Tutorials › Measure Theory › Cylinder Covers Of The Middle-Third Cantor Set

Hausdorff Measure and Dimension · Tutorial 696 of 798

Cylinder Covers Of The Middle-Third Cantor Set

Use the level-by-level cylinder covers to calculate Hausdorff costs and prove an upper bound for the Cantor set’s dimension.

Advanced 9 min read

What You'll Learn

  • Describe the level-n basic intervals of the middle-third Cantor set.
  • Compute the Hausdorff cost of a cylinder cover at an arbitrary exponent.
  • Identify the exponent at which these cover costs change from growing to shrinking.
  • Use cylinder covers to prove an upper bound for Hausdorff dimension.
  • Distinguish an upper-bound argument from a proof of the exact dimension.

Level-by-Level Covers

The middle-third Cantor set is built by repeatedly removing open middle thirds from the intervals that remain. At stage \(n\), the remaining set consists of \(2^n\) closed intervals, each of length \(3^{-n}\). These intervals are useful covers: their sizes and their number are explicit, so their total cost can be computed at any Hausdorff exponent.

Work in the usual metric on \(\mathbb{R}\). Write \(C_n\) for the union of the intervals remaining after stage \(n\), with \(C_0=[0,1]\), and let \(C=\bigcap_{n=0}^{\infty}C_n\) be the middle-third Cantor set. Each interval at stage \(n\) is indexed by a word \(w=(a_1,\ldots,a_n)\) with every \(a_j\) equal to \(0\) or \(2\). It is

$$ I_w=\left[\sum_{j=1}^{n}a_j3^{-j},\ \sum_{j=1}^{n}a_j3^{-j}+3^{-n}\right]. $$

There are \(2^n\) such words, hence \(2^n\) intervals. Each interval has diameter \(3^{-n}\), and \(C\subseteq C_n=\bigcup_w I_w\). Thus these intervals form a cover of \(C\) whose maximum diameter tends to zero as \(n\) increases.

Definition: For \(s>0\), \(\delta>0\), and \(E\subseteq\mathbb{R}\), the \(s\)-dimensional Hausdorff content at scale \(\delta\) is \[ \mathcal{H}^s_\delta(E)=\inf\left\{\sum_{i=1}^{\infty}(\operatorname{diam}U_i)^s: E\subseteq\bigcup_{i=1}^{\infty}U_i,\ \operatorname{diam}U_i\le\delta\right\}. \] The \(s\)-dimensional Hausdorff outer measure is \(\mathcal{H}^s(E)=\lim_{\delta\downarrow0}\mathcal{H}^s_\delta(E)\). The Hausdorff dimension is \(\dim_H E=\inf\{s>0:\mathcal{H}^s(E)=0\}\).

The limit defining \(\mathcal{H}^s(E)\) exists in \([0,\infty]\): reducing \(\delta\) restricts the allowed covers, so \(\mathcal{H}^s_\delta(E)\) is nondecreasing as \(\delta\) decreases. The cover cost at stage \(n\), for exponent \(s\), is obtained by multiplying the number of intervals by the \(s\)-th power of their common diameter:

$$ \sum_w(\operatorname{diam}I_w)^s =2^n(3^{-n})^s =\left(\frac{2}{3^s}\right)^n. $$

This formula is the central calculation. If \(2/3^s<1\), the stage costs tend to zero; if \(2/3^s=1\), they stay equal to one; and if \(2/3^s>1\), they grow. The equality case defines the critical exponent for this family of covers.

Computing the Critical Exponent

Proposition: For the level-\(n\) cylinder cover of \(C\), the cost at exponent \(s>0\) is \((2/3^s)^n\). The critical exponent for these costs is \[ d=\frac{\log 2}{\log 3}, \] at which every level has cost \(1\).

Proof. There are \(2^n\) intervals in the level-\(n\) cover, and each has diameter \(3^{-n}\). Therefore the cost is \(2^n(3^{-n})^s=(2/3^s)^n\). The ratio equals one precisely when \(3^s=2\). Taking logarithms gives \(s=\log 2/\log 3=d\). Substitution gives \(2/3^d=1\), so the cost is \(1^n=1\) at every level. If \(s>d\), then \(3^s>3^d=2\), and the ratio is less than one. If \(0<s<d\), then \(3^s<2\), and the ratio is greater than one. \(\square\)

Worked Example: The Cost at Exponent One

Compute the cost of the level-four cover when \(s=1\), and compare it with the general formula.

At level four there are \(2^4=16\) intervals, each of length \(3^{-4}=1/81\). Since the diameter to the first power is the interval length, the cost is

$$ 16\cdot\frac{1}{81}=\frac{16}{81}. $$

The general formula gives \((2/3^1)^4=(2/3)^4=16/81\), the same value. At level \(n\), these costs are \((2/3)^n\), which tend to zero because \(0<2/3<1\). This is consistent with \(1>d\), since \(3^1=3>2\). Check: substituting \(n=4\) into \((2/3)^n\) gives \(2^4/3^4=16/81\), agreeing with the direct interval count.

Worked Example: The Cost at the Critical Exponent

Compute the level-three cost at \(s=d=\log 2/\log 3\), and explain why the same result holds at every level.

