Derivation of approximation of Error function

607 Views Asked by At

In Abramowitz and Stegun there are some formulas for approximation of Error funtion. I am intrested in the formulas $7.1.25$ to $7.1.28$, here is one of them ($7.1.26$).

$$\operatorname{erf}(x)\approx 1 - e^{-x^2}\sum_{i=1}^5 a_i t^i +\epsilon(x),\quad 0\leq x<\infty,$$ where $t = 1/(1 + px)$. The coefficients $a_i$ and $p$ are some decimal numbers ($p = 0.3275911$, for example) and $|\epsilon(x)|\leq 1.5\cdot 10^{-7}$. The others are of similar form.

  1. How are these formulas derived?
  2. How to prove that the relative error $\epsilon(x)$ is really as small as they claim?
  3. What are the exact values of coefficients?

In the bottom of the page with formulas it is said that Abramowitz and Stegun took these formulas from C. Hastings: Approximations for digital computers. Indeed, the formulas are present there (starting at page $167$), also, there are graphs of $\epsilon(x)$, but again, no derivation or exact values.

Thank you.