Simulation and Random

Number Generation
Summary

 Discrete Time vs Discrete Event Simulation

 Random number generation
 Generating a random sequence
 Generating random variates from a Uniform distribution
 Testing the quality of the random number generator
 Some probability results
 Evaluating integrals using Monte-Carlo simulation
 Generating random numbers from various distributions
 Generating discrete random variates from a given pmf
 Generating continuous random variates from a given
distribution
 Discrete Time vs. Discrete Event
Simulation
 Solve the finite difference equation
xk +1 = min { 100, − xk + 0.1xk −1 + log xk + uk }
 Given the initial conditions 100

x0, x-1 and the input function

50

uk, k=0,1,… we can

simulate the output! 0

This is a time driven x(k)


-50

simulation.
 For the systems we are -100

interested in, this method -150

has two main problems. -10 0 10 20
k
30 40 50
 Issues with Time-Driven Simulation

 System Input  System Dynamics

1 if parcel arrives at t
u1 (t ) =   x(t ) + 1 if u1 (t ) = 1, u 2 (t ) = 0,
0 otherwise 
x(t + ) =  x (t ) − 1 if u1 (t ) = 0, u 2 (t ) = 1
1 if truck arrives at t  x (t )
u2 (t ) =   otherwise
0 otherwise
u1(t)

t
u2(t)

Δt t

x(t)

t
 Inefficient: At most steps nothing happens
 Accuracy: Event order in each interval
Time Driven vs. Event Driven
Simulation Models
 Time Driven Dynamics  Event Driven Dynamics

 x(t ) + 1 if u1 (t ) = 1, u2 (t ) = 0,
 x + 1 if e′ = a

x(t + ) =  x(t ) − 1 if u1 (t ) = 0, u2 (t ) = 1 f ( x, e ′ ) = 
 x(t ) otherwise  x − 1 if e′ = d


 In this case, time is  State is updated only at

divided in intervals of the occurrence of a
length Δt, and the state is discrete event
updated at every step
Pseudo-Random Number Generation

 It is difficult to construct a truly random number generator.

 Random number generators consist of a sequence of
deterministically generated numbers that “look” random.
 A deterministic sequence can never be truly random.
 This property, though seemingly problematic, is actually desirable
because it is possible to reproduce the results of an experiment!
 Multiplicative Congruential Method
xk+1 = a xk modulo m
 The constants a and m should follow the following criteria
 For any initial seed x0, the resultant sequence should look like random
 For any initial seed, the period (number of variables that can be
generated before repetition) should be large.
 The value can be computed efficiently on a digital computer.
Pseudo-Random Number Generation

 To fit the above requirements, m should be chosen to

be a large prime number.
 Good choices for parameters are
 m= 231 -1 and a=75 for 32-bit word machines

 m= 235 -1 and a=55 for 36-bit word machines

 Mixed Congruential Method

xk+ 1 = (a xk + c) modulo m
 A more general form:
 n 
xk =  ∑ a j xk − j  mod m
 j =1 
 where n is a positive constant
 xk/m maps the random variates in the interval [0,1)
Quality of the Random Number
Generator (RNG)

 The random variate uk=xk/m should be uniformly

distributed in the interval [0,1).
 Dividethe interval [0,1) into k subintervals and count the
number of variates that fall within each interval.
 Run the chi-square test.
 The sequence of random variates should be
uncorrelated
 Define the auto-correlation coefficient with lag k > 0 Ck and
verify that E[Ck] approached 0 for all k=1,2,…
1 n−k
Ck = ∑ ( u i − 1
) ( u i+k − 2)
1
n − k i =1 2

where ui is the random variate in the interval [0,1).

 Some Probability Review/Results

 Probability mass functions  Cumulative mass functions

Pr{X=1}=p1 FX ( x ) = Pr { X ≤ x}
Pr{X=2}=p2 x

Pr{X=3}=p3 = ∑ Pr { X = i}
i =−∞

1 2 3 X
1
p1+ p2 + p3
∞ p1+ p2
∑ Pr { X = i} = 1
i =−∞
p1

1 2 3 X
 Some Probability Results/Review

 Probability density functions  Cumulative distribution

fX(x)
FX ( x ) = Pr { X ≤ x}
x
x1
∫
x
0 = f X ( y )dy
Pr{x=x1}= 0! −∞

Pr { x1 − ε ≤ X ≤ x1 + ε } = 1
x1 + ε
FX(x)

= ∫
x1 − ε
f X ( x ) dx
0 x
Independence of Random Variables

 Joint cumulative probability distribution.

