Thank you, Mike. I have tested the dgegv you have so kindly provided on my earlier problem using the same matrix A and providing a unit matrix for B. It gives the same answers, and I checked them using Mathcad.
Regards Graham Mike Day wrote: > Message: 3 > Date: Thu, 11 Oct 2007 14:05:30 +0100 > From: Mike Day <[EMAIL PROTECTED]> > Subject: Re: [Jprogramming] Re: Implementation of DGEGV in LAPACK > To: Graham Parkhouse <[EMAIL PROTECTED]> > Cc: Programming forum <[email protected]> > Message-ID: <[EMAIL PROTECTED]> > Content-Type: text/plain; charset="iso-8859-1" > > Graham and J-LA-packers generally > > FWIW, I've had a go at scripting dggev.ijs and zggev.ijs, although you don't > need them at the moment. > > When I got round to testing them, I discovered that they > don't have entry points; ie routines dggev_ and zggev_ > are missing from jlapack.dll. So, pro tem, these scripts > call the "deprecated" routines dgegv_ and zgegv_ . > > I've modelled them on dgeev.ijs and zgeev.ijs currently supplied with the LAPACK > add-on. Just as in those paradigms, > there's no attempt to choose between none, one or both of > the sets of eigenvectors; both left and right sets of generalised eigenvectors > are returned, together with "alpha" and "beta", the generalised eigenvalues. The allotment > of work-space is possibly overgenerous. > > There are a couple of testing verbs, but on very limited > data, found at http://www.nag.co.uk/lapack-ex/node136.html > The results shown there (for right eigenvectors only) > appear consistent with those from the J scripts apart from > sign and rounding. > > I don't know how to incorporate these in the library of > LAPACK scripts, and in any case, they should be properly > tested with suitable data by experts in linear algebra. > > I'm attaching dggev.ijs and zggev.ijs as separate files, > though they might be displayed in line by some mail > readers. > > Apologies in advance for any errors... ---------------------------------------------------------------------- For information about J forums see http://www.jsoftware.com/forums.htm
