Indian J. pure appl. Math, 91) : 39-$0, January 1999 © Prinied in India NONLINEAR INTEGER GOAL PROGRAMMING NEELAM MALHOTRA AND S. R. ARORA Department of Mathematics, Hans Raj College, University of Delhi, Delhi 110 007 (Received 10 May 1998; Accepted 21 May 1998) ‘The purpose of this paper is to present a generalized approach for solving Integer Goal Programming Problems involving nonlinear multiple and conflicting. goals subject to a set of nonlinear constraints. The method is based on Goal Programming (GP) under pre-emptive priory structure. The algorithm is illustrated by an example given at the end Key Words : Fractional Programming: Non-linear Programming; Goal Programming; Multiple Objective; Integer Programming 1, INTRODUCTION Idea of Goal Programming was first introduced by Chames and Cooper!. They presented an approach to find solution of linear decision models having more than one objective functions. The work of Charnes and Cooper, Ijiri? and others has resulted in a systematic approach called Goal Programming (GP) for solving Linear Multiple Objective Programming problem in which pre-emptive priorities and weights are associated with the objectives, Basic assumptions of the procedure are (a) Establishment of pre-emptive priorities for each objective or a group of objectives. (6) Non-negative restriction on the variables. is indicated by P,, the next highest by P, and so on. The notion of The highest pri pre-emptive priorities holds that P, is preferred to P, regardless of any multiplier associated with P,, This is expressed by writing P, > P, This paper presents a pre-emptive GP model and: a procedure to solve Integer GP problem with nonlinear structural constraints. The process consists of finding feasible region of the decision variables in an augmented feasible goal programming model corresponding to the structural constraints. 2. PROBLEM FORMULATION The general priority based GP problem with nonlinear goal objectives and structural constraints is40 NEELAM MALHOTRA AND S. R. ARORA Find x(x1, x2, --.,%n) $0 as to - +t Minimize Py (Win dn) + Wy Jy) MinimizeP, Oy ding + pnt Ang) 21) Minimize Py (Wing dou + Wine doy) subject to F,(x,) +d, 1%) where it is assumed that the feasible region S(# 9) is bounded. F,(x,) is the 1th objective function associated with nth decision variable x,. h,(x) are given linear (nonlinear) functions. Term b; is the right hand side of the ith resource constraint. P, (k = 1, 2, . K, KST-N) is the kth priority factor assigned to the set of goals that are grouped together in the problem formulation. Priority factors have the relationship PL > P)> ... > Py, W., and Wy, are the numerical weights associated with deviational variable d,, and d,, renamed in as dy and dy, respectively, introduced for objective goals b,, to represent them at the kth priority level. The problem (2.1) is solved in two stages. At first stage an “auxiliary” problem is designed for generating feasible region of the decision variables. At the second stage, problem of attainment of the goal objectives is considered which may be called the “Principal” problem. Now attainment of the objectives goals b,, depend on xe S. Under the flexible nature of GP, the structural rigid constraints can also be considered as achievable goals by introducing deviational variables to each of them and considering bj, i = 1, 2, ... M as aspiration levels. This leads to solving another GP problem given by = Find x(x, 9)... %,) 80 as to Minimize P,(w3, dy + wy dy), Minimize Pw, dj, + wid) and Minimize P (wi dy +wiy d;)) subject to Afa+d, +d, =b, i= 1,2, 4MNONLINEAR INTEGER GOAL PROGRAMMING x,d,,d; 20 dd =0 and where P, (j=1,2,...,J,JSM) represents the jth priority factor assigned to the set of goals that are grouped together’ in the problem formulation, All other notations can be defined in an analogous to the problem in (2.1) 3. LINEARIZING NONLINEAR INTEGER GOAL PROGRAMMING PROBLEMS In this-section linear GP problem equivalent to the given Nonlinear GP in which the constraints are either quadratic ot fractional functions is formulated. 3.1 GP Problems with Quadratic Constraints Quadratic GP problem is given by Find x(2j,....%,) 0 as to Minimize Py (65,4 dont + Wont dont) Minimize Py(ing ding + np Ant) Minimize Pgliny ding + ina nx) subject to r= 1,2, ., Tand n= 1, 2,..,.N xe S= {re RYIA() Sb, x20, x is imeger d,.d),20d,,-d",=0. oe Ai@), i = 1, 2, Mare quadratic functions The above integer GP problem can be transformed in to a linear zero-one model and the method of integer GP* can be applied to the transformed problem. An integer variable x, is replaced by the sum of a number of zero-one variables by using the following transformation42 NEELAM MALHOTRA AND S. R. ARORA Yqr Where y,, are O or 1. Nis determined according to the upper limit on the variable x,. Any zero-one variable that has a positive exponent can be replaced by that variable to the power one. That is re y>0. Xx, is a zero-one variable. Secondly, any product of 0-1 variables may be changed to linear zero-one function as follows Let u=.,x) x, Introduce the following constraints - YX y+vS0 where u = 0, 1 3.2 GP Problems with Fractional Constraints Every Nonlinear Integer Fractional GP problem can also be transformed into a linear zero-one model as follows Let the ith goal constraints be eee x05" d,-d, or fx) +d, -g,0)-d) - g(x) =b,g, (0) with the objective Minimize d, +d, subject to x20,d;,d, 20 and dd) =0 2) where d; ,d) are deviational variables. To solve problem (3.2), solve the following problem Minimize v, + v; subject toNONLINEAR INTEGER GOAL PROGRAMMING 43 F(0 + Yj —¥, = 8,8, 0) 3) x20.) 9) 20,0) 7 vj +¥) are deviational variables given by d+ g(x), vy =d, - g(x). Find the optimal values of vj and v; and from this find the optimal values of d; and d* respectively using the following theorems. Theorem | [4] — If the minimum of problem (3.2) and (3.3) occur at x, and x respectively where x, and x, are in S. Then (i) d, =0 if and only if v, = 0 (ii) dj = 0 if and only if v;, Theorem 2 — If the minimum of problems (3.2) and (3.3) occur at x, and xy respectively, where xy and x, are in S, then @ and (ii) PROOF : Let v7 >0. Since x, is the minimum solution of (3.2) and d)>0 (‘v; #0 and d)20) Als) ae) alae ae = ny Vil) ~ baie) Coa is minimum of (3.3)44 NEELAM MALHOTRA AND S. R. ARORA Again, since x, is minimum of (3.3), we have Fh%p) — 8; 8) Sf) - 5 84) fie) = -b, soop( 20] = g(x) -d). a0 Combining I and II we get the relation (i). Proof of (ii) is similar to that of (i), From Theorem 1, we see that v; =0 if and only if d, =O. fie) Now v)=0 gives bg(x3)-f(x,)=0 => b= ae) and d; =0 gives , 1204 fen Ye Spe gi) BiG) which implies Si) _ fle) Be) 8) Same hold if we use v; =0 if and only if d; . Optimal solution x, of the related problem (3.3) is also the optimal solution of problem (3.2). Thus we conclude that vj =0, in the optimal solution of the related GP Problem (3.3), then optimal @ ity, values of d, and d. are also zero and the goal is completely achieved. (ii) In case v,>0 in the optimal solution of the related GP Problem (3.3), then d,>0 and to find the optimal value of d,, we solve the following problem Minimize d, = Minimize 8,0) subject to the constraintsNONLINEAR INTEGER GOAL PROGRAMMING 45 Fe) ~b,8,@) 59,20 where 4, SOLUTION PROCEDURE Problem (2.1) is solved in two phases. Phase 1 — Auxiliary problem (2.2) after linearization is solved to generate the set of feasible solutions and thereby giving the feasible region of the decision variables. This is called extended feasible region Phase 2 — The problem of attainment of the goal objectives in the problem (2.1) is considered. 4.1, Solution Methodology of Phase 1 Here, in the auxiliary Problem (2.2), the attainment of the resource goals generated from the resource, constraints are considered. The modified iterative solution methodology for Integet GP problem? can be used to solve Problem (2.2). The optimal solution of problem (2.2) will be a feasible solution for the attainment of the goal objective but may not be the most satisfying. This is because of the fact that objective goals and goals of the problem (2.2) are incommensurable, So, the set of feasible solutions of problem (2.1) is searched. This can be done as follows ieee Let p= (xy. 95 2g (= 1, 2, son L) be the solutions corresponding to the Mth iteration The convex comt tion of all such solutions is L x= D Ayes, L where Ay=1, OSA,S1 and A, is a scalar. 1 " 4 Let x (e 1, 2, .., 2) be the solutions obtained for different sets of values of A, (I . L) in the augmented feasible region S, which may then be called effective set of solution towards achieving the respective target levels of the objective goals. Then the feasible region of the decision variables appear as Se Sx#O and j= 1, 2 N 42. Solution Methodology of Phase 2 In phase 2, problem (2.1) of the attainment of goal objectives (to their desired levels) is considered. This problem, actually measures the effectiveness of the decision within the set of46 NEELAM MALHOTRA AND S. R. ARORA feasible solutions x* (e = 1, 2, .... J) of problem (2.2). The priority based GP model of the principal problem (2.1) can be presented in the form Find x(, 5, ..