13
$\begingroup$

Recall that the absolute value of a bounded operator $T$ on a Hilbert space $H$ is the unique positive operator $|T|$ such that $$\||T|x\|=\|Tx\|$$ for all $x\in H$. It can be defined using the continuous functional calculus, or if you have square-roots of positive operators in hand, by $|T|=(T^*T)^{1/2}$. Likewise, this can be defined in any C*-algebra.

With respect to the usual ordering on self-adjoint elements, i.e., $S\leq T$ if $T-S$ is positive, does this absolute value behave like an absolute value? For instance, does it satisfy a triangle inequality like the one below? $$||T|-|S||\leq|T-S|\leq|T|+|S|$$

If $S$ and $T$ are normal and commute, then perhaps one could use the functional calculus on the commutative C*-algebra they generate to show this, but not for general operators.

How about other inequalities with respect to this ordering?

$\endgroup$
  • 6
    $\begingroup$ The identities you mentioned are not necessarily true. See discussion in Reed and Simon Vol-1, chapter 6, exercises 16 and 17. $\endgroup$ – Piyush Grover Jul 8 '14 at 18:29
10
$\begingroup$

The answer is no. Loewner partial order is subtle. There are some simple examples given in R. Bhatia, F. Kittaneh, The matrix arithmetic–geometric mean inequality revisited, Linear Algebra and its Applications 428 (2008) 2177–2191

| cite | improve this answer | |
$\endgroup$
1
$\begingroup$

In general, if $p_1$, $p_2$ are non-commuting rank $1$ projectors, then $$p_1 + p_2 \not \succeq |p_1 - p_2|$$ Indeed, may assume $p_1 = \left(\matrix{ 1 &0\\0&0}\right)$, $p_2 = \left(\matrix{ \cos^2 \theta &\cos \theta \sin \theta\\\cos \theta \sin \theta&\sin^2 \theta}\right)$, then one checks $(p_1-p_2)^2 = \sin^2\theta \cdot I_2$, and so $$p_1 + p_2 - |p_1 - p_2| = \left(\matrix{ 1+ \cos^2 \theta - |\sin \theta| &\cos \theta \sin \theta\\\sin \theta \cos \theta&\sin^2 \theta - | \sin \theta|}\right)$$

| cite | improve this answer | |
$\endgroup$
0
$\begingroup$

The first inequality for example is false but

https://projecteuclid.org/download/pdf_1/euclid.pja/1195519395

shows that a slightly weaker inequality with logarithmic factors does hold for the spectral radii when the operators are selfadjoint. I don't know if it can be extended to a relaxed inequality for the Loewner order.

| cite | improve this answer | |
$\endgroup$

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.