Random-Number Generation

Contents
• Properties of Random Numbers
• Pseudo-Random Numbers
• Generating Random Numbers
• Linear Congruential Method
• Combined Linear Congruential Method
• Tests for Random Numbers
• Real Random Numbers
Overview
• Discuss characteristics and
the generation of random
numbers.
• Subsequently, introduce
tests for randomness:
• Frequency test
• Autocorrelation test
Overview
• Historically
• Throw dices
• Deal out cards
• Draw numbered balls
• Use digits of π
• Mechanical devices (spinning disc, etc.)
• Electric circuits
• Electronic Random Number Indicator (ERNIE)
• Counting gamma rays
• In combination with a computer
• Hook up an electronic device to the computer
• Read-in a table of random numbers
Pseudo-Random Numbers
Pseudo-Random Numbers
• Approach: Arithmetically generation (calculation) of
random numbers
• “Pseudo”, because generating numbers using a known
method removes the potential for true randomness.

Any one who considers arithmetical methods of

producing random digits is, of course, in a state of
sin. For, as has been pointed out several times,
there is no such thing as a random number — there
are only methods to produce random numbers, and
a strict arithmetic procedure of course is not such a
method.
John von Neumann, 1951
Pseudo-Random Numbers

… probably … can not be justified, but should

merely be judged by their results. Some statistical
study of the digits generated by a given recipe
should be made, but exhaustive tests are
impractical. If the digits work well on one problem,
they seem usually to be successful with others of the
same type.
John von Neumann, 1951

• Goal: To produce a sequence of numbers in [0,1] that

simulates, or imitates, the ideal properties of random
numbers (RN).
Pseudo-Random Numbers
• Important properties of good random number routines:
• Fast
• Portable to different computers
• Have sufficiently long cycle
• Replicable
• Verification and debugging
• Use identical stream of random numbers for different systems
• Closely approximate the ideal statistical properties of
• uniformity and
• independence
Pseudo-Random Numbers: Properties
• Two important statistical properties:
• Uniformity
• Independence
• Random number Ri must be independently drawn from a
uniform distribution with PDF:

f(x)
⎧1, 0 ≤ x ≤ 1
f ( x) = ⎨
⎩0, otherwise

1 x
2
1 x 1 0 1
E ( R) = xdx =
∫ = PDF for random numbers
0 2 2
0
Pseudo-Random Numbers
• Problems when generating pseudo-random numbers
• The generated numbers might not be uniformly distributed
• The generated numbers might be discrete-valued instead of
continuous-valued
• The mean of the generated numbers might be too high or too
low
• The variance of the generated numbers might be too high or
too low

• There might be dependence:

• Autocorrelation between numbers
• Numbers successively higher or lower than adjacent numbers
• Several numbers above the mean followed by several
numbers below the mean
Generating Random Numbers
Generating Random Numbers
• Midsquare method
• Linear Congruential Method (LCM)
• Combined Linear Congruential Generators (CLCG)
• Random-Number Streams
Generating Random Numbers
Midsquare method
Midsquare method
• First arithmetic generator: Midsquare method
• von Neumann and Metropolis in 1940s
• The Midsquare method:
• Start with a four-digit positive integer Z0
• Compute:Z 02 = Z 0 × Z 0 to obtain an integer with up to eight
digits
• Take the middle four digits for the next four-digit number

i Zi Ui Zi×Zi
0 7182 - 51581124
1 5811 0.5811 33767721
2 7677 0.7677 58936329
3 9363 0.9363 87665769
…
Midsquare method
• Problem: Generated i Zi Ui Zi×Zi
numbers tend to 0
0 7182 - 51581124
1 5811 0,5811 33767721
2 7677 0,7677 58936329
3 9363 0,9363 87665769
4 6657 0,6657 44315649
5 3156 0,3156 09960336
6 9603 0,9603 92217609
7 2176 0,2176 04734976
8 7349 0,7349 54007801
9 78 0,0078 00006084
10 60 0,006 00003600
11 36 0,0036 00001296
12 12 0,0012 00000144
13 1 0,0001 00000001
14 0 0 00000000
15 0 0 00000000
… random numbers should not be
generated with a method chosen at
random. Some theory should be
used.

