Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Here's my feeble attempt:

A = A* is the condition for a Hermitian matrix. So I try expanding both sides in terms of their SVD:

A = USV* and A* = (USV*)* = ... = VSU*

So equating I get:


Now maybe through this I can conclude something about how V and U relate to each other but I don't know how...

Thanks for any help.

share|cite|improve this question
It will help to conclude that $U=V$. What happens if you multiply $VSU^*$ by $USV^*$? Note that order will matter, so consider both. – Alex R. Nov 7 '12 at 6:47

A more efficient route is to consider the Schur decomposition, $A=QTQ^*$, where $T$ is upper triangular and $Q$ is unitary. Note that the diagonal of $T$ consists of the eigenvalues of $A$. The equality $A^*=A$ is $QTQ^*=QT^*Q^*$, from where we deduce $T=T^*$. As $T$ is upper triangular and $T^*$ is lower triangular, we get that $T$ is diagonal. Moreover, as the diagonal of $T^*$ is the conjugate of that of $T$ and they are equal, we have that the eigenvalues of $A$ are real.

The eigenspaces of $T$ are orthogonal to each other (easy to see since $T$ is diagonal). But $Q$ takes the eigenspaces of $T$ to the eigenspaces of $A$, so these have to be pairwise orthogonal too.

share|cite|improve this answer

The process is fairly straightforward given the fact that any square matrix has at least one eigenvalue and eigenvector.

Suppose $Ax = \lambda x$. Then $\langle x , Ax \rangle = \langle Ax , x \rangle = \overline{\langle x , Ax \rangle} = \lambda \|x\|^2$, hence $\lambda \in \mathbb{R}$.

The point about being Hermitian is that if $x$ is an eigenvector of $A$, then both $\text{sp} \{x\}$ and the subspace $\{x\}^\bot$ are invariant under $A$. To see the latter, suppose $v \bot x$, then $\langle x , Av \rangle = \langle Ax , v \rangle = \lambda \langle x , v \rangle = 0$, hence $A v \bot x$. Now let $v_k$ be a basis for $\{x\}^\bot$, then in the basis $x, v_1,...,v_k$, $A$ must have a block diagonal form: $$ A \sim \begin{bmatrix} \lambda & \\ & \tilde{A}\end{bmatrix}$$ (By $\sim$ I mean that the two matrices are similar.) Now find an eigenvalue of $\tilde{A}$ (which must also be Hermitian) and repeat the process until $A$ is diagonalized.

share|cite|improve this answer
Nice, you made that seem easier than previous proofs I've read. – littleO Nov 7 '12 at 8:11
Thanks! Well, I skipped some details like writing out the similarity transformation that reduces $A$ to block diagonal form. However, I personally find the details more straightforward when I have the 'big picture'. – copper.hat Nov 7 '12 at 8:19
Yeah, you made a nice choice of which details to skip, in order to show how simple the proof really is. – littleO Nov 7 '12 at 8:31

Use the eigenvalue decomposition of the hermitian matrix $A$.

$A=S \Lambda S^{-1}$

$ A^H = (S \Lambda S^{-1})^{H} $

$ A^H = ( S^{-1})^{H} \Lambda^H S^H$

Since all eigenvalues for a hermitian matrix are real, $\Lambda = \Lambda^H$, and since $A$ is hermitian we have

$S \Lambda S^{-1} = ( S^{-1})^{H} \Lambda S^H$

This shows $S^{-1}=S^H$ implying $SS^H=I$ as required.

share|cite|improve this answer
This assumes $S^{-1}$ exists. Not a general proof. – Parag Dighe Dec 11 '13 at 9:40

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.