18

Is it possible to evaluate this integral in closed form? $$ \int_0^1 \frac{du}{u}\text{Li}_2(u)^2\log u \stackrel{?}{=} -\frac{\zeta(6)}{3}.$$ I found the possible closed form using an integer relation algorithm.

I found several other possible forms for similar integrals, including $$ \int_0^1 \frac{du}{u}\text{Li}_2(u)^2(\log u)^2 \stackrel{?}{=} -20\zeta(7)+12\zeta(2)\zeta(5).$$

There doesn't seem to be an equivalent form when the integrand contains $(\log u)^3$, at least not just in terms of $\zeta$.

Does anybody know a trick for evaluating these integrals?

Update. The derivation of the closed form for the second integral follows easily along the ideas O.L. used in the answer for the first integral.

Introduce the functions $$ I(a,b,c) = \int_0^1 \frac{du}{u}(\log u)^c \text{Li}_a(u)\text{Li}_b(u) $$ and $$ S(a,b,c) = \sum_{n,m\geq1} \frac{1}{n^am^b(n+m)^c}. $$ Using integration by parts, the expansion of polylogarithms from their power series definition and also that $$ \int_0^1 (\log u)^s u^{t-1}\,du = \frac{(-1)^s s!}{t^{s+1}},$$ check that $$ I(2,2,2) = -\frac23 I(1,2,3) = 4S(1,2,4). $$

Now use binomial theorem and the fact that $S(a,b,c)=S(b,a,c)$ to write $$ 6S(1,2,4) + 2S(3,0,4) = 3S(1,2,4) + 3S(2,1,4)+S(0,3,4)+S(3,0,4) = S(3,3,1). $$ Now, using Mathematica, $$ S(3,3,1) = \sum_{n,m\geq1}\frac{1}{n^3m^3(n+m)} = \sum_{m\geq1}\frac{H_m}{m^6} - \frac{\zeta(2)}{m^5} + \frac{\zeta(3)}{m^4}, $$ and $$ \sum_{m\geq1}\frac{H_m}{m^6} = -\zeta(4)\zeta(3)-\zeta(2)\zeta(5)+4\zeta(7), $$ so $$ S(3,3,1) = 4\zeta(7)-2\zeta(2)\zeta(5). $$

Also, $$ S(0,3,4) = \zeta(3)\zeta(4) - \sum_{m\geq1} \frac{H_{n,4}}{m^3} = -17\zeta(7)+10\zeta(2)\zeta(5)+\zeta(3)\zeta(4), $$ from which it follows that $$ I(2,2,2) = \frac23\left(S(3,3,1)-2S(0,3,4)\right) = -20\zeta(7)+12\zeta(2)\zeta(5). $$

Ali Shadhar
  • 25,498
Kirill
  • 14,494
  • I derived this integral representations for Euler's sum. Try if you can relate your integral to one of these sums. – Mhenni Benghorbal Jun 23 '13 at 07:58
  • Check these integral techniques which they may be useful for your purpose. – Mhenni Benghorbal Jun 23 '13 at 08:01
  • Trying to evaluate some Euler sums is actually why I found these sorts of integrals in the first place, although these two don't come exactly from Euler sums as far as I can tell. – Kirill Jun 23 '13 at 21:12
  • In general you will find that: $\displaystyle\int _{0}^{1}!{\frac {{\text{ Li}_a} \left( u \right) {\text{ Li}_b} \left( u \right) \left(\log u \right) ^{c}}{u}}{d u}= \left( -1 \right) ^{c}\Gamma \left( c+1 \right) \sum _{k=1}^{ \infty } \left( \sum _{j=1}^{\infty }{\frac {1}{ \left( j+k \right) ^{ c+1}{j}^{a}{k}^{b}}} \right)$, so $I(a,b,c)= \left( -1 \right) ^{c}\Gamma \left( c+1 \right) S(a,b,c+1) $. Proof: $u=e^{-y}$, series expand polylogs, $\Gamma \left( c+1 \right) =\int _{0}^{\infty }!{y}^{c}{{\rm e}^{-y}} {dy}$, $\mathbb{R}(c)>0$. – Graham Hesketh Jun 24 '13 at 16:56
  • Sorry: $\mathbb{R}(c)\ge-1$. – Graham Hesketh Jun 24 '13 at 17:02
  • Another formula that may be useful: $\displaystyle S \left( a,b,c+1 \right) =\sum _{j=0}^{d}{n\choose j}S \left( a-j,b-d+ j,c+1+d \right)$ , for$d\ge0$. Proof: multiply the top and bottom of the summand in the definition of $S$ by $(j+k)^d$ then use the binomial expansion in the numerator. A special case of this formula for $c=-1$ leads to:$\zeta \left( a \right) \zeta \left( b \right) =\sum _{j=0}^{d}{n \choose j}S \left( a-j,b-d+j,d \right) $. – Graham Hesketh Jun 24 '13 at 19:34
  • Is that last expression invertible???? – Graham Hesketh Jun 24 '13 at 19:37
  • ${n\choose j}$ should read ${d\choose j}$ – Graham Hesketh Jun 25 '13 at 10:14
  • You might find interesting my answer here. I've used techniques from this paper. – user153012 Mar 12 '15 at 02:14

