Korovkin type theorems and approximate Hermite-Hadamard inequalities (Q452072)

Let \(X\) be a real linear space and let \(D \subset X\) be a convex subset. One can easily see that, for any constant \(\varepsilon \geq 0\), the \(\varepsilon\)-convexity of \(f\), i.e., the validity of NEWLINE\[NEWLINEf(tx+(1-t)y)\leq t f(x) + (1-t) f(y) +\varepsilon \qquad (x,y\in D, \;t\in [0,1]), NEWLINE\]NEWLINE implies the following lower and upper \(\varepsilon\)-Hermite-Hadamard inequalities NEWLINE\[NEWLINE f\left(\frac{x+y}{2}\right) \leq \int_0^1 f(tx+(1-t)y) dt +\varepsilon \qquad (x,y\in D) \tag{1}NEWLINE\]NEWLINE and NEWLINE\[NEWLINE \int_0^1 f(tx+ (1-t)y)dt \leq \frac{f(x)+f(y)}{2} +\varepsilon \qquad (x,y\in D). \tag{2}NEWLINE\]NEWLINE For \(\varepsilon =0\), the above implication was discovered by \textit{J. Hadamard} [Journ. de Math. (4) IX. 171--215 (1893; JFM 25.0698.03)] and the converse is also known to be true (see [\textit{C. P. Niculescu} and \textit{L.-E. Persson}, Real Anal. Exch. 29, No. 2, 663--686 (2003--2004; Zbl 1073.26015)]). Concerning the reversed implication for the case \(\varepsilon >0\), \textit{K. Nikodem} et al. [Math. Inequal. Appl. 10, No. 2, 359--363 (2007; Zbl 1117.39019)] have shown that the \(\varepsilon\)-Hermite-Hadamard inequalities (1) and (2) do not imply the \(c\varepsilon\)-convexity of \(f\) (with any \(c>0)\). In order to obtain results that establish implications between the approximate Hermite-Hadamard inequalities and the approximate Jensen inequality, one has to consider these inequalities with nonconstant error terms.NEWLINENEWLINE \textit{J. Makó} and \textit{Z. Páles} [Cent. Eur. J. Math. 10, No. 3, 1017--1041 (2012; Zbl 1243.39021)] established the connections between an upper Hermite-Hadamard type inequality and a Jensen type inequality (with a nonconstant error term) under some conditions, and \textit{A. Házy} and \textit{Z. Páles} [Proc. R. Soc. Lond., Ser. A, Math. Phys. Eng. Sci. 465, No. 2102, 571--583 (2009; Zbl 1186.39032)] obtained a relationship between a lower Hermite-Hadamard type inequality and a Jensen type inequality (with a nonconstant error term) under some conditions.NEWLINENEWLINEThe authors examine the implication from a lower Hermite-Hadamard type inequality to a convexity type inequality. The authors generalize the Házy and Páles' result replacing the Lebesgue integral by an integral with respect to an arbitrary Borel probability measure. This allows us to view an approximate Jensen inequality as a particular approximate Hermite-Hadamard inequality.NEWLINENEWLINE\textit{M. Bessenyei} [JIPAM, J. Inequal. Pure Appl. Math. 9, No. 3, Paper No. 63, 51 p. (2008; Zbl 1173.26007)] defined the \((w_0, w_1)\)-convexity and established the connections between the \((w_0,w_1)\)-convexity and the Hermite-Hadamard type inequality. These results are generalized in a systematic way.