2

I would like to know how to evaluate $$\sum\limits_{r=1}^\infty(-1)^{r+1}\frac{\cos(2r-1)x}{2r-1}$$

There are a couple of issues I have with this. Firstly, depending on the value of $x$, it seems, at least numerically, that this value is always $$\pm\sqrt{\frac{\sqrt5-1}{2}}$$ If this is true, how would I prove this?

Secondly, I tried using complex numbers to evalaute this, as shown briefly below:

$$C=\sum\limits_{r=1}^\infty(-1)^{r+1}\frac{\cos(2r-1)x}{2r-1}$$ $$S=\sum\limits_{r=1}^\infty(-1)^{r+1}\frac{\sin(2r-1)x}{2r-1}$$ $$\implies C+iS=\arctan e^{ix}$$ on using the power series of $\arctan x$. Differentiating yields a completely imaginary number: $$\frac{i}{2}\sec x$$ which proves that the value of $C$ is a constant. However, this gives me no information whatsoever on the value of $C$.

Thank you for your help.

  • Why would you want the exact value of this series? Or are you only interested in some "properties" of this series? –  Jan 16 '21 at 23:40
  • @mrsamy I'm just looking at it out of curiosity. – A-Level Student Jan 16 '21 at 23:41
  • 1
    That's a fair reason :-) –  Jan 16 '21 at 23:41
  • Have you studied some Fourier Series, perhaps in advanced analysis...or harmonic analysis...or stuff? – DonAntonio Jan 17 '21 at 00:41
  • @DonAntonio unfortunately not; I am still in high school and don't know a lot beyond the high school curriculum. – A-Level Student Jan 17 '21 at 09:35
  • @A-Level Student: what a coincidence. I am also calculating this series $C=\cos(x)-\dfrac{1}{3}\cos(3x)+\dfrac{1}{3}\cos(5x)...$. The sine series is just exactly the same. I got your answer. In the book "The analytic theory of heat" of Fourier, chapter III, he wrote that the sum of this series is $\dfrac{\pi}{4}$. It is an equation he used over and over. – James Warthington Feb 27 '21 at 14:08
  • @A-LevelStudent Could you tell me why your line of reasoning is that $C$ is a constant? – James Warthington Feb 27 '21 at 14:30
  • @JamesWarthington cool, that's an interesting coincidence :) I have written extremely brief working above, I don't blame you for wondering about that. Recall a couple of facts: differentiating something completely real yields a completely real derivative, and differentiating something completely imaginary yields a completely imaginary derivative. In our case, we find that $\frac{d}{dx}(C+iS)=\frac{d}{dx}\arctan (e^{ix})$ is completely imaginary. This means that either $C+iS$ is completely imaginary or $C$ is a constant, as when constant are differentiated they 'disappear'... – A-Level Student Feb 27 '21 at 19:56
  • @JamesWarthington (continued) .... As we know that $C+iS$ isn't completely imaginary we conclude that $C$ must be a constant. – A-Level Student Feb 27 '21 at 19:57
  • @JamesWarthington btw note that the value of $C$ can also be $-\frac{\pi}{4}$. – A-Level Student Feb 27 '21 at 19:59
  • @So why did we choose $\pi/4$? $C$ could take 2 values, which mean it is divergent series, is it not? – James Warthington Feb 27 '21 at 20:01
  • @A-LevelStudent BTW, I have written on the method $C+iS$ for "summing" certain kinds of trigonometric series on my blog, I credit you for being the first to introduce it to me: https://analysisbeauty.blogspot.com/2021/02/blog-post_30.html – James Warthington Feb 27 '21 at 20:02
  • @A-LevelStudent Also, for your interest $\sin(x)-\dfrac{1}{2}\sin(2x)+\dfrac{1}{3}\sin(3x)...=\dfrac{x}{2}$. This is drawn from a paper of Euler. He arrived at this result by integrating a string of other series. This is an important series since you see it a lot in Fourier's book. – James Warthington Feb 27 '21 at 20:05
  • @JamesWarthington you may be interested in my new answer to this question. – A-Level Student Mar 29 '21 at 22:16
  • @A-LevelStudent I got your answer. I will have a close look tomorrow. Thanks! It's good to see you progress when learning Math :) – James Warthington Mar 30 '21 at 00:52

3 Answers3

3

As you noted, the real part of the sum, $C$, is constant no matter what the value of $x$ (so long as $x$ is real). Thus plug in $x = 0$

