3
$\begingroup$

Assume $A, B$ are self-ajoint compact operators. Is it true that $\|A+iB\|\le \|2A+iB\|$? Do we have a stronger inequality $\prod_{k=1}^ns_k(A+iB)\le \prod_{k=1}^ns_k(2A+iB)$ or even stronger one $s_n(A+iB)\le s_n(2A+iB)$, $n=1, 2, \ldots$, where $s_n$ are s-numbers?

$\endgroup$
4
  • $\begingroup$ Even the first inequality doesn't seem to hold - take $2A + iB = 0$ for non-zero $A$. $\endgroup$
    – robibok
    Dec 27, 2012 at 15:40
  • $\begingroup$ they are self-ajoint $\endgroup$
    – Betrand
    Dec 27, 2012 at 15:48
  • 1
    $\begingroup$ could you add some motivation behind the first inequality? $\endgroup$
    – Suvrit
    Dec 27, 2012 at 19:54
  • $\begingroup$ A motivation is from the scalar case, as in this article math.pku.edu.cn/teachers/yaoy/Fall2011/Fan_Hoffman1955.pdf $\endgroup$
    – Betrand
    Dec 27, 2012 at 22:17

2 Answers 2

4
$\begingroup$

I just tried a few random matrices... $$ \begin{align} &A=\begin{bmatrix} -0.1 & -0.4\cr -0.4&0\end{bmatrix}, \qquad B=\begin{bmatrix} 1.5 & -0.5 + i\cr -0.5 - i& 3.5\end{bmatrix}, \cr &\|A+iB\| = \frac{7}{2}+\frac{\sqrt{61}}{10}\approx 4.28 \cr &\|2A+iB\| = \frac{7}{2}+\frac{\sqrt{29}}{10}\approx 4.04 \end{align} $$

$\endgroup$
4
$\begingroup$
  1. $s_n(A+iB) \not\le s_n(2A+iB)$, for example by setting \begin{equation*} A = \begin{bmatrix} -2 & -6\\\\ -6 & -2 \end{bmatrix}\qquad B = \begin{bmatrix} 10 & 4\\\\ 4 & 16 \end{bmatrix}. \end{equation*} In this case, $s_2(A+iB) = 7.7192$ and $s_2(2A+iB) = 4.9433$.

  2. $\prod_{k=1}^n s_k(A+iB) \not\le \prod_{k=1}^n s_k(2A+iB)$ for the same example as above. The lhs is 156, while the rhs is 129.2440.

  3. The operator norm version also does not hold (as shown in the nice counterexample by Gerald Edgar)


EDIT

If, however, $A$ and $B$ are assumed to be positive definite, then these inequalities probably hold. As a hint why they might hold (I have not had the time to check any of the other cases), consider $C=A+iB$ and $D=2A+iB$ with $A,B \ge 0$. Then,

\begin{equation*} \begin{split} \prod_{j=1}^n s_j(C) = |\det C| &= \det(A)\prod_{j=1}^n[1+ s_j(A^{-1/2}BA^{-1/2})^2]^{1/2}\\\\ \prod_{j=1}^n s_j(D) = |\det D| &= \det(A)\prod_{j=1}^n2\left[1+ \frac{s_j(A^{-1/2}BA^{-1/2})^2}{4}\right]^{1/2}\\\\ &= \det(A)\prod_{j=1}^n[4+ s_j(A^{-1/2}BA^{-1/2})^2]^{1/2} \ge |\det(C)|. \end{split} \end{equation*}

$\endgroup$
3
  • $\begingroup$ What if $A, B$ are positive definite? $\endgroup$
    – Betrand
    Dec 27, 2012 at 22:14
  • $\begingroup$ Thanks, but how do you define determinant? If $A, B$ are positive definite matrices, we have $|\det(2A+iB)|\ge |\det(A+iB)|$; see Lemma 5 of Kh. D. Ikramov, Determinantal inequalities for accretive-dissipative matrices, J. Math. Sci. (N. Y.), 121(2004) 2458-2464. $\endgroup$
    – Betrand
    Dec 28, 2012 at 19:29
  • $\begingroup$ I have shown above the determinant that you claim in your comment. The absolute value of the determinant of a complex matrix $X$ is the product of its singular values. Using that, my edit shows that $|\det(2A+iB)| \ge |\det(A+iB)|$, which of course, is probably the most trivial result of its type :-) $\endgroup$
    – Suvrit
    Dec 28, 2012 at 20:19

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.