IJCSI International Journal of Computer Science Issues, Vol. 6, No.

2, 2009 20
ISSN (Online): 1694-0784
ISSN (Print): 1694-0814

Lexicographic Multi-objective Geometric Programming

Problems
Dr.A. K. Ojha1 and K. K. Biswal2

1
School of Basic Sciences, IIT Bhubaneswar,
Orissa, Pin-751013, India

2
Department of Mathematics, CTTC Bhubaneswar,
B-36, Chandaka Industrial Area, Bhubaneswar, Orissa, Pin-751024, India

Abstract are reported by Behringer[1]. Impressed upon the work

A Geometric programming (GP) is a type of
mathematical problem characterized by objective and
constraint functions that have a special form. Many
methods have been developed to solve large scale
engineering design GP problems. In this paper GP
technique has been used to solve multi-objective GP
problem as a vector optimization problem. The duality
theory for lexicographic geometric programming has
been developed to solve the problems with posynomial
in objectives and constraints.
Keywords: Lexicographic minimization, Geometric
programming, duality theory, vector minimization and
vector maximization.
1. Introduction
In many real-life optimization problems, multiple
objectives have been taken into account, which may be
related to the economical, social and environmental
aspects of optimization problems. The multiple
objectives are usually incommensurate and in conflict
with one another. In general, a multiple objective
optimization problem does not have a single solution
that could optimize all objectives simultaneously. It
never search for optimal solution but for efficient
solution that can best suited compromise solution to all
multiple objectives. Though the geometric
programming technique due to Duffin et al.[3] helps to
solve various types of nonlinear single objective
posynomial problems but there are very few work have
been made in this direction to solve multiple objective
GP problems. Lexicographic optimization approach is
one such technique to handle multiple objective GP
problem. Several mathematical and game theoretic
applications of nonlinear lexicographic optimizations
of Behringer, Nijkamp[6] applied the lexicographic
optimization technique in a land use problem for
industrial activities in a newly created industrial area in
a Rhine-delta region near Rotterdam. Application of
linear lexicographic due to Isermann[4] and
Turnovec[8] strengthen to handle scalar valued
optimization problems. Biswal[2] used fuzzy
programming[10] to solve multi-objective geometric
problem where as lexicographic order and duality has
been studied by Martinez[5]. In this paper we have
applied lexicographic geometric programming
technique to solve special type of multi-objective
optimization problem.
The organization of the paper is as follows: Following
introduction the definition of multi-objective geometric
programming and lexicographic optimization have been
discussed in Section-2 and 3 respectively. Definition of
lexicographic geometric programming has been
discussed in Section 4 and the numerical examples have
been incorporated in Section 5. Finally the conclusion
has been presented in Section 6.
2. Multi-objective geometric programming
A multi-objective geometric programming problem can
be defined as:
Find x = (x1, x2,… ,xn)T so as to
Tko n
min : g k 0 ( x) = ∑ Ck 0t ∏ x j k 0 tj , k = 1,2,..., p
a
t =1 j =1
(2.1)
subject to
IJCSI International Journal of Computer Science Issues, Vol. 6, No. 2, 2009 21
ISSN (Online): 1694-0784
ISSN (Print): 1694-0814
The problem of finding lexicographically minimum
Ti n point of F with respect to X is called lexicographically
gi ( x) = ∑ Cit ∏ x j itj ≤ 1 , i = 1,2,..., m (2.2)
d
minimizing problem which is denoted as :
t =1 j =1 lex min : {F ( x) : x ∈ X } (3.1)
x j > 0, j = 1,2,..., n (2.3)
where Ck0t for all k and t are positive real numbers and Similarly a point x ∈ X is said to be lexicographically
ditj and ak0tj are real numbers for all i, k, t, j. maximum point of F with respect to X if
Tk0=number of terms present in the kth objective lex F ( x) ≥ F ( x) and this problem is called
function.
Ti =number of terms present in the ith constraint. lexicographically maximizing problem given by
In the above multi-objective geometric programming lex max : {F ( x) : x ∈ X } (3.2)
problem there are p number of minimization type If we assume
objective functions, m number of inequality type
X = {x ∈ R n : G ( x) ≤ O m } (3.3)
constraints and n number of strictly positive decision
variables. then the lexicographically minimizing problem can be
Let us define F ( x) = {g10 ( x), g 20 ( x),...., g p 0 ( x) }
defined as,
lex min : {F ( x) : G ( x) ≤ O m } (3.4)
and G ( x ) = {g1 ( x) − 1, g 2 ( x ) − 1,...., g m ( x) − 1 } A lexicographically minimum point of F with respect to
Now the above optimization problem can be rewritten (3.3) is called an optimal solution to the problem given
as: by (3.4).
{
lex min : F ( x) : x ∈ R n , G ( x) ≤ 0 m
(2.4) } The problem given by (3.4) is called convex
optimization if all the components of F and G are
which is called lexicographic geometric programming
convex functions.
(LGP) problem.

