3.

3 PROBABILITY DISTRIBUTION FUNCTIONS

Besides using the moments of data, one can use algebraic
equation as a fit for the PDF or CDF. In this section, a few
commonly used continuous and discrete functions are
introduced.
The basic conditions that these functions must satisfy are:
FX(-) = 0, FX(+) = 1,
FX(x)  0 and is continuous and non-decreasing with X
The latter implies that fX(x)  0
The functions are based on either
• their fit with data, or
• consideration of its underlying physical process and
assumptions

3.3.1 Normal (or Gaussian) Distribution

Best known and most widely used.
Its PDF for a continuous r.v. X is given by

1  1  x   2 
f X ( x)  exp       N(  ,  )    x  
 2 2
    

Parameters: ,  (mean and standard deviation)

Two possible conventions:
(1) N(,  ) (we use this, used in Matlab)
(2) N(, 2 )

Also referred to as
“bell curve” due to
its resemblance to
the shape of a bell
Standard Normal Distribution fs(s)  = 0,  = 1

1  x2 
N (0,1)   ( x )  exp   
2  2
(s) represents the CDF of standard normal variate S

 ( s )  FS ( s )    ( x)dx
s
fS(s)
 0.5

Shaded area p under the PDF as (sp) 0.4

N(0,1)
probability = p
Conversely, the value of S at a 0.3

cumulative probability p is denoted as 0.2

sp = -1(p) 0.1
area = 1-p

0.0
-4 -3 -2 -1 0
sp 1 2 3 4
s
symmetric about origin

p
Sp fS(s)
0.5

0.4
N(0,1)
probability = p 0.3

0.2

0.1
area = 1-p

0.0
-4 -3 -2 -1 0
sp 1 2 3 4
s

Round off z to 2 decimals

For example,
Sp = 1.281 ~ 1.28

Then, use z = 1.2 (row)

and 0.08 (column)

Hence, p = 0.8997
By symmetry , fS(s)
0.5

( s p )  1  (s p ) 0.4

N(0,1)
probability = 1-p
( s p )  1  p 0.3

0.2

When p  0.5 area = 1-p

0.1

s p    1(1  p ) 0.0
-4 -3 -2 -1
-sp
0
sp 1 2 3 4
s

fS(s) fS(s)
0.5 0.5

0.4 area = FX(b) - FX(a) 0.4 0.8413-(1-0.8413)

area = 0.683
0.3 0.3

0.2 0.2

0.1 0.1

0.0 0.0
-4 -3 -2 -1 0 1 2 3 4 s -4 -3 -2 -1 0 1 2 3 4 s
a b

fS(s) fS(s)
0.5
0.5

0.9772-(1-0.9772)
0.4 0.4

area = 0.954 area = 0.997

0.3 0.3

0.2 0.2

0.1 0.1

0.0 0.0
-4 -3 -2 -1 0 1 2 3 4 s -4 -3 -2 -1 0 1 2 3 4 s
 Calculate the probability of X that follows any
Gaussian distribution N(, )
a
P ( X  a )  F X (a )  
f X ( x ) dx

a 1  1  x   2 
 exp      dx

 2  2    

X  X = S + ,
Let S 
 X < a  S +  < a,
a S < (a – )/
 X aS

a
a 1  1  a 
P ( X  a)  P (S  ) 
exp   s 2  ds    
 
2  2    

Example 3.5 - use of normal distribution table

Given X ~ N(60, 15), find (a) P(X  30); (b) P(40 < X  70);
(c) xp where FX(xp) = 0.1
 30  60 
P( X  30)  1  P( X  30)  1     1   (2)
 15 
 1  [1   (2)]   (2)  0.9772

 70  60   40  60 
P(40  X  70)         (0.67)   (1.33)
 15   15 
  (0.67)  1   (1.33)  0.7486  1  0.9082  0.6568

 x p  60  x p  60
FX ( x p )     0.1    1 (0.1)  1.28
 15  15
x p  60  15(1.28)  40.8
  1 (0.9)
 A note on inequalities
For continuous random variables,
P(X  30) is the same as P(X > 30)
Because the probability that P = 30.000… (infinite no. of zeros)
is zero, i.e. P(X = 30) = 0

