1
$\begingroup$

I am trying to show that if $X_1$ and $X_2$ are independent $\mathcal N(0,1)$ random variables, then $$Z=\dfrac{X_1-X_2}{\sqrt{2}}\sim \mathcal N(0,1)$$

My attempt: Find the moment generating function: \begin{align}M_Z(t)&=E\left[\exp\left(t\dfrac{X_1-X_2}{\sqrt2}\right)\right]=E\left[\exp\left(\dfrac{tX_1}{\sqrt2}\right)\right]E\left[\exp\left(\dfrac{-tX_2}{\sqrt 2}\right)\right]\\[0.2cm]&=-2E\left[\exp(tX_1)\right]E\left[\exp({tX_2})\right]=-2M_{X_1}(t)M_{X_2}(t)=-2\exp\left(\dfrac{1}{2}t^2\right)\end{align}

Is this correct so far? How can I proceed?

$\endgroup$
  • $\begingroup$ The step $$E[exp(\dfrac{tx_1}{\sqrt 2})]E[exp(\dfrac{-tx_2}{\sqrt 2})]=-2E[exp(tx_1)]E[exp({tx_2})]$$ is (a real mystery and) incorrect. $\endgroup$ – Did Nov 4 '16 at 6:33
1
$\begingroup$

Normal distribution is uniquely determined by its moments, hence, indeed, it suffices to show that the moment generating function (MGF) of $Z$ coincides with the MGF of $\mathcal N(0,1)$ which is equal to $\displaystyle\exp\left(\frac12t^2\right)$. Now, using the following property

If $\displaystyle S_{n}=\sum _{i=1}^{n}a_{i}X_{i}$, where the $X_i$ are independent random variables and the $a_i$ are constants, then the moment-generating function for $S_n$ is given by $$M_{S_{n}}(t)=M_{X_{1}}(a_{1}t)M_{X_{2}}(a_{2}t)\cdots M_{X_{n}}(a_{n}t)$$

you obtain for $Z=\frac1{\sqrt2}X_1+\left(\frac{-1}{\sqrt2}\right)X_2$ that \begin{align}M_Z(t)&=M_{X_1}\left(\frac{t}{\sqrt2}\right)M_{X_2}\left(\frac{-t}{\sqrt2}\right)=\exp\left(\frac12\cdot\left(\frac{t}{\sqrt2}\right)^2\right)\exp\left(\frac12\cdot\left(\frac{-t}{\sqrt2}\right)^2\right)\\[0.2cm]&=\exp\left(\frac{t^2}4\right)\exp\left(\frac{t^2}4\right)=\exp\left(\frac12{t^2}\right)\end{align} which concludes the proof.

$\endgroup$
1
$\begingroup$

For a normal random variable X obviously -X is also normal. The sum of two independent normal random variables ist also normal.

So you already know that $\frac{X_1 - X_2}{\sqrt{2}}$ is normal, with expectation

$$E[\frac{X_1 - X_2}{\sqrt{2}}] = \frac{E[X_1] - E[X_2]}{\sqrt{2}} = 0$$

and variance

$$Var[\frac{X_1 - X_2}{\sqrt{2}}] = \frac{Var[X_1] + Var[X_2]}{2} = 1$$

So it's $\mathcal N(0,1)$

$\endgroup$
  • $\begingroup$ can the variance be distributed?@Gono $\endgroup$ – MAS Nov 4 '16 at 10:35
  • $\begingroup$ For independent rv it holds Var(X+Y) = Var(X) + Var(Y) $\endgroup$ – Gono Nov 4 '16 at 10:49

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.