Cone unrectifiable sets and non-differentiability of Lipschitz functions (Q2317674): Difference between revisions

Before adding more details, I state one of the main results of the paper under review: If \(E\subset \mathbb{R}^n\) is cone unrectifiable, then there is a Lipschitz function \(f: \mathbb{R}^n\to \mathbb{R}\) that is non-differentiable at any point of \(E\). I do not state the exact definition of a set \(C\) to be cone unrectifiable, but it means roughly that in each point there is a vector normal to the tangent space. Let me embed this result into some history. The starting point is Rademacher's theorem that Lipschitz functions are differentiable almost everywhere. Combining this result with a result obtained by \textit{Z. Zahorski} [Bull. Soc. Math. Fr. 74, 174--178 (1946; Zbl 0061.11302)] shows that a set \(N\subset \mathbb{R}\) is a set of measure zero if and only if there is a Lipschitz function that is not differentiable at any point of \(N\) (as the points of nondifferentiability form a \(G_{\delta\sigma}\) set, there are in general points outside \(N\) where the function is not differentiable). \textit{D. Preiss} then showed in [J. Funct. Anal. 91, No. 2, 312--345 (1990; Zbl 0711.46036)] that such a characterization does not hold for sets in \(\mathbb{R}^2\). Namely, he showed the existence of a set in the plane with vanishing measure but each Lipschitz function on the plane has a point of differentiability in this set. The above mentioned main result gives more information on the quality of such sets. The first section after the introduction deals with approximation of functions by Lipschitz functions of various sorts. These results are then used in the following section where the main results are proved. The final section looks at examples showing that some of the results are sharp. One might ask if the cone unrectifiable sets are exactly the sets of non-differentiability for Lipschitz functions. The authors state that this is unknown and provide an example showing that a possible proof cannot be along the lines of Zahorski's proof of his above mentioned result.