Hypertranscendence and linear difference equations

Boris Adamczewski; Thomas Dreyfus; Charlotte Hardouin

doi:10.1090/jams/960

Adamczewski, Boris;Dreyfus, Thomas;Hardouin, Charlotte

2019-10-04 00:00:00

HYPERTRANSCENDENCE AND LINEAR DIFFERENCE EQUATIONS BORIS ADAMCZEWSKI, THOMAS DREYFUS, AND CHARLOTTE HARDOUIN Abstract. After Hölder proved his classical theorem about the Gamma function, there has been a whole bunch of results showing that solutions to linear diﬀerence equations tend to be hypertranscendental (i.e., they cannot be solution to an algebraic diﬀerential equation). In this paper, we obtain the ﬁrst complete results for solutions to general linear diﬀerence equations associated with the shift operator x 7→ x + h ∗ ∗ (h ∈ C ), the q-diﬀerence operator x 7→ qx (q ∈ C not a root of unity), and the Mahler operator x 7→ x (p ≥ 2 integer). The only restriction is that we constrain our solutions to be expressed as (possibly ramiﬁed) Laurent series in the variable x with complex coeﬃcients (or in the vari- able 1/x in some special case associated with the shift operator). Our proof is based on the parametrized diﬀerence Galois theory initiated by Hardouin and Singer. We also deduce from our main result a general statement about algebraic independence of values of Mahler functions and their derivatives at algebraic points. 1. Introduction Number theorists have to face the following challenge. On the one hand, the ﬁelds of rational and algebraic numbers are too poor, while, on the other hand, the ﬁelds of real and complex numbers seem far too large (uncount- able). An attempt to classify transcendental numbers leads to the introduc- tion of the ring of periods P (see [KZ01]) and to the following classiﬁcation: Q ⊂ Q ⊂ P ⊂ C . The main features of P are that most of classical mathematical constants (e.g., π, log 2, ζ(3), Γ(1/3)...) belong to it, and that P is a countable set whose elements contain only a ﬁnite amount of information and can be identiﬁed in an algorithmic way. After the pioneering works of Cauchy and Riemann, analytic and holomorphic functions ﬁgure prominently in mathe- matics. As already observed by Hilbert [Hil02], their study gives rise to a similar diﬃculty: the ﬁelds of rational and algebraic functions are too limited, but the ring C[[x]] is huge. In order to overcome this deﬁciency, he suggested to classify transcendental functions according to whether they are holonomic Date: March 10, 2020. 2010 Mathematics Subject Classiﬁcation. 39A06, 12H05. This project has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation program under the Grant Agreement No 648132. However, there should be some exceptions. For instance, it is conjectured that e is not a period. arXiv:1910.01874v2 [math.NT] 9 Mar 2020 2 BORIS ADAMCZEWSKI, THOMAS DREYFUS, AND CHARLOTTE HARDOUIN (i.e., satisfy a linear diﬀerential equation with polynomial coeﬃcients), dif- ferentially algebraic (i.e., satisfy an algebraic diﬀerential equation), or not. Hence, we obtain the following classiﬁcation: (1.1) Rat ⊂ Alg ⊂ Hol ⊂ Dif ⊂ C[[x]] . Functions that are not diﬀerentially algebraic are said to be hypertranscen- dental. Note that, if we restrict our attention to power series with algebraic coeﬃcients, the rings Hol and Dif become countable, and, again, their ele- ments contain only a ﬁnite amount of information and can be identiﬁed in an algorithmic way. These rings play, in the setting of holomorphic functions, a role similar to the one played by P in the setting of complex numbers. The classiﬁcation (1.1) is also signiﬁcant in enumerative combinatorics. To some extent, the nature of a generating series reﬂects the underlying structure of the objects it counts (see the discussion in [BM06]). An appealing example, where all cases of (1.1) appear at once, is given by the study of generat- ing series associated with lattice walks (see, for instance, [DHRS18] and the references therein). Hilbert also observed that the class of diﬀerentially algebraic functions misses important functions coming from number theory. Here are few exam- ples. The very ﬁrst comes from a classical result of Hölder [Höl87] stating that the gamma function Γ(x) is hypertranscendental. Furthermore, it fol- lows from the identity πx x−1 ζ(x) = 2(2π) Γ(1 − x) sin ζ(1 − x) that the Riemann zeta function is also hypertranscendental. In an other di- rection, Moore [Moo96] shown that f (x) = x is hypertranscenden- n≥0 tal. A result reproved and extended later by Mahler [Mah30a] in connection with the so-called Mahler’s method in transcendental number theory. More recently, Hardouin and Singer [HS08] proved hypertranscendence for some q-hypergeometric series, such as 2 2 n−1 2 (1 − a) (1 − aq) ··· (1 − aq ) f (x) := x , 2 2 2 n 2 (1 − q) (1 − q ) ··· (1 − q ) n=0 ∗ Z 2 Z where q ∈ C is not a root of unity, a 6∈ q and a ∈ q . Most proofs about hypertranscendence turn out to be based on the fact that the function under consideration satisﬁes a functional equation of a diﬀerent type (i.e., not dif- ferential). For instance, the three results we just mentioned are respectively based on the following linear diﬀerence equations: Γ(x + 1) = xΓ(x) , f (x ) = f (x) − x , 1 1 and 2ax − 2 x − 1 f (q x) − f (qx) + f (x) = 0 . 2 2 2 2 2 a x − 1 a x − 1 Broadly speaking, satisfying both a linear diﬀerence equation and an alge- braic diﬀerential equation would enforce too much symmetry for a transcen- dental function. In this paper, we follow this motto: (A) A solution to a linear diﬀerence equation should be either rational or hypertranscendental. HYPERTRANSCENDENCE AND LINEAR DIFFERENCE EQUATIONS 3 Of course, this must be taken with a pinch of salt. For instance, log(x), −n(n−1)/2 n exp(x), and the Jacobi theta function θ (x) = q x are all n∈Z diﬀerentially algebraic, but they also satisfy the following simple linear dif- ference equations: log(x ) = p log x , exp(x + 1) = e exp(x) , θ (qx) = qxθ (x) . q q Nevertheless, we obtain in this paper the ﬁrst complete results about hyper- transcendence of solutions to general linear diﬀerence equations associated with the shift operator x 7→ x+h (h ∈ C ), the q-diﬀerence operator x 7→ qx ∗ p (q ∈ C not a root of unity), and the Mahler operator x 7→ x (p ≥ 2 inte- ger). The only restriction is that we constrain our solutions to be expressed as (possibly ramiﬁed) Laurent series in the variable x (or in the variable 1/x in some special case associated with the shift operator). For a precise state- ment, we refer the reader to our main result, Theorem 1.2, which conﬁrms guess (A) for these three operators. 1.1. Statement of our main result. All along this paper, our framework consists in a tower of ﬁelds extensions C ⊂ K ⊂ F ⊂ F with the following properties. • The ﬁeld K is equipped with an automorphism ρ and a derivation ∂ . • The automorphism ρ extends to F and F . • The derivation ∂ extends to F , but F is not necessarily closed under ∂. Our aim is to prove that, if f ∈ F is solution to a linear ρ-equation with coeﬃcients in K, then either f ∈ K or f is hypertranscendental over K. Speciﬁcally, we consider the following four situations, which we refer to as Cases S , S , Q, and M, respectively. 0 ∞ Case S . In this case, we let K = C(x), ∂ = , and ρ denote the automor- dx phism of K deﬁned by ρ(f(x)) = f(x + h), h ∈ C . The automorphism ρ and the derivation ∂ naturally extend to the of ﬁeld F := Mer(C) of meromorphic functions on C. Let K := C(x, exp(∗x)) denote the ﬁeld generated over C(x) by the functions exp(ℓx), ℓ ∈ C. In this setting, we let F be any ﬁeld extension of C(x) on which ρ extends and satisfying the following three conditions. (i) F ⊂ Mer(C). (ii) {f ∈ F | ρ(f) = f} = C. (iii) F ∩ K = C(x). 0 0 Remark 1.1. For h = 1, let us see that we can choose F = C(x, Γ(x)), where Γ(x) is the Gamma function. The functional equation Γ(x + 1) = xΓ(x) shows that ρ extends to F and it is well-known that Γ ∈ Mer(C). Thus, (i) holds. Now, let β ∈ F be such that ρ(β) = β. Writing β as a ratio- nal function in Γ with coeﬃcients in C(x), and then using the functional equation satisﬁed by Γ and the fact that Γ is transcendental over C(x), A derivation ∂ on K is a map from K into itself satisfying ∂(a + b) = ∂(a) + ∂(b) and ∂(ab) = ∂(a)b + a∂(b) for every a, b ∈ K. 4 BORIS ADAMCZEWSKI, THOMAS DREYFUS, AND CHARLOTTE HARDOUIN we can deduce that β ∈ C. Hence (ii) holds. Finally, to prove (iii) it is suﬃcient to show that Γ is transcendental over K . The latter property fol- lows from classical asymptotics showing that, for every α(x) ∈ K , we have lim |Γ(x)/α(x)| = +∞. Thus, Case S of Theorem 1.2 is a generaliza- x→+∞ 0 tion of Hölder’s theorem. Case S . In this case, we consider K, ρ, and ∂ as in Case S . We note that ∞ 0 the automorphism ρ and the derivation ∂ naturally extend to the of ﬁeld of −1 −1 Laurent series C((x )), and we set F = F = C((x )). 