Find this sum


My try: let

$$f(x)=\sum_{n=0}^{\infty}\dfrac{x^{3n+3}}{(3n+1)(3n+2)(3n+3)},|x|\le 1$$

then we have $$f^{(3)}(x)=\sum_{n=0}^{\infty}x^{3n}=\dfrac{1}{1-x^3}$$ then we find the $f(x)$,Following is very ugly(can you someone can post your follow solution,) have other simple methods? Thank you very much.

  • 4
    $\begingroup$ Generally speaking, $$\sum_{n=1}^\infty\frac1{(n+a)(n+b)(n+c)}=-\frac{\psi^{(0)}(a)\cdot(-b+c)+\psi^{(0)}(b)\cdot(-c+a)+\psi^{(0)}(c)\cdot(-a+b)}{(a-b)(b-c)(c-a)}$$ where $$\psi^{(0)}(t)=H_{t-1}-\gamma\qquad\qquad;\qquad\qquad H_t=\int_0^1\frac{1-x^t}{1-x}dx$$ In this case, $a=\frac13$ , $b=\frac23$ , $c=1$. $\endgroup$ – Lucian Nov 19 '13 at 13:01
  • $\begingroup$ Is this formula has any name? why dont you post this as a solution? $\endgroup$ – GA316 Nov 19 '13 at 13:46
  • $\begingroup$ this problem is from comption,so I think this methods some student don't know,Thank you $\endgroup$ – math110 Nov 19 '13 at 13:48
  • $\begingroup$ I think the integrations are all very doable. If I have time, I will detail them and produce the result. $\endgroup$ – Ron Gordon Nov 19 '13 at 13:56
  • $\begingroup$ Oh,@RonGordon Thank you,I have wait you, I fell this integral is very ugly. $\endgroup$ – math110 Nov 19 '13 at 14:03

Your try is excellent... Note that $f(0)=f'(0)=f''(0)=0$ hence $$ f(1)=\int_0^1f'(x)\,\mathrm dx=\int_0^1\int_0^xf''(y)\,\mathrm dy\,\mathrm dx=\int_0^1\int_0^x\int_0^yf'''(z)\,\mathrm dz\,\mathrm dy\,\mathrm dx, $$ that is, $$ 2f(1)=\int_0^1(1-z)^2f'''(z)\,\mathrm dz=\int_0^1\frac{1-z}{1+z+z^2}\,\mathrm dz. $$ The rest is routine. The change of variable $2z+1=\sqrt3t$ yields $$ 2f(1)=\int_{1/\sqrt3}^\sqrt3\frac{\sqrt3-t}{1+t^2}\,\mathrm dt=\left[\sqrt3\arctan t-\frac12\log(1+t^2)\right]_{1/\sqrt3}^\sqrt3. $$ Note that $\arctan\sqrt3=\pi/3$ and $\arctan1/\sqrt3=\pi/6$, hence $$ 2f(1)=\sqrt3\cdot\left(\frac\pi3-\frac\pi6\right)-\frac12\log4+\frac12\log\frac43, $$ that is, $$ f(1)=\frac14\left[\frac\pi{\sqrt3}-\log3\right]\approx0.1788. $$ Second method: The rational fraction is such that $$ \frac2{(3n+1)(3n+2)(3n+3)}=\frac1{3n+1}-\frac2{3n+2}+\frac1{3n+3}, $$ hence $$ 2f(x)=x^2g_1(x)-2xg_2(x)+g_3(x),\qquad g_k(x)=\sum_{n\geqslant0}\frac{x^{3n+k}}{3n+k}. $$ Thus, for each $k$, $$ g'_k(x)=\sum_{n\geqslant0}x^{3n+k-1}=\frac{x^{k-1}}{1-x^3}. $$ Since $g'_k(0)=0$ for every $k\geqslant1$, this yields $$ 2f(x)=x^2\int_0^x\frac{1}{1-t^3}\mathrm dt-2x\int_0^x\frac{t}{1-t^3}\mathrm dt+\int_0^x\frac{t^2}{1-t^3}\mathrm dt. $$ The change of variable $t=xu$ yields $$ 2f(x)=x^3\int_0^1\frac{1}{1-x^3u^3}(1-2u+u^2)\mathrm du, $$ that is, $$ 2f(x)=x^3\int_0^1\frac{1-u}{1-xu}\frac{1-u}{1+xu+x^2u^2}\mathrm du. $$ When $x\to1$, one obtains once again $$ 2f(1)=\int_0^x\frac{1-u}{1+u+u^2}\mathrm du. $$ More generally, for every integer $k\geqslant2$, $$ \sum_{n\geqslant0}\frac{(k-1)!}{(kn+1)(kn+2)\cdots(kn+k)}=\int_0^1\frac{(1-u)^{k-2}}{1+u+\cdots+u^{k-1}}\,\mathrm du. $$

| cite | improve this answer | |
  • $\begingroup$ This is nice methods,+1,and I think this sum have other nice methos $\endgroup$ – math110 Nov 19 '13 at 15:52
  • $\begingroup$ You might want to simplify $$ \frac{(1-u)^{k-2}}{\sum_{i=0}^{k-1}u^i}=\frac{(1-u)^{k-2}}{\frac{1-u^k}{1-u}}= \frac{(1-u)^{k-1}}{1-u^k} $$ $\endgroup$ – Tim Ratigan Nov 19 '13 at 19:11
  • $\begingroup$ @Tim.Ratigan Thanks for your comment. This does not seem to be a simplification but introduces a (false) singularity at $u=1$. $\endgroup$ – Did Nov 19 '13 at 19:33
  • $\begingroup$ I considered this, but the integral can be considered to limit at $1-\epsilon$ for $\epsilon>0$ arbitrarily close to $0$, which removes the problem of the undefined point. In any case, L'hopital indicates that $\lim_{u\to 1} \frac{(1-u)^{k-1}}{1-u^k}=0$ $\endgroup$ – Tim Ratigan Nov 19 '13 at 21:55
  • $\begingroup$ In the first displayed equation for your second method, the denominators on the right hand side should be $3n+1$, etc., not $n+1$, etc. $\endgroup$ – Barry Cipra Nov 19 '13 at 22:02

We all know that sums of the form $\displaystyle{\sum\frac1{(n+a)(n+b)}}$ or $\displaystyle{\sum\frac1{(n+a)(n+b)(n+c)}}$ etc., are telescopic in nature, and solving them is trivial: for integer values of a, b, and c, that is ! But what if a, b, and c are not integers? What then? Suddenly, things aren't so simple and trivial anymore, and the formerly banal and deceitfully tame-looking problem takes on whole new twist, gaining entirely unexpected dimensions of depth, meaning, and insight. The key, when dealing with such surprising turns of events, is to just take a step back, and try to parse or rephrase the older, well-worn solution in terms which might prove useful or relevant to circumventing the obstacles raised by the newly-encountered situation.

For instance, it is by no means difficult to show that $\displaystyle{\sum_{n=1}^\infty\frac1{(n+a)(n+b)}}=\frac{H_a-H_b}{a-b}$ , or that $$\sum_{n=1}^\infty\frac1{(n+a)(n+b)(n+c)}=\frac{(a-b)\cdot H_c\ +\ (b-c)\cdot H_a\ +\ (c-a)\cdot H_b}{(a-b)(b-c)(c-a)}$$ for natural values of a, b, c. Indeed, the user Random Variable has already proved it on this thread, though I now realize that my initial formula is slightly mistaken, in the sense that either the sum should begin at $0$, with the $\gamma$ present, or at $1$, with the $\gamma$ absent. Now, the whole question becomes how to redefine $H_m$ , so as to be able to extend its meaning to non-natural arguments. For natural arguments, we have the well-known formula $H_m=\displaystyle\sum_{k=1}^m\frac1k$ . Now, let us take the simple function $f_k(x)=\displaystyle\frac{x^k}k$ , and notice that $f_k'(x)=x^{k-1}$ . And since we know that $\displaystyle{\sum_{k=1}^mx^{k-1}=\frac{1-x^m}{1-x}}$ , this finally allows us to conclude that $\displaystyle{H_m=\int_0^1\frac{1-x^m}{1-x}dx}$ , which, unlike the previous formula, can easily be extended to non-natural arguments as well. This has already been done by Euler two and a half centuries ago, so it's hardly new territory. Now, in our case, $\{a,b,c\}=\left\{\frac13,\frac23,1\right\}$ , and the value of $H_1$ is $1$, so all that's left to do is to compute $H_\frac13$ and $H_\frac23$ using the above formula, since $$\sum_{n=0}^\infty\frac1{(3n+1)(3n+2)(3n+3)}=\frac16+\frac1{27}\cdot\sum_{n=1}^\infty\frac1{\left(n+\frac13\right)\left(n+\frac23\right)(n+1)}$$


