6

After reading the reduction formula in the post, I am curious about the closed form of the integral $$ I_n=\int_0^1 \frac{x^n}{6+x-x^2} d x $$ I first resolve the integrand into two partial fractions as $$ I_n=\frac{1}{5} \left[\underbrace{\int_0^1 \frac{x^n}{3-x} d x}_{J_n} +\underbrace{\int_0^1 \frac{x^n}{2+x} d x}_{K_n} \right] $$ Then I tackle the two integrals in general, $$ \begin{aligned} \int_0^1 \frac{x^n}{a-x} d x & =\int_{a-1}^a \frac{(a-x)^n}{x} d x \\ \\ & =\int_{a-1}^a\left(\frac{a^n}{x}-\sum_{k=1}^n\left(\begin{array}{c} n \\ k \end{array}\right) a^{n-k} x^{k-1}\right) dx \\ & =a^n \ln \left(\frac{a}{a-1}\right)-\sum_{k=1}^n\left(\begin{array}{l} n \\ k \end{array}\right) \frac{a^{n}-a^{n-k}(a-1)^k}{k} \end{aligned} $$ Hence $$J_n =3^n \ln \left(\frac{3}{2}\right)-\sum_{k=1}^n\left(\begin{array}{l} n \\ k \end{array}\right) \frac{3^{n}-3^{n-k}2^k}{k} $$ and $$K_n= -\left[(-2)^n \ln \left(\frac{2}{3}\right)-\sum_{k=1}^n\left(\begin{array}{l} n \\ k \end{array}\right) \frac{(-2)^{n}-(-1)^n2^{n-k}3^k}{k} \right]$$ We can now conclude that the closed form of the integral is $$ \begin{aligned}I_n= &\ \frac{1}{5} \left[ 3^n\left[\ln \left(\frac{3}{2}\right)-\sum_{k=1}^n\left(\begin{array}{l} n \\ k \end{array}\right) \frac{1-\left(\frac{2}{3}\right)^k}{k}\right]+ (-2)^{n}\left[\ln \left(\frac{3}{2}\right)-\sum_{k=1}^n\left(\begin{array}{l} n \\ k \end{array}\right) \frac{1-\left(\frac{3}{2}\right)^k}{k}\right]\right]\\=& \frac{3^n+(-2)^n}{5} \ln \left(\frac{3}{2}\right) -\frac{1}{5}\left[\sum_{k=1}^n\left(\begin{array}{l} n \\ k \end{array}\right) \frac{1}{k}\left(3^n\left(1-\left(\frac{2}{3}\right)^k\right)+(-2)^n\left(1-\left(\frac{3}{2}\right)^k\right)\right]\right. \end{aligned} $$


My question:

Is there any alternative method? Your comments and alternative methods are highly appreciated.

Lai
  • 31,615
  • Mathematica finds an answer with hypergeometric functions

    $$\frac{3 , _2F_1\left(1,n+1;n+2;-\frac{1}{2}\right)+2 , _2F_1\left(1,n+1;n+2;\frac{1}{3}\right)}{30 (n+1)}$$

    – userrandrand Dec 09 '23 at 05:00
  • Denoting $\Phi$ the Lerch transcendent https://en.wikipedia.org/wiki/Lerch_zeta_function, calculating first the generating function and then calculating the series coefficients of the result, Mathematica finds

    $$\begin{cases} \frac{1}{5} \left(\Phi \left(\frac{1}{3},1,n\right)-\Phi \left(-\frac{1}{2},1,n\right)\right) & n>2 \ \frac{1}{5} \log \left(\frac{3}{2}\right) & n=1 \ \frac{2}{5} \log \left(\frac{3}{2}\right) & n=0 \ \log \left(\frac{9}{4} \left(\frac{3}{2}\right)^{3/5}\right)-1 & n=2 \end{cases}$$

