Upload
sanjeet-singh
View
215
Download
0
Embed Size (px)
Citation preview
European Journal of Operational Research 160 (2005) 232–241
www.elsevier.com/locate/dsw
Continuous Optimization
Multiparametric sensitivity analysis in programmingproblem with linear-plus-linear fractional objective function
Sanjeet Singh a, Pankaj Gupta b,*, Davinder Bhatia a
a Department of Operational Research, University of Delhi, Delhi-110007, Indiab Department of Mathematics, Deen Dayal Upadhyaya College, Shivaji Marg, Karampura, New Delhi-110015, India
Received 27 June 2002; accepted 30 October 2002
Available online 14 November 2003
Abstract
In this paper, we study multiparametric sensitivity analysis for programming problems with linear-plus-linear
fractional objective function using the concept of maximum volume in the tolerance region. We construct critical re-
gions for simultaneous and independent perturbations in the objective function coefficients and in the right-hand-side
vector of the given problem. Necessary and sufficient conditions are derived to classify perturbation parameters as
�focal� and �non-focal�. Non-focal parameters can have unlimited variations, because of their low sensitivity in practice,
these parameters can be deleted from the analysis. For focal parameters, a maximum volume tolerance region is
characterized. Theoretical results are illustrated with the help of a numerical example.
� 2003 Elsevier B.V. All rights reserved.
Keywords: Linear-plus-linear fractional programming; Fractional programming; Multiparametric sensitivity analysis; Tolerance
approach; Parametric analysis
1. Introduction
A general linear-plus-linear fractional programming problem has the following form:
* Co
E-m
0377-2
doi:10
ðLLFPÞ Maximize F ðxÞ ¼ ctxþ ptxqtxþ h
subject to Ax ¼ b;
xP 0;
where A is m� n coefficient matrix with m < n; ct, pt and qt are n-dimensional row vectors; x and b are n-dimensional and m-dimensional column vectors respectively and h is a scalar quantity. It is assumed that
the feasible region of the problem (LLFP) is bounded.
rresponding author. Address: Flat No-01, Kamayani Kunj, Plot No-69, I.P. Extension, Delhi-110092, India.
ail address: [email protected] (P. Gupta).
217/$ - see front matter � 2003 Elsevier B.V. All rights reserved.
.1016/j.ejor.2002.10.001
S. Singh et al. / European Journal of Operational Research 160 (2005) 232–241 233
Teterev [13] pointed out that such problems arise when a compromise between absolute and relativeterms is to be maximized. Major applications of the problem (LLFP) can be found in; transportation,
problems of optimizing enterprise capital, the production development fund and the social, cultural and
construction fund. Teterev [13] derived an optimality criteria for (LLFP) using the simplex type algorithm.
Several authors [2,5,6,9,10,12,13] studied the problem (LLFP) and its variants and have discussed their
solution properties.
In practical applications the data collected may not be precise, we would like to know the effect of data
perturbation on the optimal solution. Hence, the study of sensitivity analysis is of great importance. In
general, the main focus of sensitivity analysis is on simultaneous and independent perturbation of theparameters. Besides this, all the parameters are required to be analyzed at their independent levels of
sensitivity. If one parameter is more sensitive than the others, the tolerance region characterized by treating
all the parameters at equal levels of sensitivity would be too small for the less sensitive parameters. If the
decision maker has prior knowledge that some parameters can be given unlimited variations without
affecting the original solution then we consider those parameters as non-focal and these non-focal
parameters can be deleted from the analysis. Wang and Huang [14] proposed the concept of maximum
volume in the tolerance region for the multiparametric sensitivity analysis of a single objective linear
programming problem. Their theory allows the more sensitive parameters called as �focal� to be investigatedat their independent levels of sensitivity, simultaneously and independently. This approach is a significant
improvement over the earlier approaches primarily because besides reducing the number of parameters in
the final analysis, it also handles the perturbation parameters with greater flexibility by allowing them to be
investigated at their independent levels of sensitivity.
In this paper, we extend the approach of Wang and Huang [14] to discuss multiparametric sensitivity
analysis for the problem (LLFP).
