23

By simple manipulating with some series, I have found the following formula for $\zeta(3)$: $$\zeta(3)=\frac27\sum_{k=0}^{\infty}(-1)^kB_{2k}\frac{\pi^{2k+2}}{(2k+2)!},$$

where $b_k$ are Bernoulli numbers, defined from the equations: $$ B_0=1,\quad B_k=-\frac{1}{k+1}\sum_{i=0}^{k-1}\binom{k+1}{i} B_i,\quad k=1,2,3,\dots $$

Questions: Is this formula for $\zeta(3)$ correct? Can somebody check it numerically? If it's correct, is this a well-known identity or not? Thanks for your help.

Added. By using my identities another interesting formula follows: $$\ln 2=\sum_{k=0}^{\infty}(-1)^kB_{2k}\frac{\pi^{2k}}{(2k+1)!}$$ $$\zeta(3)=\frac45\sum_{k=0}^{\infty}(-1)^kB_{2k}\frac{\pi^{2k+2}}{(2k+3)!}$$

Bio
  • 835
  • 6
  • 13
Marek
  • 695
  • 3
  • 14
  • 4
    It's almost impossible to verify this unless you provide your derivation. – Potato Dec 30 '13 at 23:55
  • 1
    Yes, it's correct. – Lucian Dec 31 '13 at 00:50
  • @Lucian Thank you very much. Did you prove that formula? Or it's numericaly checked? – Marek Dec 31 '13 at 01:04
  • 1
    Numerically. But I've seen enough of them to know when something's plausible or legit. – Lucian Dec 31 '13 at 01:14
  • @Lucian Nice job. Thank you very much, again. – Marek Dec 31 '13 at 01:28
  • 11
    This asks whether his formula is correct without showing any calculation, that says it is fine without any proof, this asks whether that proved it, that says he didn't but he "checked it numerically"(??)...has the site stopped being about mathematics ? – DonAntonio Dec 31 '13 at 04:45
  • 2
    @Marek Please do show your manipulation a of series... – zerosofthezeta Dec 31 '13 at 09:48
  • 8
    @DonAntonio: There are many approaches to mathematics. Yes we must be wary of non-rigorous demonstrations because mistakes may slip by unexposed, and it is fine to be sceptical, but that does not mean that the axiomatic presentation and rigorous proofs are the only worthwhile paths in the human activity we call mathematics. The results of Euler or Ramanujan (or Apery) are not worthless just because they were not always presented with watertight proof. I'm not saying that we should believe when someone says "I've checked this"; just reacting to your definition of what "about mathematics" is. – ShreevatsaR Jan 01 '14 at 18:09
  • @zerosofthezeta Ok, I'll put my derivation here, soon. It's very simple. Just one substitution to one formula and some adjustments. First I could not believe how simple, when I've invented it. – Marek Jan 01 '14 at 19:33

2 Answers2

37

Your formula is a variant of a formula from Euler ($1772$) (exposed in Ayoub's paper Euler and the Zeta Function (1974) p. $1085$) rediscovered by Ramaswami ($1934$) and more recently by Ewell ($1990$).
See too Adamchik's paper and Boros and Moll's nice book 'Irresisitible Integrals' $(11.4.6)$ :

\begin{align} \zeta(3)&=\frac{\pi^2}7\left(1-4\sum_{n=1}^\infty \frac{\zeta(2\,n)}{(2n+2)(2n+1)\,2^{2n}}\right)\qquad\text{or since}\;\zeta(0)=-\frac12\\ \tag{1}\zeta(3)&=-\frac{4\pi^2}7\sum_{n=0}^\infty \frac{\zeta(2\,n)}{(2n+2)(2n+1)\,2^{2n}}\\ \end{align} This becomes indeed, after substitution of $\,\displaystyle \frac{\zeta(2n)}{2^{2n}}=(-1)^{n+1}\frac{B_{2n}\,\pi^{2n}}{2(2n)!}$ : \begin{align} \zeta(3)&=\frac{2\,\pi^2}7\sum_{n=1}^\infty (-1)^n\frac{B_{2n}\,\pi^{2n}}{(2n+2)(2n+1)(2n)!}\\ \tag{2}\zeta(3)&=\frac27\sum_{n=0}^{\infty}(-1)^nB_{2n}\frac{\pi^{2n+2}}{(2n+2)!}\\ \end{align}

