Divided Differences, Square Functions and a Law of the Iterated Logarithm

Abstract

We study differentiability properties of functions defined in the euclidean space in terms of a conical square function which is analogue to the classical square function introduced by Stein and Zygmund in the sixties. Pointwise differentiability can be characterized, modulo sets of Lebesgue measure zero, in terms of the finiteness of the conical square function. At the complement of this set, a Law of the Iterated Logarithm describes the maximal growth of the divided differences in terms of a truncated version of the conical square function. Moreover a characterization of Sobolev spaces in terms of the conical square function is given.