1

Let $$\int_{-\infty}^{\infty}f(x) dx=1$$ Then, find $$\int_{-\infty}^{\infty}f\left(x-\frac{1}{x}\right) dx$$

I subtituted $x=t-\frac{1}{t}$ in the original equation and found $$\int_{-\infty}^{0}f(t-\frac{1}{t}) dt= \frac 12$$

But, I couldn't proceed after this.

Quanto
  • 86,284
  • 7
  • 94
  • 181
Aspirant
  • 81
  • 7
  • Can you elaborate a little bit more your result ? – Erik Satie Jul 28 '20 at 16:27
  • 2
    There must be something you're not telling us - knowing the first integral does not determine the second. (Also the you did the substitution wrong...) – David C. Ullrich Jul 28 '20 at 16:27
  • 1
    You should graph the change of variables to figure out what the new bounds will look like, and you have to figure out what the relationship between $dx$ and $dt$ – Ninad Munshi Jul 28 '20 at 16:36
  • 2
    This is a special case of the [Glasser's Master Theorem](https://en.wikipedia.org/wiki/Glasser%27s_master_theorem), although it has long been known before Glasser (such as Pólya and Szegö). – Sangchul Lee Jul 28 '20 at 17:22
  • The is question B4 from the 1968 Putnam exam. https://web.archive.org/web/20080330184108/http://www.kalva.demon.co.uk/putnam/putn68.html – Gerry Myerson Jul 31 '22 at 07:38

2 Answers2

5

Split $( -\infty, \infty)\to ( -\infty, 0) + ( 0, \infty) $ \begin{align} &\int_{-\infty}^{\infty}f\left(x-\frac{1}{x}\right) \overset{x\to -\frac1x}{ dx}\\ =& \int^{\infty}_{0}f\left(x-\frac{1}{x}\right) \frac {dx}{x^2}+ \int^{0}_{-\infty}f\left(x-\frac{1}{x}\right) \frac {dx}{x^2}\\ =&\ \frac12\int_{-\infty}^{0}f\left(x-\frac{1}{x}\right)\left( 1+\frac1{x^2}\right) \overset{x\to x-\frac1x}{dx} + \frac12\int_{0}^{\infty}f\left(x-\frac{1}{x}\right)\left( 1+\frac1{x^2}\right) \overset{x\to x-\frac1x}{dx}\\ =& \int_{-\infty}^{\infty}f(x)dx=1 \end{align}

Quanto
  • 86,284
  • 7
  • 94
  • 181
  • See https://link.springer.com/article/10.1007/s11785-010-0095-0 and https://math.stackexchange.com/questions/1690150/has-this-chaotic-map-been-studied/2578363#2578363 ; the correct answer is $1$ not $1/2$. The map $x\mapsto x-1/x$ is a 2-branched covering. – kimchi lover Jul 28 '20 at 16:55
3

Let $a, b \in \mathbb{R}$, $g(x) = x - \frac{1}{x}$. Then

$$\chi_{[a,b]} \circ g = \chi_{\left[\frac{a + \sqrt{a^2 + 4}}{2}, \frac{b + \sqrt{b^2 + 4}}{2}\right]}(x) + \chi_{\left[\frac{a - \sqrt{a^2 +4}}{2}, \frac{b - \sqrt{b^2 + 4}}{2}\right]}(x)$$

But $$\mu\left(\left[\frac{a + \sqrt{a^2 + 4}}{2}, \frac{b + \sqrt{b^2 + 4}}{2}\right] \cup \left[\frac{a - \sqrt{a^2 + 4}}{2}, \frac{b - \sqrt{b^2 + 4}}{2}\right]\right) = b - a$$ so $g$ is measure preserving and it follows that the result is 1.

cha21
  • 902
  • 3
  • 8