Fubini's theorem to find $P(X+Y \leq c)$ given the joint density

588 Views Asked by At

Given the joint density $f(x,y)$ we are asked to find $P(X + Y \leq c)$ where $c$ is some constant non-negative number.

My reasoning for this problem is thus:

$P(X+Y \leq c) = P(X \leq c - Y)$

In order to find this probability, we need the marginal probability distribution: $P(X)$. We can then integrate $P(x)$ from the lower bound (say, $0$) to $c-y$ for the desired probability.

$P(X)$ is found by taking the integral of the joint density with respect to y from the lower bound to the upper bound (say $0$ to $1$).

$P(X) = \int_{0}^{1} f(x,y)dy$

Then we just take the integral as described above and we have the following double integral:

(1) $$\int_{0}^{c - y} \int_{0}^{1} f(x,y)dydx$$

However, in all the examples I've seen thus far, the integrals are reversed:

(2) $$\int_{0}^{1} \int_{0}^{c-y} f(x,y)dxdy$$

Doing some research I discovered Fubini's theorem which at first seemed to explain why this switch is possible (if not explain why it's necessary).

However, I soon discovered that the two above integrals are not in fact equal (e.g if $f(x,y) = \frac {6}{7} (x+y)^{2}$ ). Which means that my reasoning is wrong, Fubini's theorem doesn't apply here, and I should arrive at the second integral from the very beginning.

Can some explain where my reasoning is faulty and how one can arrive at the second integral?

1

There are 1 best solutions below

2
On BEST ANSWER

Generally, for continuous real-valued random variables, $X$ and $Y$ with a joint probability density function $f_{X,Y}$, Fubini's theorem allows that: $$\begin{align}\mathsf P(X+Y\leqslant c) ~&= \int_{\{(x,y):x+y\leqslant c\}} f_{X,Y}(x,y)~\mathrm d(x,y) \\[1ex] &= \int_{-\infty}^{\infty}\int_{-\infty}^{c-x} f_{X,Y}(x,y)~\mathrm d y~\mathrm d x \\[1ex] &= \int_{-\infty}^{\infty}\int_{-\infty}^{c-y} f_{X,Y}(x,y)~\mathrm d x~\mathrm d y\end{align}$$

(Notice that the constraint is placed on the inner integral and is a function of the bound variable for the outer integral.)


In the specific case when the supports for the probability density functions of $X,Y$ are strictly non-negative, and $c$ is too, this may be expressed as:

$$\begin{align}\mathsf P(X+Y\leqslant c) ~&= \int_{\{(x,y):0\leqslant x+y\leqslant c\}} f_{X,Y}(x,y)~\mathrm d(x,y) \\[1ex] &= \int_{0}^{c}\int_{0}^{c-x} f_{X,Y}(x,y)~\mathrm d y~\mathrm d x \\[1ex] &= \int_{0}^{c}\int_{0}^{c-y} f_{X,Y}(x,y)~\mathrm d x~\mathrm d y\end{align}$$

(Notice the constraint on the outer integral is not a function of either bound variables.)


In the particular case when $f_{X,Y}(x,y) = \tfrac 67(x+y)^2~\mathbf 1_{0\leqslant x\leqslant 1, 0\leqslant y\leqslant 1}$, we have:

$$\begin{align}\mathsf P(X+Y\leqslant c) ~&=~ \int_{\{(x,y):0\leqslant x\leqslant 1,0\leqslant y\leqslant 1,0\leqslant x+y\leqslant\min\{2,c\}\}}\tfrac 67(x+y)^2~\mathrm d (x,y) \\[2ex] &=~ \int_0^{\min\{1,c\}}\int_0^{\min\{1,c-x\}} \tfrac 67 (x+y)^2~\mathrm d y~\mathrm d x \\[2ex]&=~ \begin{cases}0&:& c<0 \\ \int_{0}^{c}\int_{0}^{c-x} \tfrac 67(x+y)^2~\mathrm d y~\mathrm d x &:& 0\leqslant c < 1\\ \int_{0}^{c-1}\int_{0}^{1} \tfrac 67(x+y)^2~\mathrm d y~\mathrm d x+\int_{c-1}^{1}\int_{0}^{c-x} \tfrac 67(x+y)^2~\mathrm d y~\mathrm d x &:& 1\leqslant c < 2\\ 1 &:& 2\leqslant c \end{cases}\\[2ex] &= \begin{cases}0&:& c<0 \\ \int_{0}^{c}\int_{0}^{c-y} \tfrac 67(x+y)^2~\mathrm d x~\mathrm d y &:& 0\leqslant c < 1\\ \int_{0}^{c-1}\int_{0}^{1} \tfrac 67(x+y)^2~\mathrm d x~\mathrm d y + \int_{c-1}^{1}\int_{0}^{c-y} \tfrac 67(x+y)^2~\mathrm d x~\mathrm d y&:& 1\leqslant c < 2\\ 1 &:& 2\leqslant c \end{cases}\end{align}$$

Evaluation is left to you.