Let $$I_n = \int_0^1 \frac{\log^n (1-x) \log^{n-1} (1+x)}{1+x} dx$$ In a recently published article, $I_n$ are evaluated for $n\leq 6$: $$\begin{aligned}I_1 &= \frac{\log ^2(2)}{2}-\frac{\pi ^2}{12} \\ I_2 &= 2 \zeta (3) \log (2)-\frac{\pi ^4}{360}+\frac{\log ^4(2)}{4}-\frac{1}{6} \pi ^2 \log ^2(2) \\ I_3 &= \small 6 \zeta (3)^2+6 \zeta (3) \log ^3(2)-2 \pi ^2 \zeta (3) \log (2)+24 \zeta (5) \log (2)-\frac{23 \pi ^6}{2520}+\frac{\log ^6(2)}{6}-\frac{1}{4} \pi ^2 \log ^4(2)-\frac{1}{12} \pi ^4 \log ^2(2) \\ I_4 &= \small{-12 \pi ^2 \zeta (3)^2+288 \zeta (3) \zeta (5)+12 \zeta (3) \log ^5(2)-12 \pi ^2 \zeta (3) \log ^3(2)+168 \zeta (5) \log ^3(2)+108 \zeta (3)^2 \log ^2(2)-2 \pi ^4 \zeta (3) \log (2)-48 \pi ^2 \zeta (5) \log (2)+720 \zeta (7) \log (2)-\frac{499 \pi ^8}{25200}+\frac{\log ^8(2)}{8}-\frac{1}{3} \pi ^2 \log ^6(2)-\frac{19}{60} \pi ^4 \log ^4(2)-\frac{1}{6} \pi ^6 \log ^2(2)} \end{aligned}$$ Based on these evidences, the author (me) made the conjecture that

For positive integer $n$, $I_n$ is in the algebra over $\mathbb{Q}$ generated by $\log(2)$ and $\{\zeta(m) | m\in \mathbb{Z}, m\geq 3\}$.

The closed-form of $I_5, I_6$ also satisfy this conjecture. $I_5$ is:


$I_6$ is:


Question: How to prove the conjecture for general $n$?

Any suggestion is appreciated.

Some remarks:

  1. Even $I_3,I_4,I_5,I_6$ are extremely challenging, someone brave enough might want to embark on finding them independently.

  2. $I_n$ is not related to beta function in an obvious way, so the well-known differentiation trick does not work here.

  3. For any $I_n$, the algorithm outlined in the article should produce closed-form of $I_n$ in a finite amount of time if the conjecture is true. However, the algorithm is a bit mechanical, so benefits little toward a proof for general $n$.

