# Closed form of $2\sum_{n = 0}^{m - 1}(-1)^n\zeta(4m - 2n + 1)\zeta(2n + 2)$

Is it possible to simplify the below sum? or maybe find a nice closed form (or asymptotics)?

$$2\sum_{n = 0}^{m - 1}(-1)^n\zeta(4m - 2n + 1)\zeta(2n + 2)$$

I am very much aware about even values of Riemann zeta (Euler's formula in terms of Bernoulli numbers) and that would definitely give us a nice closed form for the term $$\zeta(2n + 2)$$, but what about the term $$\zeta(4m - 2n + 1)$$? This seems too complicated to simplify simply because we know very little about the odd values of Riemann zeta. However I'm not familiar with other closed forms for $$\zeta(2n + 1)$$, so I cannot say for sure that the sum can't be simplified more.

Out of curiosity: Can we simplify or find a closed form for the product of even zeta and odd zeta that is $$\zeta(2n + 1)\zeta(2n)$$? Or can we get a nice closed form for $$\zeta(4m + 3)$$?

Thanks.

EDIT $$\textbf{1}$$: @Claude Leibovici, in his answer claims that

If $$m$$ is even and large then

$$2\sum_{n = 0}^{m - 1}(-1)^n\,\zeta(4m - 2n + 1)\,\zeta(2n + 2) \approx \pi \coth (\pi )-2 \tag{1}\label{1}$$

and if $$m$$ is odd then $$2\sum_{n = 0}^{m - 1}(-1)^n\,\zeta(4m - 2n + 1)\,\zeta(2n + 2) \approx \pi \coth (\pi ) \tag{2}\label{2}$$

However, I have no idea on how to prove $$\eqref{1}$$ and $$\eqref{2}$$, so proving $$\eqref{1}$$ and $$\eqref{2}$$ (probably using some elementary tools) is sufficient for answering the question completely. Thanks.

• Where did you encounter this sum? Perhaps some more context would help in finding a solution. Apr 18, 2021 at 6:56
• @ParamanandSingh I encountered it while playing with the product $\zeta(2n)\zeta(2n + 1)$ Apr 18, 2021 at 6:57
• Wow great problem. I cannot wait to see who has the best answer. Apr 26, 2021 at 21:10

As is shown in this answer $$\sum_{k=1}^\infty\zeta(2k)x^{2k}=\frac12(1-\pi x\cot(\pi x))\tag1$$ substituting $$x\mapsto ix$$, we get $$\sum_{k=1}^\infty(-1)^{k-1}\zeta(2k)x^{2k}=\frac12(\pi x\coth(\pi x)-1)\tag2$$ and therefore, $$\sum_{k=1}^\infty(-1)^{k-1}(\zeta(2k)-1)x^{2k}=\frac12(\pi x\coth(\pi x)-1)-\frac1{1+x^2}\tag3$$ Taking the limit of $$(3)$$ as $$x\to1$$, we get $$\sum_{k=1}^\infty(-1)^{k-1}(\zeta(2k)-1)=\frac12(\pi\coth(\pi)-2)\tag4$$ Since $$n\le m-1$$, we have that $$4m-2n+1\ge2m+3$$. This means that $$\zeta(4m-2n+1)=1+O\!\left(4^{-m}\right)$$ \begin{align} &2\sum_{n=0}^{m-1}(-1)^n\color{#C00}{\zeta(4m-2n+1)}\color{#090}{\zeta(2n+2)}\tag{5a}\\ &=2\color{#C00}{\left(1+O\!\left(4^{-m}\right)\right)}\sum_{n=1}^m(-1)^{n-1}(\color{#090}{\zeta(2n)}-1)+2(m\bmod2)\tag{5b}\\ &=2\left(1+O\!\left(4^{-m}\right)\right)\frac12(\pi\coth(\pi)-2)+2(m\bmod2)\tag{5c}\\[6pt] &=\bbox[5px,border:2px solid #C0A000]{\pi\coth(\pi)-2+2(m\bmod2)+O\!\left(4^{-m}\right)}\tag{5d} \end{align} Explanation:
$$\text{(5b)}$$: apply $$\zeta(4m-2n+1)=1+O\!\left(4^{-m}\right)$$
$$\phantom{\text{(5b):}}$$ then substitute $$n\mapsto n-1$$
$$\phantom{\text{(5b):}}$$ then compensate for $$2\sum_{n=1}^m(-1)^n$$ with $$2(m\bmod2)$$
$$\text{(5c)}$$: apply $$(4)$$ noting that $$\zeta(2n)-1=O\!\left(4^{-n}\right)$$
$$\text{(5d)}$$: rearrange to make things look nicer

Note that since $$n\le m-1$$, $$\zeta(4m-2n+1)-\zeta(4m-2n+2)=O\!\left(4^{-m}\right)\tag6$$ and so $$\text{(5d)}$$ also works for $$\zeta(4m-2n+2)$$ in place of $$\zeta(4m-2n+1)$$.