3. Lexicographic optimization problem 4. Lexicographic geometric Programming

n
Let us denote O an n-dimensional zero vector and The lexicographic multi-objective geometric
m×n
an m×n zero matrix. An inequality of the type programming defined by
lex min :{F ( x) : x ∈ R n , G ( x) ≤ O m }
O
x ≥ O means x ≥ O n , but x ≠ O n
n

where the functions are defined by (2.1), (2.2) and (2.3).

A vector x ∈ R is said to be lexicographically non
n
Now we will prove the following theorem for the
n
negative if either x = O or its first non-zero component existence of unique optimal solution of the
is positive and we denote it by lex x ≥ O
n lexicographic optimization GP problem.

Similarly a non zero vector x ∈ R is said to be

n
Theorem:- If the primal problem of the geometric
lexicographically non positive if its first non zero programming is consistent and its dual program has a
component is negative and it is denoted by lex x < O maximizing point with strictly positive components then
n

An m×n matrix A is called lexicographically non the primal problem of geometric programming has a
negative if all its columns are lexicographically non unique optimal solution if and only if the rank of its
negative and we denote it as exponent matrix is equal to the number of columns.
lex A ≥ O m× n
Proof :-From the duality theorem due to Duffin[3], we
In the similar manner we can define lexicographically
know that for each optimization point x* for the primal
non positive and lexico-graphically negative vector and
problem there exist a maximizing point w* of dual
its corresponding matrix.
program which is given by the following equations.
Let F be an p-dimensional vector valued function
defined on R and
n
X ⊂ Rn. n
C k 0t ∏ x j k 0 tj = wk* 0t v( w* ), k = 1,2,... p, t = 1,2,....Tk 0 .
a
A vector x ∈ X is said to be a lexicographically
*

minimal point of F with respect to X if for any x ∈ X j =1

(4.1)
such that lex F ( x
*
) ≤ F ( x)
IJCSI International Journal of Computer Science Issues, Vol. 6, No. 2, 2009 22
ISSN (Online): 1694-0784
ISSN (Print): 1694-0814
Az = γ(4.7)
n *
w where A is the exponent matrix of dimension T×n and T
C it ∏ x j itj = , i = 1,2,...m, t = 1,2,....Ti .
d it
≥ n. From the basic knowledge of linear algebra the
j =1 λi ( w * ) system of equations (4.7) has a unique solution if and
(4.2) only if the rank of the matrix A is equal to the number
Taking logarithm on both sides of equation (4.1) and of its columns. With the substitution
(4.2) we have
e j = x j or z j = ln x j , j =1,2,..., n
z

wk 0t v(w* ) Our lexicographic optimization problem can be

ln(x1ak 0t1 x2ak 0t 2 ....xnak 0tn ) = ln , k = 1,2,....p, t = 1,2,....Tk 0 expressed as.
Ck 0t
lex min :{F ( z ) :G ( z ) ≤ 1m } (4.8)
wit
* where F(z) is an p-dimensional vector valued function
ln(x1d it1 x2ait 2 ....xnaitn ) = ln , i = 1,2,....m, t = 1,2,....Ti . with
Cit λi (w* ) n

which can be expressed as

Tk 0 ∑ a k 0 tj z j
g k 0 ( z ) = ∑ Ck 0t e j =1 , k=1,2,….,p
t =1
a k0t1 ln x1 + a k0t2 ln x 2 + …. + a k0tn ln x n = ln(β k0t ); n
Ti ∑ d itj z j
k = 1, 2,....p; t = 1, 2, ..., Tk0 and g i ( z ) = ∑ Cit e j =1 , i=1,2,….,m
(4.3) t =1
And Due to the monotonicity of logarithm function the
problem (4.8) can be expressed as:
d it1 ln x1 + d it2 ln x 2 + .... + d itn ln x n = ln(β it ); {
lex min : ln F ( z ) : ln G ( z ) ≤ O m }
i = 1, 2,....m; t = 1, 2,...Ti (4.9)
with
ln F ( z ) = {ln g10 ( z ), ln g 20 ( z ),...., ln g p 0 ( z )}
(4.4)
w* v( w* )
where β k 0t = kot , t = 1,2,....Tk 0 and ln G ( z ) = { ln g1 ( z ), ln g 2 ( z ),...., ln g m ( z )}
Ck 0 t
and Introducing new variables
n
* ∑ ak 0 tj z j
wit j =1
β it = , t = 1,2,....Ti , i = 1,2,...., m.
wkot =
Ck 0 t e
Cit λi ( w* ) n , k=1,2,…,p (4.10)
Tk 0 ∑ a k 0 tj z j
Now by substituting
zj = ln xj ; j = 1, 2,…., n ∑C
t =1
k 0t e j =1
ln(β k 0t ) = γ k 0t , k = 1,2,.... p, t = 1,2,.....,Tk 0 n

∑ d itj z j
Cit e j =1
and ln β it = γ it , i = 1,2,....m, t = 1,2,...., Ti And wkit = uki n , k=1,2,….p, i=
the equations (4.3) and (4.4) reduces to
Ti ∑ d itj z j
∑C e
t =1
it
j =1

ak 0t1 z1 + ak 0t 2 z2 + ... + ak 0tn zn = γ k 0t , t = 1,2,..., Tk 0 1,2,….,m (4.11)

(4.5) After suitable transformation [3] the dual problem
and associated with the lexicographic geometric
dit1z1 + dit2 z2 + ...+ ditnzn = γ it , i = 1,2,...,m, t = 1,2,...,Ti programming problem can be obtained.
(4.6) lex max : V ( w)
m Tk 0
If we assume T = ∑ Ti then we will have T number such that ∑w
t =1
k 0t = 1 , k=1,2,….,p (4.12)
i =1
of equations and n variables, which is exactly the
dimension of the exponent matrix.
Writing the equations in the matrix form we have
Tk 0 m Ti

∑ wk 0t ak 0tj + ∑∑ wkit ditj = 0, k = 1,2,...., p, j = 1,2,...., n, i = 1,2,...., m

t =1 i =1 t =1
IJCSI International Journal of Computer Science Issues, Vol. 6, No. 2, 2009 23
ISSN (Online): 1694-0784
ISSN (Print): 1694-0814
(4.13)
Ti
{ }
lex min:F(x) = g10(x) = x1−1x2−1x3−2 , g20(x) = x1−1x2−3x3−5 + x1−1x2−1
uki = ∑ wkit , k=1,2,….,p, i=
subject to x1 x 2 x3 + x 2 x3 ≤ 10
2
i =1

1,2,….,m (4.14) x1 x3 ≤ 2
where Vk ( w) is the p-dimensional vector valued x1 , x 2 , x3 > 0
function of the form
Solution Procedure of LGP problems.
Step 1:
k 0t wkit
At first the objective functions of the multi-objective GP
⎛C
Tk 0
⎞ m Ti
⎛ d it ⎞ m
problem are ranked according their priority. Let us
Vk ( w) = ∑ ⎜⎜ k 0t ⎟⎟ ∏∏ ⎜⎜ ⎟⎟ ∏u
u ki
ki , assume that the first objective function is in priority one
t =1 ⎝ wkot ⎠ i =1 t =1 ⎝ wkit ⎠ i =1 i.e. P1 and the second objective function in priority 2 i.e.
P2 and so on, and pth objective function in priority p i.e
k=1,2,….,p (4.15)
Pp.
According to the theory of geometric programming Step 2:
Then first objective function g10(x) is minimized subject
problem (4.12) is called the normality conditions, (4.13)
to all the original constraints. Let the minimum of the
is called orthogonality condition and (4.15) is the dual (1)
first objective be g10 at x(1) . Then we move to step 3.
function. Step 3:
p m Then the second objective function g20(x) is minimized
The number d = ∑T k0 + ∑ Ti − n − 1 is the degree subject to the original constraints with one additional
constraint, i.e. g10 ( x) ≥ g10
(1)
k =1 k =1

of difficulty. Let the minimum value of the second objective function

( 2)
As the problem with zero degree of difficulty is easily be g 20 at x(2). Then we move to next step.
Step 4:
solvable then the dual problem can be solved to get the
In step 4 third objective function g30(x) is minimized
maximizing vector w * . Since the vector w * is the subject to the original constraint with two additional
constraints, i.e. g10 ( x ) ≥ g10
(1)
unique solution to the dual constraints, it is also the
maximizing vector for the dual problem. g 20 ( x) ≥ g 20
( 2)

Same procedure is repeated for all the objective

Using this dual optimizing vector the optimal
functions.
solution x * to the primal problem can be determined by Step 5:
Finally, the last objective function is minimized subject
using following relationships. to all the original constraints plus additional p-1
constraints.
Ck 0t x1a k 0 t1 x2a k 0 t 2 ....xna k 0 tn = wk*0tVk ( wk*0t ) ,
g10 ( x) ≥ g10(1) , g 20 ( x) ≥ g 20
( 2)
, …. g p −1, 0 ( x ) ≥
k=1,2,….,p (4.16)
g (pp−−1,10)
*
w Let A be the exponent matrix.
Cit x1d it 1 x2d it 2 ....xnd itn = , k=1,2,…,p; i=1,2,…,m
it

⎛ −1 −1 − 2⎞
*
u ki
⎜ ⎟
(4.17) ⎜1 1 2 ⎟
A=⎜ The rank of the matrix is 2
0 1 1 ⎟
⎜ ⎟
Ti
where uki* = ∑ wkit
*
, k=1,2,….,p (4.18) ⎜1 0 1 ⎟⎠
i =1
⎝
which is less than the number of columns.
The dual program of the function g10 ( x ) is as
5. Numerical Example
Example: Let us consider a numerical example
IJCSI International Journal of Computer Science Issues, Vol. 6, No. 2, 2009 24
ISSN (Online): 1694-0784
ISSN (Print): 1694-0814
w111 w112
⎛ 1 ⎞
w101 ⎛1 ⎞ ⎛1 ⎞ ⎛ 0 .5 ⎞
w211

