Let's consider $\displaystyle f(x)=\sum_{n=1}^\infty\frac{x^n}{\sqrt{n}}$.

Where $f$ is defined, can we find a closed form for $f(x)$ ?

What would be an equivalent of $f$ near $1^-$ ?

  • 1
    $\begingroup$ This question is related, and the method proposed in my answer can be applied here. $\endgroup$ – Antonio Vargas Jan 28 '15 at 22:29
  • 1
    $\begingroup$ Such a function is just $\operatorname{Li}_{\frac{1}{2}}(x)$. Near $x=1^-$ its asymptotic behaviour is given by $\zeta\left(\frac{1}{2}\right)+\sqrt{\frac{\pi}{1-x}}.$ $\endgroup$ – Jack D'Aurizio Jan 28 '15 at 22:37
  • $\begingroup$ @AntonioVargas How would you compute the related integral, then ? $\endgroup$ – Hippalectryon Jan 28 '15 at 23:17
  • $\begingroup$ I asked Mathematica, and it said $$\int_0^\infty \frac{x^n}{\sqrt{n}}\,dn = \sqrt{\frac{\pi}{-\log x}}$$ :) $\endgroup$ – Antonio Vargas Jan 28 '15 at 23:43
  • $\begingroup$ @AntonioVargas I don't know how to show that result, but I'd love it if you could show me how if you have some time to write an answer :) $\endgroup$ – Hippalectryon Jan 28 '15 at 23:45

If we write $x^n = \exp(n\log x)$, then set $u = -n\log x$, the integral mentioned in the comments becomes

$$ \int_0^\infty \frac{x^n}{\sqrt{n}}\,dn = \frac{1}{\sqrt{-\log x}} \int_0^\infty \frac{e^{-u}}{\sqrt{u}}\,du = \frac{\Gamma(1/2)}{\sqrt{-\log x}} = \sqrt{\frac{\pi}{-\log x}}. $$

Then since $-\log x \sim 1-x$ as $x \to 1$ and the terms of the sum are strictly decreasing for $0 < x < 1$, we may conclude that

$$ \sum_{n=1}^{\infty} \frac{x^n}{\sqrt{n}} \sim \int_0^\infty \frac{x^n}{\sqrt{n}}\,dn = \sqrt{\frac{\pi}{-\log x}} \sim \sqrt{\frac{\pi}{1-x}} $$

as $x \to 1^-$.

  • $\begingroup$ Your answer seems fine, but do you know why Jack D'Aurizio has $\zeta\left(\frac{1}{2}\right)+\sqrt{\frac{\pi}{1-x}}$ as an equivalent ? (Btw the sum starts at $1$ and not $0$, shouldn't the integral start at $1$ too?) $\endgroup$ – Hippalectryon Jan 29 '15 at 0:03
  • $\begingroup$ The term $\zeta(1/2)$ is the "next" term in the approximation and is small compared to the $\sqrt{\cdots}$ term, which is dominant. Both $\sqrt{\cdots}$ and $\sqrt{\cdots} + \zeta(1/2)$ are equivalents to your function, since equivalence is only determined by leading-order behavior. The integral starts at $0$ for simplicity; the difference accrued by it starting at $1$ would be small compared to the dominant contribution of the tail. $\endgroup$ – Antonio Vargas Jan 29 '15 at 0:31
  • $\begingroup$ There's definitely a bit to unwrap in my answer. Trying to prove each $\sim$ yourself would help with understanding. $\endgroup$ – Antonio Vargas Jan 29 '15 at 3:09

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.