20-01-2016

CL 202: Introduction to Data Analysis

Fundamentals of Probability
and Random Variables
Mani Bhushan and Sachin Patawardhan
Department of Chemical Engineering
I.I.T. Bombay

1/20/2016

Fundamentals

Automation Lab
IIT Bombay

Outine

Sample Space

Borel Field and Probability Measure

Probability Space

Computing Probabilities

Concept of a Random Variable

Discrete and Continuous Random Variables

Properties of Random Variables

Appendix: Conditional Probability and Independence

1/20/2016

Fundamentals

20-01-2016

Automation Lab
IIT Bombay

Note

The material in this presentation is composed from

multiple sources. The references are listed at the
end. If you are looking for one reference text that
contains almost every concept covered here then
refer to the following standard textbook:
Papoulis, A. and Pillai, S. U., Probability, Random
Variables and Stochastic Processes, (4th Ed.),
MacGraw-Hill International, 2002.

1/20/2016

Fundamentals

Probability

Automation Lab
IIT Bombay

(Maybeck, 1979)

Intuitive approach to define probabilities of events of

interest in terms of the relative frequencies of occurrence
If the event A is observed to occur N(A) times in
a total of N trials, then P(A) is defined by

P ( A)

lim

N ( A)
N N

provided that this limit in fact exists.

Although this is a conceptually appealing basis for probability
theory, it does not allow precise treatment of many problems
and issues of direct importance.
Modern probability theory is more rigorously based
on an axiomatic definition of the probability.
1/20/2016

Fundamentals

20-01-2016

Sample Space (Maybeck, 1979)

Automation Lab
IIT Bombay

S : fundamenta l sample space containing all possible

outcomes of the experiment conducted

: single elementary outcome of the experiment

(i.e. S )

A : a specific event of interest, a specific set

of outcomes of the experiment.
Each such event A is a subset of S , i.e. A S
An event A is said to occur if the observed outcome
is an element of A, i.e. if A .
1/20/2016

Fundamentals

Sample Space

Automation Lab
IIT Bombay

Discrete Sample space: consists of a finite or

countably infinite number of elements/outcomes

Examples: (1) Coin toss or roll of a die experiments,

(2) set of manufacturing defects in a device

Continuous Sample Space: consists of

uncountable number of elements

Examples: (1) Values measurement noise can take in a

sensor, (2) Yield of a reaction, (3) Monthly profit of a
company

1/20/2016

Fundamentals

20-01-2016

Field

(Papoulous and Pillai, 2002)

Automation Lab
IIT Bombay

Field (F)
A field F is a non - empty class of sets such that if
( a ) A F then A F and
(b) A, B F then A B F
Using these properties, it can be shown that
(a) If A,B F then A B F
________

A B F A B A B F
(b) The field contains the " certain event" ( S A A )
and the " impossible event" ( A A )

Example: Consider experiment of rolling a die once

We can define sample space as S 1,2,3,4,5,6
1/20/2016

Fundamentals

Automation Lab
IIT Bombay

Borel Field (Papoulous and Pillai, 2002)

Example (contd)
Now consider a class of sets, F, defined as
F , {2},{1,3,4,5,6}, 1,3,5
, 1,2,3,5
, 2,4,6
, 4,6, S
The class F qualifies to be a field.

Borel Field (B)

Suppose A1 A2 ,...... An .... is an infinite
sequence of sets in field B. If the unions and
intersections of these sets also belong to B
then B is called a Borel field.
A probability measure is defined only on a Borel field.
1/20/2016

Fundamentals

20-01-2016

Automation Lab
IIT Bombay

Example 1: Rolling of a Die (Jazwinski, 1970)

Consider experiment of rolling a die (with faces
numbered from 1 to 6) once

We can define probability (or sample) space as

S1 1,2,3,4,5,6
Now, a Borel field can be defined in multiple ways.

Case 1 : Borel field B1 can be taken as set of all

subsets of S1 (i.e. power set of S1 )

Case 2 : If we are interested in betting on only events

(odd) and (even), then Borel field B2 can be defined as
B2 , 1,3,5
, 2,4,6, S1
1/20/2016

Fundamentals

Automation Lab
IIT Bombay

Example 1: Rolling of a Die

Now consider a set F defined as
F , {2}, 1,3,5
, 2,4,6, S1
Note : {2} 1,3,5 1,2,3,5 F
even though 1,2,3,5 S1 and S1 F.
Thus set F does not qualify as a Borel field.

Example 2: Error in temperature measurement

This is an example of a continuous RV.
Theoretically, the measurement error can take any
real value.

Sample space S R i.e. real line.

1/20/2016

Fundamentals

10

20-01-2016

Example 2: Measurement Errors

Automation Lab
IIT Bombay

Let e1 and e2 be points on R such that e1 e2 .

