I'm trying to solve the following problem

$$\text{Evaluate:} \lim_{b \to 1^+} \int_1^b \frac{dx}{\sqrt{x(x-1)(b-x)}}$$

I'm tempted to think it's 0 because the bounds would be about equal, but that's not the correct answer. But I don't know what integration techniques to use, so any hints are greatly appreciated.

I don't know how to solve for the indefinite integral itself.

  • 1
    $\begingroup$ Start with $x=\cos^2y+b\sin^2y$ $\endgroup$ – lab bhattacharjee Jul 16 '15 at 12:56
  • $\begingroup$ I simplified the integral to $\lim_{b \to 1^+} \int_{x=1}^{x=b} \frac{2}{\sqrt{\cos^2{y}+b\sin^2{y}}}$ How do I change the bounds? $\endgroup$ – Vishwa Iyer Jul 16 '15 at 13:02
  • $\begingroup$ @labbhattacharjee Just as a general question, is it possible to solve the integral using partial fractions? Or is that only applicable if there is no root present? $\endgroup$ – Gummy bears Jul 16 '15 at 13:05
  • $\begingroup$ @VishwaIyer Is that $y^2$ or is it $ycos^2y$? $\endgroup$ – Gummy bears Jul 16 '15 at 13:06
  • $\begingroup$ It's $\cos^2{y}$. I was having trouble with the formatting. $\endgroup$ – Vishwa Iyer Jul 16 '15 at 13:07

We can solve this using complex analysis: The function $x(x-1)(b-x)$ has a holomorphic square root $f$ on $\mathbb{C}\setminus\left((-\infty,0]\cup[1,b]\right)$ because every loop in this domain circles two zeroes of $x(x-1)(b-x)$ - more concretely, for $(r,\phi)\in (0,\infty)\times (-\pi,\pi)$, set $g(re^{i\phi}) = \sqrt{r}e^\frac{i\phi}{2}$. Then $g$ is the holomorphic extension of $\sqrt\cdot$ to $\mathbb{C}\setminus(-\infty,0]$, and $f(z)$ is the unique holomorphic extension of $ig(z)g(z-1)g(z-b)$ to the domain given above. Indeed, we can check directly that for $x\in (0,1)$, we have $\lim_{\epsilon\to 0} f(x\pm i\epsilon) = i\sqrt{x(1-x)(b-x)}$, so that we can extend $f$ to this interval. On the other hand, for $x\in [1,b]$, we have $\lim_{\epsilon\to 0} f(x\pm i\epsilon) = \mp\sqrt{x(x-1)(b-x)}$. Thus, we have $\int_1^b \frac{\mathrm dx}{\sqrt{x(x-1)(b-x)}} = \frac{1}{2}\lim_{\epsilon\to 0}\int_{1-\epsilon}^{b-\epsilon} f(z)\mathrm dz + \int_{b+\epsilon}^{1+\epsilon} f(z)\mathrm dz$, and in the limit we can close this path by vertical edges to get $\int_1^b \frac{\mathrm dx}{\sqrt{x(x-1)(b-x)}} = \frac{1}{2} \oint_\gamma f(z)\mathrm d z$ where $\gamma$ is any path which circles the "slit" $[1,b]$ once in counterclockwise direction. For $b$ small, we can for example take the circle of radius $\frac{1}{2}$ around $1$. Then the path of integration no longer depends on $b$, and by dominated convergence we can exchange the limit with integration to see that $ \lim_{b\to 0}\int_1^b \frac{\mathrm dx}{\sqrt{x(x-1)(b-x)}} = \frac{1}{2}\oint_\gamma \frac{\mathrm d z}{g(z)i(z-1)}$ and by the Cauchy integral formula this is $\frac{2\pi i}{2ig(1)} = \pi$.

  • $\begingroup$ This is impressive. $\endgroup$ – Axel Jul 17 '15 at 7:25

First you can notice that the $\frac{1}{\sqrt{x}}$ term is between $1$ and $\frac{1}{\sqrt{b}}$, so there is no need to bother about it. Indeed this approximation will yield a relative mistake of at most $\sqrt{b}$.


$$ \frac{1}{\sqrt{b}}I = \frac{1}{\sqrt{b}} \int_1^b \frac{dx}{\sqrt{(x-1)(b-x)}} \leq \int_1^b \frac{dx}{\sqrt{x(x-1)(b-x)}} \leq \int_1^b \frac{dx}{\sqrt{(x-1)(b-x)}} = I . $$

Then you can see by an affine change of variable that the remaining integral $I$ doesn't depend on $b$.

All you need to do then is a simple change of variable to show that the answer is $\pi$. Can you try or do you want me to spell it out?

  • $\begingroup$ i think you will not reglect the first term. Instead you approximate it as one. Which is indeed a very good approximation. Just to be totally correct...Otherwise nice (+1) $\endgroup$ – tired Jul 16 '15 at 13:27
  • $\begingroup$ Ok, I'll edit to explain. $\endgroup$ – Axel Jul 16 '15 at 13:28
  • $\begingroup$ What change of variable do you use? $\endgroup$ – Vishwa Iyer Jul 16 '15 at 14:07
  • $\begingroup$ $x = 1 + (b-1)u$ then $u=\sin^2(t)$. $\endgroup$ – Axel Jul 16 '15 at 14:25

Hint: Let $~b=1+\epsilon,~$ where $~\epsilon\to0^+,~$ and $~x=1+t,~$ where $~t\in(0,\epsilon).~$ Then let $~t=\epsilon~u.$


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.