4
$\begingroup$

I'd like to get an asymptotic expansion of $\int_0^1 \frac{x^n}{1+x} \, dx$ at order two in $\frac{1}{n}$.

I'm able to prove that $$\lim\limits_{n \to \infty} n \int_0^1 \frac{x^n}{1+x} \, dx = \frac{1}{2}$$ which provides an asymptotic expansion at order $1$. How can I go one step further? Even better, is there a way to get an asymptotic expansion at any order $m$?

$\endgroup$
  • $\begingroup$ For maximum $M_n=\max(U_1,\dots,U_n)$ of $n$ standard uniforms, $\mu_n=E(M_n)=\frac n{n+1}$ and $\sigma_n^2=var(M_n)=\frac n{(n+1)^2(n+2)}$ rewrite $$I_{n}=\frac 1 {n+1}E\left(\frac 1 {1+M_{n+1}}\right)\approx \frac 1 {n+1}\left(\frac 1{1+\mu_{n+1}}+\frac {\sigma_{n+1}^2}{4(1+\mu_{n+1})^3}\right)$$ and expand. $\endgroup$ – A.S. Dec 26 '15 at 19:41
  • $\begingroup$ Another method: Substitute $x = e^{-t}$ and apply Watson's lemma. $\endgroup$ – Antonio Vargas Dec 26 '15 at 23:56
5
$\begingroup$

Method 1. An elementary approach. You may just integrate by parts twice, $$ \begin{align} I_n&=\int_0^1\frac{x^n}{1+x}\:dx \\&=\left. \frac{x^{n+1}}{(n+1)}\frac{1}{1+x}\right|_0^1+\frac{1}{(n+1)}\int_0^1\frac{x^{n+1}}{(x+1)^2}\:dx\\ &=\frac1{2(n+1)}+\frac{1}{n+1}\int_0^1\frac{x^{n+1}}{(x+1)^2}\:dx\\ &=\frac1{2(n+1)}+\frac{1}{n+1}\left(\left. \frac{x^{n+2}}{(n+2)}\frac{1}{(1+x)^2}\right|_0^1+\frac{2}{(n+1)}\int_0^1\frac{x^{n+1}}{(x+1)^3}\:dx \right)\\ &=\frac1{2(n+1)}+\frac1{4(n+1)(n+2)}+\frac{2}{(n+1)^2}\int_0^1\frac{x^{n+1}}{(x+1)^3}\:dx \end{align} $$ but $$ 0\leq \int_0^1\frac{x^{n+1}}{(x+1)^3}\:dx\leq\int_0^1x^{n+1}dx=\frac{1}{(n+2)} $$ thus $$ \frac{2}{(n+1)^2}\int_0^1\frac{x^{n+1}}{(x+1)^3}\:dx=\mathcal{O}\left(\frac{1}{n^3} \right) $$ Finally, as $n \to \infty$,

$$ I_n=\int_0^1\frac{x^n}{1+x}dx=\frac1{2n}-\frac{1}{4n^2}+\mathcal{O}\left(\frac{1}{n^3} \right).$$

$$ $$ Method 2. One may use the standard integral representation of the digamma function and its asymptotics, as $n \to \infty$,

$$ \begin{align} I_n=\int_0^1\frac{x^n}{1+x}\:dx&=\int_0^1\frac{x^n-x^{n+1}}{1-x^2}\:dx\\ &=\frac12\int_0^1\frac{(1-t^{n/2})-(1-t^{(n-1)/2})}{1-t}\:dt\\ &=\frac12\psi\left(\frac{n}{2}+1\right)-\frac12\psi\left(\frac{n}{2}+\frac12\right)\\ &=\frac1{2n}-\frac{1}{4n^2}+\mathcal{O}\left(\frac{1}{n^3} \right). \end{align} $$

