Extremal approximately convex functions and the best constants in a theorem of Hyers and Ulam. (Q1865276)

Let \(U\) be a convex subset of \(\mathbb R^n\), \(B\geq 2\) and \(\Delta_{B-1}\) be the standard \(B-1\) simplex. One says that \(f\: U\rightarrow\mathbb R\) is \(\varepsilon\)-convex with respect to \(\Delta_{B-1}\) if \[ f\left(\sum_{i=1}^B t_ix_i\right)\leq\sum_{i=1}^B t_i f(x_i)+ \varepsilon \] for all \(x_1,\dots,x_B\in U\) and \((t_1,\dots,t_B)\in\Delta_{B-1}\). This notion generalizes the concept of \(\varepsilon\)-convexity introduced by \textit{D. H. Hyers} and \textit{S. M. Ulam} [Proc. Am. Math. Soc. 3, 821--828 (1952; Zbl 0047.29505)], which corresponds to the case \(B=2\). By adapting a proof due to \textit{P. W. Cholewa} [Aequationes Math. 27, 76--86 (1984; Zbl 0549.39006)] of the Hyers-Ulam stability theorem for \(\varepsilon\)-convex functions, the authors prove that if \(f\) is \(\varepsilon\)-convex with respect to \(\Delta_{B-1}\) then there exist convex functions \(g,g_0: U\rightarrow\mathbb R\) such that \[ g(x)\leq f(x)\leq g(x)+\kappa(n,B)\varepsilon \quad\text{and} \quad | f(x)-g_0(x)| \leq\frac{\kappa(n,B)}{2}\varepsilon \] for all \(x\in U\) and obtain the best contant \(\kappa(n,B)\) in these inequalities. They also find the pointwise largest function on \(\Delta_n\) that is \(1\)-convex with respect to \(\Delta_{B-1}\) and takes a nonpositive value on the vertices of \(\Delta_{B-1}\).