home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: sci.math
- Path: sparky!uunet!zaphod.mps.ohio-state.edu!darwin.sura.net!haven.umd.edu!decuac!pa.dec.com!oct17.dfe.dec.com!ryn.mro4.dec.com!gauss.enet.dec.com!roth
- From: roth@gauss.enet.dec.com (Jim Roth)
- Subject: Re: Bounds on Quadratic Forms ?
- Message-ID: <1992Nov18.233826.14983@ryn.mro4.dec.com>
- Sender: news@ryn.mro4.dec.com (USENET News System)
- Organization: Digital Equipment Corporation
- Date: 18 NOV 92 18:35:19
- Lines: 29
-
-
- In article <1992Nov5.021547.1@cc.newcastle.edu.au>, eemn@cc.newcastle.edu.au (Brett Ninness) writes...
-
- >I have a quadratic form f(x) = x^T A x where x is a positive
- >semidefinite matrix. I want to know the maximum value
- >that f(x) can attain subject to a norm constraint on x.
- >Now I know that if the norm I choose is the 2 norm, then
- >I can use the Rayleigh-Ritz theorm to bound via
- >the maximum eigenvalue of A. But I want to use the
- >infinity norm. I know that if I bound the infinity norm
- >then I have a bound on the 2 norm and again I
- >can use Rayleigh-Ritz, but I want something tighter than
- >this.
-
- You could test all 2^N cases but I don't think that is too efficient :-)
-
- This sounds like it's related to a problem in the geometry of
- numbers of finding the shortest vector in a lattice, where it
- can be handled by a sequence of unimodular transformations in a pretty
- efficient manner. See, for example, Knuth, Vol II, _The Art of Computer
- Programming_, the secton on the spectral test for multiplicative
- congruential random number generators. Another referece is
- Zasenhaus & Post _Algorithmic Algebraic Number Theory_ which has a
- section on the subject. I have a few papers at home on efficient
- algorithms but don't recall the exact citations off the top of my
- head. I don't think the last word has been spoken on such algorithms,
- but the ones out there are not bad.
-
- - Jim
-