$\endgroup$
  • $\begingroup$ Yes, I know. I was only suggesting that you consider adding an explanation to enhance your answer. Happy Holidays! - Mark $\endgroup$ – Mark Viola Dec 26 '15 at 19:49
  • $\begingroup$ Thanks Mark! In fact, one can see that the remainder $r_n$ is such that $0<r_n<\frac2{(n+1)^2(n+2)}$, which is sufficient to qualify it as $\displaystyle \mathcal{O}\left(\frac{1}{n^3} \right)$. $\endgroup$ – Olivier Oloa Dec 26 '15 at 19:57
  • 1
    $\begingroup$ You're welcome! Olivier I consider you one of the "good guys" here on MSE. I was suggesting that some readers might really benefit by your adding this explanation. In any case +1 for the solid post! $\endgroup$ – Mark Viola Dec 26 '15 at 20:12
  • $\begingroup$ @Dr.MV Let me return the compliment by saying that I appreciate many of your interventions here. $\endgroup$ – Olivier Oloa Dec 26 '15 at 21:06
6
$\begingroup$

First make the change of variables $x=e^{-s/n}$, so that $$I(n)=n\int_0^1\frac{x^n dx}{1+x}=\int_0^{\infty}\frac{e^{-s}ds}{e^{s/n}+1}.$$ Now it suffices to Taylor expand in $\frac1n$ to any desired order. All the resulting integrals will be of the form $\alpha_k=\int_0^{\infty}s^{k}e^{-s}ds=k!$ and therefore are easily computable. In particular, $$I(n)=\frac12-\frac1{4n}+\frac1{8n^3}+O\left(n^{-5}\right).$$ To write complete asymptotic expansion, it is helpful to notice that $$\frac{1}{e^{s/n}+1}=\frac12\left(1-\tanh\frac s{2n}\right)=\frac12-\frac12\sum_{k=1}^{\infty}{\frac{2^{2k}(2^{2k}-1)B_{2k}}{(2k)!}}\left(\frac s{2n}\right)^{2k-1},$$ which implies that (here $B_k$ denote Bernoulli numbers) $$I(n)=\frac12-\sum_{k=1}^{\infty}{\frac{(2^{2k}-1)B_{2k}}{2k}}n^{1-2k}.$$

$\endgroup$
2
$\begingroup$

Consider $\,t:=x^n\,$ then the integral becomes \begin{align} I(n)&:=\int_0^1 \frac{x^n}{1+x} \, dx\\ &=\int_0^1 \frac{t}{1+t^{1/n}} \frac{t^{1/n-1}}n\, dt\\ &=\frac 1n \int_0^1 \frac{t^{1/n}}{1+t^{1/n}}\, dt\\ \end{align} You may expand $\,\dfrac{t^{1/n}}{n\;(1+t^{1/n})}\,$ in series as $n\to \infty$ (i.e. expand $e^{\log(t)/n}$) (faster with a CAS) : $$\frac 1{2n}+\frac{\log(x)}{4n^2}-\frac{\log(x)^3}{48 n^4}+\cdots$$ and integration between $0$ and $1$ should return you : $$\frac 1{2n}-\frac{1}{4n^2}+\frac{1}{8 n^4}-\cdots$$ (the integral of $\log(x)^n$ was given not long ago as $(-1)^n\,n!$)

Explicit solution: The coefficients may be obtained from the generating function of the Bernoulli numbers so that the final result is simply : $$I(n)\sim \sum_{k>0} \frac {1-2^{k}}{k}\frac{B_{k}}{n^k}$$

$\endgroup$
  • $\begingroup$ Hi dear @Startwearingpurple! (you merit a +1btw since a little earlier! :-)) In fact it should be the expansion of $\,\dfrac 1n\dfrac{t^{1/n}}{1+t^{1/n}}\,$. Ok got it the power of $3$ was wrong, thanks for that! $\endgroup$ – Raymond Manzoni Dec 26 '15 at 19:25
  • $\begingroup$ Yes, that was it. It is not immediately obvious but the asymptotic expansion does not contain even powers of $1/n$ except for the very first term $1/2$. $\endgroup$ – Start wearing purple Dec 26 '15 at 19:32
  • $\begingroup$ @Startwearingpurple: as in the Bernoulli numbers! In fact the numerators are given by OEIS A002425 so no much doubts about this! $\endgroup$ – Raymond Manzoni Dec 26 '15 at 19:37

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.