Quasi-arithmetic means of continuous functions (Q1085353)

Let I be a proper interval, X a compact Hausdorff space and C(X,I) the set of continuous mappings from X into I. The authors define a mean on C(X,I) as a continuous functional satisfying \(M(a1_ X)=a\) for all \(a\in I\) and Mf\(\leq Mg\) if \(f\leq g\). A mean is strictly increasing if \(Mf<Mg\) for \(f<g\) and bisymmetric if \(M_ y(M_ xh(x,y))=M_ x(M_ yh(x,y))\) for all \(h\in C(X\times X,I)\), where \(M_ x\) applies to \(x\mapsto h(x,y)\), \(M_ y\) to \(y\mapsto h(x,y)\). They prove the following theorem [as they point out, a similar but not identical result is contained in a paper of the reviewer, \textit{I. Fenyö} and \textit{J. Horváth} in Port. Math. 8, 1-11 (1949; Zbl 0033.174)]. A mean on C(X,I) is strictly increasing and bisymmetric iff there exist a strictly monotonic \(\phi\) :I\(\to {\mathbb{R}}\) and a probability measure \(\mu\) with support X such that \[ M(f)=\phi^{-1}(\int_{X}\phi (f)d\mu)\text{ for all } f\in C(X,I) \] (with \(\phi\) unique up to a multiplicative constant and \(\mu\) unique). Furthermore, for \(I={\mathbb{R}}_+\) (the set of positive numbers), the mean M is said to be homogeneous if \(M(rf)=rM(f)\) for all \(r\in {\mathbb{R}}_+\), \(f\in C(X,{\mathbb{R}}_+)\) and the authors prove that the following bisymmetric means on \(C(X,{\mathbb{R}}_+)\) and only these are homogeneous: \[ (\int_{X}f^ pd\mu)^{1/p},\exp (\int_{X}\log f d\mu),\quad \max_{x\in X}(\delta (x)f(x))\quad and\quad \min_{x\in X}(\gamma (x)f(x)), \] where \(p\neq 0\) is a real constant, \(\mu\) a probability measure, \(\delta\) :X\(\to [0,1]\) and \(\gamma\) :X\(\to [1,\infty]\) are upper or lower semicontinuous functions, respectively, with \(\sup_{x\in X}\delta (x)=1=\inf_{x\in X}\gamma (x).\) (Of course, in this latter result it is not supposed that M is strictly increasing).