I want to show that the power series

$$\sum_{n = 0}^{\infty} a_{n}x^{n} = \frac{1}{3 - \cos(x)}$$

has a radius $R > 1$? I don't think the ratio test will work here. From Googling, I found the Cauchy-Hadamard Theorem, and I was wondering if I could somehow apply it to this. Can someone please help me? I'm not so familiar with $\limsup$.


This can be seen as a heuristic way which may not be necessarily correct.

There is a slightly easier way using the expansion for $\dfrac{1}{1-x}$.

$\dfrac{1}{3 - \cos(x)}=\dfrac{1}{3}\left(\dfrac{1}{1-\dfrac{\cos(x)}{3}}\right)=\dfrac{1}{3}\displaystyle\sum_{n=0}^\infty\left(\dfrac{\cos(x)}{3}\right)^n$ exists if $\left|\dfrac{\cos(x)}{3}\right|<1 \implies |\cos(x)|<3$ which is true for all $x\in\mathbb{R}$.


To find the power series we calculate using Taylor series

$f(0)=\dfrac{1}{2}, \: f'(x)=\dfrac{-\sin x}{(3-\cos x)^2}\implies f'(0)=0, \: \\f''(x)=\dfrac{(3-\cos x)(-\cos x)+2\sin^2 x(3-\cos x)}{(3-\cos x)^4}\implies f''(0)=\dfrac{1}{4}$.

Continuing this way we get $\quad$ $\dfrac{1}{3 - \cos(x)}=\dfrac{1}{2}-\dfrac{x^2}{8}+\dfrac{x^4}{24}+\cdots$

$\cos(z)=\dfrac{e^{iz}+e^{-iz}}{2}=3\implies e^{2iz}-6e^{iz}+1=0 \implies e^{iz}=\dfrac{6\pm\sqrt{36-4}}{2}=3\pm2\sqrt{2} $ $\implies z=2\pi n -i\ln(3\pm2\sqrt{2}),\: n\in\mathbb{Z}$.

Now since radius of curvature is defined such that the function is analytic in $\{z: |z - z_0| < r\}$, so the radius of convergence is at least $r$. We infer that if $f$ has a pole at $c$ then it is analytic for $|z-z_o|<c$, so radius of curvature is exactly the distance to the closest pole.

Thus $|z|=|\ln(3+2\sqrt{2})|$. Thus we see that $R=|\ln(3+2\sqrt{2})|>1$.

  • $\begingroup$ that is very clever. $\endgroup$ – joseph Nov 16 '18 at 13:21
  • 2
    $\begingroup$ I don't think this works. The power series of $1/(3- \cos x)$ is not $\frac{1}{3} \sum_n (\frac{\cos x}{3})^n$, because it is not a power series. If you expand it, it does not converge uniformly on the whole real line. Actually the power series has a finite radius of convergence. That's because otherwise it would converge uniformly on the whole $\Bbb C$: but there is the problem that on complex numbers there is a solution for $\cos z=3$. $\endgroup$ – Crostul Nov 16 '18 at 13:26
  • $\begingroup$ @Crostul: Isnt this true for $x\in\mathbb{R}$ ? How does the power series look like in the case $x\in\mathbb{C}$? $\endgroup$ – Yadati Kiran Nov 16 '18 at 13:31
  • 2
    $\begingroup$ The radius of convergence of that power series is bounded by $\ln (3+2\sqrt 2) = 1.76...$. I think that is exactly the radius of convergence, though I'm not sure. You can see it graphically at wolframalpha.com/input/?i=power+series+of+(3-cos+z)%5E(-1) . $\endgroup$ – Crostul Nov 16 '18 at 13:33
  • $\begingroup$ @Crostul: I shall check in the case for $x\in\mathbb{C}$. But what about when $x\in\mathbb{R}$ $\endgroup$ – Yadati Kiran Nov 16 '18 at 13:38

The poles of the function occur at

$$\cos z=3$$ or $$z=\pm i\text{ arcosh}(3)+2k\pi=\pm i\log(3+\sqrt8)+2k\pi.$$

As the radius of convergence is the distance to the closest pole,



You may also invoke Cauchy's integral theorem. $f(z)=\frac{1}{3-\cos(z)}$ is holomorphic over $|z|\leq\frac{3}{2}$, since

$$ \left|3-\cos(z)\right|\geq 2-\sum_{n\geq 1}\frac{|z|^{2n}}{(2n)!}=3-\cosh(|z|) $$ and $K=3-\cosh\left(\tfrac{3}{2}\right)>0$. In particular

$$ |a_n|=\left|\frac{1}{2\pi i}\oint_{|z|=\frac{3}{2}}\frac{dz}{z^{n+1}(3-\cos z)}\right|\leq \frac{1}{K\left(\frac{3}{2}\right)^n} $$ by the triangle inequality, and the radius of convergence of $\sum a_n z^n$ is at least $\frac{3}{2}$.
The same argument works also by replacing $\frac{3}{2}$ with $\frac{7}{4}$, since $\cosh\frac{7}{4}$ still is less than three.


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.