1/j Case Q. In this case, we let K = C(x ) denote the ﬁeld of ramiﬁed j≥1 1/∗ rational functions. We also use the notation C(x ) for this ﬁeld. Given q ∈ C that is not a root of unity, we let ρ denote the automorphism of K deﬁned by ρ(f(x)) = f(qx), ∂ = x , dx 1/j and we let F = F = C((x )) denote the ﬁeld of Puiseux series. We j≥1 1/∗ also use the notation C((x )) for this ﬁeld. 1/∗ Case M. In this case, we let K = C(x ) and, given a natural number p ≥ 2, we let ρ denote the automorphism of K deﬁned by ρ(f(x)) = f(x ) , d 1/∗ and we set ∂ = x and F = F = C((x )). dx In all cases, ∂ is a derivation on K and F . An element f in F is said to be diﬀerentially algebraic or ∂-algebraic over K if there exists n ∈ N such (n) that the functions f, ∂(f), . . . , ∂ (f) are algebraically dependent over K. Otherwise, f is said to be hypertranscendental over K. Our main result reads as follow. Theorem 1.2. Let K, F , and ρ be deﬁned as in Cases S , S , Q, and 0 0 ∞ M, and let n be a positive integer. Let f ∈ F be solution to the ρ-linear diﬀerence equation of order n n n−1 (1.2) ρ (y) + a ρ (y) + ··· + a ρ(y) + a y = 0 , n−1 1 0 where a , . . . , a ∈ K. Then either f ∈ K or f is hypertranscendental 0 n−1 over K. On the road to Theorem 1.2, there is already a whole bunch of partial results and a number of them are used in our proof. These results can roughly be divided into three diﬀerent types. • Those concerned with aﬃne equations of order one, that is with equations of the form ρ(y) = ay + b. References include [Höl87, Moo96, Mah30a, Nis84, Ran92, Ish98, Har08, HS08, Ngu12]. • Those concerned with equations whose diﬀerence Galois group is large (e.g., simple, semisimple, reductive). References include [HS08, DHR18, DHR16, AS17, ADR18]. In particular, such results have nice applications to equations of order 2. HYPERTRANSCENDENCE AND LINEAR DIFFERENCE EQUATIONS 5 • Those dealing with general equations, but reaching only nonholo- nomicity instead of hypertranscendence (see Theorem 1.3 and Re- mark 1.4 below). References include [Ram92, BG93, Béz94, SS19]. Among these results, we quote the following one, which may be considered as a ﬁrst step towards a proof of Theorem 1.2 (see Remark 1.4 for precise attribution). Theorem 1.3. In each of Cases S , S , Q, and M, the following property 0 ∞ holds. Any element of F that satisﬁes both a ρ-linear equation and a ∂-linear equation with coeﬃcients in K belongs to K. Remark 1.4. Theorem 1.3 can be rephrased by saying that, if f is as in Theorem 1.2, then either f ∈ K or f is not holonomic. Note that recently the authors of [SS19] give a uniform proof for all cases of Theorem 1.3, see Corollaries 3 and 5 therein. In each case, we also mention the original reference. Case S is due to Bézivin and Gramain [BG93]. Case S is due 0 ∞ to Schäfke and Singer [SS19, Corollary 5]. Furthermore, in Cases Q and M, 1/ℓ ∗ a change of variable of the form z = x , ℓ ∈ N , can be used to reduce the situation to the case where K = C(x) and F = C((x)). In the latter situation, Case Q is due to Ramis [Ram92], while Case M is due to Bézivin [Béz94]. 1.2. Strategy of proof. Elements of the rings Rat, Alg, and Hol are well-understood. We have recurrence formulas for their coeﬃcients, precise asymptotics, and a good understanding of their ﬁnitely many singularities. In contrast, our knowledge about diﬀerentially algebraic functions is much more limited (see, for instance, the survey [Rub89]). In fact, such functions can behave quite wildly. For example, the function x is diﬀerentially algebraic and admits the unit circle as a natural boundary . Thus, prov- ing hypertranscendence required more involved tools, and is substantially more diﬃcult than proving irrationality, transcendence, or nonholonomicity. In this direction, Hardouin and Singer [HS08] built a new Galois theory of diﬀerence equations, which is designed to measure the diﬀerential dependen- cies among solutions to linear diﬀerence equations. In particular, it applies to hypertranscendence. We refer the reader to [DV12] and [Har16] for an introduction to this topic. Our proof of Theorem 1.2 remains on this theory. Let us brieﬂy describe our strategy. Since f satisﬁes a linear diﬀerence equation, it is classical to associate with this equation its diﬀerence Galois group, which is a linear algebraic group that encodes the algebraic relations between the solutions to this equation. The more involved parametrized Ga- lois theory developed in [HS08] attaches to any linear diﬀerence equation, a geometric object, the parametrized Galois group, whose structure encodes the algebraic relations satisﬁed by the solutions and their derivatives. This is not a linear algebraic group anymore, but a linear diﬀerential group (see Section 2.2). Nevertheless, both groups are strongly related and, viewed on a suitable ﬁeld extension, the latter is Zariski dense in the former (see Proposi- tion 2.5). In this framework, we can conclude that f is hypertranscendental Incidentally, this example shows that the "outrageous conjecture" suggested at the P 2 −n end of [Rub89] is false. Indeed, it follows from a theorem of Nesterenko that 2 is transcendental. 6 BORIS ADAMCZEWSKI, THOMAS DREYFUS, AND CHARLOTTE HARDOUIN when the corresponding parametrized Galois group is large. Furthermore, there are several results, culminating in [AS17], showing that if the classical diﬀerence Galois group is large, then both groups are the same when viewed on a suitable ﬁeld extension. We emphasize that such results follow a strat- egy initiated in [HS08] and developed further in [DHR18]. It combines a fundamental result about classiﬁcation of diﬀerential algebraic subgroups of semisimple algebraic groups by Cassidy [Cas72, Cas89], parametrized Galois correspondence, and Theorem 1.3. We prove Theorem 1.2 by induction on the order n of the ρ-linear equation satisﬁed by f. We ﬁrst prove the result for aﬃne equation of order one. Then we show that, iterating the associated linear system if necessary, one can reduce the situation to the case where the associated diﬀerence Galois group G is connected. Now, if G acts irreducibly on C , we show how to reduce the situation to the case where G is semisimple. In that case, the diﬀerence Galois group is large enough and a recent result of Arreche and Singer [AS17] allows us to conclude that f is hypertranscendental. Finally, if G acts reducibly on C , we assume that f is diﬀerentially algebraic and we show how to construct from f a diﬀerentially algebraic function g that satisﬁes a linear equation of smaller order. Furthermore, this construction ensures that g ∈ K if and only if f ∈ K. Applying the induction assumption to g, we deduce that f ∈ K, as wanted. Thus, once the case of order one equation is solved, the proof really makes a dichotomy between irreducible and reducible diﬀerence Galois groups. 1.3. An application to number theory. At the end of the 1920s, Mahler [Mah29, Mah30b, Mah30a] introduced a new method for proving transcen- dence and algebraic independence of values at algebraic points of analytic functions satisfying diﬀerent types of functional equations associated with the operator x 7→ x . Mahler’s method aims at transferring results about algebraic independence over Q(x) of such functions to the algebraic inde- pendence over Q of their values at algebraic points. A major result in this direction is Nishioka’s theorem [Nis90], which provides an analogue of the Siegel-Shidlovskii theorem for linear Mahler systems. Let us also mention that signiﬁcant progress in this theory has been made recently [Phi15, AF17, AF18a, AF18b]. Combining Case M of Theorem 1.2 with Nishioka’s theorem and following the line of argument given in the proof of Proposition 4.1 in [AF18b], we can deduce the following general result. We recall that f(x) ∈ Q[[x]] is a p-Mahler function if there exist polynomials a (x), . . . , a (x) ∈ Q[x], not all zero, such that 0 n p p a (x)f(x) + a (x)f(x ) + ··· + a (x)f(x ) = 0 . 0 1 n We simply say that f is a Mahler function if it is p-Mahler for some integer p ≥ 2. According to Theorem 1.2, a Mahler function is thus either ratio- nal or hypertranscendental. We recall that an irrational Mahler function is meromorphic on the open unit disc and admits the unit circle as a natural boundary. Theorem 1.5. Let f(x) ∈ Q[[z]] be a Mahler function that is not rational, let r be a positive integer, and let K be a compact subset of the open unit HYPERTRANSCENDENCE AND LINEAR DIFFERENCE EQUATIONS 7 disc. Then, for all but ﬁnitely many algebraic numbers α ∈ K, the complex ′ (r) numbers f(α), f (α), . . . , f (α) are algebraically independent over Q. Results of this type have been ﬁrst obtained by Mahler [Mah30a] for solu- tions to aﬃne order one equations. The main feature of Theorem 1.5 is that it applies to any irrational Mahler function. It is also almost best possible in the sense that one cannot avoid the possibility of ﬁnding inﬁnitely many exceptions in the whole open unit disc. For instance, the transcendental 2-Mahler function f(z) = (1 − 3z ) n=0 vanishes at all algebraic numbers α such that α = 1/3 for some integer n ≥ 0. This shows that, even for r = 1, the exceptional set can depend on K. However, if K is ﬁxed, we do not know whether the exceptional set always remains ﬁnite when r tends to inﬁnity. 1.4. Organization of the paper. This article is organized as follows. In Section 2, we provide a short introduction to diﬀerence Galois theory and to parametrized diﬀerence Galois theory, following [vdPS97, HS08]. In Sec- tion 3, we describe, for each of Cases S , S , Q, and M, the suitable 0 ∞ framework that is needed to use the parametrized diﬀerence Galois theory of [HS08]. Several auxiliary results are gathered in Section 4. They concern al- gebraic groups, the behavior of the diﬀerence Galois group when considered over diﬀerent ﬁeld extensions, and Picard-Vessiot extensions associated with iterated systems. Finally, Section 5 is devoted to the proof of Theorem 1.2. 2. Difference and parametrized difference Galois theories In this section, we provide a short introduction to diﬀerence Galois theory and to parametrized diﬀerence Galois theory. 2.1. Diﬀerence Galois theory. We ﬁrst recall some notation, as well as classical results, concerning diﬀerence Galois theory. We refer the reader to [vdPS97] for more details. 2.1.1. Notation in operator algebra. A diﬀerence ring is a pair (R, ρ) where R is a ring and ρ is a ring automorphism of R. If R is a ﬁeld, then (R, ρ) is called a diﬀerence ﬁeld or a ρ-ﬁeld. An ideal I of R such that ρ(I) ⊂ I is called a diﬀerence ideal or a ρ-ideal. We say that the diﬀerence ring (R, ρ) is simple if the only ρ-ideals of R are {0} and R. Two diﬀerence rings (R , ρ ) 1 1 and (R , ρ ) are isomorphic if there exists a ring isomorphism ϕ between R 2 2 1 and R such that ϕ ◦ ρ = ρ ◦ ϕ. A diﬀerence ring (S, ρ ) is a diﬀerence 2 1 2 ring extension of (R, ρ) if S is a ring extension of R and if ρ = ρ. In this |R case, we usually keep on denoting ρ by ρ. When R is a ρ-ﬁeld, we say that S is a R-ρ-algebra. Two diﬀerence ring extensions (R , ρ) and (R , ρ) of the 1 2 diﬀerence ring (R, ρ) are isomorphic over (R, ρ) if there exists a diﬀerence ring isomorphism ϕ from (R , ρ) to (R , ρ) such that ϕ = Id . The ring 1 2 |R R of constants of the diﬀerence ring (R, ρ) is deﬁned by R := {f ∈ R | ρ(f) = f} . 8 BORIS ADAMCZEWSKI, THOMAS DREYFUS, AND CHARLOTTE HARDOUIN If R is a ﬁeld, it is called the ﬁeld of constants. If there is no risk of confusion, we usually simply say that R, instead of (R, ρ), is a diﬀerence ring (or a diﬀerence ﬁeld, or a diﬀerence ring extension...). 2.1.2. Diﬀerence equations and linear diﬀerence systems. A linear ρ- equation of order n over K is an equation of the form n n−1 (2.1) L(y) := ρ (y) + a ρ (y) + ··· + a y = 0 , n−1 0 with a , ..., a ∈ K. If a 6= 0, this relation can be written in matrix form 0 n−1 0 as (2.2) ρY = A Y where   0 1 0 ··· 0   . . 