As we know, there are many formulas of $\pi$ , one of them $$\frac{\pi ^2}{6}=\frac{1}{1^2}+\frac{1}{2^2}+\frac{1}{3^2}...... $$ and this $$\frac{\pi ^2}{8}=\frac{1}{1^2}+\frac{1}{3^2}+\frac{1}{5^2}......$$ Now,find the positive integer numbers $(a_{0}, a_{1}, a_{2}....)$ to get $$\frac{\pi^2 }{9}=\frac{1}{a_{0}^2}+\frac{1}{a_{1}^2}+\frac{1}{a_{2}^2}....$$
-
Can you specify what background you have in the relevant subjects as analysis or numbertheory? Do you already know where the first two 'formulas' come from? – flawr Nov 05 '14 at 14:32
-
2Why do you think such a sequence of integers exists? – MPW Nov 05 '14 at 14:33
-
In fact i think that writing in this form $\frac {\pi^2}{m}$ where $m$ is odd is an open problem. – Haha Nov 05 '14 at 14:46
-
Formula 2.5 in https://www.davidhbailey.com/dhbpapers/digits.pdf may be relevant here, although it has some numerators >1. – Jaume Oliver Lafont Jun 24 '20 at 11:07
6 Answers
From $$\frac{\pi^2}8=\sum_{2\nmid n}\frac1{n^2}$$ we find that $$\frac{\pi^2}{72}=\frac1{3^2}\sum_{2\nmid n}\frac1{n^2}=\sum_{2\nmid n, 3\mid n}\frac1{n^2}$$ and hence $$\begin{align}\frac{\pi^2}9=\frac{\pi^2}8-\frac{\pi^2}{72}&=\sum_{2\nmid n, 3\nmid n}\frac1{n^2}\\&=\frac1{1^2}+\frac1{5^2}+\frac1{7^2}+\frac1{11^2}+\frac1{13^2}+\frac1{17^2}+\frac1{19^2}+\frac1{23^2}+\frac1{25^2}+\ldots\end{align}$$
- 382,479
You don't actually need to know that $\sum\frac{1}{n^2}=\frac{\pi^2}{6}$ to get such a sequence. Instead, you can use this theorem (which is probably an exercise in some early chapter of Rudin):
Theorem: Let $\sum_k s_k$ be a convergent sum of positive numbers, say $S=\sum_k s_k$. Suppose that, for all $n$, we have $s_n < \sum_{k=n+1}^\infty s_k$. Then there is a subsequence of $s_k$ whose sum converges to $x$ for any positive $x < S$.
Proof: Construct the subsequence greedily. That is, given some partial sum $s_{k_1}+\dots+s_{k_{i-1}}$, let $k_i$ be the least integer greater than $k_{i-1}$ with $s_{k_1}+\dots+s_{k_i} < x$. As the $s_k$ become arbitrarily small, such a $k_i$ will always exist.
By construction, the partial sums of $\sum s_{k_i}$ are bounded above by $x$; as the $s_k$ are all positive, it follows that $\sum s_{k_i}$ converges to some number which is at most $x$.
Now, we will need the following lemma:
Lemma: The subsequence $s_{k_i}$ omits infinitely many elements of $s_k$.
Proof: As $0<x<S$, we can find $n_1$ such that $\sum_{k=1}^{n-1} s_k< x < \sum_{k=1}^{n_1} s_k$. Then by construction $s_{k_i}$ will include $s_1,\dots,s_{n_1-1}$ but not $s_{n_1}$, and so $s_{k_i}$ omits the element $s_n$. By hypothesis, $\sum_{k=n+1}^\infty s_k > s_n$, and so $$x-\sum_{k=1}^{{n_1}-1} s_k < s_{n_1} < \sum_{k=n_1+1}^\infty s_k$$ But then the same argument gives us some $n_2>n_1$ with $$ \sum_{k=n_1+1}^{n_2-1} s_k < x-\sum_{k=1}^{n_1-1} s_k < \sum_{k=n_1+1}^{n_2} s_k $$ and similarly $s_{k_i}$ will omit $s_{n_2}$. Proceeding in this manner, we obtain an infinite sequence $s_{n_j}$ of omitted terms. $\square$
With this lemma, the theorem is straightforward. Choose $\epsilon >0$. As there are infinitely many omitted $s_{n_j}$ and the sequence $s_k \to 0$, we can find $j$ so that $s_{n_j} < \epsilon$ is omitted. Now, take $i$ so that $s_{k_i}$ is the latest term in the subsequence prior to $s_{n_j}$. Then, as $s_{n_j}$ is omitted, $s_{k_1}+\dots+s_{k_i} + s_{n_j} > x$. It follows that $s_{k_1}+\dots+s_{k_i} > x-\epsilon$. As we have already shown that partial sums of $\sum_i s_{k_i}$ are all bounded above by $x$, this completes the proof that $x=\sum_{i=1}^\infty s_{k_i}$. $\blacksquare$
Now, as $\frac{1}{n^2}$ is a decreasing sequence, we can use the integral test to bound the tails of the sum $\sum \frac{1}{n^2}$: $$ \sum_n^\infty \frac{1}{k^2} \geq \int_n^\infty \frac{1}{x^2} \, dx=\frac{1}{n} \, . $$ If $n \geq 2$, $\frac{1}{n+1} > \frac{1}{n^2}$, and so the theorem holds — for the series $\frac{1}{2^2}+\frac{1}{3^2}+\dots$. But $\frac{\pi^2}{9}>1$, so that's good enough; we just apply the theorem to $x=\frac{\pi^2}{9} - 1$.
