akira Masuoka, and Mitsuhiro Takeuchi. Hopf algebraic approachto Picard-Vessiot theory. In
Handbook of algebra. Vol. 6 , volume 6 of
Handb. Algebr. ,pages 127–171. Elsevier/North-Holland, Amsterdam, 2009.[AOT] Benjamin Antieau, Alexey Ovchinnikov, and Dmitry Trushin. Galois theory of differenceequations with periodic parameters. arXiv:1009.1159.[Bou90] N. Bourbaki.
Algebra. II. Chapters 4–7 . Elements of Mathematics (Berlin). Springer-Verlag, Berlin, 1990. Translated from the French by P. M. Cohn and J. Howie.[Bui92] Alexandru Buium.
Differential algebraic groups of finite dimension , volume 1506 of
Lecture Notes in Mathematics . Springer-Verlag, Berlin, 1992.[Cas72] P. J. Cassidy. Differential algebraic groups.
Amer. J. Math. , 94:891–954, 1972.[CH] Zo´e Chatzidakis and Ehud Hrushovski. On subgroups of semi-abelian varieties definedby difference equations. arXiv:1112.0920.[CH99] Zo´e Chatzidakis and Ehud Hrushovski. Model theory of difference fields.
Trans. Amer.Math. Soc. , 351(8):2997–3071, 1999.[Cha97] Zo´e Chatzidakis. Groups definable in ACFA. In
Algebraic model theory (Toronto, ON,1996) , volume 496 of
NATO Adv. Sci. Inst. Ser. C Math. Phys. Sci. , pages 25–52. KluwerAcad. Publ., Dordrecht, 1997. 40CHP02] Zo´e Chatzidakis, Ehud Hrushovski, and Ya’acov Peterzil. Model theory of differencefields. II. Periodic ideals and the trichotomy in all characteristics.
Proc. London Math.Soc. (3) , 85(2):257–311, 2002.[Coh65] Richard M. Cohn.
Difference algebra . Interscience Publishers John Wiley & Sons, NewYork-London-Sydeny, 1965.[CS06] P.J. Cassidy and M.F. Singer. Galois theory of parameterized differential equationsand linear differential algebraic groups. In
Differential Equations and Quantum Groups ,volume 9 of
IRMA Lectures in Mathematics and Theoretical Physics , pages 113–157.2006.[CS12] Shaoshi Chen and Michael F. Singer. Residues and telescopers for bivariate rationalfunctions.
Adv. in Appl. Math. , 49(2):111–133, 2012.[Del90] P. Deligne. Cat´egories tannakiennes. In
The Grothendieck Festschrift, Vol. II , volume 87of
Progr. Math. , pages 111–195. Birkh¨auser Boston, Boston, MA, 1990.[DG70] Michel Demazure and Pierre Gabriel.
Groupes alg´ebriques. Tome I: G´eom´etriealg´ebrique, g´en´eralit´es, groupes commutatifs . Masson & Cie, ´Editeur, Paris, 1970. Avecun appendice ıt Corps de classes local par Michiel Hazewinkel.[DGS94] Bernard Dwork, Giovanni Gerotto, and Francis J. Sullivan.
An introduction to G -functions , volume 133 of Annals of Mathematics Studies . Princeton University Press,1994.[Dre] Thomas Dreyfus. A density theorem for parameterized differential Galois theory.arXiv:1203.2904.[DV12] Lucia Di Vizio. Approche galoisienne de la transcendance diff´erentielle. In
Transendanceet irrationalit´e , SMF Journ´ee Annuelle [SMF Annual Conference], pages 1–20. Soci´et´eMath´ematique de France, 2012.[DVHW] Lucia Di Vizio, Charlotte Hardouin, and Michael Wibmer. Difference algebraic relationsamong solutions of linear differential equation. arXiv:1310.1289.[Dyc] T. Dyckerhoff. The inverse problem of differential Galois theory over the field R ( z ).arXiv:0802.2897.[Har77] Robin Hartshorne. Algebraic Geometry . Number 52 in GTM. Springer, 1977.[Hru01] Ehud Hrushovski. The Manin-Mumford conjecture and the model theory of differencefields.
Ann. Pure Appl. Logic ∼ ehud/.[HS08] Charlotte Hardouin and Michael F. Singer. Differential Galois theory of linear differenceequations. Mathematische Annalen , 342(2):333–377, 2008.[Kam12] Moshe Kamensky. Tannakian formalism over fields with operators.
