On the properties of McShane's functional and their applications (Q2250845): Difference between revisions

Given a nonempty set \(E\) and a closed convex set \(K\subset\mathbb R^n\) let \(\mathcal L(E,\mathbb R)\) be a linear class of real-valued functions \(f: E\to\mathbb R\) such that \(1\in\mathcal L(E,\mathbb R)\) and let \(\mathcal F(K,\mathbb R)\) be the linear space of all real functions on \(K\). Further, let \(\mathcal L^+_0(E,\mathbb R)\) be the subset of non-negative functions from \(\mathcal L(E,\mathbb R)\) and let \(\mathcal I(\mathcal L(E,\mathbb R),\mathbb R)\) be the space of positive isotonic linear functional on \(\mathcal L(E,\mathbb R)\), i.e., linear functionals \(A\) such that \(A(f)\geq0\) for all non-negative \(f\in\mathcal L(E,\mathbb R)\). If \(A\in\mathcal I(\mathcal L(E,\mathbb R),\mathbb R)\) satisfies \(A(1)=1\) then the generalization of the Jensen inequality \(\Phi(A(f))\leq A(\Phi(f))\) holds [\textit{J. E. Pečarić} et al., Convex functions, partial orderings and statistical applications. Boston, MA etc.: Academic Press (1992; Zbl 0749.26004)]. Its higher-dimensional variant is the McShane inequality \(\Phi(A(\boldsymbol f))\leq A(\Phi(\boldsymbol f))\) for \(\boldsymbol f\in\mathcal L(E,\mathbb R)^n\). Deduced from the McShane inequality, the McShane functional \(J: \mathcal F(E,\mathbb R)\times\mathcal L(E,\mathbb R)^n\times\mathcal L^+_0(E,\mathbb R)\times\mathcal I(\mathcal L(E,\mathbb R),\mathbb R)\to\mathbb R\) is given by \(J(\Phi,\boldsymbol f,p;A)=A(p\Phi(\boldsymbol f))-A(p)\Phi (\frac{A(p\boldsymbol f)} {A(p)})\) whenever this expression is well defined. The main result states that for \(\Phi\in\mathcal F(E,\mathbb R)\) continuous and convex, \(\boldsymbol f\in\mathcal L(E,\mathbb R)^n\) and \(A\in\mathcal I(\mathcal L(E,\mathbb R),\mathbb R)\) the functional \(J\) is superadditive, increasing and non-negative in \(\mathcal L^+_0(E,\mathbb R)\), i.e., \(J(\Phi,\boldsymbol f,p+q;A)\geq J(\Phi,\boldsymbol f,p;A)+J (\Phi,\boldsymbol f,q;A)\) and \(J(\Phi,\boldsymbol f,p;A)\geq J(\Phi,\boldsymbol f,q;A)\geq0\) for \(p,q\in\mathcal L^+_0(E,\mathbb R)\) such that \(p\geq q\). The reversed inequalities hold if \(\Phi\) is concave. The result is applied to weighted general means, to inequalities of additive and multiplicative type and to the Hölder inequality.