0
$\begingroup$

If A is a $3$ by $2$ matrix,if we do the singular value decomposition (SVD) to A,that is

$A= \begin{bmatrix} \vec u_1 & \vec u_2 &\vec u_3 \end{bmatrix} \begin{bmatrix} s_1 & 0 \\ 0 & s_2 \\ 0 & 0 \\ \end{bmatrix} \begin{bmatrix} \vec v_1 \\ \vec v_2 \\ \end{bmatrix} $

$\vec u_1$ , $\vec u_2$ and $\vec u_3$ are both a column vector,that is,$3$ by $1$ vector

$\vec v_1$ and $\vec v_2$ are both a row vector,that is,$1$ by $2$ vector

$s_1$ and $s_2$ are both singular value,and $s_1>s_2$.

Now ,why must $|A\vec v_1|^2$ is the highest value ?and why must $|A\vec v_1|^2$ > $|A\vec v_2|^2$?,Is there any theory can prove it?

$|f|^2=f^Hf$,if the f is a $N$ by $1$ vector

$\endgroup$
1
$\begingroup$

It is not entirely clear from your question, but it looks like you're trying to prove that $$ \max_{x \in \Bbb C^2, |x| = 1} |Ax| = s_1. $$ To that end: we have $A = U \Sigma V^T$. Note the following:

  • For $x \in \Bbb C^2$, $|Ax| = |U (\Sigma V^Tx)| = |\Sigma V^Tx|$.
  • For any vector $x$, $y = V^Tx$ is a vector of the same magnitude.

We can therefore state that $$ \max_{x \in \Bbb C^2, |x| = 1} |Ax| = \max_{x \in \Bbb C^2, |x| = 1} |\Sigma (V^Tx)| = \max_{y \in \Bbb C^2, |y| = 1} |\Sigma y|. $$ From here, it's easy to prove the desired result by noting that if $y = (y_1,y_2)^T$, we have $$ |\Sigma y|^2 = s_1 |y_1|^2 + s_2 |y_2|^2 . $$

$\endgroup$
  • $\begingroup$ Why is $ |U (\Sigma V^Tx)| = |\Sigma V^Tx|$? $\endgroup$ – electronic component Nov 3 at 0:39
  • $\begingroup$ what does your objected function mean? $\endgroup$ – electronic component Nov 3 at 5:58
  • $\begingroup$ @electroniccomponent the columns of $U$ are orthonormal. So, for any vector $z \in \Bbb C^3$, we have $$ |Uz|^2 = |z_1 u_1 + z_2 u_2 + z_3 u_3|^2 = |z_1|^2 + |z_2|^2 + |z_3|^2 = |z|^2 $$ $\endgroup$ – Omnomnomnom Nov 3 at 7:38
  • $\begingroup$ $\max_{x \in \Bbb R^2, |x| = 1} |Ax|$ means "maximize $|Ax|$ over all $x \in \Bbb C^2$ subject to the constraint that $|x| = 1$" $\endgroup$ – Omnomnomnom Nov 3 at 7:38

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.