Int. Math. Res. Not.IMRN , 2012. doi: 10.1093/imrn/rns190.[Ked10] Kiran S. Kedlaya. p -adic differential equations , volume 125 of Cambridge Studies inAdvanced Mathematics . Cambridge University Press, Cambridge, 2010.[Kol68] E. R. Kolchin. Algebraic groups and algebraic dependence.
Amer. J. Math. , 90:1151–1164, 1968. 41Kol73] E. R. Kolchin.
Differential algebra and algebraic groups . Academic Press, New York,1973. Pure and Applied Mathematics, Vol. 54.[Kol74] E. R. Kolchin. Constrained extensions of differential fields.
Advances in Math. , 12:141–170, 1974.[Kol85] E. R. Kolchin.
Differential algebraic groups , volume 114 of
Pure and Applied Mathemat-ics . Academic Press Inc., Orlando, FL, 1985.[Kov03] Jerald J. Kovacic. The differential Galois theory of strongly normal extensions.
Trans.Amer. Math. Soc. , 355(11):4475–4522, 2003.[KP02] Piotr Kowalski and A. Pillay. A note on groups definable in difference fields.
Proc.Amer. Math. Soc. , 130(1):205–212 (electronic), 2002.[KP07] Piotr Kowalski and Anand Pillay. On algebraic σ -groups. Trans. Amer. Math. Soc. ,359(3):1325–1337 (electronic), 2007.[Lan08] Peter Landesman. Generalized differential Galois theory.
Trans. Amer. Math. Soc. ,360(8):4441–4495, 2008.[Lev08] Alexander Levin.
Difference algebra , volume 8 of
Algebra and Applications . Springer,New York, 2008.[Mac97] Angus Macintyre. Generic automorphisms of fields.
Ann. Pure Appl. Logic , 88(2-3):165–180, 1997. Joint AILA-KGS Model Theory Meeting (Florence, 1995).[Mau10] Andreas Maurischat. Galois theory for iterative connections and nonreduced Galoisgroups.
Trans. Amer. Math. Soc.
Proc. Lond. Math. Soc. (3) , 103(2):197–234, 2011.[MS12] Claude Mitschi and Michael F. Singer. Monodromy groups of parameterized lineardifferential equations with regular singularities.
Bull. Lond. Math. Soc. , 44(5):913–930,2012.[MS13] Claude Mitschi and Michael F. Singer. Projective isomonodromy and Galois groups.
Proc. Amer. Math. Soc. , 141(2):605–617, 2013.[OW] Alexey Ovchinnikov and Michael Wibmer. σ -Galois theory of linear difference equation.In preparation.[Sei56] A. Seidenberg. Contribution to the Picard-Vessiot theory of homogeneous linear differ-ential equations. Amer. J. Math. , 78:808–818, 1956.[SV99] Thomas Scanlon and Jos´e Felipe Voloch. Difference algebraic subgroups of commutativealgebraic groups over finite fields.
Manuscripta Math. , 99(3):329–339, 1999.[Tak72] Mitsuhiro Takeuchi. A correspondence between Hopf ideals and sub-Hopf algebras.
Manuscripta Math. , 7:251–270, 1972.[Tak89] Mitsuhiro Takeuchi. A Hopf algebraic approach to the Picard-Vessiot theory.
J. Algebra ,122(2):481–509, 1989.[vdPS97] Marius van der Put and Michael F. Singer.
Galois theory of difference equations .Springer-Verlag, Berlin, 1997. 42vdPS03] Marius van der Put and Michael F. Singer.
Galois theory of linear differential equations .Springer-Verlag, Berlin, 2003.[Wat79] William C. Waterhouse.
Introduction to affine group schemes , volume 66 of
GraduateTexts in Mathematics . Springer-Verlag, New York, 1979.[Wat95] G. N. Watson.
A treatise on the theory of Bessel functions . Cambridge MathematicalLibrary. Cambridge University Press, Cambridge, 1995. Reprint of the second (1944)edition.[Wib10] Michael Wibmer.
Geometric difference Galois theory . PhD thesis, Heidelberg, 2010.[Wib12a] Michael Wibmer. A Chevalley theorem for difference equations.
Math. Ann. ,354(4):1369–1396, 2012.[Wib12b] Michael Wibmer. Existence of ∂ -parameterized Picard-Vessiot extensions over fieldswith algebraically closed constants. J. Algebra , 361:163–171, 2012.[WW88] E.T. Whittaker and G.N. Watson.