1
$\begingroup$

I am interested in the distribution of Eucildean distance between two points in a $k$-dimensional space.

Fortunately I came across the following paper which is pretty good in deriving the CDF and PDF:

Distribution of Euclidean Distances Between Randomly Distributed Gaussian Points in n-Space

Unfortunately, I have troubles understanding some of the notations.

In particular the authors define the two points in a k-dimensional space as $\Gamma = (\gamma_1, \ldots, \gamma_k)$ and $\Psi = (\psi_1, \ldots, \psi_k)$. However, later on the CDF is derived the following: $$F(R, k) = 1 - \frac{\Gamma(\frac{k}{2}, \frac{R^2}{4})}{\Gamma(\frac{k}{2})}$$ in which $R$ seems to be defined as "the probability that the CDF is less than or equal to R". Taking aside the definition of $R$, the set $\Gamma$ seems suddenly to be a function with one or two inputs. The paper does not contain a definition of this function.

Hence my question, what does $\Gamma(x, y)$ and $\Gamma(x)$ return in this context?

Note: I tried to contact the original authors with out much success.

$\endgroup$
2
  • 2
    $\begingroup$ Perhaps the (generalized) Gamma functions - en.wikipedia.org/wiki/Incomplete_gamma_function? $\endgroup$
    – Calvin Lin
    Commented Nov 6, 2019 at 3:51
  • $\begingroup$ what if the Gaussians were not standard normal but some N(0,sigma^2). mean still 0 but not a unit variance. what would the result be in terms of the distrib? hope it doesnt change a tonne! $\endgroup$
    – hearse
    Commented Oct 19, 2021 at 17:48

2 Answers 2

0
$\begingroup$

As Calvin Lin suggested in the comments,

  • $\Gamma(x) = \int\limits_0^\infty t^{x-1} e^{-t}\,dt$, the Gamma function, which for integer $x$ is $\Gamma(x)=(x-1)!$
  • $\Gamma(x,y) = \int\limits_y^{\infty} t^{x-1}e^{-t}\,dt$, the upper incomplete Gamma function

The work in your link looks a little excessive. It is looking at the distribution of the distance between two independent standard multivariate normal random vectors. If this is $k$-dimensional and the distance is $D$, then you could say that

and the moments of these are well known

$\endgroup$
0
$\begingroup$

$F(R,k)=\mathbb{P}(\Vert X-Y\Vert\le R),$ where $X,\,Y\in\mathbb{R}^k$ and $X,Y\sim N(0,I_k).$ That is, it is the cdf of the distance between two $k$-dimensional standard gaussian random variables. Secondly, as pointed out in the comments, $\Gamma\left(\dfrac{k}{2},\dfrac{R^2}{4}\right)$ denotes the incomplete gamma function, and $\Gamma\left(\dfrac{k}{2}\right)$ is the standard gamma function.

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .