0
$\begingroup$

Suppose $A,E$ are $n\times n$ matrices and $A$ has singular values $\sigma_1\geq \sigma_2 \geq \cdots \geq \sigma_n >0$.

Please help me to prove that $\Vert E \Vert_2 \geq \sigma_n$ if $A+E$ is singular.

Is there any singular matrix $E$ such that $\Vert E \Vert_2 = \sigma_n$ ?

$\endgroup$
  • $\begingroup$ help please. how to prove this? $\endgroup$ – SKMohammadi Jan 10 '15 at 5:30
1
$\begingroup$

If $A+E$ is singular, then $(A+E)x=0$ for some nonzero $x$. Hence $$ x=-A^{-1}Ex\quad\Rightarrow\quad \|x\|_2\leq\underbrace{\|A^{-1}\|_2}_{=\sigma_n^{-1}}\|E\|_2\|x\|_2\quad\Rightarrow\quad\sigma_n\leq\|E\|_2. $$ The matrix $E$ of the minimum norm which makes $E$ singular is (assuming $\sigma_n<\sigma_{n-1}$) $$ E=-\sigma_n u_n v_n^*, $$ where $u_n$ and $v_n$ are left and right singular vectors associated with $\sigma_n$. Clearly, $\|E\|_2=\sigma_n$ and using the SVD of $A$, you can verify that $$A+E=\sum_{i=1}^{n-1}\sigma_i u_iv_i^*$$ is singular.

$\endgroup$
1
$\begingroup$

Hint. For your first question, if $M$ is a complex matrix, then $\|M\|_2=\sigma_1(M)=\max_{\|x\|_2=1}\|Mx\|_2$ and $\sigma_n(M)=\min_{\|x\|_2=1}\|Mx\|_2$ (Courant-Fischer minimax principle). Now, if $A+E$ is singular, $(A+E)x=0$ for some unit vector $x$.

The answer to your second question should be trivial.

$\endgroup$
0
$\begingroup$

For the first question: following where the other answer left off, suppose that we can find an $x$ such that $(A+E)x = 0$ and $\|x\| = 1$. It follows that $$ Ax = -Ex \implies \|Ax\|_2 = \|Ex\|_2 $$ What does this tell us about $\min_{\|x\| = 1} \|Ax\|_2$ and $\max_{\|x\| = 1} \|Ex \|_2$?

$\endgroup$

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.