Graphical Verification

Plotting $$4^m$$ times the absolute error of the approximation given in $$\text{(5d)}$$ shows that it is bounded and appears to tend to a limit around $$0.6$$.

I suspect that this is known but I am unable to find references

$$S_m=2\sum_{n = 0}^{m - 1}(-1)^n\,\zeta(4m - 2n + 1)\,\zeta(2n + 2)$$ If $$m$$ is even and large, it is asymptotic to $$\pi \coth (\pi )-2$$ and if $$m$$ is odd to $$\pi \coth (\pi )$$ and the values are very quickly approached.

• How did you conclude that If $m$ is odd then $$2\sum_{n = 0}^{m - 1}(-1)^n\,\zeta(4m - 2n + 1)\,\zeta(2n + 2) \approx \pi\coth(\pi)$$ Apr 18, 2021 at 7:22
• @BooleanWick. I just knew it. The problem is that I don't remember where it could be proved (problem of age !). Apr 18, 2021 at 7:32
• This is really surprising. While $\cot x$ has expansion in terms of Bernoulli numbers/zeta function, it deals with $\zeta(2n)$. Don't know how this is coming up in a sum involving both even and odd zeta values. Apr 18, 2021 at 9:03
• @ParamanandSingh Yes, indeed. Do you have any insight on how to prove it? Thanks. Apr 18, 2021 at 9:11
• (+1) for a good memory! :-)
– robjohn
Apr 20, 2021 at 13:59

Not an answer, but this is too long for a comment and basically sums up my train of thoughts.

Let's rewrite the summation with $$2j = 2n+2$$ and $$2k = 4m+3$$ (an odd number), i.e., $$2\sum_{j=1}^{\frac{2k-3}{4}-1}(-1)^{j-1}\zeta(2k-2j)\zeta(2j)=-2\sum_{j=1}^{\frac{2k-3}{4}-1}(-1)^{j}\zeta(2k-2j)\zeta(2j).\quad(1)$$ Then, let's use the relation to Bernoulli's numbers: $$\zeta(2n)=\frac{(-1)^{n+1}B_{2n}(2\pi)^{2n}}{2(2n)!}.$$ and plug it in the (1): $$-2\sum_{j=1}^{\frac{2k-3}{4}-1}(-1)^j\frac{(-1)^{k-j+1}(2\pi)^{2k-2j}B_{2k-2j}}{2(2k-2j)!}\frac{(-1)^{j+1}(2\pi)^{2j}B_{2j}}{2(2j)!}=$$ $$-\frac{2(-1)^{k}(2\pi)^{2k}}{4}\sum_{j=1}^{\frac{2k-3}{4}-1} (-1)^j{{2k}\choose{2j}}B_{2k-2j}B_{2j}=\frac{(2\pi)^{2k}}{2}\sum_{j=1}^{\frac{2k-3}{4}-1} (-1)^{k-j+1}{{2k}\choose{2j}}B_{2k-2j}B_{2j}.$$ There seems to be some treatment of this sum in the literature, e.g., in Sums of Products of Bernoulli Numbers there is a reference (Section 5.2, although with a comment "A closed form for this alternating sum doesnot appear to be known.") to article "E. Grosswald, Die Werte der Riemannschen Zetafunktion an ungeraden Argumentstellen, to which unfortunately I don't have access." Maybe there is a hint to @Claude Leibovici answer. If anyone has access to that, I would be more than happy to look at the paper.

Just to add interesting relations which could be relevant to this problem (but I am sceptical)

• $$\sum_{j=1}^{k-1} {{2k}\choose{2j}}B_{2k-2j}B_{2j}=-(2k+1)B_{2k}$$
• $$\coth(\pi)=\frac{1}{\pi}\sum_{j=0}^\infty\frac{B_{2j}(2\pi)^{2j}}{(2j)!}$$
• $2k = 4m + 3$? How do you justify that? Apr 18, 2021 at 13:49
• $k$ wouldn't be an integer here..just wanted to "fit" $2k-2j$. Then looking at $k\to\infty$ it wouldn't make too much difference at the end. But I don't justify it. There is probably a better way. (It's more a wishful thinking than a rigour). Apr 18, 2021 at 15:10
• Good to see you again (do you remember year $2013$ ?) Apr 20, 2021 at 6:53
• @Claude Leibovici You're asking me or @pisoir? Apr 20, 2021 at 8:28
• The formulas you cite are for integer $k$ and $n$. Furthermore, except for $n=0$, $B_{2n+1}=0$, so the formulas won’t work for the non-integer cases to which you are trying to apply them.
– robjohn
Apr 20, 2021 at 13:51