2. Problem formulation and sensitivity models
To address perturbations of the right-hand-side vector, and the coefficients of the objective function, in
problem (LLFP), we consider the following perturbed model:
ðPLLFPÞ Maximize F ðxÞ ¼ ðcþ DcÞtxþ ðp þ DpÞtxqtxþ h
subject to Ax ¼ bþ Db;
xP 0;
where
Dc ¼XHh¼1
cjhch
" #n�1
; Dp ¼XHh¼1
pjhch
" #n�1
; Db ¼XHh¼1
bjhch
" #m�1
are the multiparametric perturbations defined by the perturbation parameter c ¼ ðc1; c2; . . . ; cH Þt. Here H is
the total number of parameters.
Next, we give some notations that are used throughout this study:
B � f1; 2; . . . ; ng denotes the index set of basic variables.
Without loss of generality, we suppose B ¼ f1; 2; . . . ;mg.N ¼ f1; 2; . . . ; ng n B denotes the index set of non-basic variables.
AB ¼The basis matrix with inverse b ¼ A1B ¼ ½bij�.
234 S. Singh et al. / European Journal of Operational Research 160 (2005) 232–241
AN ¼The submatrix of A corresponding to the non-basic variables.
cB ¼ ½c1; c2; . . . ; cm�t and cN ¼ ½cmþ1; cmþ2; . . . ; cn�t: The vectors of the objective function coefficients corre-
sponding to the basic and non-basic variables respectively.�b ¼ A1
B bP 0: The vector of the values corresponding to xB.A1B AN ¼ ½yij�.
uh ¼ ½b1h; b2h; . . . ; bmh�t: The vector of the coefficients of parameter ch in Db.vh ¼ ½c1h; c2h; . . . ; cmh�t: The vector of the partial coefficients of parameter ch in Dc.wh ¼ ½p1h; p2h; . . . ; pmh�t: The vector of the partial coefficients of parameter ch in Dp.�Dj ¼The vector of the reduced cost corresponding to non-basic variables.
P j ¼ jth column of the matrix P .
Under the assumptions ctxP 0 and qtxþ h > 0 over the feasible region, the optimality criteria for the
problem (LLFP) using the simplex type algorithm given by Teterev [13] is stated as follows:
Let AB denote the optimal basis matrix and let x ¼ xB0
� �¼ ½x1; x2; . . . ; xm; 0; . . . ; 0�t be the corresponding
basic feasible solution to the problem (LLFP). This solution will be an optimal solution if
Dj ¼ ðcj zcjÞ þZ 00ðpj zpj Þ Z 0ðqj zqj Þ
ðZ 00Þ26 0 for j ¼ 1; 2; . . . ; n;
where
Z 0 ¼ ptBxB; Z 00 ¼ qtBxB þ h;
zc ¼ ctBA1B AN zq ¼ qtBA
1B AN ;
zp ¼ ptBA1B AN :
Here cB, pB and qB are the subvectors of c, p and q respectively that correspond to the basis B.In general, the sensitivity analysis focuses on characterizing sets called as critical regions over which the
right-hand-side vector and objective function coefficients of the problem (LLFP) may vary while still
retaining the same optimal basis B [3]. Let S be a general notation for a critical region.
In the following proposition, we construct critical region for simultaneous and independent perturba-
tions with respect to b, p and c, respectively.
