33
$\begingroup$

I am trying to prove that

$$\int_0^1 \frac{t^2-1}{(t^2+1)\log t}dt = 2\log\left( \frac{2\Gamma \left( \frac{5}{4}\right)}{\Gamma\left( \frac{3}{4}\right)}\right)$$

I know how to deal with integrals involving cyclotomic polynomials and nested logarithms but I have no idea with this one.

$\endgroup$
  • $\begingroup$ Nice question. (+1) $\endgroup$ – user 1357113 Jan 23 '13 at 17:14
24
$\begingroup$

Let's introduce the parameter $\alpha$, and then differentiate with respect to $\alpha$ that yields $$I(\alpha)=\int_0^1 \frac{t^\alpha-1}{(t^2+1)\ln t}dt $$ $$I'(\alpha)=\int_0^1 \frac{t^{\alpha}}{(t^2+1)}dt=\frac{1}{4} \left(-\psi_0\left(\frac{1 + \alpha}{4}\right) + \psi_0\left(\frac{3 + \alpha}{4}\right)\right) $$ Then $$I(\alpha)=\frac{1}{4} \int\left(-\psi_0\left(\frac{1 + \alpha}{4}\right) + \psi_0\left(\frac{3 + \alpha}{4}\right)\right) d\alpha= $$ $$I(\alpha)=\left(\ln \Gamma \left(\frac{3 + \alpha}{4}\right)- \ln \Gamma \left(\frac{1 + \alpha}{4}\right)\right)+C\tag1$$ If letting $\alpha=2$, then $$I(2)=\ln \left(\frac{\Gamma \left(\frac{5}{4}\right)}{\Gamma \left(\frac{3}{4}\right)}\right)+C$$ On the other hand, by letting $\alpha=0$ in $(1)$ we get $$C=\ln \left(\frac{\Gamma \left(\frac{1}{4}\right)}{\Gamma \left(\frac{3}{4}\right)}\right)$$ Thus $$\int_0^1 \frac{t^2-1}{(t^2+1)\ln t}dt=\ln \left(\frac{\Gamma \left(\frac{5}{4}\right)\Gamma \left(\frac{1}{4}\right)}{\Gamma^2 \left(\frac{3}{4}\right)}\right) $$

$\endgroup$
  • $\begingroup$ Nice answer (+1). $\endgroup$ – Mhenni Benghorbal Jan 23 '13 at 19:13
  • $\begingroup$ @MhenniBenghorbal: thank you! :-) $\endgroup$ – user 1357113 Jan 23 '13 at 19:14
  • $\begingroup$ Your final answer seems to be off by a factor of 2. $\endgroup$ – robjohn Jan 23 '13 at 21:07
  • $\begingroup$ @Chris'ssister: doh... I missed the square on the inside. Yes, your answer matches all the others :-) $\endgroup$ – robjohn Jan 23 '13 at 21:49
  • $\begingroup$ Thank You! This was very helpful. $\endgroup$ – Shobhit Bhatnagar Jan 24 '13 at 8:13
15
$\begingroup$

$$ \begin{align} \int_0^1\frac{t+1}{t^2+1}\frac{t-1}{\log(t)}\,\mathrm{d}t &=\int_0^1\frac{t+1}{t^2+1}\int_0^1t^x\,\mathrm{d}x\,\mathrm{d}t\\ &=\int_0^1\int_0^1\frac{t^{x+1}+t^x}{t^2+1}\,\mathrm{d}t\,\mathrm{d}x\\ &=\int_0^1\left(\frac1{x+1}+\frac1{x+2}-\frac1{x+3}-\frac1{x+4}+\dots\right)\,\mathrm{d}x\\ &=\left(\log\left(\frac21\right)+\log\left(\frac32\right)\right) -\left(\log\left(\frac43\right)+\log\left(\frac54\right)\right)+\dots\\ &=\log\left(\frac31\right)-\log\left(\frac53\right)+\log\left(\frac75\right)-\log\left(\frac97\right)+\log\left(\frac{11}9\right)-\dots\\ &=\log\left(\frac31\cdot\frac35\cdot\frac75\cdot\frac79\cdot\frac{11}9\cdots\right)\\ &=\lim_{n\to\infty}\log\left(\frac{\Gamma\left(\frac54\right)^2}{\Gamma\left(\frac34\right)^2} \frac{\Gamma\left(\frac{4n+3}4\right)^2}{\Gamma\left(\frac{4n+5}4\right)^2}(4n+3)\right)\\ &=2\log\left(2\frac{\Gamma\left(\frac54\right)}{\Gamma\left(\frac34\right)}\right) \end{align} $$ The last equality is due to Gautschi's inequality.

