My question is two parts. My book defines the gamma density as:

$$g_n(x) = \lambda\frac{(\lambda x)^{n-1}}{(n-1)!}e^{-\lambda x},\space x>0$$

Part 1: In an example my book states that the following is a gamma density with $\lambda = 1/2$ and $n = 1/2$ (density for the square of a standard normal random variable):

$$f_{X^2}(r) = \frac{1}{\sqrt{2\pi r}}e^{-r/2},\space r >0$$

I can't figure out how to get this equation from the above, given that if n = 1/2 then I get a negative factorial in the denominator of the gamma density?

Part 2: More generally, $f_{X^2}(r)$ above is the density for the square of a standard normal random variable $X$. My book states that the sum of of the square of $m$ independent standard normal random variables ($\mu_X = 0$ and $\sigma_X^2=1$) is a gamma density with $\lambda = 1/2$ and $n = m/2$ (the chi-squared density). For the case of $m=2$ the density given in my book was

$$f_{R^2}(r)=1/2e^{-r/2},\space r>0$$

I tried computing the convolution between $f_{R^2}$ and $f_{X^2}$ and ended up with:

$$f_{Z^2}(z)=\frac{1}{\sqrt{2\pi}}\sqrt{z}e^{-z/2}, \space z > 0$$

Again I couldn't arrive at this result by plugging in $\lambda = 1/2$ and $n = 3/2$ into the gamma function.


You need to use the general definition of Gamma function. For any $\alpha>0,$ $\Gamma(\alpha)$ is defined as $$\Gamma(\alpha)=\int_0^{\infty}x^{\alpha-1}e^{-x}dx.$$

An important property of the Gamma function is that $\Gamma(\alpha+1)=\alpha\Gamma(\alpha).$ This is proved here.

Another fact is that $\Gamma\left(\dfrac{1}{2}\right)=\sqrt{\pi}.$ Here is the proof.

Moreover, this function reduces to $\Gamma(\alpha)=(\alpha-1)!$ when $\alpha$ is a natural number. With this definition, the Gamma distribution is defined as $$g_{n,\lambda}(x)=\dfrac{\lambda^nx^{n-1}e^{-\lambda x}}{\Gamma(n)},\,\,x>0.$$

The pdf you have derived for sum of squares of three standard normal random variables, is correct. You can check that this is indeed the $g\left(\dfrac{3}{2},\dfrac{1}{2}\right)$ density, using the fact that $\Gamma\left(\dfrac{3}{2}\right)=\dfrac{1}{2}\times\Gamma\left(\dfrac{1}{2}\right)=\dfrac{\sqrt{\pi}}{2}.$

| cite | improve this answer | |

@ArnabAuddy has already answered part 1, so I'll talk about part 2. Let's first consider the case $m=1$, for which we just have the distribution of $X^2$ with $X\sim N(0,\,1)$. The characteristic function of $X^2$ is $$\Bbb E\exp itX^2=\int_{\Bbb R}\frac{1}{\sqrt{2\pi}}\exp\left(-\frac{1-2it}{2}x^2\right)dx=(1-2it)^{-1/2}.$$A general value of $m$ therefore obtains characteristic function $(1-2it)^{-m/2}$, as does the claimed Gamma distribution. Since distinct distributions have different characteristic functions, this completes the proof.

| cite | improve this answer | |

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.