5

Is there way to show $$\int_{1}^{0}\frac{\ln(1-x)}{x}dx=\frac{\pi ^{2}}{6}$$ without using the Riemann zeta function?

StubbornAtom
  • 17,052
esege
  • 3,621

4 Answers4

4

Note: As pointed out by Hans, this answer uses the fact that $\zeta(2) = \dfrac{\pi ^2}{6}$ and therefore isn't what the OP asked for.

$$\frac{\ln(1+x)}{x} = 1 - \frac{x}{2} + \frac{x^2}{3}...$$

$$\int\frac{\ln(1+x)}{x}dx = x - \frac{x^2}{4} + \frac{x^3}{9}...+C$$

$$\int\limits^{0}_{1}\frac{\ln(1+x)}{x}dx = -1 + \frac{1}{4} - \frac{1}{9}...$$


After question edit:

We have $$\frac{\ln(1-x)}{x} = - 1 + \frac{x}{2} - \frac{x^2}{3}...$$

$$\int\frac{\ln(1-x)}{x}dx = -x + \frac{x^2}{4} - \frac{x^3}{9}...+C$$

Therefore $$\int\limits^{0}_{1}\frac{\ln(1-x)}{x}dx = 1 + \frac{1}{4} + \frac{1}{9}...$$

$$=\frac{\pi ^2}{6}$$

Max Payne
  • 3,447
  • 3
    The idea is correct, but you need some further argument to interchange the integral with the limit. You can integrate from t to 0 and interchange the limit, since the Talor series converges uniformly. Then you can take the limit for $t \searrow 1$ by Abel's theorem. – Dominik Feb 03 '16 at 10:00
  • @Dominik Actually I don't know about abel's theorem. – Max Payne Feb 03 '16 at 10:36
  • It's a theorem that states under which conditions a power series is continuous at the border of the region of concergence (c.f. Wikipedia). Without some kind of argument like this you can't interchange the limit of the power series and the integration. The typical theorems (i.e. dominated convergence) can't be apllied here. – Dominik Feb 03 '16 at 11:40
  • 1
    I suspect that this is just what the OP meant by "using the Riemann zeta function"... (I.e., using that we know that $\zeta(2)=\pi^2/6$.) – Hans Lundmark Feb 03 '16 at 12:00
  • @HansLundmark Yes you are right. This makes my post obsolete now. – Max Payne Feb 03 '16 at 12:39
4

(I assume that what you mean is that you don't want to use the fact that $\sum 1/n^2=\pi^2/6$.)

Yes, there is a way of seeing this, and this is the basis of Mikael Passare's paper How to compute $\sum 1/n^2$ by solving triangles (free preprint on arXiv, or the published version on JSTOR for subscribers).

If you set $x=e^{-t}$, your integral becomes $$ \int_0^{\infty} -\ln(1-e^{-t}) \, dt , $$ or $$ \int_0^{\infty} -\ln(1-e^{-x}) \, dx $$ if we call the variable $x$ again. This is the area of a region in the first quadrant of the $xy$-plane, below the curve $y=-\ln(1-e^{-x})$ (or equivalently $e^{-x}+e^{-y}=1$). This is the region called $U_0$ i Passare's paper, and he shows that there are two other regions $U_1$ and $U_2$ with the same area, and then via a clever area-preserving change of variables that the combined area of $U_0$, $U_1$ and $U_2$ equals the area of a certain right triangle $T$ which is obviously $\pi^2/2$. Hence the area of $U_0$ (your integral) is $\pi^2/6$.

(See also this answer and this.)

Hans Lundmark
  • 53,395
1

To expand on Tim's solution:

The series expansion of the logarithm yields the following representation for all $x \in [-1, 1)$: $$\log(1 - x) = -\sum \limits_{n = 1}^\infty \frac{x^n}{n}$$

This means that we need to calculate the following integral: $$\int_0^1 \sum \limits_{n = 1}^\infty \frac{x^{n - 1}}{n} \; dx$$

Ususally you would apply the dominated convergence theorem to interchange the integral and the limit at this point. However, the series doesn't have an integrable majorant [I didn't check this rigorously, but I'm pretty sure this is the case].

We circumvent this by writing the integral as a limit: $$\int_0^1 \sum \limits_{n = 1}^\infty \frac{x^{n - 1}}{n} \; dx = \lim \limits_{t \nearrow 1} \int_0^t \sum \limits_{n = 1}^\infty \frac{x^{n - 1}}{n} \; dx$$

Now since the Taylor series converges uniformly on every interval $[0, t]$ for $0 < t < 1$, we are allowed to interchange the sum and the integral: $$\lim \limits_{t \nearrow 1} \int_0^t \sum \limits_{n = 1}^\infty \frac{x^{n - 1}}{n} \; dx = \lim \limits_{t \nearrow 1} \sum \limits_{n = 1}^\infty\int_0^t \frac{x^{n - 1}}{n} \; dx = \lim \limits_{t \nearrow 1}\sum \limits_{n = 1}^\infty \frac{t^n}{n^2}$$

Finally we can interchange the limit and the sum by Abel's theorem: $$\lim \limits_{t \nearrow 1}\sum \limits_{n = 1}^\infty \frac{t^n}{n^2} = \sum \limits_{n = 1}^\infty \frac{1}{n^2} = \frac{\pi^2}{6}$$

Dominik
  • 19,963
0

Using much the same method as above, for the required integral $\int_{1}^{0}\dfrac{\ln(1-x)}{x}dx$ it "feels reasonable" to first make the substitution $1-x = e^{-z}$ then $dx|_{x=1}^{x=0} = e^{-z}dz|_{z=\infty}^{z=0}$ so that the integral can be written as $\int^{\infty}_{0}\dfrac{z e^{-z}}{1-e^{-z}}dz$.

After this expand $\dfrac{1}{1-e^{-z}} = \sum_{n=0}^{n=\infty} e^{-n z}$. Eventually have to deal with the sum $\sum_{n=0}^{n=\infty} \dfrac{1}{(n+1)^2}$ which is identified as $\pi^2 / 6$.

jim
  • 1,214