Then, the following sets belong to the Borel Field, B,
which consists of events of the form

Event 1 : error is less than or equal to e1

A1 : e1 , R ( , e1 ]
Event 2 : error is less than or equal to e2
A2 : e2 , R ( , e2 ]
Event 3 : Error lies between e1 and e2
A1* : e1 , R ( , e1 ]* ( e1 , )
A1* A2 ( e1 , e2 ]
Event 4 : error is equal to e1
1/20/2016

Fundamentals

11

Automation Lab
IIT Bombay

Axioms of Probability
The probability function (or probability measure) P (.)
is defined to be a real scalar - valued function
defined on Borel field B that assigns a value, P( Ai ),
to each Ai which is a member of B ( Ai B) such that :
1. P( Ai ) 0 for all A i B
2. P( S ) 1
3. If A1,A2 ,...AN are disjoint or mutually exclusive
N

i 1

i 1

elements of B then P( Ai ) P(Ai )

for all finite and countably infinite N.
1/20/2016

Fundamentals

12

20-01-2016

Note (Papoulous and Pillai, 2002)

Automation Lab
IIT Bombay

It may be noted that, the axioms of probability are

so chosen that the resulting theory gives a
satisfactory representation of the physical world.
Probabilities as used in real problems must,
therefore, be compatible with the axioms. Using the
frequency interpretation of probability, it can be
shown that they do.
1. P ( A) 0 because N ( A) 0 and N>0.
2. P ( S ) =1 because S occurs at every trial; hence N ( S )=N .
3. If A B = { }, then N ( A B ) N ( A) N ( B ), because
if A B occurs, then A or B occurs but not both.
N ( A B ) N ( A) N ( B )
Hence P( A B )

P(A)+P(B)
N
N
1/20/2016

Fundamentals

Example: Dart Board

13

Automation Lab
IIT Bombay

Event of interest: Concentric Circles

A7

A2
Cones A1,A2 ,..., A7 ,...

A1

are disjoint or
mutually exclusive
sets : Relatively easy
to assign a probability
value to each set
1/20/2016

Sample Space (S): Set of all

Points on the Dart Board
Fundamentals

14

20-01-2016

Automation Lab
IIT Bombay

Probability Space
Probability space : Defined by the triplet ( S, B, P)
of the sample space, the underlying Borel field, and
the probability function, all defined axiomatically.

Example 1: Rolling of a Die (contd.)

Consider sample space S1 1,2,3,4,5,6
Case 1 : Borel field B1 power set of S1
Consider disjoint sets Ai {i} for i 1,2,..,6.
If we set P1 ( Ai ) 1 / 6 for i 1,2,..,6
Then we can find probability of any event in B1
1/20/2016

Fundamentals

15

Automation Lab
IIT Bombay

Example 1: Rolling of a Die (contd.)

Thus, triplet ( S1,B1,P1 ) forms a probability space.

Case 2 : If we are interested in betting

on only events (odd) and (even),
and Borel field B2 is defined as
B2 , 1,3,5
, 2,4,6, S1
Define
P2 1,3,5 p and P2 2,4,6 q
P2 ( ) 0 and P2 ( S1 ) p q 1, with p,q 0
Triplet ( S1,B 2 ,P2 ) forms a probability space
1/20/2016

Fundamentals

16

20-01-2016

Automation Lab
IIT Bombay

Example 1: Rolling of a Die (contd.)

Case 3 : We can define S2 R i.e. set of all real numbers.
Borel field B3 Set of all subsets of R
P3 ((-,a]) = (1/ 6) {No. of pts. 1,2,...,6 which (-, a] contains}
P : 2.5 2 (1/ 6) ; P : 5.3 5 (1/ 6)

P3 : 1.9 5.3 4 (1 / 6)
P3 : 1 0 ;

P3 : 10 1

Triplet ( S2 ,B 3 ,P3 ) forms a probability space

1/20/2016

Fundamentals

17

Automation Lab
IIT Bombay

Points to Note
PrA 0 does NOT imply A
PrA 1 does NOT imply A S

Definition of a probability space for a given

experiment is NOT unique.
Triplet ( S, B, P) must be 'specified' and
NOT determined by the physical experiemnt.

1/20/2016

Fundamentals

18

20-01-2016

Automation Lab
IIT Bombay

Computing Probabilities
Consider a probability sapce ( S , B, P ).
Given an event A B, in some cases, it is easier to find P ( A ) than P ( A).
Then, using the fact A A S and ( A, A ) are disjoint, we can write
P( S ) P ( A) P ( A ) 1

P( A) 1 P ( A )

Odds of event A P ( A) / P( A ) P( A) / 1 P ( A)

Non-Disjoint Events
Let A and B be two events in B that are NOT disjoint.

We can write
A A B A B
which are disjoint.
P ( A) P A B P A B ........( I )
1/20/2016

Fundamentals

19

Automation Lab
IIT Bombay

Computing Probabilities
Non-Disjoint Events (contd.)

