Convolution of distribution functions.

995 Views Asked by At

Assume that we have two independent random variables $X_1$ and $X_2$ with distribution functions $F_1$ and $F_2$ respectively. Then Theorem 6.1.1. on Kai Lai Chung's "A course in probability theory" states that:

The sum $X_1+X_2$ has distribution function $F_1*F_2$.

On this set of notes the author states that actually the distribution of the sum is NOT the convolution of the distribution functions, but rather the convolution of the $F_1$ and the density $f_2$.

For what know I about the definition of convolution

$\int F_1(x-x_1) dF_2(x_2)$ looks more like the convolution $F_1*f_2$, I think that this is more a problem of notation but still I am curious.

EDIT: Another example on which the convolution is taken between the distribution functions is "Probability Theory and examples" by Rick Durret.

1

There are 1 best solutions below

0
On BEST ANSWER

This question hinges on the definition of "convolution" and of the $*$ symbol. A glance at the wikipedia article gives one formulation for the convolution of two probability measures: it is the probability measure of the sum of two independent random variables, equivalent to $(\mu*\nu)(A)=P(X+Y\in A)$ if $X\sim\mu$ and $Y\sim\nu$. From this one has $$F_{X+Y}(t)=P(X+Y\le t)=E[(P(X+Y\le t|Y)]=E[ P(X\le t-Y|Y)] = E[ F_X(t-Y)|Y] = \int_{\mathbb{R}}F_X(t-y) F_Y(dy) = \int_{\mathbb{R}}F_X(t-y) dF_Y(y),$$ which one might as well take to be the definition of the convolution of distributions functions of probability measures, regardless of whether the random variables $X$ and $Y$ are continuous or discrete, with densities or without.

The notes the OP cites are not reliable on this issue: they adopt the dumbing-down restriction to probability laws with density functions. I do not know what Ross says, but I think Chung (the most trustworthy of the lot) does not make mistakes like this.