While studying Linear Algebra, I encountered the following exercise:


$$A = \begin{bmatrix} 0 & 1 \\ 1 & 0\end{bmatrix}$$

Write $A$ as a sum $$\lambda_{1} u_{1}{u_{1}}^T + \lambda_{2} u_{2}{u_{2}}^T$$ where $\lambda_1$ and $\lambda_2$ are eigenvalues and $u_1$ and $u_2$ are orthonormal eigenvectors.

So what I did is, I computed the eigenvalues: $1$ and $-1$. Then I computed the eigenvectors: \begin{bmatrix} 1\\1 \end{bmatrix}

and \begin{bmatrix} -1\\1 \end{bmatrix}.

But, these vectors are not an orthonormal set, thus I used Gram-Schmidt to obtain an orthonormal set of eigenvectors. This gave me: $u_1 = 1/\sqrt{2}\left( \begin{smallmatrix} 1\\ 1 \end{smallmatrix} \right)$ and $u_2 = 1/\sqrt{2}\left( \begin{smallmatrix} -1\\ 1 \end{smallmatrix} \right)$.

This outcome seems okay, however, I am wondering whether there is a faster method to do this?

  • $\begingroup$ I think you made a typo in $A$. $\endgroup$ – Gonzalo Benavides Jun 22 '18 at 18:17
  • $\begingroup$ Yup lol corrected it! $\endgroup$ – bladiebla Jun 22 '18 at 18:18
  • $\begingroup$ Why using Gram-Schmidt? You just normalized them, since they were already orthogonal. Moreover, if you orthogonalize vectors, in general, they will not be eigenvectors anymore. Hence, that is a bad ansatz. $\endgroup$ – Mundron Schmidt Jun 22 '18 at 18:22

The vectors $(1,1)^T$ and $(-1,1)^T$ are orthogonal, so you just had to normalize them (divide them by their norm) to get an orthonormal set.

  • $\begingroup$ Ah yeah makes sense, still studying the topic so didn't think about that. Thanks! $\endgroup$ – bladiebla Jun 22 '18 at 18:45

Recall that a set of vectors is defined orthogonal when

  • $u_i \cdot u_j=0$ for $i\neq j$

and it is defined orthonormal when also

  • $u_i \cdot u_j=1$ for $i=j$

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.