Amazing arctans

What is the sum of the two shaded angles?



image1

Here is a diagram that should help.


Are any other angles the same as either of the shaded angles?



image2

The purple shaded angle is formed by going along \(2\) and up or down \(\).



image3

What kind of triangle is this?



image4

The triangle is right-angled and isosceles, so the sum of the two shaded angles is \(45^\circ\).



That's a lovely geometrical approach to the problem. But trigonometry gives another way that is easier to extend, but much less intuitive.

If the two angles are \(\theta\) and \(\varphi\), then $$\begin{aligned} \tan(\theta+\varphi)&=\frac{\tan\theta+\tan\varphi}{1-\tan\theta\tan\varphi}\\[1em] &=\frac{\frac{1}{2}+\frac{1}{3}}{1-\frac{1}{2}\frac{1}{3}}\\[1em] &=\frac{\frac{5}{6}}{\frac{5}{6}}\\[1em] &1\\[1em] \Rightarrow \theta+\varphi&=45^\circ \end{aligned}$$


We can write this another way: $$\tan^{-1}\frac{1}{2}+\tan^{-1}\frac{1}{3}=\tan^{-1}1$$


Maybe we should ask: are there any other integer values of \(a\), \(b\), and \(c\) where $$\tan^{-1}\frac{1}{a}+\tan^{-1}\frac{1}{b}=\tan^{-1}\frac{1}{c}$$ that is, by applying tan to both sides and using a compound angle formula, $$\begin{aligned}&\frac{\frac{1}{a}+\frac{1}{b}}{1-\frac{1}{a}\frac{1}{b}}=\frac{1}{c}\\[1em] \Rightarrow &\frac{a+b}{ab-1}=\frac{1}{c} \end{aligned}$$

Can you find any?



What are:

\(\tan^{-1}\frac{1}{5}+\tan^{-1}\frac{1}{8}\)
and
\(\tan^{-1}\frac{1}{2}+\tan^{-1}\frac{1}{5}+\tan^{-1}\frac{1}{8}\)

Can you come up with a geometrical interpretation of this result?


First, stick with the original problem, and look at this diagram.


Can you find any other purple angles here?



image5

The two shaded triangles are similar (count the squares and diagonals).



image6

This shaded triangle is clearly isosceles.



image7

So again, $$\tan^{-1}\frac{1}{2}+\tan^{-1}\frac{1}{3}=\tan^{-1}1$$


This diagram extends to this one:



image8

to show that $$\tan^{-1}\frac{1}{5}+\tan^{-1}\frac{1}{8}=\tan^{-1}\frac{1}{3}$$ so that $$ \begin{aligned} \tan^{-1}\frac{1}{2}+\tan^{-1}\frac{1}{5}+\tan^{-1}\frac{1}{8}&= \tan^{-1}\frac{1}{2}+\tan^{-1}\frac{1}{3}\\[1em] &=\tan^{-1}1=45^\circ \end{aligned}$$

Try this: $$\tan^{-1}\frac{1}{13}+\tan^{-1}\frac{1}{21}$$ and this: $$ \tan^{-1}\frac{1}{2}+\tan^{-1}\frac{1}{5}+\tan^{-1}\frac{1}{13}+\tan^{-1}\frac{1}{21}$$

What do you notice about all the denominators of the fractions so far?


They are all Fibonacci numbers. This suggests a line of enquiry that we will follow soon.

In the meantime, lets see if there are any (other) integer solutions to $$\tan^{-1}\frac{1}{a}+\tan^{-1}\frac{1}{b}=\tan^{-1}\frac{1}{2}$$ $$\tan^{-1}\frac{1}{a}+\tan^{-1}\frac{1}{b}=\tan^{-1}\frac{1}{3}$$ $$\tan^{-1}\frac{1}{a}+\tan^{-1}\frac{1}{b}=\tan^{-1}\frac{1}{4}$$ and so on.

If $$\tan^{-1}\frac{1}{a}+\tan^{-1}\frac{1}{b}=\tan^{-1}\frac{1}{c}$$ what do we know about the relative sizes of \(a\), \(b\), and \(c\)?

