Given an $n \times n$ normal matrix $A$ over $\mathbb{R}$, show that there is an orthogonal matrix $O$ such that $O^TAO$ is either diagonal or in the form


So far, my thoughts are as follows. The characteristic polynomial $f$ of $A$ is cubic, so it has at least one real root, call it $\lambda$ with corresponding eigenvector $V_\lambda$, and $f = (x-\lambda)q$ for some polynomial $q$ of degree 2.

If $q$ has distinct real roots, then $A$ has an orthonormal basis of eigenvectors, which give the columns of a matrix $O$ such that $O^{-1}AO = O^TAO$ is diagonal.

Otherwise, ...? Essentially all of the theorems and exercises in my text work over $\mathbb{C}$, not $\mathbb{R}$. If we were working over $\mathbb{C}$, we would always have an orthonormal basis of eigenvectors $A$ would always be diagonalizable by such a matrix $O$.


You need to work out the details that the $2$-dimensional invariant subspace corresponding to eigenvalues $\lambda,\bar\lambda$ for a complex (non-real) eigenvalue $\lambda$, with the appropriate basis, gives the matrix $\begin{bmatrix} a&b\\-b&a\end{bmatrix}$ when $\lambda = a+ib$. (Hint: If $v\in\Bbb C^3$ is an eigenvector with eigenvalue $\lambda$ of the complexification of $A$, write $v=x+iy$ for $x,y\in\Bbb R^3$.)

  • $\begingroup$ Following this hint, $A$ has an orthonormal basis of eigenvectors over $\mathbb{C}^3$ with one real eigenvalue $\lambda_R$ and corresponding eigenvector $v_R$. Let $\lambda=a+bi$, $v = x+yi$; $Av = \lambda v \implies Au=au-bv$, $Av=av+bu$. Taking ordered basis $\mathcal{B}=\{v_R,u,v\}$ and using it as colunns of $O$, we find that the matrix of $A$ in$\mathcal{B}$ is $\left[\begin{matrix}{\lambda}&0&0\\0&a&b\\0&-b&a\end{matrix}\right]$. Is this correct? $\endgroup$ – user169845 Jan 4 '15 at 22:10
  • $\begingroup$ Looks good. You should stop to think about how the unitary basis for $\Bbb C^3$ translates to a basis for $\Bbb R^3$. If you follow this, do you have a (real) orthonormal basis? $\endgroup$ – Ted Shifrin Jan 4 '15 at 22:26
  • $\begingroup$ @user169845: Remember that you need to check that $u,v$ are linearly independent over the reals. $\endgroup$ – copper.hat Jan 4 '15 at 23:49

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.