3

While trying to express $\cos^n x(n\in\mathbb C)$ in terms of multiple angles, I tried to use the method of complex numbers by applying de-Moivre's theorem on $z=e^{ix}$: $$2^n\cos^nx=\left(z+\frac1{z}\right)^n$$ $$\implies2^n\cos^nx=z^{-n}(z^2+1)^n$$ Now, I got the radius of convergence of the series given by $(1+z^2)^n$ to be $1$ by the ratio test:

$$(1+z^2)^n=1+nz^2+\frac{n(n-1)}{2!}z^4+\frac{n(n-1)(n-2)}{3!}z^6+\cdots$$ $$\implies r=\lim_{t\to\infty}\left|\frac{t+1}{n-t}\right|=1$$

But $|z^2|=1$, so this is the boundary case, hence the series may be converging or diverging. So, how do I determine that for what values of $z$, this series will converge?

Please try to give answers in an easy-to-understand language from the perspective of a high-school student since as far as determining convergence or divergence of a series is concerned, I only know about the $n^{\text{th}}$ term test, the ratio test and determining the radius of convergence of a series by the ratio test.

Integreek
  • 8,530
  • We seek then for $z$ for convergence :

    $$ |1+z^2|<1 $$

    or equivalently (multiply $-1$ inside absolute value) : $$ |(iz)^2-1|<1$$

    So then (in the circle of radius one and center $(0,1)$ :

    $$ (iz)^2 \in \mathcal{C}(r=1,c=(0,1)) $$

    So the :

    $$ -z^2=1+\rho e^{i\theta}$$ Can you end from here ?

    – EDX Oct 12 '24 at 16:47
  • @EDX I am referring to the infinite binomial series represented by $(1+z^2)^n$, not a series of the form $\sum_{r=0}^{\infty} a_n(1+z^2)^r$ – Integreek Oct 12 '24 at 17:36
  • Yes same for me it is why in order for a complex expression $z^n$ to converge, you need the $|z|<1$ so for $|z^2+1|$. Noting that $(1+z^2)^n$ isn't an infinite serie in itself but rather a finite sum that can tend to an infinite serie. Indeed take $z=1/2$ it will diverge, yet being in the "supposed radius". Hence my previous development – EDX Oct 12 '24 at 18:23
  • Aren't you interested more in $z^{-n}(1+z^2)^n$ for your problem ? – EDX Oct 12 '24 at 18:35
  • @EDX By the binomial theorem for any index, shouldn't $(1+z^2)^n$ be a finite series for $n\in\mathbb N$ and infinite series otherwise? Yes, I am interested in $z^{-n}(1+z^2)^n$ but isn't finding whether it converges or not same as that of $(1+z^2)^n$? Please correct me if I am wrong. – Integreek Oct 13 '24 at 02:34
  • 1
    Oh sorry I didnt see the exponent being complex – EDX Oct 13 '24 at 09:20
  • You can still use the formula of the generalized newton binomial (see def on Wiki e.g.) it will give you the same form. Then it is interesting to understand what is $z+1/z$ – EDX Oct 13 '24 at 10:02
  • @EDX yes, I have used that only, but am doubtful if it is correct since $|z^2|=1$. – Integreek Oct 13 '24 at 10:03
  • $|z|=1$ isn't a problem. It is the problem of $z=1$. Because take $z=i$ and its works and $|i^2|=1$ ! The fundamental question is when (for which z) does the $|z+1/z|<1$ ? see the conformal transformation here https://www.youtube.com/watch?v=4bhdMM66u-E. Once you solve this question, your exercise is done, $n$ being complex or not. – EDX Oct 13 '24 at 10:05
  • But, as per Wikipedia, the binomial theorem is applicable for $(x+y)^n$ when $|x|>|y|$, and here $|z|=\left|\frac1{z}\right|=1$... – Integreek Oct 13 '24 at 10:07
  • Yes you're right, I might not have been clear. Forget the binomial, indeed not applicable and focus and the when (for which z) does the $|z+1/z|<1$ (this is based on the more elementary notion : when a geometric sequence is convergent) – EDX Oct 13 '24 at 10:08
  • @EDX but the terms involved are $z$ and $1/z$, so how does ensuring $z+1/z=1$ help here? – Integreek Oct 13 '24 at 10:32
  • 1

1 Answers1

6

Let $z\in\mathbb{C}$, $|z|=1$, and $\nu\in\mathbb{C}\setminus\mathbb{Z}_{\geqslant0}$. Then $\sum_{n=0}^\infty\binom{\nu}{n}z^n$

  • converges absolutely if $\Re\nu>0$;
  • converges if $\Re\nu>-1$ and $z\neq-1$;
  • diverges otherwise.

