I was reading Terry Tao's second volume on Analysis and he makes the following remark on page 586. I was hoping that someone could explain why this is the case since he does not give a proof.

"The unit interval has outer measure 1, but the unit interval $\{(x,0) : 0 \leq x \leq 1\}$ in $R^2$ has two dimensional outer measure 0. . . . Note that the above remarks and countable additivity imply that the entire x-axis has two-dimensional outer measure 0, despite the fact that $R$ has infinite one-dimensional measure"

I am not sure if I am understanding the terminology correctly. I assume he is talking about the two dimensional outer measure of the unit interval and not the two dimensional outer measure of a unit box.

Given that outer measure is defined as follows for some set $B$ that is covered by an interval $[a,b]$

$$ m^*_n(B) := \prod^n_{i=1}(b_i - a_i) $$

So the one dimensional outer measure$(m^*_1(\cdot))$ of the unit interval would be:

$$ m^*_1([0,1]) = (1-0) = 1 $$

When he says the two dimensional outer measure of the unit interval, is he talking about the two dimensional measure in the sense of $\{(0,1), (0,0)\}$ meaning ?:

$$ m^*_2([0,1]) = (1-0)(0-0) = 0 $$

Let me know if I am looking at this incorrectly. Thanks.

  • 2
    $\begingroup$ Your definition of outer measure is very informal to the point of being incorrect. However, yes: it is sufficient to say that the unit interval is "covered" by a rectangle of height zero. $\endgroup$ – Omnomnomnom Nov 30 '15 at 20:53
  • $\begingroup$ Yep, sorry I should have included the more detailed definition as in the response by @anomaly below. $\endgroup$ – krishnab Nov 30 '15 at 23:51

The definition of $m^*$ you give only allows for a single rectangle. With that, the disc of radius $r$ would have measure $4r^2$, for example. Instead, you want something like $m_n^*(E) = \inf \sum_i \lambda(B_i)$, where \begin{align*} \lambda([a_1, b_1] \times \cdots \times [a_n, b_n]) = \prod_i (b_i - a_i) \end{align*} and the infimum runs over finite sets of $n$-dimensional rectangles that cover $E$. (We can also consider products of open intervals, etc.)

With that, we can show that $m_2^*([0, 1]) = m_2^*([0, 1]\times \{0\})$ vanishes by considering the cover with a single rectangle $[0, 1] \times [-\epsilon, \epsilon]$ for arbitrarily small $\epsilon > 0$. Then $m_2^*([0, 1]) \leq 2\epsilon$ by definition; and since clearly $m_2^*([0, 1]) \geq 0$, we have $m_2^*([0, 1]) = 0$.

  • $\begingroup$ Okay right. Yes, thanks for your more complete answer and for validating my original intuition. You are correct, I should have included the more detailed definition of outer measure. I did not mean to leave out the important detail. $\endgroup$ – krishnab Nov 30 '15 at 23:50

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.