I have crafted an infinite product that converges to Zeta[2] very slowly!

  Product[(1296 n^4 (1 + (1 + n)^3)) / 
    ((-1 + 36 n^2)^2 (-1 + (1 +  n)^3)),{n, 1, 50000}]

I am without a computer for 2 more days and have been using Wolfram|Alpha, but it doesn't want to go much beyond the above limit. I would like someone to verify the function to make sure I haven't missed anything.

  • 2
    $\begingroup$ @ Fred Kline: how did you derive your interesting formula essentially different from the Euler product? $\endgroup$ – Dr. Wolfgang Hintze Sep 7 '15 at 10:58
  • $\begingroup$ @Dr.WolfgangHintze, A few years ago I found a product that produced $\frac{\pi}{3}$ using multiples of $6.$ Yesterday I decided to square it and insert something to multiply by $\frac{3}{2}$. It seems to work. $\endgroup$ – Fred Kline Sep 7 '15 at 12:08
  • $\begingroup$ @Dr.WolfgangHintze, It's the Euler product using multiples of 6. $\endgroup$ – Fred Kline Sep 7 '15 at 12:27
  • $\begingroup$ @ Fred Kline: as to the name "Euler product" please see my answer $\endgroup$ – Dr. Wolfgang Hintze Sep 7 '15 at 14:40
  (1296 n^4 (1 + (1 + n)^3))/((-1 + 36 n^2)^2 (-1 + (1 + n)^3))
  , {n, 1, ∞}
  ] === Zeta[2]
  • 2
    $\begingroup$ Error decreases approximately as $\frac{1}{10n}$. $\endgroup$ – rhermans Sep 7 '15 at 10:10

Amplifying on answer by @rhermans

f[m_] = Product[(1296 n^4 (1 + (1 + n)^3))/((-1 + 
        36 n^2)^2 (-1 + (1 + n)^3)), {n, 1, m}]

(* (Pi^2*Gamma[1 + m]^3*Gamma[3 + m])/
   (6*(3 + 3*m + m^2)*Gamma[5/6 + m]^
        2*Gamma[7/6 + m]^2) *)

This product converges

Limit[f[m + 1]/f[m], m -> Infinity]

(* 1 *)

Limit[f[m], m -> Infinity]

(* Pi^2/6 *)

Product[(1296 n^4 (1 + (1 + n)^3))/((-1 + 36 n^2)^2 (-1 + (1 + n)^3)), {n, 1, 

(* Pi^2/6 *)

% === Zeta[2]

(* True *)

LogLinearPlot[{f[m], Zeta[2]}, {m, 1, 100},
 Epilog -> 
  Inset[LogLinearPlot[{f[m], Zeta[2]}, {m, 75, 100000}], {Log[25], 1.55}]]

enter image description here

  • 1
    $\begingroup$ +1 for nice plot. I've learned something new. $\endgroup$ – Fred Kline Sep 7 '15 at 12:22

I tried to find an even simpler product. Here's my solution:

$$ \zeta(2) =\prod _{n=1}^{\infty } \frac{1}{\left(1-\frac{1}{4 n^2}\right) \left(1-\frac{1}{36 n^2}\right)}$$

In Mathematica

Product[ 1/(1 - 1/(4 n^2)) 1/(1 - 1/(36 n^2)), {n, 1, \[Infinity]}]

(* Out[76]= \[Pi]^2/6 *)

We can derive this from the well-known product formula of the sine

Product[1 - x^2/n^2, {n, 1, \[Infinity]}]

Out[85]= Sin[\[Pi] x]/(\[Pi] x) 

considering that

\[Pi] x/Sin[\[Pi] x] /. x -> 1/2

Out[79]= \[Pi]/2


\[Pi] x/Sin[\[Pi] x] /. x -> 1/6

Out[80]= \[Pi]/3 

The product of these two expressions give [Pi]^2/6. Finally, it is easy to transform the corresponding infinite products into the form provided above.

Remark: Although Euler used the product formula for the sine in his famous proof that the infinite sum of the inverse squares is equal to [Pi]^2/6, normally an Euler product is a product over primes, such as the one defining the zeta function:

$$\zeta(s) =\prod _{n=1}^{\infty } \frac{1}{1-p_n^{-s}}$$


It is not difficult to prove that for any positive integer m we can write $$\zeta (2 m)=\prod _{n=1}^{\infty } a (n)$$

Where the $a(n)$ are rational functions of n.


