9

Using the derivative of beta function, find

$$I=\int_0^1\frac{\ln^2x\ln^2(1-x^2)}{1-x^2}\ dx$$

setting $x^2=y$ gives

$$I=\frac18\int_0^1\frac{\ln^2y\ln^2(1-y)}{\sqrt{y}(1-y)}\ dy=\frac18\left.\frac{\partial^4}{\partial a^2\partial b^2}\text{B}(a,b)\right|_{a\mapsto 1/2\\b\mapsto0^{+}}$$

Any good software that can find the 4th derivative and also gives the final result? Wolfram fails to calculate it (or maybe I do not know how to use it well) and when I tried to do it manually, some terms involve $\psi(b)$ and if we take the limit, then $\psi(0)$ is undefined and even if I take the limit of $\psi(b)$ together with other terms, still undefined. I do not know how to avoid this problem as I am not experienced with the beta function.

Thank you.


Note: Solution should be done without using harmonic series.

Ali Shadhar
  • 25,498
  • 2
    Mathematica returns this as $$I=-\frac{7 \pi ^2 \zeta (3)}{3}+31 \zeta (5)+\frac{7}{4} \zeta (3) \log ^2(4)-\frac{1}{8} \pi ^4 \log (2).$$ – YiFan Tey Sep 29 '19 at 23:05
  • This might be helpful, look at $(1)$ and $(2)$. – Zacky Sep 30 '19 at 00:54
  • @YiFan thank you thats very helpful. – Ali Shadhar Sep 30 '19 at 01:22
  • @カカロット Thank you its an interesting solution. – Ali Shadhar Sep 30 '19 at 01:22
  • @YiFan would you mind trying $\lim_{b\mapsto 0^{+}}\frac{\partial^2}{\partial b^2}\text{B}(a,b)$ please? If mathematica gives a closed form, then the rest will be pretty easy.. – Ali Shadhar Sep 30 '19 at 01:56
  • @AliShather If my code is correct, Mathematica returns $+\infty$. – YiFan Tey Sep 30 '19 at 02:08
  • 1
    @YiFan thank you so much.. that means even mathematica can not do it. has be done manually just like M.N.C.E did here https://math.stackexchange.com/questions/1096557/calculate-int-01-frac-log21x-logx-log1-x1-xdx/1949150#1949150 – Ali Shadhar Sep 30 '19 at 02:23
  • 2
    Gamma and polygamma functions are the most developed special functions in Mathematica, you can literally get anything you want (e.g. limit, derivative, simplifications, evaluation at special arguments).

    If one cannot find the limit as simple as that posed in the question, it's the user's mathematical manipulation needs improvement rather than the CAS. Wolfram|Alpha is, of course, less powerful than Mathematica.

    – pisco Sep 30 '19 at 07:59
  • @pisco thank you for all these commands... very helpful. – Ali Shadhar Sep 30 '19 at 14:23

3 Answers3

5

All the following Mathematica commands calculate your limit, in decreasing order of time (the more naive one uses more time):

Limit[D[Gamma[a]*Gamma[b]/Gamma[a + b], {a, 2}, {b, 2}] /. {a -> 
  1/2 + x, b -> x}, x -> 0] // FunctionExpand // Expand

the above command directly calculates the limit, by choosing a path approaching $(a,b)=(1/2,0)$. It takes $32$ seconds on my machine.

D[Normal[Series[
        Gamma[a]*Gamma[b]/Gamma[a + b], {a, 1/2, 4}, {b, 0, 4}]], {a, 
       2}, {b, 2}] /. a -> 1/2 /. b -> 0 // FullSimplify // 
  Expand 

rather than calculating the limit, this one uses series expansion up to constant term. It takes $12$ seconds.

Normal[Series[
     D[Gamma[a]*Gamma[b]/Gamma[a + b], {a, 2}, {b, 2}], {a, 1/2, 
      0}, {b, 0, 0}]] // FullSimplify // Expand

this one does not even calculate derivatives, instead uses series expansion up to 4th order. It takes $3.5$ seconds.

D[Exp[Series[
        LogGamma[a] + LogGamma[b] - LogGamma[a + b], {a, 1/2, 4}, {b, 
         0, 4}]], {a, 2}, {b, 2}] /. a -> 1/2 /. b -> 0 // 
   FullSimplify // Expand

This use the well-known simple series of log gamma function, it takes only $0.5$ seconds.


It's easy to guess why the fourth one is most efficient. To see how much is used for each computation, execute ClearSystemCache[];(your command)//Timing.

Such beta limit arising from logarithm integrals is well-known, it's also not difficult to write down an recursion for it.

pisco
  • 18,983
  • Thank you so much. I tried all these commands I know how to calculate limits on mathematica now I really need it. To be honest, the first one is my fav as its easy to remember. Thanks again @pisco . – Ali Shadhar Sep 30 '19 at 14:21
2

A magnificent solution by Cornel without using the derivative of beta function.


