Differential Galois groups over Laurent series fields

DIFFERENTIAL GALOIS GROUPS OVER LAURENT SERIES FIELDS arXiv:1501.06884v1 [math.AC] 27 Jan 2015 DAVID HARBATER, JULIA HARTMANN, AND ANNETTE MAIER Abstract. In this manuscript, we apply patching methods to give a positive answer to the inverse differential Galois problem over function fields over Laurent series fields of characteristic zero. More precisely, we show that any linear algebraic group (i.e. affine group scheme of finite type) over such a Laurent series field does occur as the differential Galois group of a linear differential equation with coefficients in any such function field (of one or several variables). Introduction Differential Galois theory studies linear homogeneous differential equations by means of their symmetry groups, the differential Galois groups. Such a group acts on the solution space of the equation under consideration, and this furnishes it with the structure of a linear algebraic group over the field of constants of the differential field. In analogy to the inverse problem in ordinary Galois theory, an answer to the question of which linear algebraic groups occur as differential Galois groups over a given differential field provides information about the field and its extensions. Classically, differential Galois theory was mostly concerned with one variable function fields over the complex numbers. In that case, the inverse problem is related to the RiemannHilbert problem (Hilbert’s 21st problem) about monodromy groups of differential equations. In fact, Tretkoff and Tretkoff ([TT79]) showed that every linear algebraic group over C occurs as the differential Galois group of some equation over C(x), as a consequence of Plemelj’s solution to the (modified) Riemann-Hilbert problem ([Ple08]; see also [AB94]). The solution to the inverse differential Galois problem over C(x) for an arbitrary algebraically closed field C of characteristic zero was given in [Har05], building on decades of work by other authors, e.g. [Kov69], [Kov71], [Sin93], [MS96], [MS02]. It is worth noting that contrary to ordinary Galois theory, there seems to be no direct way of deducing this from the complex case for general groups (see [Sin93]). More recently, there have been results concerning differential Galois theory over nonalgebraically closed fields of constants; see, e.g. [AM05], [And01], [Dyc08], [CHvdP13]. (There was actually an earlier attempt, in [Eps55a] and [Eps55b], but the approach did not yield a full Galois correspondence and seems to have been dropped.) The differential Galois groups in this more general setting are still linear algebraic groups, but in the sense Date: January 27, 2015 2010 Mathematics Subject Classification. 12H05, 20G15, 14H25, 34M50. Keywords. Picard-Vessiot theory, Patching, Linear algebraic groups, Inverse differential Galois problem, Galois descent. 1 of group schemes rather than of point groups. However, only very limited results are known about the corresponding inverse problem (in particular, see [Dyc08]). In this manuscript (Theorem 4.14), we solve the inverse problem for function fields over complete discretely valued fields of equal characteristic zero, which are Laurent series fields. While such fields are never algebraically closed, they share with the complex numbers the property of being complete with respect to a metric topology, therefore allowing local calculations involving series. This relates our approach to the classical approach in [TT79], although we use different techniques. As a general result that may also be useful in other contexts, we show that in order to solve the inverse problem over finitely generated differential fields over a field of constants K, it suffices to solve it over K(x) with derivation d/dx (Corollary 4.13). This type of approach has previously been used for groups over algebraically closed fields (e.g. for connected groups in [MS96]) and is sometimes called the Kovacic trick. Our approach to solving the inverse problem over rational function fields over Laurent series fields is based on a recent version of patching methods ([HH10]) and inspired by the use of patching methods in ordinary Galois theory (see [Har03] for an overview). In ordinary Galois theory, the patching machinery reduces the realization of a given finite group to the (local) realization of generating (e.g. cyclic) subgroups. In the differential setup, there are two additional complications. The first is that in differential Galois theory, it is not possible to patch the local Picard-Vessiot rings (which are the analogs of Galois extensions), since these are not finite as algebras. Instead, we apply patching directly to the differential equations via a factorization property related to patching; this is equivalent to patching the corresponding differential modules. The second complication is that while any finite group is generated by its cyclic subgroups, an analogous statement for linear algebraic groups is true only over algebraically closed fields: Every linear algebraic group over an algebraically closed field is generated by (a finite number of) finite cyclic groups and copies of the multiplicative and additive group. To overcome this issue, we apply the patching method after base change to a finite extension of the field of constants in such a way that the result descends to the original field. The advantage of our method is that the building blocks (i.e., the local differential equations for finite cyclic, multiplicative, and additive groups) are very easy to find and verify. Because we apply the patching machinery only in a very simple case in which the process can be described explicitly, we do not require the reader to be familiar with [HH10]. We note that a related but somewhat different strategy, to handle a special case of the problem, was sketched by the second author in [Har07]. We expect that the results of this manuscript can be applied to solve the inverse differential Galois problem for function fields over constant fields other than Laurent series, for example p-adic fields (this is work in progress). Organization of the manuscript: Section 1 lists some basic facts about Picard-Vessiot theory over non-algebraically closed fields of constants. It also contains auxiliary results about the linearization of the differential Galois group, a statement and consequences of the Galois correspondence, and descent results for Picard-Vessiot rings. Section 2 describes the patching setup for differential equations, proves the main patching result and gives some 2 examples. Section 3 is concerned with the generation of linear algebraic groups by “simple” subgroups and with finding differential equations which have those groups as differential Galois groups, the so-called building blocks for patching. Finally, Section 4 solves the inverse problem over rational function fields by combining the results of the previous two sections. It also contains a result stating that solving the inverse problem over rational function fields implies a solution over arbitrary finitely generated differential fields (over the same field of constants). Combining this with the result obtained for rational function fields gives the main Theorem 4.14. Acknowledgments: The authors wish to thank Tobias Dyckerhoff and Michael Wibmer for helpful discussions. We also thank Michael Singer for comments related to the contents of this manuscript. 1. Picard-Vessiot theory Throughout this manuscript, all fields are assumed to be of characteristic zero. If F is a field, we write F̄ for its algebraic closure. If R is an integral domain, we write Frac(R) for its field of fractions. If R is a differential ring, we write CR for its ring of constants. If F is a differential field, then CF is a field which is algebraically closed in F . We first record some facts from the Galois theory of differential fields with arbitrary fields of constants. We refer to [Dyc08] for details. Let (F, ∂) be a differential field with field of constants CF = K. Given a matrix A ∈ n×n F and a differential ring extension R/F , a fundamental solution matrix for A is a matrix Y ∈ GLn (R) which satisfies the differential equation ∂(Y ) = AY . A differential ring extension R/F is called a Picard-Vessiot ring for A ∈ F n×n if it satisfies the following conditions: The ring of constants of R is CR = K, there exists a fundamental solution matrix Y ∈ GLn (R), R is generated by the entries of Y and Y −1 (we write R = F [Y, Y −1 ]), and R is differentially simple (i.e., has no nontrivial ideals which are stable under ∂). Differential simplicity implies that a Picard-Vessiot ring is an integral domain. Its fraction field is called a Picard-Vessiot extension. Note that since K is not assumed to be algebraically closed, Picard-Vessiot rings need not exist (e.g. [Sei56]) and might not be unique. The torsor theorem states that if R is a Picard-Vessiot ring, there is an R-linear isomorphism of differential rings R ⊗F R ∼ = R ⊗K CR⊗F R . The differential Galois group of a Picard-Vessiot ring R/F is defined as the group functor Aut∂ (R/F ) from the category of K-algebras to the category of groups that sends a K-algebra S to the group Aut∂ (R ⊗K S/F ⊗K S) of differential automorphism of R ⊗K S