COMPROMISE PROGRAMMING

MA7391 SEMINAR REPORT

submitted by

RAJESH BUDANIA
M180512MA

under the guidance of

Dr. SUSHAMA C M

in partial fulfillment for the award of degree of

MASTER OF SCIENCE (MATHEMATICS)

DEPARTMENT OF MATHEMATICS
NATIONAL INSTITUTE OF TECHNOLOGY CALICUT
KOZHIKODE 673601.
OCTOBER, 2019
 CERTIFICATE

This is to certify that the seminar report entitled Compromise Program-

ming submitted by Rajesh Budania to the National Institute of Technol-
ogy Calicut, in partial fulfillment for the award of degree of Master of Science
(Mathematics) is a bona-fide record of seminar work done by her under my
supervision.

Dr. SUSHAMA C M
Department of Mathematics
National Institute of Technology Calicut
Calicut 673601, Kerala.

Dr. SATYANANDA PANDA

Head, Department of Mathematics
National Institute of Technology Calicut
Calicut 673601, Kerala.

Place: NIT Calicut

Date: October, 2019.

 ACKNOWLEDGEMENT

I would like to express my deep sense of gratitude to my seminar supervisor Dr.

SUSHAMA C M, for her constant source of strength, invaluable guidance and
encouragement during the entire period of this work. I am extremely thankful
to her for giving me complete work freedom in expressing my views and then
render her experience and immense knowledge to guide me in right direction.

I would like to express my sincere thanks to Dr. SATYANANDA PANDA,

Head of the Department of Mathematics, National Institute of Technology
Calicut, for extending his support in the completion of this work.

I express my sincere gratitude to the Director, National Institute of Technology

Calicut, for providing me the necessary facilities as and when required.

I acknowledge my sincere gratitude to my parents for their constant encour-

agement and support during the seminar work. I thank almighty for guiding
my hands to success.

RAJESH BUDANIA
 ABSTRACT

Multi-objective optimization is an area of multiple criteria decision mak-

ing(DMs) that is concerned with mathematical optimization problems involv-
ing more than one objective function to be optimized simultaneously. At
present, various methods have introduced to optimize such problems.Compromise
programming(CP) model is a one of the optimization methods for multi-
objective problems. By seeking a solution as close as possible to the ideal
point,compromise programming requires less information on the decision maker
preference then goal programming and fewer computation then multi-objective
programming. It is an effective approach to highway maintenance resources
allocation problem, forest and other natural resource management and multi-
ple use planning.

Keywords: Goal Programming, Multi Objective Programming, Distance mea-

sures, Compromise Programming..

 Table of Content

Chapter 1. Introduction 1
Chapter 2. An intuitive introduction to the concept of distance measure 4
Chapter 3. A discrete approximation of the best-compromise solution 7
Chapter 4. An Illustrative Example 10
Chapter 5. Compromise Programming - a continuous setting 16
Chapter 6. The method of the displaced ideal 21
Chapter 7. Pros and cons of GP, MOP and CP 24
Chapter 8. Conclusion 26
References 27
 chapter 1

INTRODUCTION

Multiple-criteria decision-making (MCDM) or multiple-criteria decision anal-

ysis (MCDA) is a sub-discipline of operations research that explicitly evaluates
multiple conflicting criteria in decision making (both in daily life and in set-
tings such as business, government and medicine). Conflicting criteria are
typical in evaluating options: cost or price is usually one of the main criteria,
and some measure of quality is typically another criterion, easily in conflict
with the cost. In purchasing a car, cost, comfort, safety, and fuel economy
may be some of the main criteria we consider – it is unusual that the cheap-
est car is the most comfortable and the safest one. In portfolio management,
we are interested in getting high returns while simultaneously reducing risks;
however, the stocks that have the potential of bringing high returns typically
carry high risk of losing money. In a service industry, customer satisfaction
and the cost of providing service are fundamental conflicting criteria.
Multi-objective optimization (also known as multi-objective programming, vec-
tor optimization, multi-criteria optimization, multi attribute optimization or
Pareto optimization) is an area of multiple criteria decision making that is
concerned with mathematical optimization problems involving more than one
objective function to be optimized simultaneously. Multi-objective optimiza-
tion has been applied in many fields of science, including engineering, eco-
nomics and logistics where optimal decisions need to be taken in the presence
of trade-offs between two or more conflicting objectives. Minimizing cost while
maximizing comfort while buying a car, and maximizing performance whilst
minimizing fuel consumption and emission of pollutants of a vehicle are ex-