Proposition 1.When b, p and c are perturbed simultaneously, the critical region Rbpc of the problem (PLLFP)is given by
Rbpc ¼ ðc1;c2; . . . ;cH Þtj�bi
8>>><>>>:
þXHh¼1
ðbti uhÞchP0; i¼ 1;2; . . . ;m; �Djþ
XHh¼1
ðcmþj;h yt jvhÞch
þ
Pmi¼1
qixiþh
� �pj
PHh¼1
ðyt jwhpmþj;hÞch� �
qjPmi¼1
qiyij
� � PHh¼1
ðxtBwhÞch� �� �
Pmi¼1
qixiþh
� �2 60; j¼ 1;2; . . . ;nm
9>>>=>>>;: ð1Þ
S. Singh et al. / European Journal of Operational Research 160 (2005) 232–241 235
Proof. Let us assume that xB be the new basic solution, then
xB ¼ A1B ðbþ DbÞ ¼ A1
B bþ A1B Db ¼ �bþ A1
B
XHh¼1
b1hch;XHh¼1
b2hch; . . . ;XHh¼1
bmhch
" #t
¼ �bþXHh¼1
ðbt1 uhÞch;
XHh¼1
ðbt2 uhÞch; . . . ;
XHh¼1
ðbtm uhÞch
" #t:
Now ith component of xB is given by
xBi ¼ �bi þXHh¼1
ðbti uhÞch; i ¼ 1; 2 . . . ;m:
This new basic solution xB will be feasible if
�bi þXHh¼1
ðbti uhÞch P 0; i ¼ 1; 2; . . . ;m:
For the new solution xB, to satisfy optimality condition, the new values Dj�s of Dj are computed as follows:
Dj ¼ cj
þXHh¼1
cmþj;hch
! zcj þ
Z 00 pj þPHh¼1
pmþj;hch
� � zpj
� � Z 0 qj zqj
� �½Z 00�2
; j ¼ 1; 2; . . . ; n m:
Here,
Z 0 ¼Xmi¼1
pi
þXHh¼1
pihch
!xi; Z 00 ¼
Xmi¼1
qixi þ h;
zpj ¼Xmi¼1
pi
þXHh¼1
pihch
!yij; zqj ¼
Xmi¼1
qiyij;
zcj ¼Xmi¼1
ci
þXHh¼1
cihch
!yij:
Thus, Dj takes the form
Dj ¼ cj
þXHh¼1
cmþj;hch
!Xmi¼1
ci
þXHh¼1
cihch
!yij
þ
Pmi¼1
qixiþh
� �pjþ
PHh¼1
pmþj;hch
� �Pmi¼1
piþPHh¼1
pihch
� �yij
� �Pmi¼1
piþPHh¼1
pihch
� �xi
� �qj
Pmi¼1
qiyij
� �� �Pmi¼1
qixiþh
� �2
¼ cj
Xmi¼1
ciyij
!þ
Pmi¼1
qixiþh
� �pj
Pmi¼1
piyij
� �Pmi¼1
pixi
� �qj
Pmi¼1
qiyij
� �Pmi¼1
qixiþh
� �2
236 S. Singh et al. / European Journal of Operational Research 160 (2005) 232–241
þXHh¼1
cmþj;h�
yt jvh�ch þ
Pmi¼1
qixi þ h
� �pj
PHh¼1
ðyt jwh pmþj;hÞch� �
qj Pmi¼1
qiyij
� � PHh¼1
ðxtBwhÞch� �
Pmi¼1
qixi þ h
� �2
¼ �Dj þXHh¼1
cmþj;h�
yt jvh�ch þ
Pmi¼1
qixi þ h
� �pj
PHh¼1
ðyt jwh pmþj;hÞch� �
qj Pmi¼1
qiyij
� � PHh¼1
ðxtBwhÞch� �
Pmi¼1
qixi þ h
� �2 :
Now solution xB will be optimal if
�Dj þXHh¼1
cmþj;h�
yt jvh�ch
þ
Pmi¼1
qixi þ h
� �pj
PHh¼1
ðyt jwh pmþj;hÞch� �
qj Pmi¼1
qiyij
� � PHh¼1
ðxtBwhÞch� �
Pmi¼1
qixi þ h
� �2 6 0; j ¼ 1; 2; . . . ; n m:
Thus the critical region Rbpc is given by
Rbpc ¼ ðc1;c2; . . . ;cH Þtj�bi
8>>><>>>:
þXHh¼1
ðbti uhÞchP0; i¼ 1;2; . . . ;m; �Djþ
XHh¼1
ðcmþj;h yt jvhÞch
þ
Pmi¼1
qixiþh
� �pj
PHh¼1
ðyt jwhpmþj;hÞch� �
qjPmi¼1
qiyij
� � PHh¼1
ðxtBwhÞch� �� �
Pmi¼1
qixiþh
� �2 60; j¼ 1;2; . . . ;nm
9>>>=>>>;: �
It can be seen that a parameter ch is absent in Rbpc if and only if
bti uh ¼ 0 for i ¼ 1; 2; . . . ;m; cmþj;h ¼ yt jvh;
yt jwh ¼ pmþj;h for j ¼ 1; 2; . . . ; n m and xtBwh ¼ 0:
Therefore, it is not necessary that all the parameters c1; c2; . . . ; cH appear simultaneously in Rbpc. Those
parameters that are not present in Rbpc are called the non-focal parameters.
Corollary 1. When perturbation is on b only (i.e. Dc ¼ 0, Dp ¼ 0), the critical region Rb of the problem(PLLFP) is given by
Rb ¼ c
(¼ ðc1; c2; . . . ; cH Þ
tj�bi þXHh¼1
ðbti uhÞch P 0; i ¼ 1; 2; . . . ;m
): ð2Þ
Parameter ch is non-focal in Rb if and only if bti uh ¼ 0 for i ¼ 1; 2; . . . ;m. That is, vector uh is orthogonal to
bi for i ¼ 1; 2; . . . ;m. Since A1B is invertible and bt
i is the ith row of A1B , the set of column vectors
S. Singh et al. / European Journal of Operational Research 160 (2005) 232–241 237
fb1 ; b2 ; . . . ; bm g is a linearly independent set. Therefore, if bti uh ¼ 0 for i ¼ 1; 2; . . . ;m, then uh ¼ 0.
Therefore a parameter ch is non-focal in Rb if and only if uh ¼ 0.
Corollary 2.When perturbation is in c and p, (i.e. Db ¼ 0), the critical region Rpc of the problem (PLLFP) isgiven by 8
Rpc ¼ c
>>><>>>:
¼ ðc1;c2; . . . ;cH Þtj�Dj þ
XHh¼1
ðcmþj;h yt:jvhÞch
þ
Pmi¼1
qixi þ h
� �pj
PHh¼1
ðyt jwh pmþj;hÞch� �
qj Pmi¼1
qiyij
� � PHh¼1
ðxtBwhÞch� �
Pmi¼1
qixi þ h
� �2 60 for j¼ 1;2; . . . ;nm
9>>>=>>>;:
ð3Þt t
Parameter ch is non-focal in Rpc if and only if cmþj;h ¼ y jvh; y jwh ¼ pmþj;h for j ¼ 1; 2; . . . ; n m andxtBwh ¼ 0.
Remark 1. For the problem (LLFP), a local optimal solution need not be global solution [4]. However, if
the objective function of the problem is pseudoconcave over the feasible region then the local optima is alsoa global optima.
Remark 2. Moreover, if the objective function of (LLFP) is pseudoconcave and hence quasiconcave, the
optimal solution is attained at an extreme point of the feasible region, which is a compact set [11].
These features of an optimal solution are very valuable from the computational point of view. The next
result provides the conditions which ensure the pseudoconcavity of the linear-plus-linear fractional
objective function.
Lemma [1]. The function F in the problem (LLFP) is pseudoconcave over the feasible region of the problem ifand only if one of the following conditions holds:
(i) c ¼ kq, kP 0;
(ii) there is s 2 R such that p ¼ sq and shP 0.
However, in this paper, sensitivity analysis has been carried out for the local/global optimal solution of the
problem (LLFP).
Definition 1 [14]. The maximum volume region (MVR) BS in a critical region S is given by
BS ¼ rn
¼ ðr1; r2; . . . ; rHÞtjjrjj6 k�j ; j ¼ 1; 2; . . . ;Ho;
where k� ¼ ðk�1 ; k�2 ; . . . ; k�H Þ is the optimal solution of the following maximization problem:
ðP1Þ Maximizek2KðSÞ
V ðkÞ ¼ k1 k2 k3 kH ;
where
KðSÞ ¼ fk ¼ ðk1; k2; . . . ; kH Þtjjrjj6 kj; j ¼ 1; 2; . . . ;H implies r ¼ ðr1; r2; . . . ; rH Þt 2 Sg:
The volume of BS is VolðBSÞ ¼ 2Hk�1 k�2 k�H .238 S. Singh et al. / European Journal of Operational Research 160 (2005) 232–241
Let us suppose that out of H parameters only J parameters are obtained as focal parameters. Since thecritical region is a polyhedral set, there exists L ¼ ½‘ij� 2 RI�J , d ¼ fdig 2 RI , I ; J 2 N , where I and J are the
number of constraints and variables of S, respectively, such that S ¼ fr ¼ ðr1; r2; . . . ; rJ ÞtjLr6 dg. For focalparameters, it is assumed that ‘ j 6¼ 0 for j ¼ 1; 2; . . . ; J .
