The Hasudorff measure of a set $A\subset \mathbb R^n$ is defined as $$ \mathcal H^s(A) = \lim_{\delta\downarrow 0} \mathcal H^s_\delta(A)$$ where $$\mathcal H^s_\delta(A) = \inf\left\{\sum_{j=1}^\infty (\text{diam }S_j)^s : \{S_j\}_{j=1}^\infty \subset 2^{\mathbb R ^n} \text{ covers } A \text{ and } \text{diam } S_j < \delta \right\} $$ It is easy to see that $\mathcal H^s_\delta(A)$ is non-increasing in $\delta$. However, I don't quite understand why $\mathcal H^s_\delta(A)$ is not actually constant in $\delta$ for each $A$; it would seem that the infimum should be achieved as the diameters of the sets $S_j$ get smaller and smaller, meaning that bounding the diameter of $S_j$ above by $\delta$ should do nothing at all to $\inf\sum_{j=1}^\infty (\text{diam }S_j)^s $. Moreover, suppose we approach Lebesgue measure analogously. Let $\mathscr{L^n}(A)$ be $n$-dimensional Lebesgue measure, and let $$\mathscr{L_\delta^n}(A) := \inf\left\{\sum_{j=1}^\infty|B_j| : \{B_j\}_{j=1}^\infty \subset 2^{\mathbb R ^n} \text{ is a box cover of } A \text{ and } \text{diam } B_j < \delta \right\} $$ we find that $\mathscr{L^n_\delta}(A)$ is constant in $\delta$, so that for any $\delta>0$ $$\mathscr{L^n}(A) = \mathscr{L^n_\delta}(A)= \lim_{\epsilon\rightarrow 0}\mathscr{L^n_\epsilon}(A)$$
So what about switching from volumes of boxes to diameters of arbitrary sets makes $\mathcal H^s_\delta(A)$ not necessarily constant in $\delta$?

This question hints at a much more general question about the construction of outer measures via the Carathéodory method. A brief recap on this method (as I have learned it in class): Suppose $\mathcal F$ is a collection of subsets of some metric space $X$, let $A\subset X$, and consider the following set $$\mathcal C_{\delta, \mathcal F}(A) : = \left\{\{S_j\}^\infty\subset \mathcal F \mid \{S_j\}^\infty \text{ covers } A \text{ and } \text{diam }S_j< \delta\right\}$$ and consider any function $\zeta: \mathcal F \rightarrow [0,\infty]$. Then, the following functions are outer measures:

$$\mu_{\delta, \mathcal F}(A) := \inf_{\{S_j\}^\infty \in \mathcal C_{\delta, \mathcal F}(A)} \sum^\infty_{j=1} \zeta(S_j)$$ $$\mu_{\mathcal F}(A) := \lim_{\delta \downarrow 0}\mu_{\delta, \mathcal F}(A)$$ Notice that the Hausdorff measure $\mathcal H ^ s(A)$ is the special case where $\mathcal F = 2^X$ and $\zeta(S) = (\text{diam } S)^s$. Under what conditions is $\mu_{\delta,\mathcal F}(A)$ not constant in $\delta$ for every $A$, as is the case for $\mathcal H^s_{\delta}$?


Isn't it perfectly plausible that if you are allowed to use larger sets, then you get a smaller number? (because you are only summing up a function of the diameter of the set, remember)

Note that one reason for this construction is that lots of sets are not $\mathcal{H}_{\delta}^s$ measurable but are $\mathcal{H}^s$ measurable, but this isn't what you asked specifically.

The following is pretty much the simplest situation where something goes wrong: Take $\mathcal{H}_{\delta}^1$ on $\mathbf{R}^2$ for some $\delta > 0$ and then consider

Let $A := ([0,1) \times \{0\}) \cup ([0,1) \times \{\epsilon\}) \subset \mathbf{R}^2$ for some $\epsilon << \delta$. This is two parallel, horizontal line segments of length $1$, separated by a tiny gap of width $\epsilon$.

Now, how would you efficiently cover this set if you are allowed to use a covering by sets of diameter $\delta$?

You can essentially cover both lines at once, right? To cover $A$ efficiently you use very thin rectangles whose long sides are just less than delta and whose short sides are $\epsilon$. Summing their diameters gives you something close to 1. (this can be made rigorous)

(roughly speaking, since $\delta$ is much bigger than $\epsilon$, the outer measure $\mathcal{H}^1_{\delta}$ is too coarse; the set looks like a single line from the point of view of $\mathcal{H}^1_{\delta}$).

But now if you imagine changing $\delta$ so that things are the other way around, i.e. take $\delta << \epsilon$, then of course eventually the $\mathcal{H}^1_{\delta}$ measure has to get close to 2, which is the true $\mathcal{H}^1$ measure.


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.