1

The $\arctan(x)$ can be expanded as a MacLaurin series starting from the integral

$$\arctan(x) = \int \frac{1}{1 + x^2} \mathrm{d}x$$

and using

$$\frac{1}{1 + x^2} = \sum_{n = 0}^{\infty} (-1)^n x^{2n}$$

as suggested in this answer. This series converges for $x \in (-1,1)$, but, after integration, it can be shown that the resulting MacLaurin series

$$\arctan(x) = \sum_{n = 0}^{\infty} (-1)^n \frac{x^{2n + 1}}{2n + 1}$$

converges for $x \in \left[ -1, 1 \right]$.

The integration of a series is possible only when the series is evaluated within its interval of convergence, $x \in (-1,1)$: the MacLaurin series for $x = \pm 1$ shouldn't even be considered, because the above result for $\arctan(x)$ should not be available!

  • Why instead, here two more points ($x = \pm 1$) can be added to the interval of convergence?

Integration does not change the radius of convergence of a series.

  • What are then the differences between the interval of convergence and the radius of convergence? Do $(-1,1)$ and $\left[ -1, 1 \right]$ correspond to the same radius of convergence?

I read also this question, answer and comments am I am not familiar with Cauchy-Hadamard Radius Formula.

A comment here states (given an interval of convergence $(a - R, a + R)$):

The issue of convergence at the points $x= a ± R$ is independent of the convergence within the interval $(a−R,a+R)$.

My questions above essentially are: why?

BowPark
  • 1,366
  • Radius of convergence is same both before and after integration. However the results are guaranteed only in the interior of region of convergence. The analysis of convergence at boundary points depends on the specific series involved and can not be guaranteed in general. – Paramanand Singh Jun 10 '20 at 12:48
  • @ParamanandSingh So, it's not that the integrated series can not exist out of $(-1 , 1)$: rather, it is not guaranteed to exist. – BowPark Jun 10 '20 at 20:36
  • Well, it is also guaranteed that the series can't exist out of $[-1,1]$. So it is only the boundary points $-1,1$ for which there is no guarantee. – Paramanand Singh Jun 10 '20 at 22:18
  • @ParamanandSingh What really confused me is that many lectures/books state that The interval of convergence of the anti-differentiated series is the same as that of the original series (here, for example, and many other ones). It should be stated that: for $x \notin [a-R, a+R]$ the integrated series can not exist, and it could possibly exist for $x = \pm R$, as stated here ("Integration of Power Series" paragraph). – BowPark Jun 11 '20 at 09:58
  • 1
    I will suggest that you need to assume truth of statements in a math textbook only via their proof (either given in textbook or can be supplied by you). So it is not the interval of convergence which remains constant but rather the radius of convergence. – Paramanand Singh Jun 11 '20 at 10:00
  • 1
    If the radius of convergence of a power series is $R$ it means that it is absolutely convergent in $(-R, R) $ and series of absolute values diverges outside $[-R, R] $. There is no general theorem to deal with boundary points $\pm R$. Good textbooks often mention this explicitly about the boundary points. – Paramanand Singh Jun 11 '20 at 10:03
  • @ParamanandSingh I agree with you about textbooks, after dealing with this issue (IMHO, however, a book should offer a higher reliability: I am not supposed to verify any single statement). Thank you for all your help. If you would like to create an answer with all the considerations in these comments, it is welcome. – BowPark Jun 11 '20 at 10:09

2 Answers2

1

On the request of asker I am writing a full blown answer based on my comments.

Let us then note that a series of the form $$\sum_{n=0}^{\infty} a_nz^n$$ is called a power series in variable $z$. Here $a_n, z$ are complex numbers.

A fundamental feature of such a series is the existence of a non-negative real number $R$ called the radius of convergence and it has the following properties:

  • The series $\sum_{n=0}^{\infty} a_nz^n$ converges absolutely for $|z|<R$.
  • The series $\sum_{n=0}^{\infty} |a_nz^n|$ diverges to $\infty $ for $|z|>R$.

There exist special and useful cases when $R=\infty$ and then the series converges everywhere in the complex plane. Functions represented by such series are called entire functions.

The region $|z|<R$ is called the circle of convergence. In case of real variable $z$ this gets replaced by interval of convergence $(-R, R) $.

A power series can be always be integrated as well as differentiated term by term to lead to another power series with the same radius of convergence.

The convergence of a power series at boundary points $|z|=R$ requires special analysis on a case by case basis and there are no general theorems without any additional hypotheses to confirm convergence or divergence at these points.


Some textbooks define a power series in the form $\sum a_n(z-a) ^n$ and to handle these we can just replace $z$ with $z-a$ everywhere in previous part of this answer.

  • After integration or differentiation, the radius of convergence of the resulting series is the same, $R$, but the interval of convergence may be different as regards the endpoints: $x = \pm R$, in your example, in case $z = x \in \mathbb{R}$. The convergence in these endpoints must be verified individually for each case. Thank you so much for all the clarifications. – BowPark Jun 11 '20 at 10:44
-1

In general, you won't be able to say things about endpoints of an interval after integrating.

In this case, however, we can just check that the sum converges at the endpoints directly; if $x = \pm 1$, then $x^{2n+1} = x$, so the sum becomes $$x \sum_{n=0}^\infty \frac{(-1)^n}{2n+1},$$ which converges because $\frac{(-1)^n}{2n+1}$ is an alternating series with $\frac{1}{2n+1}$ decreasing and going to $0$.

hdighfan
  • 4,067
  • Yes, I know that check for endpoints. But my doubt is slightly different: why can we evaluate those checkpoints? They are not in the interval of convergence of the series which has been integrated. And what is the difference between interval and radius of convergence? – BowPark Jun 10 '20 at 08:53
  • @BowPark Abel's theorem may be relevant in this case. – Gary Jun 10 '20 at 09:29