Remark 3. It follows from Proposition 1 and Corollaries 1 and 2 that r ¼ 0 belongs to Rpc, Rb and Rbpc, and
thus we have d P 0.
Consider the case when d > 0. The set KðSÞ can be rewritten as
KðSÞ ¼ k�
¼ ðk1; k2; . . . ; kJ Þtjjrjj6 kj; j ¼ 1; 2; . . . ; J implies jLjr6 d�;
where jLj is the matrix obtained by changing the negative elements of L to positive. Since jLjP 0, followingWang and Huang [14], the equivalent form of KðSÞ is
KðSÞ ¼ k�
¼ ðk1; k2; . . . ; kJ ÞtjjLjk6 d; kP 0�:
The MVR BS of a polyhedral set S ¼ frjLr6 dg with d > 0 and ‘ j 6¼ 0 for all j is bounded and is defined by
BS ¼ rn
¼ ðr1; r2; . . . ; rJ Þtjjrjj6 k�j ; j ¼ 1; 2; . . . ; Jo;
where k� ¼ ðk�1 ; k�2 ; . . . ; k�J Þ is the unique optimal solution of the following problem:
ðP2Þ Maximize V ðkÞ ¼ k1 k2 k3 kJsubject to jLjk6 d;
kP 0:
Multiparametric sensitivity analysis of the problem (PLLFP) can now be performed as follows:
Obtain the critical regions as given in (1)–(3) by considering perturbations in the objective function
coefficients and in the right-hand-side vector. Delete all the non-focal parameters from the analysis. The
MVR of the critical regions is obtained by solving the problem (P2). The problem (P2) can be solved byexisting techniques such as Dynamic Programming. The detailed algorithm can be found in Wang and
Huang [15]. Software GINO [7] can also be used to solve the non-linear programming problem (P2).
3. Numerical example
We present the following numerical example to illustrate the theoretical results presented in this paper.
ðLLFPÞ Maximize F ðxÞ ¼ ð2x1 þ 6x2 þ 2x3Þ þ3x1 þ 5x2 þ 6x3x1 þ 3x2 þ x3 þ 2
subject to 3x1 x2 þ 2x3 þ x4 ¼ 7
2x1 þ 4x2 þ x5 ¼ 12
4x1 þ 3x2 þ 8x3 þ x6 ¼ 10
xj P 0; j ¼ 1; 2; . . . ; 6:
In view of Lemma, it can easily be seen that the objective function F is pseudoconcave. Hence, local
maxima of (LLFP) is also global.Using simplex type procedure of Teterev [13], the initial and the final simplex tables are given in Tables 1
and 2.
Table 2
Optimal simplex table
qj ! 1 3 1 0 0 0
pj ! 3 5 6 0 0 0
cj ! 2 6 2 0 0 0
qB pB cB Basic variables x1 ðz1Þ x2 ðz2Þ x3 ðz3Þ x4 ðz4Þ x5 ðz5Þ x6 ðz6Þ b
1 3 2 x1ðz1Þ 1 0 45
25
110
0 4
3 5 6 x2ðz2Þ 0 1 25
15
310
0 5
0 0 0 x6ðz3Þ 0 0 10 1 12
1 11
ðcj zjÞ ! 0 0 )2 )2 )2 0
ðpj zpj Þ ! 0 0 1.6 )2.2 )1.8 0 Z 0 ¼ 37
ðqj zqj Þ ! 0 0 )1 )1 )1 0 Z 00 ¼ 21
Dj ! 0 0 )1.84 )2.02 )2.00 0
Table 1
Initial simplex table
Basic variables x1 x2 x3 x4 x5 x6 b
x4 3 )1 2 1 0 0 7
x5 )2 4 0 0 1 0 12
x6 )4 3 8 0 0 1 10
S. Singh et al. / European Journal of Operational Research 160 (2005) 232–241 239
The optimal solution is
x� ¼ ½x�1; x�2; . . . ; x�6�t ¼ ½4; 5; 0; 0; 0; 11�t ¼ ½4; 5; 11; 0; 0; 0�t ¼ ½Z1; Z2; Z3; . . . ; Z6�t ½say�:
Here B ¼ f1; 2; 3g. The matrix of the optimal basis is
AB ¼3 1 0
2 4 0
