recurrence relation for $\zeta(2n)$

183 Views Asked by At

I found this formula. Is it correct?

For $n\in\Bbb N,\ n\geq2$, $$\zeta(2n)=\frac{2n\pi^{2n}}{\Gamma(2n+2)}+\sum_{k=0}^{n-2}(-1)^{k-n}\frac{\pi^{2n-2k-2}}{\Gamma(2n-2k)}\zeta(2k+2)$$

Here's my proof.

Assume $m\geq2$ is an even number. We start by finding the Fourier series expansion for $x^m$. We define $$a_k(m)=\frac1\pi\int_{-\pi}^{\pi}x^m\cos(kx)\mathrm dx$$ $$b_k(m)=\frac1\pi\int_{-\pi}^{\pi}x^m\sin(kx)\mathrm dx$$ Hence we know that $$x^m=\frac{a_0(m)}2+\sum_{k\geq1}a_k(m)\cos(kx)+b_k(m)\sin(kx)$$ And because $m$ is even, $$x^m=\frac{\pi^m}{m+1}+\sum_{k\geq1}a_k(m)\cos(kx)+b_k(m)\sin(kx)$$ Next, we start on $b_k(m)$. We make the substitution $x=-t$, and after a little algebra, $$b_k(m)=0$$ This is because $m$ is even, and thus $(-x)^m=x^m$. Hence we update our series $$x^m=\frac{\pi^m}{m+1}+\sum_{k\geq1}a_k(m)\cos(kx)$$ For $a_k(m)$, I integrated by parts twice to find $$a_k(m)=(-1)^{k}\frac{2m\pi^{m-2}}{k^2}-\frac{m(m-1)}{k^2}a_k(m-2)$$ Which brought me to the formula $$a_k(m)=\sum_{v=0}^{\frac{m}2-1}(-1)^{k+v}\frac{2\pi^{m-2-2v}}{k^{2+2v}}\prod_{i=1}^{2v+1}(m+1-i)$$ Then, plugging $x=\pi$ into the Fourier series, $$\pi^m=\frac{\pi^m}{m+1}+\sum_{k\geq1}(-1)^k\sum_{v=0}^{\frac{m}2-1}(-1)^{k+v}\frac{2\pi^{m-2-2v}}{k^{2+2v}}\prod_{i=1}^{2v+1}(m+1-i)$$ $$\frac{m\pi^m}{m+1}=\sum_{k\geq1}\sum_{v=0}^{\frac{m}2-1}(-1)^{v}\frac{2\pi^{m-2-2v}}{k^{2+2v}}\prod_{i=1}^{2v+1}(m+1-i)$$ $$\frac{m\pi^m}{m+1}=\sum_{v=0}^{\frac{m}2-1}(-1)^{v}2\pi^{m-2-2v}\prod_{i=1}^{2v+1}(m+1-i)\sum_{k\geq1}\frac1{k^{2+2v}}$$ $$\frac{m\pi^m}{m+1}=\sum_{v=0}^{\frac{m}2-1}(-1)^{v}2\pi^{m-2-2v}\zeta(2v+2)\prod_{i=1}^{2v+1}(m+1-i)$$ Defining $$c_v(m)=(-1)^v2\prod_{i=1}^{2v+1}(m+1-i)$$ We have $$\frac{m\pi^m}{m+1}=\sum_{v=0}^{\frac{m}2-1}\pi^{m-2-2v}c_v(m)\zeta(2v+2)$$ $$\frac{m\pi^m}{m+1}=c_{m/2-1}(m)\zeta(m)+\sum_{v=0}^{\frac{m}2-2}\pi^{m-2-2v}c_v(m)\zeta(2v+2)$$ $$c_{m/2-1}(m)\zeta(m)=\frac{m\pi^{m}}{m+1}-\sum_{v=0}^{\frac{m}2-2}\pi^{m-2-2v}c_v(m)\zeta(2v+2)$$ I then replace $m$ with $2n$ and $v$ with $k$ (because I like $k$ better): $$c_{n-1}(2n)\zeta(2n)=\frac{2n\pi^{2n}}{2n+1}-\sum_{k=0}^{n-2}\pi^{2n-2-2k}c_k(2n)\zeta(2k+2)$$ $$\zeta(2n)=\frac{2n\pi^{2n}}{(2n+1)c_{n-1}(2n)}-\sum_{k=0}^{n-2}\pi^{2n-2-2k}\frac{c_k(2n)}{c_{n-1}(2n)}\zeta(2k+2)$$ At this point I plugged some things into Wolfram Alpha and got $$\zeta(2n)=\frac{2n\pi^{2n}}{\Gamma(2n+2)}+\sum_{k=0}^{n-2}(-1)^{k-n}\frac{\pi^{2n-2k-2}}{\Gamma(2n-2k)}\zeta(2k+2)$$

I have not seen this formula anywhere, and I don't have Mathematica or Wolfram Pro, so the only way for me to feel confident in this formula is by having you look at it. I don't think I broke any rules, so is this formula valid? Are there any alternate forms for it? Thanks.

1

There are 1 best solutions below

2
On BEST ANSWER

It is correct, but it can be proved with considerably fewer efforts. Since Euler's work it is well-known that $$ \sum_{n\geq 1} \zeta(2n) z^{2n} = \frac{1-\pi z \cot(\pi z)}{2} $$ hence the values of the $\zeta$ function at the positive even integers are related to the coefficients of the Maclaurin series of $\pi z \cot(\pi z)$, or $z\coth z$, or $\frac{z}{e^z-1}$. So we have an explicit relation between $\zeta(2n)$ and the Bernoulli number $B_{2n}$. Since the reciprocal of the (exponential) generating function of Bernoulli numbers is $$ \frac{e^z-1}{z} = \sum_{n\geq 0}\frac{z^{n}}{(n+1)!} $$ by considering the Cauchy product between $\frac{e^z-1}{z}$ and $\frac{z}{e^z-1}$ we may easily derive the recurrence relation for Bernoulli numbers (with even index). This recurrence, translated in terms of $\zeta(2n)$, is exactly your identity.