sive Improvement 1 = inear Programming gramning is one ofthe most dffcul branches of ap rane mathematicians had bette remain pure mathematicians.” en ing Objectives nt gs a initial basic niora given mad improves. pienin successive stil optimal wsaecblained. improvement gueis used in linear anning, max-fow, ince | ens. The reader tbe familiar with the weg concepts by the is chapter: ssc near nar lied mathematics; the poorer —Eddsger Dijkstra 17.1 INTRODUCTION To ITERATIVE IMPROVEMENT The idea of iterative improvement is to formulate an initial solution for the given problem and to keep improving the initial solution til no further improve- ment is possible. Figure 17.1 illustrates this iterative improvement approach. Given problem [ Peper int ston | ee ee here aes Improve the solution x current solution as optimal Fig. 17.1 Iterative improvement wvement, different types of problems, namely h matching problem are To illustrate iterative impro’ rn liner programming, maximal flow ndbipeti graph matching Problem discussed in this chapter. Let us first discuss the basics ramming 17.2 LINEAR PROGRAMMING , Je ofthe iterative improvement Liner progam (Pl ang opin robles Alina ee a s oblem (LPP) can be formulated in mer domains sich rogramming problem (L! abe hare is mae patie. economics, and ies en a — programming is studied as a discipline 0 research (refer to Box 17-1),Linear programming isa mathematical technique for solving optimization problems objective function and constraints can be represented using linear ‘equations and ineg One can recollect from Chapter 11 that a solution to an optimization problem can be obtain, by finding the optimal value of a given objective function of a given optimization preg subjected to the given constraints. George Bernard Dantzig is widely regarded as the “ite srlinear programming’ (refer to Box 17.2). The objective function of the given optima problem and constraints in linear programming are expressed in the form ofa linear funtion 's linear function is of the following form: ay + a,x, +--+ @q%, =0. Here, one can ohne that all a’s whose power is | are called coefficients and all.x’s are called variables. The giv problem may have many feasible solutions, that is, any set of values that can fully sate constraints of the given problem. An optimal solution is the feasible solution that maxims or minimizes the objective function as per the given requirement. Advantages and Limitations of Linear Programming ‘The main advantages of linear programming are as follows: - Itis a flexible and scientific tool that can solve optimization problems effectively. . Linear programming can help us evaluate the alternatives and is highly Wig problems that are neither too small (ie., having smaller instances) or very very large instances). v Some of the limitations of linear programming areas follows: ; : J The objective function and all the constraints should be expressed as # liner fo 2. Alinear pro a gramme can have onl jecti ; most ofthe have multiple objectives, ly one objective function, whereas LEEEEY core serarddanniy George Bernard Dantzig and: Oregon, United States 4,""%8,00m in 1914 at Portland, Fulkerson mathematics at tov ae became a professor of nae ae i ee pr I, Dantig’s biggest contyic yan the TS? PE id after World War linear programmir ‘and solved Silene fr ecving Pe, at hedesigned the 49 cies at ‘hat tine In 1976.08 #2 Pigg 3 also contributed to the devajnre Simplex National Nel ree a ie Pent ofthe on scientists by the United Staserative Improvement and Linear Programming 3. All decision variables are assumed to be certain. This it si eee us is not true, as most of the real-world 4, All variables can have constant coefficients, 7 5. If the result obtained is a fraction, sometimes it has to aA) Toe yEll th coterie es ae 08 onde eae In the subsequent sections, let us discuss the formulation of LPPs and methods for solving them. 173 FORMULATION OF LPPS ~ Formulation of LPPs needs understanding of several terminologies, which have been discussed in the following subsections. Decision variables Decision variables or variables are unknown quantities that are required to be determined in order to solve the given problem. These variables are called decision