When $Z \sim N(0,1)$, then $P(\vert Z\vert \gt t)\le\sqrt\frac{2}{\pi}\times \frac{exp(-t^2/2)}t$.

My question is: I can get the aboving inequality, but how can I get the inequality such that $P(\vert Z\vert \gt t)\le\sqrt\frac{2}{\pi}\times \frac{t}{1+t^2} \times exp(-t^2/2)$?

  • $\begingroup$ Does this new $Z$ follow normal(0,1) $\endgroup$ – Qwerty Jun 7 '16 at 13:45
  • $\begingroup$ Yes. Z follows standard normal distribution. $\endgroup$ – Vivian Zhang Jun 7 '16 at 13:47
  • $\begingroup$ @Qwerty Yes, $Z\sim \mathcal N(0,1)$. Or what do you mean by 'new $Z$ ' ? $\endgroup$ – callculus Jun 7 '16 at 13:47
  • $\begingroup$ I dont understand your question. How can the same Z follow the 2nd inequality if it follows the first but the second is not implied by the first? $\endgroup$ – Qwerty Jun 7 '16 at 13:49
  • $\begingroup$ I can get the first inequality. But for my supervisor asking, I need to figure out the second inequality. It's the same Z, not a new one. $\endgroup$ – Vivian Zhang Jun 7 '16 at 13:55

We have $$1-\text{erf}(x)=\frac{x}{\sqrt{\pi}}\exp(-x^2)\frac{1}{x^2+(1/2)/q}$$ with $q\geq 1$. Thus $$1-\text{erf}(x)\geq\frac{x}{\sqrt{\pi}}\exp(-x^2)\frac{1}{x^2+(1/2)}$$ Now $$P(|Z|>t)=P(Z>t)+P(Z<-t)\\ =2P(Z>t)\\ =2\left(1-\frac{1}{2}(1+\text{erf}(t/\sqrt{2}))\right)\\ =1-\text{erf}(t/\sqrt{2})$$ $$1-\text{erf}(t/\sqrt{2})\geq \sqrt{\frac{2}{\pi}}\exp(-t^2/2)\frac{t}{t^2+1}$$ This looks right.

  • $\begingroup$ What I need is "less or equal". $\endgroup$ – Vivian Zhang Jun 21 '16 at 15:03
  • $\begingroup$ Your inequality is wrong. You can see this by setting $t=0$: $P(|Z|>0)=1/2$ , $\sqrt{2/\pi}\frac{0}{1+0}\exp(0)=0$. Greater than or equal is correct. $\endgroup$ – Wouter Jun 21 '16 at 15:27
  • $\begingroup$ From the first one inequality, I can derive that $t \neq 0$. Now I want to get the boundary for t such that $t \ge 1$, that's why I need to get the second inequality. $\endgroup$ – Vivian Zhang Jun 22 '16 at 14:31
  • $\begingroup$ So If do not consider $t = 0$, then how can I drive the second inequality by using error function. $\endgroup$ – Vivian Zhang Jun 22 '16 at 14:34
  • $\begingroup$ These inequalities hold for all $t$ (at least all positive $t$). Your original inequality is never true. $\endgroup$ – Wouter Jun 22 '16 at 16:18

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.