35

Some time ago I came across to the following integral:

$$I=\int_{0}^{1}\frac{1-x}{1+x}\frac{\mathrm dx}{\ln x}$$ What are the hints on how to compute this integral?

Ali Caglayan
  • 5,726
Martin Gales
  • 6,878

5 Answers5

21

Make the substitution $x=e^{-y}$ and do a little algebra to get the value of the integral to be

$$\int_0^{\infty} \frac{dy}{y} \frac{e^{-y} - e^{-2 y}}{1+e^{-y}} $$

Now Taylor expand the denominator and get

$$\int_0^{\infty} \frac{dy}{y} (e^{-y} - e^{-2 y}) \sum_{k=0}^{\infty} (-1)^k e^{-k y} $$

If we can reverse the order of sum and integral, we get

$$ \sum_{k=0}^{\infty} (-1)^k \int_0^{\infty} \frac{dy}{y} (e^{-(k+1) y} - e^{-(k+2) y}) $$

The integral inside may be evaluated exactly, and the result is the sum

$$ \sum_{k=0}^{\infty} (-1)^k \log{\frac{k+1}{k+2}} $$

$$ = \lim_{n \rightarrow \infty} \; \log{\frac{\frac{1}{2} \frac{3}{4} \ldots \frac{2 n-1}{2 n}}{\frac{2}{3} \frac{5}{6} \ldots \frac{2 n+1}{2 n+2}}} $$

$$ = \log{\left ( \frac{2}{\pi} \right )} $$

Ron Gordon
  • 138,521
  • Seems reasonable. Thank you! – Martin Gales Jan 16 '13 at 16:27
  • Thank you for asking it - this was fun! – Ron Gordon Jan 16 '13 at 16:40
  • 2
    Yeah, math is about having fun! :-) (+1) – user 1591719 Jan 16 '13 at 16:54
  • @rlgordonma: above did you just apply the known formula of Frullani's integral? http://mathworld.wolfram.com/FrullanisIntegral.html – user 1591719 Jan 16 '13 at 17:15
  • @rlgordonma: is the interchange of summation and integration dues to continuity on $[0, \infty]$? – Alex Jan 16 '13 at 17:21
  • Changing the order is due to absolute convergence of sum and integral, which appears to be the case. See T.W. Koerner, Fourier Analysis, Secs. 47 & 48. – Ron Gordon Jan 16 '13 at 18:44
  • @Chris'ssister: Indeed I did, although I didn't know it had a name. It is actually quite easy: when the integrand satisfies particular convergence properties, you may then express the integrand as the result of a definite integral, and then you change the order of integration. I picked up this result from playing around with Putnam Exam problems. – Ron Gordon Jan 16 '13 at 18:46
  • That is a result of Wallis' famous infinite product representation of $\frac{\pi}{2}$. See here, for example: http://en.wikipedia.org/wiki/Wallis_product – Ron Gordon Jan 16 '13 at 23:00
  • Thank You. though, I quickly deleted the comment. – 007resu Jan 16 '13 at 23:01
  • 1
    To reverse the order of sum and integra, I understand that you appeal to the absolute convergence of sum and integral, that is, to the convergence of the positive series $\sum\limits_{k\geqslant0}\int\limits_0^{+\infty}(e^{-(k+1)y}-e^{-(k+2)y})dy/y$. But this series diverges. Maybe I did not understand the result you are applying, could you explain? – Did Jan 21 '13 at 09:17
  • @did: No, that's not right. I appealed to the convergence of the sum and integral in the 2nd line above to produce the 3rd line. To reverse the order of integration on the quantity in the 3rd line to produce the 2nd would take some other reasoning. To be specific, see Koerner, Theorem 47.5, which is applied strictly to double integrals, but may be applied to the sum by considering the sum as an integral over the integrand times a sum of characteristic functions over $[k,k+1]$. – Ron Gordon Jan 21 '13 at 12:06
  • No problem with seeing sums as integrals. But, to apply so-called Theorem 47.5 (which most people would simply refer to as Fubini theorem), the double integral/sum of the absolute values must converge. Here, this is exactly the series of integrals I wrote in my previous comment. It diverges. Hence one cannot reverse the order of sum and integral as you do (and need to do) in your answer. – Did Jan 21 '13 at 14:29
  • 1
    Maybe we are speaking across each other. Look, do you agree that the sum in Line 2 absolutely converges (in this case, to $(1-e^{-y})^{-1})$? And that the integral absolutely converges (in this case, to $\log{\frac{k+1}{k+2}}$). Because each integral is also continuous, then we get a convergent result in exchanging the order of the sum and integral in Line 2. As I said, this does not apply to Line 3 because, as you point out, that the sum does not converge absolutely. But I am not seeking to reverse the order of integration here, just in Line 2. – Ron Gordon Jan 21 '13 at 14:38
  • I also appealed to Koerner because I really like his exposition on the topic, a really nice mixture of rigor (or an awareness of where more rigor could be used) and pedantry. – Ron Gordon Jan 21 '13 at 14:40
  • the sum in Line 2 absolutely converges... Yes, one considers functions $f_k$ such that $\sum\limits_k|f_k(x)|$ converges for every $x$ (except for $x=0$, but anyway). // the integral absolutely converges... Yes, each $\int|f_k|$ is finite (and equals $\log((k+2)/(k+1))$, not the opposite, but anyway). // Because each integral is also continuous... Eeeeh? What is continuous? The functions $f_k$? Not useful (but anyway). // we get a convergent result in exchanging the order of the sum and integral in Line 2... CERTAINLY NOT. .../... – Did Jan 21 '13 at 22:22
  • .../... Look at the hypothesis in Körner, it reads $\int\sum\limits_k|f_k|$ or $\sum\limits_k\int|f_k|$ finite. Here, by your own computation, $\sum\limits_k\int|f_k|=\sum\limits_k\log((k+2)/(k+1))$. This sums to $+\infty$. // Conclusion: Fubini/Theorem 47.5 does not apply. – Did Jan 21 '13 at 22:22
  • Clearly I am not understanding something. Maybe you could explain to me, @did, where the justification lies in my switching the order of integration. You would be doing me a huge favor. I ask this because it does in fact produces a result that we all found acceptable. How did it come about if I misunderstood it? – Ron Gordon Jan 21 '13 at 22:44
  • The best I can do is to urge you to check the conditions of Körner's Theorem 47.5. These are not met in the present situation. Other results, with different conditions, apply here, which make the final formula correct, but not the result you mention. (Hint: Alternating series.) – Did Jan 21 '13 at 22:48
  • Something along the lines of Abel's Theorem as mrf mentioned in the other thread? If so, I will outline reasoning here to that effect so we can resolve this issue. And yes, I will reexamine Koerner/Fubini. – Ron Gordon Jan 21 '13 at 23:04
9

I tried with "Differentiation under integration sign":

$$J(\alpha)=\int_0^1\frac{1-x^\alpha}{1+x}\frac{dx}{\ln x}\quad \alpha>0$$

Then, as usual, $$\frac{dJ}{d\alpha}=-\int_0^1\frac{x^\alpha}{1+x}dx=-f(\alpha),\text{(say)}$$

Then, integrating $$J(\alpha)=-\int f(\alpha)d\alpha+c$$ I used Mathematica to evaluate $f(\alpha)$ (it gives difference of two Harmonic numbers) and its integral. The integral is just $\ln\frac{\Gamma(\frac{2m+1}{2})}{\Gamma(\frac{m+1}{2})}$. Note that $J(0)=0$ and putting $\alpha=1$, the result follows.