0 0 1 .     . . . . A := ∈ GL (K) .   L . . . . n . . . . 0     0 0 ··· 0 1 −a −a ··· ··· −a 0 1 n−1 The matrix A is called the companion matrix associated with Equa- tion (2.1). It is often more convenient to use the notion of linear diﬀerence system, that is of system of the form (2.3) ρ(Y ) = AY, with A ∈ GL (K). We recall that two diﬀerence systems ρ(Y ) = AY and ρ(Y ) = BY with A, B ∈ GL (K) are said to be equivalent over K if there exists a gauge trans- −1 formation T ∈ GL (K) such that B = ρ(T )AT . In that case, ρ(Y ) = AY if and only if ρ(TY ) = B(TY ). Remark 2.1. By [HS99, Theorem B.2.], if K contains a nonperiodic ele- ment with respect to ρ, then the cyclic vector lemma holds, and any linear diﬀerence system is equivalent to one given by a companion matrix associ- ated with some linear equation. Let L be a ρ-ﬁeld extension of K and let ⊤ n (f , f , . . . , f ) ∈ L be a nonzero solution to a linear system ρ(Y ) = AY , 0 1 n−1 with A ∈ GL (K). Then each coordinate f satisﬁes some linear ρ-equation n j over K of order at most n. 2.1.3. Picard-Vessiot rings and diﬀerence Galois groups. A Picard-Vessiot ring for (2.3) over a diﬀerence ﬁeld (K, ρ) of characteristic zero is a K-ρ- algebra satisfying the following three properties. (1) There exists U ∈ GL (R) such that ρ(U) = AU. Such a matrix U is called a fundamental matrix. (2) R is generated as a ring over K by the coordinates of U and by −1 −1 det(U) , that is R = K[U, det(U) ]. (3) R is a simple diﬀerence ring. A Picard-Vessiot ring is not always an integral domain. However, it is a direct sum of integral domains which are transitively permuted by ρ. More precisely, by [vdPS97, Corollary 1.16], there exist r ≥ 1 and orthogonal idempotent elements e , . . . , e , such that 0 r−1 R = e R ⊕ ··· ⊕ e R , 0 r−1 HYPERTRANSCENDENCE AND LINEAR DIFFERENCE EQUATIONS 9 where e R is an integral domain and ρ(e ) = e . We recall that i i i+1modr Picard-Vessiot rings always exist. Indeed, it is suﬃcient to endow the poly- nomial K-algebra K[X, ] with a structure of K-ρ-algebra by setting det(X) ρ(X) = AX. Then, for any maximal ρ-ideal M of K[X, ], the quotient det(X) R = K[X, ]/M is a Picard-Vessiot ring. However, this construction det(X) does not guarantee that the ring of ρ-constant has not grown. This justiﬁes the introduction of the more convenient notion of Picard-Vessiot extension. A Picard-Vessiot extension Q for (2.3) over a diﬀerence ﬁeld (K, ρ) of characteristic zero is a K-ρ-algebra Q satisfying the following properties. (1) There exists U ∈ GL (Q) such that ρ(U) = AU. Such a matrix U is called a fundamental matrix. (2) Q is a pseudoﬁeld extension of K, that is there exist a positive integer r, orthogonal idempotent elements e , . . . , e of Q, and a ﬁeld ex- 0 r−1 tension L of K such that Q = e L⊕. . .⊕e L and ρ(e ) = e . 0 r−1 i i+1modr (3) Q is the smallest pseudoﬁeld extension of K containing U. ρ ρ (4) Q = K . When Q is a ﬁeld, we say that Q is a Picard-Vessiot ﬁeld extension. By [vdPS97, §1.1], if K is algebraically closed, then there exists a Picard-Vessiot extension and two Picard-Vessiot extensions are isomorphic as K-ρ-algebra. The relation between Picard-Vessiot-rings and Picard-Vessiot-extensions is given by the following proposition, which is a straightforward consequence of [OW15, Proposition 2.5 and Corollary 2.6]. Proposition 2.2. If K is an algebraically closed ﬁeld of characteristic zero, then the following properties hold. • Let Q be a Picard-Vessiot extension over K and let us deﬁne R := K[U, ] ⊂ Q, where U is a fundamental matrix. Then R is det(U) a Picard-Vessiot ring. • If R is a Picard-Vessiot ring over K then the total quotient ring of R is a Picard-Vessiot extension. From now on, we assume that K is a ρ-ﬁeld such that k := K is an algebraically closed ﬁeld of characteristic zero. Let Q be a Picard-Vessiot extension over K. The diﬀerence Galois group Gal(Q/K) of (2.3) over K is deﬁned as the group of K-ρ-algebra automorphisms of Q: Gal(Q/K) := {σ ∈ Aut(Q/K) | ρ ◦ σ = σ ◦ ρ}. For any fundamental matrix U ∈ GL (Q), an easy computation shows −1 that U σ(U) ∈ GL (k) for all σ ∈ Gal(Q/K). By [vdPS97, Theorem 1.13], the faithful representation Gal(Q/K) → GL (k) −1 σ 7→ U σ(U) We recall that, given a ring R, its total quotient ring is deﬁned as the localization of R at the multiplicative set formed by all nonzero divisors of R. 10 BORIS ADAMCZEWSKI, THOMAS DREYFUS, AND CHARLOTTE HARDOUIN identiﬁes Gal(Q/K) with a linear algebraic subgroup G ⊂ GL (k). Choosing another fundamental matrix of solutions U leads to a conjugate representa- tion. 2.1.4. Torsor and algebraic relations. A fundamental result in diﬀerence Ga- lois theory ([vdPS97, Theorem 1.13]) says that the Picard-Vessiot ring R is the coordinate ring of a Gal(Q/K)-torsor over K. Thereby, the diﬀerence Galois group controls the algebraic relations satisﬁed by the solutions to the underlying linear system. As a corollary of this structure of Gal(Q/K)- torsor, one obtains the fundamental equality dim Gal(Q/K) = trdeg Q , where we let trdeg Q denote the transcendence degree of the ﬁeld extension L/K. Here, the ﬁeld L is deﬁned as in (2) of the deﬁnition of a Picard-Vessiot extension. 2.1.5. The Galois correspondence. The Galois correspondence for linear dif- ference systems can be summarized as follows (see [vdPS97, Theorem 1.29]). Theorem 2.3. Let Q be a Picard-Vessiot extension over K. Let R denote the set of K-ρ-algebras F such that F ⊂ Q and such that every nonzero divisor of F is a unit of F . Let G denote the set of algebraic subgroups of Gal(Q/K). Then the following properties hold. - For any F ∈ R, the set G(Q/F ) := {σ ∈ Gal(Q/K) | σ(f) = f, ∀f ∈ F} is an algebraic subgroup of Gal(Q/K). - For any H ∈ G, the set Q := {f ∈ Q | σ(f) = f, ∀σ ∈ H} belongs to R. - The maps r : R → G g : G → R and F 7→ G(Q/F ) H 7→ Q are each other’s inverses. Remark 2.4. In the case where R is an integral domain, Q is a diﬀerence ﬁeld, and Theorem 2.3 provides a correspondence between the diﬀerence subﬁelds of Q containing K, on the one hand, and the algebraic subgroups of Gal(Q/K), on the other hand. 2.2. Parametrized diﬀerence Galois theory. We recall here some nota- tion and results concerning parametrized diﬀerence Galois theory. We refer the reader to [HS08] for more details. 2.2.1. Notation in operator algebra. A diﬀerential ring is a pair (R, δ) where R is a ring and δ is a derivation on R, that is a homomorphism of additive group from R into itself satisfying the Leibniz rule: δ(ab) = δ(a)b + aδ(b) . HYPERTRANSCENDENCE AND LINEAR DIFFERENCE EQUATIONS 11 If R is a ﬁeld, then (R, δ) is called a diﬀerential ﬁeld or a δ-ﬁeld. The ring of constants of R is deﬁned by R = {f ∈ R | δ(f) = 0} . Let L be a δ-ring extension of a δ-ring R. Given a subset S of L, we let R{S} denote the smallest R-δ-algebra generated by S. A ﬁeld K endowed with a structure of both diﬀerence and diﬀerential ﬁeld is called a (ρ, δ)-ﬁeld if ρ and δ commute, that is if ρ ◦ δ(f) = δ ◦ ρ(f) for every f ∈ K. The study of algebraic structures of diﬀerence and diﬀerential ﬁelds is the main object of the so-called diﬀerential and diﬀerence algebra. We refer the interested reader to the founding books of Kolchin ([Kol73]) for further details concerning diﬀerential algebra and to the book of Cohn for further details concerning diﬀerence algebra ([Coh65]). It is worth mentioning that these two setting are very diﬀerent. In order to avoid many analogous def- initions, we use the following convention: we add an operator preﬁx to an algebraic attribute to signify the compatibility of the algebraic structure with respect to the operator. For instance, a (ρ, δ)-ring is a ring equipped with ρ and δ, a (ρ, δ)-ﬁeld extension K ⊂ L is a ﬁeld extension such that the ﬁelds L and K are (ρ, δ)-ﬁelds and such that the action of ρ and δ are compatible with the ﬁeld extension, and so on. 2.2.2. (ρ, δ)-Picard-Vessiot ring and the parametrized diﬀerence Galois e e group. Let K be a (ρ, δ)-ﬁeld and let A ∈ GL (K). We consider the diﬀer- ence system (2.4) ρ(Y ) = AY. e e A (ρ, δ)-Picard-Vessiot ring for (2.4) over K is a (ρ, δ)-ring extension R of K satisfying the following three properties. (1) There exists U ∈ GL (R) such that ρ(U) = AU. Such a matrix U is called a fundamental matrix. e e (2) R is generated as a δ-ring over K by the coordinates of U and by −1 −1 e e det(U) , that is R = K{U, det(U) } . e e (3) R is a simple (ρ, δ)-ring, that is the only (ρ, δ)-ideals of R are {0} and R. A (ρ, δ)-Picard-Vessiot ring is not always an integral domain but it is a direct sum of integral domains closed under δ and transitively permuted by ρ. A construction similar to that of Section 2.1.3 shows that (ρ, δ)-Picard- Vessiot rings always exist. Again, this construction does not ensure that the ring of ρ-constants has not grown. This justiﬁes the introduction of the more convenient notion of (ρ, δ)-Picard-Vessiot extension. e e e A (ρ, δ)-Picard-Vessiot extension Q for (2.4) over K is a K-ρ-δ-algebra satisfying the following properties. (1) There exists U ∈ GL (Q) such that ρ(U) = AU. Such a matrix U is called a fundamental matrix. 