FXY ( x, y ) = Pr { X ≤ x, Y ≤ y}
 Independent Random Variables.
Pr { X ∈ C , Y ∈ D} = Pr { X ∈ C} Pr { Y ∈ D}
 For discrete variables
Pr { X = x, Y = y} = Pr { X = x} Pr { Y = y}
 For continuous variables
f XY ( x, y ) = f X ( x) fY ( y ) FXY ( x, y ) = FX ( x) FY ( y )
Conditional Probability

 Conditional probability for two events.

Pr { AB}
Pr { A | B} =
Pr { B}
 Bayes’ Rule.
Pr { AB} Pr { B | A} Pr { A}
Pr { B | A} = ⇒ Pr { A | B} =
Pr { A} Pr { B}

 Total Probability

Pr { X = x} = ∑ Pr { X = x | Y = yk } Pr { Y = yk }
k
Expectation and Variance

 Expected value of a random variable

∞ ∞
E[ X ] = ∑ x Pr{ X = x }
i =−∞
i i E[ X ] = ∫ xf X ( x)dx
−∞
 Expected value of a function of a random variable
∞ ∞
E[ g ( X )] = ∑ g ( x ) Pr{X = x }
i =−∞
i i E[ g ( X )] = ∫ g ( x) f X ( x )dx
−∞

 Variance of a random variable

var [ X ] = E ( X − E [ X ] ) 
 2
Covariance of two random variables

 The covariance between two random variables is defined

as
cov [ X , Y ] = E ( X − E [ X ] ) ( Y − E [ Y ] ) 

cov [ X , Y ] = E [ XY − XE [ Y ] − YE [ X ] + E [ X ] E [ Y ] ]
= E [ XY ] − E [ X ] E [ Y ]
 The covariance between two independent random
variables is 0 (why?)

E [ XY ] = ∑∑ xi y j Pr { X = xi , Y = y j }
i j

= ∑ xi Pr { X = xi } ∑ y j Pr { Y = y j } = E [ X ] E [ Y ]
i j
Markov Inequality

 If the random variable X takes only non-negative values,

then for any value a > 0.
E[ X ]
Pr { X ≥ a} ≤
a
 Note that for any value 0<a < E[X], the above inequality
says nothing!
 Chebyshev’s Inequality

 If X is a random variable having mean μ and variance σ2,

then for any value k > 0.
1
Pr { X − µ ≥ kσ } ≤ 2
k
σ

kσ kσ
fX(x)
μ x
 This may be a very conservative bound!
 Note that for X~N(0,σ2), for k=2, from Chebyshev’s inequality Pr{.}<0.25, when in fact
the actual value is Pr{.}< 0.05
The Law of Large Numbers

 Weak: Let X1, X2, … be a sequence of independent and

identically distributed random variables having mean μ.
Then, for any ε >0
 X 1 + ... + X n 
Pr  − µ > ε  → 0, as n → ∞
 n 

 Strong: with probability 1,

X 1 + ... + X n
lim =µ
n →∞ n
The Central Limit Theorem

 Let X1, X2, …Xn be a sequence of independent random

variables with mean μi and variance σι2. Then, define the
random variable X,
X = X 1 + ... + X n
 Let n
µ = E [ X ] = E [ X 1 + ... + X n ] = ∑ µ i
n i =1

σ 2 = E ( X − µ ) 2  = ∑σ ι2
i =1

 Then, the distribution of X approaches the normal

distribution as n increases, and if Xi are continuous, then
1 − ( x − µ ) / 2σ
2
fX ( x) ≈ e
2π
Chi-Square Test

 Let k be the number of subintervals, thus pi=1/k, i=1,…,k.

