11
$\begingroup$

I met this integral

$$ \int_{0}^{1} \left(\frac{1}{\ln x} + \frac{1}{1-x} -\frac{1}{2} \right) \frac{ \mathrm{d}x}{1-x} \qquad (*) $$

while evaluating this log-cosine integral. I made several attemps before being successful.

Find a closed form for

$$ I(s): = \int_{0}^{1} \left(\frac{1}{\ln x} + \frac{1}{1-x} -\frac{1}{2} \right) \frac{x^s }{1-x}\mathrm{d}x, \quad \Re (s)>-1. \qquad (**) $$

$\endgroup$
  • $\begingroup$ The first thing I would do is to consider $I'(s)$ instead of $I(s)$ in order to remove $\log x$ from the denominator of the first term of the integral, and deal with a more treatable integral. $\endgroup$ – Jack D'Aurizio Jul 21 '14 at 21:22
7
$\begingroup$

As stated in the comments, by following the Feynman's approach to integrals we have: $$\begin{eqnarray*} I'(s) &=& \int_{0}^{1}\left(1+\frac{\log x}{1-x}-\frac{\log x}{2}\right)\frac{x^s}{1-x}dx\\ &=& \frac{1}{2}\zeta(2,s+1)+\int_{0}^{1}\left(1+\frac{\log x}{1-x}\right)\frac{x^s}{1-x}dx\\&=&\frac{1}{2}\zeta(2,s+1)-\frac{1}{2s+2}\phantom{}_3 F_2\left(1,1,2;3,s+2;1\right),\end{eqnarray*}$$ but probably there is a nicer form for the second term.

As a matter of fact, by treating the second term this way: $$\int_{0}^{1}\left(1+\frac{\log x}{1-x}\right)\frac{x^s}{1-x}dx=\lim_{a\to 0^+}\int_{0}^{1}\left(1+\frac{\log x}{1-x}\right)\frac{x^s}{(1-x)^{1-a}}$$ we get: $$ I'(s)=\frac{1}{2}\psi'(s+1)+s\,\psi'(s)+\psi(s)-\psi(s+1)-1\tag{1}$$ where $\psi$ is the digamma function $\psi(s)=\frac{\Gamma'(s)}{\Gamma(s)}$. By integrating $(1)$ we have: $$ I(s) = I(1)+\frac{3\gamma+1}{2}-s+\frac{\psi(s+1)}{2}-\log\Gamma(s+1)+s\,\psi(s).\tag{2}$$

All we need to find a closed expression for $I(s)$ is now to evaluate $I(1)$ or $I(0)$, since by $(2)$: $$ I(0)=I(1)+\gamma-\frac{1}{2}$$ holds. In virtue of $(2)$ we also have: $$\lim_{s\to +\infty} I(s) = I(1) + \frac{3\gamma-\log(2\pi)}{2},$$ but since $$ f(x)=\left(\frac{1}{\log x}+\frac{1}{1-x}-\frac{1}{2}\right)\frac{x}{1-x}$$ is a continuous, positive, increasing and bounded function ($f(x)\leq\frac{1}{12}$) on $(0,1)$, as long as $s\to+\infty$ we have $I(s)\leq\frac{1}{12s}$, so $\lim_{s\to +\infty}I(s)=0$. Hence we have:

$$ I(1) = \frac{\log(2\pi)-3\gamma}{2},\qquad I(0) = \frac{\log(2\pi)-\gamma-1}{2} \tag{3}$$

and:

$$ I(s) = \frac{1+\log(2\pi)}{2}-s+\frac{\psi(s+1)}{2}-\log\Gamma(s+1)+s\,\psi(s).\tag{4}$$

As a bonus, by expanding $(4)$ in a neighbourhood of $+\infty$ and exploiting convexity, we can also show that: $$\frac{1}{12s+8}\leq I(s)\leq\frac{1}{12s+6}$$ holds for any $s\in\mathbb{R}^+$.

