Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

$A = ({a_{ij}})$ and $B = ({b_{ij}})$ are $n \times n$ symmetric real matrixes whose eigenvalue are ${\lambda _1} \le {\lambda _2} \le \cdots \le {\lambda _n}$ and ${\mu _1} \le {\mu _2} \le \cdots \le {\mu _n}$ respectively. If $\left| {{a_{ij}} - {b_{ij}}} \right| \le \varepsilon $, please show that $\left| {{\lambda _i} - {\mu _i}} \right| \le n\varepsilon $.

share|improve this question
add comment

1 Answer

up vote 10 down vote accepted

The max-norm of the "perturbation" matrix $P := A-B$ is given to be at most $\varepsilon$. For a square matrix, the max-norm is related to the operator norm as $$ \| P \|_{\max} \leqslant \| P \|_{\mathrm{op}} \leqslant n\| P \|_{\max}. $$ Therefore, the operator norm of $P$ is at most $n \varepsilon$; i.e., all its eigenvalues of $P$ are in the interval $[-n \varepsilon, +n \varepsilon]$.

With a bound on the operator norm of $P = A-B$ at hand, Weyl's inequality in matrix theory says that the eigenvalues of $A$ and $B$ cannot be too far apart. More precisely, the inequality states that for all $i$, the difference $\lambda_i - \mu_i$ is in between the smallest and largest eigenvalue of $P$; i.e., $$ - n \varepsilon \leqslant \lambda_i - \mu_i \leqslant n\varepsilon, $$ which is what we wanted to show.

share|improve this answer
    
A lemma may be help for the proof of Weyl's inequality: ${\lambda _{n - j}} = \mathop {\min }\limits_{{v_1}, \cdots {v_{j - 1}} \in {\mathbb R^n}} \mathop {\max }\limits_{(x,{v_1}) = \cdots = (x,{v_{j - 1}}) = 0,{\kern 1pt} {\kern 1pt} (x,x) = 1} x'Ax$ –  Hezudao Dec 2 '11 at 8:26
    
@Adterram Do you want me to expand my answer with a proof of Weyl's inequality? –  Srivatsan Dec 2 '11 at 17:32
    
No,thanks. I just came across the above formula and it can lead to a proof of Weyl's inequality. –  Hezudao Dec 3 '11 at 4:17
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.