Similarly, we can write

A B A B B A B B
B A B
which are disjoint.
P( A B ) P B P A B ....... ( II )

Eliminating P A B from (II) using (I), we have

P ( A B ) P A P ( B ) P ( A B )
1/20/2016

Fundamentals

20

10

20-01-2016

Example (Ross, 2009)

Automation Lab
IIT Bombay

A total of 28 percent of American males smoke

cigarettes, 7 percent smoke cigars, and 5 percent smoke
both cigars and cigarettes.
What percentage of males smoke neither cigars nor
cigarettes?
Event A: a randomly chosen male is a cigarette smoker
Event B: a randomly chosen male a cigar smoker.
P ( A B ) P( A) P( B ) P ( A B )
0.28 0.07 0.05 0.3

Thus the probability that the person is not a smoker

is .7, implying that 70 percent of American males
smoke neither cigarettes nor cigars.
1/20/2016

Fundamentals

Product of Sample Spaces

21

Automation Lab
IIT Bombay

In statistics one usually does not consider a

single experiment, but that the same experiment
is performed several times.
When we perform an experiment n times,
then the corresponding sample space is
S S1 S2 S3 .... Sn
where Si for i 1, . . . , n is a copy
of the sample space of the original experiment
Probability of the combined outcome (1 , 2 ,...., n )
P (1 , 2 ,...., n ) p1. p2 .... pn
where each i has probability pi
1/20/2016

Fundamentals

22

11

20-01-2016

Example: Repeated Coin Toss

Automation Lab
IIT Bombay

Sample space associated with coin toss experiment

S H , T
Let P ( H ) p and P (T ) q 1 p
0 p, q 1
When we perform the coin toss experiment n times
S H , T H , T ...... H , T
Consider the situation when n 3. Then
P ( H , H , H ) p 3
P ( H , T , H ) p.q. p p 2 (1 p )
Dekking et al., 2005
1/20/2016

and so on.
Fundamentals

Example: Infinite Outcomes

23

Automation Lab
IIT Bombay

Experiment with infinitely many outcomes

Toss a coin repeatedly until the first head turns up
Outcome of the experiment
No. of tosses it takes to have the first success
(i.e. occurrence of H )
S 1,2,3,......

For i 3, we have
P (3) P (T , T , H ) p(1 p )2
..........
For i n, we have
P ( n) P (T , T ,...., T , H ) p(1 p ) n 1
1/20/2016

Fundamentals

24

12

20-01-2016

Example Infinite Outcomes

Automation Lab
IIT Bombay

It may be noted that events 1,2 ,..., n,... i.e.

( H ), (T , H ), (T , T , H ),......(T , T ,...., T , H ),.....
are disjoint or mutually exclusive events.
Thus, from axioms of probability, it follows that
P ( S ) P(1) P( 2) .... P( n ) ....
p (1 p ) p .... (1 p ) n 1 p ....

p 1 (1 p ) (1 p ) 2 ......

1
p
1
1 (1 p )
1/20/2016

Fundamentals

25

Automation Lab
IIT Bombay

Random Variable
Problem 1: A sample space associated with a random
phenomenon need not consist of elements that are
not numbers.
Example 1: A die with 6 faces painted with 6
different colors
Example 2: Candidates appearing in an election held
in a constituency
How to perform numerical calculations
involving such sample spaces?
1/20/2016

Fundamentals

26

13

20-01-2016

Automation Lab
IIT Bombay

Random Variable
Problem 2: We often have to consider multiple random
phenomena simultaneously. Even if the sample spaces
associated with these random phenomenon consists of
numbers, their ranges can be widely different.
Example: Temperature, pressure and feed
concentration fluctuations in a chemical reactor.
To understand the reactor behavior, these
multiple random phenomenon have to be
considered simultaneously.
How can we treat such problems through a
unified mathematical framework?
1/20/2016

Fundamentals

27

Automation Lab
IIT Bombay

Random Variable
It is possible to define a transformation
such that we can perform all the calculations using
a generic sample space
and
a generic Borel field
defined using the generic sample space?
The concept of a random variable is introduced
because, we need a mapping from the sample space
to the set of real numbers for carrying out
quantitative analysis through
a unified mathematical framework.
1/20/2016

Fundamentals

28

14

20-01-2016

Random Variable (Maybeck, 1979)

Automation Lab
IIT Bombay

Given a sample space, S , and

an associated Borel field B.
A scalar random variable x( )
is a real - valued point 'function or mapping'
which assigns a real sclar value to each point in S,
denoted as x() , such that every set A of the form
A { : x() x }
for any value a on the real line ( x R)
is an element of the Borel field B.
1/20/2016

Fundamentals

Random Variable

29

Automation Lab
IIT Bombay

x( ) : Random variable (mapping)

: a realization of the random variable

(i.e. the value that this function assumes for a particular )

A scalar random variable

is a mapping from the sample space, S , into R
such that inverse images of
half - open intervals of the form (-, x] in R
are events in B for which
the probabilities can be defined
through a probability function P(.).
1/20/2016

Fundamentals

30

15

20-01-2016

Automation Lab
IIT Bombay

Advantage

Once we define a random variable x( )

on an original sample space, say S ,
we can start working with
the 'generic' sample space S R R
Original Borel Field B
Generic Borel Field B R
(consisting of all sub - intervals of R)
A generic elementary event A in S R
A x : x x with x, x R ( , x ]
1/20/2016

Fundamentals

The Generic Borel Field

31

Automation Lab
IIT Bombay

Let x1 and x2 be points on R such that x1 x2 .

