Probability Theory

Probability Spaces and Events

Consider a random experiment with several possible outcomes. For example, we
might roll a pair of dice, flip a coin three times, or choose a random real number
between 0 and 1. The sample space for such an experiment is the set of all possible
outcomes. For example:

• The sample space for a pair of die rolls is the set

{1, 2, 3, 4, 5, 6} × {1, 2, 3, 4, 5, 6}.

• The sample space for three coin flips is the set {0, 1}3 , where 0 represents heads
and 1 represents tails.

• The sample space for a random number between 0 and 1 is the interval [0, 1].

An event is any statement about the outcome of an experiment to which we can

assign a probability. For example, if we roll a pair of dice, possible events include:

• “Both dice show even numbers.”

• “The first die shows a 5 or 6.”

• “The sum of the values shown by the dice is greater than or equal to 7.”

From a formal point of view, events are usually defined to be certain subsets of the
sample space. Thus the event “both dice show even numbers” refers to the subset
{2, 4, 6} × {2, 4, 6}. Despite this, it is more common to write statements than subsets
when referring to a specific event.
In the special case of an experiment with finitely many outcomes, we can define
the probability of any subset of the sample space, and therefore every subset is an
event. In the general case, however, probability is a measure on the sample space,
and only measurable subsets of the sample space are events.
 2

The following definition serves as the foundation of modern probability theory:

Definition: Probability Space

A probability space is an ordered triple (Ω, E, P ), where:

• Ω is a set (the sample space). Elements of Ω are called outcomes.

• E is a σ-algebra over Ω. Elements of E are called events.

• P : E → [0, 1] is a measure satisfying P (Ω) = 1. This is the probability

measure on Ω.

In general, a measure µ : M → [0, ∞] on a
set S is called a probability measure

if µ(S) = 1, in which case the triple S, M, µ forms a probability space.

EXAMPLE 1 Finite Probability Spaces

Consider an experiment with finitely many outcomes ω1 , . . . , ωn , with corresponding
probabilities p1 , . . . , pn . Such an experiment corresponds to a finite probability space
(Ω, E, P ), where:
• Ω is the set {ω1 , . . . , ωn },
• E is the collection of all subsets of Ω, and
• P : E → [0, 1] is the probability measure on Ω defined by the formula
P ({ωi1 , . . . , ωik }) = pi1 + · · · + pik . 

EXAMPLE 2 The Interval [0, 1]

Consider an experiment whose outcome is a random number between 0 and 1. We
can model such an experiment by the probability space (Ω, E, P ), where:
• Ω is the interval [0, 1],
• E is the collection of all Lebesgue measurable subsets of [0, 1], and
• P : E → [0, 1] is Lebesgue measure on [0, 1].
Using this model, the probability that the outcome lies in a given set E ⊂ [0, 1] is
equal to the Lebesgue measure of E. For example, the probability that the outcome is
rational is 0, and the probability that the outcome lies between 0.3 and 0.4 is 1/10. 

EXAMPLE 3 Product Spaces

Let (Ω1 , E1 , P1 ) and (Ω2 , E2 , P2 ) be probability spaces corresponding to two possible
 3

(a) (b)

Figure 1: (a) Each infinite sequence of coin flips corresponds to a path down an
infinite binary tree. In this case, the sequence begins with 010. (b) The leaves of an
infinite binary tree form a Cantor set.

experiments. Now, imagine that we perform both experiments, recording the outcome
for each. The combined outcome for this experiment is an ordered pair (ω1 , ω2 ), where
ω1 ∈ Ω1 and ω2 ∈ Ω2 . In fact, the combined experiment corresponds to a probability
space (Ω, E, P ), where:

• Ω is the Cartesian product Ω1 × Ω2 .

• E is the σ-algebra generated by all events of the form E1 × E2 , where E1 ∈ E1

and E2 ∈ E2 .

• P : E → [0, 1] is the product of the measures P1 and P2 . That is, P is the unique
measure with domain E satisfying P (E1 × E2 ) = P1 (E1 ) P2 (E2 ) for all E1 ∈ E1
and E2 ∈ E2 .

For example, if we pick two random numbers between 0 and 1, the corresponding
sample space is the square [0, 1] × [0, 1], with the probability measure being two-
dimensional Lebesgue measure. 