variables, as the problem is to find the values that these variables can take to solve the problem. The objective function and constraints of a given optimization problem should be expressed as linear functions of decision variables. For example, an expected number of items of a given problem can be modelled as a variable X. A variable can have a fractional value (called a continuous variable) or only binary values. A variable that has only binary values are called binary variables. Mostly in linear programming, only continuous variables are encountered. Constraints | Constraints are the conditions or restrictions imposed on problems. These are modelled as linear equations. For example, the condition x> 0 is a constraint, as the variable x is forced to take values greater than zero only. Inequality Inequality means usage of operators *<’ or ‘>’ to model linear equations. For example, a constraint can be x; > 1. Here, the constraint is using the inequality >. Inequality provides more flexibility in linear programming. Objective function An objective function represents the goal of a given problem. Often in an optimization problem, the objective function is to maximize or minimize a factor such as profit, effort, or time. In linear programming, the objective function is also modelled as a linear function and is often determined based on the requirements of the problem. The general procedure for formulating an LPP is as follows: 1, Identify decision variables. a 2 Construct an objective function of a given problem using these decision variables. Express it in the form of a linear function s has Z= CX, + Cay + + + C,X, which can .d, In linear programming, optimization problems are always be maximized or minimize a : are alw associated with maximization or minimization of the objective function of the given problem, 3. Identify all the constra inequalities in the form of decision Variable of a linear programming pro! 4, Decision variables always take non-neg: of the linear programming, For example, * nts and express them in the form of @ linear equation, or express ariables. Constraints are the possible values that a lem can take ative values. Identify the non-negativity constraint > 0, x 2 0,x5 20.dint 8 + Iyga%o +°°°F Inn S Orn MeMpprertn ZO Here, a and b are the known coefficients that are associated with the given con The problem is to decide the unknown x’s and c’s associated with the objective ‘A vector x that satisfies the constraints is called a feasible solution. Linear progr said to be feasible if at least one solution is possible and infeasible if it has no feasible tion. A vector x is called an optimal solution if it both satisfies constraints and m j (or minimizes) an objective function. A feasible solution is said to be unbounded ifithy no optimal solutions. In short, this problem can be represented as follows: \ +; subject to the constraints Layx) sb Vi=1,2,3,...5m fl x20 CHROMEPET, CHEN MAIS O64. Vj =1,2,3,...,m / This maximization problem is said to be of canonical form. There exists a conision KB ing the terminology, as some consider this problem to be of the standard form. Let us stick term‘ canonical form’. Itcan be observed that in the canonical form, the objective functions be maximized and all the constraints are expressed in inequality forms that involve ‘< The following problems illustrate how LPPs are formulated. Example 17,1 manufacture these 90 and 80 due to th given in Table 17,1 the given constraint Let there be three items A, B, and C, The raw material rei Products are 7; and r. The respective daily allotment of 8 shortage of raw materials. The profits of manufacturing the Pt” Formulate an LPP such that the profit is maximized, taking i? s ofthe allotment of raw materials, inner ri needs to be maximized, the problem is of maximizait? | Tespectively, ies prea and Fepresent the decision variables for prods Table 17.1 are 2109. epee the Profits that can be gained. Profits for the pod © $100, %150, and %200, respectively. For example, three units opa = wand The mathematical formulation ofthis problem i given tie Maximize z= 100x, + 150x, + 20x; 3x, + 4x + 5x5 < 90 2x, + 6x, + 5x3 < 80 My Xy X32. This s the LPP formulation ofthe given problem. Once the formulation