Approximate differentiability almost everywhere (Q847020)

The authors provide a characterization of almost everywhere approximately differentiable measurable functions. To this end, a Saks Theorem-type result is also obtained. Let \(k \geq 2\), \(X \subseteq \mathbb R^{k}\) be a Lebesgue measurable set, \(x \in \mathbb R^{k}\) be a Lebesgue density point of \(X\) and let \(f : X \rightarrow \mathbb R\) be a Lebesgue measurable function. Then \(f\) is \textit{approximately differentiable at \(x\)} if there exists a Lebesgue measurable set \(U \subseteq X\) such that \(x\) is a density point of \(U\) and \(f|_{U}\) is differentiable at \(x\). The Lebesgue measure on \(\mathbb R ^{k}\) is denoted by \(\lambda\). For \(x = (x_{0}, \dots, x_{k-1})\) and \(y = (y_{0}, \dots, y_{k-1})\) in \(\mathbb R ^{k}\), we write \(x < y\) if \(x_{i} < y_{i}\) \((i < k)\). For \(x,y \in \mathbb R ^{k}\) with \(x < y\), we set \([x,y] = \prod_{i < k} [x_{i},y_{i}]\) and \(r([x,y]) = \lambda([x,y])/L^{k}(x,y)\) where \(L(x,y) = \max_{i < k} |y_{i} - x_{i}|\). For \(x \in \mathbb R ^{k}\) and \(\alpha \in (0,1)\), set \(A_{\alpha}(x) = \{y \in \mathbb R ^{k} : x < y , r([x,y]) > \alpha\}\). Analogously to a result of \textit{H.\ Whitney} [Pac.\ J.\ Math.\ 1, 143--159 (1951; Zbl 0043.05803)], the authors obtain the following theorem. Let \(k \geq 2\), \(X \subseteq \mathbb R^{k}\) be a bounded Lebesgue measurable set and \(f : X \rightarrow \mathbb R\) be a Lebesgue measurable function. Then \(f\) is approximately differentiable \(\lambda\)-almost everywhere if and only if for every \(\varepsilon > 0\) there exists a measurable set \(Y \subseteq X\) such that {\parindent=6mm \begin{itemize}\item[(i)] \(\lambda(X \setminus Y) < \varepsilon\) and \item[(ii)] for every \(\delta > 0\) there is an \(n \in \mathbb N\) and \(\alpha \in (0,1)\) such that \(\lambda(\{x \in Y : \exists y \in Y~(x < y , r([x,y]) > \alpha , |y - x| < 1/n , f(y) - f(x) > n|y-x|)\}) < \delta.\) \end{itemize}} The proof of this result uses the following Saks Theorem-type result, which may be of independent interest. Let \(\alpha \in (0,1)\), \(k \geq 2\), \(X \subseteq \mathbb R^{k}\) and \(f : \mathbb R^{k} \rightarrow \mathbb R\) be arbitrary. If for every \(x \in X\) we have \(\limsup_{y \rightarrow x, y \in A_{\alpha}(x)} f(y) < \limsup_{y \rightarrow x } f(y) \), then \(\lambda(X) = 0\).