I have been able to show that for $n\in\Bbb N_{\geq2}$ $$\phi(n)=\int_0^1\frac{dx}{x^n+1}=\sum_{k=0}^{n-1}\Gamma_{n,k}\log\frac{\lambda_{n,k}-1}{\lambda_{n,k}}$$ Where $$\lambda_{n,k}=\exp\frac{i\pi(2k+1)}{n}$$ And $$\Gamma_{n,k}=\prod_{k\neq j=0}^{n-1}\frac1{\lambda_{n,k}-\lambda_{n,j}}$$ And I was wondering: how do we simplify $\Gamma_{n,k}$ to ease the manual calculation of $\phi(n)$ values. The integral is always real, so I am sure there is a major way we can simplify $\Gamma_{n,k}$, but I have been so far unable to find it. I do suspect however that the product $$P_n=\prod_{k=0}^{n-1}\Gamma_{n,k}$$ May play a significant role in finding the simplification I seek.

For those interested, a proof.

Note that $x^n+1$ bay be factored as $$x^n+1=\prod_{k=0}^{n-1}(x-\lambda_{n,k})$$ Hence $$\phi(n)=\int_0^1\prod_{k=0}^{n-1}\frac1{x-\lambda_{n,k}}dx$$ Then define $\Gamma_{n,k}$ by saying that $$\prod_{k=0}^{n-1}\frac1{x-\lambda_{n,k}}=\sum_{k=0}^{n-1}\frac{\Gamma_{n,k}}{x-\lambda_{n,k}}$$ Multiplying both sides by $\prod_{j=0}^{n-1}(x-\lambda_{n,j})$: $$1=\sum_{k=0}^{n-1}\frac{\Gamma_{n,k}}{x-\lambda_{n,k}}\prod_{j=0}^{n-1}(x-\lambda_{n,j})$$ $$1=\sum_{k=0}^{n-1}\Gamma_{n,k}\prod_{k\neq j=0}^{n-1}(x-\lambda_{n,j})$$ So for any integer $0\leq m\leq n-1$ we may plug in $x=\lambda_{n,m}$ and simplify to get $$\Gamma_{n,m}=\prod_{m\neq j=0}^{n-1}\frac1{\lambda_{n,m}-\lambda_{n,j}}$$ And our result follows directly.

Perhaps another motivation for easing manual calculation of this product would be that $$\sum_{k=0}^{\infty}\frac{(-1)^k}{nk+1}=\phi(n)$$ Which brings about a plethora of interesting closed forms.

Edit: A little progress

We define $$c_{n,j}=\operatorname{Re}\lambda_{n,j}=\cos\frac{\pi(2j+1)}{n}$$ And $$s_{n,j}=\operatorname{Im}\lambda_{n,j}=\sin\frac{\pi(2j+1)}{n}$$ So $$\log\frac{\lambda_{n,k}-1}{\lambda_{n,k}}=\log\left(1-\lambda_{n,k}^{-1}\right)=\log\left(1-c_{n,k}+is_{n,k}\right)$$ And we also see that $$\begin{align} \prod_{k\neq j=0}^{n-1}\frac1{\lambda_{n,k}-\lambda_{n,j}}&=\prod_{k\neq j=0}^{n-1}\frac1{e^{i\pi(2k+1)/n}-e^{i\pi(2j+1)/n}}\\ &=\prod_{k\neq j=0}^{n-1}\frac{e^{-i\pi(2k+1)/n}}{1-e^{i\pi(2j-2k)/n}}\\ &=e^{i(2k+1)(2-n)/n}\prod_{k\neq j=0}^{n-1}\frac12\left(1+i\cot\frac{\pi(j-k)}n\right)\\ \Gamma_{n,k}&=\frac{\lambda_{n,k}^{2-n}}{2^{n-2}}\prod_{k\neq j=0}^{n-1}\left(1+i\cot\frac{\pi(j-k)}n\right) \end{align}$$ But the remaining product I do not know how to deal with.


Defining the polynomial \begin{align} P(x)&=x^n+1\\ &=\prod_{j=0}^{n-1}\left( x- \lambda_{n,j}\right) \end{align} we can express its derivative at $x=\lambda_{n,k}$ as: \begin{align} P'(\lambda_{n,k})&=\prod_{k\neq j=0}^{n-1}\left( \lambda_{n,k}-\lambda_{n,j} \right)\\ &=\frac{1}{\Gamma_{n,k}} \end{align} But we have also $P'(x)=nx^{n-1}=n\tfrac{x^n}{x}$. Thus, as $\left(\lambda_{n,k} \right)^n=-1$, \begin{equation} P'(\lambda_{n,k})=n\frac{-1}{\lambda_{n,k}} \end{equation} Finally, \begin{equation} \Gamma_{n,k}=-\frac{\lambda_{n,k}}{n} \end{equation} This trick comes rather naturally if the integral is evaluated by the residue method, for the function $f(z)=(1+z^n)^{-1}\ln\left(\tfrac z{1-z}\right)$ along the keyhole contour.

  • $\begingroup$ You, sir, are a genius. That was so elegant. Many thanks. $\endgroup$ – clathratus Mar 13 at 16:26
  • $\begingroup$ Too kind. Happy you like it! $\endgroup$ – Paul Enta Mar 13 at 16:28

In fact, if we may write a function $f$ as a product over its roots, i.e. $$f(x)\equiv\prod_{f(\omega)=0}(x-\omega)$$ where each root $\omega$ contributes exactly one term, then we may also write $$\frac1{f(x)}=\sum_{f(\omega)=0}\frac{b(\omega)}{x-\omega}$$ for some coefficients $b(\omega)$ which we can show to be $$b(\omega)=\prod_{{f(\alpha)=0}\atop{\alpha\ne \omega}}\frac1{\omega-\alpha}.$$ At the same time however, we have that the equality $$f'(x)=\sum_{f(\omega)=0}\frac{f(x)}{x-\omega}=\sum_{f(\omega)=0}\prod_{{f(\alpha)=0}\atop{\alpha\ne \omega}}(x-\alpha)$$ holds under the assumption that $f(x)=0\Rightarrow f'(x)\ne0$.

So for any root $\phi$ we plug in $x=\phi$ to see that $$f'(\phi)=\prod_{{f(\alpha)=0}\atop{\alpha\ne \phi}}(\phi-\alpha)$$ which implies that $$b(\omega)=\frac{1}{f'(\omega)}$$ and $$\frac1{f(x)}=\sum_{f(\omega)=0}\frac1{(x-\omega)f'(\omega)}.$$ Then @PaulEnta's results are easily derived form there.


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.