I've been learning about unitary diagonalization and this question has been on my mind for a while, unanswered by the text I'm using.

Would it be true to say that if a square matrix represents an involution, i.e. $A^2=I$ (or $A=A^{-1}$), then it is Hermitian ($A=A^*$) and Unitary ($A^{-1}=A^*$). The reverse implication is definitely true, but is it an equivalence?

If so, then why? And if not, could it be true for real matrices? (Involution $\Rightarrow$ Symmetric & Orthogonal)

Thanks for help!

  • $\begingroup$ Assuming $A$ is square, yes. $\endgroup$ – Tapu Mar 27 '13 at 8:01
  • $\begingroup$ Oh, yes, sorry, I added that to the question. Thanks for clearing that up, still, why would that be true? $\endgroup$ – Dahn Mar 27 '13 at 8:04
  • $\begingroup$ In the real case, I'd use the term orthogonal rather than unitary. $\endgroup$ – Sammy Black Mar 27 '13 at 8:12
  • $\begingroup$ Thanks, the sudden introduction of all the new names for matrices still confuses me a little. $\endgroup$ – Dahn Mar 27 '13 at 8:14
  • 1
    $\begingroup$ If $A$ satisfies the condition, then any matrix similar to it has this property. Hence, I think it is not true. $\endgroup$ – Wayson Kong Mar 27 '13 at 8:26

First, let $A \in M_n(\mathbb{C})$, and suppose that $A^2 = I$. I claim that $A$ is a self-adjoint unitary if and only if $A$ is normal. On the one hand, if $A$ is self-adjoint and unitary, then it is trivially normal. On the other hand, suppose that $A$ is normal. Then by the finite-dimensional spectral theorem, $A = U \Delta U^\ast$ for $\Delta$ a diagonal matrix (i.e., with diagonal entries the eigenvalues of $A$ counted with multiplicity), and $U$ a unitary. Since $A^2 = I$, it therefore follows that $\Delta^2 = I$, so that the diagonal entries of $\Delta$ (viz, eigenvalues of $A$) are $\pm 1$. Thus, $\Delta$ is a self-adjoint unitary, so that since $A = U \Delta U^\ast$ for $U$ unitary, $A$ too is a self-adjoint unitary.

Let us now see that the normality condition is not redundant. Consider $$ A = \begin{pmatrix} 1 & 1\\ 0 & -1\end{pmatrix}, $$ and observe that $$ A^2 = \begin{pmatrix} 1 & 1\\ 0 & -1\end{pmatrix}\begin{pmatrix} 1 & 1\\ 0 & -1\end{pmatrix} = \begin{pmatrix} 1 & 0 \\ 0 & 1 \end{pmatrix} = I_2. $$ However, $A$ is neither self-adjoint nor unitary, and indeed $$ A A^\ast = \begin{pmatrix} 1 & 1\\ 0 & -1\end{pmatrix}\begin{pmatrix} 1 & 0\\ 1 & -1\end{pmatrix} = \begin{pmatrix} 2 & -1 \\ -1 & 1 \end{pmatrix},\\ A^\ast A = \begin{pmatrix} 1 & 0\\ 1 & -1\end{pmatrix}\begin{pmatrix} 1 & 1\\ 0 & -1\end{pmatrix} = \begin{pmatrix} 1 & 1\\ 1 & 2 \end{pmatrix} \neq A A^\ast, $$ as expected.

Let us now consider $A \in M_n(\mathbb{R})$, and suppose that $A^2 = I$. I claim that $A$ is symmetric and orthogonal if and only if $A^T A = A A^T$. By viewing $A$ as a complex matrix, we see that $A$ is self-adjoint and unitary if and only if $A$ is normal; since $A$ is real, this means that $A$ is symmetric and orthogonal if and only if $A^T A = A A^T$. Once more, the explicit example above shows that the "normality" condition is not redundant.

  • $\begingroup$ Thanks a lot, an amazing answer! $\endgroup$ – Dahn Mar 27 '13 at 8:46

If $A^2=I$, you know that $A$ is diagonalizable and its eigenvalues are $\pm 1$. To find a counterexample, an idea is to take $A= P \left( \begin{array}{cc} 1 & 0 \\ 0 & -1 \end{array} \right) P^{-1}$, where $P$ is not orthogonal. For example, $P= \left( \begin{array}{cc} 1 & 2 \\ 0 & 1 \end{array} \right)$.

And indeed, $A= \left( \begin{array}{cc} 1 & -4 \\ 0 & -1 \end{array} \right)$ is a counterexample.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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