0

$X\sim N(\mu,\sigma^2)$.
$A,B$ are constants and $A\ne0$.

How to prove that $AX+B\sim N(A\mu+B,A^2\sigma^2)$ ?

Thank you!

CS1
  • 2,047
  • 1
    http://math.stackexchange.com/a/798398/31475

    There are, of course, much simpler ways :)

     – Emily May 23 '14 at 15:08
  • Just plug the new RV in the definition of $N(\mu,\sigma^2)$ and reconduct the integral to standard form. – Marco May 23 '14 at 15:09

2 Answers2

4

You can manipulate $P(aX+b \leq c)$ into $P(X \leq \frac{c-b}{a})$ for $a>0$. Then, you know the CDF of $X$, so plug this into the CDF of $X$. You also know the CDF of $N(a \mu + b , a^2 \sigma^2)$, so compare them to see that they're equal, thus the distributions are the same.

Alternatively you can use characteristic functions/moment generating functions by noting $E[e^{t (aX+b)}] = e^{bt} E[e^{(at)X}] = e^{bt} M_X(at)$, and comparing that to the moment generating function of a $N(a \mu + b , a^2 \sigma^2)$ r.v.

fin-cos
  • 3
Batman
  • 19,390
3

The standard method:

Let $Y \sim \mathcal{N}(A\mu + B, A^2\sigma^2)$. Write the CDF of $Y$:

$$F_Y(y) = \frac{1}{\sqrt{2\pi A^2\sigma^2}}\int_{-\infty}^\infty \exp \left( -\frac{(x-(A\mu+B))^2}{2A^2\sigma^2}\right)\ dx.$$

Compute the expectation of $AX+B$ using the standard methodology.

$$E[AX+B] = \frac{1}{\sqrt{2\pi \sigma^2}}\int_{-\infty}^\infty (Ax+B)\exp \left( -\frac{(x-(\mu))^2}{2\sigma^2}\right)\ dx.$$

Show that the result leads to the mean of $Y$. Do the same for the variance.

Clever way:

Use the fact that $\operatorname{erf}$ and the CDF of a Gaussian are related. Let the CDFs of $AX+B$ and $Y$ be equal to the same uniform random variable (inverse transform method). Use $\operatorname{erf}^{-1}$ to write $AX+B$ in terms of $Y$. Show that the result leads directly to $X$.

Emily
  • 35,688
  • 6
  • 93
  • 141
  • The first method doesn't work - you need to know a priori that Gaussians are from a stable family (well, something slightly weaker than that - scaling a Gaussian remains Gaussian and translating a Gaussian remains Gaussian), and that Gaussians are completely characterized by the first 2 moments [ that is, you need to know a priori that $AX+B$ is Gaussian ]. – Batman May 23 '14 at 15:18
  • @batman I assume that by using the $ N ()$ notation that that is known, but otherwise yes, you need to do that first. Not hard. – Emily May 23 '14 at 15:28