7
$\begingroup$

While browsing the post Is there any integral for the golden ratio $\phi$?, I came across this nice answer, $$ \int_0^\infty \frac{1}{1+x^{10}}dx=\frac{\pi\,\phi}5$$ it seems the general form is just

$$p \int_0^\infty \frac{1}{1+x^{p}}dx=\color{blue}{\frac{\pi}{\sin\big(\tfrac{\pi}{p}\big)}}$$

I wondered about $$\int_0^\color{red}1 \frac{1}{1+x^p}dx=\,?$$ Mathematica could find messy closed-forms for $p=5,7$. After some laborious simplification, $$5\int_0^1 \frac{1}{1+x^5}dx=\frac{\pi\sqrt{\phi}}{5^{1/4}}+\ln2+\sqrt{5}\ln\phi$$

Question 1: In general, is it true that for any $p$ ,

$$2p\,\int_0^1 \frac{1}{1+x^p}dx=\color{blue}{\frac{\pi}{\sin\big(\tfrac{\pi}{p}\big)}}+2\ln2-\psi\big(\tfrac{1}{p}\big)+\psi\big(\tfrac{p-1}{2p}\big)+\psi\big(\tfrac{p+1}{2p}\big)-\psi\big(\tfrac{p-1}{p}\big)$$

where $\psi(z)$ is the digamma function?

Note: The four digammas, implemented in Mathematica as PolyGamma[z], can be expressed as a sum of cosines x logarithms for odd $p=2m+1$. Let $k=\frac{2n-1}{p}\pi$, then, $$-\psi\big(\tfrac{1}{p}\big)+\psi\big(\tfrac{p-1}{2p}\big)+\psi\big(\tfrac{p+1}{2p}\big)-\psi\big(\tfrac{p-1}{p}\big)=-4\sum_{n=1}^m \cos (k)\ln\big(\sin\tfrac{k}{2}\big)$$

Question 2: For even $p$, can we can also avoid the digamma by using cosines and logarithms?

$\endgroup$
  • $\begingroup$ Can taylor expansion help? $\endgroup$ – vidyarthi Nov 26 '16 at 4:11
  • $\begingroup$ When the integral is evaluated by Maxima, which is similar to Mathematica, it uses factorization of denominator followed by partial fraction and some substitutions to get a complicated expression involving series of logarithms and the arctan function $\endgroup$ – vidyarthi Nov 26 '16 at 4:19
  • $\begingroup$ @vidyarthi: Can you put a simplified version of Maxima's result? $\endgroup$ – Tito Piezas III Nov 26 '16 at 4:22
  • $\begingroup$ here is the image of the final result for $p=10$ taken from integral-calculator.com which uses maxima-imgur.com/iezjQ2u $\endgroup$ – vidyarthi Nov 26 '16 at 4:29
  • 2
    $\begingroup$ You are asking, slightly in disguise, for explicit forms of the sum of the series $$\sum_{n=0}^\infty\frac{(-1)^n}{1+np}=p\sum_{n=0}^\infty\frac{1}{(1+2np)(1+(2n+1)p)}$$ which converges at least for every positive $p$. The "solutions" based on hypergeometric or digamma functions are, if only one looks at the definition of these special functions, mere rewritings of the sum of this series. $\endgroup$ – Did Nov 26 '16 at 8:46
2
$\begingroup$

I'm only going to address Question 1.

The expression proposed in Question 1 is true. However, it is a little bit too complicated than necessary. A simpler version of the expression is

$$2p\int_0^1 \frac{dx}{1+x^p} = \psi\left(\frac{1}{2p} + \frac12\right) - \psi\left(\frac{1}{2p}\right)$$


