0

I am currently working on the following improper integral:

Let $\sigma>0$ and $H>1$. I would like to show that $$ \int_0^{\infty} \frac{\frac{\ln H}{\sigma}\exp \left( \frac{- \left( \large \frac{\ln H}{\sigma}+ \frac{t \sigma}{2} \right)^2}{2t} \right)}{\sqrt{2 \pi t^3}} \,dt = \frac{1}{H}.$$

This is the result indicated in my book and also on Wolfram Alpha, but I do not know the substitution required to evaluate this. Any ideas?

Zacky
  • 27,674
Richard
  • 3,020
  • I am not sure that an antiderivative could be found (even with special functions). Effectively $$\int_0^\infty \frac{e^{-\frac{(a+b t)^2}{t}}}{t^{3/2}},dt=\frac{\sqrt{\pi } e^{-4 a b}}{a}$$ – Claude Leibovici Sep 29 '19 at 06:37
  • @ClaudeLeibovici Thanks. May you please explain how to get this formula? Contour integration? I still cannot derive this.... – Richard Sep 29 '19 at 06:41
  • No idea and I am sorry for that. This is a result given by a CAS. – Claude Leibovici Sep 29 '19 at 06:44

2 Answers2

8

$$I(a,b)=\int_0^\infty \exp{\left(-\frac{(a+bx)^2}{x}\right)}\frac{dx}{\sqrt{x^3}}\overset{x\to x^2}=2\int_0^\infty \exp{\left(-\frac{(a+bx^2)^2}{x^2}\right)}\frac{dx}{x^2}$$ $$\overset{\large \frac{1}{x}\to x}=\color{blue}{2\int_0^\infty \exp{\left(-{(ax+b/x)^2}\right)}dx}\overset{\large x\to \frac{b}{ax}}=\color{red}{\frac{2b}{a}\int_0^\infty \exp\left(-(ax+b/x)^2\right)\frac{dx}{x^2}}$$ $$\color{purple}{2I(a,b)}=\color{purple}{2\int_0^\infty \exp\left(-(ax+b/x)^2\right)}\left(\color{blue}{1}+\color{red}{\frac{b}{ax^2}}\right)dx$$

It would have been perfect if we had $(ax-b/x)$ instead of $(ax+b/x)$ since it's derivative would be found in $\left(1+b/(ax^2)\right)$, but we can adjust things:

$$(ax+b/x)^2=a^2 x^2+b^2/x^2+2ab$$ $$ (ax-b/x)^2=a^2x^2+b^2/x^2-2ab$$ $$\Rightarrow (ax+b/x)^2=(ax-b/x)^2+4ab$$

$$\Rightarrow I(a,b)=\frac{e^{-4ab}}{a}\int_0^\infty \exp\left(-(ax-b/x)^2\right)\left(a+\frac{b}{x^2}\right)dx$$ $$\overset{ax-b/x=t}=\frac{e^{-4ab}}{a}\int_{-\infty}^\infty e^{-t^2}dt=\frac{\sqrt{\pi}e^{-4ab}}{a}$$

Zacky
  • 27,674
  • 1
    This is really nice ! Thanks for posting such a beautiful anser and $+1$ for sure. – Claude Leibovici Sep 29 '19 at 08:57
  • Thank you Claude! Regarding the comments from bellow the question, wolfram can produce a closed form in terms of the error function for $\int \exp{\left(-{(ax+b/x)^2}\right)}dx$ so the original integral must have an antiderivative too. Not sure why it didn't spit one out initially. – Zacky Sep 29 '19 at 09:02
  • 1
    I got it thanks to your work. Cheers :-) – Claude Leibovici Sep 29 '19 at 09:24
  • At the price of a bunch of $\sqrt{a^2}$ and $\sqrt{b^2}$ we can cover all situations but $(a,b>0)$ give the simplest formula. – Claude Leibovici Sep 29 '19 at 12:49
  • @Zacky I'm trying to follow your solution when the limits of integration are from $0$ to $T$ (as for a CDF). When I keep track of changing the limits of integration with a new change of variables, I get different limits for the blue and for the red integral so I wouldn't be able to add them. Is this correct or am I doing something wrong? – dleal Mar 03 '20 at 01:42
  • 1
    @dleal you're not doing anything wrong, in this case this works nice because the limits are invariant under the substitution $x\to \frac{1}{x}$. For your case the result won't be pretty (check the answer from bellow involving the error function, then it can be proved by differentiating both sides). – Zacky Mar 03 '20 at 14:11
1

This is not an answer.

After カカロット's elegant solution, assuming $a>0$ and $b>0$, we can find $$\int \frac{e^{-\frac{(a+b x)^2}{x}}}{x^{3/2}} \, dx=\frac{\sqrt{\pi } e^{-4 a b}}{2 a} \left(\text{erfc}\left(\frac{a-b x}{\sqrt{x}}\right)+e^{4 a b} \text{erfc}\left(\frac{a+b x}{\sqrt{x}}\right)-2 \right)$$

Claude Leibovici
  • 260,315
  • is this based of Zacky's answer? I'm not quite clear why Zacky's answer would lead to 2 error functions, based on the last line of that other answer. – dleal Mar 03 '20 at 01:46