I'm reading Bellard's proof for his eponymous formula computing pi digits, and I can't get past the first line.

Given that:

  • $\displaystyle-\ln(1-x) = \sum_{n=1}^\infty \frac{x^n}{n}$ for $|x| < 1$,
  • $\arctan(y) = \Im[-\ln(1-iy)]$ for any real $y$

(I didn't know the last one but Wolfram Alpha agrees, so okay), the author gets

$$\arctan\left(\frac{1}{a-1}\right) = \Im\left[-\ln\left(1- \frac{1+i}{a}\right) \right] = \sum_{n=0}^\infty \frac{(-1)^n2^{2n}}{a^{4n+3}}\left( \frac{a^2}{4n+1} + \frac{2a}{4n+2} + \frac{2}{4n+3}\right)$$

In this equation, I don't even get the first equality, and assuming it to be correct, I don't see how to derive the second equality... I've tried expanding using Newton's theorem and expanding the binomial coefficients, but it did not work.

Any ideas?


A friend nearly found the solution for the second equality (up to some mysterious sign error), and the first equality is actually invalid when $a \in (0, 1)$, according to Wolfram Alpha: on this interval, we rather have $\arctan\left(\frac{1}{a-1}\right) = \Im\left[-\ln\left(1- \frac{1+i}{a}\right) \right] \color{red}{+ \pi}$ .

So I still don't know why $\arctan\left(\frac{1}{a-1}\right) = \Im\left[-\ln\left(1- \frac{1+i}{a}\right) \right]$ for $a \notin (0,1)$, but given that Bellard uses $a = 2$, let's assume it's fine. However I'm interested if someone could find the solution to this problem.

Now, from $\Im\left[-\ln\left(1- \frac{1+i}{a}\right) \right]$ to $\displaystyle\sum_{n=0}^\infty \frac{(-1)^n2^{2n}}{a^{4n+3}}\left( \frac{a^2}{4n+1} + \frac{2a}{4n+2} + \frac{2}{4n+3}\right)$:

$$\begin{align*} -\ln\left(1 - \frac{1+i}{a}\right) &= \sum_{k=1}^\infty \frac{1}{k}\left(\frac{1+i}{a}\right)^k \\ &= \sum_{n=0}^\infty \left[ \frac{1}{4n+1} \left( \frac{1+i}{a}\right)^{4n+1} + \dotso + \frac{1}{4n+4} \left( \frac{1+i}{a}\right)^{4n+4}\right]\\ &= \sum_{n=0}^\infty\left( \frac{1+i}{a}\right)^{4n+1}\left[ \frac{1}{4n+1}+ \dotso + \frac{1}{4n+4}\left(\frac{1+i}{a}\right)^3 \right] \end{align*}$$

We remark that $1 + i = \sqrt{2} e^{i\pi / 4}$ so :

$$\begin{align*} -\ln\left(1 - \frac{1+i}{a}\right) &= \sum_{n=0}^\infty \frac{\left( \sqrt{2} e^{i \pi/4}\right)^{4n+1}}{a^{4n+1}}\left[ \frac{1}{4n+1}+ \dotso + \frac{1}{4n+4}\frac{\left( \sqrt{2} e^{i \pi/4}\right)^3}{a^3} \right]\\ &=\sum_{n=0}^\infty \frac{2^{2n}(-1)^n}{a^{4n+1}}\left[ \frac{\left( \sqrt{2} e^{i \pi/4}\right)}{4n+1}+ \dotso + \frac{1}{4n+4}\frac{\left( \sqrt{2} e^{i \pi/4}\right)^4}{a^3} \right]\\ \end{align*}$$

Taking the imaginary part, knowing $\sqrt{2} e^{i \pi/4} = 1+i$ and $(1+i)^2 = 2i$, $(1+i)^3 = 2 - 2i$ and $(1+i)^4 = -2$:

$$\begin{align*} \Im \left[-\ln\left(1 - \frac{1+i}{a}\right)\right] &= \sum_{n=0}^\infty \frac{2^{2n}(-1)^n}{a^{4n+1}}\left[ \frac{1}{4n+1}+ \frac{2}{4n+2}\frac{1}{a} - \frac{2}{4n+3} \frac{1}{a^2}\right]\\ &= \sum_{n=0}^\infty \frac{2^{2n}(-1)^n}{a^{4n+3}}\left[ \frac{a^2}{4n+1}+ \frac{2a}{4n+2} \color{red}- \frac{2}{4n+3}\right]\\ \end{align*}$$

So everything is good, except for the last minus sign...


1. The $\operatorname{arctan}$ formulas.

Consider what it means for a complex number $u$ to be a logarithm of $z$: essentially, that $e^u = z$.

Since $e^u = e^{\Re u + i \Im u} = e^{\Re u} e^{i\Im u} = r e^{i \theta}$ you see that pretty much by definition, $\Im \ln(z) = \Im u = \theta$ is the angular component of $z$ in the complex plane.

Now, take the point $z = 1-iy$, is it an elementary trigonometric exercise to check that the angle forme between the real axis and $z$ is $\theta$ such that $\tan\theta = -y$. By the above remark this means exactly that $\Im \ln(1-iy) = \arctan(-y)$.

Since $\arctan$ is an odd function, this means that $-\Im \ln(1-iy) = \arctan(y)$.

The very same argument works for $z = 1 - (1+i)/a$. Indeed, this time we have

$$\begin{align*} \tan \theta & = \frac{\Im z}{\Re z} \\ & = \frac{-1/a}{1 - 1/a} \\ & = \frac{-1}{a - 1} \end{align*} $$ And therefore, $-\Im\ln(1 - (1+i)/a) = \arctan(1/(a-1))$.

2. The summation formula

You have it right, except that $(1+i)^3 = 2i(1+i) = -2 + 2i$. And not $2 - 2i$ as you wrote, which was the source of this erroneous sign.

Well done! For additional fun formulas and the rationale behind them (in French), you can have a look at this pdf from the olden days.


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.