I have a sum of a series of trig function as follows:

$\sum_{j=1}^{d} cos(2 \pi j \; t)$ where t is just a constant.

I am looking for the upper bound $f(t,d)$ such that $\sum_{j=1}^{d} cos(2 \pi j \; t) \leq f(t,d)$. Any idea or link or some papers that I can find hint about this problem will be much appreciated! Thanks ;)

Here, we can assume $t$ is a small number and $t \neq 0$. Thanks a lot!


All the cosines are bounded above by 1, so $d$ is an upper bound. It is tight because at $t=0$ the sum has value $d$.

In fact, we can do this sum exactly: multiplying by $\sin{\pi t}$, and using the prosthaphaeresis formula $$ 2\cos{A}\sin{B} = \sin{(A+B)}-\sin{(A-B)} $$ we have $$ \sum_{n=1}^d \cos{2\pi n t}\sin{\pi t} = \frac{1}{2}\sum_{n=1}^d \left(\sin{(2n+1)\pi t}-\sin{(2n-1)\pi t}\right), $$ which we can see telescopes since $2(n+1)-1=2n+1$, and hence we only have the end terms, which are $\sin{(2d+1)\pi t}$ and $-\sin{\pi t}$, giving the result $$ \sum_{n=1}^d \cos{2\pi n t} = \frac{\sin{(2d+1)\pi t}}{2\sin{\pi t}}-\frac{1}{2}. $$

Now, to get an upper bound near $t=0$, we can expand this in a power series: for $0<x<\sqrt{6}$, we have $$ x-x^3/6<\sin{x}<x, $$ from which we have $$ \frac{1}{\pi t} \leqslant \frac{1}{\sin{\pi t}} \leqslant \frac{1}{\pi t-(\pi t)^3/6} = \frac{1}{\pi t} \frac{1}{1-\pi^2 t^2/6}, $$ for $\lvert t \rvert <\sqrt{6}/\pi$, and $$ \sin{(2d+1)\pi t} \leqslant (2d+1)\pi t, $$ so the whole lot is less than $$ \frac{(2d+1)\pi t}{2\pi t}-\frac{1}{2} = d $$

For a less trivial estimate, probably your best bet is keeping the top sine intact, and taking $$ \frac{\sin{(2d+1)\pi t}}{2\pi t(1-\pi^2 t^2/6)}-\frac{1}{2} $$ as the upper limit. On the other hand, there are the bounds $x>\sin{x}>\frac{2}{\pi} x$ for $0<x<\pi/2$, which may be of use to you.

For a lower bound, you can expand the Taylor series around $t=0$ up to the negative quadratic term, which is $$\begin{align*} -&\frac{1}{2}+\frac{(2d+1)\pi t(1-(2d+1)^2\pi^2 t^2/6)+O(t^5)}{2\pi t(1-\pi^2 t^2/6)+O(t^5)} \\ &= -\frac{1}{2}+(d+1/2)\left( 1-\frac{(2d+1)^2\pi^2}{6} t^2+O(t^5) \right) \left( 1+\frac{\pi^2}{6} t+O(t^5) \right) \\ &= d + \frac{\pi^2}{12}(2d+1) \left( 1-(2d+1)^2 \right) t^2 + O(t^4) \\ &= d -\frac{\pi^2}{3}d(d+1)(2d+1) t^2 + O(t^4) \end{align*}$$ (Now, notice the quadratic term looks suspiciously like the sum of the squares? We know perfectly well why that is...)

  • $\begingroup$ Sorry, I forgot to metion $t$ is not zero, we can assume t is a small number, actully $t =t(\epsilon, p)$. I am looking for a upper limit $f(t,d)$ such that $\sum_{j=1}^{d} cos(2 \pi j \; t) \leq f(t,d)$. Thanks $\endgroup$ – Pew Mar 26 '15 at 4:27
  • $\begingroup$ Thanks! that solved my problem! $\endgroup$ – Pew Apr 10 '15 at 2:02

(Nothing really original here, though I know my use of $\mathbb{Re}$ for "real part" is wrong.)

$\begin{array}\\ s(d, t) &=\sum_{j=1}^{d} cos(2 \pi j \; t)\\ &=\sum_{j=1}^{d} \mathbb{Re} e^{2 \pi ij t}\\ &=\mathbb{Re}\sum_{j=1}^{d} e^{2 \pi ij t}\\ &=\mathbb{Re}e^{2\pi it}\sum_{j=0}^{d-1} e^{2 \pi ij t}\\ &=\mathbb{Re}e^{2\pi it}\dfrac{e^{2 \pi id t}-1}{e^{2 \pi i t}-1}\\ &=\mathbb{Re}e^{2\pi it}\dfrac{(e^{2 \pi id t}-1)(e^{-2 \pi i t}-1)}{(e^{2 \pi i t}-1)(e^{-2 \pi i t}-1)}\\ &=\mathbb{Re}e^{2\pi it}\dfrac{e^{2 \pi i(d-1) t}-e^{2 \pi id t}-e^{2 \pi i t}+1}{1-(e^{-2 \pi i t}+e^{2 \pi i t})+1)}\\ &=\mathbb{Re}\dfrac{e^{2 \pi id t}-e^{2 \pi i(d+1) t}-e^{4 \pi i t}+e^{2\pi it}}{2-2\cos(2 \pi t)}\\ &=\dfrac{\cos(2 \pi d t)-\cos(2 \pi (d+1) t)-\cos(4 \pi t)+\cos(2\pi t)}{2(1-\cos(2 \pi t))}\\ &=\dfrac{\cos(2 \pi d t)-\cos(2 \pi (d+1) t)-\cos(4 \pi t)+\cos(2\pi t)}{2(2\sin^2( \pi t))}\\ &=\dfrac{\cos(2 \pi d t)-\cos(2 \pi (d+1) t)-\cos(4 \pi t)+\cos(2\pi t)}{4\sin^2( \pi t)}\\ \end{array} $

Since $2t <\sin(\pi t)< \pi t $ (because $\sin(x) > \frac{2x}{\pi}$ for $0 < x < \frac{\pi}{2}$), and all the $\cos$ terms are at most $1$,

$|s(d, t)| \le \frac1{\sin^2(\pi t)} <\frac1{4t^2} $.

You can get better bounds that actually involve $d$ by looking at the differences $\cos(2 \pi d t)-\cos(2 \pi (d+1) t)$ and $\cos(4 \pi t)-\cos(2\pi t)$, but I'll leave that to you.

  • $\begingroup$ Thanks! That solved the problem! $\endgroup$ – Pew Apr 10 '15 at 2:02

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.