Pythagorean theorem for Lebesgue measurable sets?

69 Views Asked by At

I'm currently studying measure theory and was thinking if we can generalize the Pythagorean theorem in the context of the Lebesgue measure theory in the following way:

Let us denote $r$ to be the diagonal $y=x$ on $\mathbb{R}^2$ (or any non horizontal or vertical line). So of course $r$ is in essence $\mathbb{R}$, therefore we can talk about Lebesgue measure on $r$ as we do on $\mathbb{R}$ (note that we are not measuring things in $r$ with the "induced" Lebesgue measure of $\mathbb{R}^2$, otherwise every subset of $r$ will be a null set).

My conjecture is the following:

Let $\pi_1:\mathbb{R}^2\to\mathbb{R}$, and $\pi_2:\mathbb{R}^2\to\mathbb{R}$ the projections of the first and second coordinates respectively (for instance $\pi_2(x,y)=y$)

Then if $E\subset r$ is a measurable set, then

$$m(E)^2=m(\pi_1(E))^2+m(\pi_2(E))^2$$

Is it true? If so, is there a good reference for that? What you think about that?

1

There are 1 best solutions below

1
On BEST ANSWER

This is true. I don't have a complete answer to your question, because you asked for a reference, and I don't have that. Possibly the result would be considered too trivial for a reference (which doesn't mean that you shouldn't have asked, of course); but I also imagine that it might appear as an exercise as an application of some general theorem (such as the Carathéodory Extension Theorem, which is the first thing that I thought of, although it's overkill in this case).

We can prove this directly by an explicit calculation. Let $ r $ be an oblique line with slope $ k $, explicitly the line $ \{ x , y \mid y = k x + c \} $ for some real numbers $ k $ and $ c $ with $ k \ne 0 $. To identify $ r $ with the real line, we need a map $ \iota \colon \mathbb R \to \mathbb R ^ 2 $ that maps the origin to a specific point (I'll use $ ( 0 , c ) $ rather than $ ( - c / k , 0 ) $ or some other choice) and moves at unit speed in the direction of a vector with slope $ k $ (I'll use $ \langle 1 , k \rangle $ rather than $ \langle - 1 , - k \rangle $ or some other choice), so that $$ \iota ( t ) = ( 0 , c ) + t \widehat { \langle 1 , k \rangle } = \bigg ( \frac 1 { \sqrt { k ^ 2 + 1 } } t , \frac k { \sqrt { k ^ 2 + 1 } } t + c \bigg ) \text . $$ (You can double-check that $ \iota ( t ) $ satisfies the defining equation of $ r $ and that the distance from $ \iota ( t _ 1 ) $ to $ \iota ( t _ 2 ) $, as given by the ordinary Pythagorean Theorem, is $ \lvert t _ 2 - t _ 1 \rvert $.) Then $$ \pi _ 1 ( \iota ( t ) ) = \frac 1 { \sqrt { k ^ 2 + 1 } } t \text {, and } \pi _ 2 ( \iota ( t ) ) = \frac k { \sqrt { k ^ 2 + 1 } } t + c , $$ so that $ \pi _ 1 \circ \iota $ and $ \pi _ 2 \circ \iota $ are affine-linear. This means that if $ E $ is measurable, then $$ m ( \pi _ 1 ( \iota ( E ) ) ) = \frac 1 { \sqrt { k ^ 2 + 1 } } m ( E ) \text {, and } m ( \pi _ 2 ( \iota ( E ) ) ) = \frac k { \sqrt { k ^ 2 + 1 } } m ( E ) \text ; $$ squaring these and adding, $$ m ( \pi _ 1 ( \iota ( E ) ) ) ^ 2 + m ( \pi _ 2 ( \iota ( E ) ) ) ^ 2 = \frac 1 { k ^ 2 + 1 } m ( E ) ^ 2 + \frac { k ^ 2 } { k ^ 2 + 1 } m ( E ) ^ 2 = m ( E ) ^ 2 \text , $$ which is basically your desired result.

In the above, I've taken $ E $ to be a subset of $ \mathbb R $ rather than $ r $, but the measure on $ r $ is defined to be the measure on $ \mathbb R $ transferred by $ \iota $, so that $ m ( \iota ( E ) ) = m ( E ) $. Or taking $ E $ to be a subset of $ r $ instead, $ m ( E ) = m ( \iota ^ { - 1 } ( E ) ) $, so $$ m ( \pi _ 1 ( E ) ) ^ 2 + m ( \pi _ 2 ( E ) ) ^ 2 = m ( \iota ^ { - 1 } ( E ) ) ^ 2 = m ( E ) ^ 2 \text ; $$ this is exactly your desired result. (But the formulas giving the measures of the projections are a stronger result, as in the comment that @C-RAM made.)