Applied Mathematics and Computation 271 (2015) 489–501

Contents lists available at ScienceDirect

Applied Mathematics and Computation

journal homepage: www.elsevier.com/locate/amc

Computation of a multi-choice goal programming problem

Kanan K. Patro a,b,∗, M.M. Acharya a, M.P. Biswal c, Srikumar Acharya a
a
Department of Mathematics, School of Applied Sciences, KIIT University, Bhubaneswar, India
b
Department of Mathematics, Kendriya Vidyalaya, Bargarh, India
c
Department of Mathematics, Indian Institute of Technology, Kharagpur, India

a r t i c l e i n f o a b s t r a c t

Keywords: The standard goal programming problem allows decision maker to assign an aspiration level
Multi-criteria decision making to an objective function. In real life decision making problems, the decision maker always
Multi-choice goal programming
seeks for suitable aspiration level i.e. “the more suitable the better”. Therefore, a decision
Multiple aspiration levels
maker is allowed to assign multiple number of aspiration levels to an objective function. The
aim of the decision maker is to select an appropriate aspiration level for an objective func-
tion that minimizes the deviations between the achievement of goal and the aspiration levels.
The traditional goal programming techniques cannot be used for solving such type of multi-
choice goal programming problem. This paper presents an equivalent model of the multi-
choice goal programming problem by using Vandermonde’s interpolating polynomial, binary
variables and least square approximation method. The equivalent model is solved by exist-
ing method/software. Two illustrative examples are presented in support of the proposed
methodology.
© 2015 Elsevier Inc. All rights reserved.

1. Introduction

The term ‘decision’ carries different meanings, depending upon the nature of decision maker. A decision maker may be a
lawyer, a businessman, a psychologist or a statistical or a general person. It might be behavioral action, mathematical model, or
a specific kind of information processing. It is very difficult to represent decision making problems in mathematical models due
to conflicts of resources and incompleteness of available information. In realistic situations decision making problems require to
consider multiple objectives on one hand and various types of uncertainties on the other hand. There exists different methods to
handle different uncertainties.
Several techniques, namely utility function approach, goal programming approach, reference point method, interactive ap-
proach etc. exist for the solution of multi-criteria decision making problems. Among them the most popular is goal programming
(GP) approach. GP is an analytic approach devised to address decision making problems where targets have been assigned to all
the attributes. The decision maker (DM) is interested in minimizing the non-achievement of the goals. GP was first addressed by
Charnes and Cooper [1]. Since the mid 70s, due to the seminal works by Lee [2] and Ignizio [3] an impressive revolution of GP
applications and theoretical developments took place. Now-a-days, GP is the key-technique to work with multi-criteria decision
making problems. Lee [2] and Ignizio [4] wrote impressive books on GP. Tamiz et al. [5] provided an up to date review on GP. The

∗
Corresponding author at: Department of Mathematics, School of Applied Sciences, KIIT University, Bhubaneswar, India. Tel.: +919861414866.
E-mail addresses: kananpatro@gmail.com (K.K. Patro), mitali.me@gmail.com (M.M. Acharya), mpbiswal@maths.iitkgp.ernet.in (M.P. Biswal),
sacharyafma@kiit.ac.in, srikumarc@iitkgp.ac.in (S. Acharya).

http://dx.doi.org/10.1016/j.amc.2015.09.030
0096-3003/© 2015 Elsevier Inc. All rights reserved.
490 K.K. Patro et al. / Applied Mathematics and Computation 271 (2015) 489–501

contributions of Tamiz et al. [6], Romero et al. [7], Ijiri [8], Schniederjans [9], Zeleny and Cochrane [10] were remarkable for the
development of GP.
The distinction between various types of generalized goal programming, is made on the basis as how one actually measures
the “goodness” of any solution (the value of X) to the set of goals. This is a typical method facilitated by means of the concept of
“goal deviations” and the “achievement function”. By the philosophy of goal programming problem DM chooses a target value
and decides whether to penalize positive or negative deviations from the target. However, DMs are not interested in the specific
fixed deterministic targets associated with certain attributes in real-life.
DMs prefer flexibility and suitability. It is observed that DMs are interested in a set of deterministic targets associated with
an attribute. Keeping this in mind, Chang [11] proposed a new idea for programming the multi-choice aspiration level problem
and named it as Multi-Choice Goal Programming (MCGP) Problem. He introduced the multiplicative terms of binary variables
in order to tackle with multi-choice aspiration levels associated with each goal. The way he introduced the multiplicative terms
of binary variables is too difficult to implement and is not easily understood by industrial participants. In Chang [12], although
multiple aspiration levels are assigned to a goal, it is difficult to describe role of all aspiration levels in that goal. It replaces
multiplicative terms of the binary variables by taking the help of a continuous variable, with a range of interval values as the
lower and upper bound of each objective function. A new concept of constrained multi-choice goal programming is introduced
for constructing the relationships between goals. This paper presents a new approach to search the appropriate set of aspiration
levels from multiple sets of aspiration levels using multiplicative terms of binary variables.
Biswal and Acharya [13,14], Acharya and Biswal [15] used binary variables in order to transform a multi-choice linear pro-
gramming problem to an equivalent mathematical model. Using the concept of Chang [11], Liao [16] formulated multi-segment
goal programming. Acharya and Acharya [17] generalized the transformation technique proposed by Biswal and Acharya [14].
Biswal and Acharya [18] used interpolating polynomial approach to solve multi-choice linear programming problem. After using
the interpolation, the formulated mathematical model was a mixed integer nonlinear programming problem. Chang et al. [19]
used his own technique to select a suitable house for homebuyer. Ustun [20] used conic scalarization function for formulating
the MCGP. Fuzzy multi-choice goal programming problem was first addressed by Bankian-Tabrizi et al. [21]. Chang et al. [22]
used multi-coefficients goal programming for group pricing problem. Multi choice mixed integer goal programming problem
was carried out by Da Silva et al. [23]. They mainly focused on decisions on the choice of production process, including storage
stages and distribution.
The paper is organized as follows: Section 2 contains basic preliminaries followed by “mathematical model” in Section 3. In
Section 4 equivalent models for the proposed MCGP are presented. Two numerical examples are provided in Section 5 to justify
the methodology. In Section 6, results and discussions are presented. Finally concluding remarks are made following supporting
references.