Let's reproduce here Boros and Moll's proof of $(1)$ and start with the generating function for central binomial coefficients : $$\tag{3}\frac 1{\sqrt{1-4t}}=\sum_{n=0}^\infty \binom{2n}{n}t^n$$ Setting $\,t:=\left(\dfrac x2\right)^2$ and integrating we get : $$\tag{4}\arcsin(x)=\sum_{n=0}^\infty \frac{\binom{2n}{n}}{2^{2n}}\frac{x^{2n+1}}{2n+1}$$ (btw Euler found too this neat expression for the square : $\displaystyle \arcsin(x)^2=\sum_{n=1}^\infty \frac{2^{2n}}{\binom{2n}{n}}\frac{x^{2n}}{2\,n^2}\;$)

Dividing $(4)$ by $x$ and integrating we get : $$\tag{5}\int\frac{\arcsin(x)}xdx=\sum_{n=0}^\infty \frac{\binom{2n}{n}}{2^{2n}}\frac{x^{2n+1}}{(2n+1)^2}$$ For $x:=\sin(t)$ this becomes : $$\tag{6}\int\frac{t\cos(t)}{\sin(t)}dt=\sum_{n=0}^\infty \frac{\binom{2n}{n}}{2^{2n}}\frac{(\sin(t))^{2n+1}}{(2n+1)^2}$$

At this point we may use the well known generating function for $\zeta(2n)$ : $$\pi\;x\;\cot(\pi\;x)=1-2\sum_{n=1}^\infty \zeta(2n)\;x^{2n}$$ So that setting $t:=\pi x\,$ and integrating we get using $(6)$ : $$\tag{7}\int t\,\cot(t)\,dt=t-2\sum_{n=1}^\infty \frac{\zeta(2n)\;t^{2n+1}}{\pi^{2n}(2n+1)}=\sum_{n=0}^\infty \frac{\binom{2n}{n}}{2^{2n}}\frac{(\sin(t))^{2n+1}}{(2n+1)^2}$$ Integrating this again from $0$ to $\frac {\pi}2$ returns : $$\frac {(\pi/2)^2}2-2\sum_{n=1}^\infty \frac{\zeta(2n)\;(\pi/2)^{2n+2}}{\pi^{2n}(2n+2)(2n+1)}=\sum_{n=0}^\infty \frac{\binom{2n}{n}}{2^{2n}}\frac{1}{(2n+1)^2}\int_0^{\pi/2}\sin^{2n+1}(t)\,dt$$ The integral at the right may be evaluated with the Wallis formula : $$\frac{\binom{2n}{n}}{2^{2n}}\int_0^{\pi/2}\sin^{2n+1}(t)\,dt=\frac1{2n+1}$$ and we get $$\tag{8}\frac {\pi^2}8-\frac{\pi^2}2\sum_{n=1}^\infty \frac{\zeta(2n)}{2^{2n}(2n+2)(2n+1)}=\sum_{n=0}^\infty\frac{1}{(2n+1)^3}=\frac 78\zeta(3)$$ (separating the even and odd terms of zeta it is easy to prove the more general identity $\;\displaystyle\sum_{n=0}^\infty\frac{1}{(2n+1)^m}=\left(1-\frac 1{2^m}\right)\zeta(m)\;$)

From $(8)$ we deduce $(1)$ and thus your formula $(2)$.


GENERALIZATION

Let's define $$f(m):=\sum_{n=0}^{\infty}(-1)^nB_{2n}\frac{\pi^{2n+m-1}}{(2n+m-1)!}=-\pi^{m-1}\,\sum_{n=0}^{\infty}\frac{(2n)!\;\zeta(2n)}{(2n+m-1)!\,2^{2n-1}}$$

then using the method nicely exposed in your answer (computing further integrals of $\dfrac{x}{e^x-1}$ that is evaluating $\;\displaystyle \int \cdots \int \frac{x}{e^x-1} dx\cdots\,dx=\int_0^x \frac{t}{e^t-1}\frac{(x-t)^{n-1}}{(n-1)!} dt\,$ for $x=-\pi\,i$)

