Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

I just read in a textbook on numerical methods that you can always have that the right eigenvectors of a matrix can be taken as orthonormal to the left eigenvectors for a diagonalisable matrix. This seems suprising to me. I would have assumed that you need to have that the matrix is normal or hermitian, but is this also possible for all matrices?

share|improve this question
Can you be more explicit? What exactly does the book say? –  Brian Fitzpatrick May 17 '14 at 20:21
It says: If you have a matrix $A$ that is diagonalisable with right eigenvectors $\phi_1,...,\phi_n$, you can always find left eigenvectors $\psi_1,...,\psi_n$ such that $\phi_i \psi_j = \delta_{i,j}$ ( this is of course the real case) –  Xin Wang May 17 '14 at 20:23
Is there any hypothesis of distinct eigenvalues? Ah, I guess with geometric multiplicity $>1$ we can choose the eigenvectors to make it true. –  Ted Shifrin May 17 '14 at 20:34
no, there is nothing about distinct eigenvalues –  Xin Wang May 17 '14 at 20:37

1 Answer 1

up vote 4 down vote accepted

I've never seen this before, but it's quite true.

What is a standard exercise is that when $\lambda\ne\mu$, with $x$ a $\lambda$-eigenvector for $A$ and $y$ a $\mu$-eigenvector for $y$, then $x\cdot y = 0$ (compute $Ax\cdot y$ two ways). But it wasn't obvious to me what happened with the $\lambda$ eigenvectors for both.

Consider the diagonalization $P^{-1}AP=\Lambda$. Then, transposing this equation, we obtain $Q^{-1}A^\top Q=\Lambda$, where $Q=(P^{-1})^\top$. The columns of $P$ and $Q$ are eigenvectors of $A$ and $A^\top$, respectively. But let's look at $Q^\top P$: the $ij$-entry will compute the dot product of the $i^{\text{th}}$ eigenvector of $Q$ and the $j^{\text{th}}$ eigenvector of $P$. But $Q^\top P = P^{-1}P = I$, which is the desired result.

share|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.