Calculating the inverse of a function involving the error function

3 replies · 2K views · last reply
Science Advisor
Homework Helper
Education Advisor
Insights Author
Messages
1,098
Reaction score
1,385
I have a probability distribution over the interval ##[0, \infty)## given by $$f(x) = \frac{x^2}{2\sqrt{\pi} a^3} \exp\left(- \frac{x^2}{4a^2} \right)$$From this I want to derive a formula for the inverse cumulative density function, ##F^{-1}##. The cumulative density function is a slightly nasty-looking but doable integral involving the error function,$$F(x) = \mathrm{erf}\left( \frac{x}{2a} \right) - \frac{x}{\sqrt{\pi} a} \exp \left( -\frac{x^2}{4a^2} \right)$$So it remains to invert this. Ideally I would like to find an analytic expression, but I haven't had much success.
 
For a=1 the plot is
Wolfram Alpha input and two plots of erf(x/2) - (x/sqrt(pi)) exp(-x^2/4): top plot for x from -3.7 to 3.7 with y from -1.0 to 1.0, bottom plot for x from -22.4 to 22.4 with y from -1.0 to 1.0, both S-shaped curves rising monotonically.


So the expected plot of the inverse function is

Rotated plot of an S-shaped curve through the origin, approaching y = 1 and y = -1 at large positive and negative x, with x axis labelled x and tick marks at -50, -10, 10 and 50 and y axis at -10, -5, 5 and 10.

It seems difficult to get the anaytical form if not impossible.
 
The inverse function $$ F^{-1}(x) $$ of the cumulative density function $$ F(x) = erf(\frac{x}{2a}) – \frac{x}{\sqrt\pi a} exp(-\frac{x^2}{4a^2}) $$ can not be expressed in an analytic form.