Register to reply

Transformation Of Probability Density Functions

Share this thread:
Jan20-07, 10:46 PM
P: n/a
1. The problem statement, all variables and given/known data
Let X and Y be random variables. The pdfs are [itex]f_X(x)=2(1-x)[/itex] and [itex]f_Y(y) = 2(1-y)[/itex]. Both distributions are defined on [0,1].

Let Z = X + Y. Find the pdf for Z, [itex]f_Z(z)[/itex].

2. Relevant equations
I'm using ideas, not equations.

3. The attempt at a solution
I'm dying of curiosity about where I'm going wrong. I'm so sure of each step, but my answer can't be correct because [itex]\int_0^2 f_Z(z)\,dz[/itex] is zero! Here's my logic.

Consider the cdf (cumulative distribution function) for Z:

F_Z(z) = P(Z\le z) = P(X+Y \le z)

Here, [itex]F_Z(z)[/itex] is the volume above the triangle shown in the image I attached to this message (in case something happens to the attachment, it's the triangle in quadrant 1 bounded by x=0, y=0 and x+y=z.)

The volume above the shaded region represents [itex]F_Z(z)[/itex].

F_Z(z) = \int_{x=0}^{x=z} 2(1-x)\int_{y=0}^{y=z-x} 2(1-y)\,dy\,dx

Performing the integrals gives [itex]F_Z(z) = \frac{1}{6}z^4 - \frac{4}{3}z^3 + 2z^2[/itex]. Then taking the derivative of the cdf gives the pdf:

f_Z(z) = \partial_z F_Z(z) = \frac{2}{3}z^3 - 4z^2 +4z

Unfortunately, this can't be right because the integral of this function over [0,2] gives zero.

I also would've expected that the maximum of [itex]f_Z(z)[/itex] would be at z=0 since individually, X and Y are most likely to be zero.

I checked my algebra and calculus with Mathematica; it looked fine. I think there's a conceptual problem. There must be something I don't understand or some point I'm not clear about.

What did I do wrong?
Attached Thumbnails
Phys.Org News Partner Science news on
Experts defend operational earthquake forecasting, counter critiques
EU urged to convert TV frequencies to mobile broadband
Sierra Nevada freshwater runoff could drop 26 percent by 2100
Jan21-07, 01:03 AM
Sci Advisor
HW Helper
PF Gold
quasar987's Avatar
P: 4,771
You have to split the problem into different cases.

i) z<0
ii) z<1
iii) 1<z<2
iv) z>2

In each case the area of integration is different. But for each case, your model for the integral is

[tex]F_Z(z)=\int\int_{\{(x,y)\in [0,1]\times[0,1]:\ y\leq z-x, \}}f_{XY}(x,y)dxdy[/tex]

So you're integrating over the area that's the intersection of the square [0,1] x [0,1] with the area under the curve y=z-x.

(Btw, you never said that X and Y are independant but I assume they are?)
Jan21-07, 10:57 AM
P: n/a
Ahhh.... yes. I *completely* understand.

We're integrating that portion of the unit square (in QI with lower left corner at origin) that lies underneath x+y=1.

So when 0<z<1, there's only one function that represents the "top": the line x+y=z.

And when 1<z<2, there's two functions: y=1 and x+y=z, and therefore we need to break the region of integration into two portions.

Got it. Thanks.

And, yes. :-)

Register to reply

Related Discussions
Moment Generating Functions and Probability Density Functions Set Theory, Logic, Probability, Statistics 4
2 different probability density functions will give the same E(g(x)) or not General Math 1
Probability density functions Calculus & Beyond Homework 7
What's the difference between probability and probability density Quantum Physics 5
Transformation of non-monotonic density functions Calculus & Beyond Homework 0