is done, the problem can be solved using the methods discussed in Sections 17.4 and 17.5, : Let us consider two nutrients N, and N;. A patient requires at least 9 and Sunits of M, and N;, respectively. The problem is to mix nutrients to get food F and Fs, according to the quantity shown in Table 17.2. Formulate an LPP for the given constraints 0 that food cost is minimized. Solution Let the required decision variables for food F; and F be represented as x; and x. The requirements of nutrients N; and N, are 9 and 8, respectively, The problem can be formulated as a minimization problem, as the focus is to minimize the cost of food by mixing "with appropriate units of nutrients. The objective function can be expressed as follows: Table 17.2 Formulation of LPP Minimize z = 7x, + 10x. Here z represents the total cost of the mixture hat needs to be minimized This is subjected to the constraints of costs, as follows: 3x, +2% 29 2x, + 4x28 aym20Ce nee much items as possible. Let the decisio *ip yy and x, represent the decision products A, B, C, and D, which Tepreseal that can be gained. Therefore, the objective would be to maximize z= Gr, + 2x; +345 + that the maximum profit can be achieved by Appropriate items, , nthe is subjected to the constraint aa the items should be less than the knapsack capacity. Hence, the constraints of are as follows: 2x, + 4, + 6x; + 8x4 < 15 1, X2,X3,X4>0 Hence, the final form of the LPP is as follows: Maximize 6x, + 2xy + 3x, 2x, + 4x) + 6x, 1 Xa Xx4>0 This is the LPP fo, can be solved + Sx, + 8x4 < 15 mulation of using the method: the given problem, Once the formulation s dome S discussed in Sections 17.4 and 17.5. 17.4 GRAPHICAL METHOD For SOLVING LPps The objective of aL shical n tnd all pointy of this feasible we feasible solutions, important for understanding the graphical method further: Convex set A region (or et 8) ix convex, if any two points of the region are connected by Aline sich that the fine also falls within Redundant constraint A conatraint that doow not affect the feasible region is called # redundant constraint, ‘The optinal solution point is called an extreme point or a corner point, In other words, the extreme point represents an optimal solution, This extreme point is also known as a corner PolnL oF vertex, Theve points represent the vertices of the fewsible region that is represented 8 polygon, Thus, the focus of the graphical method is to find a corner point of this region ‘hat can maximize or minimize the objective function effectively, Thus, the corner point procedure is a graphical technique for solving LPPs by finding the Somer point, The corer point procedure is given ax follows: Step 1; Consider the inequality constraints ax equality constraints, A line ean be drawn for ‘every constraint, Step 2: Determine the feasibility region. Reco! Set of all feasible solutions, Comers of ‘ the graph or by solving the constraint equations, Heise Galop of hs cM Step 3: Dotormi f ctive function, The max rmine the valucs of the objectivi . blem, function and its corresponding vertex provide the Ree ore { 2 ae : if itis a maximization problem, On the other hand, oe a ie coonmneneiaa ‘Associated with the minimum value of the objective function, Vertex aren is the solution of the problem, a RE . 0% LPP. The following ‘example illustrates the application of the corner mel! Hleot that a feasible region is a collection or the feasible region can be solved either from fi 7 C hots kl ait ore Sitar point Ais (0,4). (©) Then by substituting x, a in this equation, one gets following relation: | 32x, =24 =x, = 12 Therefore, the coordinate point (9,4) A Bis (12,0). ea 7 (@) Plot these points as a line in a Sraph as shown in Fig, 17.3, 2. Similarly, plot the line for the ‘second constraint equation: ; 6, + 2m <4 Be Fig. 17.3 Graph for first (a) Convert the inequality to equality, as folloy ws: 1 Ox + 2x, = 24 (b) Let us assume that 1 =0; therefore, 2x, = 24 a= 12 Therefore, the Coordinate point D is (0, 12), (c) Then let US assume that = 0; hence, 6x, = 24 x1=4 Therefore, the S00rdinate int B is (@ Plot these points ag tne 1g 0) Ne to obtain to a Sraphical solution, ay shown in Fig. 17.4, .This shows that x=x273 +. The points of the vertex are (3, 3) at comer