15

How can I prove the following using integration and elementary functions?

Prove that:

$$\sum_{n=1}^{\infty} \frac{\sin(n\theta)}{n} = \frac{\pi}{2} - \frac{\theta}{2}$$

$0 < \theta < 2\pi$

  • Minus one for title and for no shown effort. – k.stm May 07 '13 at 14:04
  • I changed the title and I will try to show some effort. – please delete me May 07 '13 at 14:05
  • If you have heard of a Fourier series, you may want to check that out. – Ron Gordon May 07 '13 at 14:05
  • This was asked yesterday, already. Was it you? First, note this is false for $\theta=0,2\pi$. Second, the lhs is the Fourier series of the rhs. Which is called a sawtooth function. – Julien May 07 '13 at 14:06
  • 2
    I know that it is a simple Fourier series called sawtooth function but I want to do it using integration and elementary functions. – please delete me May 07 '13 at 14:07
  • 1
    Fourier series are much about integration, and every function here is elementary. So if you want specifically to avoid Fourier theory, it would be good to make that clear in your question. – Julien May 07 '13 at 14:11
  • I don't think gamma function and Mellin transform are elementary. – please delete me May 07 '13 at 14:14
  • The functions involved are $\sin$ and a polynomial. These are elementary functions. And when you compute the Fourier coefficients, still elementary functions. So you want to avoid Fourier theory. Please edit accordingly. – Julien May 07 '13 at 14:15
  • Ok, so how can I do it that way? By the way, in high school we don't learn Fourier series :( – please delete me May 07 '13 at 14:18
  • 1
    Is this a highschool question? How did you come accross it? – Julien May 07 '13 at 14:20
  • Homework. My teacher said the hint is INTEGRATION. I don't see how I can use high school integration to do it. – please delete me May 07 '13 at 14:21
  • Do you simply know how to prove the series converges in the first place? This is already nontrivial, and can be shown by Abel transformation for instance. – Julien May 07 '13 at 14:25
  • I'm fairly sure it can be done by elementary methods. It's just not as easy as doing it by fourier transform. – Glen O May 07 '13 at 14:25
  • @GlenO Could you be more explicit? What is your strategy? – Julien May 07 '13 at 14:33
  • @julien: I'm working on it. The "fairly sure" indicates that I'm not yet 100% certain. – Glen O May 07 '13 at 14:34
  • Alexander, you seem to conflating elementary functions (mathematical objects) and elementary methods (mathematical processes). Fourier series uses elementary functions! I believe you did something like this in your fresnel integral question too. – Aryabhata May 07 '13 at 14:34
  • How do I do it with Fourier series? Can you give me a link where I can learn all about Fourier series? – please delete me May 07 '13 at 14:36
  • Fouries series is the representation of a function in the form $f(x) = \sum_{n=0}^{\infty} (a_n \sin n x + b_n \cos nx)$. Here the suggestion is that you consider $f(x) = \frac{\pi}{2} - \frac{x}{2}$. I suggest you Google it if you don't have textbooks available in your school library. – Aryabhata May 07 '13 at 14:38
  • As I suspected, it can be done using elementary functions and integration/differentiation. My answer has been added - it requires differentiation, trig rules (specifically, the product rules for trig), summation manipulation, and integration. – Glen O May 07 '13 at 15:45

3 Answers3

11

Let, $$S_1 = \sum_{n=1}^{\infty}\frac{\cos n\theta}{n}\\ S_2 = \sum_{n=1}^{\infty}\frac{\sin n\theta}{n}$$

Then $$S_1 + iS_2 = \sum_{n=1}^{\infty}\frac{\cos(n\theta)+i\sin(n\theta)}{n}=\sum_{n=1}^{\infty}\frac{e^{in\theta}}{n}$$

Now, from the Taylor expansion, $\ln (1+x) = x -\frac{x^2}{2}+\frac{x^3}{3} ...$ $$\implies -\ln(1-x) = x+ \frac{x^2}{2}+\frac{x^3}{3} ... = \sum_{n=1}^{\infty}\frac{x^n}{n}$$ $$\begin{align} \therefore S_1+iS_2 &= -\ln(1-e^{i\theta}) \\&=-\ln(1-\cos\theta-i\sin \theta) \\ &=-\ln(2\sin^2\theta/2 - 2i\sin(\theta/2)\cos(\theta/2)) \\ &=-\ln(2\sin\theta/2)-\ln(\sin\theta/2-i\cos\theta/2) \\ &=-\ln(2\sin\theta/2)+\ln(\sin\theta/2+i\cos\theta/2) \\ &=-\ln(2\sin\theta/2)+\ln(e^{i(\pi/2-\theta/2)}) \\ &=-\ln(2\sin\theta/2)+i(\pi/2-\theta/2) \end{align} $$

