School of Mathematics and Statistics

UNSW Sydney

Introduction to Probability and Stochastic Processes

OPEN LEARNING
 Chapter 10
Brownian Motion

2 / 75
Outline:

10.1 Introduction

10.2 Standard Brownian Motion

10.3 Another Approach to Defining Brownian Motion

10.4 Invariance Properties of Brownian Motion

10.5 Other Properties of Brownian Motion

10.6 Joint Distribution of Brownian motion

10.7 Hitting Times of Brownian Motion

10.8 Maximum Value of Brownian Motion

3 / 75
Outline - continued:

10.9 Gambler’s Ruin Problem Revisited

10.10 Gaussian Processes

10.11 Brownian Motion is a Gaussian Process

10.12 Brownian Motion with Drift

10.13 Geometric Brownian Motion

10.14 Application of Geometric Brownian Motion

10.15 Martingales

10.16 Supplementary Material

4 / 75
10.1 Introduction

5 / 75
We will start with a very brief history of Brownian motion.
1827 Brown described the behaviour of random movements of particles
suspended in a medium such as a liquid or a gas.

1900 Bachelier was the first to model the return process

for the French Stock market of 1900.

1905 Einstein showed that Brownian motion could be

explained by assuming that the molecules
of the surrounding median were continually
bombarding an immersed particle.

1918 Wiener gave the first precise mathematical

definition of Brownian motion in a series of papers
starting in 1918.

6 / 75
Sample Paths of Brownian Motions

fractional Brownian motion − parameter: 0.5 fractional Brownian motion − parameter: 0.5
5 0

−2

−4

0 −6

−8

−10

−5 −12

−14

−16

−10 −18
100 200 300 400 500 600 700 800 900 1000 100 200 300 400 500 600 700 800 900 1000

7 / 75
Sample Paths of Brownian Motions

fractional Brownian motion − parameter: 0.5 fractional Brownian motion − parameter: 0.5
35 10

30
0

25
−10
20

15 −20

10 −30

5
−40
0

−50
−5

−10 −60
1000 2000 3000 4000 5000 6000 7000 8000 9000 10000 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000

8 / 75
Now, we will discuss two mathematical definitions of Brownian motion (BM).
Definition
A continuous stochastic process {X(t), t ≥ 0} is said to be a Brownian
motion with drift coefficient µ if
(1) X(0) = 0
(2) {X(t), t ≥ 0} has stationary independent increments
(3) for every t ≥ 0, X(t) is normally distributed with mean µ t and variance
σ 2 t.

Brownian motion is an example of a continuous time, continuous

state space Markov process.

9 / 75
If we write X(t + s) = X(t + s) − X(t) + X(t), then
V ar(X(t + s)) = V ar(X(t + s) − X(t) + X(t))
= V ar(X(t + s) − X(t)) + V ar(X(t))
by independent increments
= V ar(X(s)) + V ar(X(t)) by stationary increments.

Hence the solution to the above function is just V ar(X(t)) = σ 2 t.

The variance σ 2 is a function of the underlying process and can be calculated

empirically.

10 / 75
10.2 Standard Brownian Motion

11 / 75
We will write B(t) as a standard Brownian motion which means it has
zero drift (mean zero) and σ 2 = 1.

Unless otherwise stated, we will assume σ 2 = 1.

Therefore, the density of B(t) is

1 n x2 o
fB(t) (x) = √ exp − , t > 0, −∞ < x < ∞,
2π t 2t

which is normal with mean zero and variance t.

12 / 75
Hence, we have
Z ∞
1 x2
P (B(t) ≥ a) = √ e− 2t dx
2 π t Za
y2 √ √
∞
1 x
= √ √
e− 2 t dy (by letting y = √ , so dx = t dy)
2 π t a/ t t
Z ∞
1 y 2
= √ √
e− 2 dy
2 π a/ t
!
 a 
= 1−Φ √ , a ∈ R, where Φ(x) ∼ N (0, 1).
t
and
 a 
P (B(t) ≤ a) = Φ √ , a ∈ R, where Φ(x) ∼ N (0, 1).
t

13 / 75
Note that X(t) = B(t) + µt is known as a Brownian motion with drift.

Any Brownian motion X(t) with mean µ t and variance σ 2 t can be converted
to a standard Brownian motion.

That is,
X(t) − µ t
σ
is a standard Brownian motion.

14 / 75
Now, consider a Brownian motion as a particle in a liquid medium.