C. Thus, the objective function is 2= Tx, +3x, (0, 0) = 0 at the point (0, 0) E(4, 0) = 7(4) + 3(0) = 28 at the point (4, 0) (3, 3) = 7(3) + 3(3) = 30 at the point (3, 3) A(O, 4) = 7(0) + 3(4) = 12 at the point (0, 4) } The points D(O, 12) and B(12, 0) yield values 36 and 84, respectively. However, as they ate not part of the feasible region, they are ignored. Therefore, one can Sere < Maximum profit is 30 at x, = 3 and x; = 3, that is, vertex C. Thus, the optimal Problem is obtained. el Solve the following LPP using the corner mel Maximize Z = 4x, + 9x, Subjected to the following constraints: 2+ 6n< 12 10x, + 5x. S15 he it lated in a manner sini a The coordinates using the aforementioned constraints are calcul i ws: that followed in Example 17-4 and are given #8 follo\ L4 =9/5=18 Substituting in equation (17.2), 10x, +5x, =15 10x, + 5(9/5) =15 10x, =15-9=6 of consis REawenets ee in Fig.” erm i Thus, the coordinates for point C are (0.6, Le saat Now, compute the profit using the objective function, that is, us points, as follows: (0,0): Z= 0 at origin (0, 0) (point O in the graph) (0,2):Z gg Oe eet (0.6, 1.8):2= 4x, + oy. 4x 0.6 +9x 1.8 = 18.6 (point i (15, 0):Z=4r, + 9x,= 4(1.5) + 9(0) = 6 (point E in the grap?) ‘The maximum profit is obtained when x; = 0.6; x) = 1.8. Examplei76 Minimize ue following LPP using the comer ™ Minimizez=4y, — 3x, Subjectto 24 —~ x >4 4x, +3x) < 2g 220‘Some exceptional cases in LPPs are illustrated in the following sub ons. Multiple optimal solutions A case of non-unique solution arses when the line is to one of the lines that bound the feasible region, as shown in Fig. 17.8. In this case, multiple optimal solutions are possible, Infeasible solutions There may bea situation where ‘no points satisfy all the constraints of the problem. Bs This leads to infeasible solutions, | Unbounded solutions When one ormore variables keep increasing without violating the constraints of the given problem, the objective function value would also be very large. This leads to unbounded solutions. The graphical method is a simple procedure but can be used if an LPP has only two variables, This is ‘unrealistic, as real-world problems can have multiple decision variables. Hence, the graphical method is not suitable for larger sets of problems. Let us now discuss a non-graphical method, called the simplex ‘method, that can be applied to all sets of LPPs. The Bee (8,0) simplex method is a very popular technique that is "118 station or non-optial soluon se forsolving LPP and was designed by Dantzig (refer to Box 17.2). Dg SIMPLEX METHOD The simplex method is a non-graphical procedure for solving canonical LPPs in an iterative pater: The procedure starts with an initial basic feasible solution. Then it finds re = “'Solution in the next iteration. Iterations are repeated till the optimal solution is .variables are non-negative. = : The procedure for solving LPPs using the simplex method is given as follows: Step 1: Formulate the problem as an LPP. ae Step 2: Convert any inequality constraint into an equality constraint of linear po by adding slack variables. mc . : Step 3: Calculate the initial basic feasible solution and construct an initial simplex E Step 4: Check for the optimality condition (optimality test). aia ' 4a: Compute z,~ c,, where z is the objective function and c is the coe decision variables that appear in the objective function 2. 4b: Ifz)~ c, > 0, then the solution obtained in Step 4 is optimal. Others optimal solution is yet to be obtained. sc ig a set of Ul Step S: Determine the incoming or entering variable to the basis. Basis is ® ; variables associated with the given problem. Basis or Base column isthe col all basic variables, Choose a column in ; ue Step 6: the objective function row that represents @ gr js called ‘auc, Inthe case of ati, the basic variables are chosen. This colu is oes column. Similarly, identify the pi i : Bee ; milatly, Pivot row, and this determines ai Divide the independent column and the elements of the pivot col a . ‘ful in ero, hen the problem is completed, Otherwise, choose the O¥ ie we his indicates the row that goes out of the bese Te Tow’, i it ice is i : mes aL there 's a tie, then a choice is made in favour OF nn, Step 7; is the