Popoviciu's inequality for functions of several variables (Q847791)

The authors introduce the concept of a \(2D\)-convex function as follows: Let \(U\) be a convex subset of a real linear space \(E\). A function \(f\,:\,U\rightarrow \mathbb{R}\) is called \(2D\)-convex if it satisfies the inequality (1) \[ \frac{\lambda_{1}f(x_{1})+\lambda_{2}f(x_{2})+\lambda_{3}f(x_{3})}{\lambda_{1}+\lambda_{2}+\lambda_{3}}+ f\Big(\frac{\lambda_{1}x_{1}+\lambda_{2}x_{2}+\lambda_{3}x_{3}}{\lambda_{1}+\lambda_{2}+\lambda_{3}}\Big) \] \[ \begin{multlined} \geq\frac{1}{\lambda_{1}+\lambda_{2}+\lambda_{3}}\Big[(\lambda_{1}+\lambda_{2}) f\Big(\frac{\lambda_{1}x_{1}+\lambda_{2}x_{2}}{\lambda_{1}+\lambda_{2}}\Big) +(\lambda_{2}+\lambda_{3}) f\Big(\frac{\lambda_{2}x_{2}+\lambda_{3}x_{3}}{\lambda_{2}+\lambda_{3}}\Big)+(\lambda_{1}+\lambda_{3}) \\ f\Big(\frac{\lambda_{1}x_{1}+\lambda_{3}x_{3}}{\lambda_{1}+\lambda_{3}}\Big)\Big]\end{multlined} \] for all \(x_{1},x_{2},x_{3}\in U\) and \(\lambda_{1},\,\lambda_{2},\,\lambda_{3}\geq 0\) with \(\lambda_{1}+\lambda_{2}+\lambda_{3}>0\). The authors prove that every \(2D\)-convex function is convex in the usual sense. The converse holds only in the case of functions defined on intervals. They also prove a discrete analogue of Jensen's inequality in the framework of \(2D\)-convex functions and a Hermite-Hadamard type inequality for the \(2D\)-convex functions. The authors discuss a conjecture about the equivalence of \(2D\)-convexity to an analogue of midpoint convexity, built on the unweighted form of Popoviciu's inequality (that is (1) in the case where \(U\) is an interval of the real line and \(\lambda_{1}=\lambda_{2}=\lambda_{3}=1\)). Finally, the authors prove that the convolution of a \(2D\)-convex function by a nonnegative function with compact support is \(2D\)-convex, too.