10

Prove that $$\sum_{n=0}^\infty \frac{(-1)^n}{3n+1} = \frac{\pi}{3\sqrt{3}}+\frac{\log 2}{3}$$

I tried to look at $$ f_n(x) = \sum_{n=0}^\infty \frac{(-1)^n}{3n+1} x^n $$

And maybe taking its derivative but it didn't work out well.

Any ideas?

RobPratt
  • 50,938
AlonAlon
  • 2,706

3 Answers3

10

For a geometric series.

$$\sum_{n=0}^{\infty} (-1)^n(x^n) = \frac{1}{1+x}$$

Substitute $x \rightarrow x^3$

$$\sum_{n=0}^{\infty} (-1)^n (x^{3n}) = \frac{1}{1+x^3}$$

Integrate the sides.

$$\sum_{n=0}^{\infty} (-1)^n\frac{x^{3n + 1}}{3n + 1} = \int_0^x \frac{dt}{1+t^3}$$

The hard part is the integration. Then let $x=1$

GEdgar
  • 117,296
Amad27
  • 11,551
9

Hint

$$\sum_{n=0}^\infty\frac{(-1)^n}{3n+1}=\lim_{N\to\infty}\sum_{n=0}^N(-1)^n\int_0^1t^{3n}dt=\lim_{N\to\infty}\int_0^1\sum_{n=0}^N(-t^3)^ndt=\lim_{N\to\infty}\int_0^1\frac{1-(-t^3)^{N+1}}{1+t^3}dt$$

Now you need to prove two things:

  • $$\int_0^1\frac{dt}{1+t^3}=\frac{\pi}{3\sqrt{3}}+\frac{\log 2}{3}$$
  • $$\lim_{N\to\infty}\int_0^1\frac{(-t^3)^{N+1}}{1+t^3}dt=0$$
  • Thank you. Can you explain the second equality please? – AlonAlon Dec 13 '14 at 17:11
  • 1
    Which equality do you mean? Interchanging $\sum$ and $\int$? It's possible with finite sum. –  Dec 13 '14 at 17:13
  • 1
    @AlonAlon Since $0<t<1$, then as $N\to\infty$ the integrand goes to $0$ – Anastasiya-Romanova 秀 Dec 13 '14 at 17:14
  • You're welcome @AlonAlon –  Dec 13 '14 at 17:14
  • Is it legal: $\lim_{N\to\infty} \int_0^1 (-t^3)^{N+1} = \int_0^1 0 = 0$? (Meaning, talking the limit before evaluating the integral) – AlonAlon Dec 13 '14 at 17:28
  • 3
    To prove the second point use that $$\left|\int_0^1\frac{(-t^3)^{N+1}}{1+t^3}dt\right|\le\int_0^1 t^{3N+3}dt=\frac1{3N+4}$$ and then squeeze. –  Dec 13 '14 at 17:34
  • Thanks, Why doing it my way is wrong by the way? – AlonAlon Dec 13 '14 at 17:37
  • 1
    We can't interchange in general $\lim$ and $\int$ without theorems: Dominated convergence theorem, limit monotone theorem, convergence uniform theorem,... Or prove the result by hand as I did. –  Dec 13 '14 at 17:42
  • This is a very nice approach especially I like the second part where the remainder is shown as tending to $0$. This is the way such elementary problems should be be done (i.e. without any appeal to uniform convergence, dominated convergence etc). +1 – Paramanand Singh Dec 14 '14 at 04:22
  • But Fubini and Tonelli's theorems justifies it for all sums/integrals less than infinity. These are less than infinity because they converge. We can directly apply that. – Amad27 Dec 26 '14 at 08:56
4

Provided below is an alternate solution that came to my mind first that involves complex numbers as opposed to integration.