that are trivial on F ⊗K S. This functor is represented by the K-Hopf algebra CR⊗F R = K[Y ⊗ Y −1 , Y −1 ⊗ Y ]. We conclude that the differential Galois group of R/F is an affine group scheme of finite type over K which is necessarily (geometrically) reduced since the characteristic is zero ([Oor66]; see also [Car62]); i.e. it is a linear algebraic group over K with coordinate ring K-isomorphic to CR⊗F R . Moreover, the torsor theorem asserts that the affine variety Spec(R) defined by a Picard-Vessiot ring R with Galois group G is a G-torsor and R ⊗F F̄ ∼ = CR⊗F R ⊗K F̄ , which implies trdeg(Frac(R)/F ) = dim(G). It is immediate from the definitions that if K ′ /K is an algebraic extension of constants, then R ⊗K K ′ is a Picard-Vessiot ring over F ⊗K K ′ whose differential Galois group is the base change 3 of the differential Galois group of R/F from K to K ′ . We remark that the definition of the differential Galois group requires the use of a Picard-Vessiot ring rather than a PicardVessiot extension; for this reason most statements in this manuscript are phrased in terms of a Picard-Vessiot ring. When constructing Picard-Vessiot rings, we will frequently use the following well-known criterion ([Dyc08, Corollary 2.7]). Proposition 1.1. Let L/F be an extension of differential fields with constants CL = CF and let A ∈ F n×n . Assume that there exists a fundamental solution matrix Y ∈ GLn (L) for A, i.e., ∂(Y ) = AY . Then R = F [Y, Y −1 ] ⊆ L is a Picard-Vessiot ring for A. The next proposition gives a criterion to determine which elements of a Picard-Vessiot extension are contained in a Picard-Vessiot ring. Elements satisfying the condition in the proposition are also called differentially finite. Proposition 1.2. Let R/F be a Picard-Vessiot ring for some matrix A. Then an element a ∈ Frac(R) lies in R if and only if a, ∂(a), ∂ 2 (a), . . . span a finite dimensional F -vector space. Proof. In the case that CF is algebraically closed, this is a well-known statement (see [vdPS03, Cor. 1.38]), which can be applied to R ⊗K K̄ over the field L := F ⊗K K̄. Here R ⊗K K̄ is a Picard-Vessiot ring since K̄/K is an algebraic extension of constants. Let a ∈ R ⊆ R ⊗K K̄. Thus a, ∂(a), ∂ 2 (a), . . . span a finite dimensional vector space over L, so there is an r ∈ N such that a, ∂(a), ∂ 2 (a), . . . , ∂ r (a) are linearly dependent over L. Let V be the F -vector space spanned by a, ∂(a), . . . , ∂ r (a). Then dimF (V ) = dimL (V ⊗F L) = dimL (V ⊗K K̄) ≤ r; and we conclude that a, ∂(a), . . . , ∂ r (a) are linearly dependent over F , giving the forward direction. The proof of the converse direction is the same as in the case CF is algebraically closed; see [vdPS03, Cor. 1.38, proof of (3)⇒(1)].  In particular, any two Picard-Vessiot rings R, R′ with the same field of fractions (but possibly for different matrices) are necessarily equal. 1.1. Linearization of the differential Galois group. The differential Galois group of a Picard-Vessiot ring was defined above as an abstract linear algebraic group. A choice of fundamental solution matrix determines an embedding into a general linear group as follows. Proposition 1.3. Let R = F [Y, Y −1 ] be a Picard-Vessiot ring over F with field of constants CF = K. Then there is a closed embedding of linear algebraic groups ΨY = ΨY,K : Aut∂ (R/F ) ֒→ GLn,K such that for all K-algebras S: ΨY (S) : Aut∂ (R ⊗K S/F ⊗K S) ֒→ GLn (S), σ 7→ (Y ⊗ 1)−1 σ(Y ⊗ 1). 4 Proof. Since σ is a differential automorphism, ∂ ((Y ⊗ 1)−1 σ(Y ⊗ 1)) = 0, i.e., (Y ⊗1)−1 σ(Y ⊗ 1) has entries in CR⊗K S = K ⊗K S ∼ = S. Therefore, ΨY (S) is a well-defined map. Let K[Z, Z −1 ] denote the coordinate ring of GLn over K. Let ρ : Aut∂ (R/F ) ֒→ GLn,K be the closed embedding induced by the surjection on the coordinate rings K[Z, Z −1 ] → K[Y ⊗ Y −1 , Y −1 ⊗ Y ], Z 7→ Y ⊗ Y −1 . It is then easy to deduce that ρ(S) = ΨY (S) for every Kalgebra S by examining the isomorphism Aut∂ (R⊗K S/F ⊗K S) → HomK (K[Y ⊗Y −1 , Y −1 ⊗ Y ], S). Hence ΨY = ρ is in fact a closed embedding of linear algebraic groups.  We define Gal∂Y (R/F ) ≤ GLn as the image of the differential Galois group Aut∂ (R/F ) under the linearization ΨY defined in Proposition 1.3 and we also call it the differential Galois group of R/F (with respect to Y ). A different choice of fundamental solution matrix leads to a differential Galois group which is conjugate by an element of GLn (K). We will write expressions like G ≤ GLn to denote a linear algebraic group with an embedding into a fixed copy of GLn . The following example illustrates the above and will be used in our building blocks in Section 3.2. Example 1.4. Let F be a differential field with field of constants K and let a ∈ F . We  0 a . If there exists a differential consider the differential equation ∂(y) = Ay with A = 0 0 field extension L/F with CL = K and an element y ∈ L with ∂(y) =  a, then  R = F [y] 1 y ∈ GL2 (L) is a Picard-Vessiot ring for A over F by Proposition 1.1. Indeed, Y = 0 1 is a fundamental solution matrix for A and R = F [Y, Y −1 ]. Let S be a K-algebra and σ ∈ Aut∂ (R ⊗K S/F ⊗K S). As ∂(σ(y ⊗ 1)) = σ(∂(y ⊗ 1)) = σ(a ⊗ 1) = a ⊗ 1 = ∂(y ⊗ 1), there exists  an ασ  ∈ CR⊗K S = K ⊗K S ≃ S with σ(y ⊗ 1) = y ⊗ 1 + ασ . Moreover, 1 ασ , hence Gal∂Y (R/F )(S) is a subgroup of Ga (S) in its two-dimensional Y −1 σ(Y ) = 0 1 representation and Gal∂Y (R/F ) is a subgroup scheme of Ga . (Note that since Ga has no proper closed subgroups in characteristic zero, it must in fact be either the whole group or trivial.) Remark 1.5. One advantage of working within a fixed GLn is the following. Two linear algebraic groups G ≤ GLn and H ≤ GLn defined over K are equal (as subgroups of GLn ) if and only if G(K̄) = H(K̄) inside GLn (K̄), since the K̄-rational points of a linear algebraic group are dense. In particular, G and H are isomorphic over K if G(K̄) = H(K̄); whereas in general, G(K̄) ∼ = H(K̄) for abstract linear algebraic groups would imply only that G and H are isomorphic over K̄. The following lemma is immediate from the definitions (in the language of differential modules, the transformation corresponds to a change of basis). Lemma 1.6. Let R/F be a Picard-Vessiot ring for a matrix A with fundamental solution matrix Y ∈ GLn (R). Then for each B ∈ GLn (F ), R/F is also a Picard-Vessiot ring for 5 ∂(B)B −1 + BAB −1 with fundamental solution matrix BY ∈ GLn (R), and Gal∂Y (R/F ) = Gal∂BY (R/F ) inside GLn,CF . Lemma 1.7. Let R/F be a Picard-Vessiot ring for a matrix A with fundamental solution matrix Y ∈ GLn (R). Let L/F be a differential field extension such that R and L are contained in a common differential field extension with constants CF . Then the compositum LR = L[Y, Y −1 ] is a Picard-Vessiot ring for A over L, and Gal∂Y (LR/L) ≤ Gal∂Y (R/F ) as subgroups of GLn,CF . Proof. Abbreviate K = CF . By Proposition 1.1, LR is a Picard-Vessiot ring for A over L. For the second assertion, let S be any K-algebra, and let σ ∈ Aut∂ (LR/L)(S) = Aut∂ (LR ⊗K S/L ⊗K S). Then σ acts on LR ⊗K S and hence on GLn (LR ⊗K S), taking Y ⊗ 1 to (Y ⊗ 1)Z where Z = ΨY (S)(σ) ∈ Gal∂Y (LR/L)(S) ≤ GLn (S) with ΨY as in Proposition 1.3. Since R = F [Y, Y −1 ], σ restricts to an automorphism of R⊗K S. The corresponding monomorphism Gal∂Y (LR/L)(S) → Gal∂Y (R/F )(S) maps Z to Z. This defines an inclusion Gal∂Y (LR/L) ≤  Gal∂Y (R/F ) of subgroups of GLn,K . If moreover the tensor product of L and R is a compositum without new constants, we even obtain equality: Proposition 1.8. Let R/F be a Picard-Vessiot ring for a matrix A with fundamental solution matrix Y ∈ GLn (R). Let L/F be a differential field extension and assume that L ⊗F R is an integral domain such that its field of fractions has constants CF . Then L ⊗F R is a Picard-Vessiot ring for A over L with fundamental solution matrix 1 ⊗ Y . Moreover, we have an equality of differential Galois groups inside GLn,CF : Gal∂1⊗Y (L ⊗F R/L) = Gal∂Y (R/F ) Proof. Abbreviate K = CF . View L⊗F R as the compositum of L and R inside Frac(L⊗F R). Then Lemma 1.7 implies that L ⊗F R = F [1 ⊗ Y, (1 ⊗ Y )−1 ] is a Picard-Vessiot ring over L and Gal∂1⊗Y (L ⊗F R/L) ≤ Gal∂Y (R/F ) as subgroups of GLn,K . To obtain equality, it suffices to show that for any K-algebra S and any σ0 ∈ Aut∂ (R ⊗K S/F ⊗K S) there exists an element σ ∈ Aut∂ ((L ⊗F R) ⊗K S/L ⊗K S) = Aut∂ (L ⊗F (R ⊗K S)/L ⊗F (F ⊗K S)) that restricts to σ0 . But σ = idL ⊗F σ0 has that property, which concludes the proof.  