Bernstein's analyticity theorem for binary differences (Q1818924)

All functions are assumed to be real valued functions of one real variable. The differences \(\Delta^n_hf(x)\) are defined inductively by \(\Delta^1_h f(x)= f(x+ h)- f(x)\) and \(\Delta^n_h f(x)= \Delta^1_h(\Delta^{n- 1}_h f)(x)\) for \(n\geq 2\). The binary differences \(B^n_ hf\) are defined by \(B^1_h f(x)= \Delta^1_h f(x)\) and \(B^{n+ 1}_h f(x)= B^n_{2h} f(x)- 2^n\cdot B^n_h f(x)\) for \(n\geq 2\). The function \(f\) is called absolutely monotone on the real interval \(I\) if \(\Delta^n_h f(x)\geq 0\) for all \(x,x+nh\in I\), \(h>0\) and \(n= 1,2,\dots\); \(f\) is binary absolutely monotone on \(I\) if \(B^n_h f(x)\geq 0\) for all \(x,x+ 2^{n-1}h\in I\), \(h>0\) and \(n= 1,2,\dots\)\ . Theorem 1. Let \(f\) be defined on \([0,1]\). Then the following statements (a)--(c) are equivalent: (a) \(f\) is the difference between two binary absolutely monotone functions on \([0,1)\); (b) \(f\) is the difference between two absolutely monotone functions on \([0,1)\); (c) \(f\) has a power series representation \(f(x)= \sum^\infty_{n=0} c_n\cdot x^n\), \(0\leq x<1\), for some sequence \(\{c_n\}^\infty_0\) of real numbers. Theorem 2. A function \(f\) is absolutely monotone on \((-c,0)\), \(0< c\leq\infty\), if and only if \((-1)^n\cdot B^n_h\check f(x)\geq 0\) on \((0,c)\), where \(\check f(x)= f(-x)\). Bernstein's results related to \(\Delta^n_h f\) are in this way extended to the differences \(B^n_h f\) introduced by \textit{J. Marcinkiewicz} and \textit{A. Zygmund} [Fundam. Math. 26, 1-43 (1936; Zbl 0014.11102)] in the study of smoothness of functions and summability of Fourier series.