Donald E. Knuth, The Art of Computer Programming, Vol. 2

Generating Random Numbers
Linear Congruential Method
Linear Congruential Method
• To produce a sequence of integers X1, X2, … between 0 and
m-1 by following a recursive relationship:

X i +1 = (aX i + c) mod m, i = 0,1,2,...

The multiplier The increment The modulus

• Assumption: m > 0 and a < m, c < m, X0 < m

• The selection of the values for a, c, m, and X0 drastically
affects the statistical properties and the cycle length
• The random integers Xi are being generated in [0, m-1]
Linear Congruential Method
• Convert the integers Xi to random numbers

Xi
Ri = , i = 1,2,...
m

• Note:
• Xi ∈ {0, 1, ..., m-1}
• Ri ∈ [0, (m-1)/m]
Linear Congruential Method: Example
• Use X0 = 27, a = 17, c = 43, and m = 100.
• The Xi and Ri values are:

X1 = (17×27+43) mod 100 = 502 mod 100 = 2 Æ R1 = 0.02

X2 = (17×2 +43) mod 100 = 77 Æ R2 = 0.77
X3 = (17×77+43) mod 100 = 52 Æ R3 = 0.52
X4 = (17×52+43) mod 100 = 27 Æ R3 = 0.27
…
Linear Congruential Method: Example
• Use a = 13, c = 0, and m = 64 Xi Xi Xi Xi
• The period of the i X0=1 X0=2 X0=3 X0=4
0 1 2 3 4
generator is very low 1 13 26 39 52
• Seed X0 influences the 2 41 18 59 36
3 21 42 63 20
sequence
4 17 34 51 4
5 29 58 23
6 57 50 43
7 37 10 47
8 33 2 35
9 45 7
10 9 27
11 53 31
12 49 19
13 61 55
14 25 11
15 5 15
16 1 3
Linear Congruential Method:
Characteristics of a good Generator

• Maximum Density
• The values assumed by Ri, i=1,2,… leave no large gaps on [0,1]
• Problem: Instead of continuous, each Ri is discrete
• Solution: a very large integer for modulus m
• Approximation appears to be of little consequence

• Maximum Period
• To achieve maximum density and avoid cycling
• Achieved by proper choice of a, c, m, and X0

• Most digital computers use a binary representation of numbers

• Speed and efficiency are aided by a modulus, m, to be (or close to) a
power of 2.
Linear Congruential Method:
Characteristics of a good Generator

• The LCG has full period if and only if the following three
conditions hold (Hull and Dobell, 1962):
1. The only positive integer that (exactly) divides both m and c
is 1
2. If q is a prime number that divides m, then q divides a-1
3. If 4 divides m, then 4 divides a-1
Linear Congruential Method:
Proper choice of parameters

• For m a power 2, m=2b, and c≠0

• Longest possible period P=m=2b is achieved
if c is relative prime to m and a=1+4k, where k is an integer

• For m a power 2, m=2b, and c=0

• Longest possible period P=m/4=2b-2 is achieved
if the seed X0 is odd and a=3+8k or a=5+8k, for k=0,1,...

• For m a prime and c=0

• Longest possible period P=m-1 is achieved
if the multiplier a has property that smallest integer k such
that ak-1 is divisible by m is k = m-1
Characteristics of a Good Generator
Characteristics of a Good Generator
Random-Numbers in Java
• Defined in java.util.Random
private final static long multiplier = 0x5DEECE66DL; // 25214903917
private final static long addend = 0xBL; // 11
private final static long mask = (1L << 48) - 1; // 248-1 = 281474976710655

protected int next(int bits) {

long oldseed, nextseed;
...
oldseed = seed.get();
nextseed = (oldseed * multiplier + addend) & mask;
...
return (int)(nextseed >>> (48 - bits)); // >>> Unsigned right shift
}
General Congruential Generators
• Linear Congruential Generators are a special case of
generators defined by:

X i +1 = g ( X i , X i −1 ,…) mod m
• where g() is a function of previous Xi’s
• Xi ∈ [0, m-1], Ri = Xi /m

• Quadratic congruential generator

• Defined by: g ( X i , X i −1 ) = aX i2 + bX i −1 + c
• Multiple recursive generators
• Defined by: g ( X i , X i −1 ,…) = a1 X i + a2 X i −1 +  + ak X i −k
• Fibonacci generator
• Defined by: g ( X i , X i −1 ) = X i + X i −1
 Combined Linear Congruential Generators
• Reason: Longer period generator is needed because of the
increasing complexity of simulated systems.
• Approach: Combine two or more multiplicative congruential
generators.

