A class of nowhere differentiable functions satisfying some concavity-type estimate (Q2179850)

A class \(\mathcal P\) of continuous nowhere differentiable functions is described in terms of second-order differences defined with respect to dyadic rationals (the approach includes the reknown Takagi function [\textit{T. Takagi}, Tokio Math. Ges. 1, 176--177 (1903; JFM 34.0410.05)], which actually motivates such construction of \(\mathcal P\)). Namely, for a 1-periodic continuous function \(f\) on \(\mathbb R\), forward and backward differential quotients are defined, \begin{align*} \delta_{n,k}^+(y;f)&=\frac{r^n}{1-y}\cdot\left(f\left(\frac{k+1}{r^n}\right)-f\left(\frac{k+y}{r^n}\right)\right),\\ \delta_{n,k}^-(y;f)&=\frac{r^n}{1-y}\cdot\left(f\left(\frac{k+y}{r^n}\right)-f\left(\frac k{r^n}\right)\right), \end{align*} \(n\in\mathbb N\cup\{0\}\), \(k\in\mathbb Z\), \(y\in(0,1)\), which then give rise to a second-order difference \[\delta_{n,k}^+(y;f)-\delta_{n,k}^-(y;f).\] The function \(f\) belongs to \(\mathcal P_c\), \(c>0\), if, by definition, the above difference is \(\le-c\) for all \(n,k,y\). Finally, \(\mathcal P=\bigcup_{c>0}\mathcal P_c\). A relatively easy proof shows that every \(f\in\mathcal P\) is nowhere differentiable. Next, the authors consider how a ty\-pi\-cal approach to construction of continuous nowhere differentiable functions (via a series of functions; the Takagi or Weierstrass function case) fits into \(\mathcal P\). A criterion on a continuous 1-periodic function \(\psi\colon\mathbb R\to\mathbb R\) guaranteeing that \(U_\psi\) defined as \(U_\psi(x)=\sum_{j=0}^\infty\psi(r^jx)/r^j\), \(r\ge2\) an integer, belongs to \(\mathcal P\) is presented (Theorem 3.5). Moreover, a geometric condition corresponding to a mutual position of some parabolas with vertices on the graph of \(f\), is related to the class~\(\mathcal P\).