0

I ran into the series

$$ \sum_{n=0}^{\infty} \frac{3072n^2 + 3072n + 832}{4096n^6 + 12288n^5 + 14592n^4 + 8704n^3 + 2736n^2 + 432n + 27} $$

and I’m trying to figure out one things:

  1. Is there a nice closed‐form for its sum?

What I’ve tried so far:

For large $n$, the term behaves like

$$ \frac{3072n^2}{4096n^6} \;=\;\tfrac34\,n^{-4}, $$

so since $\sum n^{-4}$ converges, I’m pretty sure the original series converges absolutely.

I thought maybe the denominator factors in a helpful way (e.g.\ $(4n+\alpha)^k$) or that one could write

$$ \frac{3072n^2 + \cdots}{\text{(denominator)}} = A\bigl[f(n)-f(n+1)\bigr], $$

but I haven’t been able to match coefficients or find a telescoping pattern.

I briefly considered expressing the sum in terms of polylogarithms or known zeta‐function values, but didn’t find any standard identity that fits.

Does anyone recognize a trick to turn this into a telescoping sum, or a contour‐integral / root‐of‐unity approach, or something else that gives a closed form? Even a reference to a similar example would be really helpful. Thanks in advance!

RobPratt
  • 50,938
Frank
  • 2,434
  • 3
  • 12
  • 11
    You're the highest-reputation user with the "New contributor" decoration I've ever seen. I see that you received three 500-rep bounties within a day, and have two more answers potentially getting 900 rep. That's impressive. – Benjamin Wang Jun 19 '25 at 06:19
  • 3
    @BenjaminWang three 500-rep bounties in 1 days, over 3 old answers. :) – NN2 Jun 19 '25 at 06:22
  • 3
    Is there any context for how you got this really specific-looking sum? – Kotlopou Jun 19 '25 at 06:26
  • 1
    @NN2 That user seems to be giving away all their reputation, might be planning to leave the site.I answered questions from three different people. I’m not sure what’s suspicious? – Frank Jun 19 '25 at 06:28
  • 7
    Those bounties came from 1 person, Frank answered all of them almost exactly the day the bounties started, then shortly after that, those bounties were awarded. Totally coincidences, right? It can't get anymore suspicious. Though it is not my job to find out the truth. It might sound unfriendly if I keep speaking in this manner but I just said the facts. If there's nothing behind it, then there is nothing. Regardless, Frank's answers are good for real, and they deserve. – Quý Nhân Jun 19 '25 at 06:43
  • 2
    @QuýNhân It’s not unfriendly, if I were in your place, I might also find it suspicious. It’s okay, since I know there’s no connection between me and that user. I think the system would have flagged it if either my account or theirs were sockpuppets. – Frank Jun 19 '25 at 07:01
  • 7
    @Frank I'll be honest, I was a bit disappointed about how the bounty was awarded on that integral question, and I'm afraid I let that feeling affect my response to your question on convergence. It was immature of me to ignore your request for a proof, and I sincerely apologize. It was an unfair reaction, especially since I find your other answers to be excellent (+1). I'll get to work on that proof for you. – Bumblebee Jun 19 '25 at 07:28
  • 1
    @Bumblebee Oh please, don’t apologize, it’s totally fine. I also felt a bit weird when they awarded me the bounty for this question: https://math.stackexchange.com/questions/5018577/evaluate-int-pi-2-0-x-sin3x-ln-left-frac1-cosx-sinx-rightdx/5076146#5076146 They accepted your answer but gave the bounty to me – Frank Jun 19 '25 at 07:42
  • 1
    @QuýNhân reminds me of the great Cleo, comes out of nowhere and dips after dropping crazy results – Amrut Ayan Jun 19 '25 at 08:50
  • 1
    @AmrutAyan Cleo is https://math.stackexchange.com/users/19661/vladimir-reshetnikov – Frank Jun 19 '25 at 08:57
  • 1
    @Frank I did see that, it was unexpected haha! – Amrut Ayan Jun 19 '25 at 09:01
  • The command of Mathematica 14.2.1 Sum[(3072*n^2 + 3072*n + 832)/(4096*n^6 + 12288*n^5 + 14592*n^4 + 8704*n^3 + 2736*n^2 + 432*n + 27), {n, 0, Infinity}] results in $\pi ^3$ in a moment – user64494 Jun 19 '25 at 12:18
  • Let us compare with Sum[(3072*n^2 + 3072*n + 832)/(4096*n^6 + 12288*n^5 + 14592*n^4 + 8704*n^3 + 2736*n^2 + 432*n + 17), {n, 0, Infinity}]//ToRadicals ($17$ instead of $27$) which results in $$ -\frac{4}{3}\left(\left(13 \psi ^{(0)}\left(\frac{1}{4} \left(2-\sqrt{1+\sqrt[3]{10}}\right)\right)+\right.\right.\cdots \left.\left.16 \left(-1-\sqrt{1-4 \left(\frac{3}{16}+\frac{\sqrt[3]{5} \left(1+i \sqrt{3}\right)}{16\ 2^{2/3}}\right)}\right)^5\right)\right) $$ – user64494 Jun 19 '25 at 16:11
  • Mathematica: $\pi^3$ – David G. Stork Jun 28 '25 at 14:55
  • And, in any case, what was the context in which this arose, if you don't mind? – paul garrett Jun 28 '25 at 20:32

3 Answers3

17

$$S= \sum_{n=0}^{\infty} \frac{3072n^2 + 3072n + 832}{4096n^6 + 12288n^5 + 14592n^4 + 8704n^3 + 2736n^2 + 432n + 27} $$

