11

Find the following sum

$$S= \sum _{n=1}^{\infty} \tan^{-1} \frac{2}{n^2+n+4}$$

I am not able to make it telescopic series. Could someone help me with this?

user21820
  • 60,745
Mathematics
  • 4,052

3 Answers3

14

This can be rewritten as $$\sum_{n=1}^\infty \left(\arctan \frac2n - \arctan \frac2{n+1}\right) = \arctan 2.$$


Of course if we mysteriously find this expression, we can verify it using the formula $\arctan x - \arctan y = \arctan \frac{x-y}{1+xy}$. But how do we find this expression in the first place?

If we guess the form of $x$ and $y$, namely, $\frac1{a n + b}$ and $\frac1{a(n+1)+b}$, then we can figure out when $\frac{x-y}{1+xy} = \frac{2}{n^2+n+4}$. This gives us $$\frac{a}{a^2 n^2+a^2 n+2 a b n+a b+b^2+1} = \frac{2}{n^2+n+4}.$$ To get a $2$ in the numerator at the same time as an $n^2$ in the denominator, we want $a = \frac12$, and then $b=0$ follows by straightforward algebra.

(Guessing the form $\frac1{an+b}$ is not too unreasonable, since we want something that goes to $0$ as $n \to \infty.$ It's not the only possibility, of course.)

Another answer is "find the pattern and check". If you compute the first few partial sums and try to stuff them into one $\arctan$ (by using the corresponding addition formula), then we get $$\arctan \frac13, \arctan \frac47, \arctan \frac34, \arctan \frac89,\arctan 1, \arctan \frac{12}{11},\arctan \frac{7}{6}, \arctan \frac{16}{13}, \dots,$$ and that's how long it took me to spot the pattern: this is $$\arctan \frac26,\arctan \frac47,\arctan \frac68,\arctan \frac89,\arctan \frac{10}{10},\arctan \frac{12}{11},\arctan \frac{14}{12},\arctan \frac{16}{13},$$ so the $n^{\text{th}}$ term is $\arctan \frac{2n}{n+5}$. This doesn't let us find the telescoping form, but now we can prove that this is the correct partial sum by induction, and the limit as $n \to \infty$ is $\arctan 2$.

Misha Lavrov
  • 159,700
  • 2
    And the fact that the $5^{\text{th}}$ partial sum is $1$ gives us the identity $$\tan ^{-1}\left(\frac{1}{3}\right)+\tan ^{-1}\left(\frac{1}{5}\right)+\tan ^{-1}\left(\frac{1}{8}\right)+\tan ^{-1}\left(\frac{1}{12}\right)+\tan ^{-1}\left(\frac{1}{17}\right) = \frac \pi4.$$ – Misha Lavrov Apr 13 '17 at 22:33
10

The given series is just the argument of $$ \prod_{n\geq 1}\left(1+\frac{2i}{n^2+n+4}\right) =\left(\frac{4}{5}+\frac{8 i}{5}\right) \text{sinh}(2\pi)\, \text{sech}\left(\frac{\sqrt{15} \pi }{2}\right)$$ i.e. $\color{red}{\arctan 2}$. The above identity follows from the Weierstrass products of the (hyperbolic) sine and cosine functions.

Jack D'Aurizio
  • 361,689
  • 1
    Please, can you elaborate more your answer? I became very intrigued and now I am very curious. I found this other question https://math.stackexchange.com/questions/128181/an-arctan-series-with-a-parameter-sum-n-1-infty-arctan-left-frac2a2n?rq=1, but it seems it is not completed. Or at least give some reference in order to have a deeper understanding. Regards, – cgiovanardi Apr 13 '17 at 23:14
  • There it is! That's the right way. Genius, all the other answers are too involved, this is very distilled. –  Apr 13 '17 at 23:16
  • You mean the argument after taking log of both sides ? – Zaid Alyafeai Apr 14 '17 at 01:40
  • Big canons for a light-feathered problem. See the solution below. – uniquesolution Apr 14 '17 at 20:50
  • @uniquesolution: I know that creative telescoping works as a charm, but this approach works even when CT does not. – Jack D'Aurizio Apr 14 '17 at 20:52
7

"Recall" this trigonometric identity: $$ \arctan x + \arctan y = \arctan \frac{x+y}{1-xy} $$ Now think about applying that: \begin{align} & \arctan \frac{an+b}{cn+d} + \arctan \frac{en+f}{gn+h} \\[10pt] = {} & \arctan \frac{\dfrac{an+b}{cn+d} + \dfrac{en+f}{gn+h}}{1 - \dfrac{an+b}{cn+d} \cdot \dfrac{en+f}{gn+h}} \\[10pt] = {} & \arctan \frac{(an+b)(gn+h) + (en+f)(cn+d)}{(cn+d)(gn+h) - (an+b)(en+f)} \\[10pt] = {} & \arctan \frac{(ag+eh)n^2 + (bg+ah+ed+fe)n + (bh + fd)}{(cg-ae)n^2 + (ch+dg-af-be)n + (dh-bf)} \end{align} You'd want to choose $a,b,c,d,e,f,g,h$ in such a way that $ag+eh=0,$ $cg-ae=1,$ etc.

Then you've got two terms, $\dfrac{an+b}{cn+d} + \arctan \dfrac{en+f}{gn+h},$ where originally you had one.