Takagi functions and approximate midconvexity (Q1025054)

This paper deals with approximate (mid)convexity of functions. It is known that the bound of any continuous \((\varepsilon,p)\)-midconvex function \(h\), with \(\varepsilon\geq 0\) and \(p>0\), can be estimated in terms of \(h\) and a Takagi-like function \(\omega_p:\mathbb R\to\mathbb R\) which is given by \(\omega_p(x)=\sum_{k}\frac{1}{2^{kp}}d(2^kx)\), where \(d(x)=2\operatorname{dist}(x;\mathbb Z)\), \(x\in\mathbb R\) [see \textit{J. Tabor} and \textit{J. Tabor}, Generalized approximate midconvexity, Control Cybernet. (in press)]. Here, the authors prove that, for \(p\in[1,2]\), the bound above cannot be improved, that is, there is no \(v_p\leq \omega_p\) which can replace \(\omega_p\) in the estimate of the bound. In order to show this fact, the authors give a good estimate for the bound of the Jensen difference of the Takagi-like function \(\omega_p\) and it turns out naturally that \(\omega_p\) is \((1,p)\)-midconvex. Finally, for certain values of \(p\in[1,2]\), the authors give explicit formulae for the maximum of \(\omega_p\).