intersection of the pivot row and the piv" tep 7; Determine the new Solution as Pain om Ta: Divide the Pivide the entire pivot pow by the pivot element to get ae “ot OV pivot Tb: clement by cs cnt OY * Old row ~ dex new pivot rove Here, & 8 ee Bodie asa the pivot column of the row that is being mani vr positive © and tevise the solution tll z,~ ¢, is either 2e7 Step 8; wee"Subject to the following constrains 4x, + Ty + x5= 12 2x t4yt Xp Xp, X3, X40 Construct the initial simplex table as shown in Table 17.4. Table 17.4 Simplex table 1 Ferree omen i am at th ving attention to its sign. It is ~7. ‘be observed that the pivot element the elements of the pivot column ie for the maximum number in Z row without gi can be choose this column as the pivot column. It can 7, ~ ctlculated by dividing the independent column and Zy/xn) as follows: 022154 ae 3=12/3=4 xq = 12/4 =3 Pivot cl ltt and determines the departing variables n is 3. Therefore, choose the x, row as the pivot row. ba S° (or basis) andthe variable that goes out ofthe base or basis isa variable that indicates the ratio of independent cohumn tHe Beceaks of Oe ‘The minimum of the Hence, the variable x enters Thus, the pivot rowSimilarly, the objective row i ~ 6). This is done by malipyig . row is. (= 6). This is done by x Sart ceo come (6). This doe ty tig ~075}-[4700]+7x{05 10025) [4700), Thus, it can be seen that 7x05-4=-05 7x1-7=0 7x0-0=0 7x 0.25 -0=1.75 i “ new ups Substituting x, and x, =7in the original equation, one gets Z= 21. The is shown as Table 17.6. Table 17.6 Simplex table 2 TPivot column negativ® Tt can be observed that the pivot column is chosen based one Sel Present in the preceding table, Which is -0.5. The pivot row can be ratio, as follows: 93-325 <1 92= 310.5 = 6(4x I+(7x0)-4=0 | (4x0)+(7x1)-7=0 (4x 0.4) + (7 x (0.2) -0=0.2 (4x -0.3) + (7x 0.4)-0 1.6. Substituting the values of x, and x, in the original objective function gives the value of 2=21.6. The new table is shown as Table 17.8. Table 17.8 Simplex table 3 syrrent table is negative. Itean be observed that none of the entries of the Z ae the optimal sotton ce, the simplex method terminates. Thegefore BP 21.6, with x, = 1.2 and x2 = ical forms. ; od for non-can\ : Let us discuss the application of the simplex pan Ban 'MIZATION PROBLEMS auhee the oben funn Sof tion tyPe js called canoni nasal the problems diseussed 2 of type. It ; ties are Ho, ized. It can be observed that ae I ever in reality, all LPPs are notin Sue ietimes the non-canonical form of an LPP may have constraints having ‘~’ or > than < sign. Therefore, if any such situations arise, then all these constraints ae requ be converted to canonical forms. Therefore, we start with a non-feasible solution and towards an optimal solution. ‘Some of the ways of tackling this problem by converting all the constraints tothe cana or standard form are as follows: Surplus variables A surplus variable is a positive variable that is subtracted to co an inequality of type ‘>’ to an equality of ‘=’ form. For example, consider the follo inequality equation: 2x, + 6x, > 10 - This can be converted to an equality equation as follows: 2x, + 6x, — 1 = 10 Here, the variable x, is called a surplus variable. It can be observed that itis a negatives and iss Asurplus variable has to be introduced as a zero coefficient in the 8 Artificial variables An artificial variable j i negativity " itisfy the non-negal ope aly at le is added to satisfy ; ' ion of traints having an ‘ <” sign, Sometimes, introduction?" values may not lead to a feasible solution, In ally, a simplex starts with a 2270 § ee fh creates an artificial variable, Here, A; is called an artificial ¥ ction, it ji pape a aerial D minimization orb is taken as +4 if it is a maximization problem Equality constraints constraint. For example, Sometimes, the constraints of the given problem *1*32=3 is an equality constraint. SuPPOS® #33 -xY to get th Maximize xf x4 x3 — xf XY — xt 2x5 - 2x3 $3 A 4f'-3g 4X9 S2 a aie sO. Once the aforementioned transformation rules are applied, a problem can be solved using the simplex procedure. Two-phase simplex method: Ithas been discussed earlier that surplus variables convert constraints of the form * 2 °to equality constraints, Sometimes, the