If it converges, its sum is $(1+z)^\nu$.

The idea here is that $(-1)^n n^{\nu+1}\binom{\nu}{n}$ has a non-zero limit as $n\to\infty$, since $$ (-1)^n (n+1)^{\nu+1}\binom{\nu}{n}=\prod_{k=1}^n\left(1-\frac{\nu+1}k\right)\left(1+\frac1k\right)^{\nu+1} $$ does (the term under the product is $1+O(k^{-2})$ as $k\to\infty$).

(The limit is known as $1/\Gamma(-\nu)$, but we don't need this fact here.)

This yields the absolute convergence of $\sum_{n=0}^\infty\binom{\nu}{n}z^n$ when $\Re\nu>0$ (its term is $O(n^{-1-\Re\nu})$ as $n\to\infty$), and the divergence when $\Re\nu\leqslant-1$ (since the term doesn't tend to zero).

For $-1<\Re\nu\leqslant0$, $$ (1+z)\sum_{n=0}^N\binom{\nu}{n}z^n=1+\binom{\nu}{N}z^{N+1}+\sum_{n=1}^N\left[\binom{\nu}{n}+\binom{\nu}{n-1}\right]z^n $$ has a limit as $N\to\infty$, since $\binom{\nu}{N}=O(N^{-1-\Re\nu})$ and $\binom{\nu}{n}+\binom{\nu}{n-1}=\binom{\nu+1}{n}=O(n^{-2-\Re\nu})$ by the above. Thus, if $z\neq-1$, the series $\sum_{n=0}^\infty\binom{\nu}{n}z^n$ converges. At $z=-1$ it diverges, since otherwise, by Abel's theorem, its sum would be equal to $\lim_{r\to 1^-}(1-r)^\nu$, which doesn't exist.

The claim about the sum being equal to $(1+z)^\nu$ follows again from Abel's theorem.

metamorphy
  • 43,591
  • What does $\nu\in\mathbb{C}\setminus\mathbb{Z}_{\geqslant0}$ mean? – Integreek Oct 18 '24 at 03:41
  • 1
    @MathGuy: $\nu\in\mathbb{C}$ and $\nu$ is not a nonnegative integer. – metamorphy Oct 18 '24 at 04:17
  • And does $\binom{n}{r}=\frac{n(n-1)...(n-r+1)}{r!}$ for any $n, r\in\mathbb C$? – Integreek Oct 18 '24 at 04:20
  • 1
    @MathGuy: For $r\in\mathbb{Z}_{\geqslant 0}$. (Otherwise you have to change the numerator... but we don't need it.) – metamorphy Oct 18 '24 at 04:26
  • Can you please explain where did $(-1)^n n^{\nu+1}\binom{\nu}{n}$ come from? And what is $O(k^{-2})$? – Integreek Oct 19 '24 at 10:28
  • Multiplying this series by any non-zero complex number doesn't affect its convergence, right? – Integreek Oct 19 '24 at 11:22
  • Sorry if my questions seem dumb, but I'm still in high school; that's why I am not very acquainted with convergence and divergence of a series. – Integreek Oct 30 '24 at 04:02
  • @MathGuy: I'm using this notation. In particular, for a sequence $(c_n)$ of complex numbers, if $c_n=O(n^{-1-\epsilon})$ as $n\to\infty$, for some constant $\epsilon>0$, then $\sum_n c_n$ converges (absolutely). The fact that $(-1)^n n^{\nu+1}\binom{\nu}{n}$ has a limit (as $n\to\infty$) implies $\binom{\nu}{n}=O(n^{-1-\Re\nu})$, which is used throughout the answer. – metamorphy Nov 01 '24 at 10:49