Let $A$ be a diagonalizable $n\times n$ matrix with eigenvalues $\lambda_1,\dots, \lambda_n$, $B$ an $n\times n$ matrix, and $\lambda$ an eigenvalue of $A+B$. Show that $$\min\limits_{j=1,\dots,n}|\lambda-\lambda_j|\le ||C||_p||C^{-1}||_p||B||_p$$ where $C$ is a nonsingular matrix such that $C^{-1}AC$ is diagonal and $p=1,2,\infty$.

I'm having difficulty figuring out where to start. If given some guidance I'm sure I can easily get the rest. I know that under the assumption $A$ is diagonalizable gives $C^{-1}AC=diag(\lambda_1,\dots,\lambda_n)$, but I am failing to see how I will use the other assumptions. Any input would be greatly appreciated!

  • 1
    $\begingroup$ What is the context of this question? Is exercise contained any book? Or is it a problem whose solution contributes to some of your research? How do you know it's not a false statement? Please clarify these questions. $\endgroup$ – MathOverview Feb 2 at 14:35
  • $\begingroup$ It is a problem in the book Numerical Analysis by Rainer Kress $\endgroup$ – user906357 Feb 2 at 14:39

Here's an argument for the $p=\infty$ case.

Suppose that $(A+B)v=\lambda v$. Since $C^{-1}AC$ is diagonal, we have $C^{-1}ACe_j=\lambda_je_j$, where $e_1,\ldots,e_n$ are the canonical basis. So $Av_j=\lambda_jv_j$, where $v_j=Ce_j$.

Assume $A$ is invertible (when $A$ is not invertible, we may tweak the eigenvalues slightly so that it is, and the we approximate the estimates). Then $v_1,\ldots,v_n$ are a basis. Write $v=\sum_jr_jv_j$ for some coefficients $r_j$. Now \begin{align} Bv&=\lambda v-Av=\sum_j r_j \lambda v_j-\sum_jr_j Av_j=\sum_jr_j(\lambda-\lambda_j)v_j\\ \ \\ &=\sum_jr_j(\lambda-\lambda_j)Ce_j. \end{align} Thus $$\tag1 \sum_j r_j(\lambda-\lambda_j)e_j=C^{-1}Bv. $$ Since $v\ne0$, $r_k=\max\{|r_j|:\ j\}\ne0$. Then $$\tag2 |r_k|\,|\lambda-\lambda_k|\leq\left\|\sum_jr_j(\lambda-\lambda_j)e_j\right\|_\infty=\|C^{-1}Bv\| $$ Also, since $v=\sum_jr_jv_j=\sum_jr_jCe_j$, we have that $$\tag3 \|v\|=\left\| C\,\sum_jr_je_j\right\|\leq\|C\|\,\left\| \sum_jr_je_j\right\| _\infty=\|C\|\,|r_k| $$ From $(2)$ and $(3)$, $$ |\lambda-\lambda_k|\leq\frac1{|r_k|}\,\|C^{-1}Bv\|\leq\frac1{|r_k|}\,\|C^{-1}\|\,\|B\|\,\|v\|\leq\|C^{-1}\|\,\|B\|\,\|C\|. $$

  • $\begingroup$ Thank you for the answer. How is it we know $|r_k|\ge1$? $\endgroup$ – user906357 Feb 3 at 2:57
  • $\begingroup$ We don't. Look at $(3) $. $\endgroup$ – Martin Argerami Feb 3 at 2:58

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.