 Let Ni be the number of samples in each subinterval. Note that E[Ni]=Npi
where N is the total number of samples
 Null Hypothesis H0: The probability that the observed random variates are
indeed uniformly distributed in (0,1).
 Let T be 2
1 k k
k  N
∑( N − Npi ) = ∑  Ni − 
2
T= i
Npi i =1 N i =1  k
 Define p-value= PH0{T>t} indicate the probability of observing a value t
assuming H0 is correct.
 For large N, T is approximated by a chi-square distribution with k-1
degrees of freedom, thus we can use this approximation to evaluate the
p-value
 The H0 is accepted if p-value is greater than 0.05 or 0.01
 Monte-Carlo Approach to Evaluating
Integrals
 Suppose that you want to estimate θ, however it is rather
difficult to analytically evaluate the integral.
1
θ = ∫ g ( x ) dx  Suppose also that you don’t want to
use numerical integration.
0
 Let U be a uniformly distributed random variable in the interval [0,1)
and ui are random variates of the same distribution. Consider the
following estimator.
n
1
θˆ = ∑ g ( ui ) → E [ g ( U ) ] as n → ∞
n i =1 Strong Law of
Large Numbers
1
 Also note that: θ = ∫ g ( u ) du = E [ g ( U ) ]
0
Monte-Carlo Approach to Evaluating
Integrals
 Use Monte-Carlo simulation to estimate the following
b
integral.
θ = ∫ g ( x ) dx
a
 Let y=(x-a)/(b-a), then the integral becomes
1 1
θ = ∫ ( b − a ) g ( ( b − a ) y + a ) dy = ∫ h ( y ) dy
0 0
∞
 What if θ = ∫ g ( x ) dx
0
 Use the substitution y= 1/(1+x),
1 1 1
 What if θ = ∫ ∫ ...∫ g ( x1 ,..., xn ) dx1 ...dxn
0 0 0
 Example: Estimate the value of π.

(-1,1) (1,1)
 Let X, Y, be independent
random variables uniformly
Area of Circle π distributed in the interval [-1,1]
=
Area of square 4  The probability that a point
(X,Y) falls in the circle is given
by π
Pr { X 2 + Y 2 ≤ 1} =
(-1,-1) (1,-1) 4
 SOLUTION
 Generate N pairs of uniformly distributed random variates (u1,u2) in
the interval [0,1).
 Transform them to become uniform over the interval [-1,1), using
(2u1-1,2u2-1).
 Form the ratio of the number of points that fall in the circle over N
 Discrete Random Variates

 Suppose we would like to generate a sequence of

discrete random variates according to a probability mass
function N
{ }
Pr X = x j = p j , j = 0,1,..., N , ∑ p j = 1
j =0

 x0 if u ≤ p0
x
1  1 if p0 ≤ u ≤ p0 + p1
u  M M
X = j −1 j
 x if
X  j ∑ pi ≤ u ≤ ∑ pi
x 
i =0 i =0

Inverse Transform Method

 M M
Discrete Random Variate Algorithm
(D-RNG-1)
 Algorithm D-RNG-1
 Generate u=U(0,1)
 If u<p0, set X=x0, return;
 If u<p0+p1, set X=x1, return;
…
 Set X=xn, return;
 Recall the requirement for efficient implementation, thus
the above search algorithm should be made as efficient
as possible!
 Example: Suppose that X∈{0,1,…,n} and p0= p1 =…= pn = 1/
(n+1), then
X = ( n + 1) u 
Discrete Random Variate Algorithm

 Assume p0= 0.1, p1 = 0.2, p2 = 0.4, p3 = 0.3. What is an

efficient RNG?
 D-RNG-1: Version 1
 Generate u=U(0,1)
 If u<0.1, set X=x0, return;
 If u<0.3, set X=x1, return;
 If u<0.7, set X=x2, return;
 Set X=x3, return;
 D-RNG-1: Version 2
 Generate u=U(0,1)
 If u<0.4, set X=x2, return;
 If u<0.7, set X=x3, return; More Efficient
 If u<0.9, set X=x1, return;
 Set X=x0, return;
 Geometric Random Variables

 Let p the probability of success and q=1-p the probability

of failure, then X is the time of the first success with pmf
Pr { X = i} = pq i −1