12 BORIS ADAMCZEWSKI, THOMAS DREYFUS, AND CHARLOTTE HARDOUIN e e (2) Q is a pseudo δ-ﬁeld extension of K, that is there exist a positive integer r, orthogonal idempotent elements e , . . . , e of Q, and 0 r−1 e e e e e L a δ-ﬁeld extension of K such that Q = e L ⊕ . . . ⊕ e L and 0 r−1 ρ(e ) = e . i i+1modr e e (3) Q is equal to KhUi, that is the smallest pseudo δ-ﬁeld extension of K containing U. ρ ρ e e (4) Q = K . e e When Q is a ﬁeld, we say that Q is a (ρ, δ)-Picard-Vessiot ﬁeld exten- sion. By [Wib12, Corollary 9], if K is algebraically closed, then there e e exists a (ρ, δ)-Picard-Vessiot extension Q for (2.4) over K. The smallest e e e K-(ρ, δ)-subalgebra K{U, } of Q generated by a fundamental matrix det(U) and the inverse of its determinant is a (ρ, δ)-Picard-Vessiot ring. However, in order to ensure the uniqueness of (ρ, δ)-Picard-Vessiot rings up to K- (ρ, δ)-algebra isomorphisms, one needs the ﬁeld K to be δ-closed ([HS08, Proposition 6.16]). We recall that a diﬀerential ﬁeld (L, δ) is called diﬀer- entially closed or δ-closed if, for every set of δ-polynomials F, the system of δ-equations F = 0 has a solution in some δ-ﬁeld extension of L if and only if it has a solution in L. Note that a δ-closed ﬁeld is algebraically closed. Dif- ferentially closed ﬁelds are huge and, for instance, none of the function ﬁelds K introduced in Section 1.1 satisﬁes the assumption that K is diﬀerentially closed. Nonetheless, working with δ-closed ﬁelds allows us to simplify many arguments. Thus, we will embed K into a (ρ, δ)-ﬁeld K whose ﬁeld of ρ- constants is δ-closed, and we will use some descent argument to go back to K if necessary (see Sections 3 and 4.2). From now on, we let K denote a (ρ, δ)-ﬁeld whose ﬁeld of ρ-constants e e e e C := K is δ-closed. Let Q be a (ρ, δ)-Picard-Vessiot extension over K. e e e The parametrized diﬀerence Galois group Gal (Q/K) of (2.4) over (K, ρ, δ), also called the (ρ, δ)-Galois group, is deﬁned as the group of K-(ρ, δ)-algebra automorphisms of Q: n o e e e e Gal (Q/K) := σ ∈ Aut(Q/K) ρ ◦ σ = σ ◦ ρ and δ ◦ σ = σ ◦ δ . e e The parametrized diﬀerence Galois group Gal (Q/K) is a geometric ob- ject that encodes the diﬀerential algebraic relations between the solutions to (2.4). See [HS08, Proposition 6.24] for more details. Roughly speaking, the e e e larger Gal (Q/K), the fewer δ-algebraic relations over K that hold among the elements of Q. For any fundamental matrix U ∈ GL (Q), an easy computation shows −1 δ e e e that U σ(U) ∈ GL (C) for any element σ ∈ Gal (Q/K). By [HS08, Propo- sition 6.18], the faithful representation e e Gal (Q/K) → GL (C) −1 σ 7→ U σ(U) e e identiﬁes Gal (Q/K) with a linear diﬀerential algebraic subgroup H ⊂ e e GL (C), i.e. a subgroup of GL (C) that can be deﬁned as the vanishing n n HYPERTRANSCENDENCE AND LINEAR DIFFERENCE EQUATIONS 13 set of polynomial δ-equations with coeﬃcients in C. Furthermore, choosing another fundamental matrix U leads to a conjugate representation. 2.2.3. Comparison between diﬀerence Galois groups and parametrized diﬀer- ence Galois groups. Let R denote a (ρ, δ)-Picard-Vessiot ring for the system e e (2.4) over K, let Q denote the associated (ρ, δ)-Picard-Vessiot extension, and let U ∈ GL (R) be a fundamental matrix of solutions. By [HS08, Propo- −1 e e sition 6.21], R := K[U, det(U) ] is a Picard-Vessiot ring for (2.4) over K. Let Q denote the Picard-Vessiot extension corresponding to R. Then we e e e e have Q ⊂ Q and one can identify Gal (Q/K) with a subgroup of Gal(Q/K) e e by restricting the elements of Gal (Q/K) to Q. The following result pro- vides a fundamental link between classical diﬀerence Galois theory and its parametrized counterpart, see [HS08, Proposition 6.21]. e e Proposition 2.5. The parametrized diﬀerence Galois group Gal (Q/K) is a Zariski-dense subgroup of Gal(Q/K). 3. The parametrized framework Let us go back to our K, F , F, ρ, ∂ corresponding to Cases S , S , Q, 0 0 ∞ and M (see Section 1.1). We ﬁrst observe that, in each case, K = C is algebraically closed and K has characteristic zero. Hence the theory of [vdPS97] described in Section 2.1 applies. In this section, we describe some e e suitable ﬁeld extensions K of K, and C of C = F , that allow us to apply the parametrized diﬀerence Galois theory of [HS08] described in Section 2.2. We ﬁrst recall that any diﬀerential ﬁeld has a diﬀerential closure, i.e. a diﬀerentially closed ﬁeld extension that is minimal with respect to this property. In Case S , we set δ := ∂, and we let C denote a δ-closure of C , the subﬁeld of Mer(C) formed by those meromorphic functions that are h-periodic. Note that C is also a δ-ﬁeld for ρ and δ commute. Sometime, f e we will use the notation C instead of C. In Cases S and Q, we also set h ∞ e e e δ := ∂, and we let C denote a δ-closure of C and we set C := C. By [DHR18, Lemma 2.3], in these three cases, the ﬁeld e e K := Frac(K ⊗ C) e e is a (ρ, δ)-ﬁeld extension of K such that K = C. Case M is a bit more tricky. Indeed, in order to apply the theory of [HS08], we need to consider a derivation δ on K such that δ and ρ commute. Following [DHR18], we let log denote a transcendental element over K, and we deﬁne δ = log ×∂ to be the derivation that acts on K(log) by a a δ(log) = log and δ(x ) = ax log . Then we deﬁne a structure of ρ-ﬁeld on K(log) by setting ρ(log) = p log. We observe that K(log) = C and that ρ and δ commute. Finally, we consider e e e e C a δ-closure of C such that ρ acts trivially on C. We set C := C and by [DHR18, Lemma 2.3], the ﬁeld e e K := Frac(K(log) ⊗ C) , e e is a (ρ, δ)-ﬁeld such that K = C. 14 BORIS ADAMCZEWSKI, THOMAS DREYFUS, AND CHARLOTTE HARDOUIN In all cases, K is a ρ-ﬁeld of characteristic zero such that K = C, F is a ρ-ﬁeld extension of K with C := F , and K is a (ρ, δ)-ﬁeld of characteristic e e e zero, such that K = C, where C is diﬀerentially closed. Furthermore, ρ and δ commute. The following table summaries the diﬀerent frameworks we consider in this paper. e e Case F K C ∂ δ C K d d f f S Mer(C) C(x) C C C (x) 0 h h h dx dx −1 d d e e S C((x )) C(x) C C C(x) dx dx 1/∗ 1/∗ d d 1/∗ e e Q C((x )) C(x ) C x x C C(x ) dx dx 1/∗ 1/∗ d d 1/∗ e e M C((x )) C(x ) C x log ×x C C(x )(log) dx dx Remark 3.1. Note that the ﬁeld of δ-constants of C is the algebraically closed ﬁeld C. Then, by [CS07, Lemma 9.3], the ﬁeld of δ-constants of C is also C. At some places, we will consider some iteration of the operator ρ. In Case e f S , it might happen that we start with the ρ-ﬁeld extension K = C (x) 0 h of K, and then consider this extension as a ρ -ﬁeld extension. Note that f g f C 6= C . Nonetheless, C is a ρ -constant ﬁeld and all results of Section 2 h rh h can be applied to the (ρ , δ)-ﬁeld K. 4. Auxiliary results In this section, we let (K, ρ) be a diﬀerence ﬁeld of characteristic zero and we consider a ρ-linear diﬀerence system (4.1) ρ(Y ) = AY, with A ∈ GL (K) , where k := K is an algebraically closed ﬁeld. 4.1. Irreducible and reducible diﬀerence Galois groups. The strategy for proving Theorem 1.2 is very diﬀerent according to whether the diﬀerence Galois group associated with the diﬀerence equation over K satisﬁed by f is irreducible or not. By irreducible, we mean that the group acts irreducibly on k , see Deﬁnition 4.1. We point out that this is diﬀerent from saying that an algebraic group, viewed as an algebraic variety, is irreducible. Deﬁnition 4.1. Let G ⊂ GL (k) be a group. We say that G is irreducible n n if it acts irreducibly on k , that is, if the only k-vector subspaces of k invariant by G are {0} and k . We say that G is reducible otherwise. When G is an irreducible group, we say that G is imprimitive if there exist an integer r ≥ 2, and V , . . . , V , some k-vector spaces satisfying the following 1 r conditions. (i) k = V ⊕ ··· ⊕ V . 1 r (ii) For every g ∈ G, the mapping V 7→ g(V ) is a permutation of the set i i {V , . . . , V }. 1 r We say that G is primitive otherwise. The next result shows that a diﬀerence Galois group that is irreducible and imprimitive cannot be connected. We recall that an algebraic group G is connected if it has no proper open subgroup with respect to the Zariski HYPERTRANSCENDENCE AND LINEAR DIFFERENCE EQUATIONS 15 topology. The identity component of G is the connected component of G containing the identity. Lemma 4.2. Let G ⊂ GL (k) be an algebraic group. If G is connected and irreducible, then G is primitive. Proof. Let us assume by contradiction that G is imprimitive. Then there exist an integer r ≥ 2 and some nonzero C-vector-spaces V , . . . , V ( k 1 r such that k = V ⊕ ··· ⊕ V . Furthermore, the action induced by every 1 r element of G on k is a permutation of the sets {V , . . . , V }. Let G denote 1 r V the stabilizer of V in G. This is an algebraic subgroup of G. We claim that it has ﬁnite index in G. Let us assume by contradiction that there are an inﬁnite number of cosets g G of G in G. Since the group of permutations i V V 1 1 of a set with r elements is ﬁnite, there exist two distinct cosets g G and 1 V g G such that g and g induce the same permutation of the V ’s. In that 2 V 1 2 i −1 case, g g stabilizes V and thus belongs to G , providing a contradiction. 1 1 V 2 1 This proves the claim. Now, since G is a closed subgroup of G of ﬁnite index, it contains the identity component of G. The latter is equal to G for G is assumed to be connected. This means that V is invariant under the action of G. Hence, G is reducible. This provides a contradiction. Now, let us recall the deﬁnition of a diﬀerence module, see [vdPS97, §1.4]. Deﬁnition 4.3. A K-diﬀerence module M is a pair (M, Σ) where M is a ﬁnite-dimensional K-vector space and Σ is an additive endomorphism of M such that Σ(λY ) = ρ(λ)Σ(Y ) for every Y ∈ M and λ ∈ K. Let A ∈ GL (K). The K-diﬀerence module M attached to the system n A n −1 ρ(Y ) = AY is M = (K , Σ ), where Σ (Y ) = A ρ(Y ). Given a monic A A A linear diﬀerence operator L = a y + a ρ(y) + ··· + ρ (y) with coeﬃcients 0 1 a in K and a 6= 0, one can consider the corresponding linear diﬀerence i 0 system ρ(Y ) = A Y , where A is the companion matrix of L. If K contains L L a nonperiodic element with respect to ρ, then [HS99, Theorem B.2] asserts that any K-diﬀerence module M is isomorphic to M for some monic linear diﬀerence operator L. This is equivalent to the existence of a so-called cyclic vector in M, that is, a nonzero element e ∈ M such that the vectors n−1 e, Σ(e), . . . , Σ (e) form a K-basis of M. The next lemma characterizes linear diﬀerence systems whose diﬀerence Galois group is reducible. Lemma 4.4. Let G ⊂ GL (k) be the diﬀerence Galois group of (4.1) over K and let r be an integer with 0 < r < n. The following statements are equivalent. n ¶ • There exists a G-submodule of k of dimension r over k. • There exists a diﬀerence submodule of dimension r in the K- diﬀerence module M associated with (4.1). • There exists T ∈ GL (K) such that B B 1 2 −1 ρ(T )AT = 0 B ¶ n k is a G-module via G ⊂ GL (k). n 16 BORIS ADAMCZEWSKI, THOMAS DREYFUS, AND CHARLOTTE HARDOUIN with B ∈ GL (K). 