4 3 1
24
35 and its inverse b ¼ A1
B ¼25
110
015
310
01 1
21
24
35:
Also
�b ¼ ½4; 5; 11�t; A1B AN ¼
45
25
110
25
15
310
10 1 12
24
35:
Consider perturbations in b; c and p simultaneously
Db ¼ ½2c1 þ 5c2 þ 0c3; 0c1 þ 0c2 þ c3; c1 c2 þ c3�t;
Dc ¼ ½5c1 5c2 þ 5c3 5c4; 10c1 þ 5c2 10c3 þ 5c4; c1 c2 þ 0:1c3 0:2c4�t;
Dp ¼ ½10c1 þ 5c2 5c3 þ 5c4; 5c1 5c2 þ 10c3 5c4; c1 þ c2 þ 0:1c3 þ 5=11c4�t;
u1 ¼ ½2; 0; 1�t; u2 ¼ ½5; 0;1�t; u3 ¼ ½0; 1; 1�t; u4 ¼ ½0; 0; 0�t;
v1 ¼ ½5; 10; 1�t; v2 ¼ ½5; 5;1�t; v3 ¼ ½5;10; 0:1�t; v4 ¼ ½5; 5;0:2�t;
240 S. Singh et al. / European Journal of Operational Research 160 (2005) 232–241
w1 ¼ ½10; 5; 1�t; w2 ¼ ½5;5; 1�t; w3 ¼ ½5; 10; 0:1�t; w4 ¼ 5;
� 5;
5
11
�t;
�Dj ¼ ½1:84;2:02;2:00�t;
c41 ¼ 5; c42 ¼ 1; c43 ¼ 1; c44 ¼ 4;
c51 ¼ 2; c52 ¼ 3; c53 ¼ 3; c54 ¼ 1:2;
c61 ¼ 3; c62 ¼ 1; c63 ¼ 3; c64 ¼ 1:1;
p41 ¼ 19; p42 ¼ 10; p43 ¼ 1; p44 ¼ 6:60;
p51 ¼ 5; p52 ¼ 2; p53 ¼ 1; p54 ¼ 1:46;
p61 ¼ 1; p62 ¼ 1; p63 ¼ 3; p64 ¼ 0:77:
Here n ¼ 6, m ¼ 3.
Because yt jv4 c3þj;4 ¼ 0, yt jw4 p3þj;4 ¼ 0, for j ¼ 1,2,3, xtBw4 ¼ 0 and u4 ¼ 0, therefore, c4 is a non-
focal parameter and can be varied unlimitedly. The critical region Rbpc defined by three focal parameters c1,c2 and c3 is given by
Rbpc ¼ c�
¼ ðc1; c2; c3Þtj4þ 0:8c1 þ 2c2 þ 0:1c3 P 0; 5þ 0:4c1 þ c2 þ 0:3c3 P 0; 11þ 3c1 þ 4c2 þ 0:5c3 P 0;
1:697 14:84c1 þ 10:92c2 2:68c3 6 0; 1:78 2:78c1 0:99c2 2:98c3 6 0;
1:72 5:88c1 2:52c2 0:44c3 6 0�:
The MVR of Rbpc is obtained by solving the following maximization problem
Max V ðkÞ ¼ k1 k2 k3subject to 0:8k1 þ 2k2 þ 0:1k3 6 4;
0:4k1 þ k2 þ 0:3k3 6 5;
3k1 þ 4k2 þ 0:5k3 6 11;
14:84k1 þ 10:92k2 þ 2:68k3 6 1:697;
2:78k1 þ 0:99k2 þ 2:98k3 6 1:78;
5:88k1 þ 2:52k2 þ 0:44k3 6 1:72;
k1; k2; k3 P 0:
The optimal solution of the above problem is k� ¼ ð0:04; 0:05; 0:21Þ
MVR BRbpc ¼ c�¼ ðc1; c2; c3Þ
tjjc1j6 0:04; jc2j6 0:05; jc3j6 0:21�:
VolðBRbpcÞ ¼ 23ð0:04Þ ð0:05Þ ð0:21Þ ¼ 0:0034:
Comparison. For the above perturbations, the following tolerance region is obtained using the tolerance
approach [8,16] treating all four parameters at equal levels of sensitivity:
S ¼ c�
¼ ðc1; c2; c3; c4Þtjjc1j6 0:06; jc2j6 0:06; jc3j6 0:06; jc4j6 0:06
�:
The volume of the above region in the reduced three dimensional space is 23Æ(0.06)(0.06)(0.06)¼ 0.0017
which is less than Vol(Rbpc). Also all the four parameters are treated at equal levels of sensitivity using the
S. Singh et al. / European Journal of Operational Research 160 (2005) 232–241 241
tolerance approach while the approach adopted in the present paper enables the less sensitive parameter c4to vary in R. Thus the present approach to sensitivity analysis not only gives the maximum volume tol-
erance region but also handles the simultaneous and independent perturbations of the parameters with
greater flexibility.
