1
$\begingroup$

Let $C$ be the usual ternary cantor set, and $f$ the Cantor function, or Devil’s staircase associated to it. We know that $f$ is differentiable a.e., and on every point of the complement $C^c$, the derivative is $0$. Is there a description of the set of points of $C$ on which $f$ is differentiable? And can we identify the derivative?

$\endgroup$

1 Answer 1

4
$\begingroup$

We can characterize the Cantor set as the set of $x \in [0,1]$ which have a base-$3$ expansion $x = \sum_{i=1}^\infty x_i 3^{-i}$ where all $x_i \in \{0,2\}$. The Cantor function on $C$ is then $f(x) = \sum_{i=1}^\infty (x_i/2) 2^{-i}$. Now if $x \in C$, you can get $y \in C$ with $|x - y| = 2 \cdot 3^{-k}$ by flipping the $k$'th digit: $y_i = x_i$ except $y_k = 2-x_i$; and $|f(x) - f(y)| = 2^{-k}$. Since $\dfrac{2^{-k}}{2 \cdot 3^{-k}} \to \infty$ as $k \to \infty$, $f$ is never differentiable at any $x\in C$.

$\endgroup$
5
  • 3
    $\begingroup$ Hm, this would seem to contradict the fact that the points of nondifferentiability of the Cantor function have Hausdorff dimension $(\log 2/ \log 3)^2$ (while the Cantor set has Hausdorff dimension $\log 2/ \log 3$). See e.g. jstor.org/stable/2159830?origin=crossref&seq=1. However I can’t seem to find anything wrong with your argument either.. $\endgroup$
    – Nate River
    Commented Apr 12, 2021 at 1:32
  • 4
    $\begingroup$ The Darst paper counts both right and left derivative ${} = +\infty$ as "differentiable". Robert does not. Darst says paper reference [4] characterizes the points of differentiability. $\endgroup$ Commented Apr 12, 2021 at 10:04
  • $\begingroup$ "Can we identify the derivative?" Answer: $+\infty$ at points of $C$ where it exists. $\endgroup$ Commented Apr 12, 2021 at 10:10
  • 1
    $\begingroup$ @Nate River: Pages 28-31 of this paper gives a nice survey, and some of the references in this question/answer may also be of use. $\endgroup$ Commented Apr 12, 2021 at 10:24
  • $\begingroup$ Ah, that settles it nicely @GeraldEdgar, thanks! Also Dave L Renfro, thanks for the extensive biblography. It seems they have taken a similar approach as Robert did in the paper to compute the Dini derivatives. $\endgroup$
    – Nate River
    Commented Apr 12, 2021 at 13:20

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .