Mathematical

Modelling
Part 3: Probability Model
 Content of part 2

Chapter 1: Introduction to probability models

Chapter 2: Stochastic models

Chapter 3: Simulation of probability models

3
Introduction to probability models
• Most real-life problems contain elements of
uncertainty.
• We should consider additionally random
elements to account for uncertainties in human
behaviour.
• Probability is a familiar and intuitive idea.

4
Outline

• Discrete probability model

• Continuous probability model

5
Discrete probability models
Example: An electronics manufacturer produces a
variety of diodes. Quality control engineers attempt to
insure that faulty diodes will be detected in the factory
before they are shipped. It is estimated that 0.3% of the
diodes produced will be faulty. It is possible to test each
diode individually. It is also possible to place a number of
diodes in series and test the entire group. If this test fails,
it means that one or more of the diodes in that group are
faulty. The estimated testing cost is 5 cents for a single
diode, and 4 + 𝑛 cents for a group of 𝑛 > 1 diodes. If a
group test fails, then each diode in the group must be
retested individually to find the bad one(s). Find the most
cost-effective quality control procedure for detecting bad
diodes.

6
Step 1: Ask a question

• Here, 𝑛 is decision variable, 𝐶 is the random outcome

of the quantity control procedure.
7
Step 2: Select a modelling approach

• We select a discrete probability model for this problem.

• Consider a random variable 𝑋, which can take any of a
discrete set of values 𝑋 = {𝑥1 , 𝑥2 , … } and suppose that
𝑋 = 𝑥𝑖 occurs with probability 𝑝𝑖 . We will write
Pr 𝑋 = 𝑥𝑖 = 𝑝𝑖 . Of course, we must have ∑𝑝𝑖 = 1.
• Since 𝑋 takes the value 𝑥𝑖 with probability 𝑝𝑖 , the
average of expected value of 𝑋 should be a weighted
average of the possible values 𝑥𝑖 , weighted according
to their relative likelihoods 𝑝𝑖 . We will write 𝐸𝑋 = ∑𝑥𝑖 𝑝𝑖 .
• The probabilities 𝑝𝑖 represent what we will call the
probability distribution of 𝑋.

8
Step 3: Formulate the model

We see that the random variable 𝐶 takes on one

of two possible values for any fixed 𝑛 > 1. If all
the diodes are good, then
𝐶 =4+𝑛
Otherwise 𝐶 = 4 + 𝑛 + 5𝑛 since we have to
retest each diode.
Let 𝑝 denote the probability that all the diodes
are good, the remaining possibility must have
probability 1 − 𝑝. The expected value of 𝐶 is
𝐸𝐶 = 4 + 𝑛 𝑝 + [ 4 + 𝑛 + 5𝑛](1 − 𝑝)
9
Step 4: Solve the model
• There are 𝑛 diodes, the probability that one individual
diode is good is 0.997. Assuming independence, it
follows that all 𝑛 diodes in one test group are good is
𝑝 = 0.997𝑛 .
• The expected value of 𝐶 is

10
Step 4: Solve the model
• Hence, the average testing cost per diode is
4
𝐴 = + 6 − 5 0.997 𝑛
𝑛
• Now all we need is to minimize 𝐴 as a function
of 𝑛.
• The minimum of 𝐴 = 1.48 cents/diode occurs at
𝑛 = 17.

11
Step 5: Answer the question
Quality control procedures for detecting faulty
diodes can be made considerably more
economical by group testing methods. Individual
testing costs approximately 5 cents/unit. Bad
diodes occur only rarely, at a rate of 3 per 1,000.
By testing groups of 17 diodes each, in series,
we can reduce testing costs by a factor of three
(to 1.5 cents/diode) without sacrificing quality.

12
Sensitivity analysis
• We may test diodes in batches of 10 or 20, or
perhaps 𝑛 should be a multiple of 4 or 5
(properties of the problem)
• Fortunately, the average cost 𝐴 does not vary
significantly between 𝑛 = 10 and 𝑛 = 35.
• We also do sensitivity analysis for parameter
𝑝 = 0.003.

