[Math] An inequality related to matrix norm, inverse matrix

linear algebramatricesnormed-spaces

The question is that:

We have two matrices $A,B\in \mathbf{C}^{n\times n}$, A is nonsingular and B is singular, let $||\cdot ||$ be $\textbf{any}$ matrix norm, prove

$||A-B||\geq 1/||A^{-1}||$.

My idea is

$||(A-B)||\cdot||A^{-1}||\geq||A^{-1}(A-B)||=||I-A^{-1}B||$

And I am confused here. How to calculate the right side?

$\color{red}{Note:}$ the original question may be wrong when the norm doesn't have sub-multiplicative property. See the comments below.

Thanks for everybody's help!

Best Answer

HINT:

$B= A+ (B-A) = A\cdot( I - A^{-1}(A-B) ) = A \cdot (I - C)$

where $C = A^{-1}(A-B)$. Assume that $||A-B||< ||A^{-1}||^{-1}$. Then $||C||\le ||A^{-1} || \cdot ||A-B|| < 1$. Now use the following fact:

If $C$ is a matrix with norm $||C||<1$ then the series $\sum_{n\ge 0} C^n$ is absolutely convergent and its sum is the inverse of $I-C$. In particular: $||C||<1$ implies $(I-C)$ invertible.

Use the above to conclude $B$ is invertible, contradiction.

Related Question