In the special case when R/F is a finite Galois extension and L/F is a finite extension, the condition that L ⊗F R is an integral domain is equivalent to saying that L and R are linearly disjoint over F and the condition that the field L ⊗F R does not have new constants is equivalent to saying that L ⊗F R is regular over K. (Recall that a field extension A/K is regular if A is linearly disjoint from K̄ over K.) 1.2. The Galois correspondence. Let R/F be a Picard-Vessiot ring together with a fundamental solution matrix Y ∈ GLn (R). Set G := Aut∂ (R/F ), G := Gal∂Y (R/F ) ≤ GLn and E := Frac(R), K := CF . Then for a closed subgroup H ≤ G (defined over K), consider the inverse image H ≤ G of H under the isomorphism ΨY : G → G. The functorial invariants E H in E are defined as the elements a/b ∈ Frac(R) satisfying σ(a ⊗ 1)(b ⊗ 1) = (a ⊗ 1)σ(b ⊗ 1) 6 for all K-algebras S and all σ ∈ H(S) ⊆ Aut∂ (R ⊗K S/F ⊗K S). As H(K̄) is dense in H, it even suffices to check the above condition for S = K̄. The Galois correspondence ([Dyc08, Thm. 4.4]) then states that there is an inclusion-reversing bijection between closed subgroups of G (defined over K) and differential subfields F ⊆ L ⊆ E, given by H 7→ E H and L 7→ Aut∂ (R/L). We illustrate with an example why we need to consider functorial invariants rather than invariants on K-points. √ Example 1.9. Consider F = Q(x) with derivation d/dx and R = F [ 3 x]. Then R = Frac(R) 1 is a Picard-Vessiot ring over F for the differential equation ∂(y) = 3x y with fundamental √ 3 ∼ solution matrix y = x and differential Galois group G = µ3 , the subgroup of GL3 defined by the equation X 3 − 1 = 0. Note that µ3 (Q) = {1}; hence Aut∂ (R/F ) = Aut(R/F ) = {id}, and RAut(R/F ) strictly contains F . But RG = F . It is known ([Dyc08, Proposition 4.3]) that if H ≤ G is a closed normal subgroup, then E is the fraction field of a Picard-Vessiot ring RH over F with Aut∂ (RH /F ) ∼ = G/H. For H equal to the identity component of G, we obtain the following. H Lemma 1.10. Let F be a differential field and let R/F be a Picard-Vessiot ring with differential Galois group G. Write G0 for the identity component of G and E = Frac(R). 0 Then E G is the algebraic closure of F in E, and it is a finite field extension of degree 0 [E G : F ] = |G(K̄)/G0 (K̄)|. 0 Proof. Since trdeg(E/E G ) = dim(G0 ) = dim(G) = trdeg(E/F ) and E is finitely generated 0 over F , the extension E G /F is finite. On the other hand, every algebraic subextension F ⊆ L ⊆ E is a differential extension, and thus H := Aut∂ (R/L) is a closed subgroup of G of the 0 0 same dimension. Therefore, G0 ⊆ H and L = E H ⊆ E G . Hence E G is the algebraic closure 0 of F in E. While the finite field extension E G /F might not be Galois, the compositum 0 0 0 E G K̄ ∼ = F ⊗K K̄ of the same degree as E G /F . = E G ⊗K K̄ is a finite extension of F K̄ ∼ 0 0 0 G Moreover, Aut(E G K̄/F K̄) ∼ = (G/G0 )(K̄) ∼ = G(K̄)/G0 (K̄) and (E G K̄)Aut(E K̄/F K̄) = F K̄ 0 (because K̄ is algebraically closed). Hence E G K̄/F K̄ is a finite Galois extension of degree |G(K̄)/G0 (K̄)| and the claim follows.  1.3. Galois descent for Picard-Vessiot rings. Let F0 be a differential field with field of constants K0 . Given a linear algebraic group G ≤ GLn over K0 , it might be easier to realize GK as a differential Galois group over F0 K for some finite extension of constants K/K0 . We assume that K/K0 is a finite Galois extension with Galois group Γ. As K0 is algebraically closed in F0 , F0 K ∼ = F0 ⊗K0 K is Galois over F0 with group Γ (and the action of Γ commutes with the derivation). In our applications, we construct a Picard-Vessiot ring over F0 K such that the fundamental solution matrix is Γ-invariant. The following lemma explains that this Picard-Vessiot ring then descends to a Picard-Vessiot ring over F0 with differential Galois group G. Lemma 1.11. Let K/K0 be a finite Galois extension with Galois group Γ. Let F0 be a differential field with CF0 = K0 and set F = F0 K. Further, let L/F be an extension of 7 differential fields with CL = CF = K and such that the action of Γ on F extends to an action on L via differential automorphisms. If R = F [Y, Y −1 ] ⊆ L is a Picard-Vessiot ring over F such that Y ∈ GLn (L) is invariant under the action of Γ, then R0 := F0 [Y, Y −1 ] is a PicardVessiot ring over F0 with Gal∂Y (R0 /F0 )K = Gal∂Y (R/F ) as subgroups of GLn . In particular, if Gal∂Y (R/F ) = GK for a linear algebraic group G ≤ GLn,K0 , then Gal∂Y (R0 /F0 ) = G. Proof. Note that A := ∂(Y )Y −1 ∈ F n×n is Γ-invariant, hence A ∈ F0n×n . The field LΓ of Γ-invariants in L is a differential field extension of F0 with CLΓ = K0 = CF0 , since CL = K. As Y is contained in GLn (LΓ ), Proposition 1.1 implies that R0 is a Picard-Vessiot ring for A over F0 . Let H ≤ GLn be the linear algebraic group Gal∂Y (R0 /F0 ) defined over K0 . Since K0 is algebraically closed in Frac(R0 ), the natural map R0 ⊗K0 K → R is an isomorphism, and the induced map on matrices sends Y ⊗K0 1 to Y . Hence the natural map ι : Aut∂ (R ⊗K K̄/F ⊗K K̄) → Aut∂ (R0 ⊗K0 K̄/F0 ⊗K0 K̄) is an isomorphism. It is given by σ 7→ φ−1 σφ, where the differential K̄-isomorphism φ is the composition R0 ⊗K0 K̄ → R0 ⊗K0 K ⊗K K̄ → R ⊗K K̄, whose induced map on matrices sends Y ⊗K0 1 to Y ⊗K 1. We claim that the isomorphism ι yields the equality H(K̄) = Gal∂Y (R0 /F0 )(K̄) = Gal∂Y (R/F )(K̄) as subsets of GLn (K̄). Namely, for any σ ∈ Aut∂ (R ⊗K K̄/F ⊗K K̄), ΨY,K (K̄)(σ) = (Y ⊗K 1)−1 σ(Y ⊗K 1) = φ((Y ⊗K0 1)−1 ι(σ)(Y ⊗K0 1)) = (Y ⊗K0 1)−1 ι(σ)(Y ⊗K0 1) = ΨY,K0 (K̄)(ι(σ)). (The third equality holds since all entries of the matrix are contained in K̄.) The claim follows. We conclude that HK = Gal∂Y (R/F ) as subgroups of GLn as asserted. In particular, if Gal∂Y (R/F ) = GK for some G ≤ GLn,K0 , then H(K̄) = GK (K̄) = G(K̄) as subgroups of GLn (K̄), and hence H = G.  2. Application of patching to differential equations Throughout this section, let K = k((t)) for some field k of characteristic zero; let F = K(x); and consider a derivation ∂ on F with CF = K. Note that ∂ = ∂(x) · ∂/∂x. Moreover, F is the function field of the projective x-line P1k[[t]] over the discrete valuation ring k[[t]]. In [HH10], a collection of field extensions FP , F℘ and FU were considered. Whereas the definitions in loc. cit. are geometric, we only need a special case, in which the description is very explicit: If P ∈ A1k ⊂ P1k is a rational point defined by x = b for some b ∈ k, then we consider the fields FP = k((x − b, t)) and F℘(P ) = k((x − b))((t)), 8 where k((x−b, t)) denotes the fraction field of the power series ring k[[x−b, t]] in two variables. Given a non-empty finite set P of k-rational points, we have an index set B = {℘(P )| P ∈ P} in bijection with P. If P consists of points z = bi for some bi ∈ k and 1 ≤ i ≤ m, then we let U be the complement of P in P1k , and we write FU = Frac(k[(x − b1 )−1 , . . . , (x − bm )−1 ][[t]]). As explained in [HH10], there are inclusions F ⊆ FU ,FP ⊆ F℘(P ) for all P ∈ P. These are in fact inclusions of differential fields, where we equip FU , FP , and F℘(P ) with the derivation ∂(x) · ∂/∂x; moreover CFU = CFP = CF℘(P ) = CF = K for all P ∈ P. In particular, if A ∈ F n×n is such that there exists a fundamental solution matrix Y ∈ GLn (FU ), then F [Y, Y −1 ] is a Picard-Vessiot ring for A over F (by Proposition 1.1). The method of patching over the fields (F, FU , FP , F℘ ) relies on the following two properties. Theorem 2.1. (a) Simultaneous factorization property: Let n ∈ N. If (Y℘ )℘∈B is a collection of matrices Y℘ ∈ GLn (F℘ ) then there exist matrices BP ∈ GLn (FP ) for each P ∈ P and one matrix Y ∈ GLn (FU ) such that for each P ∈ P, Y℘(P ) = BP−1 Y in GLn (F℘(P ) ). (b) Intersection property: If x ∈ FU is such that for each P ∈ P, x is contained in FP when considered as an element of F℘(P ) , then x ∈ F . For a proof of the simultaneous factorization property, see [HH10, Thm 5.10] and [HHK11, Prop. 2.2]. The intersection property is stated in [HH10, Prop. 6.3]. Definition 2.2. In the above setup, an action of a finite group Γ on the differential patching data (P, B, U) consists of the following: (1) (2) (3) (4) a left action of Γ on F via differential automorphisms. a left action of Γ on FU via differential automorphisms, extending the action of Γ on F . a right action of Γ on the finite set P. for each σ ∈ Γ and each P ∈ P, an isomorphism σ : FP σ → FP of differential fields extending σ : F → F such that for all σ, τ ∈ Γ, στ : FP στ → FP is the composition σ ◦ τ : FP στ → FP σ → FP . (5) for each σ ∈ Γ and each P ∈ P, an isomorphism of differential fields σ : F℘(P σ ) → F℘(P ) extending both σ : FU → FU and σ : FP σ → FP such that for all σ, τ ∈ Γ, στ : F℘(P στ ) → F℘(P ) is the composition σ ◦ τ : F℘(P στ ) → F℘(P σ ) → F℘(P ) . Example 2.3. (a) Let k0 ≤ k such that k/k0 is a finite Galois extension, let e ≥ 1 be a natural number such that k contains a primitive e-th root of unity, and set t0 = te . Then K = k((t)) is a finite Galois extension of K0 = k0 ((t0 )), and Γ := Gal(K/K0 ) is the semidirect product of the cyclic group of order e and the group Gal(k/k0). In particular, Γ surjects onto Gal(k/k0 ). Note that F = K(x) is a finite Galois extension of F0 = K0 (x) whose Galois group is isomorphic to Γ and acts on F as a differential field. The action of Γ on F over F0 (from the left) induces an action of Γ on the x-line P1k[[t]] over P1k0 [[t0 ]] from the right. In particular, there is an induced action of Γ on P1k over P1k0 from the right. If P ∈ P1k is a point of the form x = b for some b ∈ k, then P σ is defined by x = σ −1 (b). The induced 9 isomorphism σ : FP σ = k((x − σ −1 (b), t)) → FP = k((x − b, t)) is a differential isomorphism, as is the induced isomorphism σ : F℘(P σ ) = k((x − σ −1 (b)))((t)) → F℘(P ) = k((x − b))((t)). Notice that if P ⊆ P1k is a finite set of closed points invariant under the action of Γ, then the action of Γ on P1k[[t]] also induces an action of Γ on FU as a differential field. Therefore, Γ acts on the differential patching data (P, B, U). (b) Observe that the Γ-orbit of a closed point x = b as above consists