2nd year Master University Amar Télidji Laghouat

Decision Support
Systems
Decision Making under
Certainty
Younes Guellouma
Chapter 3: Decision Making under Certainty

Younes Guellouma

1
Introduction

Preference Relation
Reminder on binary relations
Representations
Decision Maker Preference
Utility Fonction

Indifference Paradox

Measurement Systems and Scales

2
Introduction
Objectives

Decision-making is a process of identifying problems and

opportunities that aims to effectively choose a better option
from a well-defined list of alternatives and actions.
A condition of certainty may exist when the decision-maker
knows clearly the alternatives they face and the consequences
of each one. Thus, the decision-maker fully understands the
situations and the necessary resources, the time allocated for
decision-making, the related issues, as well as the costs of
each action leading to a decision. In almost all cases, the
solution is already known due to past experiences.

3
Objectives (2)

In this chapter, we discuss the different ways of modeling a

decision, the possible actions, as well as the preferences of a
decision-maker. We begin with various theoretical aspects to
quantify and compare actions and thus find the best solution.
This allows us to automate the decision-making process in the
certain universe. The formalisms used are also applicable to
other types of decisions, which will be the subject of the
following chapters.

4
Preference Relation
Preference Relation

When making decisions, or when choosing from a set of

alternatives, it is more natural to use carefully defined
standards to distinguish what is good from what is not. The
choice of a quantitative standard seems more ambitious.
Decision theory assumes that such a standard is admissible
and thus defines precise quantified models to respond as
adequately as possible to the needs.
To express the choices of a decision-maker, a preference
relation must be defined. We present existing models to define
a quantified comparison relation between alternatives in line
with the decision-maker’s expressions.

5
Reminder on binary relations

Let A = a1 , . . . , an be a finite set of n alternatives. In decision

theory, the exhaustive list of allowed actions or alternatives is
previously known by the decision-maker. ai for 0 ≤ i ≤ n can
be seen as a unique index of one of the alternatives that might
be the best solution to a given decision problem.
The problem of finding the best alternative can be viewed as a
search for a comparison relation among these alternatives.

6
Binary Relation

A binary relation R on A is a subset of the Cartesian product

A2 . We can say that two alternatives a, b ∈ A are in relation R.
This expression is formalized as follows.

aRb ⇔ (a, b) ∈ R (1)

7
Related Notions

Furthermore, in this comparison relation, we can define the set of

binary operations within sets. Let R and S be two relations on A:

1.

R⊆S ⇔ ∀a, b ∈ A, aRb ⇒ aSb (2)

∀a, b ∈ A, a(R ∪ S)b ⇔ aRb ∨ aSb (3)
∀a, b ∈ A, a(R ∩ S)b ⇔ aRb ∧ aSb (4)

2. The comlement of R in A2 , is: CA∈ (R) = {(a, b) ∈ A2 | (a, b) ̸∈ R}

3. The dual of R is DA (R) = {(a, b) ∈ A2 | (b, a) ∈ R}
4. The co-dual of R is CDA (R) = {(a, b) ∈ A2 | (b, a) ̸∈ R}
5. The composition relation of R and S is :
R ◦ S = {(a, b) ∈ A | ∃c, aRc ∧ cSb}

8
We can also redefine the properties of binary relations as follows:

• Reflexivity: ∀a ∈ A, aRa
• Symmetry: ∀a, b ∈ A, aRb ⇒ bRa
• Asymmetry: ∀a, b ∈ A, aRb ⇒ ¬bRa
• Anti-symmetry: ∀a, b ∈ A, aRb ∧ bRa ⇒ a = b
• Completeness: ∀a, b ∈ A, aRb ∨ bRa
• Transitivity: ∀a, b, c ∈ A, aRb ∧ bRc ⇒ aRc
• Negative transitivity: ∀a, b, c ∈ A, aCA (R)b ∧ bCA (R)a ⇒ aCA Rc
or aRb ⇒ aRc ∧ cRb
• Ferrers relation: ∀a, b, c, d ∈ A, (aRb ∧ cRd) ⇔ (aRd ∨ cRb)
• Semi-transitivity: ∀a, b, c, d ∈ A, aRb ∧ bRc ⇔ (aRd ∨ dRc)

9
It is also useful to define known types of relations:

1. Equivalence relation: R is symmetric, reflexive, and

transitive.
2. Preorder relation: R is reflexive and transitive.
3. Strict partial order: R is asymmetric and transitive.
4. Partial order: R is reflexive, antisymmetric, and transitive.
5. Total order: R is complete, antisymmetric, and transitive.

10
The most natural representation for mathematical formalisms
and proofs of a binary relation is the set-theoretic
representation. However, this representation is less efficient
when it comes to automation and implementation. For this
reason, the graphical form (which is equivalent to a matrix
representation) is more suitable for modeling a
decision-maker’s preferences.
We define a square matrix that models the direct relationship
that may exist between any two alternatives.

11
Representative Matrix

Let A = a1 , . . . , an be a finite set of alternatives. Let R be a

binary relation defined on A. The representative matrix of R,
denoted matR , with coefficients in 0, 1 and of dimension
|A| × |A|, is defined as follows.

{
1 if ai Raj
matR (i, j) = (5)
0 otherwise

12
Representative Graph

Indeed, this representation also allows for the definition of a

directed graph as follows.
Let A = a1 , . . . , an be a finite set of alternatives. Let R be a
binary relation defined on A, and let matR be its
representative matrix. The representative graph GR = (X, A) is
constructed as follows:

• X = A, the set of vertices,

• A ⊆ X2 , the set of edges, such that (ai , aj ) ∈ A ⇔ ai Raj

13
Example

Let A = a, b, c be the set of alternatives and

R = (a, a), (b, a), (b, c), (c, c) a relation on A. We observe that
the relation R:

• is not reflexive: ¬bRb

• is not symmetric (aRb but ¬bRa)
• etc.

14
Example

The representative matrix matR is:

R a b c
a 1 0 0
b 1 0 1
c 0 0 1

And the representative graph GR is:

a c

15
Decision Maker Preference

In economics and social sciences, preference is the result

obtained when a person makes a choice based on their relative
utility. The individual nature of this choice is determined
purely by a set of social factors such as taste, desire, or reason.
With the help of scientific methods, several models of
preference relations have emerged. The first model was
developed by Ragnar Frisch and was defined within an
economic context. Since then, many more sophisticated
models have been periodically proposed.

16
Decision Maker Preference

A preference relation can be seen as a comparison relation

between alternatives by a decision-maker. Consider a pair of
alternatives (a, b). A decision-maker is expected to express the
comparison as follows: either a is as good as b (or vice versa),
or a and b are both indifferent (meaning either alternative is
acceptable).
It should be noted that a preference relation is assumed to be
reflexive: each alternative can be compared to itself. By
convention, we may say that a and a are indifferent.
In a certain environment, the default assumption is that any
pair of alternatives can be compared. Lastly, it is natural to
consider a preference relation as transitive: if a is better than
b and b is better than c, we can say that a is better than c.
17
Preference Relation

Let A be a set of alternatives. A preference relation ⪯ is a

complete binary preorder relation on A (a reflexive, transitive,
and complete relation).
For clarity, we can distinguish two parts of a preference
relation ⪯ as follows:
1. The symmetric part ∼ which models indifference
∼= D(⪯)∩ ⪯
.
2. The asymmetric part ≺ that models the strict comparison
relation:
≺= CD(⪯)∩ ⪯
.
18
Preference Relation (2)

We observe that if a ≺ b then b ̸⪯ a. Similarly, if a ∼ b then

b ∼ a. We can therefore state the following:
Lemma
Let ⪯ be a preference relation on A. Then:

1. ⪯=≺ ∪ ∼
2. ≺ ∩ ∼= ∅
3. ≺ is a strict order relation.
4. ∼ is an equivalence relation.

19
EXERCISE

Exercise
Prove the previous lemma.

20
Utility Fonction

The concept of utility is well known in economics. This concept

is measured by a function called the utility function. This
function must be increasing; an alternative with a higher
income coincides with a higher level of satisfaction.
Fortunately, a property of set theory allows us to transform the
set-theoretic world into a quantitative world.

21
Theorem
Let R be a binary relation on A. Then, R is complete and
transitive if and only if there exists a function u : A → R such
that ∀a, b ∈ A, aRb ⇔ u(a) ≤ u(b).

Recall that a complete relation is always transitive.

From this fundamental result, we can associate with a
preference relation (which is reflexive and transitive) a
real-valued function that allows for comparison, ranking, and
the determination of the maximum for a set of alternatives.

22
This result can be extended to the symmetric and asymmetric
parts of a preference relation.
Definition
Let ⪯ be a preference relation on a set of alternatives A. A
utility function u : A → R is any function that satisfies for all
a, b ∈ A:

• a ⪯ b ⇔ u(a) ≤ u(b).
• a ≺ b ⇔ u(a) < u(b).
• a ∼ b ⇔ u(a) = u(b).

23
Example
Let A = {a, b, c} and ⪯= {(a, a), (b, b), (c, c), (a, b), (b, c), (a, c)} be
a preference relation. Note that ⪯ is a complete preorder. We can
compute the symmetric and asymmetric parts:

1. ≺= {(a, b), (b, c), (a, c)}

2. ∼= {(a, a), (b, b), (c, c)}

If we set u(a) = 0, u(b) = 1 and u(c) = 2, we can see that this is

indeed a utility function:

a ∼ a and u(a) = u(a), b ∼ b and u(b) = u(b), c ∼ c and u(c) = u(c),

a ≺ b and u(a) < u(b), a ≺ c and u(a) < u(c), b ≺ c and u(b) < u(c),

Therefore, the function u can replace the set-theoretic

representation of the preference relation.

24
Indifference Paradox
Indifference Paradox

In this section, we focus on the violation of the transitivity

property in a preference relation. Indeed, despite the
predominance of transitivity at the individual, social, or
economic level in decision theory, this principle may not be
respected in several real-world contexts. In some cases, the
decision-maker is unable to detect minimal changes between
two alternatives and may judge them as indifferent. The
Duncan Luce experiment is a representative example.

25
Indifference Paradox

Consider the following experiment: A person is presented with

a series of coffee cups, numbered from 1 to n, where each cup
contains i milligrams of sugar. The difference in sweetness
between two consecutive cups is so small that the person is
unable to detect it:

t1 ∼ t2 ∼ . . . ∼ tn

However, when asked to compare the first and last cups, they
can distinguish a significant difference:

t1 ≺ tn

26
Indifference Paradox

This situation illustrates a violation of the transitivity principle.

In fact, if we have a ∼ b and b ∼ c we should logically have
a ∼ c. Yet, in our example, we have t1 ∼ t2 , t2 ∼ tn but t1 ̸∼ tn .
This inconsistency is due to limitations in our sensory
perception, which prevents us from detecting very small
differences.
This phenomenon, known as the ’indifference paradox,’ is
frequently observed in behavioral sciences. It highlights the
fact that human preferences are not always rational and can
be influenced by contextual and psychological factors.”

27
Quasi-order

To accommodate these natural phenomena in

decision-making, we need to ’weaken’ the transitivity property.
The solution is to use the notion of quasi-orders to represent
preference relations with a perception threshold.
Definition
Let R be a binary relation defined on A. Then R is a
quasi-order if it is complete, Ferrers, and semi-transitive.

28
Quasi-order

A preference relation defined as a quasi-order coincides with

the definition of the perception threshold:
Theorem
Let ⪯=≺ ∪ ∼ be a binary relation defined on a set of
alternatives A. Then ⪯ is a quasi-order if and only if there
exists a utility function u : A → R and a perception threshold
q ∈ R+ such that for all a, b ∈A :

{
a ≺ b ⇔ u(a) + q < u(b)
(6)
a ∼ b ⇔ |u(a) − u(b)| ≤ q

29
Example

Suppose a, b, and c are alternatives with the following utilities:

u(a) = 3, u(b) = 5 and u(c) = 7. If the decision maker reported that
a ≺ b, b ∼ c then we have:

a ≺ b ⇔ u(a) + q < u(b)