Tapu
  • 3,496
6

Consider $$ \mathcal{I}(\alpha)=\int_0^1\frac{1-x^\alpha}{(1+x)\ln x}\ dx.\tag1 $$ Differentiating $(1)$ with respect $\alpha$ yields \begin{align} \frac{d\mathcal{I}}{d\alpha}&=\int_0^1\frac{\partial}{\partial\alpha}\left[\frac{1-x^\alpha}{(1+x)\ln x}\right]\ dx\\ &=-\int_0^1\frac{x^\alpha}{1+x}\ dx\\ &=-\int_0^1\sum_{k=0}^\infty(-1)^kx^{\alpha+k}\ dx\\ &=-\sum_{k=0}^\infty(-1)^k\int_0^1x^{\alpha+k}\ dx\\ &=-\sum_{k=0}^\infty\frac{(-1)^k}{\alpha+k+1}.\tag2 \end{align} Now consider polygamma function $$ \psi_n(z)=\frac{d^{n+1}}{dz^{n+1}}\ln\Gamma(z)\tag3 $$ and $$ \psi_n(z)=(-1)^{n+1}n!\sum_{k=0}^\infty\frac{1}{(z+k)^{n+1}}.\tag4 $$ Hence by using $(4)$ we obtain $$ \sum_{k=0}^\infty\frac{(-1)^{k}}{(z+k)^{n+1}}=\frac1{(-2)^{n+1}n!}\left[\psi_n\left(\frac{z}{2}\right)-\psi_n\left(\frac{z+1}{2}\right)\right].\tag5 $$ Using $(3)$ and $(5)$ then $(2)$ becomes \begin{align} \frac{d\mathcal{I}}{d\alpha}&=\frac12\left[\psi_0\left(\frac{\alpha+1}{2}\right)-\psi_0\left(\frac{\alpha+2}{2}\right)\right]\\ \mathcal{I}(\alpha)&=\frac12\int\left[\psi_0\left(\frac{\alpha+1}{2}\right)-\psi_0\left(\frac{\alpha+2}{2}\right)\right]\ d\alpha\\ &=\ln\Gamma\left(\frac{\alpha+1}{2}\right)-\ln\Gamma\left(\frac{\alpha+2}{2}\right)+C\\ &=\ln\Gamma\left(\frac{\alpha+1}{2}\right)-\ln\Gamma\left(\frac{\alpha+2}{2}\right)-\frac12\ln\pi,\tag6 \end{align} where $\mathcal{I}(0)=0$ and $C=-\ln\Gamma\left(\frac{1}{2}\right)$.


Thus \begin{align} \int_0^1\frac{1-x}{(1+x)\ln x}\ dx&=\mathcal{I}(1)\\ &=\ln\Gamma\left(1\right)-\ln\Gamma\left(\frac{3}{2}\right)-\frac12\ln\pi\\ &=\color{blue}{\ln\left(\frac{2}{\pi}\right)}. \end{align}

