Proving valid joint pdf is valid with unknowns

31 Views Asked by At

Question

Suppose Suppose that $Y$ and $Z$ are continuous random variables with joint probability density function $f_{y,z}(y,z) = sy^{2}e^{-(\alpha + \beta z)y}$ where $y >0 \;,\; z > 0$ and $\alpha >0 \;,\beta > 0$. Find the value of $s$ that would allow for $f_{y,z}(y,z)$ to be a valid pdf.

My approach

For $f_{y,z}(y,z)$ to be a valid distribution, the following condition must be met.

$$ \iint_{A}f_{y,z}(y,z) \; \mathrm{d}z\mathrm{dy}=1 $$

In context, for the given bounds, I have come up with this result with the given bounds.

\begin{align} \int_{0}^{\infty}\int_{0}^{\infty} sy^{2}e^{-(\alpha + \beta z)y} \; \mathrm{d}z \; \mathrm{d}y \end{align}

Integrating for $\mathrm{d}z$, I have obtained the following:

\begin{align} \int_{0}^{\infty} sy^{2}e^{-(\alpha + \beta z)y} \; \mathrm{d}z &= sy^{2}\int_{0}^{\infty} e^{-(\alpha + \beta z)y} \; \mathrm{d}z \\ &= sy^{2}\int_{0}^{\infty} e^{-y\alpha + \beta zy} \mathrm{d}z \\ &= sy^{2} \cdot -\frac{1}{\beta y} \cdot e^{-y\alpha + \beta zy} \\ \end{align}

Combining and evaluating for the bounds

\begin{align} -\frac{sye^{-y(\alpha + \beta z)}}{\beta} \Biggr|_{0}^{\infty} &= -\frac{sye^{-y(\alpha + \beta (\infty))}}{\beta} - \left(-\frac{{sye^{-y(\alpha + \beta (0))}}}{\beta}\right) \\ &= \frac{sy}{e^{y\alpha + \beta (\infty)}\beta} + \frac{sy}{e^{y(\alpha)}\beta} \end{align}

As, $\lim_{z\rightarrow\infty}$, the first fraction becomes $0$. Thus, leaving only the second result $\frac{sy}{e^{y(\alpha)}\beta}$.

Integrating for $\mathrm{d}y$ over the bounds using integration by parts. \begin{align} \int_{0}^{\infty} \frac{sye^{-y\alpha}}{\beta} \; \mathrm{d}y &= \frac{s}{\beta} \int_{0}^{\infty} ye^{-y\alpha} \; \mathrm{d}y \\ &= \frac{s}{\beta}\cdot\left(\frac{e^{-y(\alpha)}(\alpha y + 1)}{a^2} \Biggr|_{0}^{\infty}\right) \end{align}

Whereby, this yields (I am very much unsure about this component)

\begin{align} \frac{s}{\beta}\cdot\frac{e^{-y(\alpha)}(\alpha y + 1)}{a^2} \Biggr|_{0}^{\infty} &= \frac{s}{\beta}\cdot\left(\frac{e^{-\infty\alpha}(\alpha \infty)}{a^2} - \frac{e^{-\alpha(0)}(\alpha(0) + 1)}{a^2}\right) \end{align}

To satisfy the condition for a valid pdf, this must be equal to 1. \begin{align} \frac{s}{\beta}\cdot - \frac{1}{a^2} &= 1 \\ -\frac{s}{\alpha^2\beta} &= 1 \\ \alpha^2\beta &= -s \\ \therefore s &= -\alpha^2\beta \end{align}

I am very unsure if my approach is correct and my integration is not the greatest aspect. Any input would mean the world to me.

Thanks in advance!

1

There are 1 best solutions below

1
On BEST ANSWER

Everything looks correct except the sign is wrong. If $\alpha > 0$ and $\beta > 0$, then your choice $s = -\alpha^2 \beta$ would be negative, which in turn would make the joint density negative.

The marginal density of $Y$ that you obtained is correct: $$f_Y(y) = \frac{s}{\beta} y e^{-\alpha y} \, y > 0.$$ Its integral can be computed via integration by parts, or by recognizing it as a gamma function, or by recognizing it as a gamma density:

$$\begin{align} 1 &= \frac{s}{\beta} \int_{y=0}^\infty ye^{-\alpha y} \, dy \\ &= \frac{s}{\beta} \left( \left[ -\frac{1}{\alpha} y e^{-\alpha y} \right]_{y=0}^\infty + \int_{y=0}^\infty \frac{1}{\alpha} e^{-\alpha y} \, dy \right) \\ &= \frac{s}{\beta} \left( -0 + 0 + \left[-\frac{1}{\alpha^2} e^{-\alpha y} \right]_{y=0}^\infty \right) \\ &= \frac{s}{\beta} \left( -0 + \frac{1}{\alpha^2} \right) \\ &= \frac{s}{\alpha^2 \beta}. \end{align}$$

Equivalently, the substitution $$u = \alpha y, \quad du = \alpha \, dy$$ gives $$\int_{y=0}^\infty ye^{-\alpha y} \, dy = \int_{u=0}^\infty \frac{u}{\alpha} e^{-u} \cdot \frac{1}{\alpha} \, du = \frac{1}{\alpha^2} \int_{u=0}^\infty ue^{-u} \, du = \frac{1}{\alpha^2} \Gamma(2) = \frac{1}{\alpha^2}$$ where we have used the gamma function $$\Gamma(z) = \int_{u=0}^\infty u^{z-1} e^{-u} \, du.$$ Finally, we can also note that $f_Y(y)$ is proportional to a gamma density with shape parameter $a = 2$ and rate parameter $b = \alpha$; i.e., $$f_Y(y) \propto \frac{b^a y^{a-1} e^{-by}}{\Gamma(a)} = \alpha^2 y e^{-\alpha y}.$$ Therefore we require $s/\beta = \alpha^2$, or $s = \alpha^2 \beta$.