$$C = \operatorname{Re}(\arctan e^{i0}) = \frac{\pi}{4}$$

Ninad Munshi
  • 34,407
  • 1
    Thanks so much! However, it seems that $C$ can take different values; namely $$\pm\frac{\pi}{4}$$ at least according to desmos. How do I obtain the negative value? – A-Level Student Jan 16 '21 at 23:48
  • ... no matter what the value of $x$... - not quite (replace $x$ by $x+\pi$ in the original series to see what actually happens). And... actually we reinvent this... or more specifically this. – metamorphy Jan 16 '21 at 23:50
  • @metamorphy I see, thanks for the clarification! – A-Level Student Jan 16 '21 at 23:52
  • @A-LevelStudent the sum changes values at $x=\frac{\pi}{2}$ because arctan is undefined at $i$. It's like what happens with the graph of $\arctan(2\tan x)$ at the same points. But thanks for pointing it out, I would not have caught that otherwise. – Ninad Munshi Jan 16 '21 at 23:55
  • Why isn't $\arctan$ defined at $i$? Isn't it just equal to $i~\text{artanh}1$? – A-Level Student Jan 16 '21 at 23:56
  • @A-LevelStudent there are two ways of looking at it. The derivative of $\arctan x$ is $$\frac{1}{1+x^2}$$ which has poles at $x=\pm i$. The second is see what happens when you try to solve $$\tan z = i$$ with Euler's formula. Also in the formula you have, $\tanh^{-1}1$ is infinity. – Ninad Munshi Jan 16 '21 at 23:59
  • @NinadMunshi I must be tired to have missed that :) Thanks once again! – A-Level Student Jan 17 '21 at 00:00
  • @A-LevelStudent: what is $\tanh^{-1}(1)$? – robjohn Jan 17 '21 at 11:15
  • @robjohn it's $\infty$; I forgot that yesterday. – A-Level Student Jan 17 '21 at 12:17
2

The Series

This series only converges for $x\in\mathbb{R}$. If $x\not\in\mathbb{R}$, the terms do not go to $0$.$\newcommand{\sgn}{\operatorname{sgn}}\newcommand{\Re}{\operatorname{Re}}\newcommand{\Im}{\operatorname{Im}}$ $$ \begin{align} &\sum_{r=1}^\infty(-1)^{r+1}\frac{\cos((2r-1)x)}{2r-1}\tag{1a}\\ &=\frac12\sum_{r=1}^\infty(-1)^{r+1}\frac{e^{i(2r-1)x}}{2r-1}+\frac12\sum_{r=1}^\infty(-1)^{r+1}\frac{e^{-i(2r-1)x}}{2r-1}\tag{1b}\\ &=\frac12\tan^{-1}\left(e^{ix}\right)+\frac12\tan^{-1}\left(e^{-ix}\right)\tag{1c}\\[9pt] &=\Re\left(\tan^{-1}\left(e^{ix}\right)\right)\tag{1d}\\[12pt] &=\frac\pi4\,\sgn(\cos(x))\tag{1e} \end{align} $$ Explanation:
$\text{(1b)}$: $\cos(x)=\frac12\left(e^{ix}+e^{-ix}\right)$
$\text{(1c)}$: power series for $\tan^{-1}(z)$
$\text{(1d)}$: $\tan(\bar z)=\overline{\tan(z)}$
$\text{(1e)}$: $|\tan(z)\,|=1$ when $\Re(z)=\pm\frac\pi4$
$\phantom{\text{(1e):}}$ and $\Re(\tan(z))$ has the same sign as $\Re(z)$
$\phantom{\text{(1e):}}$ also, if $\cos(x)=0$, all the terms of the series are $0$


Concerning $\boldsymbol{\tan(x+iy)}$

Note that $$ \begin{align} \tan(x+iy)&=\frac{\sin(x)\cosh(y)+i\cos(x)\sinh(y)}{\cos(x)\cosh(y)-i\sin(x)\sinh(y)}\tag{2a}\\ &=\frac{\sin(x)\cos(x)+i\sinh(y)\cosh(y)}{\cos^2(x)\cosh^2(y)+\sin^2(x)\sinh^2(y)}\tag{2b} \end{align} $$ is on the unit circle when the absolute values of the numerator and denominator are equal. Since $\cosh^2(y)-\sinh^2(y)=1$, equation $\text{(2a)}$ says that is when $$ \begin{align} \sin^2(x)\cosh^2(y)+\cos^2(x)\sinh^2(y)&=\cos^2(x)\cosh^2(y)+\sin^2(x)\sinh^2(y)\tag{3a}\\[6pt] \sin^2(x)&=\cos^2(x)\tag{3b} \end{align} $$ Thus, $|\tan(z)\,|=1$ when $\Re(z)=\pm\frac\pi4$

Furthermore, equation $\text{(2b)}$ says not only that $\tan(\bar z)=\overline{\tan(z)}$, but also that, when $\Re(z)=\pm\frac\pi4$, $\Re(\tan(z))$ has the same sign as $\Re(z)$.

enter image description here

robjohn
  • 345,667
1

I've learned some new related facts since posting the question, one of which solves this very simply. The relevant formula, stated below, can be found by using the fact $\arctan z=-i\operatorname{artanh} iz$. $$\arctan(x+iy)=\frac{1}{2}\arctan\frac{2x}{1-x^2-y^2}+\frac{i}{2}\operatorname{artanh}\frac{2y}{1+x^2+y^2}$$ Plugging in $x=\cos x$ and $y=\sin x$ we find that $$\arctan e^{ix}=\frac{1}{2}\arctan(\pm\infty)+\frac{i}{2}\operatorname{artanh}\sin x$$ And thus $$C=\pm\frac{\pi}{4}$$