1
$\begingroup$

I am confused by this line from a paper:

"Let $F_1(x)$ be the cumulative distribution of the magnitude of an $n$−dimensional standard Gaussian random variable and $F_2(x)$ be the cumulative distribution of the magnitude of a random point in a Euclidean unit ball of radius $r$"

Can someone explain to me what $F_1(x)$ and $F_2(x)$ is? Thanks a lot!

The paper is here: http://www.mit.edu/~har/Dikin.pdf (page 9)

I would also really really appreciate it if someone has the time to explain to me how to sample from this Dikin ellipsoid. I think I've fully understood the definitions, but the part about the pseudo-inverse and vector scaling seems confusing.

Thank you!

Edit: I know how to sample points from the $n$-dimensional Gaussian and the $n$-dimensional Euclidean ball. But it seems to me that $F_1$ and $F_2$ are invertible functions?

$\endgroup$
2
$\begingroup$

Let $X$ be a standard $n$ dimensional Gaussian. Then $|X|$ is a random variable, and thus it has a distribution function. The author is saying let $F_1$ be that distribution function, i.e. $$ F_1(t) = P(|X| \leq t) $$

Similarly, the Euclidean ball of radius $r$ has finite volume and so we can consider the uniform measure $\mu$ on the ball (it will be a rescaling of the usual Lebesgue measure). Let $Y$ be a random vector sampled from $\mu$. Then $|Y|$ is a random variable and so it has a distribution function. The author is saying let $F_2$ be that distribution function, i.e. $$ F_2(t) = P(|Y| \leq t) $$

Edit: As for invertibilitity, you are correct that they are invertible. This is because $F_1,F_2$ are strictly increasing (a fact that you need to prove).

I suggest you make a separate, more specific, question about any technicalities that stump you in the paper.

$\endgroup$
  • $\begingroup$ Hi. Thank you so much for this! I went to the wikipedia page, and it seems to me that they both involve integrals. I thought the result would be a simple matrix. It is not clear to me how to invert these functions.. $\endgroup$ – Andy Yao Jul 28 '15 at 18:37
  • $\begingroup$ Actually $F_1$ is invertible on $[0,+\infty)$ only and $F_2$ is invertible on $[0,r]$ only. $\endgroup$ – Did Jul 29 '15 at 6:47

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.