Then, the following sets belong to the Borel Field, B R

A1 x x1 ( , x1 ]
A2 x x2 ( , x2 ]
A x x1 , R ( , x1 ]* ( x1 , )
*
1

A1* A2 ( x1 , x2 ]
Taking complement s, unions, and intersections of the sets Ai
leads to finite intervals (open, closed, half open) and point values.

The Borel field, BR , is composed of virtually all sub - intervals

of R of interest in describing a probability problem on R.
1/20/2016

Fundamentals

32

16

20-01-2016

Automation Lab
IIT Bombay

Advantage

A generic probability function, P(.), is defined

for all events of the form A ( , x ]
Fx ( x ) P(( , x ])
where 0 1
Fx ( x ) : Probability Distribution Function
Thus, from now on,
we work with probability space ( R,B R ,Fx ( x ))
for all problems.
1/20/2016

Fundamentals

33

Automation Lab
IIT Bombay

Properties of Distribution Function

Notation
Fx ( x )

lim
lim
Fx ( x ) and Fx ( x )
F (x )
0
0 x
for 0 0

Property 1
F ( ) 1 and F ( ) 0
Property 2
Fx ( x ) is a non - decreasing function of x,
i.e., if x1 x2 then Fx ( x1 ) Fx ( x2 ).
1/20/2016

Fundamentals

34

17

20-01-2016

Automation Lab
IIT Bombay

Properties of Distribution Function

Property 3
If Fx ( x0 ) 0 then Fx ( x ) 0 for all x x0 .
Property 4
The function Fx ( x ) is continuous from the right, i.e.,
Fx ( x ) Fx ( x ).

Property 5
P(x x ) Fx ( x ) Fx ( x ).

Property 6
P( x1 x x2 ) Fx ( x2 ) Fx ( x1 )
1/20/2016

Fundamentals

Points to Note

35

Automation Lab
IIT Bombay

Continuous Random Variable

A random variable is called contunuous type if
the distribution function Fx ( x ) is continuous, i.e.,
Fx ( x ) Fx ( x ) Fx ( x ) for all x.

Discrete Random Variable

If Fx (x ) is constant except for a finite number of
jump discontuities (piecewise constant, step type),
then x is said to be discrete - type RV.
It is possible to encounter a situation where
a random variable is mixed type.
1/20/2016

Fundamentals

36

18

20-01-2016

Example: Coin Toss

Automation Lab
IIT Bombay

Consider the coin toss experiment with triplet

S H , T ; B , H , T , S ;
P( H ) p, P(T ) q, p q 1, 0 ( p, q) 1
We can define a discrete RV as
x( H ) 1 and x(T ) 0
New sample space is R and associated Borel field is B R

New distribution function

0 for x 0

Fx ( x ) q for 0 x 1
1 for 1 x

1/20/2016

Fundamentals

Probability of Other Events

37

Automation Lab
IIT Bombay

Probability of event {x x} i.e. set ( x, )

P( x, ) 1 P (( , x ]) 1 Fx ( x )
(This follows from the axiom of probability P( S ) 1)

Probability of event { x1 x x2 } i.e. set ( x1 , x2 ]

( , x2 ] ( , x1 ] ( x1 , x2 ]
Note : Sets ( , x1 ] and ( x1 , x2 ] are disjoint
P( , x2 ] P( , x1 ] P( x1 , x2 ]
(follows from axioms of probability)
P( x1 , x2 ] Fx ( x2 ) Fx ( x1 )
1/20/2016

Fundamentals

38

19

20-01-2016

Automation Lab
IIT Bombay

Example 1: Rolling a Die (contd.)

Consider experiment of rolling a die (with 6 faces
painted with 6 different colors) once.

Original sample space S4 C1, C2, C3, C4 , C5 , C6

A Borel field, say B 4 , defined on the original
sample space, S , contains sets of the form

C1, C2 , C3, C4 , C5, C6

C1, C3 , C6 , C2, C4 , C5
.........
and so on
in addition to and S .
1/20/2016

Fundamentals

39

Automation Lab
IIT Bombay

Example 1: Rolling a Die (contd.)

We can define a discrete RV
x(Ci ) 10i
New sample space S R R
An event, say A, in the Borel field B4
~
can now be described using an event A in B R
Events in B R are intervals of the form
~
A x x ( , x ] where x R
or
~
A x1 x x2 [ x1 , x2 ] where x1 , x2 R, x1 x2
and so on.
1/20/2016

Fundamentals

40

20

20-01-2016

Automation Lab
IIT Bombay

Example 1: Rolling a Die (contd.)

Event {x 35} in B R Event C1, C2, C3 in B 4
because x(Ci ) 35 only if i 1,2,3
Event {x 6} in B R Event in B 4
because there is no outcome x(Ci ) 6

Event {17.5 x 46.8} in B R Event C2, C3, C4 in B4

because 17.5 x(Ci ) 46.8 only for i 2,3,4
Event {20 x 45} in B R Event C2, C3, C4 in B 4
because 20 x(Ci ) 45 only for i 2,3,4
1/20/2016

Fundamentals

41

Automation Lab
IIT Bombay

Example 1: Rolling a Die (contd.)