• Let Xi,1, Xi,2, …, Xi,k be the i-th output from k different

multiplicative congruential generators.
• The j-th generator X•,j:

X i +1, j = (a j X i + c j ) mod m j
• has prime modulus mj, multiplier aj, and period mj -1
• produces integers Xi,j approx ~ Uniform on [0, mj – 1]
• Wi,j = Xi,j - 1 is approx ~ Uniform on integers on [0, mj - 2]

Prof. Dr. Mesut Güneş ▪ Ch. 6 Random-Number Generation 6.29

 Combined Linear Congruential Generators
• Suggested form:
⎧ X i
⎛ k j −1
⎞ ⎪⎪ m , Xi > 0
X i = ⎜ ∑ (−1) X i , j ⎟⎟ mod m1 − 1
⎜ Hence, Ri = ⎨ 1
m −1
⎝ j =1 ⎠ ⎪ 1 , Xi = 0
⎪⎩ m1

(m1 − 1)(m2 − 1)...(mk − 1)

• The maximum possible period is: P=
2k −1

Prof. Dr. Mesut Güneş ▪ Ch. 6 Random-Number Generation 6.30

 Combined Linear Congruential Generators
• Example: For 32-bit computers, combining k = 2 generators with
m1 = 2147483563, a1 = 40014, m2 = 2147483399 and a2 = 40692.
The algorithm becomes:
Step 1: Select seeds
X0,1 in the range [1, 2147483562] for the 1st generator
X0,2 in the range [1, 2147483398] for the 2nd generator
Step 2: For each individual generator,
Xi+1,1 = 40014 × Xi,1 mod 2147483563
Xi+1,2 = 40692 × Xi,2 mod 2147483399
Step 3: Xi+1 = (Xi+1,1 - Xi+1,2 ) mod 2147483562
Step 4: Return
⎧ X i +1
⎪ 2147483563 , X i +1 > 0
⎪
Ri +1 = ⎨
⎪ 2147483562
⎪ 2147483563 , X i +1 = 0
⎩
Step 5: Set i = i+1, go back to step 2.
• Combined generator has period: (m1 – 1)(m2 – 1)/2 ~ 2 x 1018

Prof. Dr. Mesut Güneş ▪ Ch. 6 Random-Number Generation 6.31

 Random-Numbers in Excel 2003
• In Excel 2003 and 2007 new Random Number Generator

X, Y, Z ∈ {1,...,30000}
X = X ⋅171 mod 30269
Y = Y ⋅ 172 mod 30307
Z = Z ⋅ 170 mod 30323
⎛ X Y Z ⎞
R = ⎜ + + ⎟ mod 1.0
⎝ 30269 30307 30323 ⎠

• It is stated that this method produces more than 1013

numbers
• For more info: http://support.microsoft.com/kb/828795

Prof. Dr. Mesut Güneş ▪ Ch. 6 Random-Number Generation 6.32

 Random-Numbers Streams
• The seed for a linear congruential random-number generator:
• Is the integer value X0 that initializes the random-number sequence
• Any value in the sequence (X0, X1, …, XP) can be used to “seed” the generator

• A random-number stream:
• Refers to a starting seed taken from the sequence (X0, X1, …, XP).
• If the streams are b values apart, then stream i is defined by starting seed:

Si = X b (i −1) i = 1,2, …, ⎣Pb ⎦

• Older generators: b = 105
• Newer generators: b = 1037

• A single random-number generator with k streams can act like k

distinct virtual random-number generators

• To compare two or more alternative systems.

• Advantageous to dedicate portions of the pseudo-random number sequence
to the same purpose in each of the simulated systems.

