home *** CD-ROM | disk | FTP | other *** search
- Xref: sparky sci.math.num-analysis:2561 sci.math:10611 sci.math.stat:1759
- Path: sparky!uunet!zaphod.mps.ohio-state.edu!darwin.sura.net!Sirius.dfn.de!chx400!news.unige.ch!ugun2b!ugun2a!pfennige
- Newsgroups: sci.math.num-analysis,sci.math,sci.math.stat
- Subject: Re: Least Squares with Inequality Constraints?
- Message-ID: <1992Aug27.092419.1534@uni2a.unige.ch>
- From: pfennige@uni2a.unige.ch
- Date: 27 Aug 92 09:24:18 +0200
- References: <l9jcdhINNp98@roundup.crhc.uiuc.edu>
- Organization: University of Geneva, Switzerland
- Lines: 39
-
- In article <l9jcdhINNp98@roundup.crhc.uiuc.edu>, hougen@uirvlh.csl.uiuc.edu (Darrell Roy Hougen) writes:
- > I would like to find some code, preferably free and preferably in C,
- > that solves the general problem of linear least squares with
- > linear inequality constraints.
- >
-
- The book "Solving Least Squares Problems" by C.L. Lawson & R.J. Hanson,
- (1974), Prentice-Hall, explains (chap. 23) how to do linear least squares
- inequality contraints (LSI) (mixed with equality):
-
- minimize ||Ex-f|| subject to Gx>=h
-
- where E, G are matrices, f,h vectors, and x the unknown coefficient vector.
- Equalities constraints are equivalent to two inequalities, x>=y and y>=x, and
- unconstrained variables can be set smaller than a large positive number.
- LSI problems can be reduced to least distance problems (LDP):
-
- minimize ||x|| subject to Gx >= h
-
- LDP problems can be reduced to non-negative least squares (NNLS):
-
- minimize ||Ex-f|| subject to x>=0
-
- This book provides Fortran code for LDP and NNLS. Having used the NNLS
- algorithm extensively, I have found it robust and efficient. NNLS can be seen
- as a generalization of linear programming since when no solution exists, NNLS
- provides the nearest one, instead of nothing.
-
- > ...
- >
- > Thanks in advance.
- >
- > Darrell
-
- If anybody knows a more recent general reference about non-negative least
- squares, I would be interested to know about.
-
- Daniel Pfenniger
-
-