PBRT_V2 总结记录 <92> Transforming Between Distributions

概述：

Suppose we are given random variables Xi that are already drawn fromsome PDF px(x).
Now, if we compute Yi = y(Xi), we would like to find the distribution of the new random
variable Yi . This may seem like an esoteric problem, but we will see that understanding
this kind of transformation is critical for drawing samples frommultidimensional distribution
functions.

The function y(x) must be a one-to-one transformation; if multiple values of x mapped
to the same y value, then it would be impossible to unambiguously describe the probability
density of a particular y value. A direct consequence of y being one-to-one is that its
derivativemust either be strictly greater than zero or strictly less than zero, which implies
that

 （Pr 表示的是 CDF）

and therefore

  （这里也是CDF）

This relationship between CDFs leads directly to the relationship between their PDFs. If
we assume that y’s derivative is greater than zero, differentiating gives

（这里需要注意：

py  / dx = px / dx

py / dy * dy / dx = px / dx

这里就是以上的公式了

）

and so

 

In general, y’s derivative is either strictly positive or strictly negative, and the relationship
between the densities is

( (dy/dx)^(-1) 只是倒数  )

 

Example:

How can we use this formula? Suppose that px(x) = 2x over the domain [0, 1], and let
Y = sin X. What is the PDF of the random variable Y ? Because we know that dy/dx =
cos x,

（注意 ： y = sin x  =》 x = arcsin y）