5 Answers5

10

Now it is a proof.


Let us integrate once by parts to replace the (first) integral by $$I=\int_0^1\frac{\ln u\,\mathrm{Li}_2(u)^2du}{u}=\int_0^1\frac{\ln^2 u\ln(1-u)}{u}\mathrm{Li}_2(u)\,du.$$ Next replace $\mathrm{Li}_2(u)=\sum_{m=1}^{\infty}u^m/m^2$ and $\ln(1-u)=-\sum_{n=1}^{\infty}u^{n}/n$ by the corresponding Taylor series. Exchanging the order of summation and integration, evaluate the integrals with respect to $u$. This can be done using that $$\int_0^1 u^{s-1}\ln^2u\,du=\frac{2}{s^3}.$$ So $I$ can be written as a double series $$I=-2\sum_{m=1}^{\infty}\sum_{n=1}^{\infty}\frac{1}{m^2n(m+n)^3}.$$ Now let us introduce the following sums: \begin{align} &S_1=\sum_{m=1}^{\infty}\sum_{n=1}^{\infty}\frac{1}{m^3(m+n)^3},\\ &S_2=\sum_{m=1}^{\infty}\sum_{n=1}^{\infty}\frac{1}{n^3(m+n)^3},\\ &S_3=\sum_{m=1}^{\infty}\sum_{n=1}^{\infty}\frac{1}{m^2n(m+n)^3},\\ &S_4=\sum_{m=1}^{\infty}\sum_{n=1}^{\infty}\frac{1}{mn^2(m+n)^3}. \end{align} It is obvious that $S_1=S_2$ and $S_3=S_4$. What is more funny (but still obvious to prove) is that $$S_1+S_2+3S_3+3S_4=\sum_{m=1}^{\infty}\sum_{n=1}^{\infty}\frac{1}{m^3n^3}=\zeta(3)^2.\tag{1}$$ Therefore, if we manage to compute $S_1=S_2$, we will be able to compute $I$. But $$S_1=-\sum_{m=1}^{\infty}\frac{\psi''(1+m)}{2m^3}=\frac12\left(\zeta(3)^2-\frac{\pi^6}{945}\right).\tag{2}$$ Here the first equality follows from the recursion relation $\psi''(z+1)-\psi''(z)={2}/{z^3}$ and telescoping argument, whereas the second was obtained using Mathematica.

Now combining (1), (2) and the fact that $I=-(S_3+S_4)$, we find $$I=-\frac13\left[\left(S_1+S_2+3S_3+3S_4\right)-2S_1\right]=-\frac{1}{3}\times\frac{\pi^6}{945}=-\frac{\zeta(6)}{3}.$$

Start wearing purple
  • 53,234
  • 13
  • 164
  • 223
  • Very nice!${}{}{}$ – Kirill Jun 23 '13 at 21:28
  • @Kirill Thanks! I am now thinking about the second integral. Following the same approach, the main obstacle is that Mathematica cannot compute the analog of the formula (2). I actually think that understanding how to obtain (2) by hands would be sufficient to prove the 2nd integral. – Start wearing purple Jun 23 '13 at 21:34
  • The reason I originally considered these integrals is that $$ \sum_{m\geq1}H_{m,r}z^m = \frac{\text{Li}r(z)}{1-z}, $$ where $H{m,r+1}=\psi_{r}(1+m)-\psi_{r}(1)$. So evaluating $(2)$ is equivalent to evaluating a integrals of the form that I gave in the question. (I know this doesn't help much.) – Kirill Jun 23 '13 at 21:39
  • I guess I was hoping to evaluate these integrals without converting them back to Euler sums, using some other technique. (Although I had no way of evaluating them at all.) – Kirill Jun 23 '13 at 21:43
  • @O.L.: This technique is similar to the one I used to derive "other forms of solution 1" in this problem(+1). – Mhenni Benghorbal Jun 23 '13 at 22:36
  • @O.L. The second integral can be evaluated using exactly the same technique. I updated the question to acknowledge this. – Kirill Jun 23 '13 at 22:47
3

By Cauchy product we have

$$\operatorname{Li}_2^2(x)=\sum_{n=1}^\infty\left(\frac{4H_n}{n^3}+\frac{2H_n^{(2)}}{n^2}-\frac{6}{n^4}\right)x^n$$

Multiply both sides by $\frac{\ln^2x}{x}$ then integrate from $x=0$ to $1$ and use the fact that $\int_0^1 x^{n-1}\ln^2xdx=\frac{2}{n^3}$

we get

$$\int_0^1\frac{\operatorname{Li}_2^2(x)\ln^2x}{x}dx=8\sum_{n=1}^\infty \frac{H_n}{n^6}+4\sum_{n=1}^\infty\frac{H_n^{(2)}}{n^5}-12\zeta(7)$$

By Euler identity we have $$\sum_{n=1}^\infty \frac{H_n}{n^6}=4\zeta(7)-\zeta(2)\zeta(5)-\zeta(3)\zeta(4)$$ and in my solution here I managed to prove $$\sum_{n=1}^\infty \frac{H_n^{(2)}}{n^5}=-10\zeta(7)+5\zeta(2)\zeta(5)+2\zeta(3)\zeta(4)$$

By collecting these results we get

$$\int_0^1\frac{\operatorname{Li}_2^2(x)\ln^2x}{x}dx=-20\zeta(7)+12\zeta(2)\zeta(5)$$

Ali Shadhar
  • 25,498
2

I've decided to publish my work so far - I do not promise a solution, but I've made some progress that others may find interesting and/or helpful.

$$\text{Let } I_{n,k}=\int_{0}^{1}\frac{\text{Li}_{k}(u)}{u}\log(u)^{n}du$$ Integrating by parts gives $$I_{n,k}=\left[\text{Li}_{k+1}(u)\log(u)^{n}\right]_{u=0}^{u=1}-\int_{0}^{1}\frac{\text{Li}_{k+1}(u)}{u}n\log(u)^{n-1}du$$ $$\text{Hence, }I_{n,k}=-nI_{n-1,k+1} \implies I_{n,k}=(-1)^{r}\frac{n!}{(n-r)!}I_{n-r,k+r}$$ Taking $r=n$ gives $I_{n,k}=(-1)^{n}n!I_{0,n+k}$. $$\text{But obviously } I_{0,n+k}=\int_{0}^{1}\frac{\text{Li}_{n+k}(u)}{u}du=\text{Li}_{n+k+1}(1)-\text{Li}_{n+k+1}(0)=\zeta(n+k+1)$$

$$\text{Now consider }J_{n,k,l}=\int_{0}^{1}\frac{\text{Li}_{k}(u)}{u}\text{Li}_{l}(u)\log(u)^{n}du$$ Integrating by parts again, $$J_{n,k,l}=\left[\text{Li}_{k+1}(u)\text{Li}_{l}(u)\log(u)^{n}\right]_{0}^{1}-\int_{0}^{1}\frac{\text{Li}_{l-1}(u)}{u}\text{Li}_{k+1}(u)\log(u)^{n}-\int_{0}^{1}\frac{n\log(u)^{n-1}}{u}\text{Li}_{k+1}(u)\text{Li}_{l}(u) du$$ So $J_{n,k,l}=-J_{n,k+1,l-1}-nJ_{n-1,k+1,l}$; continuing in the spirit of the first part suggests that we ought to try to increase the first and second indices, while decreasing the third. If we can succeed in this, we have found a closed form.