Well, the angle on the right of the equation is clearly bigger that either of the angles on the left, so $$\tan^{-1}\frac{1}{a}<\tan^{-1}\frac{1}{c}$$ $$\tan^{-1}\frac{1}{b}<\tan^{-1}\frac{1}{c}$$ so $$\frac{1}{a}<\frac{1}{c}$$ $$\frac{1}{b}<\frac{1}{c}$$ so $$c< a$$ $$c< b$$ Of course \(a\) and \(b\) can be in either order, so to make life manageable, lets agree to choose \(a\) and \(b\) so that \(a< b\). That means $$c< a< b$$

If we are looking for solutions to $$\tan^{-1}\frac{1}{a}+\tan^{-1}\frac{1}{b}=\tan^{-1}\frac{1}{c}$$ then we can make our lives easier by starting with a possible value for \(a\) so that we have a definite upper bound for \(c\).

For example, take \(a=4\). Then $$\tan^{-1}\frac{1}{4}+\tan^{-1}\frac{1}{b}=\tan^{-1}\frac{1}{c}$$ which means that $$\frac{4b-1}{4+b}=c$$ We now know that \(c< 4\), so we only need to try \(2\) and \(3\) for \(c\).

Rearranging, we get $$b=\frac{4c+1}{4-c}=\frac{9}{2}\text{ or }\frac{13}{1}$$ from which we can see that the only possible values for \(b\) and \(c\) are \(13\) and \(3\), giving $$\tan^{-1}\frac{1}{4}+\tan^{-1}\frac{1}{13}=\tan^{-1}\frac{1}{3}$$

Try the same process for \(a=5,\;6,\;\dots\)


When \(a=5\), we have \(c=2, 3, \text{ or }4\) and $$b=\frac{5c+1}{5-c}=\frac{11}{3},\;\frac{16}{2},\;\frac{21}{1}$$ So $$\tan^{-1}\frac{1}{5}+\tan^{-1}\frac{1}{8}=\tan^{-1}\frac{1}{3}$$ $$\tan^{-1}\frac{1}{5}+\tan^{-1}\frac{1}{21}=\tan^{-1}\frac{1}{4}$$ and the first of those is one that we have already seen.

Arctan meets Fibonacci

Here are the Fibonacci numbers: $$0,\;1,\;1,\;2,\;3,\;5,\;8,\;13,\;21,\;34,\;55,\;80,\;\dots$$ So far, we have seen that $$\tan^{-1}\frac{1}{1}=\tan^{-1}\frac{1}{2}+\tan^{-1}\frac{1}{3}$$ $$\tan^{-1}\frac{1}{3}=\tan^{-1}\frac{1}{5}+\tan^{-1}\frac{1}{8}$$ $$\tan^{-1}\frac{1}{8}=\tan^{-1}\frac{1}{13}+\tan^{-1}\frac{1}{21}$$ and even $$\tan^{-1}\frac{1}{0}=\tan^{-1}\frac{1}{1}+\tan^{-1}\frac{1}{1}$$ So it looks as though the denominators are three consecutive Fibonacci numbers in order.
But it's not every group of three. Only every other one.

Using \(F_n\) for the \(n^{\mathrm{th}}\) Fibonacci number, and taking \(F_0=0\), can you form a hypothesis by generalising this observation?

We know that $$\tan^{-1}\frac{1}{F_2}=\tan^{-1}\frac{1}{F_3}+\tan^{-1}\frac{1}{F_4}$$ $$\tan^{-1}\frac{1}{F_4}=\tan^{-1}\frac{1}{F_5}+\tan^{-1}\frac{1}{F_6}$$ $$\tan^{-1}\frac{1}{F_6}=\tan^{-1}\frac{1}{F_7}+\tan^{-1}\frac{1}{F_8}$$ and even $$\tan^{-1}\frac{1}{F_0}=\tan^{-1}\frac{1}{F_1}+\tan^{-1}\frac{1}{F_2}$$ So to generalise this, we could try: $$\tan^{-1}\frac{1}{F_{2n}}=\tan^{-1}\frac{1}{F_{2n+1}}+\tan^{-1}\frac{1}{F_{2n+2}}$$

How could we test this hypothesis?


Remember from earlier that $$c=\frac{ab-1}{a+b}$$ So if our hypothesis is true, then $$F_{2n}=\frac{F_{2n+1}F_{2n+2}-1}{F_{2n+1}+F_{2n+2}}$$ Is this always true? If so, then $$\begin{aligned}&F_{2n}=\frac{F_{2n+1}F_{2n+2}-1}{F_{2n+1}+F_{2n+2}}\\[1em] \Rightarrow &F_{2n}\left(F_{2n+1}+F_{2n+2}\right)=F_{2n+1}F_{2n+2}-1\\ \end{aligned} $$

Simplify this as far as you can, remembering that \(F_{2n+2}=F_{2n+1}+F_{2n}\).


