I am trying to evaluate the integral $$ I_k(x)=\int_1^x \log^k t \frac{\sqrt{t-1}}{t^2} dt $$ with $x$ tending to infinity.

In fact, I wish to have an estimate $$ \sum_{k=0}^\infty \frac{1}{\log^k x} \int_1^x \log^k t \frac{\sqrt{t-1}}{t^2} dt = C+E(x), $$ where $C$ is explicit and $E(x) \to 0$ as $x \to \infty$.

The integral inside may be actually rewritten as $$ I_k(x) = \int_1^x \log^k t \sqrt{1-\frac{1}{t}} \frac{dt}{t^{3/2}} = c_k \int_{1/x^2}^1 \sqrt{1-t^2} \log^k t dt, $$ hence can be possibly attacked via multiple zeta values.

Is that possible?


First of all, your series diverges for any $x>1$, because we always have

$$I_k(x)\geq \int_{\sqrt{x}}^x \log^k t\sqrt{t-1}t^{-2}dt\geq \frac{\sqrt{\sqrt{x}-1}}{x}\int_{\sqrt{x}}^{x} \frac{\log^k t}{t} dt=\frac{\sqrt{\sqrt{x}-1}}{x}\frac{(\log^{k+1} x)(1-2^{-k-1})}{k+1}.$$

If you want to get the asymptotics for $I_k(x)$ when $k$ is fixed, then using the substitution $t=1/u$ we get

$$I_k(x)=\int_{1/x}^1 (-1)^k\log^k u \sqrt{(1-u)/u}du=(-1)^k(j_k+R_k(x)),$$


$$j_k=\int_0^1 u^{-1/2}(1-u)^{1/2}\log^k u du$$


$$R_k(x)=\int_0^{1/x} u^{-1/2}(1-u)^{1/2}\log^k u du=O_k\left(\frac{\log^{k+1} x}{x^{3/2}}\right).$$

Now lets compute $j_k$ in terms of the derivatives of Gamma-function. By the beta-integral, we have for any $s>-1$:

$$\int_0^1 u^s(1-u)^{1/2}du=B(s+1,3/2)=\frac{\sqrt{\pi}\Gamma(s+1)}{2 \Gamma(s+5/2)}.$$

Now differentiate the lhs of this equality $k$ times and set $s=-1/2$. Using $(u^s)'=u^s\log u$, we conclude that

$$j_k=\left(\frac{\partial^k}{\partial s^k} \left. \frac{\sqrt{\pi}\Gamma(s+1)}{2 \Gamma(s+5/2)}\right)\right|_{s=-1/2}.$$

  • $\begingroup$ The question arises: is $s=\frac 1 2$ small enough? $\endgroup$
    – user64494
    May 12 '18 at 14:40
  • $\begingroup$ In fact, the equality holds for $\mathrm{Re}\,s>-1$. So, yes, it is) $\endgroup$ May 12 '18 at 15:04
  • $\begingroup$ Is there a mistake in the change of variables $t=1/u$ or am I being silly? $\endgroup$
    – toshi
    May 12 '18 at 15:26
  • $\begingroup$ Numerical calculations do not confirm it: $j_2=.5204777983$, whereas NIntegrate[Log[t]^2*Sqrt[t - 1]/t^2, {t, 1, 50}] performs $4.6862 $. $\endgroup$
    – user64494
    May 12 '18 at 15:26
  • $\begingroup$ @toshi, yes there is a mistake, but I hope now everything is fine. There should be $-1/2$ instead of $1/2$ $\endgroup$ May 12 '18 at 16:14

In view of the output of the Mathematica code

Table[Series[Integrate[Log[t]^k*Sqrt[-t^2+1],{t, 1/x^2, 1},Assumptions->x> 1],{x,Infinity,2}],{k,1,3}]

$$O\left(\left(\frac{1}{x}\right)^3\right)+\frac{2 \log (x)+1}{x^2}+\left(-\frac{\pi }{8}-\frac{1}{8} \pi \log (4)\right), $$ $$\frac{1}{48} \left(\pi ^3+6 \pi +3 \pi \log ^2(4)+3 \pi \log (16)\right)+\frac{-4 \log ^2(x)-4 \log (x)-2}{x^2}+O\left(\left(\frac{1}{x}\right)^3\right), $$ $$\frac{1}{32} \left(-12 \pi \zeta (3)-\pi ^3-6 \pi -\pi \log ^3(4)-3 \pi \log ^2(4)-\pi ^3 \log (4)-\pi \log (4096)\right)+\frac{8 \log ^3(x)+12 \log ^2(x)+12 \log (x)+6}{x^2}+O\left(\left(\frac{1}{x}\right)^3\right) $$ I have doubts concerning simple asymptotics for $I_k(x)$.


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.