# Does this weird series converge?

$\sum_{n\in S}$$\frac{1}{n}, where S consists of those positive integers whose decimal expansion does not contain the digit 1. This was a part(b) question. Part (a) was an evaluation of the double integral \int \int_S (x^2 + y^2)d\sigma, where S is the unit sphere centered at (0,0,0), and \sigma is surface area. The answer is \frac{8\pi}{3}. I don't see how part(a) can be used to solve part(b). For the series evaluation, if we are summing over the positive integers that don't contain the digit 1, then I'm thinking we exclude integers such as 1, which has decimal expansion .99999.. or 1.0000, or 10, which has expansion 9.9999... or 10.0000, etc, and we sum over numbers such as 2, 3, 4, 5, ... Not sure where I can go with this idea, though. Thanks, • Part (b) is unrelated to (a). Yes, the series in part (b) converges. This sort of series actually has a name. Look at the wiki entry for Kempner series for some info. – achille hui Jun 24 '15 at 8:20 • Hi @achillehui, I think I may be counting wrong and may be interpreting "decimal expansion" incorrectly. ... does the set S contain the integers 2, 12, 22, 32, 42 ... etc? I am thinking "no", since 2's expansion is either 1.999 ... or 2.000... so that from 1 to 10, there are 7, not 8, integers contained in the set S. – User001 Jun 25 '15 at 0:59 • actually, maybe that's only the case for the number 1, which has expansion either .999... or 1.000... , but this doesn't mean that 2 has expansion 1.999 nor does it mean that 22 has expansion 21.999... Wiki says vaguely that there exist numbers with more than one decimal expansion, and the Wiki page gives .999... = 1 as a specific example. – User001 Jun 25 '15 at 1:08 ## 2 Answers For part (b), S has 8 one digit integers, 8 \cdot 9 two digit integers, 8 \cdot 9^2 three digit integers, etc. So, the sum \displaystyle\sum_{n \in S}\dfrac{1}{n} has 8 terms that are at most \dfrac{1}{2}, 8 \cdot 9 terms that are at most \dfrac{1}{20}, 8 \cdot 9^2 that are at most \dfrac{1}{200}, etc. Can you do something with this? The formula for the sum of a geometric series will also be helpful. I'm also not sure how parts (a) and (b) are related. It's entirely possible that they are not. • Your upper bounds \frac{1}{20},\frac{1}{200},\frac{1}{2000} are off by a factor of 10. – TonyK Jun 24 '15 at 8:35 • ^Thanks for catching that. – JimmyK4542 Jun 24 '15 at 16:20 • Hi @JimmyK4542, how do you count the numbers, at least in the beginning? we exclude 1 and 10 and keep 2 through 9, which gives 8 1-digit integers in S. The number 2 doesn't have expansion 1.999... , right? I think I'm incorrectly thinking this, because of the example, .999... = 1. – User001 Jun 25 '15 at 1:12 • I think I got it now, for the most part. Thanks so much for the simple intuition, @JimmyK4542. – User001 Jun 25 '15 at 1:37 Let me reformulate my answer. Now let S be the subset of \mathbb N whose decimal expansion do not contain digit 1. The answer is the series \sum_{n\in S}\frac{1}{n} is convergent. In contrast, think about the fact \sum_{p\in prime}\frac{1}{p}=\infty, and the sparsity of primes when they are large. It shows that S occupies a extremely "meager" set in \mathbb N. Before we come to the proof, here are two observations: (1)Think of this question: If we take decimal expansion of each real x\in [0,1], x=\alpha_1\alpha_2...\alpha_n... , where \alpha_i\in \{0,1,..,9\}, then remove those reals whose expansion ever contains digit 1, what set will we get？ The answer is we will get a measure zero set, since it is a variant of Cantor set whereas the Cantor set takes ternary expansion. By this observation, we have a strong feeling that S, those integer do not contain digit 1, only occupies a "measure zero" set in \mathbb N. (2)Let T_n be the set of integer contains only n-digit integer, i.e.$$T_n=\{k\in \mathbb N| 10^n\le k\le10^{n+1}-1\}.$$and let S_n=T_n \cap S. So S=\cup_{n\in \mathbb N}S_n. We have$${{|S_n|}\over{|T_n|}}={{C_8^1(C_{9}^1)^{n-1}}\over {C_9^1(C_{10}^1)^{n-1}}}={{8\times 9^{n-1}}\over{9\times 10^{n-1}}}\le ({{9}\over{10}})^n$$The proportion of |S_n| in |T_n| will be small when n goes large. Finally, let's prove the claim that \sum_{n\in S}\frac{1}{n} is convergent, notice that each n\in S_n, n\ge 10^n. Then,$$\sum_{n\in S}\frac{1}{n}=\sum_{n\in\mathbb N}\sum_{k\in S_k}\frac{1}{k} \le\sum_{n\in\mathbb N}\sum_{k\in S_k}\ ({{1}\over{10}})^n \le\sum_{n\in\mathbb N}({{9}\over {10}})^n\times ({{1}\over{10}})^n =\sum_{n\in\mathbb N}({{9}\over{100}})^n =\frac{100}{91}.$\$

So the series is convergent.

• Thanks for your detailed answer, @yilongzhang. – User001 Jun 25 '15 at 1:38