EXAMPLE 4 The Cantor Set

Suppose we flip an infinite sequence of coins, recording 0 for heads and 1 for tails.
The outcome of this experiment will be an infinite sequence (1, 0, 1, 1, 0, 1, 0, 0, 1, . . .)
of 0’s and 1’s, so the sample space Ω is the infinite product {0, 1}N .
Although the sample space {0, 1}N is infinite-dimensional, we can visualize each
outcome as a path down an infinite binary tree, as shown in Figure 1a. If we imagine
that this tree has “leaves” at infinity, then each outcome corresponds to one such
leaf. Indeed, it is possible to draw this tree so that the leaves are visible, as shown
in Figure 1b. As you can see, the leaves of the tree form a Cantor set. Indeed, under
 4

the product topology, the sample space Ω = {0, 1}N is homeomorphic to the standard
middle-thirds Cantor set.
It is not too hard to put a measure on Ω. Given a finite sequence b1 , . . . , bn of 0’s
and 1’s, let B(b1 , . . . , bn ) be the set of outcomes whose first n flips are b1 , . . . , bn , and
define

1
P0 B(b1 , . . . , bn ) = n .
2
Let B be the collection of all such sets, and let

P ∗ (E) = inf
P S
P0 (Bn )  B1 , B2 , . . . ∈ B and E ⊂ Bn
for every E ⊂ Ω. Then P ∗ is an outer measure on Ω, and the resulting measure P is
a probability measure. 

The mechanism described above for putting a measure on {0, 1}N can be modified
to put a measure on ΩN for any probability space (Ω, E, P ). For example, it is possible
to talk about an experiment in which we roll an infinite sequence of dice, or pick an
infinite sequence of random numbers between 0 and 1, and for each of these there is
a corresponding probability space.

Random Variables
A random variable is a quantity whose value is determined by the results of a
random experiment. For example, if we roll two dice, then the sum of the values of
the dice is a random variable.
In general, a random variable may take values from any set S.

Definition: Random Variable

A random variable on a probability space (Ω, E, P ) is a function X : Ω → S,
where S is any set.

In the case where S = R (or more generally if S is a topological space), we

usually require a random variable to be a measurable function, i.e. X −1 (U ) should be
measurable for every open set U ⊆ S.

EXAMPLE 5 Here are some basic examples of random variables:

• If we roll two dice, then the values X and Y that show on the dice are random
variables Ω → {1, 2, 3, 4, 5, 6}. Expressions involving X and Y , such as the sum
X + Y or the quantity X 2 + Y 3 , are also random variables.
 5

• If we pick a random number between 0 and 1, then the number X that we pick
is a random variable Ω → [0, 1].
• For an infinite number of coin flips, we can define a sequence C1 , C2 , . . . of
random variables Ω → {0, 1} by
(
0 if the nth flip is heads,
Cn = 
1 if the nth flip is tails.

Although a random variable is defined as a function, we usually think of it as a

variable that depends on the outcome ω ∈ Ω. In particular, we will often write X
when we really mean X(ω). For example, if X is a real-valued random variable, then
P (X ≤ 3)
would refer to

P {ω ∈ Ω | X(ω) ≤ 3} .

Definition: Distribution of a Random Variable

The distribution of a random variable X : Ω → S is the probability measure PX
on S defined by
PX (T ) = P (X ∈ T ).
for any measurable set T ⊂ S.

The expression P (X ∈ T ) in the definition above refers to the probability that

the value of X is an element of T , i.e. the probability of the event X −1 (T ). Thus the
distribution of X is defined by the equation
PX (T ) = P X −1 (T )

Note that the set X −1 (T ) is automatically measurable since X is a measurable func-

tion. In measure-theoretic terms, PX is the pushforward of the probability measure P
by to the set S via X.

EXAMPLE 6 Die Roll

Let X : Ω → {1, 2, 3, 4, 5, 6} be the value of a die roll. Then for any subset T of
{1, 2, 3, 4, 5, 6}, we have
|T |
PX (T ) = P (X ∈ T ) = .
6
That is, PX is the probability measure on {1, 2, 3, 4, 5, 6} where each point has prob-
ability 1/6. 
 6

EXAMPLE 7 Random Real Number

Let X : Ω → [0, 1] be a random real number between 0 and 1. Assuming X is equally
likely to lie in any portion of the interval [0, 1], the distribution PX is just Lebesgue
measure on [0, 1]. This is known as the uniform distribution on [0, 1]. 

The most basic type of random variable is a discrete variable:

Definition: Discrete Random Variable

A random variable X : Ω → S is said to be discrete if S is finite or countable.

If X : Ω → S is discrete, then the probability distribution PX for X is completely

determined by the probability of each element of S. In particular:
X
PX (T ) = PX ({x})
x∈T

for any subset T of S.