-»%,) SO as to Minimize Py (41 digg + Wa a) Minimize Pe ne dou * Mra int) , 23) Minimize Px (vinx dn * Wink Sn) subject to Ne = 1 Qi he ‘The method for solving problem (2.1) and thereby identifying the ideal solution is presented below, Ava given priority level &, the attainment problem of each goal of problem (2.1) is equivalent to the following problem : Evaluate = Minimize Py ing ding + Bint doy) subject to and d,,-d, = 0. TON tS Se fel n=l Then Z,, represents value of the objective function for eth solution at kth priority level.NONLINEAR INTEGER GOAL PROGRAMMING ar Find — Minimum Z,, ood According to the pre-emptive priority based GP procedure, the goals at the highest priority level (P1) are first considered for the achievement of their respective target levels. If solution found is unique, the optimality és reached as the goal at the lower priority level P, (k = 2, 3, .... K) can not be attained by sacrificing the degree of the achievement of the goals at the priority level P, If the solution is not unique, then goals at the second priority level (P2) are considered for their attainment with the tied solutions got at priority level (P1) and identifying the optimal solution Again if the solution found is not unique, the process is continued with the tied solutions got at the preceding priority level. Finally at the last priority level P, either unique solution or more than one solution can be found. Thus the algorithm to find ideal solution(s) is presented below. 43 The Algorithm Step 1 — Write the auxiliary problem and solve after linearlization using the technique given in section 3 Step 2 — Find the feasible region of the decision variable x,, using the convex combination of the extreme points obtained in step 1 Step 3 — Let k = 1 (where k represents the priority level under consideration and (k = 1, 2, ony RD. Step 4 — Minimize (Z,) eFh Qed Step 5 — If either one or both of the following conditions hold, go to step 7, otherwise go to step 6 (i) The solution is unique and (ii) k= K, highest priority reached (either unique or more than one optimal solution found), Step 6 — Update the feasible region of the respective decision variables. Set k = goto step 4. + Land Step 7 — Stop, as the optimality is reached. Example — When the constraints are quadratic, Find x(x.) so as to (P1) Minimize P,(D, + 2D) Minimize P,(Dj +2D3) subject to48 NEELAM MALHOTRA AND S. R. ARORA x 44xy $4 xy +35810. xjo%)20 and x),x) are integers D,,D; 20, D; -D? Auxiliary Problem is (P2): Minimize P\(d;) Minimize P,(d3) subject to 2 : ata td, xy txtd;-@=10 yo %y20, x1, ¥p are integers upper limit of x, is 2 and that of x, is 1. Let x yp +2y. and x= ys. where y= 0, 1, f= 1, 2,3, (P2) becomes equivalent to (P3) given by (P3) Minimize (d;) Minimize (d3) subject to on ae Yt 4y3 + dyyyy + 4y5 +d, ~dy y, +2) +y3+d,-d,=10 y20t, = 1,23NONLINEAR INTEGER GOAL PROGRAMMING 49 By linearization technique of section 3, (P3) is equivalent to solving. (P4) Minimize P,(d} +d, + d,) Minimize P,(d3) subject to YL + 4yy + Ay; H4U +d, —d) = 4 yy +29) +y3 +d, - dy = 10 Vy ty,-U+dy—dy=1 ~ yp -yp + 2U +d, ~dy=0 U=01,U=y y=0,1, i= 1, 2,3 G.d,20.d4 4 id, 1.234 Feasible solutions of (P4) are Op. ¥p 99) = {OO 1, © 1, 0 (1, 0, 0} Feasible solutions of (P3) are (1,5) = (@, 1, 2, 0} Feasible region of the decision variable are 4 =0,2 5 y=0,1 Solving primary problem Minimize Dj Dy atx, =0 Minimize D;=0 at x, = 0, 1 2 solution is (x),45) = {@, 0), (0, I} As there are two solutions to priority one, we go to priority two. Updated feasible region of the decision variables is50 NEELAM MALHOTRA AND S: R. ARORA. Optimal solution of (P2) is and x)=1 ACKNOWLEDGEMENT We are indebted to Professor M.C. Puri, Department of Mathematics, LT Delhi, for continuously encouraging us in our research work. REFERENCES ‘A. Chames and W. W. Cooper, Management Models and Industrial Applications of Linear Programming, Vols. | and 2, John Wiley and Sons, New York, 1961 Y. Iii, Management Goals and Accounting for Control, Chicao, Rand - McNally & Co, 1965. Ralph, E. Gomory, In : R. L. Graces and P. Wolfe (Eds), Recent Advances in Mathematical Programming, New York McGraw-Hill Book Company, (1963), 269-302, J. C. Pant and S. Shah, Opsearch 29 (1992) 297-304. J.P. Ignizio, J. Op. Res. Soc., 34 (1983) 539-42 J. P. Ignizio, Goal Programming and Extensions, Lexington Books, London 1976. LS. H. Kombluth and R. E. Steur., Euro. J. Op. Res., 8 (1981) 58-65. B. B. Pal and I. Basu, Optimization 35 (1995) 145-57.