New integral representations of \(n\)th order convex functions (Q531880)

The starting point of the paper is the definition of \(n\)th-order convexity of a real function given by Popoviciu in 1934 in terms of divided differences, that can be equivalently stated as follows: a function \(f\) is \(n\)-convex on \((a,b)\) (\(n\geq 1\)) if and only if its derivative \(f^{(n-1)}\) exists and is convex on \((a,b)\) (\(f^{(0)}(x)=f(x)\)), or its right derivative \(f^{(n)}_R\) (or left derivative \(f^{(n)}_L\)) exists and is non-decreasing on \((a,b).\) With each such \(f\) one can associate a measure \(\mu\) defined on \((a,b)\) by \(\mu([x,y])=f^{(n)}_R(y)-f^{(n)}_L(x),\) for \(a<x\leq y<b;\) if \(f^{(n)}_R(a)\) is finite, \(\mu\) can be extended to a bounded (finite) measure on \([a,c]\), for all \(c<b.\) In this case, \(f\) can be represented as follows: \[ f(x)=\sum_{k=0}^n{f^{(k)}_R(a)(x-a)^k\over k!}+{1\over n!}\int_a^b(x-t)^n_+\,d\mu(t),\qquad \forall x\in (a,b). \] This representation is valid for all \(x\in (a,b)\) if \(\mu\) is of bounded variation on \((a,b),\) otherwise this representation holds only on closed subintervals of \((a,b)\). The aim of this paper is to give a similar integral representation in the general case. In Section 2 the author provides an integral representation of an \(n\)-convex function \(f\) without additional assumptions on \(f\). Explicit formulas for \(n\)-spectral measures corresponding to an \(n\)-convex function in this representation are given. In Section 3 the author proves a characterization of an \(n\)-convex function via its decomposition as the sum of two \((n+1)\)-times monotone function and a polynomial of degree at most \(n\) (Theorem 3.2). This is used to obtain the decomposition of \(n\)-Wright-convex functions. In Section 4 the concept of relative \(n\)-convexity is introduced, extending the concept of relative convexity (\(n=1\)) studied by Karlin and Studden in 1966. Relative \(n\)-convexity induces a partial ordering in the set of \(n\)-convex functions. A measure of \(n\)-convexity of an \(n\)-convex function is characterized in terms of the \(n\)-spectral measures (see Theorem 4.3), as well as in terms of \(n\)th-order distributional derivatives and Radon-Nikodym derivatives. Furthermore, the author defines, studies and gives a characterization of strong \(n\)-convexity of an \(n\)-convex function in terms of its derivative \(f^{(n+1)}\) without additional assumptions on the differentiability of \(f.\) In Theorem 4.5 it is shown that an \(n\)-convex function \(f\) is strongly \(n\)-convex with modulus \(c\) if and only if \(f^{(n+1)}(x)\geq c\) a.e. in \((a,b).\) In Section 5 it is proved that, for any two \(n\)-convex functions \(f\) and \(g\) such that \(f\) is \(n\)-convex w.r.t. \(g,\) the function \(g\) is the support for the function \(f\) according to the notion introduced by Wąsowicz in 2007, up to a polynomial of degree at most \(n\).