1
$\begingroup$

Suppose $S$ is the sphere of radius one. And suppose $f:S\to \mathbb{R}$ is defined as follows: $$f(x_1,...,x_n)=\frac{1}{x_1^2}+\frac{1}{x_2^2}+\cdots+\frac{1}{x_n^2}$$

I am trying to calculate either of these: $$\int f(\gamma)d\gamma$$ where integrating is with respect to Haar measure over the sphere. Or maybe finding the median of the induced random variable on $\mathbb{R}$ by $f$. I appreciate any help in this direction.

$\endgroup$
5
  • $\begingroup$ You can try n-dimensional spherical coordinates to integrate it. Honestly, it's more of a math question and has a little to do with stats. I'd suggest to post it on math.stackexchange.com. $\endgroup$ Commented Mar 15, 2015 at 13:41
  • $\begingroup$ Well the truth is I did. Maybe you are right. But the reason I posted this is that $f(\gamma)$ is a random variable in the end that I calculate expected value with respect to.... I wanted to apply Levy's lemma to it :) $\endgroup$
    – Cupitor
    Commented Mar 15, 2015 at 13:49
  • 1
    $\begingroup$ If you want to calculate expected value numerically, then Stats is the right place to ask. In that case you can use Monte Carlo methods to sample points on an n-sphere. This is a good way to check your analytical calculations. $\endgroup$ Commented Mar 15, 2015 at 13:56
  • $\begingroup$ Numerical calculations are quite promising but I cannot get anywhere else analytically :| $\endgroup$
    – Cupitor
    Commented Mar 15, 2015 at 14:08
  • 2
    $\begingroup$ It looks like the integral (for n > 1) is divergent, thus the expectation doesn't exist. $\endgroup$ Commented Mar 15, 2015 at 14:15

1 Answer 1

2
$\begingroup$

Let's use spherical coordinates:

$$ x_k = \prod_{j=1}^{k-1} \sin(\phi_j) \cos(\phi_k) \\ J = \prod_{j=1}^{n-2} \sin^{n-1-j}(\phi_j) $$

I omitted $r$ because we'll be considering sphere of unit radius. Thus

$$ \int f d S = 2^n \int_{0}^{\pi/2} \dots \int_{0}^{\pi/2} \Bigl( \sum_{k=1}^n \prod_{j=1}^{k-1} \sin^{-2}(\phi_j) \cos^{-2}(\phi_k) \Bigr) \prod_{j=1}^{n-2} \sin^{n-1-j}(\phi_j) d \phi_1 \dots d \phi_n \\ = 2^n \sum_{k=1}^n \int_{0}^{\pi/2} \dots \int_{0}^{\pi/2} \prod_{j=1}^{k-1} \sin^{-2}(\phi_j) \cos^{-2}(\phi_k) \prod_{j=1}^{n-2} \sin^{n-1-j}(\phi_j) d \phi_1 \dots d \phi_n $$

The last equation looks frightening, but it's just a product on $n$ integrals of sinuses and cosinuses raised to some power. And some of them are of particular interest.

Consider $\phi_{n-1}$ for $k = n$. Corresponding integral is

$$ \int_0^{\pi/2} \sin^{-2}(\phi_{n-1}) d\phi_{n-1} $$

This integral is divergent. Clearly, the whole sum thus is unbounded.

Thus the expectation does not exist.

$\endgroup$
1
  • $\begingroup$ This is the easy part! (It doesn't really require spherical coordinates to be answered, either.) The hard part of the question is--what is the median of $f(X_1, \ldots, X_n)$? That calculation might benefit from the use of spherical coordinates. $\endgroup$
    – whuber
    Commented Mar 15, 2015 at 19:24

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.