There are \(2^3=8\) intervals, each of diameter \(3^{-3}=1/27\). At the critical exponent, \(3^d=2\), so the cost is

$$ 8\left(\frac{1}{27}\right)^d =8\cdot 3^{-3d} =8\cdot (3^d)^{-3} =8\cdot 2^{-3} =1. $$

For any level \(n\), the same substitution gives \(2^n(3^d)^{-n}=2^n2^{-n}=1\). Check: for \(n=3\), the interval count contributes \(8\), and the diameter factor is \((1/27)^d=(3^d)^{-3}=1/8\); their product is \(8\cdot1/8=1\).

The Upper Bound for Hausdorff Dimension

Theorem: For the middle-third Cantor set \(C\subseteq\mathbb{R}\), if \(s>\log 2/\log 3\), then \(\mathcal{H}^s(C)=0\). Consequently, \[ \dim_H C\le\frac{\log 2}{\log 3}. \] At the critical exponent \(d=\log 2/\log 3\), the cylinder covers also give \(\mathcal{H}^d(C)\le1\).

Proof. Fix \(s>d\) and \(\delta>0\). Since \(3^{-n}\to0\), there are arbitrarily large \(n\) for which \(3^{-n}\le\delta\). The level-\(n\) intervals are therefore an admissible cover in the definition of \(\mathcal{H}^s_\delta(C)\). By the cost calculation, their cost is \(q^n\), where \(q=2/3^s\). Because \(s>d\), the proposition gives \(0<q<1\), and hence \(q^n\to0\). For every \(\varepsilon>0\), choose such an \(n\) large enough that \(q^n<\varepsilon\). It follows that \(0\le\mathcal{H}^s_\delta(C)\le q^n<\varepsilon\). Since this holds for every \(\varepsilon>0\), \(\mathcal{H}^s_\delta(C)=0\). This is true for every \(\delta>0\), so \(\mathcal{H}^s(C)=0\). Thus every \(s>d\) belongs to the set whose infimum defines \(\dim_H C\), giving \(\dim_H C\le d\).

For \(s=d\) and any \(\delta>0\), choose \(n\) with \(3^{-n}\le\delta\). The level-\(n\) cover is admissible and has cost \(1\), so \(\mathcal{H}^d_\delta(C)\le1\). Taking the limit as \(\delta\downarrow0\) preserves this upper bound, giving \(\mathcal{H}^d(C)\le1\). \(\square\)

Worked Example: An Exponent Above the Critical Value

Take \(s=d+1\). Calculate the level-\(n\) cost and the cost at level three.

Since \(3^d=2\), we have \(3^{d+1}=3\cdot3^d=6\). The ratio in the cost formula is therefore \(2/3^{d+1}=2/6=1/3\), so the level-\(n\) cost is \((1/3)^n\). At level three it is

$$ \left(\frac{1}{3}\right)^3=\frac{1}{27}. $$

This tends to zero as \(n\) increases, as required for an exponent strictly above \(d\). Check: directly, the level-three cost is \(8(1/27)^{d+1}=8(1/27)^d/27\). Because \((1/27)^d=(3^d)^{-3}=1/8\), this equals \(8\cdot(1/8)/27=1/27\).

What These Covers Do—and Do Not—Show

The argument proves an upper bound: it constructs covers whose costs tend to zero for every \(s>d\). It does not prove that the dimension equals \(d\). In particular, the fact that the level-\(n\) costs grow when \(s<d\) does not show that \(\mathcal{H}^s(C)\) is positive. Hausdorff content takes an infimum over all admissible covers, not just the level-by-level cylinder covers. A lower bound must rule out small-cost covers of every allowed form; that requires an additional argument.

A common error is to use the ordinary interval lengths for every exponent. For example, at \(s=1/2\) and level four, an incorrect calculation might multiply the 16 intervals by \(1/81\), obtaining \(16/81\). That is the cost at \(s=1\), not at \(s=1/2\). The prescribed cost uses the diameter raised to the chosen exponent. Here each diameter contributes \((1/81)^{1/2}=1/9\), so the correct cost is \(16\cdot1/9=16/9\). The general formula confirms it: \(2^4(3^{-4})^{1/2}=16\cdot3^{-2}=16/9\). Thus the exponent changes the contribution of each interval, even though the cover itself is unchanged.

The critical-cost calculation is useful because it identifies exactly where these particular covers stop becoming cheaper with increasing level. At \(s=d\), their cost remains one, so they give a finite upper bound for the critical Hausdorff outer measure. For \(s>d\), increasingly fine cylinder covers have arbitrarily small cost. For \(s<d\), their cost increases—but that observation alone is not a lower-bound proof. Keeping these conclusions separate is essential when using explicit covers to study Hausdorff dimension.

Check Your Understanding

Use the level-\(n\) cost formula, and distinguish conclusions about cylinder covers from conclusions about all covers.

  1. How many intervals occur at level \(n\), and what is the diameter of each one?
  2. Compute the level-two cylinder-cover cost at exponent \(s=1\).
  3. Show from \(3^d=2\) that every level-\(n\) cover has cost one at \(s=d\).
  4. Why does a cost tending to zero for the level covers imply \(\mathcal{H}^s(C)=0\) when \(s>d\)?
  5. Why does the growth of level-cover costs for \(0<s<d\) not by itself prove \(\mathcal{H}^s(C)>0\)?