3

I would like to find an equivalent (first term of an asymptotic expansion) of the following function in $x=0$ and $x=1$:

$$f(x)=\int_0^{+\infty} \frac{1}{t^x\sqrt{1+t^2}}\ dt$$

My professor gave a highly unmotivated solution which doesn't invoke the usual tools to find asympto tic expansions of integrals depending on some parameter.

I would be interested in an elementary, but motivated solution.

math_lover
  • 5,826
  • 1
    There is a difference between $\displaystyle \int_0^{+\infty} \frac 1 {t^x \sqrt{1+t^2}} , dt$ and $\displaystyle \int_0^{+\infty} \frac 1 {t^x \sqrt{1+t^2}} , dx.$ – Michael Hardy Dec 20 '17 at 00:03

3 Answers3

4

Herein we present a way forward, which relies on elementary analysis only, to derive the first three terms of the "small $x$" expansion of the function of interest. To that end, we now proceed.


For $x>0$, we have

$$\begin{align} f(x)&=\int_0^\infty \frac{e^{-x\log(t)}}{\sqrt{t^2+1}}\,dt\\\\ &=\int_0^1 \frac{e^{-x\log(t)}}{\sqrt{t^2+1}}\,dt+\int_1^\infty \frac{e^{-x\log(t)}}{\sqrt{t^2+1}}\,dt\\\\ &=\int_0^1 \frac{e^{-x\log(t)}}{\sqrt{t^2+1}}\,dt+\int_0^1 \frac{e^{x\log(t)}}{t\sqrt{t^2+1}}\,dt\tag1 \end{align}$$


The first term on the right-hand side of $(1)$ is well behaved near $x=0^+$ with

$$\begin{align} \int_0^1 \frac{e^{-x\log(t)}}{\sqrt{t^2+1}}\,dt&=\text{arsinh}(1)-x\int_0^1 \frac{\log(t)}{\sqrt{1+t^2}}\,dt+O(x^2)\\\\ &=\text{arsinh}(1)-x\int_0^{\pi/4} \log(\tan(\phi))\,\sec(\phi)\,d\phi+O(x^2)\tag2 \end{align}$$


The second term on the right-hand side of $(1)$ can be written as

$$\begin{align} \int_0^1 \frac{e^{x\log(t)}}{t\sqrt{t^2+1}}\,dt&=\int_0^1 \frac{e^{x\log(t)}}{t}\,dt-\int_0^1 \frac{e^{x\log(t)}}{t}\left(1-\left(1+t^2\right)^{-1/2}\right)\,dt\\\\ &=\frac1x -\int_0^1 \frac{1-\left(1+t^2\right)^{-1/2}}{t}\,dt-x\int_0^1 \frac{(1-\left(1+t^2\right)^{-1/2})\,\log(t)}{t}\,dt+O(x^2)\\\\ &=\frac1x -\text{arsinh}(1)+\log(2)-x\int_0^{\pi/4}\log(\tan(\phi))\,\sec(\phi)\frac{1-\cos(\phi)}{\sin(\phi)}\,d\phi+O(x^2)\tag3 \end{align}$$


Substituting $(2)$ and $(3)$ into $(1)$ reveals

$$\begin{align} f(x)&=\frac1x +\log(2) -x\int_0^{\pi/4}\log(\tan(\phi))\,\sec(\phi)\left(1+\frac{1-\cos(\phi)}{\sin(\phi)}\right)\,d\phi+O(x^2)\\\\ &=\frac1x +\log(2) +\frac1{12}\left(\pi^2+6\log^2(2)\right)\,x+O(x^2) \end{align}$$

Mark Viola
  • 179,405
0

This looks like a rather tedious calculation, so resorting to computer algebra we find (for $0 < x < 1$):

$f(x) = \frac{\Gamma \left(\frac{1}{2}-\frac{x}{2}\right) \Gamma \left(\frac{x}{2}\right)}{2 \sqrt{\pi }}$.

f(x)

Notice that it is symmetric with respect to $x = 1/2$.

The first three terms in the expansion around $x=0$ are:

$f(x) \approx \frac{1}{x}+\log (2)+\frac{1}{12} x \left(\pi ^2+6 \log ^2(2)\right)+\frac{1}{12} x^2 \left(3 \zeta (3)+2 \log ^3(2)+\pi ^2 \log (2)\right)+O\left(x^3\right)$

0

$f$ is defined in $(0,1)$, and it's symmetric with respect to $x={1\over2}$ since \begin{align} f(x)&=\int_0^{1} \frac{1}{t^x\sqrt{1+t^2}}dt+\int_1^{+\infty} \frac{1}{t^x\sqrt{1+t^2}}dt\\ &=\int_0^{1} \frac{1}{t^x\sqrt{1+t^2}}dt+\int_0^{1} \frac{1}{s^{1-x}\sqrt{1+s^2}}ds\\ \end{align} thus we can study on only one side, here we choose $x\to0^+$, then we have \begin{align} f(x)&=\int_0^{1} \frac{1}{s^{1-x}\sqrt{1+s^2}}ds+\operatorname{arsinh}(1)+o(1)\\ &=[{s^x\over x}\frac{1}{\sqrt{1+s^2}}]_0^1+{1\over x}\int_0^{1}\frac{s^{x+1}}{(1+s^2)^{3/2}}ds+\ln(1+\sqrt{2})+o(1)\\ &={1\over \sqrt{2}x}+{1\over x}(1-{1\over \sqrt{2}})+o({1\over x})\\ &={1\over x}+o({1\over x}) \end{align}

Aforest
  • 2,665
  • What does it mean, ${1 \over x} + o \left( {1 \over x} \right)$? Doesn't make sense.... – David G. Stork Dec 20 '17 at 01:23
  • @DavidG.Stork You know little-o notation? It means $\forall\epsilon>0,\exists k\in(0,1), \forall x \in (0,k), |f(x)-{1\over x}|\le\epsilon|{1\over x}|$... – Aforest Dec 20 '17 at 01:32
  • Yes of course I know the ${\cal o}$ notation. But you have the same functional form inside and outside of the order. What if an answer were $g(x) = x + {\cal o}(x)$? How does this differ from $g(x) = {\cal o}(x)$? – David G. Stork Dec 20 '17 at 01:36
  • @DavidG.Stork In my answer if I write $f(x)=o({1\over x})$, it means $\forall\epsilon>0,\exists k\in(0,1), \forall x \in (0,k), |f(x)|\le\epsilon|{1\over x}|$, which is different from above.... – Aforest Dec 20 '17 at 01:40
  • @DavidG.Stork What they said makes sense... it's $1/x$ plus terms strictly less important than $1/x$ as $x\to0.$ i.e. agrees with your answer. $1/x + O(1)$ would be a more precise answer, and yours below is preciser still. With big O, $1/x + O(1/x)$ wouldn't differ from $O(1/x).$ – spaceisdarkgreen Dec 20 '17 at 01:41
  • OK. Thanks. A bit unusual, though, at least compared to the use of ${\cal O}(x)$. – David G. Stork Dec 20 '17 at 01:44