of at most |Gal(k/k0 )| ≤ |Γ| elements. To obtain an action with orbits of full length |Γ| for use in Section 4, we will use the following construction. Suppose again that k contains a primitive e-th root of unity ζ. Let z = x/t. We will work with the z-line P1k[[t]] instead of the x-line. (In other words, we perform a blow-up at the origin, and then blow down the original component.) Let σ ∈ Γ. As t is an e-th root of t0 ∈ K0 , σ(t) = ζ nσ t for some nσ ∈ N, and hence σ(z) = ζ −nσ z. Note that the induced action of Γ on the zline P1k maps a point P of the form z = b to the point P σ defined by z = ζ nσ−1 σ −1 (b); and the induced isomorphisms σ : FP σ = k((z − ζ nσ−1 σ −1 (b), t)) → FP = k((z − b, t)) and F℘(P σ ) = k((z − ζ nσ−1 σ −1 (b)))((t)) → F℘(P ) = k((z − b))((t)) are again isomorphisms of differential fields (they map z − ζ nσ−1 σ −1 (b) to ζ −nσ (z − b)). The Γ-orbit of such a point z = b then consists of all points of the form z = ζ nσ σ(b) for σ ∈ Γ. We will show later that there exist elements b ∈ k such that this orbit consists of |Γ| points. Theorem 2.4. Let n ∈ N. a) Assume that for each P ∈ P, a matrix AP ∈ FPn×n is given together with a fundamental solution matrix YP ∈ GLn (F℘(P ) ). Let GP = Gal∂YP (RP /FP ) ≤ GLn,K be the differential Galois group of the Picard-Vessiot ring RP = FP [YP , YP−1 ] for AP (see Proposition 1.1). Then there exists a matrix A ∈ F n×n and a fundamental solution matrix Y ∈ GLn (FU ) for A such that the Picard-Vessiot ring R = F [Y, Y −1 ] over F has the following property: Its differential Galois group equals the Zariski closure of the group ¯ ∈ P >. generated by the various subgroups GP of GLn,K ; i.e. Gal∂Y (R/F ) = < GP | P b) Assume that moreover a finite group Γ acts on the differential patching data (P, B, U) and assume that σ(YP σ ) = YP in GLn (F℘(P ) ) for each P ∈ P and each σ ∈ Γ. Then the fundamental solution matrix Y ∈ GLn (FU ) can be chosen such that its entries are Γ-invariant. Proof. Simultaneous factorization (Theorem 2.1.a) implies that there exists a matrix Y ∈ GLn (FU ) and matrices BP ∈ GLn (FP ) for each P ∈ P such that YP = BP−1 · Y when viewed inside GLn (F℘(P ) ) for each P ∈ P. We set A = ∂(Y )Y −1 ∈ FUn×n . For each P ∈ P, we n×n compute inside F℘(P ) , viewing FU and FP as subfields of F℘(P ) : A = ∂(Y )Y −1 = ∂(BP YP )YP−1 BP−1 = ∂(BP )BP−1 + BP AP BP−1 ∈ FPn×n , 10 using that YP is a fundamental solution matrix for AP . The intersection property (Theorem 2.1.b) implies that all entries of A are contained in F . By Proposition 1.1, R = F [Y, Y −1 ] ⊆ FU is a Picard-Vessiot ring for A ∈ F n×n over F . Set G = GalY∂ (R/F ) ≤ GLn,K . We first show that GP ≤ G for all P ∈ P. Note that RP = FP [YP , YP−1 ] = FP [Y, Y −1 ] = FP R (the compositum is taken inside F℘(P ) ) for all P ∈ P since YP = BP−1 · Y . Hence GP = Gal∂YP (RP /FP ) = Gal∂Y (RP /FP ) ≤ Gal∂Y (R/F ) = G by Lemma 1.6 and Lemma 1.7. We now let H ≤ G be the Zariski closure of < GP | P ∈ P > in GLn . We claim that G = H. Let H ≤ Aut∂ (R/F ) be the inverse image of H ≤ G under the isomorphism ΨY : Aut∂ (R/F ) → G (Proposition 1.3). By the Galois correspondence, it suffices to show that E H = F , where E = Frac(R). Suppose there exists an element a/b ∈ E H r F for some a, b ∈ R. Note that a/b ∈ E ⊆ FU ⊆ F℘ for all ℘ ∈ B. The intersection property implies that there exists a P ∈ P with a/b ∈ F℘(P ) r FP . On the other hand, GP ≤ H, hence Aut∂ (RP /FP ) ≤ H. Indeed, Aut∂ (RP /FP ) can be identified with a subgroup scheme of Aut∂ (R/F ) via restriction and ΨY : Aut∂ (R/F ) → G maps this subgroup scheme to Gal∂Y (RP /FP ) = GP ≤ H = ΨY (H). We conclude that a/b is an element of Frac(R) ⊆ Frac(RP ) that is invariant under Aut∂ (RP /FP ). The Galois correspondence applied to the Picard-Vessiot ring RP /FP implies that a/b is contained in FP , a contradiction. This proves part (a). To prove part (b), it suffices to show that there exists a B ∈ GLn (F ) such that B −1 Y ∈ GLn (FUΓ ), since then Gal∂Y (R/F ) = Gal∂B−1 Y (R/F ) by Lemma 1.6. We first claim that for each σ ∈ Γ, Y σ(Y )−1 ∈ GLn (FU ) has entries in F . Let σ ∈ Γ. By the intersection property, it suffices to show that Y σ(Y )−1 ∈ GLn (FP ) when viewed as an element in GLn (F℘(P ) ) for each P ∈ P. Let P ∈ P and set Q = P σ ∈ P. By assumption, there is a differential isomorphism σ : F℘(Q) → F℘(P ) restricting to σ : FQ → FP and restricting to σ : FU → FU . In GLn (F℘(Q) ), we have Y = BQ YQ , with notation as in the proof of part (a); hence σ(Y ) = σ(BQ )σ(YQ ) = σ(BQ )σ(YP σ ) = σ(BQ )YP in GLn (F℘(P ) ). On the other hand, Y = BP YP and we compute inside GLn (F℘(P ) ): Y σ(Y )−1 = BP YP (σ(BQ )YP )−1 = BP σ(BQ )−1 ∈ GLn (FP ), proving the claim. Therefore, we obtain a 1-cocycle χ : Γ → GLn (F ), σ 7→ Y σ(Y )−1 . By Hilbert’s Theorem 90, H 1 (Γ, GLn (F )) is trivial, hence there exists a B ∈ GLn (F ) such that for all σ ∈ Γ: Y σ(Y )−1 = Bσ(B)−1 . This implies that B −1 Y ∈ GLn (FU ) is Γ-invariant as we wanted to show.  Note that part (b) of the above theorem enables us to use Lemma 1.11 and hence to obtain Galois descent. Example 2.5. (1) In this example, we apply Theorem 2.4 to show that SL2 is a differential Galois group over F = R((t))(x) (with derivation ∂ = ∂/∂x). Let G1 , G2 ≤ SL2 be the subgroups of upper and lower unitary triangular matrices in SL2 . Let P1 , P2 be the closed points x = 1 and x = 2 on P1R and consider the patching data (P, B, U) induced by P = {P1 , P2 }. Then for j = 1, 2, FPj = R((x − j, t)) and F℘(Pj ) = R((x − j))((t)); and FU = Frac(R[(x − 1)−1 , (x − 2)−1 ][[t]]). 11 −1   0 a1 , = 0 0 −1 Set aj = −t(x − j) (x − j − t) ∈ F ⊆ FPj for j = 1, 2. Also write AP1       ∞ P 1 0 1 y1 0 0 1 tr ∈ FU ⊆ F℘(Pj ) , with yj = , Y P2 = and YP1 = AP2 = r(x−j)r y2 1 0 1 a2 0 r=1 for j = 1, 2. Then ∂(YP ) = AP YP for both points P ∈ P, and RP = FP [YP , YP−1 ] is a Picard-Vessiot ring for AP over FP . It is easy to see that yj is transcendental over FPj and to deduce that Gal∂YP (RPj /FPj ) = Gj in GL2 for j = 1, 2 (use Example 1.4). Theorem 2.4(a) j now implies that there exists an A ∈ F 2×2 and a Y ∈ GL2 (FU ) such that the Picard-Vessiot ring R = F [Y, Y −1 ] for A over F has differential Galois group Gal∂Y (R/F ) = hG1¯, G2 i = SL2 . (2) As another example, let us consider SO2 in its two-dimensional representation    a b 2 2 | a +b =1 −b a over R. We realize SO2 as a differential Galois group over R((t))(x) by using patching over F = C((t))(x) (with derivation ∂ = ∂/∂x on both fields). Consider the isomorphism over C   1 λ + λ−1 i(−λ + λ−1 ) ψ : Gm → SO2 , λ 7→ . λ + λ−1 2 i(λ − λ−1 ) Let P1 , P2 be the closed points x = i and x = −i on P1C and consider the patching data (P, B, U) induced by P = {P1 , P2 }. Then FP = C((x ± i, t)) and F℘(P ) = C((x ± i))((t)) for P = P1 , P2 , respectively, and FU = Frac(C[(x − i)−1 , (x + i)−1 ][[t]]). Note that Γ = Gal(C/R) acts on F and on FU via differential automorphisms and the non-trivial element σ ∈ Γ induces ∂-isomorphisms FP1 → FP2 and F℘(P1 ) → F℘(P2 ) . Therefore, Γ acts on the differential t patching data (P, B, U). Set y = e x−i ∈ FU and YP1 = ψ(y) ∈ GL2 (FU ) ≤ GL2 (F℘(P1 ) ). Note that RP1 = FP1 [YP1 , YP−1 ] = FP1 [y, y −1] is a Picard-Vessiot ring over FP1 for the one1 −1 −t ∈ RP2×2 is dimensional equation ∂(y) = (x−i) 2 y. It can be shown that AP1 := ∂(YP1 )YP1 1 2×2 contained in FP1 . Thus RP1 ⊆ F℘(P1 ) is also a Picard-Vessiot ring for AP1 over FP1 . Set = σ(AP1 ) ∈ FP2×2 , and RP2 = FP2 [YP2 , YP−1 ] is a YP2 = σ(YP1 ). Then AP2 := ∂(YP2 )YP−1 2 2 2 Picard-Vessiot ring for AP2 over FP2 . It is easy to see that y is transcendental over FP1 hence Gal∂y (RP1 /FP1 ) = Gm,C and thus Gal∂YP (RP1 /FP1 ) = ψ(Gm,C ) = (SO2 )C . Similarly, 1 Gal∂YP (RP2 /FP2 ) = (SO2 )C . Theorem 2.4(a) implies that there exists an A ∈ F 2×2 with 2 fundamental solution matrix Y ∈ GL2 (FU ) and Picard-Vessiot ring R = F [Y, Y −1 ] such that Gal∂Y (R/C((t))(x)) = (SO2 )C . By Theorem 2.4(b), we can moreover assume that Y is Γ-invariant. Lemma 1.11 then implies that R0 := R((t))(x)[Y, Y −1 ] is a Picard-Vessiot ring over R((t))(x) with Gal∂Y (R0 /R((t))(x)) = SO2 . (3) Let T ≤ GL√ n be a one-dimensional torus √ defined over Q((t)) that splits over the finite extension Q((t))( t) = Q((s)) with s := t, i.e., there is an isomorphism ψ : Gm → T defined over Q((s)). Set K = Q((s)) and F = K(x) with derivation ∂ = ∂/∂x. Then K/Q((t)) is Galois with Galois group Γ ∼ = C2 and F/Q((t))(x) is also Galois with Galois group Γ. Let σ ∈ Γ be the non-trivial automorphism. Consider the change of variables z = x/s ∈ F . Then F = K(z) with ∂(z) = 1/s and σ(z) = −z. Let P1 , P2 be the 12 closed points z = 1 and z = −1 on the z-line P1Q and consider the patching data (P, B, U) induced by P = {P1 , P2 }. Then FP = Q((z ± 1, s)) and F℘(P ) = Q((z ± 1))((s)) for P = P1 , P2 , respectively, and FU = Frac(Q[(z−1)−1 , (z+1)−1 ][[s]]). As explained in Example 2.3(b), Γ acts on F and on FU via differential automorphisms and σ induces ∂-isomorphisms FP1 → FP2 and F℘(P1 ) → F℘(P2 ) . Therefore, Γ acts on the differential patching data (P, B, U) s permuting P1 and P2 . We set y = e z−1 ∈ F℘(P1 ) which is transcendental over FP1 . Note −s that ∂(y)y −1 = 1s · (z−1) 2 ∈ FP1 , since ∂(z) = 1/s. We define YP1 = ψ(y) ∈ T (F℘(P1 ) ) and YP2 = σ(YP1 ) ∈ T (F℘(P2 ) ) and proceed as in the previous example. Eventually, we obtain an A ∈ F n×n with fundamental solution matrix Y ∈ GLn (FU ) and Picard-Vessiot ring R = F [Y, Y −1 ] such that Gal∂Y (R/Q((s))(x)) = TK and moreover R0 = Q((t))(x)[Y, Y −1 ] is a Picard-Vessiot ring over Q((t))(x) with Gal∂Y (R0 /Q((t))(x)) = T . 