we may obtain : \begin{align} f(1)&=0\\ f(2)&=\pi\,\log(2)\\ f(3)&=\frac 72\zeta(3)\\ f(4)&=\frac 54\pi\zeta(3)\\ \tag{9}f(5)&=-\frac {31}4\zeta(5)+\pi^2\zeta(3)\\ f(6)&= \frac{\pi^3}3\zeta(3) -\frac{49}{16}\pi\zeta(5)\\ f(7)&=\frac{381}{32}\zeta(7)-2\pi^2\zeta(5)+\frac{\pi^4}{12}\zeta(3)\\ f(8)&= \frac{\pi^5}{60}\zeta(3)-\frac 23\pi^3\zeta(5)+\frac{321}{64}\pi\zeta(7)\\ f(9)&=-\frac{511}{32}\zeta(9)+3\pi^2\zeta(7)-\frac{\pi^4}6\zeta(5)+\frac{\pi^6}{‌​360}\zeta(3)\\ \end{align} and so on and thus conjecture that for $\;n>1$ we have : $\qquad\qquad\qquad\qquad\qquad\qquad\qquad(10)$ \begin{align} f(2n+1)&=(-1)^{n+1}\left[n\left(4-2^{1-2n}\right)\zeta(2n+1)+2\sum_{k=1}^{n-1} \frac{(-1)^k(n-k)}{(2k)!}\pi^{2k}\ \zeta(2(n-k)+1)\right]\\ f(2n+2)&=(-1)^{n+1}\pi\left[\left(2n-1+2^{-2n}\right)\zeta(2n+1)+2\sum_{k=1}^{n‌​-1} \frac{(-1)^k(n-k)}{(2k+1)!}\pi^{2k}\,\zeta(2(n-k)+1)\right]\\ \end{align}

For a proof see the Theorem A of Cvijovic and Klinowski ($1997$) "New rapidly convergent series representations for $\zeta(2n+1)$" with the result (similar to the first equation $(10)$) : $$\tag{11}\zeta(2n+1)=\frac{(-1)^n\,(2\pi)^{2n}}{n(2^{2n+1}-1)}\left[\sum_{k=1}^{n-1}\frac{(-1)^{k-1}\,k\,\zeta(2k+1)}{(2n-2k)!\,\pi^{2k}}+\sum_{k=0}^\infty\frac{(2k)!\,\zeta(2k)}{(2n+2k)!\,2^{2k}}\right]$$

This allows to find whole families of formulae for $\zeta(2n+1)$ with some neat instances like these :
(cf the comments for $\zeta(5)$) \begin{align} \zeta(5)&=\frac{16}{23}\sum_{n=0}^{\infty}(-1)^nB_{2n}\frac{(2n+2)\ \pi^{2n+4}}{(2n+5)!}\\ \zeta(5)&=-\frac{2^5\,\pi^4}{23}\sum_{n=0}^\infty \frac{\zeta(2\,n)}{(2n+5)(2n+4)(2n+3)(2n+1)\,2^{2n}}\\ \zeta(7)&=-\frac{2^7\,\pi^6}{4719}\sum_{n=0}^\infty \frac{(128\,n+211)\quad\zeta(2\,n)}{(2n+7)(2n+6)(2n+5)(2n+4)(2n+3)(2n+1)\,2^{2n}}\\ \zeta(9)&=-\frac{2^{10}\,\pi^8}{77259}\sum_{n=0}^\infty \frac{(449\,n+780)\quad\quad\zeta(2\,n)}{(2n+9)(2n+8)(2n+7)(2n+6)(2n+5)(2n+3)(2n+1)\,2^{2n}}\\ \zeta(11)&=-\frac{2^{12}\,\pi^{10}}{\small{395681475}}\sum_{n=0}^\infty\frac{(2497024\,n^2+10676923\,n+11093808)\quad\zeta(2n)}{(2n+11)(2n+10)\cdots(2n+6)(2n+5)(2n+3)(2n+1)\,2^{2n}}\\ \zeta(13)&=-\frac{2^{13}\,\pi^{12}}{\small{75159854595}}\sum_{n=0}^\infty\frac{\small{(619308920\,n^2+2659184244\,n+2771839831)\quad}\zeta(2n)}{\small{(2n+13)(2n+12)\cdots(2n+8)(2n+7)(2n+5)(2n+3)(2n+1)\,2^{2n}}}\\ \zeta(15)&=-\frac{2^{15}\,\pi^{14}}{\small{1697182926260535}}\sum_{n=0}^\infty\frac{\small{(21253850808320\,n^3+165886464354888\,n^2+415352534250460\,n+332739769444737)}\zeta(2n)}{\small{(2n+15)(2n+14)\cdots(2n+8)(2n+7)(2n+5)(2n+3)(2n+1)\,2^{2n}}}\\ \end{align} The coefficients for these formulae may possibly be obtained using recurrences (by rewriting $(11)$ or using alternative relations) but I obtained them numerically with high precision (i.e. these identities are only conjectured).