1
amples of multi-objective optimization problems involving two and three ob-
jectives, respectively. In practical problems, there can be more than three
objectives.
For a nontrivial multi-objective optimization problem, no single solution ex-
ists that simultaneously optimizes each objective. In that case, the objective
functions are said to be conflicting, and there exists a (possibly infinite) num-
ber of Pareto optimal solutions. A solution is called non-dominated, Pareto
optimal, Pareto efficient or non-inferior, if none of the objective functions can
be improved in value without degrading some of the other objective values.
Without additional subjective preference information, all Pareto optimal solu-
tions are considered equally good (as vectors cannot be ordered completely).
Researchers study multi-objective optimization problems from different view-
points and, thus, there exist different solution philosophies and goals when
setting and solving them. The goal may be to find a representative set of
Pareto optimal solutions, and/or quantify the trade-offs in satisfying the dif-
ferent objectives, and/or finding a single solution that satisfies the subjective
preferences of a human decision maker (DM).
Multi-objective optimization is an area of multiple criteria decision making(DMs)
that is concerned with mathematical optimization problems involving more
than one objective function to be optimized simultaneously. At present, var-
ious methods have introduced to optimize such problems. Compromise pro-
gramming(CP) model is a one of the optimization methods for multi-objective
problems.

• the CP model, DMs are able to establish, easily and precisely, goal values
of the considered objectives.

2
• advantages of CP model are its simplicity and its capability to be used
in conditions which it is not possible to access goal values of objectives.

• The method of compromise programming identifies solution which are

closest to the ideal solution as determined by some measure of distance.

• It allows to reduce the set of efficient solutions to a more reasonable size

without demanding any information from the DMs.

• The basic idea in CP is to identify an ideal or utopian solution which is

only a point of reference for the DMs.

• CP assumes that any DMs seeks a solution as close as possible to the

ideal point, possibly the only assumption made by CP about human
preferences.

3
 chapter 2

AN INTUITIVE INTRODUCTION TO THE

CONCEPT OF DISTANCE MEASURES

If someone has to find the distance d between two points, x1 = (x11 , x12 ) and
x2 = (x21 , x22 ) , defined in a Cartesian plane, the answer according to the
Pythagorean theorem is:

1
d = [(x11 − x21 )2 + (x12 − x22 )2 ] 2 (1)

Thus, for example, the distance d between (2,3) and (10,8) using this the-
orem is:

d = [(2 − 10)2 + (3 − 8)2 ]1 /2 = 9.43

This concept of distance can easily be extended to an n-dimensional space and

the formula in 1 becomes:
n
X  12
1 2 2
d= (xj − xj ) (2)
j=1

Although this is the best-known measure of proximity between two points, it

is not necessarily the only one. Thus, for some time now mathematicians using
the notion of a family of Lp metrics or a family of distance measures have been
able to provide a generalisation of the Euclidean distance as:
n
X  p1
Lp = |(x1j − x2j )|p (3)
j=1

where the vertical lines represent absolute values. Obviously for each value
of the parameter p, a particular distance is obtained. Thus, the Euclidean

4
distance measure given above is a particular case of the family of Lp metrics
when p = 2. For p = 1, expression (3) reduces to the following L1 metric:
n
X
L1 = |(x1j − x2j )| (4)
j=1

Thus, for our numerical example the L1 distance will be:

L1 = [(2 − 10) + (3 − 8)] = 13

Geometrically, the L1 distance can be interpreted as the sum of the length

of the sides of the triangle defined in Figure 1. When the parameter p has
values greater than 2, it is not possible to give a geometrical interpretation to
the distance measure; nonetheless, for such dimensions these distances can be
computed 9 Thus, for instance, when a range of values is attached to p, the
following distance measures are obtained for the two points of our example.

L1 = [(2 − 10) + (3 − 8)] = 13

1
L2 = [(2 − 10)2 + (3 − 8)2 ] 2 = 9.43
1
L3 = [(2 − 10)3 + (3 − 8)3 ] 3 = 8.6

.
1
L8 = [(2 − 10)8 + (3 − 8)8 ] 8 = 8.2

L∞ = M ax(|2 − 10|, |3 − 8|) = 8

5
where the Max signifies obtaining the largest absolute difference within the
brackets 9 It is interesting to point out that as p increases more weight is
given to the largest deviation 9 Thus, when p = ∞ the L= distance is given
exclusively by the largest deviation (i.e. 2 - 10 = 8 for our example). In other
words the parameter p weights the deviations according to their magnitudes.

From the data of our example it is easy to see that L1 is the largest distance
and L= the shortest distance. This is a general property of the Lp metrics that
is used extensively in developing the CP approach. We should however add
that all these possible distance measures are bounded by the ’longest’, the L1
metric, and the ’shortest’, the L∞ metric, distances. This is also called the
Chebyshev distance.
Obviously, in a strict two-dimensional geometric sense the use of Lp metrics
for values of the parameter p greater than two is meaningless, because it would
mean the existence of distances shorter than the straight line! However, the use
of such metrics can be very useful, if they are used not in a geometric sense but
as a measure for human preferences. In the following sections the Lp metrics
are used to calculate’distances’ between solutions belonging to the efficient set
and an ideal or a utopian point. The use of the distance concept as a proxy
measure for human preferences makes CP approach a sound practical method
in helping the DM to choose an optimum or the best-compromise solution from
the efficient ones generated by MOP.

6
 chapter 3

A discrete approximation of the best-compromise

solution

The concept of ideal solution was introduced simultaneously by Yu (1973) and

Zeleny (1973). Once this point has been defined, it is possible to establish
the best-compromise solution as the nearest solution with respect to the ideal,
accepting the basic postulate that the DM prefers solutions as close as possible
to the ideal (Zeleny’s axiom of choice). These simple but ingenious ideas form
the basis of the CP approach.
Given the inherent conflict of multiple objectives, the ideal solution is infea-
sible; it is therefore necessary to look for compromise solutions. For that we
need to calculate distances between each solution and the ideal point. The
degree of closeness, dj , between the jth objective and its ideal is given by:

dj = Zj∗ - Zj (X) if Zj (X) is to be maximize or

dj = Zj (x) - Zj∗ if Zj (X) is to be minimize

Z being the ideal value for a given objective. The degrees of closeness between
the different objectives and their ideal values are then added into a composite
distance function. But as the units used to measure various objectives are
different, therefore, in order to avoid a meaningless summation - adding pints
of bitter to kilos of potatoes - the units of the measurement of the various
objectives must be normalised. Not only that, if the absolute values for the
achievement levels of the several objectives are different, then the scalarisation

7
or normalisation of the degrees of closeness is necessary to avoid solutions bi-
ased towards those objectives that can achieve larger values. Can be overcome
by using relative deviations rather than absolute ones. Thus, the degree of
closeness d i is given by:
|Zj∗ −Zj (X)|
dj = |Zj∗ −Z∗j |

where Zj∗ is the anti-ideal or nadir point for the jth objective. The normalised
degrees of closeness are bounded between 0 and 1; therefore, when an objective
achieves its ideal solution then the degree of closeness is 0. On the contrary,
when an objective achieves its anti-ideal solution then the degree of closeness
is 1. Hence the normalised degrees of closeness measure the percentage of
achievement of one objective with respect to its ideal value.
To measure the distance between each solution and the ideal point, CP uses
the family of Lp metrics given by expression (5) as follows:
m ∗ p
p |Zj − Zj (X)|
X
M inLp (W ) = Wj (5)
j=1
|Zj∗ − Z∗j |p

or what is equivalent to:

m ∗ p
p |Zj − Zj (X)|
X
Lp (W ) = Wj (6)
j=1
|Zj∗ − Z∗j |p

where Wp are the weights representing the importance of the discrepancy be-
tween the jth objective and the ideal point; that is, Wj measures the relative
importance of the jth objective in a given decision situation.
The family of distance functions (5.6) can be applied to a set of feasible and
efficient alternatives in order to choose the best-compromise solution. Thus,
the alternative with the lowest value for Lp (W ) will be the best-compromise
solution because it is the nearest solution with respect to the ideal point. Obvi-
ously, the best-compromise solution can change according to the values of the

8
parameter p and the weights Wj that are chosen by the DM. The parameter
p acts as a weight attached to the deviations according to their magnitudes.
Similarly Wj s become the weights for various deviations signifying the relative
importance of each objective. For different sets of values of p and Wj we can
generate different compromise solutions.
0 0 0
The distances between each extreme efficient point (A , B and C ) and the
ideal point have been calculated for the three measures of distance L1 , L2 and
L∞ and for several structures of weights.