2. Basic preliminaries

The aim of the goal programming is to minimize the deviations between the achievement of goals and their aspiration levels.
According to the philosophy of satisficing we are interested in measuring the non-achievement of each goal. This is the unwanted
deviations from the aspiration levels (i. e. the value of each goal ‘g’). We let

di = the deviation between ‘the aspiration level’ and ‘the acheivement o f goal’.
‘or’di = gi − fi (X ).
Where gi is the aspiration level and fi (X) is the objective, which is to be achieved.
Hence, we can express the general goal programming as:

min : | fi (X ) − gi | for i = 1, 2, . . . , m
subject to
X ∈R
where R is a feasible set.
The three oldest and still most widely used forms of GP are used to minimize the unwanted deviations. The methods are as
follows:

1. Lexicographic Goal Programming (LGP), also known as non-Archimedean or preemptive GP.

2. Weighted GP (WGP), also known as Archimedean GP.
3. Min–Max GP (MGP) also known as Chebyshev or Fuzzy Programming.

The mathematical formulations for LGP is expressed as follows:

lexicographically min : ā = (a1 , a2 , a3 , . . . , ak , . . . , aK )

subject to
fi (X ) + ηi − βi = bi ∀i
X, η̄, ρ̄ ≥ 0
 K.K. Patro et al. / Applied Mathematics and Computation 271 (2015) 489–501 491

where ak = gk (η̄, ρ̄) and ā is the achievement vector for which we seek the lexicographic minimum; k is the ranking or priority
level for k=1, 2, . . . , K; gk (η̄, ρ̄) is the usual linear function of goal in deviational variables that are to be minimized at priority
level k.
The mathematical formulation of WGP might be expressed as:

min : a = (αi ηi + βi ρi )
i∈G
subject to
fi (X ) + ηi − ρi = gi for i ∈ G
X ∈R
X, η̄, ρ̄ ≥ 0
Where ‘R’ is the feasible set and a is the measure of achievement (actually, of the non-achievement) of given flexible goals in G;
α i is the weight associated with the negative deviational variable (ηi ) of ith goal; β i is the weight associated with the positive
deviational variable (ρ i ) of ith goal. The weighted goal programming problem approach leads to a single linear goal programming
model which may be solved via conventional (single objective) mathematical programming methods.
Finally, let us consider the case of min–max philosophy, which leads to MGP. For convenience we first convert all goals (i. e.
f (X ) ≤ g, f (X ) ≥ g, or f (X ) = g) into the Type-1 form (i. e. f(X) ≤ g). The model and achievement function may then be written
as:
min : λ
subject to
( fi (X ) − gi ) − λ ≤ 0 ∀ i
X ∈R
X, λ ≥ 0
Here we have a feasible set ‘R’. Where λ is the maximum deviation from any goal. This problem is again in the form of conven-
tional, single objective model and thus any appropriate single objective solution methodology can be used.

3. Mathematical model

We present a general multi-choice goal programming model as:

 
(Rk )
min : |Zk (X ) − gk |, gk ∈ g(k1) , g(k2) , g(k3) , . . . , gk , k = 1, 2, 3, . . . , K (3.1)

subject to

n
ai j x j ≤ bi , i = 1, 2, . . . , m (3.2)
j=1

x j ≥ 0, j = 1, 2, . . . , n. (3.3)
The first, second, third etc. elements of the set represent the aspiration levels for first, second, third etc. objective functions
respectively.
The goal programming model given in (3.1)–(3.3) can be expressed as:

K
min : ωk (ρk + ηk ) (3.4)
k=1

subject to
 
(Rk )
Zk (X ) + ηk − ρk = gk , gk ∈ g(k1) , g(k2) , g(k3) , . . . , gk , k = 1, 2, 3, . . . , K (3.5)


n
ai j x j ≤ bi , i = 1, 2, . . . , m (3.6)
j=1


K
wk = 1 (3.7)
k=1

ρk , ηk , x j ≥ 0, j = 1, 2, . . . , n; k = 1, 2, 3, . . . , K (3.8)
ωk , ρ k , ηk represent the weights, positive and negative deviations associated with kth goal respectively.
492 K.K. Patro et al. / Applied Mathematics and Computation 271 (2015) 489–501

4. Equivalent models

In this section, three different approaches are discussed.

