Let $A\subseteq \mathbb{R}$ be Lebesgue-measurable and let $0<\alpha<1$ be its Hausdorff dimension.

For a given $0<\beta <\alpha$ can we find a subset $B\subset A$ with Hausdorff dimension $\beta$?

In case this is true, could you provide a reference for this statement?

Added: Actually I am happy if $A$ is compact.


First of all, $\dim_{H} (A) = \alpha$ iff $ H^k(A)=\infty$ for all $k<\beta$ and $H^k(A) = 0$ for all $k>\beta$. Then $H^\alpha(A) = \infty$ for all $\alpha \in (0,\beta)$.

If $A$ is closed then by Theorem 5.4 from [1] there is a compact $K\subset A$ such that $0<H^\alpha(K)<\infty$.

More generally, if $A$ is Souslin then by Theorem 5.6 from [1] again there is a compact $K\subset A$ such that $0<H^\alpha(K)<\infty$ (as @SeverinSchraven noted).

If $A$ is not Souslin then (at least assuming AC and CH) it may happen that $H^\beta(B) = 0$ for any $B\subset A$ and any $\beta \in (0, \alpha)$. In order to justify this claim let me add adopt the argument from a comment by fedja to a closely related post, where an interesting paper [2] is also mentioned.

Let $C$ denote the standard Cantor set in $[0,1]$ and take $\alpha = \dim_H(C) = \frac{\ln 2}{\ln 3}$. Take $\beta \in (0,\alpha)$.

Consider the family $\mathcal E$ of all $G_\delta$-subsets $E \subset C$ such that $H^\alpha(E)=0$. Then for any countable family $(E_i)_i\in \mathcal E$ the set $C \setminus \bigcup_{i=1}^\infty E_i$ has infinite $\beta$-dimensional Hausdorff measure and in particular is not empty.

Assuming AC+CH one can enumerate the family $\mathcal E$ using ordinals as $\mathcal E = \{E_\gamma : \gamma < \omega_1\}$, where $\omega_1$ is the first uncountable ordinal.

For each $\gamma < \omega_1$ take $x_\gamma \in C \setminus \bigcup_{\lambda < \gamma} E_\lambda$. Now we finally define the set $A:= \{x_\gamma : \gamma < \omega_1\}$.

If $H^\alpha(A)$ was zero then one could cover $A$ with $E_\gamma$ for some $\gamma$ (see e.g. Theorem 1.6 in [1]). This contradicts the definition of $A$, hence $H^\alpha(A)>0$.

Now if $B \subset A$ and $H^\beta(B)\in (0,\infty)$ then $H^\alpha(B)=0$ and then there exists $\gamma<\omega_1$ such that $B \subset E_\gamma$, hence $$ H^\beta(B) = H^\beta(B\cap E_\gamma) \le H^\beta(A\cap E_\gamma) = 0 $$ since the intersection $A\cap E_\gamma$ is at most countable for any $\gamma < \omega_1$.


  1. The Geometry of Fractal Sets by K.J. Falconer (1985)

  2. Finding subsets of positive measure by Bjørn Kjos-Hanssen and Jan Reimann (2014) https://arxiv.org/abs/1408.1999

| cite | improve this answer | |
  • $\begingroup$ That's exactly what I was looking for, thanks very much. $\endgroup$ – Severin Schraven Mar 15 '19 at 22:01
  • $\begingroup$ Actually it is worth mentioning that in the same reference in Theorem 5.6, this is generalized to Souslin spaces. $\endgroup$ – Severin Schraven Mar 16 '19 at 0:06
  • $\begingroup$ @SeverinSchraven Thank you, I have included this remark into my answer. $\endgroup$ – Skeeve Mar 16 '19 at 17:56

The answer is yes under the additional assumption that the set is compact and I do not know what happens in the general case. The result is a consequence of the following one, see [1] and references therein.

Theorem. If a compact set $A\subset\mathbb{R}^n$ has non-$\sigma$-finite measure $\mathcal{H}^\beta$, then there exists a subset $B\subset A$ such that $0<\mathcal{H}^\beta(B)<\infty$.

[1] R.O. Davies, A theorem on the existence of non-σ-finite subsets. Mathematika 15 (1968), 60–62.

| cite | improve this answer | |
  • 2
    $\begingroup$ I actually have a question (which is the reason why I wrote that my answer is partial): in the OP it is assumed that $A$ is Lebesgue measurable. Strictly saying this does not imply that $A$ is Borel, so it is not immediate that the result we both eventually refer to can be used (if $A$ were Borel or at least Souslin then yes). $\endgroup$ – Skeeve Mar 15 '19 at 20:56
  • 1
    $\begingroup$ @Skeeve Good point. I do not know what happens in general, but perhaps the compact case is sufficient for the needs of OP. Let's hear from him. I will changes my answer to emphasize that it only answers the compact case. By the way you know a lot of geometric measure theory so you should be more active :) There are not too many of us. $\endgroup$ – Piotr Hajlasz Mar 15 '19 at 21:23
  • $\begingroup$ @PiotrHajlasz Indeed, the compact case is fine for me. Thanks for the answer. $\endgroup$ – Severin Schraven Mar 15 '19 at 22:01
  • $\begingroup$ @PiotrHajlasz It seems to me that in the general case under certain axioms bad things might happen. I have updated my answer with some thoughts on that. By the way thank you, I will try to be more active :) Even though you probably overestimate my knowledge of GMT, I hope to learn it better in nice discussions on this site (in particular with you, of course!) $\endgroup$ – Skeeve Mar 16 '19 at 18:03

The following is Corollary 7 of [1].

Theorem: For $X$ (an analytic subset of) a complete separable metric space, and $ s \in [0,\infty)$, the following is true about the Hausdorff measure $\mathcal{H}^s$: For every $ l < \mathcal{H}^s(X) $ there exists a compact subset $A \subset X$ such that $$ l < \mathcal{H}^s (A) < \infty \, . $$

[1] J. D. Howroyd, On dimension and on the existence of sets of finite positive Hausdorff measure, 1993

| cite | improve this answer | |

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.