8
$\begingroup$

This extends this post.

I. For $\pi^3$:

$$\pi^6-31^2 =\sum_{k=0}^\infty\left(-\frac{63}{(2k+2)^6}+\frac{31^2}{(2k+3)^6}\right) =\sum_{k=0}^\infty P_1(k)\tag1$$

As pointed out by J. Lafont, when $P_1(k)$ is expanded out, its coefficients are all positive. Thus so is the $\text{LHS}$, implying $\pi^3>31$.

II. For $\pi^4$:

The convergents of $\pi^4$ are,

$$97,\, \frac{195}{2},\, \frac{487}{5},\, \frac{1656}{17},\, \frac{2143}{22},\dots$$

The last one, being the particularly close approximation $22\pi^4 \approx 2143.0000027$, was mentioned by Ramanujan. (See also this post.) Using,

$$\frac{\pi^8}{9450}=\sum_{k=0}^\infty \frac{1}{(k+1)^8}$$

$$\frac{17\pi^8}{161280}=\sum_{k=0}^\infty \frac{1}{(2k+1)^8}$$

and the same method to find $(1)$, we get,

$$\pi^8-\Big(\frac{487}{5}\Big)^2 =\sum_{k=0}^\infty\left(\frac{381}{5(2k+2)^8}+\frac{r_1^2}{(2k+3)^8}\right)=\sum_{k=0}^\infty P_2(k)\tag2$$

$$\pi^8-\Big(\frac{2143}{22}\Big)^2 =\sum_{k=0}^\infty\left(-\frac{181695}{11^2(2k+2)^8}+\frac{r_2^2}{(2k+3)^8}\right)=\sum_{k=0}^\infty Q_1(k)\tag3$$

$$\pi^8-\Big(\frac{2143}{22}\Big)^2 =\sum_{k=0}^\infty\left(\frac{r_2^2}{(k+2)^8}-\frac{70208}{1815(2k+1)^8} \right)=\sum_{k=0}^\infty Q_2(k)\tag4$$

where $r_1 =\frac{487}{5},\,$ $r_2 =\frac{2143}{22}$. The coefficients of $P_2(k)$ are all positive, so $5\pi^4>487$.

However, when the $Q_i(k)$ are expanded out, the constant term for both is negative, so we cannot make an analogous conclusion. (In fact, it takes several terms before the sum turns positive.)

Q: Can one find a similar series for $\pi^8-\Big(\frac{2143}{22}\Big)^2 = \sum_{k=0}^\infty R(k)$ such that all coefficients are positive and immediately implying $22\pi^4>2143$?

$\endgroup$
  • $\begingroup$ Like the phrase "extends this post" first time seen this, nice $\endgroup$ – Arjang Feb 11 '16 at 0:41
  • $\begingroup$ @AkivaWeinberger:Oops, thanks. $\endgroup$ – Tito Piezas III Feb 11 '16 at 1:53
  • 1
    $\begingroup$ If nothing comes out for $\pi^4-\frac{2143}{22}$ from$\pi^8$, how about $\pi^{12}$? $\endgroup$ – Jaume Oliver Lafont Feb 11 '16 at 17:04
  • 1
    $\begingroup$ Another solution might come from adding a suitable zero relation, similarly to changing Gregory-Leibniz series for $\frac{\pi}{4}$ into Lehmer's for $\frac{\pi}{3}$, which proves $\pi>3$ with one term. math.stackexchange.com/questions/14113/… $\endgroup$ – Jaume Oliver Lafont Feb 11 '16 at 17:13
  • 1
    $\begingroup$ @JaumeOliverLafont: You're right, $\pi^{12}$ can be used to prove $\pi^4 >\frac{2143}{22}$. Thanks. :) $\endgroup$ – Tito Piezas III Feb 11 '16 at 23:32
4
$\begingroup$

Thanks to a clever suggestion by J. Lafont, there is a series that can prove $\displaystyle\pi^4>\frac{2143}{22}$. However, it does not use $\pi^8$ but $\pi^{12}$. We start with,

$$\frac{691\pi^{12}}{638512875} = \sum_{k=0}^\infty \frac{1}{(k+1)^{12}}$$

$$\frac{691\pi^{12}}{638668800}-1 = \sum_{k=0}^\infty \frac{1}{(2k+3)^{12}}$$

Multiply them with unknowns $a,b,$ then add the two,

$$\frac{691}{420}\frac{(4096a+4095b)\pi^{12}}{13!}-b = \sum_{k=0}^\infty \left(\frac{a}{(k+1)^{12}}+\frac{b}{(2k+3)^{12}}\right)$$

Let $b=\big(\frac{2143}{22}\big)^3$, and choose $a$ such that $\pi^{12}$ has a unit coefficient. We then get,

$$\begin{aligned}\pi^{12}-\Big(\frac{2143}{22}\Big)^3 &=\sum_{k=0}^\infty\left( -\frac{52410418515}{691\cdot10648\cdot(2k+2)^{12}} +\Big(\frac{2143}{22}\Big)^3\frac{1}{(2k+3)^{12}}\right)\\ &=\sum_{k=0}^\infty R(k) \end{aligned}$$

When expanded out, the coefficients of $R(k)$ are all positive. Thus, the $\text{LHS}$ must be positive. Since it is a difference of two cubes $p^3-q^3 = (p-q)(p^2+pq+q^2)$, then that implies $\displaystyle\pi^4>\frac{2143}{22}.$

$\endgroup$
  • $\begingroup$ I revised the series for $\pi^6-961$, so now there is a version with positive coefficients that does not need any further expansion. Would you like to try that improved method on $\pi^{12}-\left(\frac{2143}{22}\right)^3$? Here it is: math.stackexchange.com/a/1651175/134791 $\endgroup$ – Jaume Oliver Lafont Feb 15 '16 at 18:14
  • $\begingroup$ @JaumeOliverLafont: I saw the edited version. Ah, quite nice. $\endgroup$ – Tito Piezas III Feb 15 '16 at 18:28
2
$\begingroup$

From the accelerated series $$\zeta(4)=\frac{\pi^4}{90}=\frac{36}{17}\sum_{n=1}^{\infty } \frac{1}{n^{4}\dbinom{2n}{n}}$$ (Convergence acceleration technique for $\zeta(4)$ (or for $\eta(4)$) via creative telescoping?)

we have the direct sum

$$\pi^4 -\frac{2143}{22}= \frac{5}{52898832} \sum_{n=10}^\infty \left( \frac{1998926767}{n^4\dbinom{2n}{n}}+\frac{17452241}{(n+1)^4\dbinom{2(n+1)}{n+1}}\right)$$

The denominator in the coefficient fraction factors into small primes: $$52898832=(2·3·7)^4·17$$

This series may be related to these.

$\endgroup$

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.