Article View: sci.math.num-analysis
Article #93Re: SVD row update
From: Andrew Mullhaupt
Date: Tue, 21 Jul 1992 15:05
Date: Tue, 21 Jul 1992 15:05
18 lines
1275 bytes
1275 bytes
In article <sjreeves.920720112016@fourier.ee.eng.auburn.edu> sjreeves@eng.auburn.edu (Stan Reeves) writes: > >Does anyone know of an efficient method for updating an SVD of a matrix >when a new row is added to the matrix? By "efficient," I mean something >more efficient than just redoing the SVD from scratch. Any help would >be appreciated. When the matrix is long and skinny, (as will eventually obtain when you keep adding rows), you can use the R-SVD and update the initial QR decomposition. You still have to do the little SVD, but the work is dominated (in the limit) by the QR update. So, to the extent that you consider QR updates 'fast' and 'reliable', you can update the SVD. It is at least faster than doing it from scratch. Hanson and Lawson, and Golub and van Loan discuss QR updates. Later, Andrew Mullhaupt
Technical Details
Message-ID:
<1100@kepler1.rentec.com>Path:
arc1-ca.pugleaf.net!sparky!uunet!zaphod.mps.ohio-state.edu!mips!pacbell.com!iggy.GW.Vitalink.COM!psinntp!psinntp!kepler1!andrewReferences:
<sjreeves.920720112016@fourier.ee.eng.auburn.edu>Show Full Headers
Complete Article Headers
Path: arc1-ca.pugleaf.net!sparky!uunet!zaphod.mps.ohio-state.edu!mips!pacbell.com!iggy.GW.Vitalink.COM!psinntp!psinntp!kepler1!andrew From: andrew@rentec.com (Andrew Mullhaupt) Newsgroups: sci.math.num-analysis Subject: Re: SVD row update Message-ID: <1100@kepler1.rentec.com> Date: 21 Jul 92 15:05:05 GMT References: <sjreeves.920720112016@fourier.ee.eng.auburn.edu> Organization: Renaissance Technologies Corp., Setauket, NY. Lines: 19