21
$\begingroup$

Sum the following: \begin{align} S &= \frac{1}{9} + \frac{1}{99} + \frac{1}{999} + \frac{1}{9999} + \cdots\\[0.1in] &= \sum_{n=1}^{\infty} \frac{1}{{10}^n - 1} \end{align}

It's fairly straightforward to show that this sum converges: \begin{align} S &= \sum_{n=1}^{\infty} \frac{1}{{10}^n - 1}\\ &= \frac{1}{9} + \sum_{n=2}^{\infty} \frac{1}{{10}^n - 1}\\ &< \frac{1}{9} + \sum_{n=2}^{\infty} \frac{1}{{10}^n - 10}\\ &= \frac{1}{9} + \frac{1}{10}\sum_{n=2}^{\infty} \frac{1}{{10}^{n-1} - 1}\\ &= \frac{1}{9} + \frac{1}{10}S\, , \end{align} which leads to $S < 10/81 = 0.\overline{123456790}$.

Numerically (Mathematica), we find that this sum is approximately $S \approx 0.122324$.

This sum can also be written as \begin{align} S &= \sum_{n=1}^{\infty} \frac{1}{{10}^n - 1}\\ &= \sum_{n=1}^{\infty} \frac{1/{10}^n}{1 - 1/{10}^n}\\ &= \sum_{n=1}^{\infty}\sum_{m=1}^{\infty} \frac{1}{{10}^{nm}} \end{align}

(The last expression above is itself quite amusing, since the coefficient of $1/{10}^k$ is the number of distinct ways of writing $k$ as a product of two positive integers.)

Analytically, Mathematica evaluates this sum as \begin{equation} S = \frac{\ln(10/9) - \psi_{1/10}(1)}{\ln(10)}\, , \end{equation} where $\psi_q(z)$ is the Q-Polygamma function. This is not really a nice, tidy, closed-form solution. Now, I can accept that this sum may not have such a nice, tidy sum, but something about it feels like it should have one. (Non-rigorous, I know!) Furthermore, I'm aware that Mathematica is by no means infallible, especially when it comes to simplification of certain expressions.

So I'm wondering if a neater solution exists.

$\endgroup$
  • 2
    $\begingroup$ Wolfram returns the same result. Perhaps if you find a closed form for $\psi_{\frac{1}{10}}(1)$, if that is even possible, then you can have your nice closed-form solution. But I doubt such a closed form exists for $\psi_{\frac{1}{10}}(1)$. $\endgroup$ – Frpzzd Jun 17 '17 at 15:30
  • 2
    $\begingroup$ If you write decimal expansions of the terms (e.g., $\frac19 = .1111111...$; or $\frac1{99} = .010101...$), you can see that the sum is equal to $\sum_1^\infty \frac{d(n)}{10^n}$ where $d(n)$ is the divisor counting function. $\endgroup$ – paw88789 Jun 17 '17 at 15:57
  • 2
    $\begingroup$ @paw88789 That is already stated by the OP in the question. $\endgroup$ – WimC Jun 17 '17 at 16:27
  • 3
    $\begingroup$ As was foreseeable, this type of series has a name and this is about all what one can say. (Why the avalanche of upvotes?) $\endgroup$ – Did Jun 17 '17 at 18:02
  • 1
    $\begingroup$ The smallest number with more than $9$ divisors is $48=2^4\cdot3^1$, so the first 46 digits of the sum are simply the number of divisors of the digit position. $\endgroup$ – robjohn Jun 17 '17 at 19:07
2
$\begingroup$

The given series is an irrational number (probably a trascendental number, too) and for its numerical evaluation it is possible to exploit some tailor-made acceleration techniques (like the one outlined here for the base-$2$ analogue). Besides that, I am not aware of any nice closed form for $\sum_{n\geq 1}\frac{d(n)}{10^n}$.

$\endgroup$
  • $\begingroup$ Is there a quick way to see why it has to be irrational? $\endgroup$ – John Barber Jun 18 '17 at 2:43
  • $\begingroup$ @JohnBarber: its decimal expansion is highly aperiodic! $\endgroup$ – Jack D'Aurizio Jun 18 '17 at 2:47

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.