home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: sci.math
- Path: sparky!uunet!cs.utexas.edu!zaphod.mps.ohio-state.edu!moe.ksu.ksu.edu!ux1.cso.uiuc.edu!news.cso.uiuc.edu!needham
- From: needham@symcom.math.uiuc.edu (James Needham)
- Subject: Re: Points on Sphere
- References: <1993Jan11.184545.18912@pinet.aip.org> <1istp4INNfgj@matt.ksu.ksu.edu>
- Message-ID: <C0pwtn.LyF@news.cso.uiuc.edu>
- Sender: usenet@news.cso.uiuc.edu (Net Noise owner)
- Organization: University of Illinois at Urbana
- Date: Tue, 12 Jan 1993 01:44:09 GMT
- Lines: 84
-
- Since there seems to be interest in placing n points on a
- sphere, I thought that I would post what little I know. I
- know that I've been working on the problem for a year,
- and it has stubbornly resisted every means of attack.
-
- The generalized problem is this: place n points on the
- unit sphere in order to extremize Sum(i<j, d(p(i), p(j)) **
- alpha), where d is the Euclidean distance. For alpha > 0,
- we want to maximize the sum, for alpha < 0, we want to
- minimize the sum. alpha = -.5 corresponds to the
- potential energy determined by n point charges. I've
- been concentrating on alpha = 1.
-
- The corresponding problem in the plane, putting points
- on the circle, has been solved for many values of alpha.
- See (1) and (2). Suffice it to say that the regular n-gon
- usually works; but for 1 < alpha < 2, I think the problem
- is still open.
-
- For the sphere, tough, most of the interesting cases are
- still unknown. There is an algorithm (see (3)) that takes
- n given points and moves them around (basically
- LaGrange multipliers) to what *seems* to be the best
- possible configurations. Unfortunately, for alpha = 1, the
- optimal configuration can only be *proved* for n = 3
- (equilateral triangle), n = 4 (regular tetrahedron), n = 12
- (regular whatever), and n = 24 (regular whatever). The
- algorithm indicates that the following are optimal:
-
- n = 5: bipyramid
- n = 6: bipyramid
- n = 7: one north pole and three pairs symmetric through
- the z axis
- n = 8: skewed cube
- n = 9: equilateral triangle on the equator with skewed
- equilateral triangles
- above and below it
-
- I have "optimal" solutions for n through about 50. Again,
- hardly any of these can be proven. The hope that there
- is one method that will take in n and spit out the best
- configuration is fading. The placement really seems to
- depend on the particular properties of n. At one point I
- even had a conjecture relating it to twin primes.
-
- The problem with this algorithm is that it behaves
- somewhat chaotically. It is possible that an initial
- configuration gets sucked into a *local* extremal
- configuration (for example, with 5 points one can start
- with configurations that converge to the pyramid rather
- than the optimal bipyramid). This seems to happen,
- though, with probability 0. Does anybody out there know
- anything about chaotic dynamical systems on higher
- dimensional manifolds? If so, you might find this
- problem interesting.
-
- One might also suspect that the optimal configurations
- remain the same as one varies alpha. This is not the case.
- With seven points , the bipyramid is minimal for alpha =
- -.5, while for alpha = 1 the configuration described above
- is better. (I think that the strange configuration
- described above for 7 points gives way to the bipyramid
- at about alpha = .6, if I remember correctly.)
-
- There are lots of interesting problems in here, but don't
- get obsessed with it. Any questions or comments will be
- cheerfully entertained!
-
- 1. R. Alexander and K. Stolarsky, Extremal problems of
- distance geometry related to energy integrals, Trans. of
- AMS 193 (1974).
-
- 2. G. Bjorck, Distributions of positive mass which
- maximize a certain generalized energy integral, Ark. Mat.
- 3 (1956), 255-269.
-
- 3. J. Berman and K. Hanes, Optimizing the arrangement of
- points on the unit sphere, Math. Comp. 31 (1977), 1006-
- 1008.
-
-
- Jim Needham
- needham@symcom.math.uiuc.edu
-
-