This is going to be an alternative answer to Dr. Wolfgang Hintze's question. Consider a limit: \begin{equation} g := \prod\limits_{n=1}^\infty \frac{1}{\left(1-\frac{A^2}{n^2}\right)\left(1-\frac{B^2}{n^2}\right)\left(1-\frac{C^2}{n^2}\right)\left(1-\frac{D^2}{n^2}\right)} \end{equation} Taking logs we have: \begin{equation} \log(g) = - \sum\limits_{n=1}^\infty \left[\log(1-\frac{A^2}{n^2}) + \log(1-\frac{B^2}{n^2})+\log(1-\frac{C^2}{n^2})+\log(1-\frac{D^2}{n^2})\right] = \sum\limits_{l=1}^\infty \frac{1}{l} (A^{2 l} + B^{2 l}+C^{2 l}+D^{2 l}) \zeta(2 l) \end{equation} Now from Faulhaber's formula we know values of the zeta function at positive integers (see for example Wikipedia page on zeta function). Therefore we have: \begin{eqnarray} \log(g) =\\ (-1) \left\{ \sum\limits_{l=1}^\infty B_{2 l} (-1)^l \frac{(2\pi A)^{2 l}}{(2 l)!}\frac{1}{2 l} + \sum\limits_{l=1}^\infty B_{2 l} (-1)^l \frac{(2\pi B)^{2 l}}{(2 l)!}\frac{1}{2 l}+ \sum\limits_{l=1}^\infty B_{2 l} (-1)^l \frac{(2\pi C)^{2 l}}{(2 l)!}\frac{1}{2 l}+ \sum\limits_{l=1}^\infty B_{2 l} (-1)^l \frac{(2\pi D)^{2 l}}{(2 l)!}\frac{1}{2 l} \right\} = \\ \sum\limits_{t=2\pi\left\{A,B,C,D\right\}} \log(\Gamma(1+\frac{t}{2 \pi})\Gamma(1-\frac{t}{2 \pi})) \end{eqnarray} Therefore the limit reads: \begin{equation} g = \prod\limits_{t=\left\{A,B,C,D\right\}} \Gamma(1+t) \Gamma(1-t) = \pi^4 \frac{A B C D}{\sin(\pi A)\sin(\pi B)\sin(\pi C)\sin(\pi D)} \end{equation}


This is not an answer. I have additional observations about the answer by Dr. Wolfgang Hintze.
In response to his first comment: From here, when we omit the first two primes from Euler's product, we get a square: $$\frac{\pi^2}{9} =\prod _{n=3}^{\infty } \frac{1}{1-p_n^{-2}},$$ then substituting $6n$ for $p_n,$ we get the square root: $$\frac{\pi}{3} =\prod _{n=1}^{\infty } \frac{1}{1-(6n)^{-2}}.$$

Now about his simplified product: When we omit the left-hand factor of the denominator, we get $\frac{\pi}{3}$ as above.

The left-hand factor of the denominator simplifies my machinations. Very nice.

However, that denominator has another interesting pattern: two odd primes bracket the $4$ as $(3,5)$ and all other odd primes intermittently bracket multiples of $6$ as $(5,7),(11,13),\dots$ So we have a product of all the primes which equals the product of all the points where odd primes may be found:

$$\zeta(2) =\prod _{n=1}^{\infty } \frac{1}{1-p_n^{-2}} =\prod _{n=1}^{\infty } \frac{1}{\left(1-\frac{1}{4 n^2}\right) \left(1-\frac{1}{36 n^2}\right)}$$

Multiples of $n$ where $0\equiv n \mod 3$ cause the left-hand denominator to collide with a previous right-hand denominator Because we omitted the prime $2,$ we will have a missing ratio: $\frac{1}{2}$ or $\frac{2}{1}.$

  • $\begingroup$ @ Fred Kline: Honestly, I don't understand your remark on the "interesting pattern". BTW you should write 2 instead of s in the Euler products. $\endgroup$ – Dr. Wolfgang Hintze Sep 8 '15 at 12:21
  • $\begingroup$ @Dr.WolfgangHintze, Thanks for the heads-up re: $s.$ The interesting pattern is that the prime product uses primes only. We use multiples of $4,6$ which are the mid-points where odd primes can occur. As soon as I get my activation code for Mathematica later today, I plan to look at both functions to see the differences/commonalities. It might not be special? $\endgroup$ – Fred Kline Sep 8 '15 at 13:02
  • 1
    $\begingroup$ @ Fred Kline: you might wish to give the proof of the theorem I presented in my EDIT #1. Try to apply it to zeta(4), for instance, before generalizing too much from one case. $\endgroup$ – Dr. Wolfgang Hintze Sep 8 '15 at 13:14

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.