Cross Validated is a question and answer site for people interested in statistics, machine learning, data analysis, data mining, and data visualization. It only takes a minute to sign up.
Sign up to join this community
The Wiki article on PDFs describes the transformations of random variables.
In deriving the formula (one-dimensional), the article states "This follows from the fact that the probability contained in a differential area must be invariant under change of variables."
$$|f_Y(y)dy|=|f_X(x)dx|$$
I do not understand what "probability contained in a differential area" means and why it must be invariant under transformation.
It means that if transformation is $Y=g(X)$, we should have $$P(X=x)=P(g(X)=g(x))=P(Y=g(x))=P(Y=y)$$ where, $g(x)=y$, where $g$ is monotonic.
Since these are continuous random variables, $P(X=x)$ is normally $0$, because it's the limit $$P(X=x) = \lim_{dx\rightarrow 0} f_X(x)|dx|$$
i.e. the integral under the curve while the limit goes to $0$. However, at infinitesimal scale, the probabilities, i.e. $P(x<X<x+dx)$ and $P(y<Y<y+dy)$ must be equal. These are $f_X(x)|dx|$ and $f_Y(y)|dy|$ respectively.
A better way to think in this scale is to have the approximation $P(X=x)\approx f_X(x)|dx|$.
