If Φ:[0,)R\Phi:[0,\infty )\rightarrow \mathbb{R} is convex and continuous with Φ(0)=0\Phi (0)=0 and if q(1,)q\in (1,\infty ),\break q:=qq1q^{\prime }:=\frac{q}{q-1}, we first prove that the inequality Φ(0f(r)dr)C0f(r)Φ(r1/q)dr\Phi \left( \int_{0}^{\infty }f(r)dr\right) \leq C\int_{0}^{\infty }f(r)\Phi ^{\prime }(r^{1/q^{\prime }})dr for every fLq(0,)f\in L^{q}(0,\infty), f0f\geq 0 with fq1||f||_{q}\leq 1 holds when C=1C=1. In general, both sides may be ±\pm \infty. Related inequalities for fL1(RN)Lq(RN)f\in L^{1}(\mathbb{R}^{N})\cap L^{q}(\mathbb{R}^{N}), f0f\neq 0 are derived. This inequality is independent of Jensen's inequality and, when q=q=\infty, it is an elaboration on an inequality of Steffensen which was discussed elsewhere by the author.\par The next goal of the paper is to identify the range of the admissible constants CC and, in particular, to characterize the optimal constant when Φ0\Phi \geq 0 or Φ0\Phi \leq 0. It turns out that C=1C=1 is ``almost always'' optimal, at least in a restricted sense, but not always when q<q<\infty: Given qq, the admissible constants lie on an interval containing 11 whose left (right) endpoint is the supremum (infimum) of a function defined on some (left/right dependent) subset of R2\mathbb{R}^{2}.\par If q=2q=2, these extrema can be calculated in a number of examples. Among other things, this reveals that C=1C=1 need not be optimal when Φ0\Phi \geq 0 and Φ+(0)=0\Phi _{+}^{\prime }(0)=0 or when Φ0\Phi \leq 0 and Φ+(0)=\Phi _{+}^{\prime} (0)=-\infty.

Contact details are reproduced from the original publication and may be historical.

Patrick J. Rabier

Dept. of Mathematics, University of Pittsburgh, Pittsburgh, PA 15260, U.S.A.

rabier@imap.pitt.edu

P. J. Rabier. “Generalized Steffensen Inequalities and Their Optimal Constants.” Journal of Convex Analysis 19 (2012), No. 2, 301–321.