9
 chapter 4

An Illustrative Example

A manufacturer of solar technology component consider mass producing two

type of Photovoltaic solar cells for energy conversions:
(1) a cell module(product 1) for applications weather satellites and
(2) a cell chip(product 2) for use in pocket calculator.
Unfortunately the production process results emissions of a pollutant in to
the atmosphere which,the company would like to minimize and president of
the company would like to maximize to profit.Trade offs between these two
objectives are sought to assist management in the decision making process.

Product 1 Product 2

No of units produced X1 X2
Unit of raw material required 1.0 5.0
Machine time required(hr) 0.5 0.25
Assembly time required(hr) 0.2 0.2
Direct material cost 0.25 0.75
Direct labor cost 2.75 1.25
sales price/unit 4.0 5.0

10
 Assumption:

• Production limited machine capacity 8 hr/period

• assembly capacity 4 hr/period

• Raw material is limited to 72 unites

• cost involve only material and labour

• Amount of pollution emissions is 3 units per unit of product and 2 units

per unit of product 2.

(i) Maximization of Net Profit of investment Z1 (X) and

(ii) Minimizing the Pollution Z2 (X).
Let the mathematical form of the problem be:

Z1 (X) = X1 + 3X2
Z2 (X) = 3X1 + 2X2
Subject to
0.5X1 + 0.25X2 ≤ 8
0.2X1 + 0.2X2 ≤ 4
X1 + X2 ≤ 72 X1 , X2 ≥ 0

Proof
In this example, involving only two decision variables and two objective func-
tions, it is possible to solve and to interpret the CP problem graphically.

11
 Table -1:
Solution in the Illustrative Example
Extreme Point Z1 (X) Z2 (X)
0
X 1 =(0,0) 0 0 A =(0,0)
0
X =(0, 72
5
) 43.2 28.8 B =(43.2,28.8)
0
X 3 =(7,13) 46 47 C =(46,47)
0
X 4 =(20,0) 20 60 D =(20,60)

The values achieved for the objectives Z1 and Z2 at the extreme points in the

decision variables space generate new extreme points in the objectives space.
The four extreme points of our example in the objective space are plotted in
Figure 2. It should be noted that the points in the objective space are algebraic
images of the points in the decision variable space. Thus, point X 1 maps point
0
A of Figure 2 through the values of Z1 and Z2 generated by point X 1 . Points
0 0 0 0
A , B , C , and D of Figure 2 are connected by straight lines.

12
 Figure 1: The image of the feasible set in the objectives space

On examining Figure 2, it is easy to deduce that the segments connected,

0 0 0
A ,B and C , represent the efficient set in the objective space for the problem
0
being analysed. In fact, the points of F that do not lie on the boundary
formed by A’, B’ and C’ are inferior or non- efficient because they offer less
Profit and equal (or more) Pollution for producing than any point belonging to
the boundary itself. In the above example, therefore, the efficient set is given
by the boundary X 1 X 2 X 3 in the decision variables space or by the boundary
0 0 0
A B C , in the objectives space.

13
 The pay-off matrix in MOP

One way of obtaining the initial and useful information of an MOP problem
is to optimise each of the objectives separately over the efficient set and then
to compute the value of each objective for each of the optimal solutions. This
procedure generates a square matrix, called the ’pay-off matrix, as shown in
Table 2 for the two objectives of our problem. The elements of the first row
mean that the maximum Profit(46) corresponds to 47 ppm pollution. The
elements of the second row mean that the minimum Pollution (0 ppm) corre-
sponds to a Profit 0.