1 r In particular, G is reducible if and only if the statements above hold for some r, with 0 < r < n. Furthermore, if the matrix A in (4.1) is the companion matrix of a diﬀerence operator L, then L admits a nontrivial right factor if and only if G is reducible. Proof. Since the category of K-diﬀerence modules is a Tannakian category, the ﬁrst two statements are nothing else than the usual Tannakian equiv- alence, see [And01, Théorème 3.2.1.1]. Completing a K-basis of a diﬀer- ence submodule M in M of dimension r, 0 < r < n, into a K-basis e = (e , . . . , e ) of K , shows that the matrix of Σ in the basis e is of the form 1 n B B 1 2 B = , with B ∈ GL (K). Let T ∈ GL (K) denote the matrix 1 r n 0 B associated with the change of basis from the canonical basis f = (f , . . . , f ) 1 n to e. Then BTf = Be = Σ(e) = Σ(Tf) = ρ(T )Af. This proves the equivalence between the second and the third statements. Now, let us assume that A is the companion matrix of some diﬀerence op- erator L and that there exists a nontrivial diﬀerence submodule N of M . Let e be a cyclic vector of M such that L is the minimal monic linear dif- ference operator annihilating e. Since N is a nontrivial diﬀerence submodule of M , then dim M /N < n and there exists a nontrivial monic linear A K A diﬀerence operator L of order smaller than or equal to dim M /N such 1 K A that L e ∈ N . The element L e is nonzero by minimality of L. Since N is a 1 1 diﬀerence submodule of M there exists a monic linear diﬀerence operator L of order smaller than or equal to dim N such that L L (e) = 0. By 2 K 2 1 minimality of L, this proves that L = L L , where L has order smaller 2 1 1 than n. Conversely, if there is a nontrivial factorization L = L L , then the 2 1 diﬀerence module corresponding to L is a nontrivial diﬀerence submodule of M. Hence G is reducible. 4.2. Going up and down. In this section, we let K, F , ρ be deﬁned as in Section 1.1, and we let K be deﬁned as in Section 3. The two following lemmas compare the diﬀerence Galois groups over K and over K. We recall that a connected linear algebraic group G is said to be semisim- ple if it is of positive dimension and has only trivial abelian normal sub- groups. Lemma 4.5. Let A ∈ GL (K) and let G (resp. G) be the Galois group of ρ(Y ) = AY over K (resp. K). The following properties hold. e e • In Cases S , S , and Q, we have G = G(C). 0 ∞ e e • In Case M, if G is semisimple, then G = G(C). Proof. In Cases S , S , and Q, the ﬁeld extension K of K is obtained by 0 ∞ extension of the ﬁeld of constants C to C. Then, [CHS08, Corollary 2.5] gives that the diﬀerence Galois group over K is just the extension of G to the C-points. Let us deal with Case M. Let G denote the diﬀerence Galois group over e e K(log). As previously, we infer from [CHS08, Corollary 2.5] that G = G(C). HYPERTRANSCENDENCE AND LINEAR DIFFERENCE EQUATIONS 17 Thus, it remains to prove that G = G when G is semisimple. By [DHR18, Proposition 1.6], we can see G as a subgroup of G such that one of the following two situations occurs. (1) G = G. (2) G is a normal subgroup of G and G/G ≃ G the multiplicative group (C , .). By [Mil17, Corollary 21.50], a semisimple algebraic group is perfect, that is equal to its derived subgroup, see [Mil17, Deﬁnition 12.45]. Thus, a semisim- ple algebraic group has no nontrivial abelian quotient. Hence, (2) cannot occur. We deduce that G = G, which ends the proof. Lemma 4.6. Let A ∈ GL (K) and let G (resp. G) be the Galois group of ρ(Y ) = AY over K (resp. K). The group G is irreducible if and only if the group G is irreducible. e e Proof. In Cases S , S , and Q, Lemma 4.5 ensures that G = G(C) and the 0 ∞ result follows directly. Let us deal with Case M. We infer from [CHS08, Corollary 2.5] that e e G = G(C), where we let G denote the Galois group of the system over K(log). Thus, it remains to prove that G is irreducible if and only if G is irreducible. As seen in the proof of Lemma 4.5, G is a subgroup of G. Thus, G is reducible as soon as G is reducible. Conversely, let us assume that G is reducible. By the cyclic vector lemma over K, the linear diﬀerence system (4.1) is equivalent to an equation n n−1 L := ρ + a ρ + ··· + a = 0 , n−1 0 with a ∈ K and a 6= 0. Since G is reducible, Lemma 4.4 gives that L i 0 admits a factorization over K(log). Multiplying L by some nonzero element if necessary, we can get rid of the denominators. Then, there exists α ∈ K[log] such that k k−1 n−k n−k−1 αL = (b ρ + b ρ + ··· + b )(c ρ + c ρ + ··· + c ) , k k−1 0 n−k n−k−1 0 where b , c ∈ K[log], b b c c 6= 0, and 0 < k < n. Rewriting the equation i i k 0 n−k 0 in terms of powers of log, one ﬁnds    X X X i κ j    log (α L) = log L log D , i κ j i≥i κ≥κ j≥j 0 0 0 where L ,D are linear diﬀerence operators over K, α ∈ K, α ∈ K , κ j i i and L := γ ρ and D are nonzero. Using the fact that log is κ r j 0 r=0 0 transcendental over K and ρ log = p log ρ, we deduce that i = κ + j . 0 0 0 Considering the terms in log in both sides of the factorization, we deduce 1 j r r that L = MD where M = γ p ρ , leading to a nontrivial fac- j r r=0 torization of L over K. By Lemma 4.4, we obtain that G is reducible, as wanted. Now, we prove the following descent lemma. Lemma 4.7. For F , K and K deﬁned as in Section 3, we have F ∩ K = K. 0 18 BORIS ADAMCZEWSKI, THOMAS DREYFUS, AND CHARLOTTE HARDOUIN 1/∗ Proof. Let us ﬁrst consider Case Q. Let a ∈ F ∩ C(x ). There exists a 1/r 1/r positive integer r such that a ∈ C((x )) ∩ C(x ). It follows that there N 1/r N exists a positive integer N such that x a ∈ C[[x ]]. Thus, x a is a for- 1/r mal power series in x with complex coeﬃcients that represents a rational fraction with coeﬃcients in C. This property can be characterized by the vanishing of its associated Kronecker-Hankel determinants (see, for instance, [Sal63, p. 5]). This condition does not depend on the ﬁeld of coeﬃcients, so N 1/r we deduce that x a ∈ C(x ). Hence, a ∈ K. 1/∗ The proof of Case S is entirely similar. For Case M, if a ∈ C((x )) ∩ 1/∗ C(x )(log), we use the fact that the logarithm is transcendental over 1/∗ 1/∗ 1/∗ C((x )) to conclude that a ∈ C((x )) ∩ C(x ). Arguing as above, we can show that a ∈ K. Now, let us deal with Case S . Let a ∈ F ∩ C(x). Since F ⊂ Mer(C), 0 0 0 one can ﬁnd x ∈ C such that a ∈ C[[x− x ]]. Once again, using Kronecker- 0 0 Hankel determinants, one can deduce that a ∈ K. 4.3. Iterating the diﬀerence operator. For every positive integer ℓ, let us set ℓ−1 A := ρ (A) × ··· × A ∈ GL (K) . [ℓ] Note that if U is a fundamental matrix of ρ(Y ) = AY , then it is also a fun- damental matrix of ρ (Y ) = A Y . In other words, a vector solution to the [ℓ] system ρ(Y ) = AY is also a solution to the iterated system ρ (Y ) = A Y . [ℓ] In this section, we show that, replacing the original system by a suitable iteration, we can reduce the situation to the more convenient case where the corresponding Galois group is connected and the Picard-Vessiot extension is a ﬁeld. This iteration process might introduce some new constants. The follow- ing lemma shows that this cannot happen when the ﬁelds of ρ-constants is algebraically closed. Lemma 4.8. Let L be a ρ-ﬁeld such that L = k, and r be a positive integer. Then, any ρ -constant of L is algebraic of degree less than or equal to r over k. In particular, if k is algebraically closed then L = k. Proof. Let a ∈ L be a ρ -constant. Then the polynomial P (X) = (X − r−1 ρ a) . . . (X −ρ (a)) belongs to L [X] = k[X]. This proves that a is algebraic over k of degree less than or equal to r. One of the main diﬃculty of diﬀerence algebra is the control of the al- gebraic diﬀerence ﬁeld extensions. This is the source of many technical diﬃculties and connected to the fact that the theory ACFA is unstable (see [CH99]). The diﬀerence ﬁelds (K, ρ) introduced in Section 1.1 have the fol- lowing very strong property with respect to ﬁnite diﬀerence ﬁelds extensions, which will allow us to bypass these diﬃculties. Lemma 4.9. Let r be a positive integer. In each of Cases S , S , Q, and 0 ∞ M, there are no proper ρ -ﬁeld extension of K of ﬁnite degree. Proof. The results in [Hen97, Hen98, Roq18] cover all cases. We are now ready to prove the following result. HYPERTRANSCENDENCE AND LINEAR DIFFERENCE EQUATIONS 19 Proposition 4.10. Let (K, ρ) be deﬁned as in Section 1.1 and let L be a ρ- ρ ⊤ n ﬁeld extension of K such that L = C. Let (f , . . . , f ) ∈ L be a nonzero 1 n solution to (4.1). Let G be the diﬀerence Galois group of (4.1) over K. There exists a positive integer r such that the following properties hold. (a) There exists a Picard-Vessiot ﬁeld extension Q for ρ (Y ) = A Y over [r] r ⊤ (K, ρ ), with a fundamental solution matrix U having (f , . . . , f ) as ﬁrst 1 n column. (b) The diﬀerence Galois group G := Gal(Q/K) of ρ (Y ) = A Y over r [r] (K, ρ ) coincides with the identity component of G and is therefore connected. Though this result can be easily deduced from the proof of [DHR16, Propo- sition 4.6] in the particular case where the parametric operator is the identity, we ﬁnd more convenient for the reader to include the proof below. ρ ρ Proof. (a) Since K(f , . . . , f ) ⊂ L = C, there exists a Picard-Vessiot ring 1 n R for ρ(Y ) = AY over K(f , . . . , f ). Since f , . . . , f ∈ R , there exists a 0 1 n 1 n 0 fundamental matrix U for R having (f , . . . , f ) as ﬁrst column. Let Q 0 1 n 0 be the total quotient ring of R . By Proposition 2.2, we ﬁnd Q = C. Set R := K[U, 1/det(U)]. The diﬀerence ring (R, ρ) is a subring of the Picard-Vessiot ring (R , ρ). Since any zero divisor in R is a zero divisor in R , the total quotient ring Q of R embeds in Q and consequently is a 0 0 pseudoﬁeld (see [Wib10, Lemma 1.3.4]). Consequently C ⊂ Q ⊂ Q = C. Then Q = C, and (Q, ρ) is a Picard-Vessiot extension for ρ(Y ) = AY over (K, ρ). Since K = C is algebraically closed, the uniqueness of Picard-Vessiot extensions above K allows to conclude that Gal(Q/K) = G. We have proved that we can embed any solution in L in a Picard-Vessiot extension above K. By Proposition 2.2, R is a Picard-Vessiot ring for ρ(Y ) = AY over (K, ρ). Let e , . . . , e denote the orthogonal idempotent relative to its ring struc- 0 r−1 ture. Since K[f , . . . , f ] is an integral domain contained in R, which is the 1 n direct sum of the integral domains e R, there exists i ∈ {0, . . . , r − 1} such that K[f , . . . , f ] ⊂ e R. Since K = C is algebraically closed, Lemma 4.8 1 n i ρ r implies that K = C. By [vdPS97, Lemma 1.26], (e R, ρ ) is a Picard- r r Vessiot ring for the ρ -system ρ (Y ) = A Y and e R is an integral domain. [r] i Its total quotient ring Q is a ﬁeld and, since K is algebraically closed, Q 1 1 is a Picard-Vessiot ﬁeld extension by Proposition 2.2. By construction, Q contains f , . . . , f , and we can thus choose a fundamental matrix having 1 n (f , . . . , f ) as ﬁrst column. 