I've been staring at this proof for a long time so any suggestions would be of great help! Prove that for any $m\times n$ matrix $A$ there is an orthonormal Basis $B =\{ v_1,\ldots,v_n\}$ of $\mathbb R^n$ such that the vectors $A v_1,\ldots,A v_n$ are orthogonal. Note that some of the vectors $A v_i$ may be zero.


Here's a suggestion: Suppose you found those vectors. Then one must have $Av_j \cdot Av_1 = 0$ for all $j = 2, \ldots, n$. That can also be written as $v_j \cdot A^TAv_1=0$ for all $j = 2, \ldots, n$. Since $A^TAv_1$ must be orthogonal to all the other vectors in the basis, it follows that $A^TAv_1$ must be a multiple of $v_1$ (possibly 0). That makes $v_1$ an eigenvector of $A^TA$. Same with the other vectors $v_2, \ldots, v_n$.

Now use the fact that $A^TA$ is a symmetric matrix to conclude that it has an orthonormal eigenbasis.

(The argument above is used in finding the Singular Value Decomposition of $A$)

  • $\begingroup$ To user330970: If my answer is not clear enough for you to accept it, please let me know what else should I add. Thank you. $\endgroup$ Nov 30 '16 at 21:07

We know that $A^tA$ is symmetric (you can easily verify this.)

Then $A^tA$ must be orthogonally diagonalizable, i.e., there is an orthonormal basis $B = \{v_1, ..., v_n\}$ of $\mathbb R^n$ that consists of normalized, pairwise-orthogonal, eigenvectors of $A^tA$, by the Spectral Theorem.

Now we return to the action of $A$ and see that, for any $i,j$, the inner product is:

\begin{align*} \langle A v_i ,A v_j \rangle &= \langle A^tA v_i, v_j \rangle \\ &= \langle \lambda_i v_i, v_j \rangle\\ &= \lambda_i \langle v_i, v_j \rangle \\ &=\lambda_i \cdot 0 \\ &= 0 \end{align*}

which proves the result.

  • 1
    $\begingroup$ This is almost correct. You should be using $A^T A$ instead, so the $v_i$ are in $\mathbb R^n$. Then your inner-product equations go through. Also, for formatting, use \langle and \rangle instead of < and >. $\endgroup$ Apr 14 '16 at 3:28
  • $\begingroup$ Ah, right @JonWarneke. I have edited my answer just now. Thanks so much for the suggestions :-) $\endgroup$
    – User001
    Apr 14 '16 at 3:42

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.