We have the identity

$$\frac{\ln^2(1-x)}{1-x}=\sum_{n=1}^\infty (H_n^2-H_n^{(2)})x^n$$

replace $x$ with $x^2$, then multiply both sides by $\ln^2x$ and integrate from $x=0$ to $1$ we get

$$\int_0^1\frac{\ln^2x\ln^2(1-x^2)}{1-x^2}\ dx=2\sum_{n=1}^\infty\frac{H_n^2-H_n^{(2)}}{(2n+1)^3}\tag1$$


By the master theorem, we have

$$3n\sum_{k=1}^\infty \frac{H_k^2-H_k^{(2)}}{(k+1)(k+n+1)}=H_n^3+3H_nH_n^{(2)}+2H_n^{(3)}$$

differentiate both sides with respect to $n$ to get

$$3\sum_{k=1}^\infty \frac{H_k^2-H_k^{(2)}}{(k+n+1)^2}=\frac{d}{dn}\left(H_n^3+3H_nH_n^{(2)}+2H_n^{(3)}\right)$$

differentiate again and let $n\mapsto -1/2$ we get

$$-6\sum_{k=1}^\infty \frac{H_k^2-H_k^{(2)}}{(k+1/2)^3}=\frac{d^2}{dn^2}\left(H_n^3+3H_nH_n^{(2)}+2H_n^{(3)}\right)_{n\mapsto-1/2}$$

or

$$\sum_{k=1}^\infty \frac{H_k^2-H_k^{(2)}}{(2k+1)^3}=\frac{31}{2}\zeta(5)-\frac{45}{8}\ln2\zeta(4)+\frac72\ln^22\zeta(3)-7\zeta(2)\zeta(3)\tag2$$


From (1) and (2) we get

$$\int_0^1\frac{\ln^2x\ln^2(1-x^2)}{1-x^2}\ dx=31\zeta(5)-\frac{45}{4}\ln2\zeta(4)+7\ln^22\zeta(3)-14\zeta(2)\zeta(3)$$

Ali Shadhar
  • 25,498
0

$\newcommand{\bbx}[1]{\,\bbox[15px,border:1px groove navy]{\displaystyle{#1}}\,} \newcommand{\braces}[1]{\left\lbrace\,{#1}\,\right\rbrace} \newcommand{\bracks}[1]{\left\lbrack\,{#1}\,\right\rbrack} \newcommand{\dd}{\mathrm{d}} \newcommand{\ds}[1]{\displaystyle{#1}} \newcommand{\expo}[1]{\,\mathrm{e}^{#1}\,} \newcommand{\ic}{\mathrm{i}} \newcommand{\mc}[1]{\mathcal{#1}} \newcommand{\mrm}[1]{\mathrm{#1}} \newcommand{\pars}[1]{\left(\,{#1}\,\right)} \newcommand{\partiald}[3][]{\frac{\partial^{#1} #2}{\partial #3^{#1}}} \newcommand{\root}[2][]{\,\sqrt[#1]{\,{#2}\,}\,} \newcommand{\totald}[3][]{\frac{\mathrm{d}^{#1} #2}{\mathrm{d} #3^{#1}}} \newcommand{\verts}[1]{\left\vert\,{#1}\,\right\vert}$ \begin{align} I & \equiv {1 \over 8}\int_{0}^{1}{\ln^{2}\pars{y}\ln^{2}\pars{1 - y} \over \root{y}\pars{1 - y}}\,\dd y \\[5mm] & = \left.{1 \over 8}{\partial^{2} \over \partial \mu^{2}\,\partial \nu^{2}}\int_{0}^{1} \pars{y^{\,\mu - 1/2} - 1}\pars{1 - y}^{\,\nu - 1} \,\dd y\, \right\vert_{\ {\Large\mu\ =\ 0} \atop {\Large\nu\ =\ 0}} \\[5mm] & = \left.{1 \over 8}{\partial^{4} \over \partial \mu^{2}\,\partial \nu^{2}}\bracks{{\Gamma\pars{\mu +1/2} \Gamma\pars{\nu} \over \Gamma\pars{\mu + \nu + 1/2}} - {1 \over \nu}}\, \right\vert_{\ {\Large\mu\ =\ 0} \atop {\Large\nu\ =\ 0}} \\[5mm] & = \left.{1 \over 8}{\partial^{4} \over \partial \mu^{2}\,\partial \nu^{2}}\braces{{1 \over \nu}\bracks{{\Gamma\pars{\mu +1/2} \Gamma\pars{\nu + 1} \over \Gamma\pars{\mu + \nu + 1/2}} - 1}}\, \right\vert_{\ {\Large\mu\ =\ 0} \atop {\Large\nu\ =\ 0}} \\[5mm] & = \left.{1 \over 24}{\partial^{5} \over \partial \mu^{2}\,\partial \nu^{3}}{\Gamma\pars{\mu +1/2} \Gamma\pars{\nu + 1} \over \Gamma\pars{\mu + \nu + 1/2}}\, \right\vert_{\ {\Large\mu\ =\ 0} \atop {\Large\nu\ =\ 0}} \\[5mm] & = \bbx{-\,{7 \pi ^2 \over 3}\,\zeta\pars{3} + 31\zeta\pars{5} + 7\zeta\pars{3}\ln^{2}\pars{2} - {1 \over 8}\,\pi^{4}\ln\pars{2}} \\[5mm] & \approx 0.0654 \end{align}

Felix Marin
  • 89,464