3. Constructing extensions 3.1. Linearizations. In this subsection, K denotes a field of characteristic zero. We begin by showing that every linear algebraic group admits a finite set of “simple” generating subgroups, after passing to a finite field extension. Proposition 3.1. Let G be a linear algebraic group defined over K. Then there exists a finite extension L/K and finitely many closed subgroups G1 , . . . , Gr ≤ GL such that GL is generated by G1 , . . . , Gm and such that each Gi is isomorphic (over L) to either Ga or Gm or a finite (constant) cyclic group. Proof. It suffices to show this for L replaced by an algebraic closure K̄ of K. By the theorem of Borel-Serre [BS64, Lemme 5.11], GK̄ is generated by its identity component together with some finite group H ≤ GK̄ . Clearly, H is generated by finitely many finite constant cyclic subgroups (defined over K̄). We may thus assume that GK̄ is connected. Theorem 6.4.5 in [Spr09] implies that GK̄ is generated by the centralizers of finitely many maximal tori T ≤ GK̄ . Such a centralizer C(T ) is a connected closed subgroup of GK̄ and it is nilpotent with maximal torus T ([Spr09, Prop. 6.4.2]). Let Cu denote the set of unipotent elements in C(T ) (which is a closed, connected subgroup). Then C = T Cu . Now T is isomorphic to a direct product of copies of Gm (over K̄) and is thus generated by finitely many subgroups that are isomorphic to Gm . It remains to show that Cu is generated by finitely many subgroups that are isomorphic to Ga . For x ∈ Cu (K̄), let G(x) ≤ Cu denote the Zariski closure of the subgroup generated by x. For each x 6= 1, G(x) is an infinite, closed, abelian, unipotent and thus also connected subgroup of Cu (defined over K̄). Since Cu is finite dimensional, there exist finitely many elements x ∈ Cu (K̄) such that the finitely many subgroups G(x) generate Cu . Now each of the groups G(x) is isomorphic to Gm a by [Spr09, Lemma 3.4.7.c] for some m ∈ N. Actually, m m = 1 since Ga does not contain a dense cyclic subgroup for m ≥ 2.  In order to construct Picard-Vessiot rings whose differential Galois groups are given subgroups of GLn that are isomorphic to Ga , Gm , or a finite cyclic group, we use the following statement that allows us to modify the representation of the differential Galois group. It is based on standard Tannakian arguments. 13 Proposition 3.2. Let R be a Picard-Vessiot ring over a differential field F with field of constants K, and let G be its differential Galois group. Suppose that ρ : G → GLn,K is any linear representation. Then there exists a Picard-Vessiot ring R′ ⊆ R over F and a fundamental solution matrix Y ′ ∈ GLn (R′ ) such that Gal∂Y ′ (R′ /F ) = ρ(G). If moreover ρ is faithful, then R′ = R. Proof. Consider the differential equation associated to R and let M be the corresponding differential module over F ([vdPS03], discussion preceding Lemma 1.7). Let {{M}} be the full subcategory of the category of differential modules over F generated by M (i.e., {{M}} is the smallest full subcategory that contains M and is closed under subquotients, finite direct sums, tensor products, and duals). Further, let ReprG denote the category of finite dimensional K-representations of G. The Picard-Vessiot extension R determines an equivalence of symmetric tensor categories S : {{M}} → ReprG by [AM05], Theorem 4.10. Here, for an object N in {{M}}, S(N) is the solution space ker(∂, R ⊗F N) with G-action induced by the action of G on R, and trivial on N. Since R is a Picard-Vessiot ring for M, dimK (S(N)) = dimF (N) for all N; hence there exists a fundamental solution matrix YN over R generating a Picard-Vessiot ring RN ≤ R, and the induced morphism G → Gal∂YN (RN /F ) is equivalent to S(N). Applying this to an object M ′ in {{M}} such that S(M ′ ) is equivalent to the given representation ρ yields the first claim. If ρ is faithful then the fraction fields of R′ and R are equal by the Galois correspondence. Since a Picard-Vessiot ring is characterized as the set of differentially finite elements in a Picard-Vessiot extension (see Proposition 1.2), this implies R′ = R.  3.2. Building blocks. As before, we fix a field k of characteristic zero. Next, we construct explicit extensions with differential Galois group Ga , Gm and cyclic group Cr of order r, via the following lemmas. For the Ga case, we have: ∂ Lemma 3.3. Let c ∈ k((t))× and consider derivation ∂ = c ∂x on the fields k((x, t)) ⊆ P∞the −r r k((x))((t)). Set y = − log(1 − t/x) := r=1 x t /r ∈ k((x))((t)). Then R =  t))[y]  k((x, 1 y , and is a Picard-Vessiot ring over k((x, t)) with fundamental solution matrix Y = 0 1   1 ∗ . Gal∂Y (R/k((x, t))) is the image of Ga in its two-dimensional representation 0 1 1 Proof. Note that y satisfies the differential equation ∂(y) = xc · 1−x/t ∈ k((x, t)), hence R ⊆ k((x))((t)) is a Picard-Vessiot ring over k((x, t)) with Gal∂Y (R/k((x, t))) ≤ Ga in its twodimensional representation (see Example 1.4). To see that this containment is an equality, it suffices to show that the differential Galois group is nontrivial. By the Galois correspondence, this is equivalent to showing that y does not lie in k((x, t)). We will show by contradiction that y does lie in the overfield K((x)) of k((x, t)) (where K = k((t)) as above). P∞not even i If y = i=m ai x with ai ∈ K, then ∂(y) = ∞ X i=m 14 iai xi−1 . Here there is no term of degree −1. But the term of lowest degree in
c 1 · = cx−1 1 + x/t + x2 /t2 + · · · x 1 − x/t has degree −1. Thus ∂(y) cannot equal c x · 1 , 1−x/t and this is a contradiction.  To treat the Gm case, we first show a generalization of Theorem 2.4 in [Völ96] (where k was assumed algebraically closed). Lemma 3.4. Let κ be a field of characteristic zero, let K = κ((π)), and let L be a finite field extension of K. Then there is a finite field extension κ′ of κ and a positive integer e such that L is contained in the field extension κ′ ((π))(π 1/e ) = κ′ ((π 1/e )) of K. Proof. After replacing L by its Galois closure over K, we may assume that L/K is Galois. Let K̄ be an algebraic closure of K that contains L. Let R = κ[[π]], let S be the integral closure of R in L, and let e be the ramification index of S over the prime πR of R. The residue field of R′ = R[π 1/e ] is again κ. Let S ′ be the integral closure of the compositum of R′ and S inside K̄. Thus R′ and S ′ are each complete discrete valuation rings, and S ′ is finite over R′ . Let κ′ be the residue field of S ′ . By Abhyankar’s Lemma ([Gro71], Lemme X.3.6), S ′ is unramified over the prime π 1/e R′ of R′ . Thus the degree of S ′ over R′ is equal to the degree d of the residue field extension κ′ /κ. Hence the inclusion κ′ [[π 1/e ]] ֒→ S ′ is an isomorphism, each ring being a degree d integrally closed extension of R′ . It follows that S is contained in κ′ [[π 1/e ]] and thus that L is contained in κ′ ((π))(π 1/e ) as asserted.  ∂ Lemma 3.5. Let c ∈ k((t))× andPconsider the derivation ∂ = c ∂x on the fields k((x, t)) ⊆ ∞ −r r k((x))((t)). Let y = exp(t/x) := r=0 x t /r! ∈ k((x))((t)). Then R = k((x, t))[y, y −1] is a Picard-Vessiot ring over k((x, t)) with Gal∂y (R/k((x, t))) = Gm . Proof. Note that y satisfies the differential equation ∂(y) = −ctx−2 y over k((x, t)), hence R ⊆ k((x))((t)) is a Picard-Vessiot ring for the one-by-one matrix A = −ctx−2 by Proposition 1.1 and Gal∂y (R/k((x, t))) ≤ GL1 = Gm . To see that this containment is an equality, it suffices by the Galois correspondence to show that y is transcendental over k((x, t)) (since the proper closed subgroups of the multiplicative group are finite). We will show by contradiction that no finite extension of the differential field extension k((t))((x)) of k((x, t)) contains an element y that satisfies ∂(y) = −ctx−2 y. (This is sufficient since derivations extend uniquely to finite extensions.) Suppose there is a finite extension L/k((t))((x)) containing such a y. Applying Lemma 3.4 with κ = k((t)), we obtain a finite field extension κ′ of κ and a positive integer e such that L is contained in κ′ ((z)), where z e = x. We consider k((t))((x)) as a differential field with respect to ∂ = c ∂/∂x. This derivation has a unique extension to a derivation on κ′ ((z)) with constant field κ′ , given by ∂(z) = cz 1−e /e. 15 Since y ∈ κ′ ((z)), we may write y = ∂(z i ) = iz i−1 ∂(z) = cie z i−e , and so P∞ ∂(y) = i=m ai z i , where each ai ∈ κ′ and am 6= 0. Now ∞ X iai c i=m e z i−e . j The coefficient of z in this expression vanishes for j < m−e, and in particular for j = m−2e. Meanwhile, the coefficient of z m−2e in −ctx−2 y = −ctz −2e y is −ctam , which is non-zero. Thus ∂(y) cannot equal −ctx−2 y. This is a contradiction.  Lemma 3.6. Let r be a positive integer, and assume that k contains a primitive r-th root of
1/r unity. Then y := x 1 −x−1 t is contained in k((x))((t)) and R = k((x, t))[y] ⊆ k((x))((t)) is a Picard-Vessiot ring over k((x, t)) with Gal∂y (R/k((x, t))) = Cr in its one-dimensional representation.
