On a problem of Matkowski (Q4939419)

Let \(\phi\) and \(\psi\) be real-valued continuous strictly monotonic functions defined on a real interval \(I\) and define on \(I^2\) \[ A_{\phi}(x,y)=\phi^{-1}(\frac{\phi(x)+\phi(y)}{2}) \] (and similarly \(A_\psi\)). The authors prove that there exist constants \(a\neq 0\), \(c\neq 0\), \(p\neq 0,\) and \(b,d\) such that \(\phi(x)=ax+b, \psi(x)=cx+d\) or \(\phi(x)=a e^{px}+b, \psi(x)= ce^{-px}+d\) if, and only if (i) \(A_{\phi}(x,y)+A_{\psi}(x,y)=x+y\), and (ii) \(A_\phi-A_\psi\) is everywhere positive or everywhere negative or everywhere 0 on \(I^2.\) They conjecture that condition (ii) can be dropped. That would answer a problem which \textit{J. Matkowski} [Aequationes Math. 57, No. 1, 87-107 (1999; Zbl 0930.26014)] posed (and solved under second-order continuous differentiability conditions). Remark: The authors follow the general malpractice of denoting \(f(x)^2\) by \(f^2(x)\) but the value of the inverse function of \(f\) at \(x\) (and not \(f(x)^{-1}\)) by \(f^{-1}(x)\).