Event {x 50} in B R Event C5 in B 4
because x(Ci ) 50 only if i 5

Event {x 28} in B R Event in B 4

because there is no outcome such that x(Ci ) 28
Event {x 80} in B R Event S in B4
Note
(1) An event in B4 can correspond to
Multiple events in B R
(2) Event in B R is NOT the same as event in B4
1/20/2016

Fundamentals

42

21

20-01-2016

Automation Lab
IIT Bombay

Probability Measured
A Probability Measure on B 4
P (Ci ) 1 / 6 for i 1,2,...,6

An equivalent probability measure on B R

No. of pts. 10,20,..., 60
Fx ( x ) P((-,x]) = (1/ 6)

which (-, x] contains

Note :
Unlike P(Ci ),
Fx ( x ) is a continuous function
defined over entire R
(with discontinuities
only at 6 points)
1/20/2016

Fundamentals

Example 1: Rolling a Die (contd.)

43

Automation Lab
IIT Bombay

Distribution function of discrete RV x(Ci ) 10i

is a staircase function.
In particular
Fx (38) P {- x 38} in B R PC1, C2 , C3 in B 4 3

1
6

1
6
1
Fx (29.5) P{- x 29.5} in B R PC1, C2 in B4 2
6
Fx (6) P {- x 6} in B R P{ } in B4 0

Fx (31.9) P{- x 31.9} in B R PC1, C2, C3 in B 4 3

Fx (82) P {- x 82} in B R P{S4 } in B 4 1

1/20/2016

Fundamentals

44

22

20-01-2016

Probability Mass Function

Automation Lab
IIT Bombay

Probability mass function of a discrete RV, x,

is a function f x ( .) : R [0,1] defined by
f x ( a ) P( x=a ) for - < a < .
1/20/2016

Fundamentals

45

Probability Mass Function

Automation Lab
IIT Bombay

If x takes values a1,a2 ,... then we have

f x ( ai ) P ( x ai ) pi 0
and p1 p2 ....=1 with 0 p1 , p2 ,.... 1
and f x ( x ) P( x x ) 0
for all other x R
Thus, the probability / Cumulative Distribution Function
of a discrete RV, x, is related to the probability mass
function of, x, as follows
F ( x ) P(x x )

( ai )

i , ai x
1/20/2016

Fundamentals

46

23

20-01-2016

Automation Lab
IIT Bombay

Probability Mass Function

The definition of probability mass function on

the previous slide is intuitively appealing.
However, it does not facilitate treatment
through calculus.

To facilitate treatment through calculus,

the probability mass function for
a discrete RV can be represented as
f x ( x ) pi ( x ai ) for - x
i

where ( x ai ) represents the Dirac delta function.

1/20/2016

Fundamentals

47

Automation Lab
IIT Bombay

Probability Mass Function

The probability mass function is related to

probability distribution function through
the following integral equation
x

Fx ( x )

x
x

( )d

p ( a )d
i

pi ( ai )d
i

( ai )

i ,ai x

(This follows from the properties of ( x ai ))

1/20/2016

Fundamentals

48

24

20-01-2016

Example: Data Transmission

Automation Lab
IIT Bombay

There is a chance that a bit transmitted through a

digital transmission channel is received in error.
Define discrete RV, x, equal to the number of bits
in error in the next four bits transmitted.

Original sample space S 0,1,2 ,3,4

Original Borel field Power set of S
Probabilities Defined with reference to S
P(1= 0)=0.6561, P(2=1)=0.2916, P(3=2)=0.0486
P(4= 3)=0.0036, P(5=4)=0.0001
1/20/2016

Fundamentals

Example: Data Transmission

49

Automation Lab
IIT Bombay

Let us define a discrete RV

x(i ) i 1 for i 1,2,3,4,5
New sample space R and new Borel field B R

Probability Mass Function

f x (0)=0.6561, f x (1)=0.2916, f x (2)=0.0486
f x (3)=0.0036,

f x (4)=0.0001

Probability/Cumulative Distribution Function

0 for x 0

Fx ( x ) 0.6561 for 0 x 1
0.9477 for 1 x 2

1/20/2016

0.9963 for 2 x 3

Fx ( x ) 0.9999 for 3 x 4
1
for 4 x

Fundamentals

50

25

20-01-2016

Example 2: Telephone Call

Automation Lab
IIT Bombay

A telephone call occurs at random

during time interval [0 ,T ].
Original Sample Space ( S ) [0,T ]

Original Borel Field

Set of all sub - intervals [t1,t2 ]
B

defined over interval [0, T ]

Probability Function on B
P (t1 t2 )
1/20/2016

t2 t1
for any 0 t1 t2 T
T
Fundamentals

Example 2: Telephone Call

51

Automation Lab
IIT Bombay

Define a continuous RV as
x( ) when [0 ,T ]
New sample space ( S R ) R
Event [0, t ] B Event x ( , t ] B R
(for 0 t T )

An event in B R , x ( , a ], where a T
is associated with the certain event, S , in B.
An event in B R , x ( , t ], where t 0
is associated with the impossible event in B.
1/20/2016