From reflection formula, take logarithm and differentiate, we get $$\Gamma(z)\Gamma(1-z) = \frac{\pi}{\sin\pi z} \implies \psi(z) - \psi(1-z) = \pi\cot\pi z$$ This leads to $$\frac{\pi}{\sin z} = \pi\cot\frac{\pi z}{2} - \pi\cot\pi z = \psi\left(\frac{z}{2}\right) - \psi\left(1-\frac{z}{2}\right) - \psi(z) + \psi(1-z) $$ From duplication formula, take logarithm and differentiate, we get $$\Gamma(z)\Gamma\left(z+\frac12\right) = 2^{1-2z}\sqrt{\pi}\Gamma(2z) \implies \psi(z) + \psi\left(z + \frac12\right) = -2\log 2 + 2\psi(2z) $$ Apply these to RHS of Question 1, we can expose RHS to following mess $$ \left[ \color{red}{\psi\left(\frac{1}{2p}\right)} - \psi\left(1 - \frac{1}{2p}\right) - \color{green}{\psi\left(\frac{1}{p}\right)} + \color{blue}{\psi\left(1 - \frac{1}{p}\right)} \right] + \left[ \color{green}{2\psi\left(\frac1p\right)} - \color{red}{\psi\left(\frac{1}{2p}\right)} - \color{magenta}{\psi\left(\frac{p+1}{2p}\right)}\right]\\ - \color{green}{\psi\left(\frac{1}{p}\right)} + \psi\left(\frac{p-1}{2p}\right) + \color{magenta}{\psi\left(\frac{p+1}{2p}\right)} - \color{blue}{\psi\left(\frac{p-1}{p}\right)} $$ After massive cancellation, we can simplify RHS to $$ \psi\left(\frac{p-1}{2p}\right) - \psi\left(1 - \frac{1}{2p}\right) = \psi\left( 1 - \left(\frac{1}{2p} + \frac12\right)\right) - \psi\left(1 - \frac{1}{2p}\right) = \psi\left(\frac{1}{2p} + \frac12\right) - \psi\left(\frac{1}{2p}\right) $$ Recall following expansion of digamma function

$$\psi(z) = \frac{1}{z} + \sum_{n=1}^\infty \left(\frac{1}{z+n} - \frac{1}{n}\right)$$ We find $$\begin{align} \text{RHS} &= \frac{1}{\frac{1}{2p}} - \frac{1}{\frac{1}{2p} + \frac12} + \sum_{n=1}^\infty\left(\frac{1}{\frac{1}{2p}+n} - \frac{1}{\frac{1}{2p} + n + \frac12}\right)\\ &= 2\sum_{n=0}^\infty\frac{(-1)^n}{\frac{1}{p}+n} = 2\sum_{n=0}^\infty\int_0^1 (-1)^n t^{\frac{1}{p}+n-1} dt = 2\int_0^1 \sum_{n=0}^\infty (-1)^n t^{\frac{1}{p}+n-1} dt\\ &= 2 \int_0^1 \frac{t^{\frac{1}{p}-1}}{1+t} dt = 2p\int_0^1 \frac{dx}{1+x^p} = \text{LHS} \end{align} $$

$\endgroup$
  • $\begingroup$ Thanks for the confirmation. I started with $p=5$ and arrived at the complicated version which shows that when $p$ is odd, the four digammas (hence the integral) is expressible in terms of elementary functions. However, when $p$ is even, perhaps one has to stay with the digammas. $\endgroup$ – Tito Piezas III Nov 26 '16 at 11:20
2
$\begingroup$

This answer is incomplete as it does not proof all the steps completely:

Let us start with the partial fraction expansion $$\tag{1}\frac{p}{1+x^p}=- \sum_{j=1}^p\frac{\omega^{2j-1}}{x-\omega^{2j-1}}$$ with $\omega= e^{i\pi/p}$.

Now, we can integrate and obtain $$\int_0^1\!dx \frac{p}{1+x^p} = -\sum_{j=1}^p \omega^{2j-1} \log\left(1-\omega^{1-2j}\right). $$

We can simplify this formula a bit taking the real part. This yields $$\int_0^1\!dx \frac{p}{1+x^p} = \pi\sum_{j=1}^{p} \frac{1-2j+p}{2p} \sin\left(\frac{\pi(1-2j)}{p}\right)+\sum_{j=1}^{p}\cos\left(\frac{\pi(1-2j)}{p}\right) \log\left[2 \sin\left(\frac{\pi(1-2j)}{2p}\right)\right].$$

The first term can be summed explicitly $$\int_0^1\!dx \frac{p}{1+x^p} =\frac{\pi}{2\sin(\pi/p)}+\sum_{j=1}^{p}\cos\left(\frac{\pi(1-2j)}{p}\right) \log\left[2 \sin\left(\frac{\pi(1-2j)}{2p}\right)\right].$$ The second term can be related to the digamma function.

Formula (1) can be proven by multiplying the expression by $1+x^p$. For the right hand side, we obtain $$- \sum_{j=1}^p\omega^{2j-1} \prod_{k\neq j} (x-\omega^{2k-1}) = -x^{p-1}\sum_{j=1}^p\omega^{2j-1}+ x^{p-2} \sum_{j=1}^p\omega^{2j-1} \sum_{k,k'\neq j} \omega^{2k-1}\omega^{2k'-1} - \cdots\\+(-1)^p \sum_{j=1}^p\omega^{2j-1} \prod_{k\neq j} \omega^{2k-1}$$ as $1+x^p=\prod_{k=1}^p (x- \omega^{2j-1})$. The result (1) follows from some not so straightforward combinatorics.

$\endgroup$

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.