Consider the Taylor series of the principal value of $\ln(1+z)$, which is:

$$ \ln(1+z)=\sum_{n=0}^\infty \frac{(-1)^n}{n+1}z^{n+1} $$

Let $\omega=\exp(2\pi i/3)$ be a third root of unity. With this, we have:

$$ \begin{align*} \ln(1+1)&=\sum_{n=0}^\infty(-1)^n\left(\frac{1}{3n+1}-\frac{1}{3n+2}+\frac{1}{3n+3}\right) \\ \omega^2\ln(1+\omega)&=\sum_{n=0}^\infty(-1)^n\left(\frac{1}{3n+1}-\frac{\omega}{3n+2}+\frac{\omega^2}{3n+3}\right) \\ \omega\ln(1+\omega^2)&=\sum_{n=0}^\infty(-1)^n\left(\frac{1}{3n+1}-\frac{\omega^2}{3n+2}+\frac{\omega}{3n+3}\right) \\ \end{align*} $$ Since $1+\omega+\omega^2=0$, we have $S=\frac13(\ln 2 + \omega^2\ln(1+\omega)+\omega\ln(1+\omega^2))$. Note that $\ln z = \ln|z|+i\arg z$ if we're considering the principal value. We can now simplify this like so: $$ \begin{align*} S &= \frac13\left(\ln 2 + \omega^2\ln(1+\omega)+\omega\ln(1+\omega^2)\right) \\ &= \frac13\left(\ln 2 + \omega^2\ln(-\omega^2)+\omega\ln(-\omega)\right) \\ &= \frac13\left(\ln 2 + \overline{\omega\ln(-\omega)}+\omega\ln(-\omega)\right) \\ &= \frac13\left(\ln 2 + 2\mathrm{Re}[\omega\ln(-\omega)]\right) \\ &= \frac13\left(\ln 2 + 2\left(-\sin\left(\frac{2\pi}{3}\right)\arg(-\omega)\right)\right) \\ &= \frac13\left(\ln 2 + 2\left(-\frac{\sqrt3}{2}\cdot-\frac\pi3\right)\right) \\ &= \frac13\left(\ln 2 + \frac{\pi}{\sqrt3}\right) \\ \end{align*} $$

As a bonus, you can extend this logic to find a general expression for all odd coefficients of $n$, though it certainly isn't pretty. I'll leave what I got here, to save other people's time:

$$ \sum_{n=0}^\infty \frac{(-1)^n}{pn+1} = \int_0^1 \frac{dx}{x^p+1} = \frac{1}{p}\left(\ln 2 + 2\sum_{k=1}^{(p-1)/2}\left(\cos\frac{2\pi k}{p}\ln\left(2\cos\frac{\pi k}{p}\right)+\pi\sin\frac{2\pi k}{p}\left(\frac{k}{p}-\left\lceil\frac{k}{p}-\frac{1}{2}\right\rceil\right)\right)\right) $$

  • 1
    Interestingly, I couldn't make popular CAS's obtain $$\sum_{n=0}^\infty\frac{(-1)^n}{5n+1}=\frac15\ln2+\frac1{\sqrt5}\ln\frac{1+\sqrt5}2+\frac\pi5\sqrt{\frac12\left(1+\frac1{\sqrt5}\right)}$$ (the most compact form I managed to get myself). – metamorphy Jan 03 '25 at 15:47
  • @metamorphy No clue how those work under the hood, but it could be related to the alternate integral solution. WolframAlpha can integrate $1/(x^3+1)$ just fine but can't give an exact answer for $1/(x^5+1)$. I tried plugging a similar $S$ expression from my method in too and it gives out a few expressions, but all contain roots of $-1$ and it approximates numerically to $0.888314-7.77156\times10^{-17}i$. – mediocrevegetable1 Jan 03 '25 at 16:05
  • "WolframAlpha can integrate $1/(x^3+1)$ just fine but can't give an exact answer for $1/(x^5+1)$." Are you kidding me? The integration form would seem simple enough to me. You generate square roots of complex numbers, true; but unlike odd order roots thsse are algebraically resolvable. – Oscar Lanzi Jan 03 '25 at 16:31
  • 2
    The last formula is essentially derived from Gauss's Digamma Theorem, which is more important to be mentioned (rather than throwing the formula). $$\sum_{n=0}^{\infty}\frac{(-1)^n}{pn+1}=\frac{1}{2p}\sum_{n=0}^{\infty}\left(\frac{1}{n+\frac{1}{2p}}-\frac{1}{n+\frac{p+1}{2p}}\right)=\frac{1}{2p}\left(\psi\left(\frac{p+1}{2p}\right)-\psi\left(\frac{1}{2p}\right)\right)$$ – Quý Nhân Jan 03 '25 at 22:12
  • @QuýNhân very interesting, not something I even thought of applying here. I would honestly say that warrants its own answer, it's pretty elegant for solving the original problem too. – mediocrevegetable1 Jan 03 '25 at 22:18