$$=3\int_0^1\left(1-\frac{1+t}{1+t+t^2}\right)dt=3\,\bigg[\int_0^11\cdot dt-\tfrac12\int_0^1\frac{2+2t}{1+t+t^2}dt\bigg]=$$




Similarly for $\displaystyle{H_{\frac23}=\tfrac32+\frac\pi{2\sqrt3}-\tfrac32\ln3}$ . Then, by substituting these values back into the original formula above, we finally arrive at the desired result $\displaystyle{I=\frac{\pi\sqrt3-3\ln3}{12}}$ .

| cite | improve this answer | |
  • $\begingroup$ An alternate expression is $~H_m~=-\displaystyle\int_0^1\ln\Big(1-\sqrt[m]x\Big)~dx$. $\endgroup$ – Lucian Aug 12 '14 at 17:19

$\newcommand{\+}{^{\dagger}} \newcommand{\angles}[1]{\left\langle #1 \right\rangle} \newcommand{\braces}[1]{\left\lbrace #1 \right\rbrace} \newcommand{\bracks}[1]{\left\lbrack #1 \right\rbrack} \newcommand{\dd}{{\rm d}} \newcommand{\isdiv}{\,\left.\right\vert\,} \newcommand{\ds}[1]{\displaystyle{#1}} \newcommand{\equalby}[1]{{#1 \atop {= \atop \vphantom{\huge A}}}} \newcommand{\expo}[1]{\,{\rm e}^{#1}\,} \newcommand{\fermi}{\,{\rm f}} \newcommand{\floor}[1]{\,\left\lfloor #1 \right\rfloor\,} \newcommand{\ic}{{\rm i}} \newcommand{\imp}{\Longrightarrow} \newcommand{\ket}[1]{\left\vert #1\right\rangle} \newcommand{\pars}[1]{\left( #1 \right)} \newcommand{\partiald}[3][]{\frac{\partial^{#1} #2}{\partial #3^{#1}}} \newcommand{\pp}{{\cal P}} \newcommand{\root}[2][]{\,\sqrt[#1]{\,#2\,}\,} \newcommand{\sech}{\,{\rm sech}} \newcommand{\sgn}{\,{\rm sgn}} \newcommand{\totald}[3][]{\frac{{\rm d}^{#1} #2}{{\rm d} #3^{#1}}} \newcommand{\ul}[1]{\underline{#1}} \newcommand{\verts}[1]{\left\vert #1 \right\vert} \newcommand{\yy}{\Longleftrightarrow}$ $\ds{ I_{N} \equiv \sum_{n = 0}^{N}{1 \over \pars{3n + 1}\pars{3n + 2}\pars{3n + 3}}\,, \qquad I = I_{\infty} =\ {\large ?}}$

\begin{align} I_{N}&\equiv {1 \over 6}\sum_{n = 0}^{N}{1 \over n + 1/3} - {1 \over 3}\sum_{n = 0}^{N}{1 \over n + 2/3} + {1 \over 6}\sum_{n = 0}^{N}{1 \over n + 1} \\[3mm]&= {1 \over 6}\sum_{n = 0}^{N}\pars{{1 \over n + 1/3} - {1 \over n + 2/3}} + {1 \over 6}\sum_{n = 0}^{N}\pars{{1 \over n + 1} - {1 \over n + 2/3}} \\[3mm]&= {1 \over 18}\sum_{n = 0}^{N}{1 \over \pars{n + 1/3}\pars{n + 2/3}} - {1 \over 18}\sum_{n = 0}^{N}{1 \over \pars{n + 1}\pars{n + 2/3}} \end{align} We'll use the well known identity: $\ds{\sum_{n = 0}^{\infty}{1 \over \pars{n + z_{0}}\pars{n + z_{1}}} ={\Psi\pars{z_{0}} - \Psi\pars{z_{1}} \over z_{0} - z_{1}}}$ where $\Psi\pars{z}$ is the $\it\mbox{digamma function}$