Tunk-Fey
  • 24,849
5

$\newcommand{\+}{^{\dagger}} \newcommand{\angles}[1]{\left\langle\, #1 \,\right\rangle} \newcommand{\braces}[1]{\left\lbrace\, #1 \,\right\rbrace} \newcommand{\bracks}[1]{\left\lbrack\, #1 \,\right\rbrack} \newcommand{\ceil}[1]{\,\left\lceil\, #1 \,\right\rceil\,} \newcommand{\dd}{{\rm d}} \newcommand{\down}{\downarrow} \newcommand{\ds}[1]{\displaystyle{#1}} \newcommand{\expo}[1]{\,{\rm e}^{#1}\,} \newcommand{\fermi}{\,{\rm f}} \newcommand{\floor}[1]{\,\left\lfloor #1 \right\rfloor\,} \newcommand{\half}{{1 \over 2}} \newcommand{\ic}{{\rm i}} \newcommand{\iff}{\Longleftrightarrow} \newcommand{\imp}{\Longrightarrow} \newcommand{\isdiv}{\,\left.\right\vert\,} \newcommand{\ket}[1]{\left\vert #1\right\rangle} \newcommand{\ol}[1]{\overline{#1}} \newcommand{\pars}[1]{\left(\, #1 \,\right)} \newcommand{\partiald}[3][]{\frac{\partial^{#1} #2}{\partial #3^{#1}}} \newcommand{\pp}{{\cal P}} \newcommand{\root}[2][]{\,\sqrt[#1]{\vphantom{\large A}\,#2\,}\,} \newcommand{\sech}{\,{\rm sech}} \newcommand{\sgn}{\,{\rm sgn}} \newcommand{\totald}[3][]{\frac{{\rm d}^{#1} #2}{{\rm d} #3^{#1}}} \newcommand{\ul}[1]{\underline{#1}} \newcommand{\verts}[1]{\left\vert\, #1 \,\right\vert} \newcommand{\wt}[1]{\widetilde{#1}}$ \begin{align} I & \equiv \bbox[#ffd, 5px]{\int_{0}^{1}{1 - x \over 1 + x} \,{\dd x \over \ln\pars{x}}} = \int_{0}^{1}{1 \over 1 + x}\ \overbrace{\bracks{-\int_{0}^{1}x^{t}\,\dd t}} ^{\ds{1 - x \over \ln\pars{x}}}\ \,\dd x = -\int_{0}^{1}\int_{0}^{1}{x^{t} \over 1 + x}\,\dd t\,\dd x \\[5mm] & = -\int_{0}^{1}\int_{0}^{1}{x^{t} - x^{t + 1} \over 1 - x^{2}}\,\dd x\,\dd t \,\,\,\stackrel{x^{2}\ \mapsto\ x}{=}\,\,\, -\,{1 \over 2}\int_{0}^{1}\int_{0}^{1}{x^{t/2 - 1/2}\,\,\, -\, x^{t/2} \over 1 - x}\,\dd x\,\dd t \\[5mm] & = {1 \over 2}\int_{0}^{1}\bracks{% \int_{0}^{1}{1 - x^{t/2 - 1/2} \over 1 - x}\,\dd x - \int_{0}^{1}{1 - x^{t/2} \over 1 - x}\,\dd x}\,\dd t \\[5mm] & = {1 \over 2}\int_{0}^{1}\bracks{% \Psi\pars{{t \over 2} + {1 \over 2}} - \Psi\pars{{t \over 2} + 1}}\,\dd t\label{1}\tag{1} \\[5mm] & = \left.\ln\pars{\Gamma\pars{t/2 + 1/2} \over \Gamma\pars{t/2 + 1}} \right\vert_{0}^{1} = \ln\pars{{\Gamma\pars{1} \over \Gamma\pars{3/2}} \,{\Gamma\pars{1} \over \Gamma\pars{1/2}}} = \ln\pars{{1 \over \root{\pi}/2}\,{1 \over \root{\pi}}} \\[5mm] & = \bbox[10px, border:1px groove navy]{\ln\pars{2 \over \pi}} \approx -0.4514 \\ & \end{align} (\ref{1}): See $\ds{\color{black}{\bf 6.3.22}}$ in A & S Table.

Felix Marin
  • 89,464
3

You may like this method. Note $$ \int_0^1x^ada=-\frac{1-x}{\ln x}, \text{ for }x>0, H_a=\sum_{k=1}^\infty\frac{a}{k(a+k)}, $$ and hence \begin{eqnarray} \int_0^1\frac{1-x}{1+x}\frac{dx}{\ln x}&=&-\int_0^1\int_0^1\frac{x^a}{1+x}dadx\\ &=&-\int_0^1\int_0^1\frac{x^a}{1+x}dxda\\ &=&-\frac{1}{2}\int_0^1(H_{\frac{a}{2}}-H_{\frac{a-1}{2}})da\\ &=&-\frac{1}{2}\int_0^1\sum_{k=1}^\infty\left(\frac{a}{k(a+2k)}-\frac{a-1}{k(a+2k-1)}\right)da\\ &=&-\frac{1}{2}\sum_{k=1}^\infty\left(4\ln k-2\ln(k^2-\frac{1}{4})\right)\\ &=&-\frac{1}{2}\ln\prod_{k=1}^\infty\frac{k^4}{(k^2-\frac{1}{4})^2}\\ &=&-\prod_{k=1}^\infty\frac{k^2}{k^2-\frac{1}{4}}\\ &=&-\ln\left(\frac{\pi}{2}\right). \end{eqnarray} In the last step, we used the Wallis Formula from here.

xpaul
  • 44,000