4.1. Interpolating polynomial approach

A polynomial that passes through a given set of points is called an interpolating polynomial. There are ‘n’ points in a plane
(i, yi ), i = 0, 1, 2, 3 . . . , n − 1 with distinct i then there exists a unique polynomial in x whose degree is n − 1.
Let the polynomial be
PRk −1 (zi ) = a0 + a1 zi + a2 (zi )2 + a3 (zi )3 + · · · + an−2 (zi )n−2 + an−1 (zi )n−1 (4.9)
after substituting the n points in the polynomial we have,
P (0) = a0 + a1 (0) + a2 (0)2 + a3 (0)3 + . . . + an−2 (0)n−2 + an−1 (0)n−1 = y0 (4.10)

P (1) = a0 + a1 (1) + a2 (1)2 + a3 (1)3 + . . . + an−2 (1)n−2 + an−1 (1)n−1 = y1 (4.11)

P (2) = a0 + a1 (2) + a2 (2)2 + a3 (2)3 + . . . + an−2 (2)n−2 + an−1 (2)n−1 = y2 (4.12)

..
. (4.13)

P (n − 1) = a0 + a1 (n − 1) + a2 (n − 1)2 + a3 (n − 1)3 + . . . + an−1 (n − 1)n−1 = yn−1 (4.14)

The system of equations can be written as AX = B where
⎛ ⎞ ⎛a ⎞ ⎛y ⎞
1 0 0 ··· 0 0 0 0
⎜1 1 12 ··· 1n−2 1n−1 ⎟ ⎜a1 ⎟ ⎜ y1 ⎟
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎜1 ···
⎟ ⎜ ⎟ ⎜ ⎟
A=⎜ ⎟
2 n−2 n−1
⎜
2 2 2 2
⎟ X = ⎜a2 ⎟ B = ⎜ y2 ⎟ (4.15)
⎜.. .. .. .. .. .. ⎟ ⎜. ⎟ ⎜. ⎟
⎝. . . . . . ⎠ ⎝.. ⎠ ⎝.. ⎠
1 (n − 1) (n − 1) 2
··· (n − 1)n−2 (n − 1)n−1 an−1 yn−1
The value of a1 , a2 , . . . , an can be found out by solving AX = B. If A is non singular then the system can be solved.
Now, we formulate a multi-choice goal programming model for the multi-objective linear programming problem by using
Vandermonde’s interpolating polynomial as:


K
min : ωk (ρk + ηk ) (4.16)
k=1

subject to
Zk (X ) + ηk − ρk = PRk −1 (z), k = 1, 2, 3, . . . , K (4.17)


n
ai j x j ≤ bi , i = 1, 2, . . . , m (4.18)
j=1

z is an integer and 0 ≤ z ≤ Rk − 1 (4.19)

ρk , ηk , x j , z ≥ 0, j = 1, 2, . . . , n; k = 1, 2, 3, . . . , K (4.20)

4.2. Binary variable approach

Theorem 1. Every natural number can be expressed as sum of 2k number of terms and each term is a power of 2, where k ∈ N ∪ {0}.

The proof of the above theorem is obvious.

Now, we formulate a multi-choice goal programming model for the multi-objective linear programming problem by using
binary variables as:


K
min : ωk (ρk + ηk ) (4.21)
k=1
 K.K. Patro et al. / Applied Mathematics and Computation 271 (2015) 489–501 493

Table 1
Nodes.

x 0 1 2 … n−1

f(x) f0 f1 f2 … fn−1

subject to
 
Zk (X ) + ηk − ρk = Pk θk(1) , θk(2) , k = 1, 2, 3, . . . , K (4.22)


n
ai j x j ≤ bi , i = 1, 2, . . . , m (4.23)
j=1

θk(1) , θk(2) ∈ {0, 1} (4.24)

ρk , ηk , x j , z ≥ 0, j = 1, 2, . . . , n; k = 1, 2, 3, . . . , K (4.25)

4.3. Linear least square approximation approach

Let a function f(x) be given by the following Table 1 at a discrete set of points i; i = 0(1)n − 1.
By using the set of points we want to find out a least square line such that the sum of the square of the vertical distance of the
points (i, fi ), i = 0(1)n − 1 from the line is the minimum. In this case the degree of the least square polynomial is 1 and

li (z) = a0 + a1 z (4.26)

the least square error E is given by


n−1
E= ( fi − a0 − a1 i)2 (4.27)
i=0

The necessary conditions for E to be minimum are

 

n−1 
n−1 
n−1
2
a0 i + a1 i = ifi (4.28)
i=0 i=0 i=0



n−1 
n−1
a0 n + a1 i = fi (4.29)
i=0 i=0

which is a system of two linear equations in a0 , a1 and are the normal equations for the least square linear polynomial. By solving
the normal equation given by (4.28) and (4.29), we get the two coefficients a0 and a1 and the linear least square polynomial
li (z) = a0 + a1 z.
Now, we formulate a multi-choice goal programming model for the multi-objective linear programming problem by using
linear least square approximation approach as:


