I have come across a problem which I can't properly solve:

Let $\tau$ be a linear operator on $\mathbb{C}^n$ and let $\lambda_{1},...,\lambda_{n}$ be the eigenvalues of $\tau$, each one written a number of times equal to its algebraic multiplicity. I should show that:

$\sum_{i}|\lambda_{i}|^2\leq tr(\tau^*\tau)$

Also, one should show that the equality holds iff $\tau$ is normal.

First I felt that this might use singular values, but I have no success with this. My idea then was that Cauchy-Schwarz may be useful. (I work with matrices, this is clearly not a restriction to the problem.) So I defined the inner-product $\langle A,B\rangle=tr(B^*A)$, which I know to be acceptable. Elementary operations on Cauchy-Schwarz inequality

$|\langle A,A^*\rangle|^2$$\leq \langle A,A\rangle\langle A^*,A^*\rangle$

then give that $|\sum_{i}\lambda_{i}^2|^2$$\leq (tr(A^*A))^2$ (I may be mistaken). This is not what I want. In the question ''equality holds iff $\tau$ is normal'', one way (right to left) is easy.

I highly appreciate any suggestion!


Let's do a simpler thing.

We have a Schur decomposition $\tau=u^*\sigma u$ with $u$ unitary and $\sigma$ an upper triangular matrix. Since both sides of the inequality take the same value for $\tau$ and for $\sigma$ we may assume that $\tau$ is in fact itself upper triangular. Now the inequality is obvious, because the right hand side is a sum of the left hand side and some other, non-negative terms.

  • $\begingroup$ I fail to follow your argument. For diagonal matrices, the inequality in the question is an equality. Why and how would that extend as a (proper) inequality to more general matrices? $\endgroup$ – Martin Argerami Sep 6 '12 at 1:46
  • $\begingroup$ As an example of where the inequality is strict, consider the matrix representation of the right shift operator $R$, $e_1 \rightarrow e_2 \rightarrow e_3 \rightarrow 0$ and it's adjoint the left shift operator $L=R^*$, $e_3 \rightarrow e_2 \rightarrow e_1 \rightarrow 0$. The composition $RL$ acts as the identity on $e_1$ and $e_2$, and sends $e_3$ to zero, so it has trace 2, whereas there are no eigenvalues of $R$ or $L$. $\endgroup$ – Nick Alger Sep 6 '12 at 6:03
  • $\begingroup$ N.B. I have just edited my answer, and the two comments aobve refer to the old version. $\endgroup$ – Mariano Suárez-Álvarez Sep 6 '12 at 6:27

It wouldn't surprise me if I'm missing something, but I couldn't think of a straightforward proof. But this inequality is a particular case of Weyl's Majorant Theorem (see Theorem II.3.6 in Bhatia's Matrix Analysis, or there are surely many other references).

  • $\begingroup$ I agree with you about my retracted "answer." $\endgroup$ – Kevin Carlson Sep 6 '12 at 5:10
  • $\begingroup$ You are welcome. I couldn't address the implication "left implies right" with this idea. But it does follow easily from Mariano's argument (the case of equality forces $\sigma$ to be diagonal). $\endgroup$ – Martin Argerami Sep 6 '12 at 14:17

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.