4. Conclusion
In the present paper, we discuss multiparametric sensitivity analysis for the problem (LLFP) by clas-
sifying the perturbation parameters as �focal� and �non-focal�. This approach not only reduces the number of
parameters in the final analysis but also allow the different parameters to be investigated at their inde-
pendent levels of sensitivity. The study of perturbations in the constraint matrix of the problem (LLFP)
would be investigated by the authors in their forthcoming research paper.
Acknowledgements
The authors are thankful to the unknown referees for their critical evaluation of the paper and sug-
gestions for the improvement in presentation of the paper. The authors also wish to express their deep
gratitude to Professor R.N. Kaul (Retd.), Department of Mathematics, University of Delhi, Delhi and
Professor M.C. Puri, Department of Mathematics, I.I.T., Delhi for their encouragement and inspiration to
complete the work.
References
[1] A. Cambini, J.-P. Crouzeix, L. Martein, On the pseudoconvexity of a quadratic fractional function, Optimization 51 (4) (2002)
677–687.
[2] S.S. Chadha, Dual of the sum of a linear and linear fractional program, European Journal of Operational Research 67 (1993) 136–
139.
[3] T. Gal, Postoptimal Analysis, Parametric Programming and Related Topics, McGraw-Hill, New York, 1979.
[4] J. Hirche, On programming problems with a linear-plus-linear fractional objective function, Cahiers du Centre d�Etudes de
Recherche Op�erationelle 26 (1–2) (1984) 59–64.
[5] J. Hirche, A note on programming problems with linear-plus-linear fractional objective functions, European Journal of
Operational Research 89 (1996) 212–214.
[6] H.C. Joksch, Programming with fractional linear objective functions, Naval Research Logistics Quarterly 11 (1964) 197–204.
[7] J. Liebman, L. Ladson, L. Scharge, A. Waren, Modeling and Optimization with GINO, The Scientific Press, San Francisco, CA,
1986.
[8] N. Ravi, R.E. Wendell, The tolerance approach to sensitivity analysis of matrix coefficients in linear programming, Management
Science 35 (1989) 1106–1119.
[9] S. Schaible, A note on the sum of a linear and linear fractional functions, Naval Research Logistics Quarterly 24 (1977) 691–693.
[10] S. Schaible, Fractional programming: Applications and algorithms, European Journal of Operational Research 7 (1981) 111–120.
[11] S. Schaible, Simultaneous optimization of absolute and relative terms, Zeitschrift fur Angewandte Mathematik und Mechanik 64
(8) (1984) 363–364.
[12] S.M. Sinha, G.C. Tuteja, On fractional programming, Opsearch 36 (4) (1999) 418–424.
[13] A.G. Teterev, On a generalization of linear and piecewise linear programming, Matekon 6 (1970) 246–259.
[14] H.F. Wang, C.S. Huang, Multi-parametric analysis of the maximum tolerance in a linear programming problem, European
Journal of Operational Research 67 (1993) 75–87.
[15] H.F. Wang, C.S. Huang, The maximal tolerance analysis on the constraint matrix in linear programming, Journal of the Chinese
Institute of Engineers 15 (5) (1992) 507–517.
[16] R.E. Wendell, The tolerance approach to sensitivity analysis in linear programming, Management Science 31 (1985) 564–578.