K
min : ωk (ρk + ηk ) (4.30)
k=1

subject to

Zk (X ) + ηk − ρk = lk (z), k = 1, 2, 3, . . . , K (4.31)


n
ai j x j ≤ bi , i = 1, 2, . . . , m (4.32)
j=1

z is an integer and 0 ≤ z ≤ Rk − 1 (4.33)

ρk , ηk , x j , z ≥ 0, j = 1, 2, . . . , n; k = 1, 2, 3, . . . , K (4.34)
494 K.K. Patro et al. / Applied Mathematics and Computation 271 (2015) 489–501

Table 2
Investment opportunities (Eatman and Sealey Jr [24]).

Investment category Rate of return (%) Liquid part (%) Required capital (%) Risk asset

1. Cash 0.0 100.0 0.0 No

2. Short term 4.0 99.5 0.5 No
3. Government: over 1 to 5 years 4.5 96 4.0 No
4. Government: over 1 to 5 years 5.5 90 5.0 No
5. Government: over 10 years 7.0 85.0 7.5 No
6. Installment loans 10.5 0.0 10.0 Yes
7. Mortgage loans 8.5 0.0 10.0 Yes
8. Commercial loans 9.2 0.0 10.0 Yes

5. Numerical example

Example 1. Consider a weighted goal programming problem which was proposed by Eatman and Sealey Jr [24]. Bank Three has
a modest $20 million capital, with $150 million in demand deposits(checking accounts) and $80 million in time deposits (saving
accounts and certificates of deposit). Table 2 displays the categories among which the Bank must divide its capital and deposited
funds. Rates of return are also provided for each category together with other information related to risk.

We model Bank Three’s investment decisions with a decision variable A, B, C, D, E, F, G and H for each category of investment
in the given table.
The first goal of any private business is to maximize profit. Using rates of return from the Table 2, we have the objective
function
max: Z1 = 0.04B + 0.045C + 0.055D + 0.07E + 0.105F + 0.085G + 0.092H
It is less clear how to quantify investment risk. We employ two common ratio measures. One is capital-adequacy ratio, expressed
as the ratio of required capital for bank solvency to actual capital. A low value indicates minimum risk. Bank Three’s present
capital is $20 million. Thus, we express a second objective as
min: Z2 = 0.05(0.005B + 0.04C + 0.05D + 0.075E + 0.1F + 0.1G + 0.1H )
Another measure of risk focuses on liquid risk assets. A low risk asset/capital ratio indicates a financially secure institution. For
our example, this third measure of success is expressed as
min Z3 = 0.05(F + G + H )
To complete the model of Bank Three’s investment plans, we must describe the relevant constraints. Our example will assume
five types:

1. Investments must sum to the available capital and deposit funds.

2. Cash reserves must be at least 14% of demand deposit plus 4% of time deposits.
3. The portion of investments considered liquid should be at least 47% of demand deposit plus 36% of time deposits.
4. At least 5% of funds should be invested in each of the eight categories, for diversity.
5. At least 30% of funds should be invested in commercial loans, to maintain the bank’s community status.

Combining the three objective functions above with these five systems of constraints gives rise to multi-objective linear pro-
gramming model
max : Z1 = 0.04B + 0.045C + 0.055D + 0.07E + 0.105F + 0.085G + 0.092H (5.35)

min : Z2 = 0.05(0.005B + 0.04C + 0.05D + 0.075E + 0.1F + 0.1G + 0.1H ) (5.36)

min : Z3 = 0.05(F + G + H ) (5.37)

subject to
A + B + C + D + E + F + G + H = 250 (5.38)

A ≥ 24.2 (5.39)

1.0A + 0.995B + 0.96C + 0.9D + 0.85E ≥ 99.3 (5.40)

A, B, C, D, E, F, G ≥ 12.5 (5.41)

H ≥ 75 (5.42)
 K.K. Patro et al. / Applied Mathematics and Computation 271 (2015) 489–501 495

The above multi-objective programming model can be written in goal format as follows:
min : |0.04B + 0.045C + 0.055D + 0.07E + 0.105F + 0.085G + 0.092H − g1 |,
g1 ∈ {17, 20, 22, 24, 27} (5.43)

min : |0.05(0.005B + 0.04C + 0.05D + 0.075E + 0.1F + 0.1G + 0.1H ) − g2 |,

g2 ∈ {0.6, 0.7, 0.8, 0.9} (5.44)

min : |0.05(F + G + H ) − g3 |, g3 ∈ {5, 6, 7} (5.45)

subject to
A + B + C + D + E + F + G + H = 250 (5.46)

A ≥ 24.2 (5.47)

1.0A + 0.995B + 0.96C + 0.9D + 0.85E ≥ 99.3 (5.48)

A, B, C, D, E, F, G ≥ 12.5 (5.49)

H ≥ 75 (5.50)