So many subscripts! Let's make life easier. Use: $$F_{6}\left(F_{7}+F_{8}\right)=F_{7}F_{8}-1$$ Then $$\begin{aligned} &F_{6}\left(F_{7}+F_{8}\right)=F_{7}F_{8}-1\\[1em] \Rightarrow &F_{6}\left(F_{7}+F_{8}\right)=F_{7}\left(F_{6}+F_{7}\right)-1\\[1em] \Rightarrow &F_6F_8=F_7^2-1\\[1em] \Rightarrow &F_7^2-F_6F_8=1\\[1em] \end{aligned}$$ or more generally $$F_{2n+1}^2-F_{2n}F_{2n+2}=1$$ So if we can prove that this is always true for Fibonacci numbers, then we will have proved that our hypothesis is correct.

This formula goes by the name "Cassini's Identity". Cassini had many interests in mathematics, astronomy, and engineering, but he is now best known for his work in astronomy at a time (he was just a little older than Newton) when our knowledge of the world was being firmly placed within a rational, scientific framework. This was the scientific revolution.

So much for Cassini the man. Look him up! But what about his identity. Since I've already told you it has a name, it must certainly be true, but proving it is not totally straightforward.



First of all, let's check that \(F_7^2-F_6F_8=1\).

Well, \(F_7^2-F_6F_8=13^2-8*21=1\), so yes, it is true. And we could try as many other cases as we like and find it to always be true. But that's not a proof.

This is a classic case for induction. To see how this is going to work, well start with a relatively easy case:

Using the equation \(F_7^2-F_6F_8=1\) and the recurrence relation for Fibonacci numbers, prove (without resorting to raw calculation) that \(F_9^2-F_8F_{10}=1\)


Of course it's easy with raw calculation, but that will not help us see how to carry out the induction. I must admit, I did not find this at all easy! But eventually I figured out that I shouldn't try to do too many things in one go, and should start just by dealing with \(F_6\), which is the one furthest from the solution. So here goes: $$\begin{aligned}F_7^2-F_6F_8&=F_7^2-(F_8-F_7)F_8\\[1em] &=F_7^2-F_8^2+F_7F_8\\[1em] &=F_7(F_7+F_8)-F_8^2\\[1em] &=F_7F_9-F_8^2\\[1em] &=(F_9-F_8)F_9-F_8^2\\[1em] &=F_9^2-F_8(F_9+F_8)\\[1em] &=F_9^2-F_8F_{10} \end{aligned}$$ And that, in a nutshell, is the inductive step of the proof by induction. All you need to do is to replace \(6\) with \(2k-2\), replace \(7\) with \(2k-1\), replace \(8\) with \(2k\) and so on. Then of course you will need the logical niceties of a proof by induction, but I am sure you can take care of those details!

So what have we achieved? We have discovered that $$\tan^{-1}\frac{1}{F_{2n}}=\tan^{-1}\frac{1}{F_{2n_+1}}+\tan^{-1}\frac{1}{F_{2n+2}}$$ but that there are also plenty of other integer solutions to $$\tan^{-1}\frac{1}{F_c}=\tan^{-1}\frac{1}{F_a}+\tan^{-1}\frac{1}{F_b}$$ The final flourish, though, is perhaps the most amazing of all.

$$\begin{aligned} 45^\circ&=\tan^{-1}\frac{1}{2}+\tan^{-1}\frac{1}{3}\\[1em] &=\tan^{-1}\frac{1}{2}+\tan^{-1}\frac{1}{5}+\tan^{-1}\frac{1}{8}\\[1em] &=\tan^{-1}\frac{1}{2}+\tan^{-1}\frac{1}{5}+\tan^{-1}\frac{1}{13}+\tan^{-1}\frac{1}{21} \end{aligned}$$

Can you extend this to find an infinite series that sums to \(45^\circ\) or, if you prefer (which you do), to \(\displaystyle\frac{\pi}{4}\)?



$$\frac{\pi}{4}=\sum_{r=1}^\infty \tan^{-1}\frac{1}{F_{2r+1}}$$