1 n (b) By [Roq18, Theorem 12], G = Gal(Q /K) is a normal algebraic sub- r 1 group of G and the quotient G/G is ﬁnite. To conclude, it remains to prove that G is connected. Let G denote its identity component. The Galois correspondence, see Theorem 2.3, gives that the ρ -ﬁeld Q is a ﬁnite ex- tension of K. Lemma 4.9 implies that Q = K and, applying the Galois correspondence again, we deduce that G = G . Remark 4.11. The proof of (b) shows that, if G is connected, then, for every positive integer ℓ, the Galois group of ρ (Y ) = A Y over K coincides with [ℓ] G. In the parametrized framework, we have the following similar result. 20 BORIS ADAMCZEWSKI, THOMAS DREYFUS, AND CHARLOTTE HARDOUIN e e Proposition 4.12. Let L be a (ρ, δ)-ﬁeld extension of K with L = C, and ⊤ n let (f , . . . , f ) ∈ L be a nonzero solution to (4.1). Then there exist a 1 n positive integer r and a parametrized Picard-Vessiot ﬁeld extension Q for r ⊤ ρ (Y ) = A Y over K, with a fundamental matrix U having (f , . . . , f ) 1 n [r] as ﬁrst column. Proof. The proof is similar to Proposition 4.10 and [DHR18, Lemma 3.7], e e e just noting that, by Lemma 4.8, K = C for C is algebraically closed. The following result is obtained by combining the two previous proposi- tions. ⊤ n Corollary 4.13. Let (f , . . . , f ) ∈ F be a nonzero solution to (4.1) and 1 n let G be the diﬀerence Galois group of (4.1). Then, there exists a positive integer r such that (a) There exists a (ρ , δ)-Picard-Vessiot ﬁeld extension Q for r ⊤ ρ (Y ) = A Y with a fundamental matrix U having (f , . . . , f ) as [r] 1 n e e g ﬁrst column over K (where K = C (x) for some positive integer s sh dividing r in Case S ). (b) The diﬀerence Galois group of ρ (Y ) = A Y over K coincides with [r] the identity component of G. Proof. By Proposition 4.10, there exists a positive integer s such that the diﬀerence Galois group of ρ (Y ) = A Y is equal to the identity component [s] of G. In order to apply Proposition 4.12 to the system ρ (Y ) = A Y , we [s] s ρ e e need to embed F into a (ρ , δ)-ﬁeld extension L of K with L = C. We proceed as follows. • Case S . We let F denote the smallest (ρ , δ)-subﬁeld of Mer(C) 0 1 containing F and C . Since Mer(C) = C , we ﬁnd that F = 0 sh sh 1 C . Let C denote a δ-closure of C , considered as a constant ρ - sh sh ﬁeld. The ﬁelds extensions C and F are linearly disjoint above C 1 sh (see [Wib10, Lemma 1.1.6]). Thereby, their compositum L is the fraction ﬁeld of C ⊗ F (see [Bou90, A.V.13]). Arguing as in the C 1 sh proof of [DHR18, Lemma 2.3], we obtain that L has the required properties. • Cases S and Q. We let L be the fraction ﬁeld of F ⊗ C. ∞ 0 C 1/∗ • Case M. We let L be the ﬁeld C((x ))(log). Applying Proposition 4.12 to the system ρ (Y ) = A Y over K, one can [s] sl perform a second iteration to obtain a system ρ (Y ) = A Y satisfying (a). [sl] sl By Remark 4.11, the diﬀerence Galois group of ρ (Y ) = A Y over K is [sl] the identity component of G. This ends the proof. 5. Proof of Theorem 1.2 This section is devoted to the proof of our main result. Before proving Theorem 1.2 in full generality, we ﬁrst consider the following two particular cases. - The function f is solution to an inhomogeneous equation of order one. HYPERTRANSCENDENCE AND LINEAR DIFFERENCE EQUATIONS 21 - The diﬀerence Galois group of the equation associated with f is both connected and irreducible. All along this section, we keep on with the notation of Sections 1.1 and 3. 5.1. Aﬃne order one equations. Various hypertranscendence criteria for solutions to inhomogeneous order one equations have already been obtained, see for instance [Ran92, Ish98, HS08, Ngu12]. We ﬁrst deduce from these criteria the following result. Proposition 5.1. If f ∈ F is solution to an equation of the form ρ(f) = af + b, with a, b ∈ K, then either f is ∂-transcendental over K or f belongs to K. −1 Proof of Proposition 5.1. Let us ﬁrst note that if a = 0, then f = ρ (b) ∈ K. We can thus assume that a 6= 0. Furthermore, we observe that, in Cases 1/ℓ Q and M, we can use a change of variable of the form z = x to reduce the situation to the case where K = C(x). From now on, we thus assume that K = C(x). Let us also assume that f is ∂-algebraic over K. It remains to prove that f ∈ K. We ﬁrst note that Case M corresponds to [Ran92, Théorème 5.2]. Now, let us consider Cases S , S , and Q, which are partially covered by [HS08]. 0 ∞ We recall the following deﬁnition from [HS08, §6.1] (see also [vdPS97, §2]). We say that α ∈ K is standard if, for all positive integers ℓ, α and ρ (α) have no common zero or pole. By [HS08, Lemma 6.2], there exists a standard ∗ ∗ element a and a nonzero e ∈ K such that a = a ρ(e)/e. It follows that f/e is solution to the equation ρ(f/e) = a f/e + b/ρ(e) . Set g = f/e and b = b/ρ(e). Since e ∈ K, it suﬃces to prove that g ∈ K. Let us note that g = f/e is ∂-algebraic over K. In Cases S and S , [HS08, Proposition 3.9] implies that a ∈ C and 0 ∞ there exists h ∈ K such that b = ρ(h) − a h. We obtain that ρ(g) = a g + b ∗ ∗ = a g + ρ(h) − a h . Hence ρ(g − h) = a (g − h). If g = h, then g ∈ K and we are done. Thus, we can assume that g 6= h. In that case, setting f = g − h, we ˜ ˜ ˜ ˜ ˜ ˜ obtain ρ(∂(f)) = a ∂(f). It follows that ρ(∂(f)/f) = ∂(f)/f, and thus ˜ ˜ ˜ ∂(f)/f ∈ C ⊂ K. Since f ∈ F satisﬁes both a linear ρ-equation and a linear ∂-equation with coeﬃcients in K, Theorem 1.3 implies that f ∈ K. Since h ∈ K, we deduce that g ∈ K, as expected. ∗ α ∗ In Case Q, [HS08, Proposition 3.10] implies that a = cx for some c ∈ C and α ∈ Z, and one of the following conditions holds. ∗ r ∗ r (a) a = q for some r ∈ Z, and b = ρ(h) − a h + dx for some h ∈ K and d ∈ C. ∗ Z ∗ (b) a ∈/ q and b = ρ(h) − a h for some h ∈ K. 22 BORIS ADAMCZEWSKI, THOMAS DREYFUS, AND CHARLOTTE HARDOUIN In Case (a), we deduce that g − h g − h −r (5.1) ρ = + dq · r r x x Since (g − h)/x ∈ F , we can consider its expansion in Puiseux series. Since q is not a root of unity, we easily deduce from Equation (5.1) that (g − h)/x ∈ C ⊂ K. Hence g ∈ K, as wanted. ∗ Z ∗ In Case (b), we have that a ∈/ q and b = ρ(h) − a h for some h ∈ K. It follows that ρ(g − h) = cx (g − h), where g − h ∈ F is a Puiseux series. Arguing as in Case (a), we obtain that α = 0 and g − h = c x for some r ∈ Q and c ∈ C. Hence g ∈ K. This ends the proof. For order one equations over K whose solutions do not necessarily belong to F , one has the following criterion for diﬀerential algebraicity. ρ ∗ e e Lemma 5.2. Let L be a (ρ, δ)-ﬁeld extension of K with L = C. Let f ∈ L such that ρ(f) = af with a ∈ K . Then, f is δ-algebraic over K if and only ρ(g) α ∗ ∗ if a = cx for some c ∈ C , α ∈ Q, and g ∈ K . Moreover, α = 0 in Cases S , S , and M. 0 ∞ Proof. Cases S , S , and Q are given by [HS08, Corollary 3.4], while Case M 0 ∞ is given by [DHR18, Proposition 3.1] combined with some descent argument to go from C to C. Such argument is similar to the one given in the proof of [HS08, Corollary 3.2]. 5.2. Connected and irreducible Galois groups. As a second step, we consider the situation where the diﬀerence Galois group is both connected and irreducible. Proposition 5.3. Let us assume that n ≥ 2. If f ∈ F is a nonzero solution to Equation (1.2) and if the corresponding diﬀerence Galois group G over K is both connected and irreducible, then f is ∂-transcendental over K. Proposition 5.3 is obtained as a direct consequence of the following more general result. Proposition 5.4. Let us consider a linear system of the form (4.1) with n ≥ 2. Let us assume that the diﬀerence Galois group G for (4.1) over K is connected and irreducible. Let Q be a (ρ, δ)-Picard-Vessiot ﬁeld extension for (4.1) over K, with fundamental matrix of solutions U. Then every column of U contains at least one element that is δ-transcendental over K. Proof of Proposition 5.3. We argue by contradiction, assuming that f is ∂- n−1 ⊤ algebraic over K. Then all coordinates of the vector (f, . . . , ρ (f)) are also ∂-algebraic over K. This follows from the fact that ρ and ∂ almost commute, that is ∂ρ = cρ∂ with c = 1 in Cases S , S , and Q, and c = p 0 ∞ in Case M. We deduce from the construction of δ with respect to ∂ given in Section 3 (and the fact that log is δ-algebraic for Case M) that all coordinates n−1 ⊤ of the vector (f, . . . , ρ (f)) are also δ-algebraic over K. Now, let A denote the companion matrix associated with Equation (1.2). n−1 ⊤ n Since (f, . . . , ρ (f)) ∈ F is nonzero, Corollary 4.13 ensures the exis- tence of a positive integer r and a (ρ , δ)-Picard-Vessiot ﬁeld extension Q HYPERTRANSCENDENCE AND LINEAR DIFFERENCE EQUATIONS 23 r n−1 ⊤ for ρ (Y ) = A Y over K such that the vector (f, . . . , ρ (f)) is the ﬁrst [r] column of a fundamental matrix U. Furthermore, Remark 4.11 ensures that the Galois group of ρ (Y ) = A (Y ) over K is equal to G, for the latter is [r] connected. Thus, Proposition 5.4 applies with ρ replaced by ρ , providing a contradiction. It remains to prove Proposition 5.4. As a key argument, we will use the following result due to Arreche and Singer [AS17, Lemma 5.1]. It says that the parametrized Galois group must be as big as possible when the diﬀerence Galois group has an identity component that is semisimple. Let us recall that a connected algebraic group is said to be semisimple if it is of positive dimension and has only trivial abelian normal subgroups. Proposition 5.5. Let us consider a linear system ρ(Y ) = AY , where A ∈ 1/ℓ GL (C(x)) in Cases S and S , and where A ∈ GL (C(x )) for some n 0 ∞ n positive integer ℓ in Cases Q and M. In Cases S , S , and M, we let 0 ∞ e e e G ⊂ GL (C) denote the diﬀerence Galois group over K, and H ⊂ GL (C) n n e e denote the (ρ, δ)-Galois group over K. In Case Q, we let G ⊂ GL (C) 1/ℓ e e denote the diﬀerence Galois group over C(x ), and H ⊂ GL (C) denote 1/ℓ the (ρ, δ)-Galois group over C(x ). As in Proposition 2.5, we see H as a subgroup of G. If the identity component of G is semisimple, then H = G. Proof. Cases S and S are explicitly proved in [AS17, Lemma 5.1], since, 0 ∞ by Remark 3.1, the ﬁeld of diﬀerential constants of C is C. Cases Q and M are not explicitly proved in [AS17, Lemma 5.1] for we consider the slightly 1/ℓ more general situation where A ∈ GL (C(x )). However, the result follows easily from the argument given in the proof of [AS17, Lemma 5.1]. We are now ready to prove Proposition 5.4. Proof of Proposition 5.4. We let H denote the (ρ, δ)-Galois group of (4.1) over K, that is e e e H = Gal (Q/K) ⊂ GL (C) . e e We also let G denote the Galois group of ρ(Y ) = AY