EXAMPLE 8 Difference of Dice

Let X and Y be random die rolls, and let Z = |X − Y |. Then Z is a random variable
Ω → {0, 1, 2, 3, 4, 5}, with the following probability distribution:
1 10 8
PZ ({0}) = , PZ ({1}) = , PZ ({2}) = ,
6 36 36
6 4 2
PZ ({3}) = , PZ ({4}) = , PZ ({5}) = . 
36 36 36

We end with a useful formula for integrating with respect to a probability distri-
bution. This is essentially just a restatement of the formula for the Lebesgue integral
with respect to a pushforward measure.

Proposition 1 Integration Formula for Distributions

Let X : Ω → S be a random variable, and let g : S → R be a measurable function.

Then Z Z
g dPX = g(X) dP ,
S Ω

where g(X) denotes the random variable g ◦ X : Ω → R.

 7

Continuous Random Variables

One particularly important type of random variable is a continuous variable on R.

Definition: Continuous Random Variable

Let X : Ω → R be a random variable with probability distribution PX . We say that
X is continuous if there exists a measurable function fX : R → [0, ∞] such that
Z
PX (T ) = fX dm
T

for every measurable set T ⊂ R, where m denotes Lebesgue measure. In this case,
the function fX is called a probability density function for X.

That is, X is continuous if PX is absolutely continuous with respect to Lebesgue

measure, i.e. if
dPX = fX dm
for some non-negative measurable function fX . Recall the following formula for inte-
grals with respect to such measures.

Proposition 2 Weighted Integration

Let X : Ω → R be a continuous random variable with probability density fX , and

let g : R → R be a measurable function. Then
Z Z
g dPX = gfX dm.
R R

EXAMPLE 9 Standard Normal Distribution

Consider a random variable X : Ω → R with probability density function defined by
 
1 1 2
fX (x) = √ exp − x .
2π 2

In this case, X is said to have the standard normal distribution. A graph of the
function fX is shown in Figure 2a.
For such an X, the probability PX (T ) that the value X lies in any set T is given
by the formula Z
PX (T ) = fX dm.
T
 8

0.4

P Ha < X < bL
0.3

0.2

0.1

-3 -2 -1 0 1 2 3 a b
(a) (b)

Figure 2: (a) The probability density fX for a standard normal distribution. (b) Area
under the graph of fX on the interval (a, b).

For example, if (a, b) is an open interval, then

Z Z b  
1 1 2
P (a < X < b) = fX dm = √ exp − x dx.
(a,b) a 2π 2

This is illustrated in Figure 2b. 

The probability density function fX can be thought of as describing the probability

per unit length of PX . The following theorem gives us a formula for this function:

Theorem 3 Density Formula

Let X : Ω → R be a random variable, and define a function fX : R → [0, ∞] by

PX [x − h, x + h]
fX (x) = lim+ .
h→0 2h
Then fX (x) is defined for almost all x ∈ R and fX is measurable. Moreover:
Z
1. If fX dm = 1, then X is a continuous random variable.
R

2. If X is continuous, then fX is a probability density function for X.

PROOF This is related the Lebesgue differentiation theorem, though it does not
follows from it immediately. 
 9

0.0 0.2 0.4 0.6 0.8 1.0

Figure 3: Probability density for X 2 , where X is chosen uniformly from [0, 1].

EXAMPLE 10 Density of X 2
Let X : Ω → [0, 1] be uniformly distributed, and let Y = X 2 . Then for any interval
[a, b] ⊂ [0, 1], we have
√ √ 
Y ∈ [a, b] ⇔ X∈ a, b

so
√ √
√ √
PY [a, b] = PX [ a, b ] = b − a.
Therefore,

√ √
PY [x − h, x + h] x+h− x−h 1
fY (x) = lim+ = lim+ = √ .
h→0 2h h→0 2 2 x

A plot of this function is shown in Figure 3. Note that the total area under the graph
of the function is 1, which proves that Y is continuous, and that fY is a probability
density function for Y . 

Finally, it should be mentioned that most of the discussion of continuous random

variables generalizes to random variables X : Ω → Rn . Such a variable is called
continuous if there exists a function fX : Rn → [0, ∞] so that
Z
PX (T ) = fX dµ
T

for every measurable T ⊂ Rn , where µ denotes Lebesgue measure on Rn . The prob-

ability density for such a variable is determined by the formula


PX B(p, r)
fX (p) = lim+
,
r→0 µ B(p, r)

where B(p, r) denotes a ball of radius r in Rn centered at the point p.