We may compare them with the Euler formula $(1)$ and its variants : \begin{align} \zeta(3)&=-\frac{4\pi^2}7\sum_{n=0}^\infty \frac{\zeta(2\,n)}{(2n+2)(2n+1)\,2^{2n}}\\ \zeta(3)&=-\frac{8\pi^2}5\sum_{n=0}^\infty \frac{\zeta(2\,n)}{(2n+3)(2n+2)(2n+1)\,2^{2n}}\\ \zeta(3)&=\frac{2\pi^2}7\left(\log(2)+\sum_{n=0}^\infty \frac{\zeta(2\,n)}{(n+1)\,2^{2n}}\right)\\ \end{align} These two last formulae were obtained by Chen and Srivastava in "Some Families of Series Representations for the Riemann $\zeta(3)$" and reproduced in Srivastava and Choi's book "Zeta and q-Zeta Functions and Associated Series and Integrals" (see p.$405$ "$4.2$ Rapidly Convergent Series for $\zeta(2n + 1)$") that contains many other series for $\zeta$ and general formulae like this one : \begin{align} \zeta(2n+1)&=\frac{(-1)^{n-1}(2\pi)^{2n}}{2^{2n+1}-1}\left[\frac{H_{2n}-\log(\pi)}{(2n)!}+\sum_{k=1}^{n-1}\frac{(-1)^k\,\zeta(2k+1)}{(2n-2k)!\,\pi^{2k}}+2\sum_{k=1}^\infty\frac{(2k-1)!\,\zeta(2k)}{(2n+2k)!\,2^{2k}}\right]\\ \end{align} A review is proposed in Katsurada's paper "Rapidly convergent series representations for $\zeta(2n+1)$ and their $\chi$-analogue".


OTHER GENERALIZATIONS

The OP asked what would happen with the upper bound of the integral $x=-\pi\,i\,$ replaced by $\,x=-\frac{\pi}2i\,$ so let's investigate this by defining $$g(m):=\sum_{n=0}^{\infty}(-1)^nB_{2n}\frac{\left(\frac{\pi}2\right)^{2n+m-1}}{(2n+m-1)!}=-\left(\frac{\pi}2\right)^{m-1}\,\sum_{n=0}^{\infty}\frac{2\,(2n)!\;\zeta(2n)}{(2n+m-1)!\;2^{4n}}$$

then evaluating $\;\displaystyle \int \cdots \int \frac{x}{e^x-1} dx\cdots\,dx=\int_0^x \frac{t}{e^t-1}\frac{(x-t)^{n-1}}{(n-1)!} dt\,$ for $x=-\frac{\pi}2\,i\;$ returns if $\;\displaystyle \beta(s):=\sum_{n=0}^\infty \frac{(-1)^n}{(2n+1)^s}$ is the Dirichlet beta function : \begin{align} g(1)&=\frac{\pi}4\\ g(2)&=\frac{\pi}4\log(2)+\beta(2)\\ g(3)&=\frac{35}{16}\zeta(3) -\frac 12\pi\,\beta(2)\\ \tag{12}g(4)&=\frac{61}{64}\pi\,\zeta(3)-3\,\beta(4)\\ g(5)&=\frac 14\pi^2\,\zeta(3)-\frac{527}{128}\zeta(5)+\frac 12\pi\,\beta(4)\\ g(6)&=\frac 1{24}\pi^3\,\zeta(3)-\frac{2033}{1024}\pi\,\zeta(5)+5\,\beta(6)\\ g(7)&=\frac 1{192}\pi^4\,\zeta(3)-\frac 12\pi^2\,\zeta(5)+\frac{24765}{4096}\zeta(7)-\frac 12\pi\,\beta(6)\\ \end{align}

