Hello all,

while calculating the eigenvalues and eigenvectors of a (n x n) Hermitian matrix - let it be called R - I encountered a problem I was not able to solve so far.

According to NRs' proposal, I map the complex matrix at first to a (2n x 2n) real matrix Rr:
Rr = [ (real(R)), (imag(R)); (-imag(R)), (real(R)) ];
Since R is Hermitian, its main diagonal elements must be real-only, so Rr contains some zero-elements, one in each row and in each column.

Unfortunately, with those zeros in the input matrix passed to tred2, tqli's results seem not to comply with Matlab's results, so I guess they're wrong: Although the eigenvalues do always correspond, the eigenvectors differ.

As soon as there are no zero-elements in the input matrix, tred2 and tqli produce the same eigenvectors as Matlab. Did anybody else run across this occurrence?

Thanks in advance,
yours sincerely
Leonhard Brandl, Munich, Germany