 10

Expected Value

Definition: Expected Value

Let X : Ω → R be a random variable on a probability space (Ω, E, P ). The ex-
pected value of X is defined as follows:
Z
EX = X dP.
Ω

The expected value is sometimes called the average value or mean of X, and is
also denoted X or hXi.

It is possible to calculate the expected value of a random variable directly from

the distribution:

Proposition 4 Expected Value for a Distribution

Let X : Ω → R be a random variable with distribution PX . Then

Z
EX = x dPX (x).
R

PROOF Let g : R → R be the function g(x) = x. By the integration formula for

distributions (Proposition 1), we have
Z Z Z Z
x dPX (x) = g dPX = g(X) dP = X dP = EX. 
R R Ω Ω

EXAMPLE 11 Expected Value of a Discrete Variable

Let X : Ω → S be a discrete random variable, where S is a finite subset of R. Then
it follows from the above theorem that
X
EX = x P ({x}).
x∈S

For example, if X is the value of a die roll, then

           
1 1 1 1 1 1
EX = 1 +2 +3 +4 +5 +6 = 3.5. 
6 6 6 6 6 6
 11

EXAMPLE 12 Expected Value of a Continuous Variable

Let X : Ω → R be a continuous random variable with probability density function fX .
Then it follows from the weighted integration formula (Proposition 2) that
Z
EX = x fX (x) dm(x).
R

For example, if X is a random number from [0, 1] with uniform distribution, then
Z
1
EX = x dm(x) = .
[0,1] 2

For another example, consider √the random variable Y = X 2 . The probability density
function for Y is fY (x) = 1/(2 x) (see Example 10), so
Z
1 1
EY = x √ dm(x) = . 
[0,1] 2 x 3

EXAMPLE 13 Infinite Expected Value

It is possible for a random variable to have infinite expected value. For example, let
X : Ω → [1, ∞) be a continuous random variable with
1
fX (x) = .
x2
R
Note that [1,∞) fX dm = 1, so this is indeed a legitimate probability density function.
Then the expected value of X is
Z Z ∞
1
EX = x fX (x) dm(x) = dx = ∞.
[1,∞) 1 x
The same phenomenon can occur for a discrete random variable. For example, if
X : Ω → N satisfies
1
PX ({n}) = ,
Cn2
where C = ∞ 2 2
P
n=1 1/n = π /6, then
∞ ∞
X X 1
EX = n PX ({n}) = = ∞.
n=1 n=1
Cn

Note that both of these examples involve positive random variables. For a general
variable X : Ω → R, it is also possible for EX to be entirely undefined. 

The following proposition lists some basic properties of expected value. These
follow directly from the corresponding properties for the Lebesgue integral:
 12

Proposition 5 Properties of Expected Value

Let X : Ω → R and Y : Ω → R be random variables with finite expected value.

Then:

1. If C ∈ R is constant, then EC = C.

2. E[αX] = α EX for any α ∈ R.

3. E[X + Y ] = EX + EY .

4. |EX| ≤ E|X|.

5. If X ≥ 0, then EX ≥ 0, with equality if and only if P (X = 0) = 1.

Variance and Standard Deviation

There is much more to the distribution of a variable than its expected value. For
example, the average length of an adult blue whale is about 80 feet, but this tells
us very little about whale lengths: are most blue whales about 75–85 feet long, or
is the range more variable, say 20–140 feet? The tendency of a random variable to
take values far away from the mean is called dispersion. Two common measures of
dispersion are the variance and standard deviation:

Definition: Variance and Standard Deviation

Let X : Ω → R be a random variable with finite expected value µ. The variance
of X is the quantity
Var(X) = E (X − µ)2 .
 

The square root of Var(X) is called the standard deviation of X.

Though the variance has nicer theoretical properties, the standard deviation has
more meaning, since it is measured in the same units as X. For example, if X is a
random variable with units of length, then the standard deviation of X is measured
in feet, while the variance is measured in square feet. The standard deviation in the
length of an adult blue whale is about 10 feet, meaning that most whales are about
70–90 feet long.
To give you a feel for these quantities, Figure 4a shows several normal distributions
with different standard deviations. The standard normal distribution has standard
deviation 1, and normal distributions with standard deviations of 0.5 and 2 have also
been plotted. As a general rule, a normally distributed random variable has roughly
 13

¾ = 0.5
Area
¾=1 0.682
¾=2
-4 -2 0 2 4 -3 -2 -1 0 1 2 3

(a) (b)

