Random Number Tests

Here are the key formulas and an example from your document regarding testing random number generators:

1. Chi-Square (χ²) Test for Uniformity

Formula:

To test if numbers u1 , u2 , … , un are uniformly distributed:

​ ​ ​

1. Divide the interval (0, 1) into k equal segments.

2. Oi : Observed count of numbers in segment i

​

3. Ei = n/k : Expected count in each segment (assuming uniformity)

​

The Test Statistic (TS) is:

k
(Oi − Ei )2
χ2 = ∑
​ ​

​ ​

i=1
Ei ​

H0 : Numbers are uniformly distributed

​

Ha : Numbers are not uniformly distributed

​

Reject H0 if χ2 ​ > χ2α,k−1 (from critical table)

​

Example:

100 random numbers generated.

Divide the interval (0, 1) into k = 10 equal segments.

Compute Oi , Ei ​ ​ = 10, and the test statistic:

(Oi −Ei )2
Interval Oi Ei
​ ​

Ei
​
​ ​

[0.0, 0.1) 12 10 0.4

[0.1, 0.2) 9 10 0.1

[0.2, 0.3) 10 10 0.0

[0.3, 0.4) 13 10 0.9

[0.4, 0.5) 12 10 0.4

[0.5, 0.6) 8 10 0.4

[0.6, 0.7) 16 10 3.6

[0.7, 0.8) 5 10 2.5

[0.8, 0.9) 5 10 2.5

[0.9, 1.0] 10 10 0.0

Test Statistic:

χ2 = 0.4 + 0.1 + 0 + 0.9 + 0.4 + 0.4 + 3.6 + 2.5 + 2.5 + 0 = 10.8

Critical Value for α = 0.05 and k − 1 = 9 is χ29,0.05 = 16.9.

​

Since 10.8 < 16.9, do not reject H0 . ​

Conclusion: At the 5% significance level, the numbers appear uniformly distributed.

1/4
2. Runs Test for Independence

Formula:

Test if numbers are independent based on the number of "runs" (increasing or decreasing sequences):

Mean of runs:

2n − 1
E[R] =
3
​

Variance of runs:

16n − 29
σR2 =
90
​ ​

Test Statistic:

r − E[R]
z= ​

σR ​

Reject H0 if ∣z∣ ​
> zα/2 (from standard normal table)
​

Let me know if you need a more detailed breakdown or help with another example!

Here is an example for the Runs Test for Independence:

Problem:
Suppose we have a sequence of 20 numbers:
0.15, 0.78, 0.22, 0.95, 0.33, 0.85, 0.11, 0.68, 0.25, 0.91, 0.42, 0.76, 0.08, 0.61, 0.37, 0.88, 0.45, 0.93, 0.12, 0.74

Step 1: Determine Runs

A "run" is a sequence of consecutive increasing or decreasing numbers.

Compare each number with the next one:

If ui ​ < ui+1 , it is an increasing trend.

​

If ui ​ > ui+1 , it is a decreasing trend.

​

Mark increasing trends as + and decreasing trends as −:

0.15(+)0.78(−)0.22(+)0.95(−)0.33(+)0.85(−)0.11(+)0.68(−)0.25(+)0.91(−)0.42(+)0.76(−)0.08(+)0.61(−)0.37(+)0.88(−)0.45(+
Runs: +, −, +, −, +, −, +, −, +, −, +, −, +, −, +, −, +, −, +, −

Number of Runs (r): 19

Step 2: Compute Test Statistic

Given n = 20, use the formulas for the expected number of runs E[R] and the variance σR2 : ​

1. Expected Number of Runs:

2/4
 2n − 1 2(20) − 1 39
E[R] = = = = 13
3 3 3
​ ​

2. Variance of Runs:

16n − 29 16(20) − 29 320 − 29 291

σR2 = = = = ≈ 3.23
90 90 90 90
​ ​ ​ ​ ​

3. Standard Deviation:

σR = ​ σR2 = ​ ​ 3.23 ≈ 1.8 ​

4. Test Statistic:

r − E[R] 19 − 13 6
z= = = ≈ 3.33
1.8 1.8
​ ​ ​

σR ​

Step 3: Decision Rule

At significance level α = 0.05, the critical value for z from the standard normal table is z0.025 = ±1.96. ​

Since z = 3.33 > 1.96, reject H0 . ​

Conclusion:
At the 5% significance level, there is sufficient evidence to conclude that the sequence is not independent.