However for discrete random variables, e.g. integers,

P(X  30) implies X = 30, 31, 32, etc (includes 30)
But
P(X > 30) implies X = 31, 32, etc (does not include 30)

Example 3.6 - settlement of shell structure

Settlements at three supports A, B, C are independent
normal r.v. XAXB and XC with means 2, 2.5 and 3 cm and
COVs 0.2, 0.2 and 0.25 respectively
(reminder: COV,  = /)

Find (a) P(Xmax> 4 cm); (b) P(> 1.5 cm | XA= 2.5  XB= 3.5)
where  = max. differential settlement.
P(Xmax> 4) = 1 – P(Xmax 4)
A
P(Xmax 4) = P[(XA 4)(XB 4)(XC 4)]

= P(XA 4)  P(XB  4)  P(XC  4)

 4 2  4  2 .5   4 3
          0.9075 B C
 0. 4   0. 5   0.75 

P(Xmax> 4) = 1 – 0.9075
 Using De Morgan’s Rule

Let Event A = XA > 4

B = XB > 4

C = XC > 4

Let Xmax = max (XA, XB, XC)

X max  4 = (X A  4)  ( X B  4)  ( X C  4)
= ABC = ABC
= ABC

P(> 1.5 cm | XA= 2.5  XB= 3.5) where  = max. differential

settlement.
Consider XA and XC: for XA= 2.5 (XC< 1)(XC>4)

Consider XB and XC: for XB= 3.5 (XC< 2)(XC > 5)

Combining both cases: [(XC<1)(XC>4)][(XC<2)(XC>5)]

gives (XC< 2)(XC>4)

XA XB
XA and XC
XB and XC

1 2 3 4 5
2.5 3.5


P[(XC < 2)(XC > 4)] = P(XC < 2) + P(XC > 4) … m.e.
 23 43
    1    0.75   0.1836
 0. 75   
3.3.2 Lognormal Distribution (positive and skew)
1  1  ln x    2 
f X ( x)  exp      LN ( ,  ) 0 x
x 2  2    

Parameters:  = E(ln X), 2 = var(ln X)

E[ X ]   X  exp(  0.5 2 )    ln  X  0.5 2
  X2 
  ln  1  2   ln 1   X2    X2
2
(for small  X ,  X  0.3)
 X 
Median, xm  e  fX(x)
median = exp () = 1
4

=0.1
3

If ln X is normal, X has a
2

lognormal probability
=0.3
distribution 1

=0.5
0 x
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8

CDF of LN(, ):

ln X   ln a  
Let S  then X  a implies that S 
 
ln a    ln a   
FX (a )  P ( X  a )  P( S  )   
   

If X is lognormally distributed with parameters , 

ln X  
then S  is normally distributed with mean 0, s.d. 1.

Or, Y = ln X is normally distributed with mean  and standard
deviation 
Example 3.7 - Use of normal distribution table for LN
For X,  = 60 and  = 15, X follows LN, find (a) P(X  30); (b)
P(40 < X  70); (c) xp where FX(xp) = 0.1

  X2   15 2 
  ln  1  2   ln  1  2     0 .2462
2

 X   60 
  ln  X  0 .5 2  ln 60  0 .5  0 .06062  4 .064

 ln 30  4 .064 
P ( X  30 )  1      1   (  2 .692 )  0 .996
 0 .2462 
 ln 70  4 .064   ln 40  4 .064 
P ( 40  X  70 )       
 0 .2462   0 .2462 
  ( 0 .7494 )   (  1 .524 )  0 .71
 ln x p  4 .064  ln x p  4 .064
FX x p   
( )  
  0 .1     1 ( 0 .1 )
 0 .246  0 .2462
x p  exp[ 4 .064  0 .2462 (  1 .28 )]  42 .5
  (0.9)  1.28
1