3
$\begingroup$

Calculate the following limit

$$\lim_{n\to\infty}\int_{0}^{\infty} \frac{\sqrt x}{1+ x^{2n}} dx$$

I tried to apply dominated convergence theorem but I could not find the dominating function. even I broke down the integration from $0$ to $1$ and $1$ to infinity. then found only integration from $0$ to $1$ is possible. Do you have any ideas?

$\endgroup$
  • $\begingroup$ Can you find a dominating function on $[1,+\infty)$? $\endgroup$ – Daniel Fischer Nov 13 '15 at 20:11
  • $\begingroup$ no I could not that my problem @DanielFischer $\endgroup$ – user145993 Nov 13 '15 at 20:19
  • 1
    $\begingroup$ Split the interval at $1$, treat both parts separately. Now, look at the integrand for $n = 1$. Then, what happens when $n$ increases? $\endgroup$ – Daniel Fischer Nov 13 '15 at 20:21
1
$\begingroup$

Hint: you only care about $n$ big and $x$ big.

$\endgroup$
  • $\begingroup$ Why don't we care about $n$ big and $x$ small? $\endgroup$ – mrf Nov 13 '15 at 22:17
  • $\begingroup$ Because, for the limit you only want to consider $n$ arbitrarily big; if you start with $n$ equal to ten trillions of trillions instead of $1$, the limit will not notice. Same with the integral: a continuous function will have a finite integral over any interval: so all you care if what happens for very big $x$. $\endgroup$ – Martin Argerami Nov 13 '15 at 23:11
  • $\begingroup$ That makes no sense. The integrand tends (sufficiently fast) to $0$ for $x > 1$ when $n\to\infty$. It's the contribution on $[0,1]$ that will matter. $\endgroup$ – mrf Nov 13 '15 at 23:20
  • $\begingroup$ It makes a lot of sense. Your question was about using dominated convergence and that you couldn't find the function. What I'm telling you, you only care to find a function for big $n$ and big $x$. For small $x$, you could take a constant for all that matters. You bounding function could be for instance $13\times\,1_{[0,10]}+\frac{\sqrt x}{1+x^4}\,1_{(10,\infty)}$. $\endgroup$ – Martin Argerami Nov 13 '15 at 23:44
3
$\begingroup$

A dominating function is $\sqrt x \cdot \chi_{[0,1]} + x^{-3/2}\cdot\chi_{[1,\infty)},$ which is in $L^1([0,\infty)).$ Since $f_n(x) \to \sqrt x \cdot\chi_{[0,1]}$ pointwise a.e., the DCT shows the limit is $\int_0^1\sqrt x \, dx = 2/3.$

$\endgroup$
  • $\begingroup$ this one is completely flawless. @ zhw $\endgroup$ – user145993 Nov 13 '15 at 20:59
3
$\begingroup$

We have

$$\int_0^\infty \frac{x^{1/2}}{1+x^n}\,dx=\int_0^1 \frac{x^{1/2}}{1+x^n}\,dx+\int_1^\infty \frac{x^{1/2}}{1+x^n}\,dx$$

For $0\le x\le 1$, we have

$$\frac{x^{1/2}}{1+x^n}\le x^{1/2}$$

For $1\le x$, we have for $n\ge 1$

$$\frac{x^{1/2}}{1+x^n}\le \frac{x^{1/2}}{1+x^2}$$

Therefore, the Dominated Convergence Theorem guarantees that

$$\begin{align} \lim_{n\to \infty}\int_0^\infty \frac{x^{1/2}}{1+x^n}\,dx&=\int_0^1 \lim_{n\to \infty}\frac{x^{1/2}}{1+x^n}\,dx+\int_1^\infty \lim_{n\to \infty} \frac{x^{1/2}}{1+x^n}\,dx\\\\ &=2/3+0\\\\ &=2/3 \end{align}$$

$\endgroup$
  • 2
    $\begingroup$ For $x < 1$, the integrands converge (monotonically) to $\sqrt{x}$, and $\int_0^1 \sqrt{x}\,dx = \frac{2}{3}$. $\endgroup$ – Daniel Fischer Nov 13 '15 at 20:58
  • $\begingroup$ @DanielFischer oh ... of course. Thank you for the catch! +1 $\endgroup$ – Mark Viola Nov 13 '15 at 21:01
1
$\begingroup$

No problem when $x$ is $0$ (check by yourself, the function is well defined for $x \geq 0$). So your function is integrable in any set $[0, a]$, with $a>0$.

