6-ConceptsProbability Page 193 Wednesday, February 4, 2004 3:00 PM
Concepts of Probability 193
This relationship can be extended to any number of variables.
In probability theory, given a random variable X, the following
expectation is called the characteristic function (c.f.) of the variable X
φX()t = Ee[ itX]= E[cos tX ]+ iE[sin tX]
If the variable X admits a d.f. FX(y), it can be demonstrated that the
following relationship holds:
∞ ∞ ∞
∫
φ itX
X()t = Ee[ x x
itX]= e dF
X()x = ∫ cos tx dFX()+ ∫sin tx dFX()
- ∞ –∞ –∞
In this case, the characteristic function therefore coincides with the Fou-
rier-Stieltjes transform. It can be demonstrated that there is a one-to-one
correspondence between c.d.s and d.f.s. In fact, it is well known that the
Fourier-Stieltjes transform can be uniquely inverted.
In probability theory convolution is defined, in a more general way,
as follows. Given two d.f.s FX(y) and FY(y), their convolution is defined
as:
∞
F ()u = (FXFY)()u = ∫ FX(uy– )dFY ()y
- ∞
It can be demonstrated that the d.f. of the sum of two variables X,Y
with d.f.s FX(y) and FY(y) is the convolution of their respective d.f.s:
∞
PX ( + Y ≤u) = FXY+ ()u = F ()u = (FXFY)()u = ∫ FX(uy– )dFY ()y
- ∞
If the d.f.s admits p.d.f.s, then the inversion formulas are those estab-
lished earlier. Inversion formulas also exist in the case that the d.f.s do
not admit densities but these are more complex and will not be given
here.^13
We can therefore establish the following property: the characteristic
function of the sum of n independent random variables is the product of
the characteristic functions of each of the summands.
(^13) See Chow and Teicher, Probability Theory.