over K. Let us argue by contradiction, assuming the existence of one column of U whose coordinates are all δ-algebraic over K. We ﬁrst show that all entries of U are δ-algebraic over K. Since G is ir- reducible, Lemma 4.6 implies that G is irreducible too. By Proposition 2.5, H is Zariski dense in the irreducible group G. This implies that H is irre- ducible, since otherwise, H would be conjugated to a group formed by block upper triangular matrices, as well as its Zariski closure, contradicting the e e irreducibility of G. Now, let S denote the C-vector space generated by all columns of U whose coordinates are all δ-algebraic over K. By assumption, S is not reduced to {0}. Furthermore, S is invariant under the action of any σ ∈ H. Since H is irreducible, we deduce that S contains n linearly independent solutions to (4.1). In other words, all columns of U belong to S, which implies that all entries of U are δ-algebraic over K, as claimed. We observe that the determinant det(U) is solution to the equation (5.2) ρ(y) = det(A)y 24 BORIS ADAMCZEWSKI, THOMAS DREYFUS, AND CHARLOTTE HARDOUIN and that the diﬀerence Galois group of this equation over K is the group det(G). Since all entries of U are δ-algebraic over K, we obtain that det(U) is also δ-algebraic over K. We ﬁrst consider the particular case where det(U) ∈ K, and then we move to the general case. Let us ﬁrst assume that det(U) belongs to K. In that case, det(G) = {1} and therefore G ⊂ SL (C). We recall that G is assumed to be connected and irreducible. According to Lemma 4.2, we have that G is primitive. By [SU93, Proposition 2.3], we ﬁnally obtain that G is semisimple. By Lemma 4.5, e e G = G(C) is semisimple too. In Cases S , S , and M, we infer from Proposition 5.5 that H = G. 0 ∞ Since all entries of U are δ-algebraic over K, the δ-dimension of the (ρ, δ)- Picard-Vessiot extension is zero , and therefore the δ-dimension of H is zero by [HS08, Proposition 6.26]. By a result of Kolchin [Kol73, Chap. IV, Proposition 10], the dimension of an algebraic group G over C is the same as the δ-dimension of G viewed as a diﬀerential group over C. This proves that the dimension of H equals the dimension of G(C) . Since C is algebraically e e closed, the dimension of G(C) over C equals the dimension of G over C. e e Thus, the above equality H = G = G(C) implies that the dimension of the algebraic group G over C is zero. We recall that an algebraic group of dimension zero is just a ﬁnite group, and that a ﬁnite connected group has cardinality one. Thus, we deduce that G = {I }, where we let I denote n n the identity matrix of size n. Since by assumption n ≥ 2, this provides a contradiction with the assumption that G is irreducible. This ends the proof in these cases. 1/ℓ In Case Q, the system ρ(Y ) = AY has coeﬃcients in C(x ) and the 1/ℓ entries of U are δ-algebraic over C(x ), for some positive integer ℓ. Let 1/ℓ G denote the diﬀerence Galois group of ρ(Y ) = AY over C(x ). Since K,ℓ 1/ℓ e e K is an algebraic extension of C(x ), we deduce from [Roq18, Theorem 7] that G and G have the same identity component, which is G since e e e K K,ℓ K the latter is connected. Hence the identity component of the identity of G is semisimple. By Proposition 5.5, it follows that H = G , where e ℓ e K,ℓ K,ℓ 1/ℓ H denote the (ρ, δ)-Galois group over C(x ). Furthermore, the entries of 1/ℓ U are δ-algebraic over C(x ) which implies that the δ-dimension of H is zero. Since H = G , we get as previously that the algebraic group G e e K,ℓ K,ℓ has dimension zero. Since G ⊂ G , this proves that the dimension of G e e e K K,ℓ K is also zero, and we can argue as previously to get a contradiction. Now, let us consider the general case. We remind that det(U) is δ-algebraic over K and ρ(det(U)) = det(A)det(U). By Lemma 5.2, there exist some rational number α and nonzero elements c ∈ C and g ∈ K such that det(A) = cx ρ(g)/g. Furthermore, α = 0 in Cases S , S , and M. 0 ∞ Let us consider the rank one linear diﬀerence system −1/n −α/n (5.3) ρ(y) = c x y We refer the reader to [HS08, P. 374] for a deﬁnition of the notion of δ-dimension of a (ρ, δ)-Picard-Vessiot ring and of a (ρ, δ)-Galois group. HYPERTRANSCENDENCE AND LINEAR DIFFERENCE EQUATIONS 25 e e e Since Q is a (ρ, δ)-ﬁeld with Q = C, there exists a (ρ, δ)-Picard-Vessiot e e e extension Q for (5.3) over Q. Let (λ) ∈ GL (Q ) be a fundamental matrix 1 1 1 associated with this system. Then −1/n −α/n ρ(λ) = c x λ and λ is invertible in Q . Using the commutativity of δ and ρ, we obtain that δ(λ) e e δ ∈ Q = C , which shows that λ is δ-algebraic over C. In particular, it is δ-algebraic over e e e K. Thus, all entries of the matrix λU ∈ GL (Q ) are δ-algebraic over K. n 1 On the other hand, we set −1/n −α/n B = c x A ∈ GL (K) and U := λU. n B Note that ρ(U ) = BU . Let G be the diﬀerence Galois group of the B B B system ρ(Y ) = BY over K. Our choice of B ensures that det(B) = ρ(g)/g. Thus, the equation ρ(y) = det(B)y has a solution in K. It follows that its diﬀerence Galois group det(G ) is reduced to {1}. Hence, G ⊂ SL (C). For k ≥ 1, we let G denote the B n B [k] diﬀerence Galois group of ρ (Y ) = B Y over K. By Proposition 4.10, there [k] exists r ≥ 1 such that G is connected. Furthermore, ρ (g)/g = det(B ) B [r] [r] and the diﬀerence Galois group of ρ (Y ) = B Y over K remains included [r] in SL (C). Let us notice that A = α B [r] [r] for some α ∈ K . We claim that G is irreducible. Let us argue by contradiction, assuming [r] that G is reducible. By Lemma 4.4, there exists T ∈ GL (K) such that B n [r] r −1 ρ (T )B T is a block upper triangular matrix, and we deduce that [r] r −1 r −1 α ρ (T )B T = ρ (T )A T . r [r] [r] By Lemma 4.4, we obtain that G , the diﬀerence Galois group of the system ρ (Y ) = A Y over K, is conjugated to a group of block upper triangular [r] matrices, which implies that G is reducible. By Remark 4.11, G = G, r r providing a contradiction with the assumption that G is irreducible. This proves that G is irreducible. [r] Finally by Lemma 4.2, the irreducible connected group G is primitive. [r] Again, since G ⊂ SL (C), we infer from [SU93, Proposition 2.3] that B n [r] e e G is semisimple. Since Q s a pseudo δ-ﬁeld that contains K and all B 1 [r] e e coordinates of U , we have KhU i ⊂ Q (see Section 2.2.2 for the notation). B B 1 e e e Furthermore, since Q = C, the pseudo δ-ﬁeld KhU i is a (ρ , δ)-Picard- Vessiot extension for ρ (Y ) = B Y over K. [r] Since all entries of U are δ-algebraic over K, we can apply Proposition 5.5 and, arguing as in the ﬁrst part, we deduce that G = {I }. Since n ≥ 2, B n [r] we obtain a contradiction with the fact that G is irreducible. [r] 26 BORIS ADAMCZEWSKI, THOMAS DREYFUS, AND CHARLOTTE HARDOUIN 5.3. The general case. We are now ready to prove Theorem 1.2. Proof of Theorem 1.2. We argue by induction on n. More precisely, our induction assumption reads as follows. (H ) For all positive integers k and all f ∈ F that is solution to a ρ -linear n 0 equation of order at most n with coeﬃcients in K, we have either f is ∂-transcendental over K or f ∈ K. Proposition 5.1 implies that (H ) hold true. Let n ≥ 2 and let us assume (H ). Let f ∈ F be solution to a ρ -linear equation of order n with n−1 0 coeﬃcients in K. Without any loss of generality, we can assume that f 6= 0 and k = 1. Considering the companion matrix associated with this equation, Corollary 4.13 ensures the existence of a positive integer r such that the following properties hold. n−1 ⊤ r (a) The vector (f, . . . , ρ (f)) is solution to the system ρ (Y ) = AY for some A ∈ GL (K). r r (b) There exists a (ρ , δ)-Picard-Vessiot ﬁeld extension Q for ρ (Y ) = n−1 ⊤ AY over K, such that the vector (f, . . . , ρ (f)) is the ﬁrst column of a fundamental matrix U ∈ GL (Q). (c) The diﬀerence Galois group G of the system ρ (Y ) = AY over K is connected. If G is irreducible, Proposition 5.3 shows that f is ∂-transcendental. Hence (H ) holds. From now on, we assume that G is reducible. Furthermore, we assume that f is ∂-algebraic over K. Thus, it remains to prove that f ∈ K. Without loss of generalities, we can assume that r = 1. By Lemma 4.4, there exists a gauge transformation T = (t ) ∈ GL (K) i,j n such that A A −1 1 1,2 ρ(T )AT = , 0 A where A ∈ GL (K), n + n = n, and n < n. Furthermore, let us assume i n 1 2 2 that n is minimal with respect to this property. Set ⊤ i−1 ⊤ n (5.4) (g ) := T (ρ (f)) ∈ F . i≤n i≤n 0 ⊤ n The vector (g ) ∈ F is solution to the system ρ(Y ) = A Y . Fur- i 2 n +1≤i≤n 0 thermore, since f is ∂-algebraic over K, the g are also ∂-algebraic over K. By (H ) and Remark 2.1, we obtain that (5.5) (g ) ∈ K . n +1≤i≤n Let G denote the diﬀerence Galois group of the system ρ(Y ) = A Y over K. 1 1 We claim that G is connected and irreducible. Indeed, if G were reducible, 1 1 then by Lemma 4.4, there would exist a gauge transformation changing A into a block upper triangular matrix, contradicting the minimality of n . Furthermore, the Galois group of ρ(Y ) = A Y over K is a quotient of the connected group G and thereby a connected group. The main step of the proof consists in showing the following result. Claim. One has n = 1. 1 HYPERTRANSCENDENCE AND LINEAR DIFFERENCE EQUATIONS 27 Proof of the claim. By assumption, f is ∂-algebraic over K. Arguing as in the proof of Proposition 5.3, we get that all coordinates of the vector n−1 (f, . . . , ρ (f)) are δ-algebraic over K. Then, (5.4) implies that all the g ’s are also δ-algebraic over K. We have to distinguish two cases. Let us ﬁrst assume that all the g ’s belong to K. By (5.4) and Lemma 4.7, we obtain that ∀i ∈ {1, . . . , n}, g ∈ F ∩ K = K i 0 n−1 ⊤ −1 ⊤ n and hence (f, . . . , ρ (f)) = T (g , . . . , g ) ∈ K . Thus the vector 1 n n−1 ⊤ (f, . . . , ρ (f)) is ﬁxed by the diﬀerence Galois group G. By Lemma 4.4, there exists P ∈ GL (K) such that b B 1 1,2 −1 ρ(P )AP = , 0 B for some matrices B and B with coeﬃcients in K and a nonzero b ∈ K. 1,2 2 1 By minimality of n , we obtain that n = 1, as wanted. 1 1 Now, let us assume that at least one the g ’s does not belong to K. Note that, by assumption, the f ’s all belong to Q and thus the g ’s all belong to i i e e Q too. We let H denote the (ρ, δ)-Galois group of ρ(Y ) = AY over K. By the parametrized Galois correspondence [HS08, Theorem 6.20], we deduce the existence of some σ ∈ H, such that ⊤ ⊤ (σ(g )) 6= (g ) , i i i≤n i≤n while (5.5) implies that σ(g ) = g for every i, n + 1 ≤ i ≤ n. Set i i 1 ⊤ ⊤ ⊤ u := (g ) , u := (g ) and v := (σ(g )) . 