🚀 go-pugleaf

  I am attempting to find the spectral radius of a larger
matrix (241x241).  It's ~98.5% sparse, not banded and asymmetric.

  Since I'm only after the largest eigenvalues,
my first attempt was a power method.  However, this was divergent
in some cases (in retrospect, I realize that there is no
guarantee that the eigenvalue is real of that it doesn't appear as
both the negative and positive values).

  I then implemented IMSL (routine: DEVLRG), which would not
run to completion because of a large predicted performance index.

  Trying EISPACK (subroutine: RG) gave results that looked reasonable,
but when I transpose the matrix and calculate the eigenvalues of the
transposed matrix (with EISPACK) I get different results than the
untransposed matrix...

  The questions are:
    1) Which method is most "believable"?
    2) As there any recommendations on another algorithm that could
       be attempted?

  Any help/suggestions would be appreciated.

  Thanx in advance.


-- 
                                     |
Michael Zika (zika@ecn.purdue.edu)   |    Hey don't ask me "why?",
Purdue University                    |    I'm still working on "how?" ! 
School of Nuclear Engineering        |