Liouvillian solutions of linear difference-differential equations (Q846351)

A most important theorem in differential Galois theory states that a linear differential equation over \(\mathbb C(x)\) can be solved by ``elementary means'' (that is, roughly said,successive extensions by integrals, exponentiation and algebraic functions) if and only if its Galois group (which is an algebraic group over the complex numbers) has a solvable identity component. In that case, it is proved that the differential equation admits a solution of the form \(e^{\int f}\), where \(f\) is an algebraic function. This, in turn, is the source of algorithms that can decide for the existence of such solutions. A similar theory for difference equations has been developed by van der Put and Singer (about the Galois group), then by Hendricks and Singer (about the algorithms). In both cases, the results extend to the case of linear systems \(dY(x)/dx = A(x) Y(x)\) or \(Y(x+1) = A(x) Y(x)\). In the present paper, the authors build the corresponding theory for difference-differential equations. These involve vector valued functions of two variables \(Y(t,x)\) satisfying two functional (matricial) equations: \(dY(t,x)/dx = A(t,x) Y(t,x)\) and \(Y(t+1,x) = B(t,x) Y(t,x)\) (maybe subject to some compatibility condition). In terms of the variable \(t\), the authors rather mean this as a discrete difference system: prominent examples are then sequences \(Y(n,x)\) of orthogonal polynomials or other special functions (e.g. Bessel). The authors then define a notion of Liouvillian solution and give a Galois theoretic characterization of equations admitting such solutions. In a subsequent paper [J. Symb. Comput. 45, No. 3, 306--323 (2010; Zbl 1233.12005)], the authors explore the algorithmic side of the theory.