$\endgroup$
  • $\begingroup$ A powerful answer! I love the use of the double integrals technique that usually works great to Frullani's integrals (+1) $\endgroup$ – user 1357113 Jan 23 '13 at 21:24
  • $\begingroup$ @Chris'ssister: I actually started using $\lim\limits_{n\to\infty}n(x^{1/n}-1)=\log(x)$, but $$\frac1n\frac{t-1}{t^{1/n}-1}=\frac1n\left(t^{(n-1)/n}+t^{(n-2)/n}+t^{(n-3)/n}+\dots+1\right)$$ is the Riemann sum for $\int_0^1t^x\,\mathrm{d}t$, so I just used that. :-) $\endgroup$ – robjohn Jan 23 '13 at 21:58
  • $\begingroup$ Nice. I was only saying that $\int_0^1 \frac{t-1}{\log(t)}\,\mathrm{d}t$ is a Frullani integral and then may be nicely tackled by double integrals. It's exactly what you did. :) $\endgroup$ – user 1357113 Jan 23 '13 at 22:04
13
$\begingroup$

There have been similar integrals such as this with a factor of $1/\log{t}$ in the integrand. The way I have attacked these is to use the substitution $t=e^{-x}$; here, this produces

$$-\int_0^{\infty} dx \: \frac{e^{-x}}{x} \frac{1-e^{-2 x}}{1+e^{-2 x}} $$

$$ = -\int_0^{\infty} dx \: \frac{1}{x} (e^{-x} - e^{-3 x}) \sum_{k=0}^{\infty} (-1)^k e^{-2 k x} $$

Now, we reverse the order of the sum and integral. This is justified by Abel's theorem, although I will leave the details for another time or another person to fill in:

$$ = -\sum_{k=0}^{\infty} (-1)^k \int_0^{\infty} dx \: \frac{1}{x} (e^{-(2 k+1) x} - e^{-(2 k+3) x}) $$

The integrals have a simple closed form, and the result is now a sum:

$$ = \sum_{k=0}^{\infty} (-1)^{k+1} \log{ \left [ \frac{(2 k+1)}{(2 k+3)} \right ] } $$

You can form a log of a Wallis-type product from this sum:

$$ = \log{\left [ \prod_{k=0}^{\infty} \frac{4 k+3}{4 k+1} \frac{4 k+3}{4 k+5} \right ]} $$

Partial products of the above product may be evaluated:

$$ \prod_{k=0}^{n} \frac{4 k+3}{4 k+1} \frac{4 k+3}{4 k+5} = \frac{9 \Gamma{\left ( \frac{5}{4} \right )} \Gamma{\left ( \frac{9}{4} \right )} \Gamma{\left ( n + \frac{7}{4} \right )}^2}{5 \Gamma{\left ( \frac{7}{4} \right )}^2 \Gamma{\left (n+ \frac{5}{4} \right )} \Gamma{\left ( n + \frac{9}{4} \right )}}$$

You may show that the above expression converges as $n \rightarrow \infty$. A little manipulation produces the stated result.

EDIT

@MikeSpivey points my attention to an equation out of Whittaker & Watson that applies here:

$$ \prod_{k=0}^{\infty} \frac{4 k+3}{4 k+1} \frac{4 k+3}{4 k+5} = \prod_{k=0}^{\infty} \frac{(k+\frac{3}{4})^2}{(k+\frac{1}{4})(k+\frac{5}{4})} = \frac{\Gamma{\left ( \frac{1}{4} \right )} \Gamma{\left ( \frac{5}{4} \right )}}{\Gamma{\left ( \frac{3}{4} \right )^2}} = \frac{2^2 \Gamma{\left ( \frac{5}{4} \right )^2}}{\Gamma{\left ( \frac{3}{4} \right )^2}} $$

The result immediately follows.

$\endgroup$
  • 5
    $\begingroup$ +1. Also, the formula $$\prod_{k=0}^{\infty} \frac{(k+a_1) \cdots (k+a_j)}{(k + b_1) \cdots (k+b_j)} = \frac{\Gamma(b_1) \cdots \Gamma(b_k)}{\Gamma(a_1) \cdots \Gamma(a_k)},$$ where $a_1 + a_2 + \cdots + a_j = b_1 + b_2 + \cdots + b_j$ and no $b_j$ is zero or a negative integer, plus a little manipulation, will finish off your answer. Sondow and Yi give Whittaker and Watson (Section 12.13) as a reference for the formula and state that it is a corollary of the Weierstrass infinite product for the gamma function. $\endgroup$ – Mike Spivey Jan 23 '13 at 18:38
  • $\begingroup$ @MikeSpivey: thanks. Should have dusted off that Whittaker & Watson reference before I typed! $\endgroup$ – Ron Gordon Jan 23 '13 at 18:41

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.