Prof. Dr. Mesut Güneş ▪ Ch. 6 Random-Number Generation 6.33

 Random Numbers in OMNeT++
• OMNeT++ releases prior to 3.0 used a linear congruential
generator (LCG) with a cycle length of 231-2.
• By default, OMNeT++ uses the Mersenne Twister RNG
(MT) by M. Matsumoto and T. Nishimura.
• MT has a period of 219937-1, and 623-dimensional
equidistribution property is assured.
• This RNG can be selected from omnetpp.ini
• OMNeT++ allows plugging in your own RNGs as well. This
mechanism, based on the cRNG interface.

Prof. Dr. Mesut Güneş ▪ Ch. 6 Random-Number Generation 6.34

 Tests for Random Numbers

Prof. Dr. Mesut Güneş ▪ Ch. 6 Random-Number Generation 6.35

 Tests for Random Numbers
• Two categories:
• Testing for uniformity:
H0: Ri ~ U[0,1]
H1: Ri ≁ U[0,1]
• Failure to reject the null hypothesis, H0, means that evidence of non-
uniformity has not been detected.
• Testing for independence:
H0: Ri ~ independent
H1: Ri ≁ independent
• Failure to reject the null hypothesis, H0, means that evidence of
dependence has not been detected.

• Level of significance α, the probability of rejecting H0 when it is

true:
α = P(reject H0 | H0 is true)

Prof. Dr. Mesut Güneş ▪ Ch. 6 Random-Number Generation 6.36

 Tests for Random Numbers
• When to use these tests:
• If a well-known simulation language or random-number generator is
used, it is probably unnecessary to test
• If the generator is not explicitly known or documented, e.g.,
spreadsheet programs, symbolic/numerical calculators, tests should
be applied to many sample numbers.

• Types of tests:
• Theoretical tests: evaluate the choices of m, a, and c without actually
generating any numbers
• Empirical tests: applied to actual sequences of numbers produced.
• Our emphasis.

Prof. Dr. Mesut Güneş ▪ Ch. 6 Random-Number Generation 6.37

 Tests for Random Numbers
Frequency tests: Kolmogorov-Smirnov Test

Prof. Dr. Mesut Güneş ▪ Ch. 6 Random-Number Generation 6.38

 Kolmogorov-Smirnov Test
• Compares the continuous CDF, F(x), of the uniform distribution
with the empirical CDF, SN(x), of the N sample observations.
F(x)
• We know: F ( x) = x, 0 ≤ x ≤ 1

• If the sample from the RNG

is R1, R2, …, RN, then the empirical x
CDF, SN(x) is: 0 1

Number of Ri where Ri ≤ x
S N ( x) =
N

• Based on the statistic: D = max | F(x) - SN(x)|

• Sampling distribution of D is known

Prof. Dr. Mesut Güneş ▪ Ch. 6 Random-Number Generation 6.39

 Kolmogorov-Smirnov Test
• The test consists of the Kolmogorov-Smirnov Critical Values
following steps
• Step 1: Rank the data from
smallest to largest
R(1) ≤ R(2) ≤ ... ≤ R(N)

• Step 2: Compute
⎧ i ⎫
D + = max ⎨ − R(i ) ⎬
1≤i ≤ N N
⎩ ⎭
⎧ i − 1⎫
D − = max ⎨ R(i ) − ⎬
1≤i ≤ N
⎩ N ⎭
• Step 3: Compute D = max(D+, D-)
• Step 4: Get Dα for the
significance level α

• Step 5: If D ≤ Dα accept,
otherwise reject H0

Prof. Dr. Mesut Güneş ▪ Ch. 6 Random-Number Generation 6.40

 Kolmogorov-Smirnov Test
• Example: Suppose N=5 numbers: 0.44, 0.81, 0.14, 0.05, 0.93.
i 1 2 3 4 5 Arrange R(i) from
smallest to largest
Step 1: R(i) 0.05 0.14 0.44 0.81 0.93

i/N 0.20 0.40 0.60 0.80 1.00 D+ = max{i/N – R(i)}

i/N – R(i) 0.15 0.26 0.16 - 0.07
Step 2: D - = max{R(i) - (i-1)/N}
R(i) – (i-1)/N 0.05 - 0.04 0.21 0.13

Step 3: D = max(D+, D-) = 0.26

Step 4: For α = 0.05,
Dα = 0.565 > D = 0.26

Hence, H0 is not rejected.