 Using the previous discrete random variate algorithm, X=j if

j −1 j

∑ Pr { X = i} ≤ U < ∑ Pr { X = i}
i =1 i =1
⇒ 1 − q j −1 ≤ U < 1 − q j
j −1

∑ Pr { X = i} = 1 − Pr { X > j − 1} = 1 − q
i =1
j −1
⇒ q j < 1 − U ≤ q j −1

⇒ X = min { j : q < 1 − U }j
As a result:
X = min { j : j log ( q ) < log ( 1 − U ) }
 log ( 1 − U ) 
 log ( 1 − U )  X =  +1
= min  j : j >
log ( ) 
q
  log ( q ) 

 Poisson and Binomial Distributions

 Poisson Distribution with rate λ.

i −λ
λe
Pr { X = i} = , i = 0,1,...
λ
i!
 Note: pi +1 = pi
i +1
 The binomial distribution (n,p) gives the number of successes
in n trials given that the probability of success is p.
n!
Pr { X = i} = p (1− p)
n −i
i
, i = 0,1,..., n
i !( n − i ) !
n −1 p
 Note: pi +1 = pi
i +1 1− p
 Accept/Reject Method (D-RNG-AR)

 Suppose that we would like to generate random variates

from a pmf {pj, j≥0} and we have an efficient way to
generate variates from a pmf {qj, j≥0}.
 Let a constant c such that
pj
≤ c for all j such that p j > 0
qj
 In this case, use the following algorithm
 D-RNG-AR:
1. Generate a random variate Y from pmf {qj, j≥0}.
2. Generate u=U(0,1)
3. If u< pY/(cqY), set X=Y and return;
4. Else repeat from Step 1.
 Accept/Reject Method (D-RNG-AR)

 Show that the D-RNG-AR algorithm generates a random

variate with the required pmf {pj, j≥0}.
∞

pi = Pr { X = i} = ∑ Pr { not stop for 1,...,k − 1} Pr { X = i and stop at iteration k}

k =1
Pr { X = i and stop after k} =
pi pi
Pr { X = i | Y = i and is accepted} Pr { Y accepted|Y = i} Pr { Y = i} = i
q =
cqi c
1 pi/cqi qi
k −1
 
Pr { Not stop up to k − 1} =  ∑ Pr { Y not accepted|Y = j} Pr { Y = j} 
 j k −1 
  pj    1
k −1

=  ∑  1 −  q j  =  1 − 
 j  c
  cq j  
Therefore k −1
∞
 1 pi
pi = ∑ 1 −  = pi
k =1  c c
 D-RNG-AR Example

 Determine an algorithm for generating random variates for

a random variable that take values 1,2,..,10 with
probabilities 0.11, 0.12, 0.09, 0.08, 0.12, 0.10, 0.09, 0.09,
0.10, 0.10 respectively.
 pi 
c = max   = 1.2
i
 qi 
 D-RNG-1:  D-RNG-AR:
 Generate u=U(0,1)  u1=U(0,1), u2=U(0,1)
 k=1;  Y=floor(10*u1 + 1);
 while(u >cdf(k))  while(u2 > p(Y)/c)
 k=k+1;  u1= U(0,1); u2=U(0,1);

 x(i)=k;  Y=floor(10*rand + 1);

 y(i)=Y;
 The Composition Approach

 Suppose that we have an efficient way to generate

variates from two pmfs {qj, j≥0} and {rj, j≥0}
 Suppose that we would like to generate random variates
for a random variable having pmf, a ∈(0,1).
Pr { X = j} = p j = aq j + (1 − a) r j , j ≥ 0
 Let X1 have pmf {qj, j≥0} and X2 have pmf {rj, j≥0} and
define  X 1 with probability a
X =
X2 with probability 1- a
 Algorithm D-RNG-C:
 Generate u=U(0,1)
 If u <= a generate X1
 Else generate X2
 Continuous Random Variates
Inverse Transform Method
 Suppose we would like to generate a sequence of
continuous random variates having density function FX(x)
 Algorithm C-RNG-1: Let U be a random variable uniformly
distributed in the interval (0,1). For any continuous
distribution function, the random variate X is given by
X = FX−1 ( U )
FX(x) 1

X x
 Example: Exponentially Distributed
Random Variable
 Suppose we would like to generate a sequence of random
variates having density function
f X ( x ) = λ e−λ x
 Solution
 Find the cumulative distribution
x
FX ( x ) = ∫ λ e− λ y dy = 1 − e− λ x
0

