Tried to prove that the following property related to Hausdorff dimension holds: $H_{dim}([0,1]) = 1$, where:

  • $H_{dim}(A):=inf\{s\in [0,+\infty] : H^{s}(A)=0 \}$ ,
  • $H^{s}(A):= \lim\limits_{\delta \to 0 }H_{\delta}^{s}(A) $
  • $H_{\delta}^{s}:=inf\{\alpha(s)\sum\limits_{i=1}^{+\infty} (\frac{diamC_{i}}{2})^{s}: A \subseteq \bigcup\limits_{i=1}^{+\infty}C_{i}, diamC_{i} \leq \delta \} $
  • $\alpha(s):= \frac{\sqrt\pi^{s}}{\Gamma(\frac{s}{2}+1)}$

Tried to derive this proof by the use of proof of contradiction by deriving contradiction in the cases $s\in [0,1)$ and $s\in (1,+\infty)$ but I failed to obtain any results.

Help in deriving this proof very appreciated!

  • $\begingroup$ I wonder why this is tagged "contest-math"? $\endgroup$ – Mark McClure Jan 28 '17 at 21:15
  • $\begingroup$ I am preparing for Putnam. That is why. $\endgroup$ – user410985 Jan 28 '17 at 21:15

Not sure if that is the easiest way: to calculate $H^s_{\delta}$, it suffices to consider those $\{ \widetilde C_i\}$ which are all intervals and are pairwise disjoint: If $\{ C_i\}$ is any covering of $[0,1]$. Let $\widetilde C_1$ be the smallest interval containing $C_1$. Then $\widetilde C_1$ and $C_1$ has the same diameter. Let $C^1_i = C_i\setminus \widetilde C_1$ and do the same as $n=2$: Let $\widetilde C_2$ be the smallest intervals containing $C^1_2$.

Then $\{\widetilde C_i\}$ will be covering $[0,1]$ and pairwise disjoint (possibly not, but they intersect only at the endpoints) and $\text{diam} (\widetilde C_i) \le \text{diam} (C_i)$.

Not you are essentially finding up to constant

$$H_\delta^s([0,1]) = \inf\left\{ \sum_{i=1}^\infty x_i^s : \delta\ge x_i \ge 0, \sum_{i=1}^\infty x_i = 1\right\}.$$

For $s=1$, it implies $H^1_\delta([0,1]) = 1$ for all $\delta$, so $H^1([0,1]) = 1$. When $s<1$, $x_i^s \ge x_i$. So

$$H^s_\delta \ge 1$$

for all $\delta$ and so $H^s([0,1]) \neq 0$. When $s>1$, choose the following covering (where $1/n \le \delta$)

$$\{ [0,1/n], [1/n, 2/n], \cdots, [(n-1)/n, 1]\}.$$


$$H^s_\delta([0,1]) \le \sum_{i=1}^n \frac{1}{n^s} = \frac{1}{n^{s-1}} \to 0$$

as $n\to\infty$. This implies

$$H^s_\delta([0,1]) = 0$$

for all $\delta $ and so $H^s([0,1]) = 0$ for all $s>1$. Thus the Hausdorff dimension is one.


If you are familiar with measure theory then you may note that the restriction to $[0,1]$ of Lebesgue measure $(\lambda)$ is $1$-Ahlfors regular, i.e. for $x\in [0,1]$, $0<r<1/2$: $$ r\leq \lambda(B(x,r)) \leq 2r $$ Given any countable open cover you may replace by a cover using intervals (without changing Hausdorff measure) and then by countable sub-additivity: $$ 1\leq \sum_{k\geq 1} \lambda(U_i) \leq \sum_k |U_i|$$ which implies that dimH is at least 1. Taking an equipartition gives an upper bound equal to 1.


Your Answer

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