Prof. Dr. Mesut Güneş ▪ Ch. 6 Random-Number Generation 6.41

 Tests for Random Numbers
Frequency tests: Chi-square Test

Prof. Dr. Mesut Güneş ▪ Ch. 6 Random-Number Generation 6.42

 Chi-square Test
• Chi-square test uses the sample statistic:
n is the # of classes Oi is the observed # in the i-th class

Ei is the expected # in the i-th class

n 2
(O − E )
χ 02 = ∑ i i

i =1 Ei
• Approximately the chi-square distribution with n-1 degrees of
freedom
• For the uniform distribution, Ei, the expected number in each class
is:
N
Ei = , where N is the total number of observations
n

• Valid only for large samples, e.g., N ≥ 50

Prof. Dr. Mesut Güneş ▪ Ch. 6 Random-Number Generation 6.43

 Chi-square Test: Example
• Example with 100 numbers from [0,1], α=0.05
• 10 intervals Interval Upper Limit Oi Ei Oi-Ei (Oi-Ei)^2 (Oi-Ei)^2/Ei
• χ 20.05,9 = 16.9 1 0.1 10 10 0 0 0

• Accept, since 2 0.2 9 10 -1 1 0.1

• X20=11.2 < χ 20.05,9 3 0.3 5 10 -5 25 2.5

4 0.4 6 10 -4 16 1.6
5 0.5 16 10 6 36 3.6
6 0.6 13 10 3 9 0.9
7 0.7 10 10 0 0 0
8 0.8 7 10 -3 9 0.9
9 0.9 10 10 0 0 0
10 1.0 14 10 4 16 1.6
Sum 100 100 0 0 11.2

2 (Oi − Ei ) 2
n
χ =∑
0 X20=11.2
i =1 Ei

Prof. Dr. Mesut Güneş ▪ Ch. 6 Random-Number Generation 6.44

 Tests for Random Numbers
Tests for autocorrelation

Prof. Dr. Mesut Güneş ▪ Ch. 6 Random-Number Generation 6.45

 Tests for Autocorrelation
• Autocorrelation is concerned with dependence between
numbers in a sequence
• Example:

0.12 0.01 0.23 0.28 0.89 0.31 0.64 0.28 0.83 0.93
0.99 0.15 0.33 0.35 0.91 0.41 0.60 0.27 0.75 0.88
0.68 0.49 0.05 0.43 0.95 0.58 0.19 0.36 0.69 0.87

• Numbers at 5-th, 10-th, 15-th, ... are very similar

• Numbers can be
• Low
• High
• Alternating

Prof. Dr. Mesut Güneş ▪ Ch. 6 Random-Number Generation 6.46

 Tests for Autocorrelation
• Testing the autocorrelation between every m numbers (m is
a.k.a. the lag), starting with the i-th number
• The autocorrelation ρi,m between numbers: Ri, Ri+m, Ri+2m, Ri+(M+1)m
• M is the largest integer such that i + ( M + 1)m ≤ N

• Hypothesis:
H 0 : ρi ,m = 0, if numbers are independent
H1 : ρi ,m ≠ 0, if numbers are dependent

• If the values are uncorrelated:

• For large values of M, the distribution of the estimator of ρi,m,
denoted ρ̂ i,m is approximately normal.

Prof. Dr. Mesut Güneş ▪ Ch. 6 Random-Number Generation 6.47

 Tests for Autocorrelation
• Correlation at lag j
Cj
ρj =
C0
C j = Cov( X i , X i + j ) = E ( X i X i + j ) − E ( X i ) E ( X i + j )
C0 = Cov( X i , X i ) = E ( X i X i ) − E ( X i ) E ( X i ) = E ( X i2 ) − [ E ( X i )]2 = Var ( X i )
E( X i X i+ j ) − E( X i )E( X i+ j )
⇒ ρj =
Var ( X i )
• Assume Xi = Ui
1 1
E (U i ) = and Var (U i ) =
2 12
1
E (U iU i + j ) −
ρj = 4 = 12 E (U U ) − 3
i i+ j
1
12