1/r Proof. First note that the element y/x = 1 − x−1 t lies in the subring k[x−1 ][[t]] of k((x))((t)) because it is a power series in x−1 t, hence y ∈ k((x))((t)). Now y r lies in k[[x, t]]. We claim that r is minimal for this property. To show this, let m be minimal such that y m ∈ k[[x, t]]. Then m divides r; let d = r/m. Thus y r = xr − xr−1 t is a d-th power of some element f in k[[x, t]]. Since f d = xr − xr−1 t, it follows that f has no term of total degree less than m; and among its terms of total degree m there is a nonvanishing xm term as well as some xi tj term with i + j = m and j > 0. Let j be maximal for this property. Then xr − xr−1 t = f d has a non-vanishing xid tjd term. Hence i = r − 1 and j = d = 1. Thus r = m, as claimed. Since k contains a primitive r-th root of unity, it then follows from Kummer theory that k((x, t))(y) is a cyclic extension of k((x, t)) of degree r. The derivation ∂ extends uniquely to R = k((x, t))(y) = k((x, t))[y], and y satisfies the differential equation ∂(y) = ay for a := ∂(y)/y = ∂(y r )/(ry r ) ∈ k((x, t)). Hence R is a Picard-Vessiot ring over k((x, t)) by Proposition 1.1, and Gal∂y (R/k((x, t))) = Cr .  Proposition 3.7. Let k be a field of characteristic zero. Let P be a k-point of the affine x-line A1k ⊂ P1k , and let FP and F℘(P ) be as defined in Section 2. Endow these fields with ∂ for some c ∈ k((t))× . Suppose that G ⊆ GLn is a linear algebraic group the derivation c ∂x over K = k((t)) that is K-isomorphic to Gm , Ga , or Cr where r is such that k contains a primitive r-th root of unity. Then there exist a matrix A ∈ FPn×n , a Picard-Vessiot ring R/FP for A that is contained in F℘(P ) , and a fundamental solution matrix Y ∈ GLn (R) for A, such that Gal∂Y (R/FP ) = G. Proof. After a change of variables, we may assume that P is the point of P1k where x = 0. We then have inclusions FP = k((x, t)) ⊂ F℘(P ) = k((x))((t)). By Lemma 3.3, 3.5, and 3.6, there exists a Picard-Vessiot ring over FP contained in F℘(P ) with differential Galois group Gm , Ga , or Cr as in the statement, respectively. The result then follows from Proposition 3.2.  4. The differential inverse Galois problem In this section, we solve the inverse problem over function fields over k((t)), where as before k is any field of characteristic zero. We begin by making the following 16 Definition 4.1. Let F be a differential field with field of constants K. We say that a linear algebraic group G defined over K is a differential Galois group over F if there exists a Picard-Vessiot ring R/F with Galois group K-isomorphic to G. Note that if G is a differential Galois group over F , this also implies that every faithful representation G ≤ GLn of G arises as a differential Galois group over F by Proposition 3.2; namely, there exists a Y ∈ GLn (R) such that Gal∂Y (R/F ) = G. We next note that in order to solve the inverse problem, we may modify the derivation. Lemma 4.2. Let F be a field, and let ∂ and ∂ ′ be derivations on F such that there is an a ∈ F × with ∂ ′ = a∂. Let G ≤ GLn be a linear algebraic group over CF , and suppose that G is a differential Galois group over (F, ∂). Then G is also a differential Galois group over (F, ∂ ′ ). Proof. By assumption, there exists a matrix A ∈ F n×n with Picard-Vessiot ring R over (F, ∂) and fundamental solution matrix Y ∈ GLn (R) such that Gal∂Y (R/F ) = G in the given representation G ≤ GLn . Then ∂ ′ = a∂ extends to a derivation on R with ∂ ′ (Y ) = aAY . By Proposition 1.1, R is a Picard-Vessiot ring over (F, ∂ ′ ) for the matrix aA with funda′ ′ mental solution matrix Y . Note that Aut∂ (R/F ) = Aut∂ (R/F ) and thus GalY∂ (R/F ) = Gal∂Y (R/F ) = G.  4.1. The inverse problem over k((t))(x). As before, k denotes a field of characteristic zero. In this section, we show that every linear algebraic group over k((t)) is a differential Galois group over the rational function field k((t))(x), equipped with any nontrivial derivation with field of constants k((t)). Lemma 4.3. Let k0 ≤ k such that k/k0 is a finite Galois extension of degree d. Assume further that k contains a primitive e-th root of unity ζ for some e ∈ N. Then there exist infinitely many elements a ∈ k such that there are d·e distinct elements among the Gal(k/k0 )conjugates of a, ζa, . . . , ζ e−1a. Proof. We may assume that d > 1. Since k/k0 is separable, there exists some primitive element b ∈ k. Thus there are exactly d conjugates of b. For any c ∈ k0 , consider the Gal(k/k0 )-conjugates of ζ i (c + b) for 0 ≤ i < e. If the number of conjugates is less than d · e, then there exists a non-trivial σ ∈ Gal(k/k0 ) such that ζ i (c + b) = σ(ζ j (c + b)) for some 0 ≤ i ≤ j < e. Note that ζ i 6= σ(ζ j ), since σ(c) = c and σ(b) 6= b. Hence there exists i 0 < l < e with σ(ζζ j ) = ζ l . We obtain c(ζ l − 1) = σ(b) − ζ l · b, hence c is contained in the l ·b | 0 < l < e, 1 6= σ ∈ Gal(k/k0 )} ⊆ k. Therefore, almost all choices of finite set S = { σ(b)−ζ ζ l −1 c yield an element a = c + b with the desired property.  Suppose that k/k0 is a finite Galois extension of degree d and that k contains a primitive e-th root of unity, for a positive integer e. Consider the Laurent series field extension k((t))/k0 ((t0 )) with t0 := te . Define Γ = Gal(k((t))/k0 ((t0 ))). Recall from Example 2.3(a)
that Γ has order d · e, and it is a semi-direct product of the cyclic group Gal k((t))/k((t0 )) of order e with the group Gal(k/k0 ). The action of Γ on K = k((t)) over K0 = k0 ((t0 )) extends to an action of F = K(x) over K0 (x) by taking x to x. Consider the variable change z = x/t. 17 Lemma 4.4. In the above setup, consider the induced action of Γ on the z-line P1k (as explained in Example 2.3(b)). Then for any positive integer r, there exist r k-points P1 , . . . , Pr ∈ A1k ⊂ P1k whose orbits P1Γ, . . . PrΓ are disjoint and are each of order |Γ|. Proof. For each σ ∈ Γ, there exists an integer 0 ≤ nσ ≤ e − 1 with σ(t) = ζ nσ t and thus σ(z) = ζ −nσ z. Recall from Example 2.3(b) that if σ ∈ Γ and P ∈ P is the k-point z = b, −1 then P σ is the point z = ζ nσ σ(b). We apply Lemma 4.3 and obtain an a1 ∈ k such that there are d · e distinct elements among the Gal(k/k0 )-conjugates of ζ i a1 (0 ≤ i ≤ e − 1). Let P1 be the point z = a1 . Then P1Γ consists of the points z = ζ nσ σ(a1 ). We claim that these |Γ| = de points are distinct. It suffices to check that {ζ nσ σ(a1 ) | σ ∈ Γ} = {τ (ζ i a1 ) | 0 ≤ i < e, τ ∈ Gal(k/k0 )}.
Given an element τ (ζ i a1 ) in the right hand side, let ψ be the element of Gal k((t))/k0 ((t)) ≤ Γ that lifts τ ∈ Gal(k/k0 ). Thus nψ = 0. Since
τ (ζ i ) is an e-th root of unity, there is an element π in the cyclic group Gal k((t))/k((t0 )) with ζ nπ = τ (ζ i ). Define σ = π ◦ ψ ∈ Γ. As nψ = 0, ζ nσ = ζ nπ = τ (ζ i ). Therefore ζ nσ σ(a1 ) = τ (ζ i a1 ), and the claim follows. To find a second point P2 , we apply Lemma 4.3 to obtain an element a2 ∈ k with the same property and such that a2 is not contained in the finite set of Gal(k/k0 )-conjugates of ζ i a1 . Therefore, P2 ∈ / P1Γ. Thus P1Γ and P2Γ are disjoint orbits of order |Γ|. By induction, we get points P1 , . . . , Pr as asserted.  Theorem 4.5. Let k be a field of characteristic zero, let K = k((t)), and let F be a rational function field of transcendence degree one over K. Let ∂ be a non-trivial derivation on F with field of constants K. Then every linear algebraic group defined over K is a differential Galois group over F . Proof. In order to make the notation in the proof consistent with the notation used in Section 2, we rename our base field as follows: We consider the rational function field F0 = K0 (x) over K0 = k0 ((t0 )), the field of formal Laurent series over a field k0 of characteristic zero. By Lemma 4.2 we may assume without loss of generality that the derivation ∂ on F0 satisfies ∂(x) = 1, with CF0 = K0 (hence ∂ = ∂/∂x). Let G ≤ GLn be a linear algebraic group defined over K0 . We will show that there exists a differential equation A ∈ F0n×n with a Picard-Vessiot ring R and fundamental solution matrix Y such that Gal∂Y (R/F0 ) = G. By Proposition 3.1, there exists a finite extension K/K0 and finitely many K-subgroups G1 , . . . , Gr ≤ GK that generate GK and such that each Gi is either a finite cyclic group or is K-isomorphic to either Gm or Ga . After enlarging K, we may assume it contains a primitive m-th root of unity for each m that is the order of one of the finite cyclic groups among G1 , . . . , Gr . We may further assume that K = k((t)) for a finite Galois extension k/k0 and an e-th root t of t0 for some e ∈ N, by Lemma 3.4. Without loss of generality, we may assume that k contains a primitive e-th root of unity ζ. Then K/K0 is a finite Galois extension (see Example 2.3) and we set Γ = Gal(K/K0 ). We first construct a Picard-Vessiot ring over F = K(x) with Galois group GK , using patching. Let z = x/t. By Lemma 4.4, we can fix k-points P1 , . . . , Pr on the affine zline A1k ⊂ P1k such that |PiΓ| = |Γ| for all i and such that these orbits are disjoint. We set 18 P = P1Γ ∪· · ·∪PrΓ . Let (P, B, U) be the corresponding differential patching data as explained in the beginning of Section 2. As explained in Example 2.3(b), Γ acts on the differential patching data (P, B, U) (see also Definition 2.2). By Proposition 3.7 (with c = ∂(z) = t−1 ), we can now fix Picard-Vessiot rings RPj /FPj for each j = 1, . . . , r with fundamental solution matrices YPj ∈ GLn (F℘(Pj ) ) such that Gal∂YP (RPj /FPj ) = Gj ≤ GK ≤ GLn,K . Now let P ∈ P be arbitrary. Then there exists j a unique σ ∈ Γ and a unique integer 1 ≤ j ≤ r such that P σ = Pj . Recall that there is a differential isomorphism σ : F℘(Pj ) → F℘(P ) restricting to σ : FPj → FP (see Definition 2.2 and Example 2.3). Set YP = σ(YPj ) ∈ GLn (F℘(P ) ) and RP = FP [YP , YP−1 ] = σ(RPj ) ⊆ F℘(P ) . We define AP = ∂(YP )YP−1 . Note that AP = σ(APj ) ∈ GLn (FP ). Hence RP is a Picard-Vessiot ring over FP for AP with fundamental solution matrix YP . Fix an extension of σ from K to K̄. Then σ induces a differential isomorphism σ : RPj ⊗K K̄ → RP ⊗K K̄. Hence Aut∂ (RP /FP )(K̄) = σAut∂ (RPj /FPj )(K̄)σ −1 and thus Gal∂YP (RP /FP )(K̄) = Gal∂σ(YP ) (σ(RPj )/σ(FPj ))(K̄) j = σ(Gal∂YP (RPj /FPj )(K̄)) j = σ(Gj (K̄)). We conclude by Remark 1.5 that Gal∂YP (RP /FP ) = Gjσ , the linear algebraic group obtained from Gj by applying σ to the defining equations. Theorem 2.4 then yields a matrix A ∈ F n×n and a Picard-Vessiot ring R = F [Y, Y −1 ] for A over F (Y ∈ GLn (FU ) a fundamental solution matrix) such that Gal∂ (R/F ) = < G σ | 1 ≤ j¯≤ r, σ ∈ Γ > = < G σ | ¯σ ∈ Γ > = GK . Y j K Moreover, we constructed the fundamental solution matrices (YP )P ∈P such that σ(YP σ ) = YP for all P ∈ P, σ ∈ Γ. Therefore, Theorem 2.4(b) asserts that we can assume that Y is Γstable, which in turn implies that all entries of A are contained in F Γ = F0 . Therefore, R0 := F0 [Y, Y −1 ] is a Picard-Vessiot ring for A over F0 with differential Galois group Gal∂Y (R0 /F0 ) = G ≤ GLn,K0 (via Lemma 1.11 with L = FU ).  