Let x0 be the particle’s initial position at time t0 . That is, X(t0 ) = x0 .

Let ft (x|x0 ) be the conditional probability density of X(t + t0 ) given

X(t0 ) = x0 .

We assume that the probability distribution governing the transitions is

stationary in time, so ft (x|x0 ) does not depend on time t and is only a
density function in x.

Therefore, we have
Z ∞
ft (x|x0 ) ≥ 0 and ft (x|x0 ) dx = 1. (1)
−∞

15 / 75
In addition, we say that for small t, X(t + t0 ) is likely to be near X(t0 ) = x0 .
That is,

lim ft (x|x0 ) = 0 for x ̸= x0 (2)

t→0

From the physical principles, Einstein (1905) showed that ft (x|x0 ) must
satisfy the following partial differential equation

∂f ∂ 2f
= D 2. (3)
∂t ∂x

The above is known as the diffusion equation, and D is called the diffusion
coefficient.

16 / 75
Actually, D = 2RT
Nh
, where R is the gas constant, T is temperature, N is
Avogadro’s number and h is a coefficient of friction.

When D = 1/2, we can see that,

1 1
ft (x|x0 ) = √ exp{− (x − x0 )2 } (4)
2π t 2t

is a solution to Equation (3).

In fact, it is the unique solution under the boundary conditions (1) and (2).

Proof of this is beyond this course.

17 / 75
10.3 Another Approach to Defining Brownian
Motion

18 / 75
Another approach to Equation (3) is an approximation using a discrete
random walk.

Specifically, we consider a symmetric random walk on the integers:


+1 with probability 1/2
ξi =
−1 with probability 1/2,
where ξi , i = 1, 2, . . . , are independent and identically distributed.

Let pk (n) = P ( ni=1 ξi = k) be the probability that the random walk finds
P
itself k steps to the right of its starting point at time n.

19 / 75
Then by the Chapman-Kolmogorov equation and the Central Limit
Theorem , we have

1 1
pk (n + 1) = pk+1 (n) + pk−1 (n)
2 2
1
pk (n + 1) − pk (n) = [pk+1 (n) − 2pk (n) + pk−1 (n)]. (5)
2

Let the length between transitions be denoted by ∆ and the length of each
step η.

Then the analog of Equation (5) is

1
pkη ((n + 1)∆) − pkη (n∆) 2 [p(k+1)η (n∆) − 2pkη (n∆) + p(k−1)η (n∆)]
= . (6)
∆ ∆

20 / 75
Now let ∆ and η shrink to zero preserving the relationship ∆ = η 2 and at the
same time let n and k increase to ∞ so kη → x while n ∆ → t.

Then pk η (n ∆) → ft (x|x0 = 0) and Equation (6) passes formally into

Equation (4) .

21 / 75
Another limiting process for pk (n) requires the CLT. If we write
pk (n) = P (ξ1 + · · · + ξn = k), where

+1 with probability 1/2
ξi =
−1 with probability 1/2.

Then by the CLT ,

√
nx Z x
X 1 2 /2
lim pk (n) = √ e−u du. (7)
n→∞
k=−∞
2π −∞

The convergence of the sequence of stochastic processes above to a standard

Brownian motion is termed the invariance principle.
The limiting equation of Equation (6) and that of Equation (7) are
essentially the same and are connected by the “invariance principle of
stochastic processes”.
22 / 75
10.4 Invariance Properties of Brownian Motion

23 / 75
Before we continue, let us state some other invariance principles. Let X(t),
t ≥ 0 be a Brownian motion starting at x (i.e. X0 = x).

Translation Invariance: If a ∈ R, then a + X(t), t ≥ 0 is a

Brownian motion starting at a + x.

Scale Invariance: If c > 0, then c−1/2 X(ct), t ≥ 0 is a Brownian

motion starting at c−1/2 x.

Orthogonal Invariance: If Q is an orthogonal linear transformation

on Rd , then Q X(t), t ≥ 0 is a Brownian motion starting at Qx (in
particular, −X(t), t ≥ 0 is a Brownian motion starting at −x).

24 / 75
10.5 Other Properties of Brownian Motion

25 / 75
Without any proofs, some other properties of Brownian Motion are
sample paths of Brownian motion are nowhere differentiable.

