0
$\begingroup$

Let $\theta=(\theta_1,\theta_2,\cdots \theta_n)$, and $a_{ij}$ are constants. There is no condition on the positiveness of $a_{ij}$.

Under which condition on $\theta$, such that the following function only has one type of root, which is all $\theta_i$ equals, i.e. $\theta_i-\theta_j=0$, for any $(i,j)$ pair?

\begin{equation}\label{1} \sum_{1\leq i,j\leq n} a_{ij} (\theta_i-\theta_j)\sin((\theta_i-\theta_j)t)=0,\text{ for }t\in(0,1]. \end{equation}

What I already know is the case $a_{ij}\geq 0$

If $a_{ij}$ is element in the adjacency matrix of a connected graph (thus, $a_{ij}\geq 0$), then by imposing the constriant $\theta_i-\theta_j\in[-\frac{\pi}{2},\frac{\pi}{2}]$, we have $$(\theta_i-\theta_j)t\sin((\theta_i-\theta_j)t)\geq\frac{2}{\pi}(t^2(\theta_i-\theta_j)^2),\text{ for }t\in(0,1]$$ since $x\sin x\geq \frac{2}{\pi}x^2$ for $x\in[-\frac{\pi}{2},\frac{\pi}{2}]$ (see the figure of $x\sin x$ below)

Thus $$\frac{1}{t}a_{ij} (\theta_i-\theta_j)t\sin((\theta_i-\theta_j)t)\geq a_{ij}\frac{2}{\pi}t(\theta_i-\theta_j)^2,\text{ for }t\in(0,1]$$ which allows us to conclude that $\theta_i-\theta_j=0$.

My question specifically regarding the case that $a_{ij}$ can be both positive, zero and negative.

Any idea would be much appreciated!

Update: Since writing in comments about the following is too long, I write here as answer. Thanks for @Alexandre Eremenko pushing forward. I still have no idea how to proceed the case (2) and (3), which means Sines are not linearly independent.

(1) If there is no $\sin((\theta_i-\theta_j)t)$ equals, then all $\sin((\theta_i-\theta_j)t)$ are linearly independent.

Under this condition, since $\sin((\theta_i-\theta_j)t)$ are linearly independent, then only when $a_{ij}(\theta_i-\theta_j)=0$ (equivalently $(\theta_i-\theta_j)=0$ since we consider $a_{ij}\neq 0$) it holds that $\sum a_{ij}(\theta_i-\theta_j)\sin((\theta_i-\theta_j)t)=0$.

(2) If there is $\sin((\theta_i-\theta_j)t)$ equals, equivalently, there exists pair of $\theta_i-\theta_j$ equals and there exists $i_1,j_1,i_2,j_2$ such that $(\theta_i-\theta_j)+(\theta_i-\theta_j)=\pi+2k\pi$ where $k\in\mathbb{Z}$. Then Sines are not linearly independent. We split into two cases:

(2.1) If there exists pair of $\theta_i-\theta_j$ equals, then for those $i^*,j^*$ that $\theta_{i^*}-\theta_{j^*}\neq 0$, we do not have further information and this needs further analysis.

(2.2) If there exists $i_1,j_1,i_2,j_2$ such that $(\theta_i-\theta_j)+(\theta_i-\theta_j)=\pi+2k\pi$ where $k\in\mathbb{Z}$, then we do not have further information and this needs further analysis.

enter image description here

$\endgroup$
7
  • $\begingroup$ The main identity that you wrote, before the words "What I already know", is a linear combination of sines (not functions of the form $t\sin(ct)$ as the rest of this post suggests. Is this a misprint? $\endgroup$ Mar 7, 2023 at 13:01
  • $\begingroup$ In any case, this problem is easily solved by noticing that functions $t^n\sin(kt)$ with any distinct pairs $(n,k)$ are linearly independent. $\endgroup$ Mar 7, 2023 at 13:04
  • $\begingroup$ @AlexandreEremenko It's not a misprint, I rewrote $(\theta_i-\theta_j)\sin((\theta_i-\theta_j)t)$ as $\frac{1}{t}(\theta_i-\theta_j)t\sin((\theta_i-\theta_j)t)$ $\endgroup$
    – M.K
    Mar 7, 2023 at 13:18
  • $\begingroup$ @AlexandreEremenko Hi Alexandre, Could you elaborate your idea more? Thank you very much! $\endgroup$
    – M.K
    Mar 7, 2023 at 13:21
  • $\begingroup$ I understand that $t\sin(kt)$ with different $k$ are linearly independent, which means $a_1t\sin(k_1t)+a_2t\sin(k_2t)+\cdots+a_mt\sin(k_mt)=0$ only has one solution which is $(a_1,\cdots,a_m)=(0,\cdots,0)$. However here I am solving the equation for $k_1,\cdots,k_m$ instead of $(a_1,\cdots,a_m)$. $\endgroup$
    – M.K
    Mar 7, 2023 at 13:42

1 Answer 1

1
$\begingroup$

The equation you wrote is $$\sum_{i,j}a_{i,j}(\theta_i-\theta_j)\sin((\theta_i-\theta_j)t)=0$$ for $t\in(0,1).$. Since sines are linearly independent, this is only possible either when $a_{i,j}(\theta_i-\theta_j)=0$ for all $i,j$ or if $$\theta_{i_1}-\theta_{j_1}=\theta_{i_2}-\theta_{j_2}$$ for some $i_1,j_1,i_2,j_2$. Conversely, if this holds for these indices, set $a_{i_1,j_1}=-a_{i_2,j_2}$ and the rest of $a_{i,j}=0$ and you have your identity.

$\endgroup$
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.