3
$\begingroup$

The SVD is defined as:

$$A=U\Sigma V^T$$

So, for each row of the matrix $a_i$, singular vector $u_i, v_i$ and singular value $\sigma_i$

$$a_i = \sigma_i u_i v_i^T$$

with
$$u_i = \frac{1}{\sigma_i}a_iv_i$$

so, my problem is the following....

$$a_i = \sigma_i\frac{1}{\sigma_i}a_i v_i v_i^T$$ It seems to me that: $$a_i \ne a_i v_i v_i^T$$

What's my mistake?

$\endgroup$
  • $\begingroup$ If $u_i$ and $v_i$ are rows of the matrices $U$ and $V$, then $u_i v_i^T$ would be a $1\times n$ matrix multiplied by an $n\times 1$ matrix, resulting in a $1\times1$ matrix, not in an entire row. $\qquad$ $\endgroup$ – Michael Hardy Dec 23 '16 at 17:36
2
$\begingroup$

The mistake is in the first formula. You are saying that each row of $A$ depends only on one singular value, which is not true.

For instance let $$ V=U=\frac1{\sqrt2}\begin{bmatrix}1&1\\1&-1\end{bmatrix},\ \ \ \Sigma=\begin{bmatrix}\sigma_1&0\\0&\sigma_2\end{bmatrix}. $$ Then $$ A=\frac12\,\begin{bmatrix}\sigma_1+\sigma_2&\sigma_1-\sigma_2\\ \sigma_1-\sigma_2&\sigma_1+\sigma_2\end{bmatrix}. $$ You can check that your formula does not apply, as you could fix $\sigma_1$ and let $\sigma_2$ vary to make it fail.

| cite | improve this answer | |
$\endgroup$
2
$\begingroup$

In

$a_i = \sigma_i u_i v_i^T$ with $u_i = \dfrac{1}{\sigma_i}a_iv_i$

the dimensions do not match. Instead, one has

enter image description here

| cite | improve this answer | |
$\endgroup$

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.