and so on so that we may conjecture that for $\;n>1$ : $\qquad\qquad\qquad\qquad\qquad\qquad\qquad(13)$ \begin{align} g(2n+1)&=(-1)^n\left[\frac {\pi}2\beta(2n)-n\frac{2^{4n+1}+2^{2n}-1}{2^{4n}}\zeta(2n+1)-\\2\sum_{k=1}^{n-1}(-1)^k\frac{n-k}{(2k)!}\left(\frac{\pi}2\right)^{2k}\zeta(2(n-k)+1))\right]\\ g(2n+2)&=(-1)^n\left[(2n+1)\beta(2n+2)-\left(n-\frac{2^{2n}-1}{2^{4n+2}}\right)\pi\,\zeta(2n+1)-\\2\sum_{k=1}^{n-1}(-1)^k\,\frac{n-k}{(2k+1)!}\left(\frac{\pi}2\right)^{2k+1}\zeta(2(n-k)+1)\right]\\ \end{align}

Returning some expressions of interest for the Catalan constant $G$ and other $\beta(2n)$ constants :

\begin{align} G=\beta(2)&=-\frac{\pi}4\log(2)-\pi\,\sum_{n=0}^{\infty}\frac{\zeta(2n)}{(2n+1)\;2^{4n}}\\ \beta(4)&=-\frac {\pi^3}{840}\left[\frac{61}4\log(2)+\sum_{n=0}^\infty \frac{(244(n+2)(n+1)-9)\,\zeta(2n)}{(2n+3)(2n+2)(2n+1)\,2^{4n}}\right]\\ \beta(6)&=-\frac{\pi^5}{3541440}\left[\frac{50345}8\log(2)+\\\sum_{n=0}^\infty\frac{(402760n^4 + 3020700n^3 + 8300346n^2 + 9777801n + 4077233)\zeta(2n)}{(2n+5)(2n+4)(2n+3)(2n+2)(2n+1)\,2^{4n}}\right]\\ \end{align} as well as fast series for $\zeta(\text{odd})$ : \begin{align} \zeta(3)&=-\frac{2\;\pi^2}{35}\left[\log(2)+4\sum_{n=0}^\infty\frac{(2n+3)\quad\zeta(2n)}{(2n+2)(2n+1)\;2^{4n}}\right]\\ \zeta(5)&=-\frac{2\;\pi^4}{55335}\left[157\,\log(2)+8\sum_{n=0}^\infty\frac{(628\,n^3+3140\,n^2+5111\,n+2581)\;\zeta(2n)}{(2n+4)(2n+3)(2n+2)(2n+1)\;2^{4n}}\right]\\ \end{align}


We could get slower converging results with the upper bound of the integral replaced by $\,x:=-2\pi\,i$. For this let's define $$h(m):=\sum_{n=0}^{\infty}(-1)^nB_{2n}\frac{(2\,\pi)^{2n+m-1}}{(2n+m-1)!}=-(2\,\pi)^{m-1}\,\sum_{n=0}^{\infty}\frac{2\,(2n)!\;\zeta(2n)}{(2n+m-1)!}$$

then evaluating $\;\displaystyle \int \cdots \int \frac{x}{e^x-1} dx\cdots\,dx=\int_0^x \frac{t}{e^t-1}\frac{(x-t)^{n-1}}{(n-1)!} dt\,$ for $x=-2\,\pi\,i$)

we may obtain : \begin{align} h(3)&=0\\ h(4)&=6\,\pi\,\zeta(3)\\ h(5)&=4\,\pi^2\,\zeta(3)\\ \tag{14}h(6)&=\frac 83\pi^3\,\zeta(3)-10\,\pi\,\zeta(5)\\ h(7)&=\frac 43\pi^4\,\zeta(3)-8\,\pi^2\,\zeta(5)\\ h(8)&= \frac 8{15}\pi^5\,\zeta(3)-\frac{16}3\pi^3\,\zeta(5)+14\,\pi\,\zeta(7)\\ h(9)&= \frac 8{45}\pi^6\,\zeta(3)-\frac 83\,\pi^4\,\zeta(5)+12\,\pi^2\,\zeta(7)\\ h(10)&=\frac{16}{315}\pi^7\,\zeta(3)-\frac{16}{15}\pi^5\,\zeta(5)+8\,\pi^3\,\zeta(7)-18\,\pi\,\zeta(9)\\ \end{align} And find further expressions for $\zeta(\text{odd})$ (using these expressions or combining them with the earlier results) but since the subject appears endless I'll stop here.

