0
$\begingroup$

I came across the following claim:

The multiplication of an orthogonal matrix $A \in \Bbb R^{m\times n}$ by a vector $x$ with $||x||_2=1$ must have a sum of squared elements (i.e. Frobenius norm) at most $1$.

This can be cast into a clumsy-looking inequality of real numbers: We assume $x_1+...+x_n=1$ and $\sum_i a_{ij}^2=1$ and $\sum _ka_{ij}a_{kj}=0$ for all $k,j,i$. Then we claim $\sum _i (\sum_j a_{ij}x_j)^2 \leq 1$.

Cauchy's inequality only gives me a weak bound (of $m$ instead of $1$), but I saw this version used and it seems true. How can I show this?

$\endgroup$
  • $\begingroup$ An orthogonal matrix respects norms, so $\;\left\|Ax\right\|=\left\|x\right\|\;$ and we're done $\endgroup$ – DonAntonio Feb 19 '17 at 23:38
1
$\begingroup$

It's because $$\|Ax\|_2^2 = x^T A^T A x = x^T x = \|x\|_2^2 = 1.$$ The result always has sum of squared elements exactly $1$.

$\endgroup$
1
$\begingroup$

The result you are looking for is a classical property of orthogonal matrices:

An orthogonal matrix gives an isometry, i.e., preserves the norm:

$$\forall X, \ \ \|AX\|=\|X\|$$

$\endgroup$

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.