13
Outline

• Discrete probability model

• Continuous probability model

14
Continuous probability models
Example: A “type I counter” is used to measure
the radioactive decay in a sample of fissionable
material. Decays occur at random, at an unknown
rate, and the purpose of the counter is to measure
the decay rate. Each radioactive decay locks the
counter for a period of 3 × 10−9 seconds, during
which time any decays that occur are not counted.
How should the data received from the counter be
adjusted to account for the lost information?

15
Step 1: Ask a question

16
Step 2: Select a modelling approach

• We use a continuous probability model for this

problem.
• Suppose that 𝑋 is a random variable that takes
values on the real line. A convenient way to
describe the probability structure of 𝑋 is to specify
the function 𝐹 𝑥 = Pr 𝑋 ≤ 𝑥 called the
distribution function of 𝑋. If 𝐹 𝑥 is differentiable,
we call the function 𝑓 𝑥 = 𝐹′(𝑥) the density
function of 𝑋. Then for any real numbers 𝑎, 𝑏 we
have
𝑏
Pr 𝑎 < 𝑋 ≤ 𝑏 = 𝐹 𝑏 − 𝐹 𝑎 = න 𝑓 𝑥 𝑑𝑥
𝑎

17
Step 2: Select a modelling approach

• In other word, the area under the density curve

represents probability. The mean or expected
value of 𝑋 is defined by
+∞
𝐸𝑋 = න 𝑥𝑓 𝑥 𝑑𝑥
−∞

18
Step 3: Formulate the problem
We assume that radioactive decays occur at random at
an unknown rate 𝜆. We will model this process by
assuming that the times between successive radioactive
decays are independent and identically distributed with
an exponential distribution with rate parameter 𝜆. Let

𝑋𝑛 = 𝑇𝑛 − 𝑇𝑛−1
denote the times between successive observations of a
radioactive decay. Of course, 𝑋𝑛 does not have the same
distribution as the time between successive decays,
because of the lock time.

19
Step 3: Formulate the problem
• The random time 𝑋𝑛 consists of two parts.
First, we must wait 𝑎 = 3 × 10−9 seconds while
the counter is locked, and then we must wait
an additional 𝑌𝑛 seconds until the next decay.
Now 𝑌𝑛 is not simply the time between two
decays, because it begins at the end of the
lock time, not at a decay time. However, the
memoryless property of the exponential
distribution guarantee that 𝑌𝑛 is still exponential
with rate parameter 𝜆.

20
Step 4: Solve the model
• Since 𝑋𝑛 = 𝑎 + 𝑌𝑛 , we have 𝐸𝑋𝑛 = 𝑎 + 𝐸𝑌𝑛 where
∞
𝐸𝑌𝑛 = න 𝑡𝜆𝑒 −𝜆𝑡 𝑑𝑡
0
1 1
• Integrate by parts to find 𝐸𝑌 = . Thus, 𝐸𝑋𝑛 = 𝑎 + .
𝜆 𝜆
𝑋1 +⋯+𝑋𝑛 1
• From the law of large numbers, so lim =𝑎+
𝑛 →∞ 𝑛 𝜆
𝑇𝑛 1
with probability 1. In other words, →𝑎+ .
𝑛 𝜆
𝑇 1
• For large 𝑛 it will be approximately true that 𝑛 = 𝑎 + .
𝑛 𝑛
𝑛
• Solving for 𝜆, we obtain 𝜆 =
𝑇𝑛 −𝑛𝑎

21
Step 5: Answer the question
We have obtained a formula for decay rate that corrects
for the decays missed while the counter is locked. All
that is required is to record the length of observation and
the number of decays recorded. The distribution of those
decays in the observation interval is not required to
determine 𝜆.

22
Sensitivity analysis
• We consider the lock time 𝑎, which must be
determined empirically.

23
Exercises

24