The simple question is: what is the correct way to calculate the series expansion of $\arctan(x+2)$ at $x=\infty$ without strange (and maybe wrong) tricks?

Read further only if you want more details.

The first (obvious) problem is that infinite is not a number, thus $\sum_{k=0}^{\infty}\frac{f^{(k)}(\infty)}{k!}(x-\infty)^k$ doesn't make sense.

So I've tried to solve $\lim_{x_0\to \infty} ({\sum_{k=0}^{\infty}\frac{f^{(k)}(x_0)}{k!}(x-x_0)^k})$ one term at a time. The first term (with $k=0$) can be easily calculated and the second term is: $$\lim_{x_0\to \infty} (\frac{x-x_0}{1+(x_0+4x_0+4)} \sim -\frac{x_0}{x_0^2}=\frac{-1}{x_0}) = 0$$

The asymptotic function $\frac{-1}{x_0}$ is similar to what I should get as the second term (which is $\frac{-1}{x}$). But to get the correct result I should do some dangerous stuff: I should replace $x_0$ with $x$ (why?) and do not calculate the limit (otherwise I cannot get the desired precision of the series expansion).

I've also tried to solve $\arctan(1/t+2)$ with $t=0$, but the argument of arctg is still an infinite and the difficulties I encounter are the same.

Is there any way to calculate the series expansion of $\arctan(x+2)$ at $x=\infty$ in a cleaner way without all these problems?


Substituting $x=\frac 1t$ and expanding around $t=0$ is the correct thing to do. The first derivative of $\tan^{-1}(\frac{1}{t}+2)$ is, by the chain rule $$ -\frac{1}{t^2}\frac{1}{(\frac 1t +2)^2} = -\frac{1}{(1+2t)^2}~, $$ which is perfectly finite at $t=0$. The same goes for all the other derivatives.

After you construct your Taylor expansion, just replace $t$ with $\frac 1x$.

Edit: Perhaps the zeroth-order term concerned you. Although the argument does indeed become singular, $\tan^{-1}(1/t)$ is well-behaved as $t\to 0^+$; it just approaches $\pi/2$. So this is the zeroth-order term for your expansion.

Edit 2: Marc Palm correctly points out in his short answer that the Taylor series does not converge (one easy way to see this is that $\tan^{-1}$ approaches different values at $x \to \pm \infty$, but these both correspond to $t\to 0$, so the function cannot be analytic there; this is why I wrote $t\to 0^+$ above). However, the procedure above still gives you a sensible expansion to some finite order, with error bounded by Taylor's theorem.

  • $\begingroup$ Thanks! So I was right with the replacement... I just got confused! :) $\endgroup$ – collimarco Sep 20 '13 at 15:04

The poles accumulate at $\infty$, no Taylor expansion at $\infty$ is possible. $artcan$ is a transcendental function and has non-removable singularity at $\infty$.

  • $\begingroup$ +1: This is worth pointing out; I added to my answer to emphasise that the Taylor series does not converge. $\endgroup$ – Rhys Sep 20 '13 at 15:11

We can write $$\arctan (x+2) = \frac{\pi}{2} - \int_x^\infty \frac{dt}{1+(t+2)^2}.$$

Now we can expand the integrand into a "power series" in $\frac1t$,

$$\begin{align} \frac{1}{1+(t+2)^2} &= \frac{1}{2i} \left(\frac{1}{t+2-i} - \frac{1}{t+2+i}\right)\\ &= \frac{1}{2it} \left(\frac{1}{1+\frac{2-i}{t}} - \frac{1}{1+\frac{2+i}{t}}\right)\\ &=\sum_{\nu=1}^\infty \frac{(-1)^\nu\bigl((2-i)^\nu - (2+i)^\nu\bigr)}{2i}\frac{1}{t^{\nu+1}}. \end{align}$$

That series converges uniformly on $[c,\infty)$ for $c > \sqrt{5}$, and all summands as well as the sum are integrable, so

$$\begin{align} \arctan(x+2) &= \frac{\pi}{2} - \sum_{\nu=1}^\infty \frac{(-1)^\nu\bigl((2-i)^\nu - (2+i)^\nu\bigr)}{2i} \int_x^\infty \frac{dt}{t^{\nu+1}}\\ &= \frac{\pi}{2} - \sum_{\nu=1}^\infty \frac{(-1)^\nu\bigl((2-i)^\nu - (2+i)^\nu\bigr)}{2i\nu} \frac{1}{x^\nu}. \end{align}$$

Expanding the first few coefficients yields

$$\arctan (x+2) = \frac{\pi}{2} - \frac{1}{x} + \frac{2}{x^2} - \dotsb$$

  • $\begingroup$ A general approach to such problems is described in N.G. de Bruijn, "Asymptotic methods in analysis". $\endgroup$ – user64494 Sep 20 '13 at 14:45

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.