Simply Explicitly Invertible Approximations to 4 Decimals of Error Function and Normal Cumulative Distribution Function
We improve the Modified Winitzki's Approximation of the error function $erf(x)\cong \sqrt{1-e^{-x^2\frac{\frac{4}π+0.147x^2}{1+0.147x^2}}}$ which has error $|\varepsilon (x)| < 1.25 \cdot 10^{-4}$ $\forall x \ge 0$ till reaching 4 decimals of precision with $|\varepsilon (x)| < 2.27 \cdot 10^{-5}$; also reducing slightly the relative error. Old formula and ours are both explicitly invertible, essentially solving a biquadratic equation, after obvious substitutions. Then we derive approximations to 4 decimals of normal cumulative distribution function $Φ(x)$, of erfc$(x)$ and of the $Q$ function (or cPhi).
stat.CO↗