Profit Pollution(ppm)

Profit 46 47
Pollution 0 0

The degree of conflict between the two objectives can be investigated from the
above pay-off matrix; thus in our example there is a clear conflict between the
objectives Profit and Pollution.
In the literature on MOP, the elements of the main diagonal in the pay-off
matrix are referred to as the ’ideal point’; that is, the solution where all the
objectives achieve their optimum value. In our example the’ideal point’ con-
tains 46 of Profit and 0 ppm Pollution. When the objectives are in conflict, as
is the case in our example, the ’ideal point’ is infeasible.
When we take the worst element- the maximum element of the objective is
minimised, or the minimum element of the objective is maximised - from each
row of the pay-off matrix, then we have what is called the ’anti-ideal’ or ’nadir
point’. This is the situation where all the objectives achieve their worst val-

14
ues. In our example the ’anti-ideal’ point is 0 of Profit and 47 ppm Pollution.
Although the ’anti-ideal’ point is non-efficient, it has an important use in
normalising objective functions measured in different units and with different
absolute values. Moreover, the difference between the ideal and the anti-ideal
values define a range of values for each objective function, which is useful to
operationalise the constraint method.
Table -2:
Compromise Programming(discrete approximation)
0 0 0
A B C Z∗j (X) Z∗j (X)
Maximum 0 43.2 46 46 0
Profit
Minimum 0 28.8 47 0 47
Pollution
d1 1 0.06 0
d2 0 0.61 1
L 1 W1 = 1 1 0.67 1
L 2 W2 = 1 1 0.6129 1
L∞ 1 0.61 1
L 1 W1 = 2 2 1.34 2
L 2 W2 = 2 2 1.2258 2
L∞ 2 1.22 2

15
 chapter 5

Compromise programming - a continuous setting

The discrete version of compromise programming (CP) is a useful device to

rank a finite set of alternatives. It has, however, two possible weaknesses.
First, such an application of CP requires that the efficient set (or at least
the subset of extreme efficient points) has been determined already. Second,
the best-compromise point should always be an extreme efficient point, which
is not always the case as in many instances the best-compromises could be
Pareto-interior points. Hence, in the discrete approximation a considerable
amount of information is missed; therefore, in our example the interior points
0 0
belonging to the efficient segment A B (see Figure 2) cannot be considered
as compromises easily, particularly when some of these points could have been
the best-compromise for a given metric and a set of weights indicating the
preferences of the DM.
Both of these problems are avoidable if CP is used in a continuous setting.
An added bonus is that the best-compromise solutions are obtained straight-
forwardly from conventional LP models. Thus, for the L1 metric (i.e. p = 1)
expression (5) gives the minimisation of the relative deviation with respect to
the ideal. But, as Zj∗ ≥Zj (X) for every j, because Zj∗ is a component of the
ideal vector, then the absolute value signs of (5) can be dropped and, therefore,
for the Ll metric the best-compromise or the closest solution to the ideal point
can be obtained by solving the following LP problem:

m
X |Zj∗ − Zj (X)|
M inL1 (W ) = Wj (7)
j=1
|Zj∗ − Z∗j |

16
 Subject to
0.5X1 + 0.25X2 ≤ 8

0.2X1 + 0.2X2 ≤ 4
X1 + 5X2 ≤ 72
X1 , X2 ≥ 0

Applying the model (7) to the data of our example the best compromise
solution for the L1 metric is obtained by solving the following LP problem:

Min L1 = W1 |46−Z 1 (X)|

|46−0| + W2 |0−Z 2 (X)|
|0−47|
Subject to

x∈ f

where f is the feasible set.The optimum solution of the above LP problem for
W1 = W2 (that is, when the objectives are equally important), is given by
0 0
the point B ; therefore, the point B is the best- compromise solution and this
means that B’ is the efficient point closest to the ideal point when the metric
L1 is used.
For the L∞ metric (p = ∞), the maximum deviation from among the individual
deviations is minimised. That is, when p = ∞ only the largest deviation
counts. For this metric, the best-compromise solution is obtained by solving
the following LP problem:
∗
First, X∞ = a∗1 X 1 + a∗2 X 2 + a∗3 X 3 + ... + a∗m X m
Pm
Where X j , j = 1, 2, ..., m are non-dominated extreme points and j=1 a∗j = 1

Hence,the compromise solution convex combination of non-dominated extreme

points.

17
 The a∗j are defined

a
a∗j = Pm j
j=1 aj

Where the aj are the optimal solution to the following problem:

Pm
j=1 aj

subject to

∗ j
Pm |Zi −Zi (X )|
j=1 Wi aj |Z ∗ −Z∗i | ≤ 1
i

j = 1, 2, 3, ..., m a1 , a2 , a3 , ..., am ≥ 0 i=1,2.

