[R] Computing generalized eigenvalues

Peter Dalgaard p.dalgaard at biostat.ku.dk
Fri Jun 17 11:48:46 CEST 2005


Prof Brian Ripley <ripley at stats.ox.ac.uk> writes:

> On Thu, 16 Jun 2005, Joshua Gilbert wrote:
> 
> > I need to compute generalized eigenvalues. The eigen function in base
> > doesn't do it and I can't find a package that does.
> 
> They are very rarely used in statistics, so this is not surprising.

An aside, going a bit off-topic:

However, there's the related  generalized singular value decomposition:

 K = U Sigma inv(T) 
 L = V M inv(T)

U'U = V'V = I ;  Sigma and M diagonal (Sigma^2 + M^2 = I by
convention) ; T regular

(Look at K'K and L'L to see the connection.)

This is used in Tikhonov regularization, which is penalized least
squares, which is a statistical issue (whether numerical analysts
realize it or not). Smoothing splines is a special case. Deconvolution
is another. 

 
> I presume you mean solving Ax = lambda B x: if B is non-singular this 
> reduces to a conventional eigenproblem for B^{-1}A.

(There are some complications if both A and B are singular. That's
why the GSVD has that peculiar-looking convention.) 

-- 
   O__  ---- Peter Dalgaard             Øster Farimagsgade 5, Entr.B
  c/ /'_ --- Dept. of Biostatistics     PO Box 2099, 1014 Cph. K
 (*) \(*) -- University of Copenhagen   Denmark          Ph: (+45) 35327918
~~~~~~~~~~ - (p.dalgaard at biostat.ku.dk)                  FAX: (+45) 35327907




More information about the R-help mailing list