Serial Test Overview

The Serial Test is used to evaluate independence and uniformity of pseudo-random numbers in higher dimensions. It extends the Chi-Square (
χ2 ) test by examining d-tuples of random numbers instead of testing individual numbers.

How It Works
1. Data Preparation:

Arrange the random numbers u1 , u2 , u3 , … , un into d-dimensional tuples. For example, for d
​ ​ ​ ​ = 2, the tuples are:

(u1 , u2 ), (u3 , u4 ), (u5 , u6 ), …

​ ​ ​ ​ ​ ​

For d = 3, tuples would be (u1 , u2 , u3 ), (u4 , u5 , u6 ), ….

​ ​ ​ ​ ​ ​

2. Partition the d-Dimensional Space:

Divide the d-dimensional unit hypercube [0, 1]d into k d smaller regions or cells (e.g., k = 5).
3. Count Observations in Each Cell:

Oj1 ,j2 ,…,jd : Number of d-tuples falling into the (j1 , j2 , … , jd )-th cell.
​ ​ ​
​ ​ ​ ​

E= n
kd
: Expected number of tuples in each cell (assuming uniformity).

4. Calculate Test Statistic: The Serial Test Statistic is based on the Chi-Square formula:

k k k
(Oj1 ,j2 ,…,jd − E)2
χ2 = ∑ ∑ ⋯ ∑
​

​ ​ ​

​ ​ ​ ​

E
j 1 =1 j 2 =1
​ ​ j d =1
​

H0 : The random numbers are IID and uniformly distributed over [0, 1]d .
​

Ha : The random numbers are not IID or uniformly distributed.

​

5. Decision Rule:

3/4
 Reject H0 if χ2
​ > χ2α,kd −1 , where χ2α,kd −1 is the critical value for the given significance level α.
​ ​

Example: Serial Test for d =2

Problem:

A sequence of 20 random numbers:

0.12, 0.34, 0.56, 0.78, 0.23, 0.67, 0.89, 0.45, 0.10, 0.90, 0.13, 0.38, 0.59, 0.76, 0.27, 0.65, 0.88, 0.41, 0.15, 0.80.
1. Form Tuples: For d = 2:

(0.12, 0.34), (0.56, 0.78), (0.23, 0.67), …

2. Divide the Space: Suppose we divide the [0, 1]2 space into k = 2 intervals:
[0, 0.5) and [0.5, 1]. This creates k 2 = 4 cells:
Cell 1: [0, 0.5) × [0, 0.5)

Cell 2: [0, 0.5) × [0.5, 1]

Cell 3: [0.5, 1] × [0, 0.5)

Cell 4: [0.5, 1] × [0.5, 1]

3. Count Observations: Assign each tuple to a cell:

(0.12, 0.34): Cell 1

(0.56, 0.78): Cell 4
(0.23, 0.67): Cell 2
Continue for all tuples.

Counts:

O1 = 5, O2 = 5, O3 = 6, O4 = 4
​ ​ ​ ​

10
4. Compute Test Statistic: E = 4
​
= 5 (Expected count per cell):

(5 − 5)2 (5 − 5)2 (6 − 5)2 (4 − 5)2

χ2 = + + +
5 5 5 5
​ ​ ​ ​

1 1
χ2 = 0 + 0 + + = 0.4
5 5
​ ​

5. Decision: Critical value χ20.05,3

= 7.81. ​

Since 0.4 < 7.81, do not reject H0 . ​

Conclusion:

At a 5% significance level, there is insufficient evidence to reject the hypothesis that the numbers are IID and uniformly distributed.

The Serial Test can be extended to higher dimensions d, but as d increases, the complexity and required sample size n also increase. Let me know
if you'd like to explore this further!

4/4