Table -3:
Non-dominated Solution in the Illustrative Example
|Z1∗ −Z1 (X j )| |Z2∗ −Z2 (X j )|
Non-dominated |Z1∗ −Z∗1 | |Z2∗ −Z∗2 |
Σ
Extreme Point
X 1 = (0, 0) 1 0 1
X 2 = (0, 72
5
) 0.06 0.61 0.67
X 3 = (7, 13) 0 1 1

• For the example at hand,the system assumes the following form:

M ax a1 + a2 + a3

subject to

a1 + 0.06a2 ≤ 1

18
 0.61a2 + a3 ≤ 1

a1 , a2 , a3 ≥ 0

Solution of the above problem yields

a1 = 0.9016

a2 = 1.639

a3 = 0

then

X
aj = 2.54
j

and

a1
a∗1 = Σaj = 0.355

a2
a∗2 = Σaj = 0.645

a3
a∗3 = Σaj =0

Therefore,the compromise solution for p → ∞ is equal to

∗
X∞ = 0.355X 1 + 0.645X 2 + 0X 3

∗
X∞ = (0, 9.29)

19
 0
• For W1 = W2 = 1 solution is the point Z

– X1 = 0 and X2 = 9.29

∗ Maximum Profit = 27.9

∗ Minimum Pollution = 18.6

0
∗ For W1 = W2 = 1 solution is the point Z .

Figure 2: The ideal point and compromise set

20
For different sets of values for the weights W1 and W2 the structure of the
compromise sets can be modified. A sensitivity analysis with the weights can
furnish the DM with worthwhile information related to the stability of the
solution and the range within which the compromise sets can be defined.

21
 chapter 6

The method of the displaced ideal

Taking the preceding fact into account, it is inconvenient to work with compro-
mise sets that range over a big portion of the efficient set. In fact, with a large
compromise set it is not easy for the DM to choose the optimum solutions from
the efficient compromises; and also too large a neighbourhood of the tangency
point provides little interesting information. It is therefore worthwhile to ex-
plore operational methods that allow us to reduce the size of the compromise
sets. Zeleny (1973) has suggested one such method called the displaced ideal
which helps to reduce the compromise set to a manageable size.Let us now ex-
amine the basic ideas that underlie this method.Assume that the compromise
set Z’B’ is considered to be too large; that is, to state that the tangency be-
tween utility function and the efficient set will presumably occur at some point
of this segment may be regarded as a piece of information which is fuzzy and
hardly valuable for the DM. The situation can however be redeemed if the size
of the compromise set is reduced. The DM is asked to discard those portions
of the efficient set that do not interest him at all. For instance, assume that
the DM decides to discard efficient solutions belonging to the segment B 0 C 0
because they require too many pollution (more than 47 ppm), and also the so-
lutions along A0 Z 0 as they provide too little profit (less than s 27.9).Discarding
these efficient points implies obtaining a new ideal point. In other words, the
ideal point is displaced. Thus, in our example the discarding of that part of
the efficient set gives the new ideal point Z ∗∗ (43.2,18.6) represented in Figure
3. Obviously, the displacements of the ideal and the anti-ideal will produce a
new compromise set. As it was explained in the preceding section, the bounds

22
of the compromise set are established by solving two LP problems. For the L1
metric the LP problem to be solved now is the following one:
Min L1 = W1 |43.2−Z 1 (X)|
|43.2−27.9|
+ W2 |28.8−Z 2 (X)|
|28.8−18.6|

subject to
x∈f
0
W1 = W2 the L1 bound B of the compromise set is obtained again; that is,
the displacement of the ideal has not produced any displacement in the L1
bound of the compromise set. The L∞ bound of the new compromise set is
now obtained by solving the following problem:
Pm
j=1 aj

subject to

Pm |Z ∗∗j −Zj (X)|

j=1 Wi aj |Z ∗ ∗j −Z∗∗j | ≤d=1

∗
We have only two non dominated extreme point,X 2 and X∞ . Solving the
above LP problem for W1 = W2 , the new L= bound of the compromise set
00
is obtained. That bound is represented in Figure 3 as point Z . That point
corresponds to a Profit value of 35.5 and Pollution 23.7 ppm.