⇔ q < u(b) − u(a)
⇔q<2
but

b ∼ c ⇔ |u(c) − u(b)| ≤ q
⇔q≥2

Therefore, the preference cannot be represented as quasi-orders in

this situation. 30
Interval Orders

However, it is common to encounter situations where the

perception threshold is not fixed but varies depending on the
alternative in question. For example, the human eye can
distinguish fine variations in light in the dark but is unable to
differentiate the same variation in brighter places.
To overcome this situation, a generalization of quasi-orders
allows for a further weakening of the constraints on a
preference relation. In fact, interval orders allow for the
definition of utility functions with a variable threshold.
Definition
Let A be a finite set. A relation R defined on A is an interval
order if and only if it is complete and of Ferrers.

31
Interval Orders

In the same way, we can define a utility function with a

variable perception threshold.
Théorème
Let ⪯=≺ ∪ ∼ be a binary relation defined on a set of
alternatives A. ⪯ is an interval order if and only if there
exists a utility function u : A → R and a perception function
q : A → R such that for a, b ∈A :

{
a ≺ b ⇔ u(a) + q(a) < u(b)
a ∼ b ⇔ (u(a) ≤ u(b) + q(b)) ∧ (u(b) ≤ u(a) + q(a))
(7)

32
Measurement Systems and Scales
Measurement Systems and Scales

We have seen that a utility function allows for a quantitative

representation of a decision maker’s expressed preference.
This concept is highly useful; for example, a fitness function is
used in various fields to evaluate alternatives. Recall that the
list of alternatives contains only the ”indices” of actions that a
decision maker may take. An alternative a can therefore index
an action described by a vector a(a1 , . . . , am ) belonging to a
description space defined by a set of variables known to the
decision maker. An evaluation function u(a1 , . . . , am ) can thus
represent a utility function.
However, we note that the uniqueness of a utility function is
not guaranteed. Indeed, an infinite number of utility functions
can be derived since for any increasing function ϕR → R, ϕ ◦ u
is also a utility function.” 33
Measurement Systems and Scales

Therefore, a rather general criterion is needed to determine

which utility function is better suited to describe the
relationships that may exist among alternatives.
To address this issue, we refer to measurement systems and
scales.
Definition
Let A be a set of alternatives and ⪯ a preference relation
defined on that set. Let u : A → R be the associated utility
function. A measurement system is defined as the pair
M = (A, u). Let M = (A, u) ube a measurement system. A
scale is the tuple (M, N, φ) with φ : R → R where
N = (A, ϕ ◦ u).

34
Now, the question is whether a scale produces a measurement
system that respects the decision-maker’s preferences. This
leads us to define the notion of an admissible scale.
Definition
Let M = (A, u) be a measurement system. (M, N, φ) is an
admissible scale if and only if

∀a, b ∈ A, u(a) ≤ u(b) ⇔ φ(u(a)) ≤ φ(u(b)) (8)

35
Some common Scales

A sufficient condition to a scale to be admissible is the

monotonicity of the function φ.
Indeed, there exists a large number of scales. Here, we are
particularly interested in real (quantitative) scales. Below are
some widely used scales in practice:

Ration Scale φ(x) = ax, a ≥ 0 physical conver-

sions, monetary ...
Interval Scale φ(x) = ax + b, a ≥ 0 Temperature, dates
...
Ordinal Scale φ monotonically increasing //

36
Example
Let = {a, b, c} and ⪯= {(a, a), (b, b), (c, c), (a, b), (b, c), (a, c)} ube
a preference relation. We define u(a) = 0, u(b) = 1 and u(c) = 2.
Then (A, u) constitutes a measurement system. If we apply the
following ratio scale: φ(x) = 5x − 2 and define v = φ ◦ u, we get

v(a) = −2, v(b) = 3, v(c) = 7

We observe that the measurement system (A, v) respects the

preference ⪯. Therefore, the scale φ is admissible.
1
However, if we take a decreasing scale on ] − 1, +∞[: φ′ = x+1
(defining w = φ′ ◦ u), we obtain:

w(a) = 1, w(b) = 0.5, u(c) = 0.33

. We observe that the measurement system (A, w) does not respect

the preference relation ⪯ since φ′ is not admissible.

37