8
$\begingroup$

Let $m$ be an odd and $n$ an even positive integers. I need to estimate the maximum value of $\left|\sin(m\theta)\sin(n\theta)\right|$: $$ c_{m,n}:=\max_{\theta\in[0,\pi]} \left| \sin(m\theta) \sin(n\theta) \right| $$ It is not difficult to show that $c_{1,n}\leq \cos\left(\frac{\pi}{2(n+1)}\right)$. I would like to see if this sort of inequality is true for $c_{m,n}$.

Any answer or reference will be appreciated.

$\endgroup$
7
  • $\begingroup$ Do I understand it correctly that you are not claiming that $c_{1, n} = \cos\left(\frac{\pi}{2(n+1)}\right)$? Then what precisely do you want as a "this sort of inequality for $c_{m, n}$"? Clearly $c_{m, n} \le 1$ but that's probably not what you're after... $\endgroup$ Commented Jun 16 at 2:54
  • 2
    $\begingroup$ To maximize this you want $m \theta, n \theta$ close to $k\pi + \frac{\pi}2$ for an integer $k$. This means that you want to approximate $\frac nm$ by $\frac{k_1\pi + \frac{\pi}2}{k_2\pi + \frac{\pi}2}$. Cancelling $\pi$ and multiplying by 2, you want to approximate $\frac nm$ by $\frac{2k_1 + 1}{2k_2 + 1}$ subject to $0 \leq k_1 < n, 0 \leq k_2 < m$. This seems like a fairly standard Diophantine approximation problem, there may be some results on it. $\endgroup$ Commented Jun 16 at 3:14
  • $\begingroup$ Dear Aleksei: Many thanks for your response. $c_{1,n}$ can be less than $\cos(\pi/(2(n+1)))$ so I am not claiming equality but rather an inequality. Of course $c_{m,n}\leq 1$, but one can show that $c_{m,n}<1$ and I wonder if something like "$c_{m,n}\leq \cos(m\pi/(2(n+1)))$" or something in this format is true. I am not sure what is not clear in the question. $\endgroup$
    – MO B
    Commented Jun 16 at 3:16
  • 2
    $\begingroup$ Certainly not $\cos (m\pi/(2(n+1))$, as for large $m, n$ this product of sines may be made close to 1. Possibly, some upper bound like $\cos (\pi/(2(n+m)))$ holds. $\endgroup$ Commented Jun 16 at 6:33
  • 1
    $\begingroup$ Dear @GHfromMO: Thank you. $\endgroup$
    – MO B
    Commented Jun 18 at 23:33

2 Answers 2

7
$\begingroup$

$\newcommand{\ep}{\varepsilon}\newcommand{\de}{\delta}\newcommand{\al}{\alpha}\newcommand{\be}{\beta}\renewcommand{\th}{\theta}$The upper bound $\exp-\frac{d^2}{2(m^2+n^2)}$ on $c_{m,n}$, obtained in the previous answer, can be improved as follows: \begin{equation*} c_{m,n}\le\cos\frac{\pi d}{2(m+n)}, \tag{100}\label{100} \end{equation*} where $d$ is the greatest common divisor of $m$ and $n$.

Indeed, reasoning as in the previous answer (cf. (10), (20), (40), and (50) there) and using the substitutions \begin{equation*} u:=\frac n{m+n}\in(0,1),\quad \th:=\frac{\pi d}{2(m+n)}, \end{equation*} and also using the interchangeability of $m$ and $n$, we reduce \eqref{100} to \begin{equation*} (1+\cos\ep)(1+\cos\de)\le4\cos^2\th \tag{110}\label{110} \end{equation*} given that \begin{equation*} 0\le\ep\le\de\le\pi,\quad u\ep+(1-u)\de\ge2\th. \tag{120}\label{120} \end{equation*}

But \eqref{120} implies \begin{equation*} (1+\cos\ep)(1+\cos\de)\le2(1+\cos\de)=4\cos^2\tfrac\de2 \le4\cos^2\th, \end{equation*} since $0\le\th\le u\frac\ep2+(1-u)\frac\de2\le\tfrac\de2\le\frac\pi2$. So, \eqref{110} follows. $\quad\Box$

$\endgroup$
2
  • $\begingroup$ Dear @Iosif: This is beautiful. Thanks. Another point that might be related to this: I am wondering if your argument can be applied to the problem discussed in this paper: matwbn.icm.edu.pl/ksiazki/aa/aa49/aa4944.pdf $\endgroup$
    – MO B
    Commented Jun 17 at 18:32
  • $\begingroup$ @MOB : Thank you for your appreciation. Having quickly looked at that linked paper, I don't see right away how my answer, concerning the product of two sines, can help there, with the product of many sines. $\endgroup$ Commented Jun 17 at 18:46
5
$\begingroup$