Taking the imaginary part of both sides, $$S_2 = \frac{\pi}{2} - \frac{\theta}{2}$$

Milind Hegde
  • 3,914
  • You can do it a little more directly by using $$\sin(x)=\frac{e^{ix}-e^{-ix}}{2i}$$ – Glen O May 07 '13 at 14:38
  • Yes, I'm not very sure how rigorous this is. With the complex logarithm and the Taylor expansion of natural logarithm, I too doubt this is high school level, but it was the simplest I could get it. – Milind Hegde May 07 '13 at 14:39
  • Note in particular that you are on the boundary of the disk of convergence. Where convergence is the tricky part. – Julien May 07 '13 at 14:40
  • Thanks a lot for the answer, it doesn't use integration though. – please delete me May 07 '13 at 14:42
  • @julien: It is known that $\log (1-z)$ converges everywhere on the boundary except at $z=1$. But, perhaps it would be circular logic, as a proof of that fact might involve proving the above two series converge. – Aryabhata May 07 '13 at 14:44
  • 4
    @AlexanderJones: It can be converted to integration by integrating $\frac{1}{1-x}$... – Aryabhata May 07 '13 at 14:45
  • @Aryabhata I know where it converges... I was indeed slightly pointing at some circularity. – Julien May 07 '13 at 14:46
  • @julien: Yeah, but Abel's test does indeed prove it (as you mentioned earlier). So there is probably no circularity. I was only trying to clarify. – Aryabhata May 07 '13 at 14:48
  • @Aryabhata I mentioned Abel above and the OP does not seem to be aware of that. – Julien May 07 '13 at 14:51
  • @julien: I think we agree :-) – Aryabhata May 07 '13 at 14:52
  • @Aryabhata I think so too. Regards. – Julien May 07 '13 at 14:55
4

$$ I=\sum_{n=1}^\infty \frac{\sin(n\theta)}{n} $$ Now, \begin{align} \frac{dI}{d\theta} &= \sum_{n=1}^\infty \cos(n\theta)\\ \frac{dI}{d\theta}\cos(\theta)&=\sum_{n=1}^\infty \cos(n\theta)\cos(\theta)\\ &=\sum_{n=1}^\infty \frac{\cos((n-1)\theta)+\cos((n+1)\theta)}2\\ &=\sum_{n=0}^\infty \frac{\cos(n\theta)}{2}+\sum_{n=2}^\infty\frac{\cos(n\theta)}{2}\\ &=\frac{1}2+\sum_{n=1}^\infty \frac{\cos(n\theta)}{2}+\sum_{n=1}^\infty\frac{\cos(n\theta)}{2}-\frac{\cos(\theta)}2\\ &=\frac{1-\cos(\theta)}{2}+\frac{dI}{d\theta}\\ \frac{dI}{d\theta}(\cos(\theta)-1) &= -\frac{\cos(\theta)-1}{2}\\ \frac{dI}{d\theta} &= -\frac12 \end{align} Noting that $$ I(\pi) = \sum_{n=1}^\infty \frac{\sin(n\pi)}n = 0 $$ we integrate around $\theta=\pi$ to get $$ I = -\frac\theta2 + \frac\pi2 = \frac\pi2-\frac\theta2 $$ Note that this doesn't strictly require that the $\cos$ sum converges, as we may alter the summation process to obtain convergence. What is important is which terms may be extracted for the purposes of the integration.


There are two obvious ways to handle the nonconvergent nature of the sum for $\frac{dI}{d\theta}$.

Option 1: use $$I = \sum_{n=1}^\infty \frac{\sin(n\theta)}n z^n$$ and then take the limit as $z\to1^{-}$. For any $|z|<1$, the sum in the derivative will converge, and in the limit it will be $\frac12$.

Option 2: Change the order of summation. Let $$ I = \sum_{n=1}^\infty \frac{\sin(n\theta)}n \sum_{k=1}^\infty 2^{-k} $$ which is the same as multiplying by $1$. Now change the order of summation to $n+k=m$ first, as $$ I = \sum_{m=2}^\infty \sum_{k=1}^{m-1} \frac{\sin((m-k)\theta)}{m-k}2^{-k} $$ When summed in this order, the derivative converges.

Glen O
  • 12,425
  • I think it should be $I(\pi) = 0$, not $\pi/2$. – Milind Hegde May 07 '13 at 15:41
  • @milind - good point. A minor typo (repeated when I said "around $\theta=\pi/2$"). Now fixed. – Glen O May 07 '13 at 15:42
  • Your answer is a manipulation of a series which converges nowhere. It would be great if you explained what you mean by altering the summation process. – Julien May 07 '13 at 18:16
  • @julien: I've edited in an explanation. – Glen O May 07 '13 at 20:36
  • You realize that when there is no absolute convergence, all these manipulations are not "obvious", right? – Julien May 07 '13 at 20:47
  • @julien: That's debatable. In the case of this sum, it's easy to see that it's the equivalent of the average point for a sequence that moves in a circle (in the case of $\frac\pi2$, it's direct alternation between 0 and 1). All you're doing with these methods is accelerating convergence, not unlike shortening the step length in Newton's Method to improve convergence for cases like $\sqrt[3]{x}=0.001$. – Glen O May 07 '13 at 20:59
  • As you know, there are theorems to allow you to swap limits/sums and derivatives/integrals. I believe option 1 can be made rigorous. But it implies using at least Abel's theorem, which is not an obvious fact per se. – Julien May 07 '13 at 21:12
  • But hey, that's your answer, and some people seemed to already be convinced by your original manipulations. So I'll move on. – Julien May 07 '13 at 21:17
4

In this answer, I show

$$ \begin{align} \sum_{k=1}^\infty\frac{\sin(2kx)}{k} &=\sum_{k=1}^\infty\frac{e^{i2kx}-e^{-i2kx}}{2ik}\\ &=\frac1{2i}\left(-\log(1-e^{i2x})+\log(1-e^{-i2x})\right)\\ &=\frac1{2i}\log(-e^{-i2x})\\[4pt] &=\frac\pi2-x\quad\text{for }x\in\left(0,\pi\right) \end{align} $$

which is, in essence, milind's answer. However, the question asks about integration. This sounds as if the question is asking to find the Fourier Series of $f(\theta)=\frac\pi2-\frac\theta2$. First, note that $f(\theta)$ is odd; that is, $$ \begin{align} f(2\pi-\theta) &=\frac\pi2-\frac{2\pi-\theta}2\\ &=\frac\theta2-\frac\pi2\\ &=-\left(\frac\pi2-\frac\theta2\right)\\[6pt] &=-f(\theta)\tag{1} \end{align} $$ Equation $(1)$ implies that $$ \begin{align} \color{#00A000}{\int_0^{2\pi}f(\theta)\cos(n\theta)\,\mathrm{d}\theta} &=\int_0^{2\pi}f(2\pi-\theta)\cos(n 2\pi-n\theta)\,\mathrm{d}\theta\\ &=-\color{#00A000}{\int_0^{2\pi}f(\theta)\cos(n\theta)\,\mathrm{d}\theta}\\[6pt] &=0\tag{2} \end{align} $$ because $\color{#00A000}{x}=-\color{#00A000}{x}\implies\color{#00A000}{x}=0$.

Now the question is $$ \begin{align} \int_0^{2\pi}f(\theta)\sin(n\theta)\,\mathrm{d}\theta &=\int_0^{2\pi}\left(\frac\pi2-\frac\theta2\right)\sin(n\theta)\,\mathrm{d}\theta\\ &=-\frac1n\int_0^{2\pi}\left(\frac\pi2-\frac\theta2\right)\,\mathrm{d}\cos(n\theta)\\ &=\left.-\frac1n\left(\frac\pi2-\frac\theta2\right)\cos(n\theta)\right]_0^{2\pi}\\ &\hphantom{=\,}+\frac1n\int_0^{2\pi}\cos(n\theta)\,\mathrm{d}\left(\frac\pi2-\frac\theta2\right)\\[4pt] &=\frac\pi{n}\tag{3} \end{align} $$ $(3)$ says that the Fourier series for $f(\theta)$ on $(0,2\pi)$ is $$ f(\theta)=\sum_{n=1}^\infty\frac{\sin(n\theta)}{n}\tag{4} $$

robjohn
  • 345,667
  • Detail: the Fourier series converges to $(f(\theta^-)+f(\theta^+))/2$. That's $f(\theta)$ on $(0,2\pi)$, but $0$ at $0$ and $2\pi$. – Julien May 07 '13 at 18:01
  • @julien: indeed, but the question only considers $0\lt\theta\lt2\pi$. – robjohn May 07 '13 at 18:03
  • Yes, +1. I guess the simplest justification for pointwise convergence would be that $f$ is piecewise $C^1$ (here meaning continuously differentiable except at two points where right and left derivatives exist), which yields Dirichlet. – Julien May 07 '13 at 18:11
  • @julien: Since $0\equiv2\pi$ on $\mathbb{R}/(2\pi\mathbb{Z})$ there is only one point of discontinuity, or are you talking about something else? – robjohn May 08 '13 at 17:36
  • No, you're right. I somehow meant $0$ and $2\pi$, not considering the quotient. But that does not make much sense. – Julien May 08 '13 at 17:39