Fundamentals

52

26

20-01-2016

Example 2: Telephone Call

Automation Lab
IIT Bombay

0 if - t 0

Define Fx (t ) P( x t ) (t/T ) if 0 t T
1
if T t

Now consider t1 , t2 [0, T ] such that t1 t2

Suppose we want to find out the probability
of occurrence of a call in interval (t1 x t2 )
Since ( x t2 ) ( x t1 ) (t1 x t2 )
and events ( x t1 ) and (t1 x t2 ) are disjoint,
from the axioms of probability, it follows that
P ( x t2 ) P ( x t1 ) P (t1 x t2 )
1/20/2016

Fundamentals

53

Example 2: Telephone Call

Automation Lab
IIT Bombay

P (t1 x t2 ) P( x t2 ) P (x t1 )
Thus, P(t1 x t2 ) Fx (t2 ) Fx (t1 )

t2 t1
,
T

Note
Now, suppose t1 t and t2 t
where 0 is a small number, then
2
P(t x t )
0 as 0
T
This is an example of a continuous random variable.
Which attributes qualify a RV to be called continuous?
1/20/2016

Fundamentals

54

27

20-01-2016

Continuous Random Variable

Automation Lab
IIT Bombay

A random variable, x, is called continuous

if there exists a density function f X ( x ) : R R
such that for any a,b R with a b
b

P ( a x b) f X ( x )dx ......(1)
a

f X ( x ) must satisfy f X ( x ) 0 x and

( x )dx 1

Distribution function of a continuous RV, x,

is a function F (.):R [0 ,1] defined by
a

F (a ) P( x a )

( x )dx

1/20/2016

Fundamentals

Probability Density Function

55

Automation Lab
IIT Bombay

Typical f x (x )

Area under the probability

density function f x ( x ) in [a, b]
probability that the RV x
will lie in [a,b]
1/20/2016

Fundamentals

56

28

20-01-2016

Automation Lab
IIT Bombay

Note (Dekking et al., 2005)

If the interval gets progressively smaller, then

the probability will tend to zero.

For any arbitrarily small >0, we have

a

P(a x a )

( x )dx

As 0, it follows that P( a ) 0
This implies that for a continuous random variable,
we need not be precise about of the intervals :
P(a x b)=P (a<x b)=P (a<x<b)=P (a x<b)
1/20/2016

Fundamentals

57

Automation Lab
IIT Bombay

Note

For any arbitrarily small >0, we can write

a

( x )dx 2 f X ( x )

Thus, f X ( x ) can be interpreted as a (relative) measure

of how likely it is that RV, x, will be near a
Important to note :
In general, f x (a ), can have a very large value
and is NOT the value of probability of x at a.
1/20/2016

Fundamentals

58

29

20-01-2016

Uniform Distribution

Automation Lab
IIT Bombay

Generalization of PDF appearing in Telephone Call example

0 if - x a

Fx ( x ) ( x/ (b a )) if a x b

1
if b x

The distribution is not

differentiable at a and at b.

Differentiating Fx ( x ), we get
0

f x ( x ) 1 /(b a )

xa
a xb
xb
Uniform Density Function

1/20/2016

Fundamentals

Example: Chemical Reactor

59

Automation Lab
IIT Bombay

Consider steady state operation of

a Continuously Stirred Tank
Reactor (CSTR)
V Reactor Volume
q effluent volumetric flow rate
inlet feed flow rate
Random Variable of interest
x residence time of a
particle in the vessel

1/20/2016

Assumption: perfect mixing,

particle's position is uniformly
distributed over the volume
Fundamentals

60

30

20-01-2016

Example: Chemical Reactor

Automation Lab
IIT Bombay

Suppose an elemental volume, v, entered the CSTR at t 0

Question : How long does v stay in the reactor?

Consider interval [0, t ] when t is not too large.

Let the interval [0,t] be divided into n small intervals
each of equal length t t/n
Consider a volume element v qt which enters the
reactor during an interval t.
(Since the system is at steady state, equal amount leaves
the reactor during t and the total volume is still V .)
1/20/2016

Fundamentals

Example: Chemical Reactor

61

Automation Lab
IIT Bombay

Let us assume that what happens in each interval t is

qualitatively similar to the repeated coin toss experiments .

Thus, during each t , the sample space S consists of

1 success ( v leaves the reactor)

2 failure ( v stays in the reactor)
Since the reactor is assumed to be well mixed,
the probability that v leaves the vessel during
any of the n intervals of length t is
P (1 ) p ( v / V ) qt / V qt /(nV )
1/20/2016

Fundamentals

62

31

20-01-2016

Automation Lab
IIT Bombay

Example: Chemical Reactor

Thus, the probability that v stays in the vessel during

any of the n intervals of length t is
P (2 ) 1 p
Sample space (assumption n ) :

No. of intervals (or " tosses" ) it takes to have

the first occurrence of 1 S 1,2,3,......
Thus, the probability that an elemental volume, v,
which entered the CSTR at t 0 is still in the vessel
at least upto time t is, for large n, well approximat ed by
qt
P( x t ) P (2 , 2 ,...., 2 ) (1-p ) 1
nV

1/20/2016

Fundamentals

Example: Chemical Reactor

63

Automation Lab
IIT Bombay

By letting n
P (x t )

lim qt 1 n
qt
1 exp
n V n
V

It follows that the distribution function of x equals

for t 0
0

qt

Fx (t ) P(x t )
1 exp for t 0

V
and associated probability density function is
for t 0
0

qt
f x (t ) q
exp for t 0
V
V

This is an example of the exponentia l distribution.