When $x$ goes to infinity, then: $$\frac{\sqrt x}{(1+ x^{2n})} \sim \frac{x^\frac{1}{2}}{x^{2n}} = x^{\frac{1}{2} - 2n}$$

It is well known that the integral

$$\int_a^{+\infty}x^{-b}dx$$

converges when $b > 1$. In your case:

$$b = -\frac{1}{2} + 2n > 1 \Rightarrow n > \frac{3}{4}$$

(last part fixed thanks to the suggestion of Daniel Fischer)

$\endgroup$
  • $\begingroup$ very good way, I loved this @ the_candyman $\endgroup$ – user145993 Nov 13 '15 at 20:31
  • $\begingroup$ Sign error. $b = 2n-\frac{1}{2}$, so $n > \frac{3}{4}$. $\endgroup$ – Daniel Fischer Nov 13 '15 at 20:44
  • $\begingroup$ @DanielFischer thanks a lot, I fixed it. $\endgroup$ – the_candyman Nov 13 '15 at 20:59
1
$\begingroup$

Suppose we are interested in $$I = \int_0^\infty \frac{\sqrt{x}}{1+x^{2n}} dx$$

and evaluate it by integrating $$f(z) = \frac{\sqrt{z}}{1+z^{2n}} = \frac{\exp(1/2\log(z))}{1+z^{2n}}$$

around a keyhole contour with the branch cut of the logarithm on the positive real axis and the range of the argument of the logarithm from $0$ to $2\pi.$

Introduce $$\rho_k = \exp(\pi i/2/n + \pi i k/n)$$ where $0\le k\lt 2n$ to obtain

$$ (1-\exp(\pi i)) \times I = 2\pi i\sum_{k=0}^{2n-1} \mathrm{Res}_{z=\rho_k} f(z).$$

We get $$I = \pi i \sum_{k=0}^{2n-1} \frac{\exp(1/2\log(\rho_k))}{2n \rho_k^{2n-1}} = \pi i \sum_{k=0}^{2n-1} \rho_k \frac{\exp(1/2\log(\rho_k))}{2n \rho_k^{2n}} \\ = - \pi i \sum_{k=0}^{2n-1} \rho_k \frac{\exp(1/2\log(\rho_k))}{2n} \\ = -\frac{\pi i}{2n} \sum_{k=0}^{2n-1} \exp(\pi i/2/n + \pi i k/n) \exp(\pi i/4/n + \pi i k/2/n) \\ = -\frac{\pi i}{2n} \exp(3\pi i/4/n) \sum_{k=0}^{2n-1} \exp(3 \pi i k/2/n) \\ = -\frac{\pi i}{2n} \exp(3\pi i/4/n) \frac{\exp(2n \times 3\pi i/2/n)-1}{\exp(3\pi i/2/n)-1} \\ = \frac{\pi i}{n} \exp(3\pi i/4/n) \frac{1}{\exp(3\pi i/2/n)-1} \\ = \frac{\pi i}{n} \frac{1}{\exp(3\pi i/4/n)-\exp(-3\pi i/4/n)} \\ = \frac{\pi}{2n} \frac{1}{\sin(3\pi/4/n)}.$$

It is convenient to write this as $$\frac{2}{3} \frac{3\pi}{4n} \frac{1}{\sin(3\pi/4/n)}.$$

Now the Taylor series for $\frac{x}{\sin(x)}$ starts $$\left.\frac{x}{\sin(x)}\right|_{x=0} + \left.\left(\frac{x}{\sin(x)}\right)'\right|_{x=0} + \frac{1}{2}\left.\left(\frac{x}{\sin(x)}\right)''\right|_{x=0} \\ = 1 + \frac{1}{6} x^2 + \cdots$$

With $x=3\pi/4/n$ we get for the limit as $n\rightarrow\infty$ the value $$\frac{2}{3}.$$

Remark. The integral along the large circle vanishes because we have by the ML bound $$\lim_{R\rightarrow\infty } 2\pi R \frac{\sqrt{R}}{R^{2n}-1} \rightarrow 0$$

and along the small circle enclosing the origin we get applying ML a second time $$\lim_{\epsilon\rightarrow 0} 2\pi\epsilon \sqrt{\epsilon} \rightarrow 0.$$

Remark II. Actually we do not need the Taylor series of $\frac{x}{\sin(x)}$ as we can obtain the constant term with L'Hopital to get $$\lim_{x\rightarrow 0} \frac{x}{\sin(x)} = \lim_{x\rightarrow 0} \frac{1}{\cos(x)} = 1.$$

$\endgroup$

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.