5

Calculate $$\sum_{k=1}^{\infty}\frac{1}{2^{k+1}-1}.$$

I used Wolfram|Alpha to compute it and got it to be approximately equal to $0.6$. How to compute it? Can someone give me a hint or a suggestion to do it?

Clayton
  • 24,751
Simple
  • 3,593
  • @JackD'Aurizio: While I did NOT downvote, perhaps it was because of not "answering" the question. In your defense, however, I feel like you made the right decision to post it as an answer instead of a long comment. Anyways, it seems like good work. +$1$ :) – Clayton Oct 21 '16 at 08:34
  • About an explicit form, that is quite hopeless. It is not difficult to prove that $S$ is an irrational number (the infinitude of primes has a precise effect on the binary representation of $S$), and probably it is a trascendental number, too. – Jack D'Aurizio Oct 21 '16 at 08:34
  • @Jack D'Aurizio I am very surprised also by the downvote to your excellent answer. There is a general problem with downvoting freed on SE. Downvoters should not be anonymous. Besides, why have you deleted your answer ? – Jean Marie Oct 21 '16 at 08:36
  • 1
    @Clayton: it looks to me that I actually provided a way for computing seven figures. I do not care about the silly downvote, it is enough the OP is aware of the contents of my comment. – Jack D'Aurizio Oct 21 '16 at 08:36
  • 1
    @JeanMarie: the principle of anonymity of upvotes/downvotes cannot be violated, in order to prevent voting wars. But I think that anonymous explanations should be made compulsory. I do not know the reason behind the downvote, such phenomenon occurs to me quite often since a while. (I undeleted my answer, by the way) – Jack D'Aurizio Oct 21 '16 at 08:39

2 Answers2

14

It is a fast-convergent series, and a Lambert series, too, since

$$ S=\sum_{k\geq 1}\frac{1}{2^{k+1}-1}=-1+\sum_{k\geq 1}\frac{1}{2^k-1}=-1+\sum_{k\geq 1}\sum_{m\geq 1}\frac{1}{2^{mk}}=-1+\sum_{n\geq 1}\frac{d(n)}{2^n}$$ where $d(n)$ is the number of divisors of $n$. Since $d(n)\leq n$ (this is a very crude bound) $$ S+1-\sum_{n=1}^{N}\frac{d(n)}{2^n}\leq \sum_{n>N}\frac{n}{2^n}=\frac{N+2}{2^N} $$ hence by choosing $N=30$ we get that $$ -1+\sum_{n=1}^{30}\frac{d(n)}{2^n} = \color{red}{0.6066951}49\ldots $$ is an extremely good approximation of $S$, with the correct red digits.


As suggested by Yves Daoust, another good strategy comes from noticing that $\frac{1}{2^k-1}$ is pretty close to $\frac{1}{2^k}$ if $k$ is large, hence we may perform a series acceleration in the following way:

$$ S=\sum_{k\geq 2}\frac{1}{2^k-1}=\frac{1}{2^2-1}+\sum_{k\geq 3}\frac{1}{2^k}+\sum_{k\geq 3}\frac{1}{2^k(2^k-1)}$$ turning $S$ into $$ S = \frac{7}{12}+\frac{1}{2^3(2^3-1)}+\sum_{k\geq 4}\frac{1}{4^k}+\sum_{k\geq 4}\frac{1}{4^k(2^k-1)} $$ or $$ S = \frac{815}{1344}+\frac{1}{4^4(2^4-1)}+\sum_{k\geq 5}\frac{1}{8^k}+\sum_{k\geq 5}\frac{1}{8^k(2^k-1)} $$ so that $S$ equals $\frac{260927}{430080}$ plus $\sum_{k\geq 5}\frac{1}{8^k(2^k-1)}$. With just three iterations of this technique we get already $S=\color{red}{0.60669}41\ldots$, and and the fourth step we get $S\approx \frac{1391613}{2293760}=\color{red}{0.6066951}20\ldots$

In a compact form, this acceleration technique leads to: $$ S = \sum_{k\geq 1}\left(\frac{1}{2^{(k^2-1)}(2^{k+1}-1)}+\frac{1}{2^{(k^2+k)}(2^k-1)}\right) $$ collapsing to: $$\boxed{S=\frac{1}{4}+\sum_{k\geq 2}\frac{8^k+1}{(2^k-1)\,2^{k^2+k}}=0.6066951524152917637833\ldots} $$

with a significant convergence boost.
Now the main term of the sum behaves like $2^{-k^2}$ instead of $2^{-k}$.

Jack D'Aurizio
  • 353,855
  • I really don't understand the downvote so (+1). – Marco Cantarini Oct 21 '16 at 08:41
  • By ignoring the $-1$ at the denominator, the tail of the summation of the initial series after $n$ terms is close to $2^{1-n}$. This gives a very fast converging estimate ($12$ exact decimals with $19$ terms). –  Oct 21 '16 at 09:06
  • @YvesDaoust: that is a very good remark. For instance, it is not difficult to design a tailor-made acceleration technique for our series based on that. Answer updated, many thanks. – Jack D'Aurizio Oct 21 '16 at 09:18
  • Tremendous answer +1 (for what it's worth) –  Oct 21 '16 at 09:22
10

Maybe it is interesting to see that there is a “closed form” of this series in terms of the $q$-polygamma function. We have that $$S=\sum_{k\geq1}\frac{1}{2^{k+1}-1}=-1-\sum_{k\geq1}\frac{1}{1-2^{k}} $$ and recalling the definition of the $q$-polygamma function we have $$S=\color{red}{-1-\frac{\psi_{1/2}\left(1\right)+\log\left(1/2\right)}{\log\left(2\right)}}\approx0.60669515.$$

Marco Cantarini
  • 33,062
  • 2
  • 47
  • 93