max :V ( w) = ⎜⎜ ⎟⎟ ⎜ 10 ⎟ ⎜ 10 ⎟ ⎜⎜ ⎟⎟ (w111 + w112 )w 111 + w112

⎜⎜ w ⎟⎟ ⎜⎜ w ⎟⎟
⎝ 101 ⎠
w
⎝ 111 ⎠ ⎝ 112 ⎠ ⎝ 211 ⎠
w
w112 = 0.3333334 , w211 = 0.3333334 and the
Subject to: w101 = 1 mean value of dual v(w*) = 0.15. Using the primal dual
− w101 + w111 + w211 = 0 relationship we have the optimal solution of g10(x) are
x1* =0.9086967, x2* = 1.514494, x3* = 2.200954 and its
− w101 + w111 + w112 = 0 primal optimal solution is 0.15.
Similarly the dual program of g20(x) is defined as,
− w101 + 2w111 + w112 + w211 = 0
The solution of the dual program gives
w101 = 1 , w111 = 0.666... ,
w211 w212
⎛ 1 ⎞
w201
⎛ 1 ⎞
w202 ⎛1 ⎞ ⎛1 ⎞ ⎛ 0.5 ⎞
w211

max:V (w) = ⎜⎜ ⎟⎟ ⎜⎜ ⎟⎟ ⎜ 10 ⎟ ⎜ 10 ⎟ ⎜⎜ ⎟⎟ (w211 + w212 )w 211+ w212

⎜⎜ w ⎟⎟ ⎜⎜ w ⎟⎟
⎝ 201 ⎠
w ⎝ 202 ⎠
w
⎝ 211 ⎠ ⎝ 212 ⎠ ⎝ 211 ⎠
w
References
Subject to: w201 + w202 = 1 [1] Behringer,F.A., Lexicographic quasiconcave
multiobjective program-ming, Zeitschrift fier operation
− w201 − w202 + w211 + w221 = 0 research 21,
− 3w201 − w202 + w211 + w221 = 0 103-116, 1977.
[2] Biswal,M.P., Fuzzy Programming technique to solve
− 5w201 + 2w211 + w212 + w221 = 0 multi-objective geometric programming problem, Fuzzy
The solution of the problem gives its optimal value sets and
V2 ( w* ) = 0.4316470 × 10−1 systems 51,67-71 ,1992.
[3] Duffin R.J.,Peterson E.L.and Zener C.M.,
* * *
with w201 = 0.6666667, w202 = 0.3333…, w211 = 1, Programming Theory and Application, Wiley, New
* * York. ,1967
w212 = 1.3333…, w221 = 0 [4] Isermann,H., Linear lexicographic optimization,
Using primal dual relationship we have x1 = 3.020273, OR-spektrum 4,223-228. ,1982
x2 = 23.01163, x3 = 0.2483217 [5] Martinez-Legaz,J.E., Lexicographic order and
duality in multiobjectve programming, European
6. Conclusions Journal of operational
In this chapter solution procedure of lexicographic GP research 33, 342-348, 1988.
has been presented. Unless the objective functions [6] Nijkamp,P. Environmental policy analysis in
ranked properly, lexicographic solution may not be operational methods and models, Wiley, New York.
acceptable to a design engineer. If there are p number of ,1980
objective functions one may formulate p! no of ways [7] Saaty T.L., A Scaling Method for Priorities in
priority, which is a very difficult task for a design Hierarchical Structures, Journal of Mathematical
engineer. Also sometimes more than one objective pshychology, 15,
functions remain in a priority. Unless the priority is 234-281, 1977
proper solution of a real life problem gives some [8] Turnovec F., Lexicographic optimization problems
abnormal result. To set the proper ranking, method of in production scheduling in optimization, Theory,
Analytic Hierarchy Process (AHP) may be adopted. methods,
Two popular method of AHP by Saaty[7] namely row- Applications, Vol-I Dum Tech- niky CSVTS Praha,
column Addpotion method and Eigen-value method is Prague, 295-306, 1985
used to find the proper ranking (weights) of the [9] Zimmermann,H.J. ,Fuzzy programming and linear
objective function. If the weights of the objective programming with several objective functions. Fuzzy
function can be estimated, then using the weights multi- sets and
objective GP problem can be converted to a single systems 1, 46-55. ,1978
objective GP problem and solved. [10] Zimmermann,H.J. Fuzzy set theory and its
applications, 2nd ed Kluwer Academic ,publishers,
Dordrecht-Boston,
1990.