1 i 2 i 1 i i≤n n +1≤i≤n i≤n 1 1 1 Hence w := u − v is a nonzero vector. Since the coordinates of u are δ- 1 1 1 algebraic over K and σ belongs to the (ρ, δ)-Galois group H, the coordinates of v are also δ-algebraic over K. Hence the coordinates of w are δ-algebraic over K. Furthermore, u and v are both solution to the system 1 1 ρ(Y ) = A Y + A u . 1 1,2 2 It follows that ρ(w) = A w . Since we have ρ ρ ρ e e e e e C = K ⊂ K(w) ⊂ Q = C , Corollary 4.13 ensures the existence of a positive integer s and a (ρ , δ)- e e Picard-Vessiot ﬁeld extension Q for the system ρ (Y ) = (A ) Y over K 1 1 [s] such that w is the ﬁrst column of a fundamental matrix. Furthermore, the diﬀerence Galois group of ρ (Y ) = (A ) Y is equal to G for the latter is 1 1 [s] connected. The coordinates of w being δ-algebraic over K and G being connected and irreducible, Proposition 5.4 implies that n = 1. Now, let us prove that (g , . . . , g ) ∈ K . Set 1 n u := (g ) . 2 i 2≤i≤n n−1 n−1 By (5.5), we have u ∈ K and thus A u ∈ K . Furthermore, g ∈ F 2 1,2 2 1 0 is solution to the inhomogeneous order one equation ρ(Y ) = A Y + A u . 1 1,2 2 28 BORIS ADAMCZEWSKI, THOMAS DREYFUS, AND CHARLOTTE HARDOUIN Since g is ∂-algebraic over K, Proposition 5.1 implies that g ∈ K, and 1 1 hence (g , . . . , g ) ∈ K . By (5.4), we obtain that 1 n n−1 ⊤ −1 ⊤ (f, . . . , ρ (f)) = T (g , . . . , g ) . 1 n Since the coeﬃcients of T belong to K, it follows that f ∈ K, as wanted. References [ADR18] C. E. Arreche, T. Dreyfus, and J. Roques. On the diﬀerential transcendence of the elliptic hypergeometric functions. arXiv preprint arXiv:1809.05416, 2018. [AF17] B. Adamczewski and C. Faverjon. Méthode de Mahler: relations linéaires, tran- scendance et applications aux nombres automatiques. Proc. Lond. Math. Soc., 115:55–90, 2017. [AF18a] B. Adamczewski and C. Faverjon. Mahler’s method in several variables I: The theory of regular singular systems. arXiv preprint arXiv:1809.04823, 2018. [AF18b] B. Adamczewski and C. Faverjon. Mahler’s method in several variables II: Applications to base change problems and ﬁnite automata. arXiv preprint arXiv:1809.04826, 2018. [And01] Y. André. Diﬀérentielles non commutatives et théorie de Galois diﬀérentielle ou aux diﬀérences. Ann. Sci. École Norm. Sup., 34:685–739, 2001. [AS17] C. E. Arreche and M. F. Singer. Galois groups for integrable and projectively integrable linear diﬀerence equations. J. Algebra, 480:423–449, 2017. [Béz94] J.-P. Bézivin. Sur une classe d’equations fonctionnelles non linéaires. Funkcial. Ekvac., 37:263–271, 1994. [BG93] J.-P. Bézivin and F. Gramain. Solutions entières d’un système d’équations aux diﬀérences. Ann. Inst. Fourier, 43:791–814, 1993. [BM06] M. Bousquet-Mélou. Rational and algebraic series in combinatorial enumer- ation. In International Congress of Mathematicians. Vol. III, pages 789–826. Eur. Math. Soc., Zürich, 2006. [Bou90] N. Bourbaki. Algebra. II. Chapters 4–7. Elements of Mathematics. Springer- Verlag, Berlin, 1990. [Cas72] P. J. Cassidy. Diﬀerential algebraic groups. Amer. J. Math., 94:891–954, 1972. [Cas89] P. J. Cassidy. The classiﬁcation of the semisimple diﬀerential algebraic groups and the linear semisimple diﬀerential algebraic Lie algebras. J. Algebra, 121:169–238, 1989. [CH99] Z. Chatzidakis and E. Hrushovski. Model theory of diﬀerence ﬁelds. Trans. Amer. Math. Soc., 351:2997–3071, 1999. [CHS08] Z. Chatzidakis, C. Hardouin, and M. F Singer. On the deﬁnitions of diﬀerence Galois groups. In Model Theory with Applications to Algebra and Analysis. Vol. 1, volume 349 of London Math. Soc. Lecture Note Ser., pages 73–109. Cambridge Univ. Press, Cambridge, 2008. [Coh65] R. M. Cohn. Diﬀerence Algebra. Interscience Publishers John Wiley & Sons, New York-London-Sydeny, 1965. [CS07] P. J. Cassidy and M. F. Singer. Galois theory of parameterized diﬀerential equations and linear diﬀerential algebraic groups. In Diﬀerential Equations and Quantum Groups, volume 9 of IRMA Lect. Math. Theor. Phys., pages 113–155. Eur. Math. Soc., Zürich, 2007. [DHR16] T. Dreyfus, C. Hardouin, and J. Roques. Functional relations of solutions of q-diﬀerence equations. arXiv preprint arXiv:1603.06771, 2016. [DHR18] T. Dreyfus, C. Hardouin, and J. Roques. Hypertranscendance of solutions of mahler equations. J. Eur. Math. Soc. (JEMS), 20:2209–2238, 2018. [DHRS18] T. Dreyfus, C. Hardouin, J. Roques, and M. F. Singer. On the nature of the generating series of walks in the quarter plane. Invent. Math., 213:139–203, [DV12] L. Di Vizio. Approche galoisienne de la transcendance diﬀérentielle. In Tran- scendance et Irrationalité, volume 2012 of SMF Journ. Annu., pages 1–20. Soc. Math. France, Paris, 2012. HYPERTRANSCENDENCE AND LINEAR DIFFERENCE EQUATIONS 29 [Har08] C. Hardouin. Hypertranscendance des systèmes aux diﬀérences diagonaux. Compos. Math., 144:565–581, 2008. [Har16] C. Hardouin. Galoisian approach to diﬀerential transcendence. In Galois The- ories of Linear Diﬀerence Equations: an Introduction, volume 211 of Math. Surveys Monogr., pages 43–102. Amer. Math. Soc., Providence, RI, 2016. [Hen97] P. A. Hendriks. An algorithm for computing a standard form for second-order linear q-diﬀerence equations. J. Pure Appl. Algebra, 117:331–352, 1997. [Hen98] P. A. Hendriks. An algorithm determining the diﬀerence galois group of second order linear diﬀerence equations. J. Symbolic Comput., 26:445–462, 1998. [Hil02] D. Hilbert. Mathematical problems. Bull. Amer. Math. Soc., 8:437–479, 1902. [Höl87] O. Hölder. über die eigenschaft der gammafunction keiner algebraischen diﬀer- entialgleichung zu genügen. Math. Ann., 28:1–13, 1887. [HS99] P. A. Hendriks and M. F. Singer. Solving diﬀerence equations in ﬁnite terms. J. Symbolic Comput., 27:239–259, 1999. [HS08] C. Hardouin and M. F. Singer. Diﬀerential galois theory of linear diﬀerence equations. Math. Ann., 342:333–377, 2008. [Ish98] K. Ishizaki. Hypertranscendency of meromorphic solutions of a linear functional equations. Aequationes Math., 56:271–283, 1998. [Kol73] E. R. Kolchin. Diﬀerential Algebra and Algebraic Groups. Academic Press, New York-London, 1973. Pure and Applied Mathematics, Vol. 54. [KZ01] M. Kontsevich and D. Zagier. Periods. In Mathematics Unlimited—2001 and Beyond, pages 771–808. Springer, Berlin, 2001. [Mah29] K. Mahler. Arithmetische Eigenschaften der Lösungen einer Klasse von Funk- tionalgleichungen. Math. Ann., 101:342–366, 1929. [Mah30a] K. Mahler. Arithmetische Eigenschaften einer Klasse transzendental- transzendenter Funktionen. Math. Z., 32:545–585, 1930. [Mah30b] K. Mahler. Uber das Verschwinden von Potenzreihen mehrerer Veränderlichen in speziellen Punktfolgen. Math. Ann., 103:573–587, 1930. [Mil17] J. S. Milne. Algebraic Groups. The Theory of Group Schemes of Finite Type over a Field, volume 170. Cambridge University Press, 2017. [Moo96] E. H. Moore. Concerning transcendentally transcendental functions. Math. Ann., 48:49–74, 1896. [Ngu12] P. Nguyen. Équations de Mahler et Hypertranscendance. Thèse de l’Institut de Mathématiques de Jussieu, 2012. [Nis84] Ke. Nishioka. A note on diﬀerentially algebraic solutions of ﬁrst order linear diﬀerence equations. Aequationes Math., 27:32–48, 1984. [Nis90] Ku. Nishioka. New approach in Mahler’s method. J. Reine Angew. Math., 407:202–219, 1990. [OW15] A. Ovchinnikov and M. Wibmer. σ-Galois theory of linear diﬀerence equations. Int. Math. Res. Not. IMRN, 12:3962–4018, 2015. [Phi15] P. Philippon. Groupes de Galois et nombres automatiques. J. Lond. Math. Soc., 92:596–614, 2015. [Ram92] J.-P. Ramis. About the growth of entire functions solutions of linear algebraic q-diﬀerence equations. Ann. Fac. Sci. Toulouse Math., 1:53–94, 1992. [Ran92] B. Randé. Équations Fonctionnelles de Mahler et Applications aux Suites p- Régulières. Thèse de l’Université Bordeaux 1, 1992. [Roq18] J. Roques. On the algebraic relations between mahler functions. Trans. Amer. Math. Soc., 370:321–355, 2018. [Rub89] L. A. Rubel. A survey of transcendentally transcendental functions. Amer. Math. Monthly, 96(9):777–788, 1989. [Sal63] R. Salem. Algebraic Numbers and Fourier Analysis. D. C. Heath and Co., Boston, Mass., 1963. [SS19] R. Schäfke and M. Singer. Consistent systems of linear diﬀerential and diﬀer- ence equations. J. Eur. Math. Soc. (JEMS), 21:2751–2792, 2019. [SU93] M. F. Singer and F. Ulmer. Galois groups of second and third order linear diﬀerential equations. J. Symbolic Comput., 16:9–36, 1993. 30 BORIS ADAMCZEWSKI, THOMAS DREYFUS, AND CHARLOTTE HARDOUIN [vdPS97] M. van der Put and M. F. Singer. Galois Theory of Diﬀerence Equations, volume 1666 of Lecture Notes in Mathematics. Springer-Verlag, Berlin, 1997. [Wib10] M. Wibmer. Geometric Diﬀerence Galois Theory. PhD thesis, Heidelberg Uni- versity, 2010. [Wib12] M. Wibmer. Existence of ∂-parameterized Picard-Vessiot extensions over ﬁelds with algebraically closed constants. J. Algebra, 361:163–171, 2012. Univ Lyon, Université Claude Bernard Lyon 1, CNRS UMR 5208, Institut Camille Jordan, 43 blvd. du 11 novembre 1918, F-69622 Villeurbanne cedex, France E-mail address: boris.adamczewski@math.cnrs.fr Institut de Recherche Mathématique Avancée, U.M.R. 7501 Université de Strasbourg et C.N.R.S. 7, rue René Descartes 67084 Strasbourg, France E-mail address: dreyfus@math.unistra.fr Institut de Mathématiques de Toulouse, Université Paul Sabatier, 118, route de Narbonne, 31062 Toulouse, France E-mail address: hardouin@math.univ-toulouse.fr

http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png

Mathematics arXiv (Cornell University)

http://www.deepdyve.com/lp/arxiv-cornell-university/hypertranscendence-and-linear-difference-equations-rFwnL8h0MT

Hypertranscendence and linear difference equations

Loading next page...

ISSN: 0894-0347
eISSN: ARCH-3343
DOI: 10.1090/jams/960
Publisher site: See Article on Publisher Site

Abstract

Journal

Mathematics – arXiv (Cornell University)

Published: Oct 4, 2019

Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Hypertranscendence and linear difference equations

Hypertranscendence and linear difference equations

Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Hypertranscendence and linear difference equations

Hypertranscendence and linear difference equations

Abstract

Journal

Recommended Articles

There are no references for this article.

Our policy towards the use of cookies