1
$\begingroup$

I am trying to compute the following Integral $$ I = \int_{0}^\infty x \exp \left(-2 x \right) \operatorname{erf}\left(\frac{x}{t^{H}\sqrt[4]{2}}-\frac{t^H}{2^{3/4}}\right) \, dx $$ where $\operatorname{erf} \left(z\right)$ is the error function given by \begin{equation} \operatorname{erf}(z) = \frac{2}{\sqrt{\pi}}\int_0^z \mathrm{e}^{-t^2} \,dt . \end{equation}

Edit To provide some context to the problem I have a probability density function, which I am trying to verify that it is well behaved and integrates to 1. I am also interested in computing the expectation and the integral above is the part I am struggling to integrate. In order to reduce the noise, I only posted part of the formula I was struggling with. The below integral should work out to be 1, but I have not been able to establish that, as of yet. \begin{eqnarray} I &=& \int_0^{\infty}\mathrm{e}^{-2 x} \left(\Psi\left(x, t\right) + \frac{2^{3/4} \mathrm{e}^{x-\frac{t^{-2 H} \left(t^{4 H}+2 x^2\right)}{2 \sqrt{2}}}}{t^{H}\sqrt{\pi }}\right){\kern 1pt} \,dx \\ &=& \int_0^{\infty} \mathrm{e}^{-2 x}\Psi\left(x, t\right){\kern 1pt} \,dx + \frac{2^{3/4}}{t^{H}\sqrt{\pi }}\int_0^{\infty}\mathrm{e}^{-x-\frac{t^{-2 H} \left(t^{4 H}+2 x^2\right)}{2 \sqrt{2}}}{\kern 1pt} \,dx\\ &=& \int_0^{\infty} \mathrm{e}^{-2 x}\Psi\left(x, t\right){\kern 1pt} \,dx + \operatorname{erfc}\left(\frac{t^H}{2^{3/4}}\right)\\ \end{eqnarray} where \begin{equation} \Psi\left(x, t\right) = \operatorname{erfc} \left(\frac{x_0}{t^{H}\sqrt[4]{2}}-\frac{t^H}{2^{3/4}}\right), \end{equation}

Formulation 2 $$\lim_{x\to\infty} \frac{1}{2}\left(\Xi\left(x, t\right) - \mathrm{e}^{2 x} \Psi\left(x, t\right)+1\right) = 1$$ where $$\begin{equation} \Xi\left(x_0, t\right) = \operatorname{erf} \left(\frac{x_0}{t^{H}\sqrt[4]{2}}+\frac{t^H}{2^{3/4}} \right), \end{equation}$$ Any help would be much appreciated.

$\endgroup$
  • 1
    $\begingroup$ @Michael Hardy, Thank you very much for the edit. I found it educational. $\endgroup$ – Comic Book Guy May 6 '16 at 5:19
  • $\begingroup$ As $x \to -\infty$ the first factor inside the integral tends to $+\infty$ and the second to $-2$. This smells wrong to me. Isn't some bad sign somewhere? $\endgroup$ – leonbloy May 6 '16 at 12:44
  • $\begingroup$ @leonbloy actually I did find an error in my derivation a few steps, before I got to this integral. I have revised version. $\endgroup$ – Comic Book Guy May 6 '16 at 12:46
  • $\begingroup$ Did you try to change the order of integration? Actually, what did you try? $\endgroup$ – Did May 6 '16 at 12:56
  • $\begingroup$ @Did, I am not sure how to solve it in this case by order of integration. I do have another formulation, of my problem I will add, maybe that is easier. $\endgroup$ – Comic Book Guy May 6 '16 at 13:10
1
$\begingroup$

If $a=\frac1{t^H\sqrt[4]2}$, $b=\frac{t^H}{2^{3/4}}$, then we need $$\begin{align}\int_0^{\infty}xe^{-2x}\cdot\frac2{\sqrt{\pi}}\int_0^{ax-b}e^{-y^2}dy&=\left.\frac2{\sqrt{\pi}}\left(-\frac x2-\frac14\right)e^{-2x}\int_0^{ax-b}e^{-y^2}dy\right|_0^{\infty}\\ &+\frac a{\sqrt{\pi}}\int_0^{\infty}\left(x+\frac12\right)e^{-2x}e^{-(ax-b)^2}dx\\ &=\frac14\frac2{\sqrt{\pi}}\int_0^{-b}e^{-y^2}dy\\ &+\frac a{\sqrt{\pi}}e^{-\frac{2b}a+\frac1{a^2}}\int_0^{\infty}\left(x-\frac ba+\frac1{a^2}+\frac ba-\frac1{a^2}+\frac12\right)e^{-a^2\left(x-\frac ba+\frac1{a^2}\right)^2}dx\\ &=\frac14\text{erf}(-b)+\frac a{\sqrt{\pi}}e^{-\frac{2b}a+\frac1{a^2}}\int_{-\frac ba+\frac1{a^2}}^{\infty}\left(u+\frac ba-\frac1{a^2}\right)e^{-a^2u^2}du\\ &=\frac14\text{erf}(-b)+\frac a{\sqrt{\pi}}e^{-\frac{2b}a+\frac1{a^2}}\left\{\left.-\frac1{2a^2}e^{-a^2u^2}\right|_{-\frac ba+\frac1{a^2}}^{\infty}+\frac1a\left(\frac ba-\frac1{a^2}\right)\int_{-b+\frac1a}^{\infty}e^{-v^2}dv\right\}\\ &=\frac14\text{erf}(-b)+e^{-\frac{2b}a+\frac1{a^2}}\left\{\frac1{2a\sqrt{\pi}}e^{-\left(b-\frac1a\right)^2}+\frac12\left(\frac ba-\frac1{a^2}\right)\text{erfc}\left(-b+\frac1a\right)\right\}\end{align}$$ Is that the kind of stuff you're looking for?

