home *** CD-ROM | disk | FTP | other *** search
- Xref: sparky comp.theory:2353 sci.math:14501
- Newsgroups: comp.theory,sci.math
- Path: sparky!uunet!zaphod.mps.ohio-state.edu!cs.utexas.edu!qt.cs.utexas.edu!yale.edu!spool.mu.edu!news.nd.edu!mentor.cc.purdue.edu!pop.stat.purdue.edu!hrubin
- From: hrubin@pop.stat.purdue.edu (Herman Rubin)
- Subject: Re: Uniform noise in a d-sphere
- Message-ID: <BxAt7A.8D5@mentor.cc.purdue.edu>
- Sender: news@mentor.cc.purdue.edu (USENET News)
- Organization: Purdue University Statistics Department
- References: <1992Nov5.211723.26238@bnlux1.bnl.gov>
- Date: Fri, 6 Nov 1992 14:21:09 GMT
- Lines: 35
-
- In article <1992Nov5.211723.26238@bnlux1.bnl.gov> murphy@sscdaq.phy.bnl.gov (Michael Murphy) writes:
-
- >I am trying to compute uniformly random noise inside a d-dimensional
- >sphere. I have identified two ways of doing so:
-
- >(1) (Straightforward Monte Carlo method) Choose a random d-vector from
- >the d-hypercube, centered at the origin with extreme points -1 and 1.
- >Accept this vector only if it falls inside the unit d-sphere. This
- >method is adequate for low dimensions, but as the dimension increases,
- >the volume of the d-sphere is significantly less than the volume of
- >the hypercube. Therefore many samples must be taken until a 'hit' is
- >found. In a high dimension, such as 25, the wait is unbearable.
-
- >(2) A faster, although perhaps wrong approach is:
- > 1. Choose an arbitrary d-vector
- > 2. Normalize it so it is a point on the d-dimensional sphere
- > 3. Multiply this normalized vector by a random number in [0,1]
- > raised to the 1/dth power.
-
- >I believe that method two works. However, when I project a large number of
- >points in a high dimensional space (e.g. 25) onto the plane (by ignoring all
- >but two coordinates), I get something that resembles a square. My advisor
- >is not so sure that this is correct and I can only give an intuitive
- >argument as to why it may be true.
-
- This method works if the d-vector consists of independent normal random
- variables, but is likely not to work otherwise. Another method is to
- find both one coordinate and the norm of the remaining coordinates and
- iterate; this method is more efficient than it looks, can be done without
- square roots, but is likely to cost more than the independent normals.
- --
- Herman Rubin, Dept. of Statistics, Purdue Univ., West Lafayette IN47907-1399
- Phone: (317)494-6054
- hrubin@pop.stat.purdue.edu (Internet, bitnet)
- {purdue,pur-ee}!pop.stat!hrubin(UUCP)
-