First off, in case anyone is wondering, no, this is not for a homework problem.
Now then, suppose I have a non-Hermitian matrix M. It’ll have eigenvalues w, and right and left side eigenvectors X and Y, so that:
MX=Xw YM=wY YX=1
We also know that I can do a similarity transformation of M so that M’=A[sup]-1[/sup]MA; this will obviously change the eigenvectors (X’=A[sup]-1[/sup]X and Y’=YA), but won’t affect the eigenvalues.
So now suppose that I have real eigenvalues. I would think that there ought to be a similarity transformation that makes M’ Hermitian and hence X’ unitary. If there is such a transformation, how do I go about finding it?
Write your first equation in matrix form: M X = X W, where X is the matrix of eigenvectors (each column is an eigenvector) and W is the diagonal matrix of eigenvalues. If the eigenvectors span the space (note that this is not necessarily true for non-Hermitian matrices), X is invertible, so X[sup]-1[/sup] M X = W is a similarity transformation from M to the real and diagonal (and hence Hermitian) matrix W.
:smack:
Oh, of course. Given that any unitary transformation of W would give me some Hermitian matrix M" with eigenvalues W, that would explain why I can’t find a unique similarity transform to do what I want.
… Er, I’m not sure what you’re saying here. Which matrix do I assume is symmetric? The method I give will work for any non-defective matrix (one whose eigenvectors span the space) with real eigenvalues. Neither the original matrix M nor the eigenvector matrix X need to be real for the diagonalization to work. Can you elaborate on what you mean?