23
Figure 3: The trajectory of the displacement of the ideal and the new
compromise set

0
Displacing the ideal has produced the segment Z ” B , which is a smaller
and much more manageable compromise set. It reduces the landing area for
the utility function considerably and thus it is now much easier for the DM
0
to choose a best-compromise solution from the new compromise set Z ” B .
Obviously, this method can be implemented iteratively. If the size of the
compromise set obtained in the first iteration is not small enough, then the
DM is asked to discard new portions of the efficient set, producing a new

24
displacement of the ideal, continuing the iterative procedure until a satisfactory
compromise set is obtained. Thus, if in our example the DM discards efficient
0
solutions outside the new compromise set Z ” B , the new ideal Z ∗∗∗ in Figure
3 is obtained, which will lead to a new contraction of the compromise set.
Optimal solution X1 = 0 and X2 between 9.29 and 14.5

25
 chapter 7

Pros and cons of GP, MOP and CP

Having discussed the nature and functioning of the various MCDM techniques,
it is worth critically assessing the individual pros and cons of each method in
the light of their different attributes.
The first attribute to consider is the amount of computational time required in
using each technique. In this respect, GP is most efficient as it requires a single
’computer run’. If sensitivity analysis is carried out say for targets, weights,
order of priorities, etc. then, of course, the computing time increases corre-
spondingly. However, the MOP approach involves even more computation,
as with both the weighting and constraint method algorithms, the number
of’computer runs’ required is an exponential function of the number of objec-
tives involved. These demanding computing time requirements of MOP can
be reduced significantly if CP (in a continuous setting) is used instead; as it is
only necessary to solve two LP problems (for the metrics L1 and L∞ ) for each
set of weights; thus giving us a part of the efficient set as the compromise set to
work with. This however implies a possible weakness. We lose the information
on the trade-offs between objectives that was contained in those parts of the
efficient set that are excluded from the compromise set.
In terms of the quantity of information and the precision with which it is re-
quired from the DM, perhaps GP is the most difficult approach. The DM has
to provide precise target values, weights attached to each deviational variable,
pre-emptive ordering of preferences, etc. Much of this is very difficult to ob-
tain in some cases, as with the target values, it could be argued that this is
the kind of information that the model should provide to the DM rather than

26
it being an input requirement. The MOP approach is at the opposite end of
the scale as regards the information needed from the DM. To build an MOP
model, it is not necessary to know anything about the DM’s preferences. The
mathematical expression of the objectives being considered is sufficient. For
CP we need to know only the relative preferences of the DM for each objective
(that is,the weights attached to the discrepancies between each objective and
their ideal values) in order to approximate the compromise set for the different
metrics. As regards the information produced by the model for use by the DM,
GP is clearly inferior to the other two techniques. In fact, the different GP
variants provide only a single solution; that is, the set of decision variables
that permits the closest satisfaction of the different goals. Athough a sensitiv-
ity analysis subsequent to obtaining the optimal solution may be sufficient in
many instances, the fact remains that GP provides rather meagre information
compared to either MOP or CP.

27
 chapter 8

Conclusion

• As Ignizio (1983) says: “there is not now, and probably never shall be,
one single ”best“ approach to all types of multi- objective mathematical
programming problems”.

• No definite superiority of one MCDM approach relative to others.

• Success of MCDM technique depends on DMs on formation about targets

and priorities.

• DMs are required to make detailed analyses of various solu- tions.

28
References

[1] Carlos Romero,Tahir Rehman. Multiple Criteria Analysis for Agriculture

Decision. Elsevier Science B.V, 2003.

[2] D.E. Bell, H. Raiffa, A. Tversky. Decision making: Descriptive norma-

tive and prescriptive Interactions. Cambridge University Press, Cambridge,
England (1988).

[3] E. Ballestero, C. Romero. Portfolio selection: A compromise programming

solution.Journal of Operational Research Society, 47 (1996).

[4] M. Amiri, M. Ekhtiari, and M. Yazdani. Nadir compromise programming: a

model for optimization of multi-objective portfolio problem.Expert Systems
with Applications,Iran(2011).

[5] J. L. Cochrane, M. Zeleny. Compromise Programming, in: Multiple Crite-

ria Decision Making. University of South Carolina Press, Columbia(1973).

29