 Let a uniformly distributed random variable u

1
u = FX ( x ) = 1 − e ⇒ ln ( 1 − u ) = −λ x ⇒ x = − ln ( 1 − u )
−λ x

λ
 Equivalently, since 1-u is also uniformly distributed in (0,1)
1
x = − ln ( u )
λ
 Convolution Techniques and the Erlang
Distribution
 Suppose the random variable X is the sum of a number of
independent identically distributed random variables
n
X = ∑ Yi
i =1
 Algorithm C-RNG-Cv:
 Generate Y1,…,Yn from the given distribution
 X=Y1+Y2+…+Yn.

 An example of such random variable is the Erlang with order n

which is the sum of n iid exponential random variables with rate λ.

( λ x ) n e−λ x
f X ( x) =
n!
 Accept/Reject Method (C-RNG-AR)

 Suppose that we would like to generate random variates

from a pdf fX(x) and we have an efficient way to generate
variates from a pdf gX(x).
 Let a constant c such that
fX ( x)
≤ c for all x
gX ( x)
 In this case, use the following algorithm
 C-RNG-AR:
1. Generate a random variate Y from density gX(x).
2. Generate u=U(0,1)
3. If u< fX(Y)/(cgX(Y)), set X=Y and return;
4. Else repeat from Step 1.
 Accept/Reject Method (C-RNG-AR)

 The C-RNG-AR is similar to the D-RNG-AR algorithm

except the comparison step where rather than comparing
the two probabilities we compare the values of the density
functions.
 Theorem
 The random variates generated by the Accept/Reject method
have density fX(x).
 The number of iterations of the algorithm that are needed is a
geometric random variable with mean c

 Note: The constant c is important since is implies the number of

iterations needed before a number is accepted, therefore it is
required that it is selected so that it has its minimum value.
 C-RNG-AR Example

 Use the C-RNG-AR method to generate random variates

X that are normally distributed with mean 0 and variance
1, N(0,1).
 First consider the pdf of the absolute value of |X|.
2 − 12 x 2
f| X | ( x ) = e
2π
 We know how to generate exponentially distributed random
variates Y with rate λ=1.
gY ( x ) = e − x , x ≥ 0
 Determine c such that it is equal to the maximum of the ratio
f| X | ( x ) 2 x − 12 x2 2e
= e ⇒c=
gY ( x ) 2π π
 C-RNG-AR Example

 C-RNG-AR for N(0,1):

 u1=U(0,1), u2=U(0,1);
 Y= -log(u1);
 while(u2 > exp(-0.5(Y-1)*(Y-1)))
 u1= U(0,1); u2=U(0,1);

 Y= -log(u1);

 u3= U(0,1);
 If u3 < 0.5 X=Y;
 Else X= -Y;

 Suppose we would like Z~N(μ, σ2), then

Z := σ X + µ
 Generating a Homogeneous Poisson
Processes
 A homogenous Poisson process is a sequence of points
(events) where the inter-even times are exponentially
distributed with rate λ (The Poisson process will be studied in detail during
later classes)

 Let ti denote the ith point of a Poisson process, then the

algorithm for generating the first N points of the sequence {ti,
i=1,2,…,N} is given by
 Algorithm Poisson-λ:
 k=0, t(k)=0;
 While k<N
 k= k+1;

 Generate u=U(0,1)

 t(k)= t(k-1) – log(u)/lambda;

 Return t.
 Generating a Non-Homogeneous
Poisson Processes
 Suppose that the process is non-homogeneous i.e., the
rate varies with time, i.e., λ(t) ≤λ, for all t<T.
 Let ti denote the ith point of a Poisson process, and τ the
actual time, then the algorithm for generating the first N points
of the sequence {ti, i=1,2,…,N} is given by
 Algorithm Thinning Poisson-λ:
 k=0, t(k)=0, tau= 0;
 While k<N
 Generate u1=U(0,1);

 tau= tau – log(u1)/lambda;

 Generate u2= U(0,1);