Notice the first term of denominator, $4096n^6$ is $64n^3\times 64n^3$, that was a hint to the factoring of $4n+\text{something}$; $$4096n^6 + 12288n^5 + 14592n^4 + 8704n^3 + 2736n^2 + 432n + 27=(4n+1)^3(4n+3)^3$$

Using partial fractions, $$\frac{3072n^2 + 3072n + 832}{4096n^6 + 12288n^5 + 14592n^4 + 8704n^3 + 2736n^2 + 432n + 27}=\frac{32}{(4n+1)^3}-\frac{32}{(4n+3)^3}$$

$$S=32\sum_{n=0}^{\infty}\frac{1}{(4n+1)^3}-32\sum_{n=0}^{\infty}\frac{1}{(4n+3)^3}=32S_1-32S_2$$

Using the definition of Hurzwitz Zeta function, $$\color{blue}{\displaystyle \zeta (s,a)=\sum _{n=0}^{\infty }{\frac {1}{(n+a)^{s}}}}$$

$$S_1=\frac{1}{64}\sum_{n=0}^{\infty}\frac{1}{\left(n+\frac14\right)^3}=\frac{1}{64}\zeta \left(3,\frac14\right)$$

$$S_2=\frac{1}{64}\sum_{n=0}^{\infty}\frac{1}{\left(n+\frac34\right)^3}=\frac{1}{64}\zeta \left(3,\frac34\right)$$

$$\therefore S=\frac{1}{2}\zeta \left(3,\frac14\right)-\frac{1}{2}\zeta \left(3,\frac34\right)=\pi^3$$

An interesting identity,

$$\text{Cl}_m\left(\frac{\pi}2\right) = \sum_{k=1}^\infty\frac{\sin\left(k\,\frac{\pi}2\right)}{k^m}=\sum_{n=0}^\infty\left(\frac1{(4n+1)^m}-\frac1{(4n+3)^m}\right)$$

$$\text{Cl}_3\left(\frac{\pi}2\right) = \sum_{k=1}^\infty\frac{\sin\left(k\,\frac{\pi}2\right)}{k^3}=\sum_{n=0}^\infty\left(\frac1{(4n+1)^3}-\frac1{(4n+3)^3}\right)$$

$$\therefore S=32\text{Cl}_3\left(\frac{\pi}2\right)$$

Amrut Ayan
  • 8,887
3

Doing the same as @Amrut Ayan,

$$S_p=\sum_{n=0}^p \Bigg(\frac{32}{(4n+1)^3}-\frac{32}{(4n+3)^3}\Bigg)$$ $$S_p=\frac{1}{4} \left(\psi ^{(2)}\left(p+\frac{5}{4}\right)-\psi ^{(2)}\left(p+\frac{7}{4}\right)+\psi ^{(2)}\left(\frac{3}{4}\right)-\psi ^{(2)}\left(\frac{1}{4}\right)\right)$$

Asymptotically $$S_p=\frac{1}{4} \left(\psi ^{(2)}\left(\frac{3}{4}\right)-\psi ^{(2)}\left(\frac{1}{4}\right)\right)-\frac{1}{4 p^3}+\frac{3}{4 p^4}+O\left(\frac{1}{p^5}\right)$$ and $$\frac{1}{4} \left(\psi ^{(2)}\left(\frac{3}{4}\right)-\psi ^{(2)}\left(\frac{1}{4}\right)\right)=\pi ^3$$ Using the truncated series, the relative error is $3.21\times 10^{-3}$ percent if $p=4$.

2

Knowing that these values come from the Hurwitz zeta function, we can use the Euler-Maclaurin formula. Let $f(n)$ be the summand. By the Euler–Maclaurin formula, $$S=\frac{32}{9}+\frac{832}{54}+\int_0^{+\infty}f^{(1)}(t)\tilde B_1(t)\ {\rm d}t,\quad f^{(1)}(t) = -6144\frac{(2t+1)(16t^2+16t+5)}{(16t^2+16t+3)^4}.$$ Now the Euler–Maclaurin formula on $\zeta(3,\tfrac{1}{4})$ and $ \zeta(3,\tfrac{3}{4})$ yields, $$\begin{align*}\zeta(3,\tfrac{1}{4})=\sum_{n=0}^\infty \frac{1}{(n+1/4)^3}=8+32+\int_0^{+\infty} g_1^{(1)}(t)\tilde B_1(t)\ {\rm d}t&,\quad g^{(1)}_1(t) = -\frac{3}{(t+1/4)^4},\\ \zeta(3,\tfrac{3}{4})=\sum_{n=0}^\infty\frac{1}{(n+3/4)^3} = \frac{8}{9}+\frac{32}{27}+\int_0^{+\infty}g_2^{(1)}(t)\tilde B_1(t)\ {\rm d}t &,\quad g_2^{(1)}(t) = -\frac{3}{(t+3/4)^4}.\end{align*}$$ One can check $\frac{1}{2}(g_1^{(1)}(t) - g_2^{(1)}(t)) = f^{(1)}(t)$, so it turns out that, $$\frac{1}{2}\zeta(3,\tfrac{1}{4})-\frac{1}{2}\zeta(3,\tfrac{3}{4}) = \frac{32}{9}+\frac{832}{54}+\int_0^{+\infty} f^{(1)}(t)\tilde B_1(t)\ {\rm d}t=S.$$

bob
  • 3,079