  4. Perhaps I am missing something, this conjecture is elementary to state, so it might have an easy proof and I was being negligent.
  • 1
    $\begingroup$ Are you interested to see $I_{3-6}$ evaluated independently here in this question, or you're only after the conjecture? $\endgroup$ – Zacky Oct 29 '19 at 11:34
  • 7
    $\begingroup$ I would try making the change of variable $(1+x)/2=y$ and then connect the result with the fact that $$ \int_0^{1/2} \frac{\log^p(1-x)\log^{p+1}(x)}{1-x}\textrm{d}x=-\frac{1}{2(1+p)}\log^{2(p+1)}(2)+\frac{1}{2}\lim_{\substack{x\to0 \\ y \to 1}}\frac{\partial^{2p+1}}{\partial x^p \partial y^{p+1}}\operatorname{B}(x,y),$$ which appears in the book (Almost) Impossible Integrals, Sums, and Series, on page $9$. $\endgroup$ – user97357329 Oct 29 '19 at 13:59
  • 2
    $\begingroup$ $$I_3=\int_0^{1/2} \left(\frac{\log ^5(2)}{1-t}+\frac{2 \log ^4(2) \log (1-t)}{1-t}+\frac{3 \log ^4(2) \log (t)}{1-t}+\frac{6 \log ^3(2) \log (1-t) \log (t)}{1-t}+\frac{2 \log (2) \log (1-t) \log ^3(t)}{1-t}+\frac{6 \log ^2(2) \log (1-t) \log ^2(t)}{1-t}+\frac{3 \log ^2(2) \log ^2(1-t) \log (t)}{1-t}+\frac{3 \log (2) \log ^2(1-t) \log ^2(t)}{1-t}+\frac{\log ^3(2) \log ^2(1-t)}{1-t}+\frac{3 \log ^3(2) \log ^2(t)}{1-t}+\frac{\log ^2(2) \log ^3(t)}{1-t}+\frac{\log ^2(1-t) \log ^3(t)}{1-t}\right) \textrm{d}t.$$ $\endgroup$ – user97357329 Oct 29 '19 at 15:20
  • 1
    $\begingroup$ @pisco for higher weights like $I_4$ we may also need to deal with the integrals of the type, $\displaystyle \int_0^{1/2} \frac{\log ^n(1-x) \log ^n(x)}{1-x} \textrm{d}x$, where we can use again a result inspired from (Almost) Impossible Integrals, Sums, and Series, that is $$\int_0^{1/2} \frac{\log ^n(2 (1-x)) \log ^n(2 x)}{1-x}\textrm{d}x=\int_0^1 \frac{\log ^n(1+x) \log ^n(1-x)}{1+x} \textrm{d}x.$$ Combined with the algebraic identities we should get rid of all troublesome series. $\endgroup$ – user97357329 Oct 29 '19 at 15:44
  • 1
    $\begingroup$ @AliShather Well, it's a tricky integral. Then one can do it without using Beta function as Cornel did here facebook.com/… $\endgroup$ – user97357329 Oct 30 '19 at 16:15

Denote $f(k,j)=\int_0^{\frac{1}{2}} \frac{\log ^j(1-y) \log ^k(y)}{1-y} \, dy$. Then for $j, k>1$ (RHS denotes Beta derivatives)

$$U(k,j):=jf(k,j-1)+kf(j,k-1)=-(-\log(2))^{j+k}+ k \left( \partial_a^{k-1} \partial_b^j B\right) (0,1)$$

Which is direct by IBP, separation, Beta derivatives and reflection $y\to 1-y$: $$\small jf(k,j-1)= -(-\log (2))^{j+k}+ k \int_0^{\frac{1}{2}} \frac{\log ^j(1-y) \log ^{k-1}(y)}{y} \, dy$$ $$\small =-(-\log (2))^{j+k}+ k \left(\int_0^{1}-\int_{\frac{1}{2}}^1 \right) \frac{\log ^j(1-y) \log ^{k-1}(y)}{y} \, dy$$ $$\small =-(-\log(2))^{j+k}+ k \left( \partial_a^{k-1} \partial_b^j B\right) (0,1)-kf(j,k-1)$$ Thus taking $\frac{\binom{n-1}{j-1} \binom{n}{k}}{\binom{n}{j} \binom{n-1}{k-1}}=\frac{j}{k}$ into account yields the important $\color{blue}{formula}$

$$\small \binom{n}{k} \binom{n-1}{m-k} f(k,m-k)+\binom{n}{m+1-k} \binom{n-1}{k-1} f(m+1-k,k-1)=\frac{\binom{n}{k}\binom{n-1}{m-k} }{-k+m+1}U(k,m+1-k)$$

Now let $y\to\frac{1-x}{2}$ $$I_n=\int_0^{\frac{1}{2}} \frac{\log ^n(2 y) \log ^{n-1}(2 (1-y))}{1-y} \, dy$$ Apply Binomial thm twice, extract $k=0$ $$I_n=\sum _{k=1}^n \sum _{j=0}^{n-1} \binom{n}{k} \binom{n-1}{j} f(k,j) \log ^{2n-j-k-1}(2)+\int_0^{\frac{1}{2}} \frac{\log ^n(2) \log ^{n-1}(2 (1-y))}{1-y} \, dy$$ Take Cauchy product $$I_n=\sum _{m=1}^{2n-1} \sum _{k+j=m}\binom{n}{k} \binom{n-1}{j} f(k,j) \log ^{2n-m-1}(2)+\frac{\log ^{2 n}(2)}{n}$$ Take care of range of $j,k$ $$\scriptsize I_n=\sum _{m=1}^n \sum _{k=1}^m \binom{n}{k} \binom{n-1}{m-k} f(k,m-k) \log ^{2n-m-1}(2)+ \sum _{m=n+1}^{2 n-1} \sum _{k=m-n+1}^n \binom{n}{k} \binom{n-1}{m-k} f(k,m-k) \log ^{2n-m-1}(2)+\frac{\log ^{2 n}(2)}{n}$$ Let $k\to m+1-k$, take average $$\scriptsize I_n=\frac{1}{2} \sum _{m=1}^n \sum _{k=1}^m \left(\binom{n}{k} \binom{n-1}{m-k} f(k,m-k)+\binom{n}{m+1-k} \binom{n-1}{k-1} f(m+1-k,k-1)\right) \log ^{2n-m-1}(2)+\frac{1}{2} \sum _{m=n+1}^{2 n-1} \sum _{k=m-n+1}^n \left(\binom{n}{k} \binom{n-1}{m-k} f(k,m-k)+\binom{n}{m+1-k} \binom{n-1}{k-1} f(m+1-k,k-1)\right) \log ^{2n-m-1}(2)+\frac{\log ^{2 n}(2)}{n}$$ Use the $\color{blue}{formula}$ to simplify $$\scriptsize I_n=\frac{1}{2} \sum _{m=1}^n \sum _{k=1}^m \frac{\binom{n}{k}\binom{n-1}{m-k} \log ^{-m+2 n-1}(2) }{-k+m+1} U(k,m+1-k)+\frac{1}{2} \sum _{m=n+1}^{2 n-1} \sum _{k=m-n+1}^n \frac{\binom{n}{k}\binom{n-1}{m-k} \log ^{-m+2 n-1}(2) }{-k+m+1}U(k,m+1-k)+\frac{\log ^{2 n}(2)}{n}$$ Expand $U(k,m+1-k)$

$$ \scriptsize I_n=\frac{1}{2} \left(\sum _{m=1}^n \sum _{k=1}^m +\sum _{m=n+1}^{2 n-1} \sum _{k=m-n+1}^n\right) \frac{\binom{n}{k}\binom{n-1}{m-k} \log ^{-m+2 n-1}(2) }{-k+m+1}\left(k \underset{a\to 0}{\text{lim}}\underset{b\to 1}{\text{lim}}\frac{\partial ^{m}B(a,b)}{\partial a^{k-1}\, \partial b^{-k+m+1}}+(-1)^m \log ^{m+1}(2)\right)+\frac{\log ^{2 n}(2)}{n}$$

This is the final expression of $I_n$. According to Lemma $2.3$ in OP's article, all Beta derivatives in this expression lie in the algebra $\mathbb{Q}(\pi^2, \zeta(3), \zeta(5), \zeta(7), \cdots)$, whence after adding up $\log(2)$ terms, $I_n$ lies in the extended $\mathbb{Q}(\log(2), \pi^2, \zeta(3), \zeta(5), \zeta(7), \cdots)$. QED.

  • $\begingroup$ Creative (+1). I'm sure this generalization doesnt exist in the literature. Congrats $\endgroup$ – Ali Shadhar Jun 13 '20 at 19:00

Too long for a comment: By using the start I described in comments and then algebraic identities, I could reduce $\mathcal{I_4}$ to

$$\mathcal{I_4}=\log ^8(2)+\frac{31}{420} \log^2(2)\pi^6+4 \log (2) \underbrace{\int_0^1 \frac{\log ^3(1-t) \log ^3(t)}{t} \textrm{d}t}_{\text{Beta function}}+\log ^7(2)\int_0^{1/2} \frac{1}{1-t}\textrm{dt}\\+3 \log ^6(2)\int_0^{1/2}\frac{ \log (1-t)}{1-t}\textrm{d}t+4 \log ^6(2)\int_0^{1/2}\frac{ \log (t)}{1-t}\textrm{d}t+12 \log ^5(2) \int_0^{1/2}\frac{\log (1-t) \log (t)}{1-t}\textrm{d}t+12 \log ^3(2)\underbrace{\int_0^{1/2}\frac{ \log (1-t) \log ^3(t)}{1-t}\textrm{d}t}_{\text{Reducible to K}}+4 \log ^3(2)\int_0^{1/2} \frac{\log ^3(1-t) \log (t)}{1-t}\textrm{d}t\\+3 \log ^5(2)\int_0^{1/2}\frac{ \log ^2(1-t)}{1-t}\textrm{d}t+6 \log ^5(2)\int_0^{1/2}\frac{ \log ^2(t)}{1-t}\textrm{d}t+\frac{3}{5} \log ^2(2)\int_0^{1/2}\frac{ \log ^5(t)}{1-t}\textrm{d}t\\-\frac{3}{5} \log ^2(2)\int_0^{1/2}\frac{ \log ^5(1-t)}{1-t}\textrm{d}t+\log ^4(2) \int_0^{1/2} \frac{\log ^3(1-t)}{1-t}\textrm{d}t+4 \log ^4(2)\int_0^{1/2}\frac{ \log ^3(t)}{1-t}\textrm{d}t\\+\log ^3(2) \int_0^{1/2}\frac{\log ^4(t)}{1-t}\textrm{d}t+\underbrace{\int_0^{1/2}\frac{\log ^3(1-t) \log ^4(t)}{1-t}\textrm{d}t}_{\text{Reducible to $J_3$}}+18 \log ^4(2) \underbrace{\int_0^{1/2}\frac{ \log (1-t) \log ^2(t)}{1-t}\textrm{d}t}_{\textrm{Reducible to $J_1$}}+12 \log ^4(2)\int_0^{1/2}\frac{ \log ^2(1-t) \log (t)}{1-t}\textrm{d}t+3 \log ^2(2)\int_0^{1/2}\frac{ \log ^4(1-t) \log (t)}{1-t}\textrm{d}t\\+18 \log ^3(2)\underbrace{\int_0^{1/2}\frac{ \log ^2(1-t) \log ^2(t)}{1-t}\textrm{d}t}_{\text{Reducible to $K$}}+18 \log ^2(2) \underbrace{\int_0^{1/2}\frac{\log ^2(1-t) \log ^3(t)}{1-t} \textrm{d}t}_ {\text{Reducible to $J_2$}}.$$

I considered the auxiliary results

$$J_n=\int_0^{1/2} \frac{\log^n(1-x)\log^{n+1}(x)}{1-x}\textrm{d}x=-\frac{1}{2(1+n)}\log^{2(n+1)}(2)+\frac{1}{2}\lim_{\substack{x\to0 \\ y \to 1}}\frac{\partial^{2n+1}}{\partial x^n \partial y^{n+1}}\operatorname{B}(x,y)$$ and $$ K=\int_{0}^{1/2} \frac{\log^2(x)\log^2(1-x)}{x}\textrm{d}x$$ $$=\frac{1}{8}\zeta(5)-2\zeta(2)\zeta(3)-\frac{2}{3}\log^3(2)\zeta(2)+\frac{7}{4}\log^2(2)\zeta(3)-\frac{1}{15}\log^5(2) $$ $$+4\log(2)\operatorname{Li}_4\left(\frac{1}{2}\right)+4\operatorname{Li}_5\left(\frac{1}{2}\right),$$ which are both calculated in the book (Almost) Impossible Integrals, Sums, and Series.

A short note: For the generalization the key is to figure out which groups of integrals to take together for transformations after the first step I described in comments, where to further use algebraic identities to get those expected magical cancellations like in the case above. The rest is trivial. Also, I skipped giving references for the trivial integrals above.

  • $\begingroup$ outstanding approach (+1). $\endgroup$ – Ali Shadhar Oct 29 '19 at 19:09
  • $\begingroup$ @AliShather Thank you! $\endgroup$ – user97357329 Oct 29 '19 at 19:11
  • $\begingroup$ A nice solution of $I_4$, I wish your method could be generalized to arbitrary $n$ though. $\endgroup$ – pisco Oct 30 '19 at 3:31
  • $\begingroup$ @pisco Thank you. The way used above should work for every $n$, but to put all in a generalized form requires some tedious work. $\endgroup$ – user97357329 Oct 30 '19 at 6:47
  • 1
    $\begingroup$ This is indeed an inspiring way. Actually using IBP, the calculation of nontrivial part $K$ is not needed. By generalizing your method I proved the general proposition. Much appreciation. $\endgroup$ – Iridescent Jun 2 '20 at 15:12

My approach to $I_3$:

Starting with the algebraic identity $20a^3b^2=(a+b)^5+(a-b)^5-2a^5-10ab^4$ we can write

$$20\int_0^1\frac{\ln^3(1-x)\ln^2(1+x)}{1+x}\ dx\\=\int_0^1\frac{\ln^5(1-x^2)}{1+x}+\int_0^1\frac{\ln^5\left(\frac{1-x}{1+x}\right)}{1+x}-2\int_0^1\frac{\ln^5(1-x)}{1+x}-10\int_0^1\frac{\ln(1-x)\ln^4(1+x)}{1+x}\ dx$$

The first integral can be calculated the same way Cornel did here

$$\int_0^1\frac{\ln^5(1-x^2)}{1+x}dx=\int_0^1(1-x)\frac{\ln^5(1-x^2)}{1-x^2}dx\overset{x^2=y}{=}\frac12\int_0^1\frac{1-\sqrt{y}}{\sqrt{y}}.\frac{\ln^5(1-y)}{1-y}dy$$ $$\overset{IBP}{=}-\frac1{24}\int_0^1\frac{\ln^6(1-y)}{y^{3/2}}dy=-\frac{1}{24}\lim_{x\mapsto-1/2\\y\mapsto1}\frac{\partial^6}{\partial y^6}\text{B}(x,y)$$


The second integral can be simplified via subbing $\frac{1-x}{1+x}=y$:

$$\int_0^1\frac{\ln^5\left(\frac{1-x}{1+x}\right)}{1+x}\ dx=\int_0^1\frac{\ln^5 y}{1+y}\ dy\\=-\sum_{n=1}^\infty (-1)^n\int_0^1 y^{n-1}\ln^5 y\ dy=5!\sum_{n=1}^\infty\frac{(-1)^n}{n^6}=\boxed{-\frac{465}{4}\zeta(6)}$$

and lets set $1-x=y$ for the third integral:

$$\int_0^1\frac{\ln^5(1-x)}{1+x}\ dx=\int_0^1\frac{\ln^5y}{2-y}\ dy\\=\sum_{n=1}^\infty\frac1{2^n}\int_0^1 y^{n-1}\ln^5 y\ dy=-5!\sum_{n=1}^\infty\frac{1}{2^n n^6}=\boxed{-120\operatorname{Li}_6(1/2)}$$

For the last integral, we set $1+x=y$

$$\int_0^1\frac{\ln(1-x)\ln^4(1+x)}{1+x}\ dx=\int_1^2\frac{\ln(2-y)\ln^4y}{y}\ dy\\=\ln2\int_1^2\frac{\ln^4y}{y}\ dx+\int_1^2\frac{\ln(1-y/2)\ln^4y}{y}\ dy\\=\frac15\ln^62-\sum_{n=1}^\infty\frac{1}{n2^n}\int_1^2 y^{n-1}\ln^4y\ dy\\=\frac15\ln^62-\sum_{n=1}^\infty\frac1{n2^n}\left(24\frac{2^n}{n^5}-24\frac{2^n\ln2}{n^4}+12\frac{2^n\ln^22}{n^3}-4\frac{2^n\ln^32}{n^2}+\frac{2^n\ln^42}{n}-\frac{24}{n^5}\right)\\\boxed{=\frac15\ln^62-24\zeta(6)+24\ln2\zeta(5)-12\ln^22\zeta(4)+4\ln^32\zeta(3)-\ln^42\zeta(2)+24\operatorname{Li}_6(1/2)}$$

By combining the boxed results, the closed form of $I_3$ follows:

  • 1
    $\begingroup$ (+1) for the nice approach. $\endgroup$ – user97357329 Oct 30 '19 at 19:12
  • 1
    $\begingroup$ thank you for reminding me how to solve the tricky integral :) $\endgroup$ – Ali Shadhar Oct 30 '19 at 19:18
  • $\begingroup$ @ali shather I noticed an error in $$\int_0^1\frac{\ln^5(1-x^2)}{1+x}\ dx=-\frac{8295}{28}\zeta(6)-40\zeta(2)Log^4(2)+...=-30.382721029123626...$$ The other terms are correct. $\endgroup$ – user178256 Oct 30 '19 at 23:17
  • $\begingroup$ @user178256 thank you.. i forgot the term $-40\ln^42\zeta(2)$.. edited now $\endgroup$ – Ali Shadhar Oct 31 '19 at 1:31

Generalization using the Stirling number

Start with $1-x\to x$



$$=\int_0^1\frac{\ln^n(x)\sum_{k=0}^{n-1}{n-1\choose k}\ln^{n-k-1}(2)\ln^k(1-x/2)}{2-x}dx$$

$$=\frac12\sum_{k=0}^{n-1}{n-1\choose k}\ln^{n-k-1}(2)\int_0^1\frac{\ln^n(x)\ln^k(1-x/2)}{1-x/2}dx$$

$$\overset{IBP}{=}n\sum_{k=0}^{n-1}{n-1\choose k}\ln^{n-k-1}(2)\int_0^1\frac{\ln^{n-1}(x)\ln^{k+1}(1-x/2)}{x}dx$$

Recall Stirling number of first kind

$$\frac{\ln^k(1+x)}{k!}=\sum_{j=k}^\infty(-1)^{j-k} \begin{bmatrix} j \\ k \end{bmatrix}\frac{x^j}{j!}$$

where if we replace $x$ by $-x/2$ and $k$ by $k+1$ we have

$$\frac{\ln^{k+1}(1-x/2)}{(k+1)!}=\sum_{j=k+1}^\infty(-1)^{j-k-1} \begin{bmatrix} j \\ k+1 \end{bmatrix}\frac{x^j}{j!}$$


$$I_n=\sum_{j=k+1}^\infty\sum_{k=0}^{n-1}(-1)^{j-k-1} \begin{bmatrix} j \\ k+1 \end{bmatrix}\frac{n}{j!}{n-1\choose k}\ln^{n-k-1}(2)\int_0^1\ln^{n-1}(x)x^{j-1}dx$$

$$=\sum_{j=k+1}^\infty\sum_{k=0}^{n-1}(-1)^{j-k-1} \begin{bmatrix} j \\ k+1 \end{bmatrix}\frac{n}{j!}{n-1\choose k}\ln^{n-k-1}(2)\left[\frac{(-1)^{n-1}(n-1)!}{j^n}\right]$$

$$=\sum_{j=k+1}^\infty\sum_{k=0}^{n-1}(-1)^{j-k+n} \frac{n!}{j!j^n}\begin{bmatrix} j \\ k+1 \end{bmatrix}{n-1\choose k}\ln^{n-k-1}(2)$$

Not sure if this double sum can be further simplified.


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.