2.
Transformation of variables
Sec 6.5 of GW, 6.7 of HPS, 6.7 of Ross
Example 1. Let X ∼ U nif orm(0, 1) and let Y = X 2 . What is the density for Y ?
Theorem 1. Let U ∼ U nif orm(0, 1) random variable. Let X be a continuous ran-
dom variable such that its distribution function, FX , is a strictly increasing continuous
function. Then
−1
1. Y = FX (U ) has the same distribution as X.
2. Z = FX (X) has the same distribution as U .
Example 2. Find a function g, such that g(U ) has Exp(λ) distribution, where U ∼
U nif (0, 1).
Suppose X has density fX (x) and Y = g(X), where g : Rk → Rk is differentiable
and has a well-defined inverse. We are interested in determining the density of Y . Note
that ġ(x) is a k × k matrix, namely, the matrix of partial derivatives ∂gi (x)/∂xj . This
square matrix is called the Jacobian matrix. We will use the term “Jacobian” to refer to
the determinant of this matrix. Note that the Jacobian of g(x) is a real-valued function
def
of x. We denote this function Jg : Rk → R. Thus, Jg (x) = Det{ġ(x)}.
Suppose now that h(y) denotes the inverse of g(x), so that h ◦ g(x) = x for all x and
g ◦ h(y) = y for all y.
Then we may write the density function for Y in terms of the density function for X
as follows:
fY (y) =| Jh (y) | fX ◦ h(y) (1)
It is important to note, that the Jacobian referred to is the Jacobian of the inverse
transformation, not the transformation itself.
Example 3. Let X and Y be independent N (0, 1) random variables. Find the joint
distribution of (X + Y )/2 and (X − Y )/2.
Example 4. Let R and Θ be independent random variables such that R has the Rayleigh
density 2
re−r /2 , r ≥ 0
fR (r) =
0, r < 0,
and Θ is uniformly distributed on (−π/2, π/2). Find the joint distribution of X = RcosΘ
and Y = RsinΘ. This is the Box-Muller method of generating observations from normal
distribution.
Example 5. Suppose X1 , · · · , Xn+1 are iid standard exponential random variables. For
j = 1, · · · , n define
Pj
i=1 Xi
Yj = Pn+1
i=1 Xi
1
We derive the joint density of (Y1 , · · · , Yn ) as follows. First, we observe that the joint
density of (X1 , · · · , Xn+1 ) is
( n+1 )
X
fX (x) = exp − xi I{x1 > 0, · · · , xn+1 > 0}
i=1
Pj
As an intermediate step, define Zj = i=1 Xi for j = 1, · · · , n + 1. Then the Xi may be
expressed in terms of the Zi as
Zi if i = 1
Xi = (2)
Zi − Zi − 1 if i > 1
The Jacobian is clearly 1 for this transformation, since the Jacobian matrix is lower
triangular with ones on the diagonal. Therefore, we obtain FZ (z) = exp{−zn+1 }I{0 <
z1 < z2 < · · · < zn+1 } as the density of Z. If we define Yn+1 = Zn+1 , then we may
express the Zi in terms of the Yi as
Yn+1 Yi if i < n + 1
Zi = (3)
Yn+1 if i = n + 1
The Jacobian matrix of the transformation in equation (3) is upper triangular, with yn+1
along the diagonal except for a 1 in the lower right corner. Thus, the Jacobian equals
n
yn+1 , so the density of Y is
n
fY (y) = yn+1 exp{−yn+1 }I{yn+1 > 0}I{0 < y1 < · · · < yn < 1}. (4)
Thus, (Y1 , · · · , Yn ) is independent of Yn+1 and the density of (Y1 , · · · , Yn ) is proportional
to I{0 < y1 < · · · < yn < 1}. Note that Yn+1 may be seen to have a Gamma(n + 1, 1)
density, which is also evident when we consider that Yn+1 is the sum of n + 1 iid standard
exponential random variables.
Example 6. Let R1 and R2 be independent, absolutely continuous, positive random
variables and let R3 = R1 R2 . Find the density of R3 .
