wu :: forums (http://www.ocf.berkeley.edu/~wwu/cgi-bin/yabb/YaBB.cgi)
riddles >> putnam exam (pure math) >> hypersphere packing
(Message started by: JocK on Jul 25th, 2004, 1:30pm)

Title: hypersphere packing
Post by JocK on Jul 25th, 2004, 1:30pm
In N dimensions, N+1 equally sized hyperspheres can be configurated in a cluster such that each hypersphere touches all other hyperspheres.

For N[to][infty] calculate the radius of the smallest hypersphere that can contain such a cluster (assign a radius of unity to the spheres in the cluster).

J8)CK

Title: Re: hypersphere packing
Post by Eigenray on Jul 26th, 2004, 6:59am
First let's inductively find (n+1) equally spaced vectors in Sn-1 [subset] [bbr]n.
[hide]For n=1 we have of course v0=-1, v1=1.
Suppose v0, ..., vn are equally spaced in Sn-1.  Now let's rotate each of these up into the (n+1)th dimension, giving vectors vi' = (vi*sqrt(1-t2),t) [in] Sn, which gives us room to insert the vector (0, -1).
Let dn = <vi,vj>, i [ne] j.  d1 = -1.
We pick t in such a way to make them equally spaced, i.e.,
dn+1 = <vi', vj'> = <vk',(0,-1)>
dn(1-t2) + t2 = -t
t = -1, dn/(dn-1).
dn+1 = -dn/(dn-1)
Inductively we find dn = -1/n.
The distance rn between two of these vectors is given by
rn2 = <vi-vj, vi-vj> = 1 + 1 - 2(-1/n) = 2 + 2/n.
So, with our equally spaced vectors in place, we scale them such that each pair are a distance 2 apart to make room for the hyperspheres:
wi = vi*2/rn
The radius of the containing hypersphere is |wi|+1 = [sqrt](2n/(n+1)) + 1 [to] 1+[sqrt]2[/hide]

Title: Re: hypersphere packing
Post by Eigenray on Jul 27th, 2004, 6:14am
Compare to this packing (http://www.ocf.berkeley.edu/~wwu/cgi-bin/yabb/YaBB.cgi?board=riddles_putnam;action=display;num=1065293213), where the radius of the inscribed hypersphere is unbounded as n[to][infty].

Title: Re: hypersphere packing
Post by JocK on Jul 31st, 2004, 12:15pm
Correct Eigenray!

Your reference to the inscribed hypersphere problem is interesting, as it is (losely) related to the follow-up question:


Things get a bit more complicated as we now investigate unequally sized hyperspheres:

In n dimensions, if properly sized, n+2 hyperspheres can be configurated in a cluster such that each hypersphere touches all other hyperspheres. In particular, for n dimensions this holds for properly chosen [alpha][subn]>1 if the hyperspheres have radii R,  [alpha][subn] R,  [alpha][subn]^2 R, .. ,  [alpha][subn]^(n+1) R.

This means that an infinite nesting of hyperspheres can be constructed with each subsequent hypersphere being a factor 1/[alpha] smaller in size, such that within this nesting each group of n+2 subsequently sized hyperspheres consist of hyperspheres that all touch each other. For such a cluster of n+2 subsequent hyperspheres, the smallest hypersphere is the inscribed hypersphere of the n+1 larger spheres.

For n=3,  [alpha] < 2. Can you prove this? What is the exact value?

What happens if we increase n beyond 3? What is the value for [alpha][subn] for n [to] [infty]? And how does [alpha][subn]^n (the volume-ratio of subsequent hyperspheres) behave for n [to] [infty]?

J8)CK



Powered by YaBB 1 Gold - SP 1.4!
Forum software copyright © 2000-2004 Yet another Bulletin Board