1/20/2016

Fundamentals

64

32

20-01-2016

Automation Lab
IIT Bombay

Exponential Distribution

fx (t )

Fx (t )

t
Probability Density Function

1/20/2016

Probability Distribution Function

Fundamentals

Exponential Distribution

65

Automation Lab
IIT Bombay

Probability distribution function

for t 0
0
Fx (t ) P(x t )
1 exp t for t 0
Probability density function
for t 0
0
f x (t )
exp t for t 0
Useful in describing probabilities
associated with many engineering problems
For example, x = lifetime of an equipment/component
(i.e. time before which the equipment/component fails)
1/20/2016

Fundamentals

66

33

20-01-2016

Automation Lab
IIT Bombay

Points to Note

A discrete RV does not have a probability density function

and
continuous RV does not have a probability mass function.
However, both have a distribution function and

F ( a ) P ( x a ) P x ( , a ]

( x )dx

For a continuous or discrete RV, it follows from

the integral calculus that
dFx ( x )
dx
when the derivative exists.
f x ( x)

1/20/2016

Fundamentals

Points to Note

67

Automation Lab
IIT Bombay

A continuous RV is defined using the integral

equation (1) on slide 51. The RV definition does
not require the density function to be
continuous and differentiable at all points on R.
Thus, a valid probability density function may
have discontinuities at isolated points on the
real line.
Histogram of a continuous RV can be viewed as
an approximation of the probability density
function. The relative frequency is an estimate
of the probability that a measurement falls in
the interval.
1/20/2016

Fundamentals

68

34

20-01-2016

Automation Lab
IIT Bombay

Histogram

A typical histogram of a continuous Random Variable [1]

1/20/2016

Fundamentals

Automation Lab
IIT Bombay

Summary

69

The modern axiomatic definition of the probability

facilitates rigorous mathematical treatment of
the random phenomenon.
A probability space consists of the triplet (i)
sample space, (ii) a Borel field defined on the
sample space and (iii) a probability measure
defined on each event in the Borel field.
The concept of a random variable is introduced
because, we need a mapping from the sample space
to the set of real numbers for carrying out
quantitative analysis through a unified
mathematical framework.

1/20/2016

Fundamentals

70

35

20-01-2016

Automation Lab
IIT Bombay

Summary

After defining a random variable mapping

we work with probability space ( R,B R ,Fx ( x ))
for all problems.

Moreover
a generic probability function, P(.), is defined
for all events of the form A ( , x ]
Fx ( x ) P(( , x ])

where 0 1

for all problems.

1/20/2016

Fundamentals

71

Automation Lab
IIT Bombay

Appendix
Conditional Probabilities
and Independence
(Dekking et al., 2005)

1/20/2016

Fundamentals

72

36

20-01-2016

Conditional Probability

Automation Lab
IIT Bombay

Consider a probability sapce ( S , B, P ).

Let A and B represent two events in B.
Conditional Probability
Knowing that an event B has occurred sometimes
forces us to reassess the probability of event A. The
new probability is the conditional probability.
Independence
If the conditional probability of A equals what the
probability of A was before, then events A and B
are called independent.
1/20/2016

Fundamentals

Conditional Probability

73

Automation Lab
IIT Bombay

The conditional probability of event A

given event B has occurred is defined as
P( A B )
P( B )
provided P( B ) 0.
P( A | B )

Note
P( A | B )
P( A | B ) P( A | B )

1/20/2016

P( A B)
P( B )
P( A B ) P( A B )

P( B )
P( B )

P ( A B ) ( A B ) P( B )

1
P( B )
P( B )
Fundamentals

74

37

20-01-2016

Conditional Probability

Automation Lab
IIT Bombay

Note
The conditional probability function
satisfies all the axioms of probability, and,
thus, is a valid probability function in itself.
Multiplication rule for any events A and B
P( A B ) P( A | B ) P( B ) P( B | A) P( A)
Consider event A, which can be expressed as
A ( A B) ( A B )
the probability of event A
P( A) P ( A B ) P ( A B )
P( A | B ) P( B ) P( A | B ) P( B )
1/20/2016

Fundamentals

Example: Mad Cow Disease

75

Automation Lab
IIT Bombay

Consider a test in which a cow is tested to determine

infection with the mad cow disease. It is known that,
using the specified test, an infected cow has a 70%
chance of testing positive, and a healthy cow just 10%. It
is also known that 2% cows are infected. Find probability
that an arbitrary cow tests positive.

