Tuesday, April 26, 2016

Wallis sieve, and lp n-balls III

This is the third installment, do read the previous posts. I left some standing questions in the last post, and I am going to answer at least one of them: Can we squeeze an $l_p$ ball´s volume out of a generalized Wallis sieve?. At the same time I will generalize and simplify the proof from the xkcd post. First, lets generalize the Wallis sieve.

In $d$ dimensions, starting with a $p$ sided hypercube and cutting it appropriately (I leave to the reader to draw it), we get the product,
$$A_n^d = \prod_{n=1}^\infty \frac{p^d n^{d-1} \Big(n+\frac{d}{p}\Big)} {(pn+1)^d},$$
which can be rewritten as the limit
$$A_n^d = lim_{n\to\infty} \frac{p^{nd}\Gamma(n+1)^{d-1}\Gamma\Big(n+1+\frac{d}{p}\Big)\Gamma\Big(1+\frac{d}{p}\Big)^d}{p^{nd}\Gamma(1+\frac{d}{p})\Gamma\Big(n+1+\frac{1}{p}\Big)^d}$$
where I have made use of the Euler gamma function recurrent properties (see previous posts).
Note that the $d$ in $A_n^d$ is an index, not an exponent.
 The limit can be separated into two factors (after cancelling the $p^{nd}$),
 $$A_n^d = lim_{n\to\infty}\Bigg[ \frac{\Gamma(n+1)^{d-1}\Gamma\Big(n+1+\frac{d}{p}\Big)}{\Gamma\Big(n+1+\frac{1}{p}\Big)^d} \Bigg]\Bigg[\frac{\Gamma\Big(1+\frac{1}{p}\Big)^d}{\Gamma(1+\frac{d}{p})}\Bigg]$$.
Remember that $\frac{\Gamma(n+a)}{\Gamma(n)}\sim n^a$, so the first term when $n\to\infty$
$$ \frac{\Gamma(n+1)^{d-1}\Gamma\Big(n+1+\frac{d}{p}\Big)}{\Gamma\Big(n+1+\frac{1}{p}\Big)^d} = \frac{\Gamma(n+1)^{d-1}\Gamma\Big(n+1+\frac{d}{p}\Big)}{\Gamma\Big(n+1+\frac{1}{p}\Big)^{d-1}\Gamma\Big(n+1+\frac{1}{p}\Big)} \sim \frac{n^{\frac{d-1}{p}}}{n^{\frac{d-1}{p}}}\sim 1.$$
So we obtain, finally,
 $$A_n^d = lim_{n\to\infty} \frac{\Gamma\Big(1+\frac{1}{p}\Big)^d}{\Gamma(1+\frac{d}{p})} = V_d^p\Big(\frac{1}{2}\Big).$$
The value of the limit is the volume of the $l_p$ ball, with the case of the hypersphere $p=2$ being a particular case.
Another remarkable case happens when $p=1$ and we obtain the volume of the $d$-dimensional cross-polytope,  of radius $R=\frac{1}{2},$ which is $\frac{1}{d!},$ . The cross-polytope is the generalization of the octahedron to $n$ dimensions and is the dual of the hypercube we start with.

This formula lets us also interpret the volume of various fat Cantor and other Smith-Cantor-Volterra sets.
The question still standing from last post is: Is there a geometrical interpretation for the intermediate $A_n^d$? And of course, What more can we learn from this relation between $l_p$ and these sets?

No comments:

Post a Comment