Proposition 4.6. The Picard-Vessiot ring R in Theorem 4.5 can be constructed such that R ⊆ K((x)). Proof. We switch to the notation as in the proof of Theorem 4.5, so we need to show that R0 ⊆ K0 ((x)) = k0 ((t0 ))((x)). We refine the choice of P such that the point z = 0 is not contained in P. Set m = |Γ| · r, where r is as in the proof of Theorem 4.5, and let α1 , . . . , αm ∈ k × such that P is the set of all points z = αi . Now R0 = F0 [Y, Y −1 ] and Y ∈ GLn (FU ) is Γ-invariant. As all αi are non-zero, FU = Frac([(z − α1 )−1 , . . . , (z − αm )−1 ][[t]]) ⊆ Frac(k[[z, t]]). Moreover k[[z, t]] is contained in k((t))[[x]], as can be seen by replacing z by x/t in any power series in k[[z, t]]. Thus Frac(k[[z, t]]) ⊆ Frac(k((t))[[x]]) = k((t))((x)). Hence FU can be 19 regarded as a differential subfield of k((t))((x)); and it is easy to check that the embedding FU ⊆ k((t))((x)) is Γ-equivariant. Therefore, all entries of Y lie in FUΓ ⊆ k((t))((x))Γ = k0 ((t0 ))((x)) and the assertion follows.  Remark 4.7. Theorem 4.5, in conjunction with [Hru02], can be used to obtain a new proof of the fact that every linear algebraic group G over Q̄ is a differential Galois group over Q̄(x). (In [Har05], the second author gave an earlier proof of this, which also handled the case of C(x) for C an arbitrary algebraically closed field of characteristic zero). Namely, we may regard G as defined over Q̄((t)); and then by Theorem 4.5, there is a Picard-Vessiot extension R of Q̄((t))(x) with differential Galois group GQ̄((t)) , for some matrix A ∈ Q̄((t))(x)n×n . This data descends to some finitely generated Q̄-subalgebra D of Q̄((t)); and by [Hru02, Section V.1], infinitely many specializations to closed points of Spec(D) yield the same differential Galois group G. Since Q̄ is algebraically closed, these closed points are all Q̄-rational; and so G is a differential Galois group over Q̄(x). Remark 4.8. Theorem 4.5 in particular asserts that every finite étale group scheme G over a characteristic zero Laurent series field K = k((t)) is a differential Galois group over K(x). Equivalently, for each such G, there is a finite morphism of smooth connected projective K-curves C → P1K , with a faithful action of G on C, such that the generic point of C is a G-torsor over K(x). In the case that G is a finite constant group, this says that G is a Galois group over K(x); and that was proven (in fact in arbitrary characteristic) in [Har87, Theorem 2.3], using formal patching methods. In the case of finite étale group schemes G that need not be constant, this was shown (again in arbitrary characteristic) in [MB01]. That paper in fact showed more, viz. that this holds if K is more generally a large field in the sense of Pop ([Pop96]); and also that one fiber of the morphism C → P1K can be given in advance; this extended results of Pop ([Pop96, Main Theorem A]) and Colliot-Thélène ([CT00, Theorem 1]) for finite constant groups. 4.2. Passage to finitely generated extensions. In this subsection, we prove our main result (Theorem 4.14). This is a consequence of Theorem 4.5, together with the following result (see Theorem 4.12): If F is any differential field of characteristic zero with the property that every linear algebraic group defined over CF is a differential Galois group over F , then every finitely generated extension L/F of differential fields with CL = CF also has this property. To prove the latter, we first require some preparation. Lemma 4.9. Let L/F be an extension of differential fields. If x1 , . . . , xn ∈ CL are algebraically independent over CF , then they are algebraically independent over F . Proof. We prove the contrapositive. Assume that x1 , . . . , xn ∈ CL are algebraically dependent over F . Let r be the smallest positive integer such that there exists a polynomial expression over F , vanishing on (x1 , . . . , xn ), with exactly r monomial terms. Let p be such a polynomial. After dividing by an element in F × , we can assume that some coefficient equals one. Differentiating the coefficients of p yields a polynomial with less than r monomial terms that also vanishes on (x1 , . . . , xn ) and must be the zero polynomial by minimality of p. Hence all coefficients of p are contained in CF , and thus x1 , . . . , xn are algebraically dependent over CF .  20 Lemma 4.10. Let F be a differential field, let R/F be a Picard-Vessiot ring and set G = Aut∂ (R/F ) and E = Frac(R). Let H1 , . . . , Hr be closed subgroups of G (defined over CF ) r T and set H = Hi . Then i=1 E H1 · · · E Hr = E H ⊆ E. Proof. The compositum of E H1 , . . . , E Hr is a differential subfield of E, so it equals E H̃ for some closed subgroup H̃ of G, by the Galois correspondence. Then E Hi ⊆ E H̃ , so H̃ ⊆ Hi for all i, and thus H̃ ⊆ H. On the other hand, every element in the compositum is invariant under H, hence H̃ ⊇ H.  Lemma 4.11. Let F1 , . . . , Fr be finite field extensions of a field F . Embed F into F1 ⊗F · · · ⊗F Fr via 1 7→ 1 ⊗ · · · ⊗ 1. Let V ⊆ F1 ⊗F · · · ⊗F Fr be an F -subspace of dimension d < r with F ⊆ V . If 1 ≤ s ≤ r − d + 1, then after relabeling the indices, (F1 ⊗F · · · ⊗F Fs ) ∩ V = F, where the intersection is taken inside F1 ⊗F · · · ⊗F Fr . Proof. Since F is contained in each Fi , the right hand side is contained in the left hand side. For the other containment, it suffices to prove the assertion for s = r − d + 1. For each i ≤ r, fix an F -basis {vi,j | 1 ≤ j ≤ [Fi : F ]} of Fi with vi,1 = 1. We use induction on r. If r = 2, then d = 1 and V = F , and the claim follows. Let r > 2. If (F1 ⊗F · · ·⊗F Fr )∩V = F , there is nothing to prove. Otherwise, fix an x ∈ (F1 ⊗F · · ·⊗F Fr )∩V that is not contained in F . Then we can write x as an F -linear combination of the elements / F , this sum contains a term v1,j1 ⊗ · · · ⊗ vr,jr v1,j1 ⊗ · · · ⊗ vr,jr in a unique way. Since x ∈ with ji 6= 1 for some i. After relabeling the indices we may assume i = r, which implies x∈ / F1 ⊗F · · · ⊗F Fr−1 . Hence dimF ((F1 ⊗F · · · ⊗F Fr−1 ) ∩ V ) ≤ d − 1 and the claim follows by induction.  As before, let F be a differential field, and let G be a linear algebraic group defined over K = CF . We now prove the above mentioned general result, using a strategy sometimes called the Kovacic trick: Assume there is a Picard-Vessiot extension R over F with differential Galois group Gr . The idea is to show that R ⊗F L must contain a Picard-Vessiot ring over L with differential Galois group G if r is sufficiently large. Theorem 4.12. Let F be a differential field of characteristic zero and write K = CF . Let L/F be a differential field extension that is finitely generated over F with CL = K. Let G be a linear algebraic group defined over K with the property that for every r ∈ N, Gr is a differential Galois group over F . Then G is a differential Galois group over L. Proof. Let L′ be the algebraic closure of F in L and let L′′ be the normal closure of L′ in F̄ . Set d = [L′′ : F ] and m = trdeg(L/F ) + 1. Note that 1 ≤ d, m < ∞. First step: We show that there is a Picard-Vessiot ring R/F with differential Galois group Aut∂ (R/F ) isomorphic to G2m , and with the following property: If F ′ denotes the algebraic closure of F in Frac(R), then (1) F ′ ⊗F L′ is a field in which (2) K is algebraically closed. By assumption, there is a Picard-Vessiot ring R/F with differential Galois group G2m+2d . We show that we can achieve (1) and (2) by replacing R with a suitable subring. 21 Set E = Frac(R) and let Ei be the fixed field E G×···×G×1×G×···×G (omitting the i-th factor) for 1 ≤ i ≤ 2m + 2d. Thus CE = CEi = K. Since G × · · · × G × 1 × G × · · · × G is a normal subgroup of G2m+2d , the Galois correspondence implies that Ei = Frac(Ri ) for a Picard-Vessiot ring Ri /F with differential Galois group Aut∂ (Ri /F ) ∼ = G for each i. ′ ′ We let F ⊆ E be the algebraic closure of F in E and similarly let Fi ⊆ Ei the algebraic clo0 0 0 sure of F in Ei . Then Lemma 1.10 implies F ′ = E G ×···×G and Fi′ = E G×···×G×G ×G×···×G . By 0 0 ′ Lemma 4.10, the compositum F1′ · · · F2m+2d equals E G ×···×G = F ′ . Again by Lemma 1.10, [F ′ : F ] = |G(K̄)/G0 (K̄)|2m+2d and [Fi′ : F ] = |G(K̄)/G0 (K̄)| for each 1 ≤ i ≤ 2m + 2d. A dimension count yields ′ F′ ∼ . = F1′ ⊗F · · · ⊗F F2m+2d Let V = F ′ ∩ L′′ . Thus dim(V ) ≤ dim(L′′ ) = d. An application of Lemma 4.11 yields ′ (F1′ ⊗F · · · ⊗F F2m+d ) ∩ L′′ = F , possibly after renumbering the indices. We now change notation: we replace E by E 1×···×1×G×···×G (with d copies of G on the right); we replace R with the unique Picard-Vessiot ring in E 1×···×1×G×···×G (hence Aut∂ (R/F ) ∼ = 2m+d ′ ′ ′ 1×···1×G×···×G G ); and we replace F by the algebraic closure F1 ⊗F · · ·⊗F F2m+d of F in E . (Note that this tensor product is indeed equal to the relative algebraic closure, since both have the same degree over F by Lemma 1.10 and since the former field extension is contained in the latter.) We similarly replace Ei and Ri . In this new situation, F ′ ∩ L′′ = F . As L′′ /F is a finite Galois extension, F ′ ∩ L′′ = F implies that F ′ ⊗F L′′ is a field. In particular, F ′ ⊗F L′ is a field which proves (1) for this choice of R. To establish (2), let K ′ be the algebraic closure of K in the field F ′ ⊗F L′ . Thus K ′ = CF ′ ⊗F L′ . Moreover, K ′ and F ′ are linearly disjoint over K, since K ′ /K is algebraic whereas K is algebraically closed in F ′ (using CF ′ ⊆ CE = K). Hence a K-basis of K ′ is linearly independent over F ′ . Therefore, [K ′ : K] = [F ′ K ′ : F ′ ] ≤ [F ′ ⊗F L′ : F ′ ] = [L′ : F ] ≤ d and thus [K ′ L′ : L′ ] ≤ d, where all composita are taken inside F ′ ⊗F L′ . Recall that ′ F′ ∼ , hence = F1′ ⊗F · · · ⊗F F2m+d ′ ⊗F L′ ). F ′ ⊗F L′ ∼ = (F1′ ⊗F L′ ) ⊗L′ · · · ⊗L′ (F2m+d Since the L′ -vector space V = L′ K ′ has dimension at most d, by Lemma 4.11 we can relabel the indices such that ′ ⊗F L′ ) ∩ L′ K ′ = L′ . (F1′ ⊗F L′ ) ⊗L′ · · · ⊗L′ (F2m ′ ⊗F L′ ) ∩ K ′ = L′ ∩ K ′ = K, where the last equality In particular (F1′ ⊗F L′ ) ⊗L′ · · · ⊗L′ (F2m uses that CL = K. We again change notation, replacing E by E 1×···×1×G×···×G (with d copies of G on the right) and replacing R, F ′ , Ei , and Ri correspondingly as above. In this new notation, we obtain that F ′ ⊗F L′ is a field in which K is algebraically closed. Note that Aut∂ (R/F ) ∼ = G2m for this new R. Second step: We claim that if R is as constructed in the first step and E = Frac(R), then E ⊗F L is an integral domain and K is algebraically closed in Ẽ = Frac(E ⊗F L). As E/F ′ and L/L′ are regular, both E ⊗F ′ (F ′ ⊗F L′ ) and (F ′ ⊗F L′ ) ⊗L′ L are regular over F ′ ⊗F L′ . Therefore, their tensor product over F ′ ⊗F L′ is regular over F ′ ⊗F L′ ([Bou90, Proposition V.17.3b]). But this tensor product is isomorphic to E ⊗F L, and we conclude 22 that E ⊗F L is a regular (F ′ ⊗F L′ )-algebra. In particular, E ⊗F L is an integral domain and its fraction field Ẽ is a regular extension of the field F ′ ⊗F L′ ([Bou90, Proposition V.17.4]). Recall that K is algebraically closed in the field F ′ ⊗F L′ (by the first step). But F ′ ⊗F L′ is algebraically closed in Ẽ, as Ẽ is regular over F ′ ⊗F L′ . We conclude that K is algebraically closed in Ẽ which completes the second step. Third step: With notation as above, for 1 ≤ i ≤ 2m we define R̃i = Ri ⊗F L; this is an integral domain, being contained in Ẽ. We also let Ẽi = Frac(R̃i ). Thus Ẽi is the compositum of Ei and L inside Ẽ. We claim that CẼi = K for some i. Suppose to the contrary that CẼi strictly contains K for all 1 ≤ i ≤ 2m. Then for each i, Ẽi contains a constant that is transcendental over K (since K is algebraically closed in Ẽi ⊆ Ẽ by the second step) and thus transcendental over L (by Lemma 4.9, using CL = K). Therefore, trdeg(CẼi L/L) ≥ 1 for all i and thus trdeg(Ẽi /CẼi L) ≤ trdeg(Ẽi /L) − 1 ≤ trdeg(Ei /F ) − 1 = dim(G) − 1, where the last equality follows from the fact that Ri /F is a Picard-Vessiot ring with differential Galois group G. Base change from CẼi L to CẼ L then yields trdeg(Ẽi CẼ /CẼ L) ≤ dim(G) − 1 for all 1 ≤ i ≤ 2m. By Lemma 4.10, the compositum E1 · · · E2m equals E. Therefore, Ẽ equals the compositum of Ẽ1 , . . . , Ẽ2m inside Ẽ and we conclude that (1) trdeg(Ẽ/CẼ L) ≤ 2m(dim(G) − 1). On the other hand, any subset of CẼ that is algebraically independent over K remains algebraically independent over E by Lemma 4.9, since CE = K. Therefore, trdeg(CẼ /K) = trdeg(ECẼ /E) ≤ trdeg(Ẽ/E) and we obtain (2) trdeg(CẼ L/L) ≤ trdeg(CẼ /K) ≤ trdeg(Ẽ/E) ≤ trdeg(L/F ) = m − 1, where we used that Ẽ is a compositum of E and L to obtain the last inequality. Equations (1) and (2) together yield trdeg(Ẽ/L) ≤ 2m dim(G) − m − 1, and therefore trdeg(E/F ) ≤ trdeg(Ẽ/F ) ≤ 2m dim(G) − 2. But R/F is a Picard-Vessiot ring with differential Galois group G2m , so trdeg(E/F ) = 2m dim(G), a contradiction, which completes the third step. Conclusion of the proof: We conclude that for some 1 ≤ i ≤ 2m, R̃i = Ri ⊗F L is an integral domain and Ẽi = Frac(R̃i ) satisfies CẼi = K. By Proposition 1.8, R̃i is a Picard-Vessiot ring over L with Aut∂ (R̃i /L) ∼  = Aut∂ (Ri /F ) ∼ = G. A field of characteristic zero with a nontrivial derivation is always a transcendental extension of its field of constants (since derivations extend uniquely to separable algebraic extensions); hence we may find a copy of a rational function field within the given field. This gives the following Corollary 4.13. Let L be a differential field that is finitely generated over its field of constants K 6= L. Let G be a linear algebraic group defined over K with the property that for any r ∈ N, Gr is a differential Galois group over K(x); here K(x) denotes a rational function field with derivation d/dx. Then G is a differential Galois group over L. 23 Proof. Let z ∈ L be transcendental over K. By Lemma 4.2, we may assume that ∂(z) = 1 by replacing ∂ with ∂(z)−1 · ∂. Then F = K(z) is a differential subfield of L differentially isomorphic to K(x) and the claim follows from Theorem 4.12.  Combining Theorem 4.5 and Corollary 4.13, we obtain our main result: Theorem 4.14. Let K = k((t)) be a field of Laurent series over a field k of characteristic zero. Let L/K be a finitely generated field extension with a non-trivial derivation ∂ on L such that CL = K. Then every linear algebraic group defined over K is a differential Galois group over L. References [AM05] Katsutoshi Amano and Akira Masuoka. Picard-Vessiot extensions of Artinian simple module algebras. J. Algebra, 285 no. 2 (2005), 743–767. [And01] Yves André. Différentielles non-commutatives et théorie de Galois différentielle ou aux différences. Annales Scient. E.N.S. 34 (2001), 685–739. [AB94] Dmitri Anosov and Andrei Bolibruch. The Riemann-Hilbert problem. Aspects of Mathematics E22. Friedr. Vieweg & Sohn, Braunschweig, 1994. [Bou90] Nicolas Bourbaki. Algebra. II. Chapters 4–7. Elements of Mathematics (Berlin). Springer, Berlin, 1990. [BS64] Armand Borel and Jean-Pierre Serre. Théorèmes de finitude en cohomologie galoisienne. Comment. Math. Helv. 39 (1964), 111–164. [Car62] Pierre Cartier. Groupes algébriques et groupes formels. Colloq. Théorie des Groupes Algébriques (Bruxelles, 1962), 87–111. [CT00] Jean-Louis Colliot-Thélène. Rational connectedness and Galois covers of the projective line. Ann. of Math. 151 (2000), 359–373. [CHvdP13] Teresa Crespo, Zbigniew Hajto and Marius van der Put. Real and p-adic Picard-Vessiot fields. 2013 manuscript. Available at arXiv:1307.2388. [Dyc08] Tobias Dyckerhoff. The inverse problem of differential Galois theory over the field R(z). 2008 manuscript, available at arXiv:0802.2897. [Eps55a] Marvin P. Epstein. On the theory of Picard-vessiot extensions. Ann. of Math. 62 (1955), 528– 547. [Eps55b] Marvin P. Epstein. An existence theorem in the algebraic study of homogeneous linear ordinary differential equations. Proc. Amer. Math. Soc. 6 (1955), 33–41. [Gro71] Alexander Grothendieck. Revêtements étales et groupe fondamental. Séminaire de Géométrie Algébrique (SGA) 1. Lecture Notes in Math. 224, Springer-Verlag, Berlin, Heidelberg and New York, 1971. [Har87] David Harbater. Galois coverings of the arithmetic line. In Number Theory: New York, 1984-85. Springer LNM, vol. 1240 (1987), 165–195. [Har03] David Harbater. Patching and Galois theory. In: Galois Groups and Fundamental Groups (L. Schneps, ed.), MSRI Publications series 41, Cambridge University Press (2003), 313–424. [HH10] David Harbater and Julia Hartmann. Patching over fields. Israel J. Math., 176 (2010), 61–107. [HHK11] David Harbater, Julia Hartmann, and Daniel Krashen. Local-global principles for torsors over arithmetic curves. 2011 manuscript, available at: ArXiv: 1108.3323. To appear in Amer. J. Math. [Har05] Julia Hartmann. On the inverse problem in differential Galois theory. J. reine angew. Math. 586 (2005), 21–44 [Har07] Julia Hartmann. Patching and differential Galois groups (joint work with David Harbater). In: Arithmetic and Differential Galois Groups. Oberwolfach reports 4, No.2, European Mathematical Society (2007). 24 [Hru02] [Kov69] [Kov71] [MS96] [MS02] [MB01] [Oor66] [Ple08] [Pop96] [Sei56] [Sin93] [Spr09] [TT79] [vdPS03] [Völ96] Ehud Hrushovski. Computing the Galois group of a linear differential equation. Differential Galois theory (Bȩdlewo, 2001), 97–138, Banach Center Publ., vol. 58, Polish Acad. Sci., Warsaw, 2002. Jerald Kovacic. The inverse problem in the Galois theory of differential fields. Ann. of Math. 89 (1969), 583–608. Jerald Kovacic. On the inverse problem in the Galois theory of differential fields. Ann. of Math. 93 (1971), 269–284. Claudine Mitschi and Michael Singer. Connected Linear Groups as Differential Galois Groups. J. Algebra 184 (1996), 333–361. Claudine Mitschi and Michael Singer. Solvable-by-Finite Groups as Differential Galois Groups. Ann. Fac. Sci. Toulouse Math. (6) 11/3 (2002), 403–423. Laurent Moret-Bailly. Construction de revêtements de courbes pointées. J. Algebra 240 (2001), 505–534. Frans Oort. Algebraic group schemes in characteristic zero are reduced. Invent. Math. 2 (1966), 79–80. Josip Plemelj. Riemannsche Funktionenscharen mit gegebener Monodromiegruppe. Monatsh. Math. Phys. 19 (1908), no. 1, 211–246. Florian Pop. Embedding problems over large fields. Ann. of Math. 144 (1996), 1–34. Abraham Seidenberg. Contribution to the Picard-Vessiot theory of homogeneous linear differential equations. Amer. J. Math. 78 (1956), 808–818. Michael Singer. Moduli of Linear Differential Equations on the Riemann Sphere with Fixed Galois Groups. Pac. J. of Math. 106(2) (1993), 343–395. Tonny Albert Springer. Linear algebraic groups. Birkhäuser Boston Inc., second edition, 2009. Carol Tretkoff and Marvin Tretkoff. Solution of the Inverse Problem of Differential Galois Theory in the Classical Case. Amer. J. Math. 101, No. 6 (1979), 1327–1332. Marius van der Put and Michael F. Singer. Galois theory of linear differential equations. Springer, Berlin, 2003. Helmut Völklein. Groups as Galois groups: An introduction. Volume 53 of Cambridge Studies in Advanced Mathematics. Cambridge University Press, Cambridge, 1996. Author Information: David Harbater: Department of Mathematics, University of Pennsylvania, Philadelphia, PA 19104-6395, USA; email: [email protected] Julia Hartmann: Lehrstuhl für Mathematik (Algebra), RWTH Aachen University, 52056 Aachen, Germany; current address: Department of Mathematics, University of Pennsylvania, Philadelphia, PA 19104-6395, USA; email: [email protected] Annette Maier: Fakultät für Mathematik, Technische Universität Dortmund, D-44221 Dortmund, Germany; email: [email protected] The first author was supported in part by NSF grants DMS-0901164 and DMS-1265290, and NSA grant H98230-14-1-0145. The second author was supported by the German Excellence Initiative via RWTH Aachen University and by the German National Science Foundation (DFG). 25