Log-convex solutions of the second order to the functional equation \(f(x+1)=g(x)f(x)\) (Q878509)

It is quite plausible that some solutions of the functional equation \[ f(x+ 1)= g(x)\,f(x)\tag{1} \] are related to the solutions of the linear non-autonomous recurrence \[ y_{n+1}= g(n)\,y_n,\tag{2} \] interpreted as a point-map and/or as a difference equation. Studying limits of infinite products Euler considered the case \(g(n)=(n+ 1)\) and guessed that if \(n\) is restricted to positive integers, the solution of (2) is \[ y_n= \prod^{n-1}_{m=0} (n- m)\triangleq n!\;(\text{Fakultät, factorial}).\tag{3} \] It was also plausible that there is no unique extension of (3) if \(n\) is allowed to take real values. For analytical consistency Euler choosed the integral `imbedding' \[ n!= \int^\infty_0 e^{-z}\cdot z^n\cdot dz.\tag{4} \] In 1814 Legendre introduced the notation \(n!=\Gamma(n+1)\) [for a more detailed context see \textit{N. Nielsen}, Handbuch der Theorie der Gammafunktion, Teubner, Leipzig, 326 p. (1906; JFM 37.0450.01)]. Using the same approach as Euler (limits of infinite products), but restricting severely the apparently free choice of \(g(x)\), the authors establish an existence and uniqueness theorem for a real-valued solution of (1) without constructing a usable explicit solution. It is thus not yet possible to determine whether and how this solution approaches (4) as \(g(n)\) is made to approach \((n+ 1)\). The study of (1) serves to classify log-convex functions according to their order. No other, practical significant application is claimed.