5.1. Solution by binary variable approach

By using the concept of multi-choice goal programming, it can be represented as:

min : η1 + ρ2 + ρ3 (5.51)
subject to
0.04B + 0.045C + 0.055D + 0.07E + 0.105F + 0.085G + 0.092H + η1 − ρ1
= 17θ11 (1 − θ12 )θ13 + 20(1 − θ11 )θ12 θ13 + 22θ11 (1 − θ12 )(1 − θ13 ) + 24(1 − θ11 )θ12 (1 − θ13 )
+ 27(1 − θ11 )(1 − θ12 )θ13 (5.52)

0.05(0.005B + 0.04C + 0.05D + 0.075E + 0.1F + 0.1G + 0.1H ) + η2 − ρ2

= 0.6θ21 θ22 + 0.7θ21 (1 − θ22 ) + 0.8(1 − θ21 )θ22 + 0.9(1 − θ21 )(1 − θ22 ) (5.53)

0.05(F + G + H ) + η3 − ρ3 = 5θ31 θ32 + 6θ31 (1 − θ32 ) + 7(1 − θ31 )θ32 (5.54)

A + B + C + D + E + F + G + H = 250 (5.55)

A ≥ 24.2 (5.56)

1.0A + 0.995B + 0.96C + 0.9D + 0.85E ≥ 99.3 (5.57)

A, B, C, D, E, F, G ≥ 12.5 (5.58)

H ≥ 75 (5.59)

θ11 + θ12 + θ13 ≥ 1 (5.60)

θ11 + θ12 + θ13 ≤ 2 (5.61)

θ11 + θ12 ≤ 1 (5.62)

θ31 + θ32 ≥ 1 (5.63)

A, B, C, D, E, F, G, H ≥ 0 (5.64)
496 K.K. Patro et al. / Applied Mathematics and Computation 271 (2015) 489–501

ρi , ηi ≥ 0; i = 1, 2, 3 (5.65)

θ11 , θ12 , θ13 ∈ {0, 1} (5.66)

θ21 , θ22 ∈ {0, 1} (5.67)

θ31 , θ32 ∈ {0, 1} (5.68)

The above mathematical model (5.51)–(5.68) is solved with the help of LINGO 11.0 [25] software to obtain the optimal solution
as (A, B, C, D, E, F, G, H) = (39.7, 12.5, 12.5, 12.5, 32.8, 12.5, 12.5, 115), (θ 11 , θ 12 , θ 13 , g1 ) = (1.0, 0.0, 1.0, 17), (θ 21 , θ 22 , g2 ) = (0.0, 0.0,
0.9), (θ 31 , θ 32 , g3 ; Z1 , Z2 , Z3 ) = (0.0, 1.0; 17.0, 0.88, 7.0).

5.2. Solution by Vandermonde’s interpolating polynomial approach

By using the concept of multi-choice goal programming, it can be represented as:

min : η1 + ρ2 + ρ3 (5.69)

subject to

0.04B + 0.045C + 0.055D + 0.07E + 0.105F + 0.085G + 0.092H + η1 − ρ1

= 17 + 3.83z1 − z12 + 0.17z13 (5.70)

0.05(0.005B + 0.04C + 0.05D + 0.075E + 0.1F + 0.1G + 0.1H ) + η2 − ρ2

= 0.6 + 0.1z2 (5.71)

0.05(F + G + H ) + η3 − ρ3 = 5 + z3 (5.72)

A, B, C, D, E, F, G, H ≥ 0 (5.73)

A + B + C + D + E + F + G + H = 250 (5.74)

A ≥ 24.2 (5.75)

1.0A + 0.995B + 0.96C + 0.9D + 0.85E ≥ 99.3 (5.76)

A, B, C, D, E, F, G ≥ 12.5 (5.77)

H ≥ 75 (5.78)

ρi , ηi ≥ 0, i = 1, 2, 3 (5.79)

0 ≤ z1 ≤ 4 (5.80)

0 ≤ z2 ≤ 3 (5.81)

0 ≤ z3 ≤ 2. (5.82)

zi ∈ Z, i = 1, 2, 3 (5.83)

The above mathematical model (5.69)–(5.83) is solved with the help of LINGO 11.0 [25] software to obtain the solution as (A, B,
C, D, E, F, G, H) = (39.89, 12.5, 12.5, 12.5, 32.6, 13.46, 12.5, 114.03); (z1 , g1 ) = (0.0, 17); (z2 , g2 ) = (3, 0.9); (z3 , g3 ; Z1 , Z2 , Z3 ) = (2.0,
7.0; 17.0, 0.88, 7.0).
 K.K. Patro et al. / Applied Mathematics and Computation 271 (2015) 489–501 497

5.3. Solution by linear least square approximation approach

By using the concept of multi-choice goal programming, it can be represented as:

min : η1 + ρ2 + ρ3 (5.84)
subject to
0.04B + 0.045C + 0.055D + 0.07E + 0.105F + 0.085G + 0.092H + η1 − ρ1
= 17.2 + 2.4z1 (5.85)

0.05(0.005B + 0.04C + 0.05D + 0.075E + 0.1F + 0.1G + 0.1H ) + η2 − ρ2

= −1.5 + 1.5z2 (5.86)

0.05(F + G + H ) + η3 − ρ3 = 5 + z3 (5.87)

5A + 7B + 8C + η3 − ρ3 = 56.8 + 5.4z2 (5.88)

