Convex functions with respect to an arbitrary mean (Q4357532)

If \(M\) is a two variable mean on an interval \(I\subset\mathbb{R}\), then the authors call a function \(\varphi:I\to\mathbb{R}\) to be \textit{\(M\)-convex} if, for all \(x,y\in I\), \[ f(M(x,y))\leq M(f(x),f(y)). \] Analogously, the notions of \(M\)-concavity, \(M\)-affinity can also be defined. Taking the arithmetic mean, \(M\)-convexity reduces to Jensen-convexity. The authors study the class \(\text{Conv}_M(I)\) of functions which consists of all \(M\)-convex functions defined on the interval \(I\). It is proved that the superadditivity of the mean \(M\) is sufficient to obtain that \(\text{Conv}_M(I)\) is a semigroup. Conversely, if the class of \(M\)-convex functions is large enough and if \(\text{Conv}_M(I)\) is a semigroup, then \(M\) must be superadditive. In the case when \(M\) is a positively homogeneous mean, \(\text{Conv}_M(I)\) turns out to be a cone, that is to be closed under multiplication by positive scalars. As the main application, the authors take the Stolarski means defined by \[ E_{r,s}(x,y)=\left({r\over s}\cdot {x^s-y^s\over x^r-y^r}\right)^{1\over s-r} \] if \(rs(s-r)(x-y)\not=0\) and by a limiting argument in the remaining cases. Using the characterization of superadditivity of the means \(E_{r,s}\) due to \textit{L. Losonczi} and \textit{Zs. Páles} [\textit{Minkowski's inequality for two variable difference means}, Proc. Am. Math. Soc. (in print)], the main results are specialized to Stolarski means. In this way, the authors prove that polynomials with nonnegative coefficients, exponential functions with base bigger than one are \(E_{r,s}\)-convex if \(r+s\leq3\) and \(\min(r,s)\leq 1\). For the entire collection see [Zbl 0864.00057].