\begin{align} I & = I_{\infty} = {1 \over 18}\,{\Psi\pars{1/3} - \Psi\pars{2/3} \over 1/3 - 2/3} - {1 \over 18}\,{\Psi\pars{1} - \Psi\pars{2/3} \over 1 - 2/3} \\[3mm]&= {1 \over 6}\,\bracks{\Psi\pars{2 \over 3} - \Psi\pars{1 \over 3}} - {1 \over 6}\,\bracks{\Psi\pars{1} - \Psi\pars{2 \over 3}} = -\,{1 \over 6}\,\Psi\pars{1} - {1 \over 6}\,\Psi\pars{1 \over 3} + {1 \over 3}\,\Psi\pars{2 \over 3} \end{align} However, \begin{align} \Psi\pars{1} &= -\gamma\quad\mbox{where}\ \gamma\ {\it\mbox{is the Euler-Mascheroni constant}}\ \pars{~\gamma = 0.5772\ldots~} \\[3mm] \Psi\pars{1 \over 3}&=-\gamma - {\root{3} \over 6}\,\pi - {3 \over 2}\,\ln\pars{3} \\[3mm] \Psi\pars{2 \over 3}&=-\gamma + {\root{3} \over 6}\,\pi - {3 \over 2}\,\ln\pars{3} \\ I & = I_{\infty} = -\,{1 \over 6}\pars{-\gamma} - {1 \over 6}\bracks{-\gamma - {\root{3} \over 6}\,\pi - {3 \over 2}\,\ln\pars{3}} + {1 \over 3}\bracks{-\gamma + {\root{3} \over 6}\,\pi - {3 \over 2}\,\ln\pars{3}} \\[3mm]&= {\root{3} \over 12}\,\pi - {1 \over 4}\,\ln\pars{3} \end{align}

$$\color{#0000ff}{\large \sum_{n = 0}^{N}{1 \over \pars{3n + 1}\pars{3n + 2}\pars{3n + 3}} = {\root{3} \over 12}\,\pi - {1 \over 4}\,\ln\pars{3}} \approx 0.1788 $$
| cite | improve this answer | |

OK, so let's start with

$$\frac{d^3}{dx^3} f(x) = \frac{1}{1-x^3}$$

where $f(0)=f'(0)=f''(0)=0$. Note that we may use partial fractions and a little rearranging to find that

$$\frac{1}{1-x^3} = \frac13 \left [\frac{1}{1-x} + \frac{x+\frac12}{\left (x+\frac12\right)^2+\frac{3}{4}}+\frac{3}{2} \frac{1}{\left (x+\frac12\right)^2+\frac{3}{4}}\right ]$$

Integrating and using $f''(0)=0$ we get that

$$f''(x) = -\frac13 \log{(1-x)} + \frac16 \log{(1+x+x^2)} + \frac{1}{\sqrt{3}} \arctan{\frac{2 x+1}{\sqrt{3}}} - \frac{\pi}{6 \sqrt{3}}$$

From here the challenge will be managing the various terms rather than individual integrations, which are easily done through parts. Note that, for example,

$$\int dx \, \log{(1+x+x^2)} = x \log{(1+x+x^2)} - \int dx \frac{2 x^2+x}{x^2+x+1}$$

You may use partial fractions on the latter integral and determine that

$$\int dx \frac{2 x^2+x}{x^2+x+1} = 2 x - \frac12 \log{(1+x+x^2)} - \sqrt{3} \arctan{\frac{2 x+1}{\sqrt{3}}} +C$$