AKP2002
  • 156
Raymond Manzoni
  • 43,021
  • 5
  • 86
  • 140
  • Wow, nice and interesting techniques! Thank you very much. My derivation is very very simple. But I'm not sure if my steps are fully correct. May be I'll add it here, later. – Marek Jan 01 '14 at 19:10
  • Thanks @Marek. Your derivation would of course be welcome! Cheers, – Raymond Manzoni Jan 01 '14 at 20:39
  • Raymond, does this generalize to $\zeta(n)$ for odd $n>3$? – Tito Piezas III Jan 02 '14 at 01:01
  • @Tito: Using the method exposed by Marek and defining : $$f(m):=\sum_{n=0}^{\infty}(-1)^nB_{2n}\frac{\pi^{2n+m-1}}{(2n+m-1)!}$$ you'll get :

    \begin{align} f(1)&=0\ f(3)&=\frac 72\zeta(3)\ f(5)&=-\frac {31}4\zeta(5)+\pi^2\zeta(3)\ f(7)&=\frac{381}{32}\zeta(7)-2\pi^2\zeta(5)+\frac{\pi^4}{12}\zeta(3)\ f(9)&=-\frac{511}{32}\zeta(9)+3\pi^2\zeta(7)-\frac{\pi^4}6\zeta(5)+\frac{\pi^6}{360}\zeta(3)\ \end{align} (and so on... I didn't search the general formula yet)
    so that conversely we may write $\zeta(2n+1)$ as a sum of $f(m)$ terms with $m\le 2n+1$.