$\newcommand{\ep}{\varepsilon}\newcommand{\de}{\delta}\newcommand{\al}{\alpha}\newcommand{\be}{\beta}$Note that \begin{equation*} c_{m,n}^2=\max_{t\in[0,\pi]}f(t), \tag{10}\label{10} \end{equation*} where \begin{equation*} f(t):=f_{m,n}(t):=\tfrac14\,(1-\cos2mt)(1-\cos2nt). \tag{20}\label{20} \end{equation*} Next, if $\frac{2i+1}{m}=\frac{2j+1}{n}$ for some integers $i$ and $j$, then the even number $(2i+1)n$ equals the odd number $(2j+1)m$, a contradiction. Also, the smallest common denominator for the fractions $\frac{2i+1}{m}$ and $\frac{2j+1}{n}$ is $mn/d$, where $d$ is the greatest common divisor of $m$ and $n$. So, for any integers $i$ and $j$, \begin{equation*} \Big|\frac{2i+1}{2m}\,\pi-\frac{2j+1}{2n}\,\pi\Big|\ge h:=\frac\pi{2mn/d}\in(0,\pi/2]. \tag{30}\label{30} \end{equation*}

Take any $t\in[0,\pi]$. Then \begin{equation*} t\in I\cap J, \end{equation*} where \begin{equation*} I:=\pi[\tfrac{2a-1}{2m},\tfrac{2a+1}{2m}],\quad J:=\pi[\tfrac{2b-1}{2n},\tfrac{2b+1}{2n}] \end{equation*} for some $a\in\{0,\dots,m\}$ and $b\in\{0,\dots,n\}$, so that $|2mt-2\pi a|\le\pi$ and $|2nt-2\pi b|\le\pi$.
Letting \begin{equation*} \ep:=\pi-|2mt-2\pi a|\in[0,\pi],\quad\de:=\pi-|2nt-2\pi b|\in[0,\pi], \tag{40}\label{40} \end{equation*} we get
\begin{equation*} t=\pi\tfrac{2a+\al}{2m}-\al\tfrac\ep{2m}=\pi\tfrac{2b+\be}{2n}-\be\tfrac\de{2n} \end{equation*} for some $\al$ and $\be$ in the set $\{-1,1\}$. So, by \eqref{30} and the triangle inequality, \begin{equation*} h\le\pi|\tfrac{2a+\al}{2m}-\tfrac{2b+\be}{2n}|=|\al\tfrac\ep{2m}-\be\tfrac\de{2n}| \le\tfrac\ep{2m}+\tfrac\de{2n}, \end{equation*} so that \begin{equation*} \frac\ep{2m}+\frac\de{2n}\ge h. \tag{50}\label{50} \end{equation*} Using now the inequality \begin{equation*} 1-\cos u\le2-2(1-|u|/\pi)^2 \end{equation*} for $|u|\le\pi$, and recalling \eqref{40} and \eqref{50}, we see that \begin{equation*} f(t)\le(1-\ep^2/\pi^2)(1-\de^2/\pi^2)\le\exp-\frac{\ep^2+\de^2}{\pi^2} \\ \le\exp-\frac{4h^2m^2n^2}{\pi^2(m^2+n^2)}=\exp-\frac{d^2}{m^2+n^2}. \end{equation*} Thus, by \eqref{10} and \eqref{30}, \begin{equation*} c_{m,n}\le\exp-\frac{d^2}{2(m^2+n^2)}, \end{equation*} where $d$ is the greatest common divisor of $m$ and $n$.

$\endgroup$
7
  • $\begingroup$ Dear Iosif: Many thanks for your answer. Can you please elaborate more on $\epsilon+\delta\geq h$? Also how do you get $-(h/\pi)^4/4$? Because from the previous term you get $+c^2(\epsilon\delta)^2\leq +c^2(\frac{\epsilon+\delta}{2})^4$. $\endgroup$
    – MO B
    Commented Jun 16 at 23:15
  • $\begingroup$ @MOB : I have added details. Also, the bound got improved, because the inequality $\varepsilon+\delta\ge h$ got improved to $\frac\varepsilon{2m}+\frac\delta{2n}\ge h$. $\endgroup$ Commented Jun 17 at 3:17
  • $\begingroup$ Dear @Iosif: I am extremely thankful for your answer and I admit I would never be able to come up with this nice solution. Just a remark: it seems to me that (after some calculation) $\cos\left(\frac{\pi}{2\left(\frac{n}{d}+\frac{m}{d}\right)}\right)$ is also an upper bound which is slightly better than your bound. I wonder if your method would produce this bound (if it is correct) $\endgroup$
    – MO B
    Commented Jun 17 at 4:28
  • $\begingroup$ @MOB : Thank you for your appreciation. At this point, I don't see how to get this $\cos(\cdots)$ bound. Actually, your comment about that bound is unclear to me: it looks like you got the bound but unsure if it is correct, right? Perhaps, I could say more about this if I saw your calculation. $\endgroup$ Commented Jun 17 at 12:17
  • $\begingroup$ Dear @Iosif: Yes, I am not sure if my bound is correct. For calculation, I just picked some $m$ and $n$ and I plot the function $|\sin(nx)\sin(mx)|$. For instance $n=10$ and $m=9$ and then compare your bound with the cos(...) bound. $\endgroup$
    – MO B
    Commented Jun 17 at 13:43

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .