12

I'm searching for a "simple" proof of: \begin{align}\int_0^1 \frac{\arctan^2 x\ln x}{1+x}dx=-\frac{233}{5760}\pi^4-\frac{5}{48}\pi^2\ln ^2 2+\text{Li}_4\left(\frac{1}{2}\right)+\frac{7}{16}\zeta(3)\ln 2+\frac{1}{24}\ln^4 2+\pi \Im\left(\text{Li}_3\left(1+i\right)\right)-\frac{1}{4}\text{G}\pi\ln 2\end{align}

The context: I have written a script for Pari GP to search heuristically for certain close-forms. Testing my script with the above integral i was lucky enough to find something. The script searchs for integer linear relation between the integral and some constants: \begin{align}\pi^4,\pi\ln^3 2,\pi^2\ln^2 2,\pi^3\ln 2,\text{Li}_4\left(\frac{1}{2}\right),\zeta(3)\ln 2,\zeta(3)\pi,\ln^4 2,\pi \Im\left(\text{Li}_3\left(1+i\right)\right), \Im\left(\text{Li}_3\left(1+i\right)\right)\ln 2,\text{G}^2,\text{G}\ln^2 2,\text{G}\pi^2,\text{G}\pi\ln 2\end{align}

NB: Using the algorithm with $\displaystyle\int_0^1 \dfrac{\ln^2(1+x^2)\ln x}{1+x}dx$ gives someting too.

Addendum:

PARI GP script:

beta(n)={intnum(x=0,1,(-log(x))^(n-1)/(1+x^2))}; lindep4(x)={ NAME=["x","Pi^4","Pilog(2)^3","Pi^2log(2)^2","Pi^3log(2)","polylog(4,1/2)","zeta(3)log(2)","zeta(3)Pi","log(2)^4","Piimag(polylog(3,1+I))","log(2)imag(polylog(3,1+I))","Catalan^2","Catalanlog(2)^2","CatalanPi^2","Catalanlog(2)Pi","beta(4)","imag(polylog(4,1+I))"]; VAL=[x,Pi^4,Pilog(2)^3,Pi^2log(2)^2,Pi^3log(2),polylog(4,1/2),zeta(3)log(2),zeta(3)Pi,log(2)^4,Piimag(polylog(3,1+I)),log(2)imag(polylog(3,1+I)),Catalan^2,Catalanlog(2)^2,CatalanPi^2,Catalan*log(2)*Pi,beta(4),imag(polylog(4,1+I))]; L=lindep(VAL); for(i=2,length(L),if(-L[i]/L[1]>0,print1("+",-L[i]/L[1],NAME[i]));if(-L[i]/L[1]<0,print1(-L[i]/L[1],NAME[i]))); }

for example:

\p 100
lindep4(intnum(x=0,1,atan(x)^2*log(x)/(1+x)))

NB: To improve the script add $\beta(4)$ value and some polygamma values.

PS: the PARI GP script has been updated to take into account more integrals.

Try this one: $\displaystyle \int_0^1\frac{\ln^2(1+x^2)\ln x}{1+x^2}dx$

FDP
  • 13,647
  • 1
    Possible approach: Integration by parts (repeated) could get rid of arctan. – herb steinberg Jul 11 '21 at 21:26
  • 2
    Just to remark that $\arctan^3 x$ is "nice" too: $$ \int_0^1 \frac{\arctan^3 x\ln x}{1+x}dx = \frac{21 G \zeta (3)}{64}-\frac{3}{32} \pi ^2 G \log (2)-\frac{3}{8} \pi ^2 \Im\left(\text{Li}_3\left(\frac{1}{2}+\frac{i}{2}\right)\right)-\frac{3}{4} \beta(4) \log (2)-\frac{3}{8} \pi \text{Li}_4\left(\frac{1}{2}\right)-\frac{21}{64} \pi \zeta (3) \log (2)+\frac{391 \pi ^5}{30720}-\frac{1}{64} \pi \log ^4(2)+\frac{7}{256} \pi ^3 \log ^2(2)$$ I call it "nice" because many other similar integrals like $\int_0^1 \frac{\arctan^2 x\ln^2 x}{1+x}dx$ don't have result of comparable simplicity. – pisco Jul 11 '21 at 21:29
  • @Pisco: Probably the idea is a good one but the computation is rather complicated. – FDP Jul 11 '21 at 23:57
  • $\displaystyle\int_0^1 \dfrac{\ln(1+x^2)\ln(1+x)\ln x}{1+x}dx$ is nice as well. – FDP Jul 12 '21 at 00:07
  • @FDP This integral $\int_0^1 \frac{\ln(1+x^2)\ln(1+x)\ln x}{1+x}dx$ is nice because it has weight $4$, here weight of $$\int_0^1 \dfrac{\ln^i(1+x^2)\ln^j(1+x)\ln^k x \log^m (1-x) \arctan^l x}{1+x}dx$$ means the number $i+j+k+l+m+1$. Difficulty on closed-form evaluation increases exponentially with weight. All weight $4$ integrals are nice. While weight $5$ being nice is a very special occurrence. The above example I gave has weight $5$ – pisco Jul 12 '21 at 07:42
  • @Pisco: That is noticeable is the weight can be read on the values in the closed form. $\pi,\ln 2$ weight is $1$, $\text{Li}_n\left(r\right),\beta(n)$, $r$ is a rational, weight is $n$. $\text{G}\pi$ weigth is $2+1=3$. All terms in the sum are homogeneous, that is, they have the same weight. – FDP Jul 12 '21 at 08:47
  • @FDP Yes, good observation. This is indeed a very general phenomenon. For a highly-nontrivial example, see here, note that every term in the result is homogenous of weight $8$. A general weight $8$ integral is more difficult than a weight $7$, which is in term more difficult than weight $6$ etc. – pisco Jul 12 '21 at 08:51
  • One have homogeneity when one develops expression like $(a_1+...+a_k)^n$ where $n$ is our weight. What are $a_1,...a_k?$ – FDP Jul 12 '21 at 08:57
  • @Pisco: is your package for Mathematica can be programmed for Maxima? – FDP Jul 12 '21 at 09:17
  • @FDP Most likely no, because the program is long and complicated. Conversion of such long code between any two programming languages is arduous. – pisco Jul 12 '21 at 09:27
  • @Pisco: have you an article about your algorithm to teach the complete idiot i am? – FDP Jul 12 '21 at 09:53
  • @FDP A brief outline of the algorithm used in my Mathematica package is described in https://arxiv.org/pdf/2007.03957.pdf. A more detailed outline will be published later. – pisco Jul 12 '21 at 13:23

2 Answers2

3

A framework proposed by Cornel (answer to the main integral)

The skeleton of the solution may be immediately obtained by using the strategy given for the generalization in Sect. $1.24$, page $14$, (Almost) Impossible Integrals, Sums, and Series (more precisely, see pages $142-145$). So, following the suggested strategy, we have that

$$\int_0^1 \frac{\arctan^2(x)\log(x)}{1+x}\textrm{d}x$$ $$=\frac{1}{2}\int_0^1 \frac{\arctan^2(x)\log(x)}{x}\textrm{d}x-\frac{1}{2}\int_0^{\infty} \frac{\arctan^2(x)\log(x)}{x(1+x)}\textrm{d}x+\frac{\pi}{2}\int_0^1 \frac{\arctan(x)\log(x)}{1+x}\textrm{d}x$$ $$-\frac{\pi^2}{8}\int_0^1\frac{\log(x)}{1+x}\textrm{d}x.$$

The resulting integrals are either known or manageable.

  • Behind the first resulting integral lies a very difficult harmonic series which is calculated by real methods here https://math.stackexchange.com/q/3803762.

  • The integral $\displaystyle \int_0^{\infty} \frac{\arctan^2(x)\log(x)}{x(1+x)}\textrm{d}x$ is reducible to the following integrals, $\displaystyle \int_0^1 \frac{x\log(1-x)}{1+x^2}\textrm{d}x$, $\displaystyle \int_0^1 \frac{x\log(1+x)}{1+x^2}\textrm{d}x$, $\displaystyle \int_0^1 \frac{x\log^2(x)\log(1-x)}{1+x^2}\textrm{d}x$, $\displaystyle \int_0^1 \frac{x\log^2(x)\log(1+x)}{1+x^2}\textrm{d}x$, $\displaystyle \int_0^1 \frac{\log(x)\log(1-x)}{1+x^2}\textrm{d}x$,$\displaystyle \int_0^1 \frac{\log(x)\log(1+x)}{1+x^2}\textrm{d}x$, where most of them are known or easily reducible to known harmonic series like the ones given at the previous link (see the third resulting integral). Here one also needs to calculate $\displaystyle \sum_{n=1}^{\infty}(-1)^{n-1}\frac{H_{2n}^{(3)}}{n}$ which is derived in a generalized form in the second theorem of the paper https://www.researchgate.net/publication/333339284_A_simple_strategy_of_calculating_two_alternating_harmonic_series_generalizations.

End of story

user97357329
  • 5,319
  • $\displaystyle \int_0^{\infty} \frac{\arctan^2(x)\log(x)}{x(1+x)}\textrm{d}x$ is probably the harder to compute among the four integrals. – FDP Jul 12 '21 at 13:40
  • @FDP Do you mind if I ask Cornel to include this integral in the sequel of (Almost) Impossible Integrals, Sums, and Series? (I bet he will do it) This solution will be presented in detail with all steps that were omitted here. – user97357329 Jul 12 '21 at 14:31
  • @ user97357329: if you want, i'm not the owner of integrals, integrals own me. It's not that hard to get some nice integrals (i have provided a way for weight 4) but it's harder to give proofs. – FDP Jul 12 '21 at 14:41
  • @FDP It is a nice integral and given some magical transformations on the way, I think it would be a cool entry. The integral you are worried about reduces exactly to the ones mentioned in the edited post. So, we are done. – user97357329 Jul 12 '21 at 14:45
2

Another framework proposed by Cornel (answer to the second integral, $\displaystyle \int_0^1\frac{\ln^2(1+x^2)\ln x}{1+x^2}\textrm{d}x$)

Observe that $$\int_0^1 \frac{1}{1+x^2}\log^3\left(\frac{2x}{1+x^2}\right)\textrm{d}x$$ $$=\log^3(2)\int_0^1\frac{1}{1+x^2}\textrm{d}x+3\log^2(2)\int_0^1\frac{\log(x)}{1+x^2}\textrm{d}x+3\log(2)\int_0^1\frac{\log^2(x)}{1+x^2}\textrm{d}x$$ $$+\int_0^1\frac{\log^3(x)}{1+x^2}\textrm{d}x-3\log^2(2)\int_0^1\frac{\log(1+x^2)}{1+x^2}\textrm{d}x-6\log(2)\int_0^1\frac{\log(x)\log(1+x^2)}{1+x^2}\textrm{d}x$$ $$-3\int_0^1\frac{\log^2(x)\log(1+x^2)}{1+x^2}\textrm{d}x+3\log(2)\int_0^1\frac{\log^2(1+x^2)}{1+x^2}\textrm{d}x$$$$-\int_0^1\frac{\log^3(1+x^2)}{1+x^2}\textrm{d}x+3\color{blue}{\int_0^1\frac{\log(x)\log^2(1+x^2)}{1+x^2}\textrm{d}x}.$$

  • Note that the integral in the left-hand side may be beautifully reduced by the variable change $\displaystyle x\mapsto \frac{2x}{1+x^2}$ to $\displaystyle \int_0^1 \frac{1}{1+x^2}\log^3\left(\frac{2x}{1+x^2}\right)\textrm{d}x=\frac{1}{2}\int_0^1 \frac{\log^3(x)}{\sqrt{1-x^2}}\textrm{d}x$, where the last integral is a form involving the derivative of the Beta function.

  • Note that all the other resulting integrals in the right-hand side are already known.

  • To easily make the connection with the known integrals, for the integrals $\displaystyle \int_0^1\frac{\log^2(1+x^2)}{1+x^2}\textrm{d}x$ and $\displaystyle \int_0^1\frac{\log^3(1+x^2)}{1+x^2}\textrm{d}x$ make the variable change $x \mapsto \tan(x)$ to have a view in terms of trigonometric functions. A relevant link: About the integral $\int_{0}^{\pi/4}\log^4(\cos\theta)\,d\theta$

  • Also, you might like to know the following generalized integral, $\displaystyle \int_0^1\frac{\log^{2n}(x)\log(1+x^2)}{1+x^2}\textrm{d}x$, is nicely presented and calculated by Ali Shadhar in his book, An Introduction To The Harmonic Series And Logarithmic Integrals: For High School Students Up To Researchers (see page $149$). The integral easily and naturally reduces to forms involving derivatives of the Beta function.

End of story

user97357329
  • 5,319