5
$\begingroup$

For $\sigma>0$, how can we prove that $$\frac{1}{2}\int_{-1}^1 \text{erf}\left(\frac{\sigma}{\sqrt{2}}+\text{erf}^{-1}(x)\right) \, \mathrm{d}x= \text{erf}\left(\frac{\sigma}{2}\right)$$ where erf is the error function, $$\text{erf}(x)=\frac{2}{\sqrt{\pi}}\int_{0}^x e^{-t^2}dt.$$ The result was obtained by tinkering, and I was wondering if there a concise derivation.

$\endgroup$
  • 1
    $\begingroup$ First thing that comes to mind is to change variable (y = invert of erf). Then look at the integral as a convolution of erf and its derivative, then use convolution theorem. Overall it requires around 5 steps. $\endgroup$ – user227136 Mar 28 '15 at 22:36
8
$\begingroup$

Let $u=\text{erf}^{-1}(x)$ so that $x=\text{erf}\left(u\right)$, and $dx=\frac{2}{\sqrt{\pi}}e^{-u^{2}}du.$ Then $$I(\sigma)=\frac{1}{2}\int_{-1}^{1}\text{erf}\left(\frac{\sigma}{\sqrt{2}}+\text{erf}^{-1}(x)\right)dx=\frac{1}{\sqrt{\pi}}\int_{-\infty}^{\infty}e^{-u^{2}}\text{erf}\left(\frac{\sigma}{\sqrt{2}}+u\right)du.$$ Differentiating with respect to $\sigma$ we find that $$\frac{\partial I(\sigma)}{\partial\sigma}=\frac{\sqrt{2}}{\pi}\int_{-\infty}^{\infty}e^{-u^{2}}e^{-(u+\sigma/\sqrt{2})^{2}}du,$$ and so substituting $u=y/\sqrt{2}$ we have $$\frac{\partial I(\sigma)}{\partial\sigma}=\frac{1}{\pi}\int_{-\infty}^{\infty}e^{-y^{2}+y\sigma-\sigma^{2}/2}dy=\frac{1}{\pi}\int_{-\infty}^{\infty}e^{-(y+\sigma/2)^{2}-\sigma^{2}/4}dy=\frac{1}{\sqrt{\pi}}e^{-\sigma^{2}/4}.$$ Hence $$I(\sigma)=\frac{1}{\sqrt{\pi}}\int_{0}^{\sigma}e^{-t^{2}/4}dt=\frac{2}{\sqrt{\pi}}\int_0^{\sigma/2}e^{-t^2}dt=\text{erf}\left(\sigma/2\right),$$ as desired.

$\endgroup$
  • 1
    $\begingroup$ Thanks! That's about the shortest possible. $\endgroup$ – Nero Mar 28 '15 at 23:16
3
$\begingroup$

I did it fast without checking it too much. Might give inspiration.$$\int erf( \frac{\sigma}{\sqrt{2}} + erf^{-1}(x)) dx$$ To solve, we first do the variable exchange $u = \frac{\sigma}{\sqrt{2}} - erf^{-1}(x)$; By the inverse function derivative we find that $du = \frac{\sqrt\pi}{2}e^{erf^{-1}(x)} dx$. Solving for x from above equation and plugging it in yields us with $du =\frac{\sqrt{\pi}}{2}e^{u-\frac{\sigma}{2}}$ Therefore the integral is transformed to: $$\int erf(u)*e^{-{u-\frac{\sigma}{\sqrt{2}}}} du$$ Using partial differentiation with $ s = erf (u)$ and $ dt = e^{-({u-\frac{\sigma}{2}})}$ should solve quite easily.

$\endgroup$

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.