3
$\begingroup$

Can there be a set whose Hausdorff dimension gradually changes?

For instance, a set of real numbers contained in an interval, whose Hausdorff dimension is 0 at the beginning and 1 closer to the end, and changes without jumps?

$\endgroup$
19
$\begingroup$

I assume you want a set $A\subseteq [0,1]$ such that $\dim (A\cap [0,x])=x$ for all $x$. We can define $A_1$ by taking the union of a (Borel) subset of dimension $0$ of $[0,1/2]$ with a subset of dimension $1/2$ of $[1/2,1]$.To obtain $A_2$, we then make our sets larger on $[1/4,1/2]$, $[3/4,1]$ and again, we make the dimension on each current interval equal to its left endpoint. Then $A_1\subseteq A_2\subseteq \ldots$, and $A=\bigcup A_n$ works: by monotone convergence, $h^d(I\cap A)=\lim h^d(I\cap A_n)$, so $A\cap [0,x]$ has the right dimension at each $x=k2^{-n}$ and thus everywhere.

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ Can we Lebesgue-integrate over such set? $\endgroup$ – Anixx Jan 18 at 23:06
  • 10
    $\begingroup$ @Anixx Each $A_n$ has Hausdorff dimension below $1$, so their (1-dimensional) Lebesgue measure is zero. So $A$ has Lebesgue measure zero. You can certainly integrate functions over null sets, but the results are not too interesting... $\endgroup$ – Wojowu Jan 18 at 23:11
  • 1
    $\begingroup$ In fact for all $0<x\le1$ $A\cap[0,x]$ is an $x$-dimensional set of $x$-dimensional null measure, for the same reason. $\endgroup$ – Pietro Majer Jan 19 at 20:05
  • $\begingroup$ Cool. I think a similar idea (reversing subintervals of $[k/2^n,(k+1)/2^n]$ as needed, for $n\rightarrow \infty, 0\le k<n$) can also be used to produce a set $A$ with density $x$ at $x$, i.e. $\mu(A\cap[0,x])=x^2/2$. $\endgroup$ – Yaakov Baruch Jan 20 at 8:31
  • 2
    $\begingroup$ @YaakovBaruch Are you sure? Doesn't that contradict Lebesgue Density Theorem? $\endgroup$ – Wojowu Jan 20 at 8:46
5
$\begingroup$

For an example of a different sort, you can consider the record set $R$ of a fractional Brownian motion with varying Hurst parameter $H(t)$. In this short paper it is shown that when $H(t)$ is constant, it equals the Hausdorff dimension of $R$. The properties of the fBM with varying Hurst parameter $H(t)$ are given here for measurable functions $H(t)$ taking values in $(\tfrac12,1)$. One expects that the Hausdorff dimension of $R\cap (t-\epsilon,t+\epsilon)$ converges to $H(t)$ as $\epsilon\to 0$.

| cite | improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.