1
$\begingroup$

I want to calculate the following integral: $$I=\int_{1/\phi^2}^{1}\frac{\log^2x}{1-x}\ \log^2\left(\frac{1}{x}\left(\frac{1-x}{1+x}\right)^2\right)\ \mathrm{d}x$$ where $\phi=(\sqrt{5}+1)/2$ is the golden ratio, so $1/\phi^2 = (3-\sqrt 5)/2$.

Can Mathematica give a closed form answer for $I$ (even if it is in terms of some special functions like the Riemann zeta function)?

The following code returns unevaluated:

Integrate[
 Log[x]^2/(1 - x) Log[1/x ((1 - x)/(1 + x))^2]^2, {x, 1/GoldenRatio^2, 1}]
$\endgroup$
0

1 Answer 1

1
$\begingroup$

This is as close as I've been able to get:

First, I rewrite your integrand a little:

$\frac {\mathrm {log}[x]^2} {(1 - x)} \mathrm {log}[\frac {1} {x} (\frac {1 - x} {1 + x})^2]^2 $ = $\frac {\mathrm {log}[x]^2} {(1 - x)} (-\mathrm {log}[x] + 2 (\mathrm {log}[1 - x] - \mathrm {log}[1 + x]))^2 $

I input this as:

part1=Log[x]^2/(1 - x);
part2 = (-Log[x] + 2 (Log[1 - x] - Log[1 + x]))^2;
integrand= part1*part2;

Then define this alternating function (it will come in handy in a second):

altFun[n_] = 1/2 (1 - (-1)^n (-1 + Log[x]) + Log[x]);

observe the coefficients of the expansion of part2 at $x=0$ follow a kind of regular pattern when divided by altFun:

expansionPart2 = (Table[{n, 
    SeriesCoefficient[(-Log[x] + 2 (Log[1 - x] - Log[1 + x]))^2, {x, 
       0, n}]/altFun[n]}, {n, 0, 30}])

(*{{0, Log[x]^2}, {1, 8}, {2, 16}, {3, 8/3}, {4, 32/3}, {5, 8/5}, {6, 
  368/45}, {7, 8/7}, {8, 704/105}, ... *)

The odd n terms are just $\frac{8}{n}$. We can find the even terms by using FindSequenceFunction:

odds[n_] = 8/n;

evens[n_] = FindSequenceFunction[no[[3 ;; All ;; 2]], n]

(*-((16 (-2 + PolyGamma[0, 3/2] - PolyGamma[0, 1/2 + n/2]))/n)*)

and the $n=0$ term is $\mathrm{log[x]^2}$.

we can now create a function that generates the $n$th expansion term at $x=0$ for $part2$:

seqFun[n_] = 
 FullSimplify[
  PiecewiseExpand[
   x^n*altFun[n]*
    Piecewise[{{odds[n], Mod[n, 2] == 1}, {evens[n], 
       Mod[n, 2] == 0 && n > 0}, {Log[x]^2, n == 0}}]]]

This results in

$\begin{equation} \mathrm{seqFun[n]} = \left\{ \begin{array}{lr} 8 x^n \frac{\mathrm{log[x]}}{n}, & \text{odd n } \\ (16 x^n (\mathrm{HarmonicNumber}[1/2 (-1 + n)] + \mathrm{Log[4]}))/n, & \text{even n, n>0 } \\ (\mathrm{Log[x]}^2 & \text{n = 0 } \end{array} \right\} \end{equation}$

we now can multiply seqFun by part1 and get integral for each term:

seqInt[n_] = 
 PiecewiseExpand[part1*seqFun[n]];
indef = Integrate[serInt[n], x];
x0=1/GoldenRatio^2;
defInt = 
 FullSimplify[PiecewiseExpand[(indef /. x -> 1) - (indef /. x -> x0)],
   n >= 1]