+ −
56.8 + 5.4z2 + d21 − d22 =0 (5.89)

A, B, C ≥ 0 (5.90)

A + B + C + D + E + F + G + H = 250 (5.91)

A ≥ 24.2 (5.92)

1.0A + 0.995B + 0.96C + 0.9D + 0.85E ≥ 99.3 (5.93)

A, B, C, D, E, F, G ≥ 12.5 (5.94)

H ≥ 75 (5.95)

ρi , ηi ≥ 0, i = 1, 2, 3 (5.96)

0 ≤ z1 ≤ 2 (5.97)

0 ≤ z2 ≤ 4; z1 , z2 ∈ Z (5.98)
The above mathematical model (5.84)–(5.98) is solved with the help of LINGO 11.0 [25] software to obtain the solution as (A, B,
C, D, E, F, G, H) = (39.89, 12.5, 12.5, 12.5, 32.6, 13.46, 12.5, 114.03); (z1 , g1 ) = (0.0, 17);(z2 , g2 ) = (3, 0.9); (z3 , g3 ; Z1 , Z2 , Z3 ) = (2.0,
7.0; 17.0, 0.88, 7.0).

Example 2. Let us consider a multi-choice goal programming problem [11] with the following goals and constraints, which
cannot be solved by the known GP techniques.
Goals : (G1)3A + 2B + C ≤ g1 , g1 ∈ {100, 120}
(G2)4A + 3B + 2C ≥ g2 , g2 ∈ {80, 100}
(G3)3.5A + 5B + 3C ≥ g3 , g3 ∈ {70, 90, 110}
subject to
B + C ≥ 10; B ≥ 4; A + B + C ≥ 15.

5.4. Solution by binary variable approach

Based on the proposed binary variable approach, this problem can be formulated as:
min : ρ1 + η2 + η3 (5.99)
subject to
3A + 2B + C + ρ1 − η1 = 100θ41 + 120(1 − θ41 ) (5.100)
498 K.K. Patro et al. / Applied Mathematics and Computation 271 (2015) 489–501

4A + 3B + 2C + ρ2 − η2 = 80θ51 + 100(1 − θ51 ) (5.101)

3.5A + 5B + 3C + ρ3 − η3 = 70θ61 θ62 + 90θ61 (1 − θ62 ) + 110(1 − θ61 )θ62 (5.102)

A, B, C ≥ 0 (5.103)

B + C ≥ 10 (5.104)

B≥4 (5.105)

A + B + C ≥ 15 (5.106)

θ41 ≤ 1 (5.107)

θ51 ≤ 1 (5.108)

ρi , ηi ≥ 0; i = 1, 2, 3 (5.109)

θ61 + θ62 ≥ 1 (5.110)

θ61 + θ62 ≤ 2 (5.111)

This problem is solved by using LINGO 11.0 [25] to obtain the optimal solution as (A, B, C, θ 41 , θ 51 , θ 61 , θ 62 ) = (17.55̄, 7.77̄,
3.22̄, 1, 0, 0, 1). From the results we get that goal G1 has achieved 71.44 and reached the aspiration level 100, whereas goals G2
and G3 reached the aspiration level 100 and 110 respectively.

5.5. Solution by Vandermonde’s interpolating polynomial approach

Based on the proposed Vandermonde’s interpolating polynomial approach, this problem can be formulated as:
min : ρ1 + η2 + η3 (5.112)
subject to
3A + 2B + C + ρ1 − η1 = 100 + 20z1 (5.113)

4A + 3B + 2C + ρ2 − η2 = 80 + 20z2 (5.114)

3.5A + 5B + 3C + ρ3 − η3 = 70 + 20z3 (5.115)

A, B, C ≥ 0 (5.116)

B + C ≥ 10 (5.117)

B≥4 (5.118)

A + B + C ≥ 15 (5.119)

zi ∈ Z, i = 1, 2, 3 (5.120)

0 ≤ z1 ≤ 1 (5.121)

0 ≤ z2 ≤ 1 (5.122)

0 ≤ z3 ≤ 2 (5.123)

ρi , ηi ≥ 0; i = 1, 2, 3 (5.124)
This problem is solved by using LINGO 11.0 [25] to obtain the optimal solution as (A, B, C, z1 , z2 , z3 ) = (17.77̄, 8.88̄, 1.11̄, 0, 1, 2).
From the results we realized that goal G1 has reached 72.22 and achieved the aspiration level 100, goals G2 and G3 reached the
aspiration level respectively 100 and 110.
 K.K. Patro et al. / Applied Mathematics and Computation 271 (2015) 489–501 499

5.6. Solution by linear least square approximation approach

Based on the proposed linear least square approximation approach, this problem can be formulated as:

min : ρ1 + η2 + η3 (5.125)

subject to

3A + 2B + C + ρ1 − η1 = 100 + 20z1 (5.126)

4A + 3B + 2C + ρ2 − η2 = 80 + 20z2 (5.127)

3.5A + 5B + 3C + ρ3 − η3 = 70 + 20z3 (5.128)

A, B, C ≥ 0 (5.129)

B + C ≥ 10 (5.130)

B≥4 (5.131)

A + B + C ≥ 15 (5.132)

zi ∈ Z, i = 1, 2, 3 (5.133)

0 ≤ z1 ≤ 1 (5.134)

0 ≤ z2 ≤ 1 (5.135)

0 ≤ z3 ≤ 2 (5.136)

ρi , ηi ≥ 0; i = 1, 2, 3 (5.137)

This problem is solved by using LINGO 11.0 [25] to obtain the optimal solution as (A, B, C, z1 , z2 , z3 ) = (17.77̄, 8.88̄, 1.11̄, 0, 1, 2).
From the results we realized that goal G1 has reached 72.22 and achieved the aspiration level 100, goals G2 and G3 reached the
aspiration level exactly 100 and 110 respectively.

