I would like to show the following:

$$\lim_{y \to +\infty} \left[\frac{1}{\sqrt{\pi}e^{y^2}(1-\text{erf}(y))}-y\right]=0.$$

Here $$\text{erf}(y)= \frac{2}{\sqrt{\pi}} \int_{0}^{y}e^{-x^2}dx$$ denotes the error-function. In fact I would like to have that $\lim\limits_{y \to +\infty} [\mathbb{E}(X\mid X \geq y)- y]=0$ provided that $X \sim \mathcal{N} (0,\frac{1}{\sqrt{2}})$. Since $$\mathbb{E}(X\mid X \geq y)= \frac{\int_{y}^\infty xe^{-x^2} dx}{\int_{y}^\infty e^{-x^2} dx}=\frac{\frac{1}{2}e^{-y^2}}{\frac{2}{\sqrt{\pi}}(1-\text{erf}(y))}=\frac{1}{\sqrt{\pi}e^{y^2}(1-\text{erf}(y))}$$ the second limit should then reduce to the first one in the yellow box.

Since $erf(y)$ cannot be stated explicitely with elementary functions I tried to use approximizations of it but it seems that it is not enough to allow errors up to a constant (although the constant is tiny) to show that the limit is zero. I tried the ones given here: https://en.wikipedia.org./wiki/Error_function#Approximation_with_elementary_functions.

I think that intuitively the statments should be correct because of the strong descent of the Gaussian function $e^{-x^2}$ when approaching infinity: The bigger $y$ is the more I expect $\mathbb{E}(X\mid X \geq y)$ to be close to $y$.

  • $\begingroup$ By L'Hospital's rule, as $y \to \infty$, $$\frac{\frac{e^{-y^2}}{\sqrt{\pi}}}{(1-\text{erf}(y))}=y+\varepsilon(y)$$ with $\lim_{y \to \infty}\varepsilon(y)=0$, see my answer below. $\endgroup$ – Olivier Oloa Feb 15 '17 at 15:44

Hint. One may observe that, as $y \to \infty$, $$ \frac{1}{\sqrt{\pi}e^{y^2}(1-\text{erf}(y))} =\frac{\frac{e^{-y^2}}{\sqrt{\pi}}}{(1-\text{erf}(y))}\to \frac00 $$ then one is allowed to apply L'Hôpital's rule getting, as $y \to \infty$, $$ \frac{\frac{e^{-y^2}}{\sqrt{\pi}}}{(1-\text{erf}(y))}\sim\frac{\frac{-2ye^{-y^2}}{\sqrt{\pi}}}{-\frac{2}{\sqrt{\pi}}e^{-y^2}}=y. $$


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.