3
$\begingroup$

Let $E$ be a complex Hilbert space. Let $T_1,T_2\in \mathcal{L}(E)$. Let \begin{align} W_{\max}(T_1,T_2) =\big\{ (\lambda_1,\lambda_2)\in \mathbb{C}^2; & \;\exists\,(x_n)_n;\;\|x_n\|=1,\;(\langle T_1 x_n\; ,\;x_n\rangle,\,\langle T_2 x_n\; ,\;x_n\rangle)\to (\lambda_1,\lambda_2),\\ & \text{ and }\displaystyle\lim_{n\rightarrow+\infty}(\|T_1x_n\|^2+\|T_2x_n\|^2)\rightarrow \|T_1\|^2+\|T_2\|^2\;\big\}. \end{align}

Why $W_{max}(T_1,T_2)$ is not convex?

$\endgroup$

1 Answer 1

4
$\begingroup$

Consider the matrices $$ T_1 = \left[\begin{array}{ccc} 1&0&0 \\ 0& 0&0 \\0&1&0 \end{array}\right] \ \ \textrm{and} \ \ T_2 = \left[\begin{array}{ccc} 0&0&0 \\ 1& 0&0 \\0&1&0 \end{array}\right]. $$ Note that because we are working in finite-dimensional Hilbert space we do not need the limits in the definition of $W_\max(T_1,T_2)$.

Any unit vector $x \in \mathbb C^3$ that norms both $T_1$ and $T_2$ will be of the form $x = \left( \begin{smallmatrix}a\\ b\\ 0\end{smallmatrix}\right)$ where $|a|^2 + |b|^2 = 1$. Hence, $$ (\langle T_1 x \ | \ x\rangle, \langle T_2 x \ | \ x\rangle) = (|a|^2, a\overline b) $$ and we see that $(1,0)$ and $(0,0)$ are in $W_\max(T_1,T_2)$. If $|a| = 1/2$ then $|b| = \sqrt 3/2$ and so $(|a|^2, a\overline b) = (1/4, \lambda \sqrt 3/4)$ for some $|\lambda| = 1$. Thus, it is clear that $(1/4, 0) \notin W_\max(T_1,T_2)$ even though $(1/4,0) = 1/4(1,0) + 3/4(0,0)$. Therefore $W_\max(T_1,T_2)$ is not convex.

$\endgroup$
0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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