Event B : A randomly picked cow is infected

Event T : Test comes positive
P(T | B ) 0.7 and P (T | B ) 0.1
Note: As no test is 100% accurate, most tests have the
problem of false positives and false negatives. A false
positive means that according to the test the cow is
infected, but in actuality it is not. A false negative means an
infected cow is not detected by the test.
1/20/2016

Fundamentals

76

38

20-01-2016

Example: Mad Cow Disease

Automation Lab
IIT Bombay

Note : S B B
P( B ) 0.02 P( B ) 0.98
Problem is to find P(T )
Since T (T B ) (T B )
P (T ) P(T B ) P(T B )
P (T ) P(T | B ) P ( B ) P (T | B ) P ( B )
0.7 0.02 0.1 0.98 0.112
This is an application of the law of total probability.
1/20/2016

Fundamentals

Law of Total Probability

77

Automation Lab
IIT Bombay

Computing a probability through conditioning on

several disjoint events that make up the whole
sample space

Suppose C1, C2 , . . . , Cm are disjoint events

such that C1 C2 . . . . Cm S
Then, the probability of an arbitrary event A
can be expressed as :
P( A) P ( A C1 ) P( A C2 )..... P( A Cm )
P ( A | C1 ) P(C1 ) P( A | C2 ) P (C2 )
......P( A | Cm ) P(Cm )
1/20/2016

Fundamentals

78

39

20-01-2016

Law of Total Probability

Automation Lab
IIT Bombay

The law of total probability (illustration for m = 5).

1/20/2016

Fundamentals

Mad Cow Example (contd.)

79

Automation Lab
IIT Bombay

A more pertinent question about the mad cow

disease test is the following:
Suppose a cow tests positive; what is the
probability it really has the mad cow disease?

In mathematical terms, what is P ( B | T ) ?

(T B )
(T B )

P(T )
P(T B ) P(T B )
P(T | B ) P( B )

P(T | B ) P( B ) P(T | B ) P( B )
0.7 0.02

0.125
0.7 0.02 0.1 0.98

P( B | T )

(Dekking et al., 2005)

1/20/2016

Fundamentals

80

40

20-01-2016

Example: Mad Cow Disease

Automation Lab
IIT Bombay

Interpretation
If we know nothing about a cow, we would say
that there is a 2% chance it is infected.
However, if we know it tested positive, then we
can say there is a 12.5% chance the cow is
infected.
Finding P( B | T ) using P(T | B ) is an
application of Bayes' Rule derived by English
clergyman Thomas Bayes in the 18th century.

1/20/2016

Fundamentals

81

Automation Lab
IIT Bombay

Bayes Rule

Suppose C1, C2 , . . . , Cm are disjoint events

such that C1 C2 . . . . Cm S
Then, the conditiona l probability of Ci , given an
arbitrary event A, can be expressed as :
P (Ci | A)

1/20/2016

P ( A | Ci ) P (Ci )
P ( A | C1 ) P (C1 ) ...... P ( A | Cm ) P (Cm )
P ( A | Ci ) P (Ci )
P ( A)

Fundamentals

82

41

20-01-2016

Independence

Automation Lab
IIT Bombay

An event A is called independent of B if

P ( A|B ) P( A)
Result 1 : A independent of B A independent of B
P( A |B ) 1 P ( A|B) 1 P( A) P( A )
Result 2 : A independent of B P ( A B ) P( A) P ( B )
By application of the multiplication rule,
if A is independent of B
P ( A B ) P( A|B ) P( B ) P ( A) P( B )
1/20/2016

Fundamentals

Independence

83

Automation Lab
IIT Bombay

Result 3 : A independent of B B independent of A

P ( B | A)

P ( A B ) P( A) P ( B )

P( B )
P( A)
P ( A)

To show that A and B are independent it suffices

to prove just one of the following
P( A|B ) P( A)
P ( B | A) P( B )
P ( A B ) P( A) P( B )
( A may be replaced by A and B may be replaced by B .)
If one of these statements holds, all of them are true.
If two events are not independen t, they are called dependent.
1/20/2016

Fundamentals

84

42

20-01-2016

Independence Multiple Events

Automation Lab
IIT Bombay

An event A1 , A2 ,..., An are called independent if

P( A1 A2 .... An ) P( A1 ) P ( A2 ).... P( An )
and this statement also holds when any number
of the events A1,. . . , An are replaced by their
complements throughout the formula.
Note :
If A and B are independen t
and B and C are independen t, then
it does NOT imply that A and C are independent
1/20/2016

Fundamentals

Automation Lab
IIT Bombay

References
1.
2.

3.
4.
5.
6.

85

Papoulis, A. Probability, Random Variables and Stochastic

Processes, MacGraw-Hill International, 1991.
Dekking, F.M., Kraaikamp, C., Lopuhaa, H.P., Meester, L. E., A
Modern Introduction to Probability and Statistics:
Understanding Why and How, Springer, 2005.
Montgomery, D. C. and G. C. Runger, Applied Statistics and
Probability for Engineers, John Wiley and Sons, 2004.
Ross, S. M., Introduction to Probability and Statistics for
Engineers and Scientists, Elsevier, 4th Edition, 2009.
Maybeck, P. S., Stochastic models, Estimation, and Control:
Volume 1, Academic Press, 1979.
Jazwinski, A. H., Stochastic Processes and Filtering Theory,
Academic Press, 1970.

1/20/2016

Fundamentals

86

43