EDIT: I should have gone a little farther since $ab=\frac12$. Then $\frac1a=2b$, so we can write this result as $$\int_0^{\infty}xe^{-2x}\cdot\frac2{\sqrt{\pi}}\int_0^{ax-b}e^{-y^2}dy= \frac14\text{erf}(-b)+\frac1{2a\sqrt{\pi}}e^{-b^2}-b^2\text{erfc}(b)$$ Also by popular demand, $$\begin{align}\int_0^{\infty}e^{-2x}\text{erfc}(ax-b)dx+\text{erfc}(b)&= \text{erfc}(b)+\frac2{\sqrt{\pi}}\int_0^{\infty}e^{-2x}\int_{ax-b}^{\infty}e^{-y^2}dy\\ &=\text{erfc}(b)+\frac2{\sqrt{\pi}}\left\{\left.-\frac12e^{-2x}\int_{ax-b}^{\infty}e^{-y^2}dy\right|_0^{\infty}-\frac a2\int_0^{\infty}e^{-2x}e^{-(ax-b)^2}dx\right\}\\ &=\text{erfc}(b)+\frac12\text{erfc}(-b)-\frac a{\sqrt{\pi}}e^{-\frac{2b}a+\frac1{a^2}}\int_0^{\infty}e^{-a^2\left(x-\frac ba+\frac1{a^2}\right)^2}dx\\ &=\text{erfc}(b)+\frac12\text{erfc}(-b)-\frac a{\sqrt{\pi}}e^{-\frac{2b}a+\frac1{a^2}}\int_{-\frac ba+\frac1{a^2}}^{\infty}e^{-a^2u^2}du\\ &=\text{erfc}(b)+\frac12\text{erfc}(-b)-\frac1{\sqrt{\pi}}e^{-\frac{2b}a+\frac1{a^2}}\int_{-b+\frac1{a}}^{\infty}e^{-v^2}vu\\ &=\text{erfc}(b)+\frac12\text{erfc}(-b)-\frac12e^{-\frac{2b}a+\frac1{a^2}}\text{erfc}\left(-b+\frac1a\right)\\ &=\text{erfc}(b)+\frac12\text{erfc}(-b)-\frac12\text{erfc}\left(b\right)\\ &=\frac12\text{erfc}(b)+\frac12\text{erfc}(-b)\\ &=\frac1{\sqrt{\pi}}\int_b^{\infty}e^{-y^2}dy+\frac1{\sqrt{\pi}}\int_{-b}^{\infty}e^{-y^2}dy\\ &=\frac1{\sqrt{\pi}}\int_b^{\infty}e^{-v^2}dv-\frac1{\sqrt{\pi}}\int_b^{-\infty}e^{-v^2}dv\\ &=\frac1{\sqrt{\pi}}\int_b^{\infty}e^{-v^2}dv+\frac1{\sqrt{\pi}}\int_{-\infty}^be^{-v^2}dv\\ &=\frac1{\sqrt{\pi}}\int_{-\infty}^{\infty}e^{-v^2}dv\\ &=\frac1{\sqrt{\pi}}(2)\left(\frac12\right)\Gamma\left(\frac12\right)=1\end{align}$$ Verification complete, but I noticed that my error rate was kind of high in the above. Please check and let me know of any further edits required.

$\endgroup$
  • $\begingroup$ Thank you, Did you also have any thoughts about the verification of the density part ? $\endgroup$ – Comic Book Guy May 6 '16 at 14:48
  • $\begingroup$ OK, I'll edit some more junk in there... $\endgroup$ – user5713492 May 6 '16 at 15:10
  • $\begingroup$ Thank you very much for your help. I will verify your calculations and accept in the near future $\endgroup$ – Comic Book Guy May 7 '16 at 0:18
  • $\begingroup$ Firstly there are some amazing steps here. really hats off to your skills. I do have one question though in the density verification steps, Half way down there is a step which combines the exp and erfc, to get a simplified erfc term. I am not able to follow how that is determined. Can u please help me with that. $\endgroup$ – Comic Book Guy May 7 '16 at 2:33
  • $\begingroup$ Are you referring to the step $$-\frac12e^{-\frac{2b}a+\frac1{a^2}}\text{erfc}\left(-b+\frac1a\right)=-\frac12\text{erfc}\left(b\right)$$ That step is actually a no-op given that $\frac1a=2b$: $$e^{-\frac{2b}a+\frac1{a^2}}=e^{-4b^2+4b^2}=e^0=1$$ and $$\text{erfc}\left(-b+\frac1a\right)=\text{erfc}(-b+2b)=\text{erfc}(b)$$ $\endgroup$ – user5713492 May 7 '16 at 2:41

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.