That is, for every t, and fixed ω ∈ Ω,

Bt+h (ω) − Bt (ω)

Bt′ (ω) = lim does not exists.
h→0 h

26 / 75
 (i)
Let Bt , i = 1, 2 be two independent Brownian motions and a1 , a2 ∈ R,
where both are non-zero.
Then,
(1) (2
B̃t (ω) ≡ (a21 + a22 )−1/2 (a1 Bt (ω) + a2 Bt (ω)), t ≥ 0,

is a Brownian motion.
The random variable σ B(t) + a t, t ≥ 0, is Gaussian random variable
N (a t, σ 2 t) for any constants σ and a.
This process is called a Brownian motion with linear drift or Brownian
motion with drift.
The constant σ is called the volatility and it contributes to the order of
the variance function σ(t) = σ 2 t.

27 / 75
10.6 Joint Distribution of Brownian Motion

28 / 75
Since we know the distribution of B(t), t ≥ 0, we can also determine the joint
distribution of

B(t1 ), B(t2 ), . . . , B(tn ), t1 < t2 < · · · < tn ∈ [0, t].

By the independent increment property ,

B(t1 ), B(t2 ) − B(t1 ), B(t3 ) − B(t2 ), . . . , B(tn ) − B(tn−1 )

are all independent.

By the stationary increment property ,

B(tk ) − B(tk−1 )

is normally distributed with mean zero and variance tk − tk−1 for

each k = 1, 2, . . . , n.

29 / 75
Notation

B(t1 ) = B(t1 ) − B(t0 ) since B(t0 ) = B(0) = 0.

It follows from the stationary increment property

D
B(tk ) − B(tk−1 ) −
→ B(tk + s) − B(tk−1 + s),

D
B(tk ) − B(tk−1 ) −
→ B(tk − tk−1 ).

D
Recall the definition of −
→ Convergence in Distribution .

30 / 75
It follows that the joint distribution is

fB(t1),B(t2), ... ,B(tn)(x1, x2, . . . , xn)

= fB(t1)(x1) fB(t2−t1)(x2 − x1) fB(t3−t2)(x3 − x2) · · · fB(tn−tn−1)(xn − xn−1)

  
x21 2 2
exp − 12 + (xt22−x
t1 −t1
1)
+ · · · + (xtnn−x n−1 )
−tn−1
= p , (8)
(2 π)n/2 t1 (t2 − t1 ) · · · (tn − tn−1 )

−∞ < xi < ∞, i = 1, 2, . . . , n, 0 < t1 < t2 < · · · < tn = t.

From this equation, we can calculate any desired probability/distribution.

31 / 75
It can be shown that the conditional distribution of B(s) given B(t) = a,
when s < t, is Normal with mean st a and variance st (t − s). That is,
  s s 
P B(s) B(t) = a ∼ N a, (t − s) .
t t

The conditional distribution of B(t) given B(t1 ) = a and B(t2 ) = b, when

−t
t1 < t < t2 , is Normal with mean tt22−t1
a + tt−t 1
2 −t1
b and variance (t2 −t)(t−t
t2 −t1
1)
.
That is,
! !
t −t t − t1 (t2 − t)(t − t1 )
P B(t) B(t1 ) = a, B(t2 ) = b ∼ N 2 a+ b, .
t2 − t1 t2 − t1 t2 − t1

32 / 75
10.7 Hitting Times of Brownian Motion

33 / 75
Let τa be the first time at which the Brownian motion B(t) reaches a.
Suppose that a > 0.

τa is called a hitting time or first passage time

τa = min{s ≥ 0 : B(s) = a}.

We want to compute the distribution of τa , i.e., P (τa ≤ t), that the process
will hit level a before time t.

34 / 75
If τa ≤ t, the process hits a at some point in [0, t], and by symmetry, it is just
as likely to be above a or below a at time t:
  1
P B(t) ≥ a τa ≤ t = .
2

This fact is often called the reflection principle.

35 / 75
Consider P (B(t) ≥ a) by conditioning on whether or not τa ≤ t (by the law
of total probability ).

P (B(t) ≥ a) = P (B(t) ≥ a τa ≤ t) P (τa ≤ t)+P (B(t) ≥ a τa > t) P (τa > t).

Note that if τa > t, the process value cannot be greater than a without
having yet hit a (by continuity of sample paths), so that

P (B(t) ≥ a τa > t) = 0.

