Given the integral $(1)$

$$\int_{0}^{\infty}e^{-x}\sin^n x\left({1\over x}+{1\over x^2}+{1\over x^3}+\cdots+{1\over x^n}\right)dx =\arctan(n)\tag1$$

An attempt:

Rewrite $(1)$ as

$$\sum_{j=1}^{n}\int_{0}^{\infty}e^{-x}\sin^n x\cdot{\mathrm dx\over x^j}\tag2$$

Maybe if we know the closed form for $(3)$, it could be useful

$$\sum_{j=1}^{n}{1\over x^j}?\tag3$$

Employing $e^{-x}$ series $(2)$ becomes

$$\sum_{j=1}^{n}\int_{0}^{\infty}\sin^n x\left(x^{-j}-x^{1-j}+{x^{2-j}\over 2!}-{x^{3-j}\over 3!}+\cdots \right)\mathrm dx\tag4$$

I guess we may use integration by parts to integrate term by term, it may results in a lengthy process.

How else may integral $(1)$ be prove?

  • 1
    $\begingroup$ $\sin^nx\left({1\over x}+{1\over x^2}+\cdots+{1\over x^n}\right)=\left(\sin x\over x\right)^n\left(x^n-1\over x-1\right)$, but I'm not sure how much that helps. $\endgroup$ – Barry Cipra Mar 21 '17 at 16:50
  • 1
    $\begingroup$ For $n=1$, $n=2$ and $n=3$ the equality seems to be true, but for $n=4$ the integral equals (according to Mathematica) $$\frac{1}{48} \left(\log \left(\frac{625}{17}\right)-8 \arctan(2)+52 \arctan(4)\right)\approx 1.32687$$ which is not equal to $$\arctan(4)\approx 1.32582.$$ $\endgroup$ – mickep Mar 21 '17 at 17:43
  • $\begingroup$ Through the Laplace transform it is straightforward to check that the given identity cannot hold for any even $n\geq 4$. We have a simple pole that contributes with a logarithmic term in the closed form. $\endgroup$ – Jack D'Aurizio Mar 21 '17 at 18:12

The inverse Laplace transform of $e^{-x}\left(\frac{1}{x}+\ldots+\frac{1}{x^n}\right)$ is given by $$ \mathbb{1}_{s\geq 1}(s)\sum_{k=0}^{n-1}\frac{(s-1)^k}{k!} $$ and the Laplace transform of $\sin^n(x)$ is given by $$ \frac{n!}{s(4+s^2)(16+s^2)\cdots(n^2+s^2)} $$ if $n$ is even and by $$ \frac{n!}{(1+s^2)(9+s^2)\cdots (n^2+s^2)} $$ if $n$ is odd. The given integral boils down to the integral of a rational function over $(1,+\infty)$,
but for $n=4$

$$ \int_{1}^{+\infty}\frac{4!}{s(4+s^2)(16+s^2)}\sum_{k=0}^{3}\frac{(s-1)^k}{k!}\,ds \neq \arctan(4)$$ since the LHS depends on $\log\frac{625}{17}$. Anyway, through the previous representation we may simply compute the given integral through partial fraction decomposition. The given identity holds for $n\in\{1,2,3\}$, but this problem rightfully belongs to the Mathematical coincidence Wikipedia page.

  • 3
    $\begingroup$ nice counterexample (+1). it is strange that OP didn't check his conjecture for a few more values by a CAS $\endgroup$ – tired Mar 21 '17 at 18:17

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.