Implications between approximate convexity properties and approximate Hermite-Hadamard inequalities (Q424133): Difference between revisions

Let \(D\) be a convex subset of a real linear space \(X\) and let \(f:D \to \mathbb R\) be a convex function, then it satisfies the lower and upper Hermite--Hadamard inequalities \[ f\Big(\frac{x+y}{2}\Big) \leq \int_0^1 f(tx+(1-t)y) dt, \qquad x, y \in D; \] \[ \int_0^1 f(tx+(1-t)y) dt \leq \frac{f(x)+f(y)}{2}, \qquad x, y \in D. \] Conversely, if a function \(f:D \to \mathbb R\) is continuous over the segments of \(D\) and satisfies the previous inequalities, then it is convex. It is known that the \(\varepsilon\)--convexity of \(f\), i.e., the validity of \[ f(tx+(1-t)y) \leq tf(x)+(1-t)f(y)+\varepsilon, \qquad x, y \in D, \] implies the \(\varepsilon\)-Hermite-Hadamard inequalities \[ f\Big(\frac{x+y}{2}\Big) \leq \int_0^1 f(tx+(1-t)y) dt+\varepsilon, \qquad x, y \in D; \] \[ \int_0^1 f(tx+(1-t)y) dt \leq \frac{f(x)+f(y)}{2}+\varepsilon, \qquad x, y \in D. \] The reversed implication is not true. In order to obtain a positive result in this direction a nonconstant error term must be considered. The authors investigate the connection between the following inequalities: \[ f\Big(\frac{x+y}{2}\Big) \leq \frac{f(x)+f(y)}{2}+\alpha_J(x-y), \qquad x, y \in D; \] \[ f\Big(\frac{x+y}{2}\Big) \leq \int_0^1 f(tx+(1-t)y) dt+\alpha_H(x-y), \qquad x, y \in D; \] \[ \int_0^1 f(tx+(1-t)y)\rho(t) dt \leq \lambda f(x)+(1-\lambda)f(y)+\alpha_H(x-y), \qquad x, y \in D. \] where \(\alpha_J, \alpha_H\) are given functions, \(\lambda \in [0,1]\), and \(\rho:[0,1] \to \mathbb R_+\) is an integrable nonnegative function with \[ \int_0^1 \rho(t) dt=1. \]