36 / 75
Distribution of τa

37 / 75
We obtain,

P (τa ≤ t) = 2 P (B(t) ≥ a)
Z ∞
2 x2
= √ e− 2t dx
2 π t Za
∞ 2 √ √
2 − y2 x
= √ e t dy (by letting y = √ , so dx = t dy)
2 π t a/√t t
Z ∞
2 y2
= √ √
e− 2 dy
2 π a/ t
!
 a 
= 2 1−Φ √ , a > 0, where Φ(x) ∼ N (0, 1).
t

38 / 75
For a < 0, the distribution of τa is the same as that of τ−a by symmetry. We
obtain Z ∞
2 2
− y2
P (τa ≤ t) = √ e dy, for a < 0.
2 π |a|/√t

39 / 75
10.8 Maximum Value of Brownian Motion

40 / 75
Another variable of interest is maximum value of a Brownian motion
attains in [0, t], max0≤s≤t B(s).

Note that B(0) = 0, so the maximum value of Brownian motion over [0, t]
must be non-negative.

For a > 0,


P max B(s) ≥ a = P ( τa ≤ t ).
0≤s≤t

41 / 75
Therefore, the cumulative distribution function of max0≤s≤t B(s) is

P max B(s) ≤ a = 1 − P max B(s) ≥ a
0≤s≤t 0≤s≤t
= 1 − P (τa ≤ t)
!
 a 
= 1−2 1−Φ √
t
 a 
= 2 Φ √ − 1, a > 0, Φ(x) ∼ N (0, 1).
t

42 / 75
10.9 Gambler’s Ruin Problem - Revisited

43 / 75
Let us now consider the probability that a Brownian motion hits a
before −b, where a, b > 0.
We will interpret a Brownian motion as being a limit of a symmetric random
walk.
Recall from the result of gambler’s ruin problem that the probability that
the symmetric random walk goes up to a steps before going down b steps
when each step is equally likely to be either up or down a distance ∆ x is (by
formula Pi = Ni with N = a+b
∆x
and i = ∆bx ) equal to
i b
Pi = = .
N a+b
Hence, letting ∆ x → 0, we see that
b
P (up a before down b) = .
a+b

44 / 75
10.10 Gaussian Processes

45 / 75
Definition
A random vector X1 , X2 , . . . , Xn is said to have a multivariate normal
distribution or a joint normal distribution, if every linear combination of

α 1 X1 + α 2 X2 + · · · + α n X n

is normally distributed, where αi , i = 1, 2, . . . , n, are real constants.

Now, the multivariate normal distribution is specified by its two parameters,

mean value µi = E(Xi ), for every i = 1, 2, . . . , n, and covariance matrix with
entries Cov(Xi , Xj ), i, j = 1, 2, . . . , n.

The condition Cov(Xi , Xj ) = 0 for i ̸= j implies sufficiently that Xi and Xj

are independent random variables.

46 / 75
Definition
A stochastic process {X(t), t ≥ 0} is called a Gaussian process if
X(t1 ), X(t2 ), . . . , X(tn ), with t1 < t2 < · · · < tn , has a multivariate normal
distribution.

Every Gaussian process is uniquely described by its two

parameters: mean and covariance functions.

Conversely, given an arbitrary mean function and covariance

function, then there exists a corresponding Gaussian process.

47 / 75
10.11 Brownian motion is a Gaussian Process

48 / 75
A Brownian motion is a unique Gaussian process with zero mean (i.e.
E(B(t)) = 0) and covariance function, for 0 ≤ s ≤ t,
h

i
Cov(B(s), B(t)) = E B(s) − E( B(s) ) B(t) − E( B(t) )
h

i
= E B(s) − 0 B(t) − 0
h 
= E B(s) B(t)
h i
= E B(s) (B(t) − B(s) + B(s))
h i
= E B 2 (s) + E B(s) (B(t) − B(s))
 

= E B 2 (s) + E B(s) E B(t) − B(s)

     

= σ 2 s since E(B(s)) = 0, and 0 ≤ s ≤ t.

49 / 75
Similarly, if 0 ≤ t ≤ s, we see that Cov(B(s), B(t)) = σ 2 t.

Both cases can be treated as a single expression by

Cov(B(s), B(t)) = σ 2 min(s, t) for s, t ≥ 0

= σ 2 (s ∧ t), where ∧ denotes the smaller of s and t.

50 / 75
Example
If B(t) is a standard Brownian motion, then X(t) = c B(t/c2 ) for a fixed
c > 0 is a version of standard Brownian motion.

51 / 75
Example
If B(t) is a standard Brownian motion, then X(t) = c B(t/c2 ) for a fixed
c > 0 is a version of standard Brownian motion.