This yields a very complicated expression that doesn't format well here: $\begin{array}{cc} \{ & \begin{array}{cc} \frac{32 \left(\sqrt{5}+1\right)^{-2 (n+1)} \left(H_{\frac{n-1}{2}}+\log (4)\right) \left(-4^{n+1} \left(\Phi \left(\frac{1}{\phi ^2},3,n+1\right)+2 \text{csch}^{-1}(2) \Phi \left(\frac{1}{\phi ^2},2,n+1\right)\right)-\left(\left(\sqrt{5}+3\right) \left(\sqrt{5}+1\right)^{2 n} \left(4 \text{csch}^{-1}(2)^2 B_{\frac{1}{\phi ^2}}(n+1,0)+\psi ^{(2)}(n+1)\right)\right)\right)}{n} & (n \bmod 2)=0 \\ \frac{8 \left(3\ 2^{2 n+3} \left(\sqrt{5}+1\right)^{-2 (n+1)} \left(\Phi \left(\frac{2}{\sqrt{5}+3},4,n+1\right)+2 \text{csch}^{-1}(2) \left(\Phi \left(\frac{2}{\sqrt{5}+3},3,n+1\right)+\text{csch}^{-1}(2) \Phi \left(\frac{2}{\sqrt{5}+3},2,n+1\right)\right)\right)+8 \text{csch}^{-1}(2)^3 B_{\frac{1}{\phi ^2}}(n+1,0)-\psi ^{(3)}(n+1)\right)}{n} & (n \bmod 2)=1 \\ \end{array} \\ \end{array}$

This is a screenshot of the output which looks a little easier to read: enter image description here You may notice that I ran FullSimplify on defInt with the assumption $n\geq1$. This is because indef/.{x->1,n->0} results in Indeterminate, but we can evaluate the $n=0$ case separately and take the limit as $x->1$:

int0 = Integrate[seqInt[0], x];

p0 = Limit[int0, x -> 1] - (int0 /. x -> x0)
(* 16 Log[1 - 1/GoldenRatio^2] Log[GoldenRatio]^4 - 
 32 Log[GoldenRatio]^3 PolyLog[2, 1/GoldenRatio^2] - 
 48 Log[GoldenRatio]^2 PolyLog[3, 1/GoldenRatio^2] - 
 48 Log[GoldenRatio] PolyLog[4, 1/GoldenRatio^2] - 
 24 PolyLog[5, 1/GoldenRatio^2] + 24 Zeta[5] *)

So now we have calculated the definite integral defInt for each term, and we just need to sum from $n=1,...,\infty$ to get the result:

result = p0 + Sum[defInt, {n, 1, Infinity}]

The infinite sum unfortunately does not evaluate for me. But we can plot the partial sum and see it approaches the value of NIntegrate:

partSum := p0 + Sum[defInt, {n, 1, nMax}];
nResult = NIntegrate[integrand, {x, x0, 1}];
DiscretePlot[{nResult, partSum}, {nMax, 1, 300, 10}, Joined -> True, 
 Filling -> None]

enter image description here

I am not sure how feasible it is to find a closed form to DiscreteLimit[partSum, nMax -> Infinity], but this is at least much closer to a closed form.

$\endgroup$
5
  • 1
    $\begingroup$ Thank you for your answer. +1. I need a simplified answer. Please simplify your answer. I will with utmost respect accept your answer $\endgroup$
    – Max
    Jun 27 at 2:19
  • 1
    $\begingroup$ I will update answer if I make anymore progress. I am also looking at if I can use the series coefficients from AsymptoticIntegrate and plug them into FindGeneratingFunction to get a closed form for the indefinite integral, but I have not had luck with this yet. $\endgroup$
    – ydd
    Jun 27 at 2:21
  • $\begingroup$ Okay. Thank you so much. I will wait for your updated answer. Please comment after updating the answer $\endgroup$
    – Max
    Jun 27 at 2:38
  • $\begingroup$ I haven't been able to make any more progress, and I don't think I will be able to find a closed form for the infinite sum. Have you considered posting to the mathematics SE? There are some really talented integrators over there $\endgroup$
    – ydd
    Jun 27 at 16:21
  • $\begingroup$ No problem. Thanks for your effort. I will try it in MSE. Thanks. $\endgroup$
    – Max
    Jun 27 at 16:24

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.