    – Raymond Manzoni Jan 02 '14 at 20:58
  • (the trick is to compute $;\displaystyle \int \int \int \int \frac{x}{e^x-1} dx,dx,dx,dx=\int_0^x \frac{t}{e^t-1}\frac{(x-t)^{4-1}}{(4-1)!} dt,$ instead of $;\displaystyle \int \int \frac{x}{e^x-1} dx,dx,$ and so on with $6$ integrals, $8$ integrals...). – Raymond Manzoni Jan 02 '14 at 21:12
  • I will conjecture the general expression : $$f(2n+1)=(-1)^{n+1}\left[n\left(4-2^{1-2n}\right)\zeta(2n+1)+2\sum_{k=1}^{n-1} \frac{(-1)^k(n-k)}{(2k)!}\pi^{2k}\ \zeta(2(n-k)+1))\right]$$ – Raymond Manzoni Jan 02 '14 at 23:53
  • @Raymond Manzoni, I'm glad, that the method inspires you. Btw, very nice conjecture! – Marek Jan 03 '14 at 14:55
  • @Marek: To be complete I'll add the formula for the even terms (for $n>1$) : $$f(2n+2)=(-1)^{n+1}\left[\left(2n-1+2^{-2n}\right)\pi\zeta(2n+1)+2\sum_{k=1}^{n-1} \frac{(-1)^k(n-k)}{(2k+1)!}\pi^{2k+1}\ \zeta(2n+1-2k)\right]$$ while $\displaystyle f(2)=\log(2)\pi$. From your other thread if seems that you'll become the specialist of the derivatives and integrations of $\dfrac{x}{e^x-1}$. Excellent continuation ! – Raymond Manzoni Jan 03 '14 at 21:52
  • @Raymond Manzoni, very nice formula for even terms! I love this moments, if something new arises. Thank you very much. I'm not so specialist, but math is my passion. – Marek Jan 03 '14 at 22:15
  • Glad you liked them @Marek ! Let's obtain two expressions for $\zeta(5)$ for example. We could use $,\displaystyle \frac{31}4\zeta(5)=\frac27\pi^2 f(3)-f(5),$ (and $\frac {2(2n+4)(2n+3)}7-1=\frac{8n^2+28n+17}7$) to get : $$\zeta(5)=\frac 4{31}\sum_{n=0}^{\infty}(-1)^nB_{2n}\frac{(8n^2+28n+17),\pi^{2n+4}}{7,(2n+4)!}$$ But it may be more interesting to use $f($even$)$ in$,\displaystyle \frac{23}{16}\zeta(5)=f(5)-3\frac{f(6)}{\pi}$ (since $,\displaystyle f(6)= \frac{\pi^3}3\zeta(3) -\frac{49}{16}\pi\zeta(5)$) to get the somewhat nicer : – Raymond Manzoni Jan 04 '14 at 00:38
  • $$\zeta(5)=\frac{16}{23}\sum_{n=0}^{\infty}(-1)^nB_{2n}\frac{(2n+2)\ \pi^{2n+4}}{(2n+5)!}$$ (we could clearly obtain many other expressions for $\zeta(5)$ by composing these two...) – Raymond Manzoni Jan 04 '14 at 00:39
  • Oh no!:) Ican't believe how many formulas can be obtained by using one simple method! The nicer is really beautiful! Excelent job! You should publish it! Thank you very much. My dream is, if somebody will express by simple formula all odd zetas. May be it can be done, but using $\zeta(3)$ as a constant(accepted as new basic constant) Try it. Math need some new formulas. To push knowledge a bit more forward... – Marek Jan 04 '14 at 05:35
  • @Marek: Yes so many jewels are shining in the $\zeta$ mine (as you may see in my updated answer). There could even be an Arkenstone hidden there but... we should really take care to the Dragon's illusions :-). All the best, – Raymond Manzoni Jan 04 '14 at 18:13
  • @Raymond Manzoni: Thank you very much for your updated answer. It's helpful for me. I think, some new formulas must be discovered, if we can derive even nicer resullts for $\zeta(2n+1)$. There are so many years from time of Euler, so many nice results, but all are very complicated. E.g. integrals, that nobody can calculate with simple results, or infinite sums, or infinite products or something similar. Nobody can solve this with simple results. Or nobody can prove, that it has no simple expressions. What a pity. So I appreciate any help to this problem. – Marek Jan 05 '14 at 07:55
  • @Marek (and Tito) : Under my expression for $\zeta(5)$ I added expressions I got for $\zeta(7)$ and $\zeta(9)$. To come back at $\dfrac{x}{e^x-1}$ let's just add that many series involving $\sum_k \dfrac 1{k^n,(e^{2\pi k}\pm 1)}$ were obtained by Ramanujan and others as you may see at MathWorld (starting with $(92)$). Excellent continuation, – Raymond Manzoni Jan 06 '14 at 00:01
  • Many thanks to Andrew of course for the generous Bonus !! – Raymond Manzoni Jan 06 '14 at 22:07
  • @Raymond Manzoni: Thank you very much, again. – Marek Jan 07 '14 at 20:50
  • @Marek: I updated my answer with some new (for me) expressions for $\beta(\text{odd})$ and $\zeta(\text{even})$. Cheers, – Raymond Manzoni Feb 10 '14 at 22:48
  • Can you please name the "Ayoub's paper", the link you gave seems to redirect to this page – AKP2002 Jan 10 '22 at 08:20
  • @AKP2002: the paper was Ayoub (1974) "Euler and the zeta function" available now here. – Raymond Manzoni Jan 10 '22 at 12:08
  • @RaymondManzoni Thanks a lot! – AKP2002 Jan 11 '22 at 09:47
  • 1
    Thanks for the edit @AKP2002 and excellent continuation! – Raymond Manzoni Jan 11 '22 at 12:40
11

So,I put here my derivation, but I'm not sure, if all my steps are fully correct. The key idea is very simple. We just put $x=-i\pi$ in this formula: \begin{equation} 2\zeta(3)+\zeta(2)x+\frac{x^3}{12}+\sum_{k=0}^{\infty}B_{2k}\frac{x^{2k+2}}{(2k+2)!}=2\sum_{k=1}^{\infty}\frac{e^{kx}}{k^3}-x\sum_{k=1}^{\infty}\frac{e^{kx}}{k^2},\quad x\in(-2\pi,0\,\rangle\end{equation} plus some simplifacations. That's all.

