hello

i have to compute eigenvalues and eigenvectors of a symmetric matrix. for this purpos i am using the subroutines tred2 and tqli respectively. everzthing is working fine.

however for the problem i have to solve i get the error message:

too many iterations in tqli.

now the question. what's the reason for this message? i.e matrix to big, not symmetric...

and why this message comes for iter>30 seems quite arbitrary to me.

is it possible that tqli can only be used for matrixes up to a cetrain size???

thanks for any kind of suggestions

ueli

I hope that bellow words will be useful for you.

See thread under the title:
Possible convergence problems in svdcmp, jacobi, tqli, hqr
by Saul Teukolsky
in Forum: Official Bug Reports
with known bugs

May be this is a reason of slow convergency.

It is good check, that matrix is symmetric and to work with double accuracy.

I have known also versions with increased number of iterations (200 for example). But I think that this experimental number is right in any case: if you have not convergency for 30 iterations, there is no convergency at all.

SVD method used in book is an intrinsic iterative procedure,
30 iterations is a good number to convergency up to numerical accuracy.

Evgeny

thanks for your fast reply i will check it out!!!