Solution:
Assume that B(t) is a standard Brownian motion. Our goal is to show that
X(t) = c B(t/c2 ) for a fixed c > 0 is a version of standard Brownian motion.

That is, we need to show that

1 {Xt , t ≥ 0} is a Gaussian process
2 E(X(t)) = 0 for every t ≥ 0
3 Cov(X(s), X(t)) = s ∧ t for every s, t ≥ 0

51 / 75
Example

Solution - continued:
1 {X(t), t ≥ 0} is a Gaussian process because {B(t), t ≥ 0} is a standard
Brownian and hence Gaussian .



2 E(X(t)) = E c B(t/c2 ) = c E B(t/c2 ) = 0 since for any t ≥ 0,
B(t) ∼ N (0, t) and B(t/c2 ) ∼ N (0, t/c2 ).

Cov(X(s), X(t)) = Cov c B(s/c2 ), c B(t/c2 )

= c2 Cov B(s/c2 ), B(t/c2 )

 s t 
= c2 2 ∧ 2 = s ∧ t.
c c

52 / 75
10.12 Brownian Motion with Drift

53 / 75
Definition
A stochastic process {X(t), t ≥ 0} is said to be a Brownian Motion with
drift coefficient µ and variance parameter σ 2 if
1 X(0) = 0
2 {X(t), t ≥ 0} has stationary and independent increments
3 for every t ≥ 0, X(t) is normally distributed with mean µ t and variance
σ 2 t.

An equivalent definition is to let {B(t), t ≥ 0} be a standard Brownian

motion and define
X(t) = σ B(t) + µ t.

54 / 75
10.13 Geometric Brownian Motion

55 / 75
Definition
Let {Y (t), t ≥ 0} be a Brownian motion with drift coefficient µ and variance
parameter σ 2 . Then
X(t) = eY (t)
is called a geometric Brownian motion.

56 / 75
If Y (0) = 0, the X(0) = eY (0) = e0 = 1.

Since Y (t) ∼ N (µ t, σ 2 t), its moment-generating function is

u2
σ2 t
E(eu Y (t) ) = eu µ t+ 2 .

By setting u = 1 we have
1 2
E(X(t)) = E(eY (t) ) = eµ t+ 2 σ t

and

V ar(X(t)) = E(X 2 (t)) − (E(X(t)))2

= E(e2 Y (t) ) − (E(eY (t) ))2
2 1 2
= e2µ t+2 σ t
− (eµ t+ 2 σ t )2 .

57 / 75
10.14 Application of Geometric Brownian
motion

58 / 75
Samuelson1 (1965) first introduced geometric Brownian motion to represent
the stock price changes.

Geometric Brownian motion is useful in modelling stock prices over time

when the percentage changes are considered independent and identically
distributed.

1
Samuelson, Paul A. (1965) Rational Theory of Warrant Pricing, Industrial
Management Review, 6(2), 13-39.

59 / 75
Given the initial price of the stock price, S(0) = s0 , we can assume that the
price changes follow a geometric Brownian motion,

S(t) = s0 e(µ t+σ B(t)) .

σ2 σ2
Then E(S(t)) = s0 eµ t+ 2
t
= s0 e(µ+ 2
)t
.
σ2
Thus, the expected price grows at µ + 2
under geometric Brownian motion.

60 / 75
Example
Suppose that S(t) is a geometric Brownian motion with drift parameter
µ = 0.01 and volatility parameter σ = 0.2. If S(0) = 100, find
1 E(S(10))
2 P (S(10) > 100)
3 P (S(10) < 110)

61 / 75
Example

Solution:
From the question, we have S(t) = 100 e0.01 t+0.2 B(t) .
2
1 E(S(10)) = 100 exp{(0.01 + 0.22 ) 10} = 100 e0.3 .

P (S(10) > 100) = P (100 exp{0.01(10) + 0.2 B(10)} > 100)

= P (exp{0.1 + 0.2 B(10)} > 1)
= P (0.1 + 0.2B(10) > ln(1))
= P (B(10) > −0.1/0.2)
1 1
= P (B(10) > − ) = 1 − Φ(− √ ) = 0.5628165.
2 2 10
See P (B(t) > a) formula and use the RStudio command 1 − pnorm(−1/(2 ∗ sqrt(10)))
62 / 75
Example

Solution- continued
➌ Similarly,

P (S(10) < 110) = P (B(10) < (log(11/10) − 0.1)/0.2)

 −0.0234491 
= Φ √
10

= pnorm(−0.0234491, 0, sqrt(10)) from RStudio

= 0.4970418

63 / 75
10.15 Martingales

64 / 75
Definition
A stochastic process {Y (t), t ≥ 0} is said to be a martingale process if, for
s<t  
E Y (t) Y (u), 0 ≤ u ≤ s = Y (s). (9)

Y (u), 0 ≤ u ≤ s is called the history or information up to time s.

