home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!paladin.american.edu!gatech!rpi!uwm.edu!psuvax1!psuvm!hdk
- Organization: Penn State University
- Date: Fri, 8 Jan 1993 12:54:09 EST
- From: H. D. Knoble <HDK@psuvm.psu.edu>
- Message-ID: <93008.125409HDK@psuvm.psu.edu>
- Newsgroups: comp.lang.fortran
- Subject: Re: inverse matrix
- References: <C0I49C.Jrr@athena.cs.uga.edu>
- Lines: 15
-
- Regardless of the kind of matrix that is being dealt with, the code
- (e.g., IMSL, NAG) should check for computational singularity (or in
- the case of generalized inverse check that the generalized inverse is
- also a least squares inverse and set an appropriate flag). The code
- should also effect as accurate an inverse as feasible on each platform
- it is run on. Else one can get an "inverse" that is just plain wrong
- too easily.
-
- If this matrix inversion is for an important project, it might be wise
- to stay away from Numerical Recipes algorithms (and textbook code
- in general with a few exceptions) as they are not usually robust in
- the above sense. What this amounts to is that good algorithms are most often
- more than taking the algebra and applying FORmula TRANslation to it.
- So know WHO you're getting code from, and what algorithm(s) are being
- used.
-