 If(lambda(tau)\lambda < u2)

 k= k+1, t(k)= tau;
 Return t.
 Generating a Non-Homogeneous
Poisson Processes
 Again, suppose that the process is non-homogeneous
i.e., the rate varies with time, i.e., λ(t) ≤λ, for all t<T but
now we would like to generate all points ti directly, without
thinning.
 Assuming that we are at point ti, then the question that we
need to answer is what is the cdf of Si where Si is the time until
the next event

{ s
}
F| Si | ( s ) = Pr { Si < s | t = ti } = 1 − exp − ∫ λ ( ti + y ) dy
0

 Thus, to simulate this process, we start from t0 and generate S1

from FS1 to go to t1=t0+S1. Then, from t1, we generate S2 from FS2
to go to t2=t1+S2 and so on.
 Example of Non-Homogeneous
Poisson Processes
 Suppose that λ(t)= 1/(t+α), t ≥0, for some positive constant
a. Generate variates from this non-homogeneous
Poisson process.
 First, let us determine the rate of the cdf
s 1  s + ti + a 
∫0 λ ( ti + y ) dy = ∫0 ( a + ti + y ) dy = log  ti + a 
s

  s + ti + a   ti + a s
FSi = 1 − exp − log    = 1− =
  ti + a   s + ti + a s + t i + a
 Inverting this yields
−1 ( ti + a ) u
FS =
i
1− u
 Example of Non-Homogeneous
Poisson Processes
−1 ( ti + a ) u
 Inverse Transform FS =
i
1− u

 Thus we start from t0=0

( t0 + a ) u1 au1
t1 = t0 + =
1 − u1 1 − u1
( t1 + a ) u2 t1 + au2
t2 = t1 + =
1 − u2 1 − u2
…
( ti −1 + a ) ui ti −1 + aui
ti = ti −1 + =
1 − ui 1 − ui
 The Composition Approach

 Suppose that we have an efficient way to generate variates

from cdfs G1(x),…, Gn(x).
 Suppose that we would like to generate random variates for a
random variable having cdf
n n
FX ( x ) = ∑ ri G i ( x), ∑ r = 1,
i ri > 0, i = 1,..., n
i =1 i =1

 Algorithm C-RNG-C:
 Generate u=U(0,1)
 If u<p1, get X from G1(x), return;
 If u<p1+p2, get X from G2(x), return;
…
 Polar Method for Generating Normal
Random Variates
 Let X and Y be independent normally distributed random
variables with zero mean and variance 1. Then the joint
density function is given by
1 − 12 x2 1 − 12 y2 1 − 12 ( x2 + y 2 )
f XY ( x, y ) = e e = e
2π 2π 2π
 Then make a variable change
Y y
r=x +y 2 2
θ = arctan
R x
 The new joint density function is
θ 1 1 − 12 r
X f RΘ ( r , θ ) = e
2π 2
Uniform in the Exponential
interval [0,2π] with rate 1/2
 Polar Method for Generating Normal
Random Variates
 Algorithm C-RNG-N1:
Generates 2
 Generate u1=U(0,1), u2=U(0,1);
independent
 R= -2*log(u1); W= 2*pi*u2; RVs
 X= sqrt(R) cos(W); Y= sqrt(R) sin(W);
 But, sine and cosine evaluations are inefficient!
 Algorithm C-RNG-N2:
1. Generate u1=U(0,1), u2=U(0,1);
2. Set V1= 2*u1-1, V2= 2*u2-1; (-1,1) (1,1)
3. S=V1*V1+V2*V2; (V1,V2)
4. If S > 1, Go to 1
5. R= sqrt(-2*log(S)/S);
6. X= R*V1;
7. Y= R*V2;
(-1,-1) (1,-1)
Simulation of Discrete Event Systems

INITIALIZE

TIME
STATE
EVENT CALENDAR
Update State
e1 t1 Update Time
e2 t2 t’=t1
x’=f(x,e1) …

Delete
Infeasible
CLOCK Add New
STRUCTURE Events
Feasible
RNG Events
Verification of a Simulation Program

 Standard debugging techniques

 Debug “modules” or subroutines
 Create simple special cases, where you know what to
expect as an output from each of the modules
 Often choosing carefully the system parameters, the
simulation model can be evaluated analytically.
 Create a trace which keeps track of the state
variables, the event list and other variables.