I have a little question regarding the error function that disturbs me a lot:

Let us consider the classical error function defined as:

erf($z$) $= \frac{2}{\sqrt{\pi}}\int_{0}^{z} e^{-t^2} dt$

It is well-known [1] that for $z= \infty,$ erf($z$) = 1. My question may be idiotic but is the following one: is the error function only equals to 1 at $z= \infty$ ? In other words, is the error function never reaches the value of 1 before the strict value $z= \infty$ ? Don't tell me to refer to tables of values of erf($z$) because the latter are based on numerical integrations and thus subjected to numerical uncertainties.

Thank you very much for your help,


[1] http://mathworld.wolfram.com/Erf.html

  • 6
    $\begingroup$ Yes for the question in title, as the error function is strictly increasing since the integrand is strictly positive. $\endgroup$ – GNUSupporter 8964民主女神 地下教會 May 10 '18 at 0:26
  • 1
    $\begingroup$ The value of the integral starts at zero and grows toward the asymptote one. It reaches one only when z reaches infinity (so to speak, b/c/ infinity is not a number). $\endgroup$ – Christopher Marley May 10 '18 at 0:26
  • $\begingroup$ It does take on the value $1$ in the complex plane, however: the roots with smallest absolute value are at roughly $-1.35 \pm 1.99i$. $\endgroup$ – Chappers May 10 '18 at 0:46
  • $\begingroup$ @GNUSupporter I wanted to write an answer, but it is basically your comment. Maybe you should answer? $\endgroup$ – SK19 May 10 '18 at 0:51
  • 1
    $\begingroup$ @SK19 By writing $z$, I think OP wants the domain to be complex. As there's already a answer for the complex case, I think there's no need to add one for the real case. $\endgroup$ – GNUSupporter 8964民主女神 地下教會 May 10 '18 at 3:49

If you see that: $$\frac{d}{dz} \text{erf}(z) = \frac{2e^{-z^2}}{\sqrt{\pi}}>0$$ Therefore the function is strictly increasing.

  • As $\text{erf}(0)=0$ then $\text{erf}(x)>0$.
  • $\frac{2}{\sqrt{\pi}}\int_{0}^{\infty} e^{-t^2} dt=1$

Take $x>0$.

$$\frac{2}{\sqrt{\pi}}\int_{0}^{\infty} e^{-t^2} dt = \frac{2}{\sqrt{\pi}}\int_{0}^{x} e^{-t^2} dt + \frac{2}{\sqrt{\pi}}\int_{x}^{\infty} e^{-t^2} dt $$ $$1 = \text{erf}(x) + \frac{2}{\sqrt{\pi}}\int_{x}^{\infty} e^{-t^2} dt$$ $$1- \text{erf}(x) = \frac{2}{\sqrt{\pi}}\int_{x}^{\infty} e^{-t^2} dt >0$$



Probably not very useful.

In the linked page, they give the asymptotic series $$\text{erf}(x)=1-\frac{e^{-x^2}}{\sqrt{\pi }}\sum_{n=0}^\infty (-1)^n \frac{(2n-1)!!}{2^n} x^{-(2n+1)}$$ and a very accurate asymptotics would be $$\text{erf}(x)=1-e^{-x^2} \left(\frac{1}{\sqrt{\pi } x}+O\left(\frac{1}{x^2}\right)\right)$$

For $x=10$, the "exact" value is $$0.99999999999999999999999999999999999999999999791151$$ while the above truncated expression gives $$0.99999999999999999999999999999999999999999999790117$$


After GNU Supporter's comment, consider $$a_n=(-1)^n \frac{(2n-1)!!}{2^n} x^{-(2n+1)}$$ This gives $$\left|\frac{a_{n+1}}{a_n}\right|=\frac{2 n+1}{2 x^2}\approx \frac n {x^2}$$ which decreases very fast.

  • $\begingroup$ I find it useful as it gives a second-order estimation. To answer OP's question, it remains to set $x$ large enough so that $$\frac{1}{\sqrt{\pi } x}+O\left(\frac{1}{x^2}\right) > 0.$$ $\endgroup$ – GNUSupporter 8964民主女神 地下教會 May 10 '18 at 3:41

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.