Suppose $A$ is an $n\times n$ matrix, and $\lambda_1, \lambda_2, \ldots, \lambda_n$ are its eigenvalues. Prove that

$$ \sum_{i = 1}^n \lvert{A_{ii}} \rvert \leq \sum_{i = 1}^n \lvert \lambda_i\rvert. $$

This should be true because I have verified it on 100,000 $10\times 10$ matrices. However, I failed to prove it since I cannot find a convenient representation of sum of these absolute values.

UPDATE: It is not true for general matrices, but does hold for normal matrices.


This isn't true. Every nilpotent matrix $A$ with a nonzero diagonal element can serve as a counterexample, such as $A=\pmatrix{1&-1\\ 1&-1}$.


This does hold if we assume $A$ is normal

Let $D$ be a unitary diagonal matrix where $d_{i,i}$ has the same polar angle as $\bar{a_{i,i}}$
(and if $a_{i,i}=0$ then set $d_{i,i}=1$)

$\sum_{i = 1}^n \lvert{a_{ii}} \rvert = \text{trace}\Big(DA\Big)\leq \sum_{i=1}^n 1 \cdot \sigma_i = \sum_{i = 1}^n \lvert \lambda_i\rvert$
by the von-Neumann trace inequality and the fact that $\sigma_i=\vert \lambda_i\vert$ for normal matrices.


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.