(Formula can be easily obtaided from two expansions: $$\frac{x}{e^x-1}=-\frac{x}2+\sum_{k=0}^{\infty}B_{2k}\frac{x^{2k}}{(2k)!},\quad x\in(-2\pi,2\pi),$$ $$\frac{x}{e^x-1}=-x\frac{1}{1-e^x}=-x\sum_{k=0}^{\infty}e^{kx},\quad x\in(-\infty,0),$$ so formal double integration gives: $$\int_0^x\!\!\int_0^x\frac{x}{e^x-1} dx dx=-\frac{x^3}{12}+\sum_{k=0}^{\infty}B_{2k}\frac{x^{2k+2}}{(2k+2)!}$$ $$\int_0^x\!\!\int_0^x\frac{x}{e^x-1} dx dx=2\sum_{k=1}^{\infty}\frac{e^{kx}}{k^3}-x\sum_{k=1}^{\infty}\frac{e^{kx}}{k^2}-\frac{x^{3}}{6}-\zeta(2)x-2\zeta(3)$$ which implies the desired identity.)

Now back to my derivation of formula for $\zeta(3)$. Putting $x=-i\pi$ we have: $$ 2\zeta(3)-i\pi\zeta(2)+\frac{(-i\pi)^3}{12}+\sum_{k=0}^{\infty}B_{2k} \frac{(-i\pi)^{2k+2}}{(2k+2)!}=2\sum_{k=1}^{\infty}\frac{e^{(-i\pi)k}}{k^3}+i\pi\sum_{k=1}^{\infty}\frac{e^{(-i\pi)k}}{k^2} $$ and because $$e^{-i\pi}=\cos(-\pi)+i\sin(-\pi)=-1$$ $$\sum_{k=1}^{\infty}\frac{(-1)^k}{k^2}=-\frac12\zeta(2)$$ $$\sum_{k=1}^{\infty}\frac{(-1)^k}{k^3}=-\frac34\zeta(3),$$ we have step by step: \begin{align} 2\zeta(3)-i\pi\zeta(2)+\frac{i\pi^3}{12}-\sum_{k=0}^{\infty}B_{2k}\frac{(-1)^{k}\pi^{2k+2}}{(2k+2)!}&=2\sum_{k=1}^{\infty}\frac{(-1)^k}{k^3}+i\pi\sum_{k=1}^{\infty}\frac{(-1)^k}{k^2}\\ 2\zeta(3)-i\pi\zeta(2)+\frac{i\pi^3}{12}-\sum_{k=0}^{\infty}B_{2k}\frac{(-1)^{k}\pi^{2k+2}}{(2k+2)!}&=-\frac32\zeta(3)-i\pi\frac12\zeta(2)\\ \frac72\zeta(3)-\frac{i\pi}2\zeta(2)&=\sum_{k=0}^{\infty}B_{2k}\frac{(-1)^{k}\pi^{2k+2}}{(2k+2)!}-\frac{i\pi^3}{12}, \end{align} so we immediately obtain even two formulas: $$\zeta(2)=\frac{\pi^2}6$$ $$\zeta(3)=\frac{2}7\sum_{k=0}^{\infty}(-1)^{k}B_{2k}\frac{\pi^{2k+2}}{(2k+2)!}.$$

Marek
  • 695
  • 3
  • 14
  • +1: Nice result @Marek! Our derivations have common points (noticing that $; \displaystyle\pi;x;\cot(\pi;x)=\pi;x;i\frac {e^{2\pi i x}+1}{e^{2\pi i x}-1}=\pi;x;i+\frac{2\pi ;x;i}{e^{2\pi i x}-1},$)
    but your $,\displaystyle\int\cdot\int\frac{x}{e^x-1} dx\cdots dx,$ integrals are easier to evaluate (with Wolfram/MMa at least) allowing more general results (as you may see in my comment to Tito). Cheers,
    – Raymond Manzoni Jan 02 '14 at 21:04
  • 2
    Thank you very much.I'am glad that the method inspires other people to find interesting results. – Marek Jan 03 '14 at 14:38