    – userrandrand Dec 09 '23 at 05:22
  • I should note that I did not verify the hypothesis of switching integral and sum in the calculation mentioned before but using your fraction decomposition, it seems (I did not check the details) that you can find that result using the integral representation of the lerch transcendent given here https://functions.wolfram.com/ZetaFunctionsandPolylogarithms/LerchPhi/07/01/01/01/0003/ – userrandrand Dec 09 '23 at 05:39
  • $\newcommand{\d}{,\mathrm{d}}$We have: $$I_n=\frac{1}{6}\sum_{m\ge0}\frac{(-1)^m}{6^m}\int_0^1x^{n+m}(1-x)^m\d x=\frac{1}{6}\sum_{m\ge0}\frac{(-1)^m}{6^m}\frac{(n+m)!m!}{(n+2m+1)!}$$ – FShrike Dec 09 '23 at 14:08
  • Fast and elegant! – Lai Dec 10 '23 at 05:10

3 Answers3

3

Here is a method to arrive at a simpler closed-form result. Note that

\begin{align} J_n(a)=&\int_0^1 \frac{x^n}{x+a}dx = \frac1n - a J_{n-1}(a)\\ = &\ (-a)^n\ln\frac{a+1}a+\sum_{k=0}^{n-1}\frac{(-a)^k}{n-k} \end{align} Then \begin{align} &\int_0^1 \frac{x^n}{6+x-x^2} d x\\ =&\ \frac{1}{5} \int_0^1 \frac{x^n}{2+x}-\frac{x^n}{x-3}\ d x = \frac15[J_n(2)-J_n(-3)] \\ =&\ \frac{3^n+(-2)^n}5\ln\frac32 +\frac15 \sum_{k=0}^{n-1}\frac{(-2)^k-3^k}{n-k} \end{align}

Quanto
  • 120,125
0

Define the function $I : [1,\infty) \to \mathbb{R}_{\ge 0}$ by $I(t) = \int_{0}^{1} \frac{x^t}{6+x-x^2} dx$. I am aware that you are searching for a closed form of this integral, I think the best we have currently is @userrandrand 's answer in the comments and your solution. I present a numerical approximation which is close to the true solution and gets better with larger $t$. First observe that $0 \le x(1-x) \le \frac{1}{4}$ for $x\in [0,1]$, therefore $$\frac{4}{25(t+1)} = \int_{0}^{1} \frac{x^t}{6+\frac{1}{4}} dx \le I(t) \le \frac{1}{6}\int_{0}^{1} x^t dx = \frac{1}{6(t+1)}.$$ We can do even better by realizing that the contribution in the integral is only from a neighborhood near $1$, as $x^t$ is near zero away from $1$. This implies that we can observe the leading behaviour of the integral by taylor expanding the rest of the integrand at $x=1$. Taylor expanding $1/(6+x-x^2)$ near $x = 1$, we have $$\frac{1}{6+x-x^2} = \frac{1}{6}+\frac{1}{36}(x-1)+\frac{7}{216}\left(x-1\right)^{2} + O(x-1)^3.$$ With this, the integral can be approximated as following: \begin{align*} I(t) &\approx \int_{0}^{1} \left(\frac{1}{6}+\frac{1}{36}(x-1)+\frac{7}{216}\left(x-1\right)^{2}\right)x^t dx. \\ &= \frac{1}{216}\left(\frac{37}{t+1}-\frac{8}{t+2}+\frac{7}{t+3}\right). \end{align*} For $t=1$, you already have two digits of accuracy, which rapidly improves with a larger value of $t$. You can take more terms in the Taylor expansion for better accuracy, but the approximation becomes more convoluted.

Sam
  • 3,763
0

Thank @userrandrand for introducing the LerchPhi function $\Phi(z, s, a) $with an integral representation as below: $$ \Phi(z, s, a)=\frac{1}{\Gamma(s)} \int_0^1 \frac{[-\ln (t)]^{s-1} t^{a-1}}{1-z t} d t, $$ we have $$ I_n=\frac{1}{5}\left[\frac{1}{3} \int_0^1 \frac{x^n}{1-\frac{1}{3} x} d x+\frac{1}{2} \int_0^1 \frac{x^n}{1-\left(-\frac{1}{2}\right) x} d x\right]= \frac{1}{30}\left[2 \Phi\left(\frac{1}{3}, 1, n+1\right)+3 \Phi\left(-\frac{1}{2}, 1, n+1\right)\right] $$

Lai
  • 31,615