This definition tells us the best guess of the expected value of Y (t), at
time t, given the history (or information) up to time s is the current
value of Y (s) at time s.

65 / 75
There are other conditions to the definition of a martingale, but it is
beyond the level of this course.
Martingales are known as a type of betting strategy.
There exists a rich and growing theory on martingales. They are very
useful in studying stochastic processes, especially in mathematical
finance.

66 / 75
Example
Let {B(t), t ≥ 0} be a standard Brownian motion. Show that B(t) is a
martingale. That is, for s < t,

E(B(t) B(u), 0 ≤ u ≤ s) = B(s).

67 / 75
Example

Solution:
For s < t,
 
E(B(t) B(u), 0 ≤ u ≤ s) = E B(t) − B(s) + B(s) B(u), 0 ≤ u ≤ s

= E B(t) − B(s) B(u), 0 ≤ u ≤ s
 
+ E B(s) B(u), 0 ≤ u ≤ s


= E B(t) − B(s) by independent increments
+ B(s) expectation of a constant is a constant.

= 0 + B(s) the expected value of of standard Brownian motion is zero

= B(s).

Therefore, {B(t), t ≥ 0} is a martingale.

68 / 75
Supplementary Material

69 / 75
Supplementary Material - Chapman-Kolmogorov Equation

Recall that we established

P(X0 = i0 , X1 = i1 , . . . , Xn = in ) = Pin−1 ,in Pin−2 ,in−1 · · · Pi0 ,i1 P(X0 = i0 ).

Write Pijn as the probability of the process going from state i to state j in n
steps. That is,

Pijn = P(Xn+m = j|Xm = i).

This is known as the n step transition probabilities. The next proposition is

known as the Chapman-Kolmogorov equation, and it is a method of
calculating n transitional probabilities.

70 / 75
Supplementary Material - Chapman-Kolmogorov Equations

For any r ≤ n,
∞
X
n−r
Pijn = Pikr Pkj . (10)
k=0

71 / 75
Supplementary Material - Chapman-Kolmogorov Equations - Matrix Form

If we let P(n) denote the matrix of n-step transition probability Pijn , then

P(n) = P(r) · P(n−r) , for any r ≤ n.

In particular, P(2) = P(1) · P(1) = P · P = P2 .

By induction, P(n) = P(n−1) · P(1) = P(1) · P(n−1) = Pn .

That is, the n-step transition matrix is the nth power of the one-step
transition matrix.

Note: P(0) = P0 = I, where I is the identity matrix.

72 / 75
Supplementary Material - CLT

Central Limit Theorem

Suppose X1 , X2 , . . . are independent and identically distributed random
variables with common mean µ = E(Xi ) and common variance
σ 2 = V ar(Xi ) < ∞.

For each n ≥ 1, let X̄n = ni=1 Xi . Then

P

X̄n − µ D
√ − → Z,
σ/ n
where Z ∼ N (0, 1). It is common to write

X̄n − µ D
√ − → N (0, 1).
σ/ n

Memorise this! 73 / 75
Supplementary Material - Law of Total Probability

Law of Total Probability

Suppose that A1 , A2 , . . . , Ak are mutually exclusive events (i.e. Ai ∩ Aj = ∅
for all i ̸= j) and exhaustive events (i.e. ki=1 Ai = S, in order words,
S
A1 , A2 , . . . , Ak form a partition of S.) Then, for any event B,
k
X
P (B) = P (B Ai )P (Ai ).
i=1

74 / 75
Supplementary Material - Convergence in Distribution

Definition
Let X1 , X2 , . . . be a sequence of random variables. We say that Xn
converges in distribution to X if

lim FXn (x) = FX (x)

n→∞

for all x, where FX is continuous.

A common shorthand is
D
Xn −
→ X.

We say that FX is the limiting distribution of Xn .

Note that both X and Xn are random variables.

Memorise this! 75 / 75