Figure 4: (a) Normal distributions with standard deviations of 0.5, 1, and 2. (b) A
normally distributed variable has a 68.2% chance of being within one standard devi-
ation of the mean.

a 68.2% probability of being within one standard deviation of the mean, as shown
in Figure 4b. In particular,
Z 1  
1 1 2
√ exp − x dx ≈ 0.682689.
−1 2π 2

For example, if we assume that the length of a random adult blue whale is nor-
mally distributed with a mean of 80 feet and a standard deviation of 10 feet, then
approximately 68% of blue whales will be between 70 and 90 feet long.
As with expected value, we can compute the variance of a random variable directly
from the distribution:

Proposition 6 Variance of a Distribution

Let X : Ω → R be a random variable with finite mean µ and distribution PX .

Then Z
Var(X) = (x − µ)2 dPX (x).
R

PROOF Let g : R → R be the function g(x) = (x − µ)2 . By the integration formula

for distributions (Proposition 1), we have
Z Z Z
(X − µ)2 dP = E (X − µ)2 = Var(X).
 
g dPX = g(X) dP = 
R Ω Ω
 14

EXAMPLE 14 Variance of a Discrete Variable

Let X : Ω → S be a discrete random variable, where S ⊂ R, and let µ = EX. Then
X
Var(X) = (x − µ)2 P ({x}).
x∈S

For example, if X is the value of a die roll, then EX = 3.5, so

6
X (n − 3.5)2
Var(X) = ≈ 2.92.
n=1
6

Taking the square root yields a standard deviation of approximately 1.71. 

EXAMPLE 15 Variance of a Continuous Variable

Let X : Ω → R be a continuous random variable with probability density function fX .
Then it follows from the weighted integration formula (Proposition 2) that
Z
Var(X) = (x − µ)2 fX (x) dm(x).
R

For example, if X is a random number from [0, 1] with uniform distribution, then
EX = 1/2, so
Z  2
1 1
Var(X) = x− dm(x) = .
[0,1] 2 12
Taking the square root yields a standard deviation of approximately 0.29. 

EXAMPLE 16 Normal Distributions

A random variable X : Ω → R is said to be normally distributed if it has a prob-
ability density function of the form.
  
1 1 x − µ 2
fX (x) = √ exp −
σ 2π 2 σ

Such a variable has mean µ and standard deviation σ, as can be seen from the
following integral formulas:
  
1 x − µ 2
Z
x
√ exp − dm(x) = µ.
R σ 2π 2 σ

(x − µ)2
  
1 x − µ 2
Z
√ exp − dm(x) = σ 2 . 
R σ 2π 2 σ
 15

EXAMPLE 17 Infinite Variance

As with the expected value, it is possible for the variance of a random variable to
be infinite or undefined. For example, let X : Ω → R be a random variable with
distribution
C
fX =
1 + |x|3
√
where C = 3 3/(4π). Then X has finite expected value, with
Z
Cx
EX = 3
dm(x) = 0.
R 1 + |x|

But
Cx2
Z
Var(X) = dm(x) = ∞.
R 1 + |x|3
From a measure-theoretic point of view, this random variable is an L1 function on
the probability space Ω, but it is not L2 . 

We end by stating another standard formula for variance.

Proposition 7 Alternative Formula for Variance

Let X : Ω → R be a random variable with finite expected value. Then

Var(X) = E[X 2 ] − (EX)2 .

PROOF Let µ = EX. Then

Var(X) = E (X − µ)2 = E X 2 − 2µX + µ2

   

= E[X 2 ] − 2µ(EX) + µ2 = E[X 2 ] − 2µ2 + µ2 = E[X 2 ] − µ2 . 

It follows from this formula that Var(X) is finite if and only if E[X 2 ] < ∞, i.e. if
and only if X ∈ L2 (Ω).

Exercises
1. Let X : Ω → (0, 1] have the uniform distribution, and let Y = 1/X.
a) Find the probability density function fY : [1, ∞) → [0, ∞] for Y .
 16

b) What is the expected value of Y ?

2. Let X : Ω → [0, 1] have the uniform distribution, and let Y = sin(8X). Use the
integration formula for distributions (Proposition 1) to compute EY .

3. Let X and Y be the values of two die rolls, and let Z = max(X, Y ).
a) Find PZ ({n}) for n ∈ {1, 2, 3, 4, 5, 6}.
b) Determine EZ, and find the standard deviation for Z.

4. Let X : Ω → [1, ∞) be a continuous random variable with probability density

function
3
fX (x) = 4 .
x
Compute EX and Var(X).