  • Thank you. When I was looking at this myself, I also found some nice closed forms for $J_{n,k,l}$ for small $n$, $k$ and $l$; $J_{1,2k,0}$ is especially simple because of integration by parts. – Kirill Jun 20 '13 at 01:19
  • Since the differentiation and integration of $\text{Li}_{k}^{m}$ introduces other indices, I think analysing $J$ may be the more sensible of the two options. I'll try to put in a bit more work on paper to make more progress, and it does seem that the closed form you seek exists. The goal, it seems, is to increase both $k$ and $l$ while decreasing $l$. – preferred_anon Jun 20 '13 at 16:42
  • That should say "increase $k$ and $l$ while decreasing $n$" – preferred_anon Jun 23 '13 at 21:12
1

\begin{align}J&=\int_0^1 \frac{\ln^2 x\left(\text{Li}_2(x)\right)^2}{x}dx\\ &\overset{\text{IBP}}=\frac{1}{3}\Big[\left(\text{Li}_2(x)\right)^2\ln^3 x\Big]_0^1+\frac{2}{3}\int_0^1 \frac{\text{Li}_2(x)\ln(1-x)\ln^3 x}{x}dx\\ &\overset{\text{IBP}}=\frac{1}{6}\Big[\text{Li}_2(x)\ln(1-x)\ln^4 x\Big]_0^1+\frac{1}{6}\int_0^1 \ln^4 x\left(\frac{\ln^2(1-x)}{x}+\frac{\text{Li}_2(x)}{1-x}\right)dx\\ &\overset{\text{IBP}}=\left(\frac{1}{30}\Big[\ln^5 x\ln^2(1-x)\Big]_0^1+\frac{1}{15}\int_0^1 \frac{\ln(1-x)\ln^5 x}{1-x}dx\right)+\frac{1}{6}\int_0^1 \frac{\text{Li}_2(x)\ln^4 x}{1-x}dx\\ &=\frac{1}{15}\underbrace{\int_0^1 \frac{\ln(1-x)\ln^5 x}{1-x}dx}_{J_1}+\frac{1}{6}\underbrace{\int_0^1 \frac{\text{Li}_2(x)\ln^4 x}{1-x}dx}_{J_2}\\ C&=\int_0^1\frac{\ln^5 t}{1-t}dt\\ J_1&\overset{\text{IBP}}=\left[\left(\int_0^x\frac{\ln^5 t}{1-t}dt-C\right)\ln(1-x)\right]+\int_0^1 \frac{1}{1-x}\left(\int_0^x\frac{\ln^5 t}{1-t}dt-C\right)dx\\ &=\int_0^1 \int_0^1 \left(\frac{x\ln^5(tx)}{(1-t)(1-x)}-\frac{C}{1-x}\right)dt dx\\ &=\int_0^1 \int_0^1 \left(\frac{\ln^5(tx)}{(1-t)(1-x)}-\frac{\ln^5(tx)}{(1-t)(1-tx)}-\frac{C}{1-x}\right)dt dx\\ &=-240\zeta(2)\zeta(5)-240\zeta(3)\zeta(4)+\int_0^1 \frac{1}{1-t}\left(\int_t^1 \frac{\ln^5 u}{1-u}du\right)dt-\\&\int_0^1 \frac{1}{t}\left(\int_0^t \frac{\ln^5 u}{1-u}du\right)dt+\int_0^1 \int_0^1 \left(\frac{\ln^5 t}{(1-t)(1-x)}-\frac{C}{1-x}\right)dt dx\\ &=-240\zeta(2)\zeta(5)-240\zeta(3)\zeta(4)-J_1+\int_0^1 \frac{\ln^6 t}{1-t}dt+\\&\underbrace{\int_0^1 \int_0^1 \left(\frac{\ln^5 t}{(1-t)(1-x)}-\frac{C}{1-x}\right)dt dx}_{K}\\ &0\leq A<1\\ K(A)&=\int_0^A \int_0^A \left(\frac{\ln^5 t}{(1-t)(1-x)}-\frac{C}{1-x}\right)dt dx\\ &=\ln(1-A)\left(AC-\int_0^A \frac{\ln^5 t}{1-t}dt\right)\\ &=-(1-A)\ln(1-A)C+\ln(1-A)\int_A^1 \frac{\ln^5 t}{1-t}dt\\ K&=\lim_{A\rightarrow 1}K(A)\\ &=0 \end{align} Therefore,

$\boxed{\displaystyle J_1=360\zeta(7)-120\zeta(2)\zeta(5)-120\zeta(3)\zeta(4)}$ \begin{align*}J_2&\overset{\text{IBP}}=\left[\left(\int_0^x\frac{\ln^4 t}{1-t}dt\right)\text{Li}_2(x)\right]_0^1+\int_0^1 \frac{\ln(1-x)}{x} \left(\int_0^x\frac{\ln^4 t}{1-t}dt\right)dx\\ &\overset{\text{IBP}}=24\zeta(5)\zeta(2)+\left[\left(\int_0^x\frac{\ln^4 t}{1-t}dt\right)\ln(1-x)\ln x\right]_0^1-\\&\int_0^1 \ln x\left(\frac{\ln^4 x\ln(1-x)}{1-x}-\frac{1}{1-x}\left(\int_0^x\frac{\ln^4 t}{1-t}dt\right)\right)dx\\ &=24\zeta(5)\zeta(2)-J_1+\underbrace{\int_0^1 \frac{\ln x}{1-x}\left(\int_0^x\frac{\ln^4 t}{1-t}dt\right)dx}_{J_3}\\ J_3&\overset{\text{IBP}}=-24\zeta(5)\zeta(2)-\int_0^1 \frac{\ln^4 x}{1-x}\left(\int_0^x\frac{\ln t}{1-t}dt\right)dx\\ &=-24\zeta(5)\zeta(2)-\int_0^1\int_0^1 \frac{x\ln^4 x\ln(tx)}{(1-x)(1-tx)}dt dx\\ &=-24\zeta(5)\zeta(2)-\int_0^1\int_0^1 \left(\frac{\ln^4 x\ln(tx)}{(1-t)(1-x)}-\frac{\ln^4 x\ln(tx)}{(1-t)(1-tx)}\right)dt dx\\ &=-\int_0^1 \frac{1}{1-t}\left(\int_t^1 \frac{\ln^5 u}{1-u}du\right)dt+\int_0^1 \frac{1}{t}\left(\int_0^t \frac{\ln^5 u}{1-u}du\right)dt-4\underbrace{\int_0^1 \frac{\ln t}{1-t}\left(\int_0^t \frac{\ln^4 u}{1-u}du\right)dt}_{J_1+J_2-24\zeta(2)\zeta(5)}-\\&4\int_0^1 \frac{\ln t}{t}\left(\int_0^t \frac{\ln^4 u}{1-u}du\right)dt+6\int_0^1 \frac{\ln^2 t}{1-t}\left(\int_0^t \frac{\ln^3 u}{1-u}du\right)dt+6\int_0^1 \frac{\ln^2 t}{t}\left(\int_0^t \frac{\ln^3 u}{1-u}du\right)dt-\\&4\int_0^1 \frac{\ln^3 t}{1-t}\left(\int_0^t \frac{\ln^2 u}{1-u}du\right)dt-4\int_0^1 \frac{\ln^3 t}{t}\left(\int_0^t \frac{\ln^2 u}{1-u}du\right)dt+\underbrace{\int_0^1 \frac{\ln^4 t}{1-t}\left(\int_0^t \frac{\ln u}{1-u}du\right)dt}_{-24\zeta(2)\zeta(5)-J_3}+\\& \int_0^1 \frac{\ln^4 t}{t}\left(\int_0^t \frac{\ln u}{1-u}du\right)dt\\ &\overset{\text{IBP}}=\underbrace{\Big(360\zeta(7)-120\zeta(2)\zeta(5)-120\zeta(3)\zeta(4)\Big)}_{J_1}- 720\zeta(7)-\\&4\Big(\big(360\zeta(7)-120\zeta(2)\zeta(5)-120\zeta(3)\zeta(4)\big)+J_2-24\zeta(2)\zeta(5)\Big)+4\times \frac{1}{2}\times 720\zeta(7)+\\&6\int_0^1 \frac{\ln^2 t}{1-t}\left(\int_0^t \frac{\ln^3 u}{1-u}du\right)dt-6\times \frac{1}{3}\times 720\zeta(7)-4\times -6\zeta(4)\times 2\zeta(3)+\\&4\int_0^1 \frac{\ln^2 t}{1-t}\left(\int_0^t \frac{\ln^3 u}{1-u}du\right)dt+4\times\frac{1}{4}\times 720\zeta(7)+\Big(-24\zeta(2)\zeta(5)-J_3\Big)-\frac{1}{5}\times 720\zeta(7)\\ &=-1224\zeta(7)+432\zeta(2)\zeta(5)+408\zeta(3)\zeta(4)-4J_2-J_3+10\int_0^1 \frac{\ln^2 t}{1-t}\left(\int_0^t \frac{\ln^3 u}{1-u}du\right)dt\\ J_3&=-612\zeta(7)+216\zeta(2)\zeta(5)+204\zeta(3)\zeta(4)-2J_2+5\underbrace{\int_0^1 \frac{\ln^2 t}{1-t}\left(\int_0^t \frac{\ln^3 u}{1-u}du\right)dt}_{J_4}\\ \end{align*} \begin{align*} J_4&=\int_0^1\int_0^1 \frac{t\ln^2 t\ln^3(tu)}{(1-t)(1-tu)}dtdu\\ &=\int_0^1\int_0^1\left(\frac{\ln^2 t\ln^3(tu)}{(1-t)(1-u)}-\frac{\ln^2 t\ln^3(tu)}{(1-u)(1-tu)}\right)dt du\\ &=-48\zeta(3)\zeta(4)-72\zeta(2)\zeta(5)+\\&\int_0^1\int_0^1\left(\frac{\ln^5 t}{(1-t)(1-u)}-\frac{\ln^2 u\ln^3(tu)-2\ln u\ln^4(tu)+\ln^5(tu)}{(1-u)(1-tu)}\right)dt du\\ &=-48\zeta(3)\zeta(4)-72\zeta(2)\zeta(5)+\int_0^1 \frac{1}{1-u}\left(\int_u^1 \frac{\ln^5 t}{1-t}dt\right)du-\\&\int_0^1 \frac{1}{u}\left(\int_0^t\frac{\ln^5 t}{1-t}dt\right)du-\underbrace{\int_0^1 \frac{\ln^2 u}{1-u}\left(\int_0^t\frac{\ln^3 t}{1-t}dt\right)du}_{J_4}-\\&\int_0^1 \frac{\ln^2 u}{u}\left(\int_0^t\frac{\ln^3 t}{1-t}dt\right)du+2\underbrace{\int_0^1 \frac{\ln u}{1-u}\left(\int_0^t\frac{\ln^4 t}{1-t}dt\right)du}_{J_3}+\\&2\int_0^1 \frac{\ln u}{u}\left(\int_0^t\frac{\ln^4 t}{1-t}dt\right)du\\ J_4&=-60\zeta(7)+24\zeta(2)\zeta(5)+36\zeta(3)\zeta(4)+J_3\\ J_3&=228\zeta(7)-84\zeta(2)\zeta(5)-96\zeta(3)\zeta(4)+\frac{1}{2}J_2\\ J_2&=\boxed{-264\zeta(7)+120\zeta(2)\zeta(5)+48\zeta(3)\zeta(4)}\\ \end{align*} Therefore, \begin{align*}\boxed{J=12\zeta(2)\zeta(5)-20\zeta(7)}\end{align*}

FDP
  • 13,647
0

You might find Two New Zeta Constants: Fractal String, Continued Fraction, and Hypergeometric Aspects of the Riemann Zeta Function interesting . It relates hypergeometric functions, polylogarithms, and the zeta function

crow
  • 1