7
$\begingroup$

Let $X \sim N (0, 1)$ and $Y ∼ N (0, 1)$ be two independent random variables, and define $Z = \min(X, Y )$. Prove that $Z^2\sim\chi^2(1),$ i.e. Chi-Squared with degree of freedom $1.$

I found the density functions of $X$ and $Y,$ as they are normally distributed. How would one use the fact that $Z = \min(X,Y)$ to answer the question? Thanks!

$\endgroup$
6
  • 1
    $\begingroup$ Maybe it would it be useful to express $Z$ as a liner combination of $X+Y$ and $|X-Y|$, using properties of maximum and minimum. $\endgroup$ – AnyAD May 22 '18 at 2:52
  • 4
    $\begingroup$ Can you clarify what the subscript $1$ in $X_1^2$ refers to? $\endgroup$ – Mike Earnest May 22 '18 at 3:21
  • $\begingroup$ @Programmer : Your comment on $X_1$ as "the degree of freedom 1" does not make sense. That is because it is not clear what that means, and how $X_1$ would differ from $X$. I believe you are really trying to show that $Z^2$ has the same distribution as $X^2$. Thus, I believe the "1" subscript is just a typo. $\endgroup$ – Michael May 22 '18 at 3:46
  • $\begingroup$ I doubt the correctness of the conclusion... Because the p.d.f. of $Z^2$ can be explicitly figured out, and when I figured it out, it was not the one of $\chi^2_1$ variable... $\endgroup$ – Wanshan May 22 '18 at 4:11
  • 2
    $\begingroup$ @Wanshan If $X, Y$ are iid and continuous random variables with the property that $X$ and $-X$ have the same distribution, then $(\min[X,Y])^2$ has the same distribution as $X^2$. You can condition on events $\{X<0, Y<0\}, \{X<0, Y>0\}, \{X>0, Y<0\}, \{X>0, Y>0\}$. $\endgroup$ – Michael May 22 '18 at 4:13
0
$\begingroup$

Let $\Phi$ be the standard normal cdf. First, find the cdf of $Z^2$. For any $z>0$,

$$ \begin{align} P(Z^2\le z) &=P(Z>-\sqrt{z})-P(Z>\sqrt{z})\\ &=P(X>-\sqrt{z})P(Y>-\sqrt{z})-P(X>\sqrt{z})P(Y>\sqrt{z})\\ &=(1-\Phi(-\sqrt{z}))^2-(1-\Phi(\sqrt{z}))^2. \\&=(\Phi(\sqrt{z}))^2-(1-\Phi(\sqrt{z}))^2. \\&=2\Phi(\sqrt z)-1 \end{align} $$

On the other hand, $$ P(X^2\le z)=P(X\le \sqrt{z})-P(X<-\sqrt{z})=\Phi(\sqrt{z})-\Phi(-\sqrt{z})=\Phi(\sqrt{z})-(1-\Phi(\sqrt{z})) $$ As you can see, $P(X^2\le z)=P(Z^2\le z)$ for all $z$, QED.

$\endgroup$
1
  • $\begingroup$ I see this is same as existing answers, for some reason they had not loaded when I wrote this. $\endgroup$ – Mike Earnest May 22 '18 at 14:26
1
$\begingroup$

$1-F_Z(t) = P(Z>t) = P(X>t)P(Y>t) =\frac{1}{2\pi}\left[ \int_t^{\infty}\exp(-x^2/2) \, dx \right]^2$. Take derivative w.r.t. $t$ and we can get $$ f_Z(t) = -\frac{d}{dt}\frac{1}{2\pi} \left[ \int_t^\infty \exp(-x^2/2)\,dx \right]^2 = \frac{1}{\pi}\exp(-t^2/2)\left[\int_t^{\infty}\exp(-x^2/2)\,dx\right]. $$ Now let $W = Z^2$ \begin{align} 1-F_W(t) = {} & P(W>t) = P(Z>\sqrt{t})+P(Z<-\sqrt{t})\\[10pt] = {} & \int_{\sqrt{t}}^{\infty}\frac{1}{\pi}\exp(-s^2/2) \left[\int_s^\infty \exp(-x^2/2)\,dx\right]\,ds \\[10pt] & {} + \int_\infty^{-\sqrt{t}}\frac{1}{\pi}\exp(-s^2/2) \left[ \int_s^{\infty}\exp(-x^2/2)\,dx\right]\,ds\\[10pt] = {} & \int_{\sqrt{t}}^{\infty}\frac{1}{\pi}\exp(-s^2/2)\left[ \int_s^\infty \exp(-x^2/2)\,dx \right] \, ds \\[10pt] & {} + \int^\infty_{\sqrt{t}}\frac{1}{\pi}\exp(-s^2/2) \left[ \int^{s}_{-\infty}\exp(-x^2/2)\,dx\right]\,ds\\[10pt] = {} & \int_{\sqrt{t}}^\infty \frac{1}{\pi}\exp(-s^2/2)\frac{\sqrt{2\pi}}{2}\,ds \end{align} Taking derivative we can get $f_W(t) = \frac{1}{\sqrt{2\pi t}}\exp(-t/2)$, which is the same as $f_{\chi^2_1}(t) = \frac{1}{\sqrt{2\pi t}}\exp(-t/2)$.

$\endgroup$
4
  • 1
    $\begingroup$ @Michael Yes I just noticed that and fixed it. Thanks for your discussion! It's really helpful. $\endgroup$ – Wanshan May 22 '18 at 4:53
  • 1
    $\begingroup$ Since you've shown that $P(Z>z)=P(X>z)^2$, you could also just write $$P(Z^2>t)=P(Z>\sqrt{t})+P(Z<-\sqrt{t})\\ = P(Z>\sqrt{t})+1-P(Z>-\sqrt{t})\\ =P(X>\sqrt{t})^2 + 1-P(X>-\sqrt{t})^2\\ =P(X>\sqrt{t})^2+1-(1-P(X<-\sqrt{t}))^2\\ =2P(X>\sqrt{t}),$$ hence $P(Z^2\le t) = 1-2P(X>\sqrt{t})=P(-\sqrt{t}<X<\sqrt{t})=P(X^2<t).$ QED $\endgroup$ – r.e.s. May 22 '18 at 5:34
  • $\begingroup$ @r.e.s. Good idea! That's far more elegant. $\endgroup$ – Wanshan May 22 '18 at 13:31
  • $\begingroup$ @Programmer Yes I agree with you. Maybe he meant to write chi square distribution. But since X is a standard normal variable, it's also correct to say that $Z^2$ has the same distribution as $X^2$ :) $\endgroup$ – Wanshan May 22 '18 at 19:21
0
$\begingroup$

Since $X$ and $Y$ are independent, we can see that $f(x,y) = f(x)f(y)$.

Now, if we transform our axis from $X,Y$ to $X_1,Y_1$, where $X_1 = \frac{1}{\sqrt{2}}(X+Y)$ and $X_2 = \frac{1}{\sqrt{2}}(X-Y)$ - essentially, we are rotating the $XY$ space by a +45 degrees.

Since the distribution $f(x,y)$ has radial symmetry, we can see that $f(x_1,x_2) =f(x,y) $. From this we can see that, even $X_2$ $∼ N (0, 1)$.

Now, the variable $Z = min(X,Y)$ lies to the space right of the line $\frac{1}{\sqrt{2}}(X-Y)$ in the $XY$ space.

Therefore, distribution of $Z^2$ is the same as the distribution of $X_2^{2}$ - which is essentially a Chi-square distribution with degrees of freedom = 1.

$\endgroup$

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.