Estimate the condition number in the second norm of the matrix An

172 Views Asked by At

Let $A_{n}$ be a matrix of size $n$ for $n \geq 1$, and a structure:

$$ A_{n}=\left[\begin{array}{cccccccc} \sqrt{21} & 1 & 0 & 0 & \ldots & 0 & 0 & 0 \\ 0 & \sqrt{21} & 1 & 0 & \ldots & 0 & 0 & 0 \\ 0 & 0 & \sqrt{21} & 1 & \ldots & 0 & 0 & 0 \\ 0 & 0 & 0 & \sqrt{21} & \ldots & 0 & 0 & 0 \\ \vdots & \vdots & \vdots & \vdots & \ddots & \vdots & \vdots & \vdots \\ 0 & 0 & 0 & 0 & \ldots & \sqrt{21} & 1 & 0 \\ 0 & 0 & 0 & 0 & \ldots & 0 & \sqrt{21} & 1 \\ 0 & 0 & 0 & 0 & \ldots & 0 & 0 & \sqrt{21} \end{array}\right] $$

Estimate the condition number in the second norm of the matrix $A_{n}$.

My solutions is as follows

$$ \|A\|_{2}=\sqrt{\lambda_{\max }\left(A^{*} A\right)}=\sigma_{\max }(A) . $$

where $\sigma_{\max }(A)$ represents the largest singular value of matrix $A$. So

$$ \begin{aligned} &\left\|A_{n}\right\|_{2}=\sqrt{21} \\ &\left\|A_{n}^{-1}\right\|_{2}=\frac{1}{\sqrt{21}} \end{aligned} $$

Condition number

$$ \mu=|| A_{n} \| \cdot|| A_{n}^{-1}||=1 $$

I want to check if this is correct and I'm used to only solving this for symmetric matrices

2

There are 2 best solutions below

3
On

Estimate the condition number in the second nom of matrix $A_{n}$

You can now bound the eigenvalues of this matrix using the Gershgorin circle theorem, which tells you that they must be in

$|\lambda-\sqrt{21}| \leqslant 1 \Rightarrow \lambda \in\langle\sqrt{21}-1, \sqrt{21}+1\rangle$

$\operatorname{cond}_{2}(A)=\sqrt{\frac{\lambda_{\max }\left(A^{\top} A\right)}{\lambda_{\min }\left(A^{\top} A\right)}}$

note:

$\sqrt{\lambda_{\max }\left(A^{T} A\right)}=\sigma_{\max }(A)$

$\sqrt{\lambda_{\min }\left(A^{T} A\right)}=\sigma_{\min }(A)$

$=\frac{\sigma_{\max }(A)}{\sigma \min (A)} \mid \begin{gathered}\sigma_{\operatorname{max}}(A) \text { represents the } \\ \text { largest singular value } \\ \text { of matirx } A \text {. }\end{gathered}$

$=\frac{\sqrt{21}+1}{\sqrt{21}-1}$

0
On

As you indicated, you essentially need to estimate the eigenvalues of $A^*A$.

Fortunately, in this case it is easy to compute $A^*A$ explicitly. $$ A^*A=\left[\begin{array}{cccccccc} 22 & \sqrt{21} & 0 & 0 & \ldots & 0 & 0 & 0 \\ \sqrt{21} &22 & \sqrt{21} & 0 & \ldots & 0 & 0 & 0 \\ 0 & \sqrt{21} &22 & \sqrt{21} & \ldots & 0 & 0 & 0 \\ 0 & 0 & \sqrt{21} &22 & \ldots & 0 & 0 & 0 \\ \vdots & \vdots & \vdots & \vdots & \ddots & \vdots & \vdots & \vdots \\ 0 & 0 & 0 & 0 & \dots &22 & \sqrt{21} & 0 \\ 0 & 0 & 0 & 0 & \ldots & \sqrt{21} &22 & \sqrt{21} \\ 0 & 0 & 0 & 0 & \ldots & 0 & \sqrt{21} &22 \end{array}\right] $$

You can now bound the eigenvalues of this matrix using the Gershgorin circle theorem, which tells you that they must be in $[22-2\sqrt{21},22+2\sqrt{21}]$.

Now, $$ 22-2\sqrt{21} = (\sqrt{21} - 1)^2, \text{ and}\\ 22+2\sqrt{21} = (\sqrt{21} + 1)^2. $$ Therefore, an estimate for the condition number of $A$ is $$ \frac{\sqrt{21} + 1}{\sqrt{21} - 1}. $$