Error function and integrals

What constitutes an explanation will depend on your background. Do you know about integration techniques such a changing variables? And do you know that the definition of erfc(x) itself involves an integral?

 

Well, the definition of erfc(z) is actually:

[tex]erfc(z) \equiv 1 - erf(z)[/tex]

However, this can be pretty easily changed to the integral definition if we remember that

[tex]erf(z) \equiv \frac{2}{\sqrt{\pi}} \int_0^z exp\left(-t^2\right) dt[/tex]

and furthermore that

[tex]\frac{2}{\sqrt{\pi}} \int_0^\infty exp\left(-t^2\right) dt = 1[/tex]

By substituting these definitions in for 1 and erf(z), we get this expression:

[tex]erfc(z) = \frac{2}{\sqrt{\pi}} \int_0^\infty exp\left(-t^2\right) dt - \frac{2}{\sqrt{\pi}} \int_0^z exp\left(-t^2\right) dt[/tex]

Which is easily proven using basic integral properties to be equal to the expression in the OP.

 

By definition of erfc :
http://mathworld.wolfram.com/Erfc.html

 

We usually just find values numerically. The only value that I know of where erf(x) takes a simple, closed-form value is x=0. Other than that, it's all approximation.

 

The erfc(x) is a function which is used exactly like the functions exp(x), ln(x), cos(x) or many others. All these functions have to be found numerically (except for a few particular values of x).
For example, what is the analytical expressiion of :
integrate (1/t)*dt from t=1 to t=x ?
Of course, the answer is ln(x). Then would you say "knowing that it has to be found numerically makes me slightly unhappy" ?

The only difference between erfc(x) and ln(x) is that one is familiar to you and the other not.

An almost similar question arised elsewhere about the integration of x^x. The consequence was a funny discussion reported in the paper "The Sophomores Dream Function" :
http://www.scribd.com/JJacquelin/documents