Prof. Dr. Mesut Güneş ▪ Ch. 6 Random-Number Generation 6.48

 Tests for Autocorrelation
• Test statistics is: ρˆ i ,m
Z0 =
σˆ ρˆ i ,m

• Z0 is distributed normally with mean = 0 and variance = 1, and:

1 ⎡ M ⎤
ρˆ i ,m = ⎢ ∑ Ri + km × Ri +(k +1 )m ⎥ − 0.25
M + 1 ⎣ k =0 ⎦
13M + 7
σˆ ρi ,m =
12( M + 1)

• After computing Z0 do not reject

the hypothesis of
independence if –zα/2 ≤ Z0 ≤ –zα/2

Prof. Dr. Mesut Güneş ▪ Ch. 6 Random-Number Generation 6.49

 Tests for Autocorrelation
• If ρi,m > 0, the subsequence has positive autocorrelation
• High random numbers tend to be followed by high ones, and vice versa.

• If ρi,m < 0, the subsequence has negative autocorrelation

• Low random numbers tend to be followed by high ones, and vice versa.

Prof. Dr. Mesut Güneş ▪ Ch. 6 Random-Number Generation 6.50

 Example
• Test whether the 3rd, 8th, 13th, and so on, for the numbers on
Slide 38.
• Hence, α = 0.05, i = 3, m = 5, N = 30, and M = 4

1 ⎡(0.23)(0.28) + (0.28)(0.33) + (0.33)(0.27 )⎤

ρˆ 35 = ⎢ ⎥ − 0.25
4 + 1 ⎣+ (0.27 )(0.05) + (0.05)(0.36) ⎦
= −0.1945
13(4) + 7
σ ρˆ 35 = = 0.128
12( 4 + 1 )
0.1945
Z0 = − = −1.516
0.1280

• z0.025 = 1.96
• Since -1.96 ≤ Z0 = -1.516 ≤ 1.96, the hypothesis is not rejected.

Prof. Dr. Mesut Güneş ▪ Ch. 6 Random-Number Generation 6.51

 Shortcomings
• The test is not very sensitive for small values of M, particularly
when the numbers being tested are on the low side.
• Problem when “fishing” for autocorrelation by performing
numerous tests:
• If α = 0.05, there is a probability of 0.05 of rejecting a true
hypothesis.
• If 10 independence sequences are examined:
• The probability of finding no significant autocorrelation, by chance
alone, is 0.9510 = 0.60.
• Hence, the probability of detecting significant autocorrelation
when it does not exist = 40%

Prof. Dr. Mesut Güneş ▪ Ch. 6 Random-Number Generation 6.52

 Real Random Numbers

Prof. Dr. Mesut Güneş ▪ Ch. 6 Random-Number Generation 6.53

 Real Random Numbers
• There are also sources for
real random numbers in the
Internet
• www.random.org
„RANDOM.ORG offers true
random numbers to anyone
on the Internet. The
randomness comes from
atmospheric noise, which for
many purposes is better than
the pseudo-random number
algorithms typically used in http://www.random.org/analysis/
computer programs. People
use the numbers to run
lotteries, draws and
sweepstakes and for their
games and gambling sites.”

Prof. Dr. Mesut Güneş ▪ Ch. 6 Random-Number Generation 6.54

 Real Random Numbers
• http://www.randomnumbers.info/
„It offers the possibility to download true random
numbers generated using a
quantum random number generator upon demand. “

Prof. Dr. Mesut Güneş ▪ Ch. 6 Random-Number Generation 6.55

 Real Random Numbers

• Hardware based generation

of random numbers

• http://www.comscire.com

Prof. Dr. Mesut Güneş ▪ Ch. 6 Random-Number Generation 6.56

 Summary
• In this chapter, we described:
• Generation of random numbers
• Testing for uniformity and independence
• Sources of real random numbers

• Caution:
• Even with generators that have been used for years, some of which
still in use, are found to be inadequate.
• This chapter provides only the basics
• Also, even if generated numbers pass all the tests, some underlying
pattern might have gone undetected.

Prof. Dr. Mesut Güneş ▪ Ch. 6 Random-Number Generation 6.57