$\endgroup$
  • $\begingroup$ Thanks for your interest! Please, there might be a mistake, you have $I(s) \sim -\ln s$ as $s$ tends to $0^+$ in (2). $\endgroup$ – Olivier Oloa Jul 21 '14 at 23:18
  • $\begingroup$ You also have two extra terms in your line (1), namely:$\psi(s)$ and $-\psi(s+1)$, the rest is correct. $\endgroup$ – Olivier Oloa Jul 22 '14 at 0:45
  • $\begingroup$ @OlivierOloa: yes, there were two mistakes. The first one was a forgotten half when bringing back the Hurwitz zeta into its $\psi'$ form, the second one was a typo for which a $\psi'(s)$ became a $\psi'(s+1)$. Now the formula is right, I checked it numerically, too. $\endgroup$ – Jack D'Aurizio Jul 22 '14 at 0:46
  • $\begingroup$ The term $\psi(s)-\psi(s+1)=-\frac{1}{s}$ is right. $\endgroup$ – Jack D'Aurizio Jul 22 '14 at 0:48
  • $\begingroup$ OK. So what is the value of I:=I(0)? $(*)$ $\endgroup$ – Olivier Oloa Jul 22 '14 at 0:49
5
$\begingroup$

Recall Binet's formula $$ \log \Gamma(z)= \left( z-\frac{1}{2}\right)\log z - z + \frac{1}{2}\log(2\pi) + \int_0^{\infty} \! \left(\frac{1}{2} - \frac{1}{x} + \frac{1}{e^{x}-1} \right)\frac{e^{-zx}}{x} \mathrm{d}x,\quad \Re z >0 $$ which, upon making $x = - \log v$, can be written as

$$ \log \Gamma(z)= \! \left( z-\frac{1}{2}\right)\log z - z + \frac{1}{2}\log(2\pi) - \!\! \int_0^{1} \! \left(\frac{1}{\log v}+\frac{1}{1-v}-\frac{1}{2}\right)\frac{v^{z-1}}{\log v}\mathrm{d}v, \, \Re z >0 $$

and Gauss' formula

$$ -\psi(z)+\log z = \int_0^{1} \left(\frac{1}{\log v}+\frac{1}{1-v}\right)v^{z-1} \mathrm{d}v. $$

One may check that \begin{multline} \displaystyle \left(\frac{1}{\log u} + \frac{1}{1-u} - \frac{1}{2} \right) \frac{u^s}{1-u} = u^{s} \frac{d}{du}\left\{u \left(\frac{1}{\log u} + \frac{1}{1-u}\right)\right\} \, \\ + \left(\displaystyle \frac{1}{\log u} + \frac{1}{1-u} - \frac{1}{2} \right) \frac{u^s}{\log u} - \frac{1}{2} \left(\displaystyle \frac{1}{\log u} + \frac{1}{1-u}\right)u^s. \end{multline}

Hence our integral $I(s)$ is the sum of three integrals.

The first, \begin{multline} I_{1}(s) =\displaystyle \left.\left(\frac{1}{\log u} + \frac{1}{1-u} \right)u^{s+1} \right|_{0}^{1} - s \int_{0}^{1}\! \displaystyle \left(\displaystyle \frac{1}{\log u} + \frac{1}{1-u}\right)u^{s} \mathrm{d}u, \end{multline} using Gauss' formula, $$ I_{1}(s) = \frac{1}{2} + s \left( \psi(s+1)-\log \left(s+1\right) \right). $$ The second, applying Binet's formula, \begin{multline} I_{2} (s)= \displaystyle \int_{0}^{1}\! \left(\displaystyle \frac{1}{\log u} + \frac{1}{1-u} -\frac{1}{2} \right) \frac{u^{s}}{\log u} \mathrm{d}u \\ = -\log \Gamma(s+ 1)+ \left(s + 1/2 \right)\log \left(s+ 1 \right) - s - 1 + \frac{1}{2}\ln(2\pi). \end{multline} The third, using Gauss' formula once more \begin{equation} I_{3}(s) = - \frac{1}{2} \displaystyle \int_{0}^{1}\! \left(\displaystyle \frac{1}{\log u} + \frac{1}{1-u}\right)u^{s} \mathrm{d}u = \frac{1}{2}\left( \psi(s+1)-\log \left(s+1\right) \right). \end{equation} Consequently, $ I(s)=I_{1}(s)+I_{2}(s)+I_{3}(s)$ is given by

\begin{multline} I(s):= \displaystyle \int_0^{1} \left(\frac{1}{\log u}+ \frac{1}{1-u}-\frac{1}{2}\right) \frac{u^s}{1-u} \mathrm{d}u \\ = -\log \Gamma(s+1) + \left( s +\frac{1}{2}\right) \psi(s+1) - s - \frac{1}{2} + \frac{1}{2}\ln(2\pi). \end{multline}

We have, with $s=0$,

$$ \int_{0}^{1} \left(\frac{1}{\log x} + \frac{1}{1-x} -\frac{1}{2} \right) \frac{ \mathrm{d}x}{1-x} = -\frac{1}{2}+\frac{1}{2} \ln (2 \pi)-\frac{1}{2} \gamma. $$

$\endgroup$

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.