6. Results, discussions and conclusions

This paper uses three approaches, namely: Binary variable approach, Vandermonde’s interpolating polynomial approach and
linear least square approximation approach. From the computational results, it is concluded that in all the three methods, goals
are the same. The objective values in all the three cases are approximately equal. In Chang [11] model, multiplicative terms of
binary variables were also used but repetitions occur during solution procedure, whereas, in our model the binary variable do
not allow any repetition. The use of binary variables in multi-choice goal programming model increases the complexity. The
complexity of solution also increases if the number of objective functions and the number of choice increases. Here we use two
sets of binary variable. The first set of binary variables searches a group which the appropriate choice belongs. Second set of
binary variables searches the appropriate parameter from the searched group. Comparison of results of three approaches for
Example 1 presented in Table 3.
Chang [11] model represents the time complexity in non-polynomial time(exponential/logarithmic) whereas Chang [12]
model maintain the time complexity in polynomial time. Biswal and Acharya [14] used binary variable and some auxiliary
constraints to deal with multi-choice parameters. But the proposed binary variable approach does not need any extra con-
straints. It is observed that our proposed Vandermonde’s interpolating polynomial approach has equal computational com-
plexity as that of Acharya et al. [26]. Although the proposed linear least square approximation approach gives approximate
choices of the multi-choice parameter, but computational complexity is equal to simple linear integer programming problem.
It can also be solved by linear integer programming method. The method suggested by Chang [12] also do the same thing but
unable to take care of all the multi-choice parameters. Here a comparative study of time complexity for 8, 9 and 10 number of
goals having 2, 3, 4, 5, and 8 choices of multiple aspiration levels is presented in Tables 4–6. It shows the proposed methods
perform better. An attempt is to be made to linearize binary variable approach and Vandermonde’s interpolating polynomial
approach.
500 K.K. Patro et al. / Applied Mathematics and Computation 271 (2015) 489–501

Table 3
Comparative analysis of three approaches for Example 1.

Investment category PM1 PM2 PM3

1. Cash(A) 39.7 39.89 17.7

2. Short term(B) 12.5 12.5 12.5
3. Government: over 1 to 5 years(C) 12.5 12.5 54.7
4. Government: over 5 to 10 years(D) 12.5 12.5 12.5
5. Government: over 10 years(E) 32.8 32.6 12.5
6. Installment loans(F) 12.5 13.46 12.5
7. Mortgage loans(G) 12.5 12.5 52.5
8. Commercial loans(H) 115 114.03 75.0
9. Profit(Z1 ) 17 17 17.2
10.Capital - adequacy ratio(Z2 ) 0.88 0.88 0.89
11.Risk - asset ratio(Z3 ) 7.0 7.0 7.0

PM1 : Binary variable approach

PM2 : Vandermonde’s interpolating polynomial approach
PM3 : Linear least square approach

Table 4
Relative performance of 8 number of Goals(CPU Time)(hh:mm:ss).

No. of A.L. Chang [11] Chang [12] Biswal and Acharya [14] Biswal and Acharya [18] PM1 PM2 PM3

2 00:00:01 00:00:01 00:00:01 00:00:01 00:00:01 00:00:01 00:00:01

3 00:00:06 00:00:01 00:00:04 00:00:04 00:00:02 00:00:04 00:00:01
4 00:00:05 00:00:01 00:00:05 00:00:05 00:00:03 00:00:05 00:00:01
5 00:03:45 00:00:01 00:00:15 00:00:15 00:00:10 00:00:15 00:00:01
8 00:03:31 00:00:01 00:00:12 00:00:12 00:00:07 00:00:12 00:00:01

Table 5
Relative performance of 9 number of Goals(CPU Time)(hh:mm:ss).

No. of A.L. Chang [11] Chang [12] Biswal and Acharya [14] Biswal and Acharya [18] PM1 PM2 PM3

2 00:00:01 00:00:01 00:00:01 00:00:01 00:00:01 00:00:01 00:00:01

3 00:00:07 00:00:01 00:00:05 00:00:05 00:00:03 00:00:05 00:00:01
4 00:00:06 00:00:01 00:00:06 00:00:06 00:00:01 00:00:06 00:00:01
5 01:55:37 00:00:01 00:07:38 00:07:38 00:06:43 00:07:38 00:00:01
8 01:43:34 00:00:01 00:06:31 00:06:31 00:00:01 00:06:31 00:00:01

Table 6
Relative performance of 10 number of Goals(CPU Time)(hh:mm:ss).