In fact, if we now allow ourselves to know that $\sum\frac{1}{n^2}=\frac{\pi^2}{6}$ again, this method implies that $x$ can be written as an infinite sum of the reciprocals of distinct squares whenever $x \in \left(0,\frac{\pi^2}{6}-1\right] \cup \left(1, \frac{\pi^2}{6}\right]$.
- 38,733
-
The greedy strategy produces an interesting but unfamiliar sequence of denominators: $\frac{\pi^2}{9} = 1 + \frac{1}{4^2} + \frac{1}{6^2}+\frac{1}{13^2}+\frac{1}{49^2}+\frac{1}{298^2}+\ldots$. My spreadsheet produced more terms, but I begin to doubt their accuracy. – hardmath Nov 05 '14 at 21:23
-
I get the same sequence in Sage, and the next term seems to be 7106. After that I start running into overflow problems. (OEIS doesn't seem to know anything about this sequence, but I guess that's probably to be expected...) – Micah Nov 05 '14 at 22:24
-
-
The most theoretically sound of the three algorithms I tried produced 433244 and then crashed, so that's probably right... – Micah Nov 05 '14 at 22:40
-
1Mathematica is pretty good at this sort of thing. It gives the sequence
$${1, 4, 6, 13, 49, 298, 7106, 433244, 202600705, 2524354628746}$$
– Alexander Vlasev Nov 13 '14 at 03:54 -
2The bound for the tail of $\sum\frac1{n^2}$ can of course also be obtained by comparison with the telescope $\sum(\frac1n-\frac1{n+1})=\sum\frac1{n^2+n}$ – Hagen von Eitzen Nov 17 '14 at 21:04
$$\frac{\pi ^2}{9}=\frac{\pi ^2}{6}\times 6 \times \frac{1}{9}=(\frac{1}{3^2}+\frac{1}{6^2}+\frac{1}{9^2}+ \cdots)\times 6$$
- 16,048
-
1That doesn't work, though—he wants the fractions to be of the form $\frac1{a^2}$. – Akiva Weinberger Nov 05 '14 at 15:34
-
4
-
Ah. I think I was under the impression that they had to be distinct. – Akiva Weinberger Nov 05 '14 at 15:38
-
@columbus8myhw So what about expanding Next's answer to $\frac1{3^2}+\frac1{3^2}+\frac1{3^2}+\frac1{3^2}+\frac1{3^2}+\frac1{3^2}+\frac1{6^2}+\frac1{6^2}+\frac1{3^2}+\frac1{9^2}+\frac1{9^2}+\frac1{9^2}+\frac 1{9^2}+\frac 1{9^2}+\frac1{9^2}+\ldots $ – Hagen von Eitzen Nov 05 '14 at 15:38
-
This is equal to
$$\frac1{72} \sum_{n=-\infty}^{\infty} \left [\frac1{(n+1/6)^2} + \frac1{(n+5/6)^2}\right ] $$
This may be treated using the following formula from residue theory:
$$\sum_{n=-\infty}^{\infty} f(n) = -\pi \sum_k \operatorname*{Res}_{z=z_k} f(z) \cot{\pi z} $$
where $z_k$ are the poles of $f$. In this case
$$f(z) = \frac1{(z+1/6)^2} + \frac1{(z+5/6)^2} $$
which has double poles at $z=-1/6$ and $z=-5/6$. Then the sum is equal to
$$\pi \left [-\frac{\cot{(\pi/6)}}{(4/6)^2} + \pi \csc^2{(\pi/6)} -\frac{\cot{(5\pi/6)}}{(-4/6)^2} + \pi \csc^2{(5\pi/6)}\right ] = 8 \pi^2$$
The result follows.
- 141,538
Hint: $\frac 8 9 = \frac 1 2 + \frac 1 3 + \frac 1 {18}$ Now try using your second formula.
- 16,931
$$\begin{align}\frac{\pi^2}{9}&=1+\frac{1}{3}\sum_{k=0}^\infty\frac{1}{(k+1)^2(k+2)^2}\\&=1+\frac{1}{3(1·2)^2}+\frac{1}{3(2·3)^2}+\frac{1}{3(3·4)^2}+\frac{1}{3(4·5)^2}+\ldots\\&=1+\frac{1}{12}+\frac{1}{108}+\frac{1}{432}+\frac{1}{1200}\ldots...\end{align}$$
- 5,485