Using a substitution ($u=(2 x+1)/\sqrt{3}$) and integration by parts, we similarly find that

$$\int dx \, \arctan{\frac{2 x+1}{\sqrt{3}}} = \frac{\sqrt{3}}{2}\left [\frac{2 x+1}{\sqrt{3}} \arctan{\frac{2 x+1}{\sqrt{3}}} - \frac12 \log{(1+x+x^2)}\right ] + C$$

Omitting a lot of algebra, I find that, using $f'(0)=0$:

$$f'(x) = \frac13 (1-x) \log{(1-x)} - \frac16 (1-x) \log{(1+x+x^2)} + \frac{1+x}{\sqrt{3}} \arctan{\frac{2 x+1}{\sqrt{3}}} - \frac{\pi}{6 \sqrt{3}} (1+x)$$

Now we integrate once more. Again, the challenge is bookkeeping, as all the integrals may be done by parts. Using $f(0)=0$, I get that

$$f(x) = -\frac16(1-x)^2 \log{(1-x)} + \frac{1}{12} (x^2-2 x-2) \log{(1+x+x^2)} + \frac{(x^2+2 x)}{2 \sqrt{3}} \arctan{\frac{2 x+1}{\sqrt{3}}} - \frac{\pi}{12 \sqrt{3}} (x^2+2 x)$$

The sum is then

$$f(1) = \frac{\sqrt{3}}{12} \pi - \frac14 \log{3} \approx 0.178$$

which checks out with, e.g., WA.

| cite | improve this answer | |

I undeleted my answer and fixed it up since one of the other answers refers to it.

In the comments, Lucian mentioned the generalization $$\sum_{n=0}^{\infty} \frac{1}{(n+a)(n+b)(n+c)} = \frac{(b-c) \, \psi(a)+ (c-a) \, \psi(b)+(a-b) \, \psi(c)}{(a-b)(b-c)(c-a)} \tag{1}$$ where $\psi(x)$ is the digamma function; $a,b,c >0$; $a \ne b$; $b \ne c$; and $a \ne c$.

We can prove $(1)$ by using the series representation $$\psi(x) = - \gamma + \sum_{n=0}^{\infty} \left(\frac{1}{n+1}-\frac{1}{n+x} \right) \, , \quad x>0.$$

Starting from the right side of $(1)$, we get

$$ \begin{align} &\frac{(b-c) \, \psi(a)+ (c-a) \, \psi(b)+(a-b) \, \psi(c)}{(a-b)(b-c)(c-a)} \\ &= \small\left[ (b-c)\left(-\gamma + \sum_{n=0}^{\infty} \left( \frac{1}{n+1} - \frac{1}{n+a} \right) \right)+ (c-a)\left(- \gamma + \sum_{n=0}^{\infty} \left( \frac{1}{n+1} - \frac{1}{n+b} \right) \right) + (a-b) \left( -\gamma +\sum_{n=0}^{\infty} \left( \frac{1}{n+1} - \frac{1}{n+c} \right) \right) \right] /{\left((a-b)(b-c)(c-a) \right)} \\ &= \frac{1}{(a-b)(b-c)(c-a)}\sum_{n=0}^{\infty} \left( \frac{c-b}{n+a}+\frac{a-c}{n+b} + \frac{b-a}{n+c} \right) \\ &=\frac{1}{(a-b)(b-c)(c-a)} \sum_{n=0}^{\infty} \ \frac{bc^{2}-b^{2}c+a^{2}c-ac^{2}+ab^{2}-a^{2}b}{(n+a)(n+b)(n+c)} \\ &= \frac{1}{(a-b)(b-c)(c-a)}\sum_{n=0}^{\infty} \frac{(a-b)(ac-c^{2}-ab+bc)}{(n+a)(n+b)(n+c)} \\ &=\frac{1}{(a-b)(b-c)(c-a)}\sum_{n=0}^{\infty} \frac{(a-b)(b-c)(c-a)}{(n+a)(n+b)(n+c)} \\ &= \sum_{n=0}^{\infty} \frac{1}{(n+a)(n+b)(n+c)} \end{align}$$

| cite | improve this answer | |

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.