No. of A.L. Chang [11] Chang [12] Biswal and Acharya [14] Biswal and Acharya [18] PM1 PM2 PM3

2 00:00:01 00:00:01 00:00:01 00:00:01 00:00:01 00:00:01 00:00:01

3 00:00:07 00:00:01 00:00:05 00:00:05 00:00:03 00:00:05 00:00:01
4 00:00:06 00:00:01 00:00:05 00:00:05 00:00:03 00:00:05 00:00:01
5 15:53:25 00:00:01 01:02:14 01:02:14 57:03:12 01:02:14 00:00:01
8 15:13:05 00:00:01 00:58:45 00:58:45 55:01:35 00:58:45 00:00:01

Acknowledgments

The authors would like to thank the Editor-in-Chief, Associate Editor, and the reviewers for their helpful suggestions and
comments which have led to an improvement in both quality and clarity of this paper.

References

[1] A. Charnes, W.W. Cooper, Management Models and Industrial Applications of Linear Programming, vol. 1, John Wiley & Sons, 1961.
[2] S.M. Lee, Goal Programming for Decision Analysis, Auerbach Philadelphia, 1972.
[3] J.P. Ignizio, Goal Programming and Extensions, vol. 26, Lexington Books Lexington, MA, 1976.
[4] J.P. Ignizio, Linear Programming in Single-& Multiple-Objective Systems, Prentice-Hall Englewood Cliffs, NJ, 1982.
[5] M. Tamiz, D. Jones, E. El-Darzi, A review of goal programming and its applications, Ann. Oper. Res. 58 (1) (1995) 39–53.
[6] M. Tamiz, D. Jones, C. Romero, Goal programming for decision making: an overview of the current state-of-the-art, Eur. J. Oper. Res. 111 (3) (1998) 569–581.
[7] C. Romero, M. Tamiz, D. Jones, Goal programming, compromise programming and reference point method formulations: linkages and utility interpretations,
J. Oper. Res. Soc. 49 (9) (1998) 986–991.
[8] Y. Ijiri, Management Goals and Accounting for Control., Chicago, IL: Rand McNally, 1965.
[9] M.J. Schniederjans, Goal Programming: Methodology and Applications, Kluwer Academic Pub, 1995.
[10] M. Zeleny, J.L. Cochrane, Multiple Criteria Decision Making, vol. 25, McGraw-Hill New York, 1982.
[11] C.-T. Chang, Multi-choice goal programming, Omega 35 (4) (2007) 389–396.
 K.K. Patro et al. / Applied Mathematics and Computation 271 (2015) 489–501 501

[12] C.-T. Chang, Revised multi-choice goal programming, Appl. Math. Model. 32 (12) (2008) 2587–2595.
[13] M.P. Biswal, S. Acharya, Multi-choice multi-objective linear programming problem, J. Interdiscip. Math. 12 (5) (2009a) 606–637.
[14] M.P. Biswal, S. Acharya, Transformation of a multi-choice linear programming problem, Appl. Math. Comput. 210 (1) (2009b) 182–188.
[15] S. Acharya, M.P. Biswal, Solving probabilistic programming problems involving multi-choice parameters, OPSEARCH 48 (3) (2011) 217–235.
[16] C.-N. Liao, Formulating the multi-segment goal programming, Comput. Ind. Eng. 56 (1) (2009) 138–141.
[17] S. Acharya, M.M. Acharya, Generalized transformation techniques for multi-choice linear programming problems., Int. J. Optim. Control: Theor. Appl. 3 (1)
(2013) 45–54.
[18] M.P. Biswal, S. Acharya, Solving multi-choice linear programming problems by interpolating polynomials, Math. Comput. Model. 54 (5) (2011) 1405–1412.
[19] C.-T. Chang, C.-Y. Ku, H.-P. Ho, C. Liao, A mcgp decision aid for homebuyers to make the best choice, Qual. Quantity 45 (4) (2011) 969–983.
[20] O. Ustun, Multi-choice goal programming formulation based on the conic scalarizing function, Appl. Math. Model. 36 (3) (2012) 974–988.
[21] B. Bankian-Tabrizi, K. Shahanaghi, M. Saeed Jabalameli, Fuzzy multi-choice goal programming, Appl. Math. Model. 36 (4) (2012) 1415–1420.
[22] C.-T. Chang, H.-M. Chen, Z.-Y. Zhuang, Multi-coefficients goal programming, Comput. Ind. Eng. 62 (2) (2012) 616–623.
[23] A.F. Da Silva, F.A.S. Marins, J.A.B. Montevechi, Multi-choice mixed integer goal programming optimization for real problems in a sugar and ethanol milling
company, Appl. Math. Model. 37 (9) (2013) 6146–6162.
[24] J.L. Eatman, C. Sealey Jr, A multiobjective linear programming model for commercial bank balance sheet management, J. Bank Res. 9 (4) (1979) 227–236.
[25] L. Schrage, Lingo release 11.0 lindo system, inc.(2008).
[26] S. Acharya, S. Nanda, B.B. Mishra, Solving multi-choice linear goal programming problem with preemptive priorities, J. OMS 30 (2011) 52–70.