1
$\begingroup$

Let

  • $R = \mbox{diag} (r_1,\dots,r_n)$, where $r_1, \dots, r_n > 0$, be a (positive) diagonal matrix.

  • $1_n \in \mathbb{R}^n $ denote the $n$-dimensional vector of all ones.

  • $S$ be a matrix defined by $[S]_{jk}=\sin (\theta_{j}-\theta_k +\delta)$, where $[\theta_1, \dots, \theta_n]^\mathrm{T} \in \mathbb{R}^n$ and $\delta \in \mathbb R$.

Is it true that $$\left( 1^\mathrm{T}_n S^T R S 1_n -1^\mathrm{T}_n R S R^{-1} S 1_n \right) \geq 0$$ always holds?


My thoughts so far:

Since $S^T R S = (R^{1/2} S)^\mathrm{T} (R^{1/2}S)$, therefore it is positive semidefinite and so $1^\mathrm{T}_n S^T R S 1_n \geq 0$.

Also, I see that when $1^\mathrm{T}_n S^T R S 1_n = 0$, then $1^\mathrm{T}_n R S R^{-1} S 1_n =0$.

$\endgroup$

1 Answer 1

3
$\begingroup$

Consider $R = {\rm diag}(r, 1/r)$ and $\delta = \pi/4, \theta_1 = \pi/4, \theta_2 = 0$. Then $$ S = \left[\begin{array}{cc} \sin(\theta_1 - \theta_1 + \delta) & \sin(\theta_1 - \theta_2 + \delta) \\ \sin(\theta_2 - \theta_1 + \delta) & \sin(\theta_2 - \theta_2 + \delta) \end{array}\right] = \left[ \begin{array}{cc} \sin(\pi/4) & \sin(\pi/2) \\ \sin(0) & \sin(\pi/4) \end{array}\right]= \left[\begin{matrix} 1/\sqrt 2 & 1 \\ 0 & 1/\sqrt 2\end{matrix}\right].$$ Now $$ S^TRS = \left[\begin{matrix} r/\sqrt 2 & 0 \\ r & 1/(\sqrt 2 r)\end{matrix}\right]\left[\begin{matrix} 1/\sqrt 2 & 1 \\ 0 & 1/\sqrt 2\end{matrix}\right] = \left[\begin{array}{cc} r/2 & r/\sqrt 2 \\ r/\sqrt 2 & r + 1/(2r) \end{array}\right] $$ $$ RSR^{1/2}S = \left[\begin{matrix} 1/\sqrt 2 & r^2 \\ 0 & 1/\sqrt 2\end{matrix}\right]\left[\begin{matrix} 1/\sqrt 2 & 1 \\ 0 & 1/\sqrt 2\end{matrix}\right]= \left[\begin{array}{cc} 1/2 & (1+r^2)/\sqrt 2 \\ 0 & 1/2 \end{array}\right] $$ and so $$ 1_2^TS^TRS1_2 = (3/2 + \sqrt 2)r + 1/(2r) \ \ \textrm{and} \ \ 1_2^TRSR^{-1}S1_2 = 1 + 1/\sqrt 2 + r^2/\sqrt 2. $$ Therefore, $$\lim_{r\rightarrow \infty} 1_2^TS^TRS1_2 - 1_2^TRSR^{-1}S1_2 = -\infty$$ and so your desired inequality does not hold in general.

$\endgroup$
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.