constraints are unable to provide the basic variables Tequired for initiating a simplex algorithm. In such a case, another procedure called the two hase simplex method, is used, The phases of two-phase simplex method areas follows: Phase 1 Artificial variables are introduced to provi Minimum value of the objective function is zero, Phase 11 tn Phase I, the inital basi solution of Phase 1 wed, the org! Se OS pai is reused, and the normal simplex procedure is used to ee nate sa following numerical example illustrates the application lod, ide an initial basic feasible solution. If then Phase 1 is initiated. Exam, Ol — ; 17.8 Solve the following minimization LPP: Minimize 5x, + 6x,see! is associated | 3c hc arc nim in mand Sin Zn i aie hence, choose this column as a pivot column. The a eee : Itcan be grt (1000/520) 520, There fire’ choke tie row ofx;as a aa get the that the pivot element is 5, Divide the entire row by the pivot element “ = Now, x3 enters the base, The new row ig calculated as painiee, a is, [320 new Pivot row is computed as [1600/5 2/5 5/5 -1/5 oa 00.2}. The new row can be Computed as shown in Table 17.10. Table 17.19 ‘Computation of new row In the new table, the Sntering Variable ig x, and the leaving table is Table 17.peead Itcan be seen that there is no negative number in Z row variables x,,x>,x,andx,. Similarly, the z now is also updated. Therefore, negate the Z value and the optimal solution is Z= 1920, With the values of the variables being x, = 0 and x, = 320. 57 PRINCIPLE OF DUALITY Itis possible to convert a maximization problem toa minimization LPP. The original maximization problem is called a primal problem. The associated minimization Problem is called its dual. Similarly, if the primal LPP is a minimization problem, then its ‘dual LPP is a maximization problem. This principle provides extreme flexibility such that, if Necessary, it Ive it. ‘one can convert a problem to its dual form and can so ; The procedure for converting a primal LPP to a dual LPP iss follows: = : ds the variables of the given primal LPP. The number of variables will be eq umber of constraints of the given problem. Sf the pri a "jective ction of the dual problem will be writen based on the RHS of the primal Teen yw coefficients of The column coefficients of the primal problem will now become the ro . Primal problems. She oes ae 5 Mininiaton aie primal problem will become maximization in the dual probl ice Versa ox s : : and vice ve * Operator Cn and Yon dy. ed Example 17.9 Consider the following problem: Maximize z= 4x, + 6x, Subject to the following Constraints: 2 +3x < 10 41 +2, <30 2x, + 4x, <35 420Ay, + yt) 3yy +292 + 4y3 26 Yn YaIs20 This is the dual of the given problem. Consider the following problem: Minimize z= 2x, + 4x, Subject to the following constraints: ® x) + 6x,>4 7; 4427 | m2 Solution The given problem is called a primal and is of minimization type. Therefore, its ‘tual is of maximization type. Variables of the given problem can be given as follows: Primal = x x 2 Value 1 ene Dia Bead <2 b>2+ Tx = 10 where x;, x)> 0 and x; is unrestricted with respect to sign fame tm? gi is is of an asymmetric fr Solution The constraint involves an ‘=" sign. Therefore, this is of a Express 1 as x3 =x} —ax{’ , Now introduce this in the objective function and constant: following is the first constraint: 2x, 12 Temain as itis, as it does not involve the variable». The third constraint would be as follows: “4x, +322 +7005 — 2) =10 Now, the dual LPP can be expressed as follows: Maximize ~7y, + 12y, + 10y, Subject to the following constraints: ~29 +29, 4y, <1 Y1~ 82 +3y, 3-2 Wty, s2 Here, y, ant ste 0, andy, is unrestricted with regpect to Sgt17.2. There are three nutrients IN, Ne, and Ny, The need of a patientis atleast 14, 18, and 24 units of N;, Ny, Maximize z= 2hx, + 14 ‘and Ny, How to mix the following nutrients, given the angen He ‘cost considerations? ‘Subject to 8x, + 8x, $48 Ox, + 12, $64 X20 s ‘Solve the following LPP using Minimize z = 12x, + 8x2 Subject to 8x, + 2x, $23 9x, + 12,548 %20 20Xy, Xp XZO. “Find the dual of the following LPP and verify it: “Maximize 2 = 2x, + 4x2 Subject to x, + 2x, $15 2x, + xp $28 X20 d the dual ofthe following LPP and verity I: dimize 2 = 12x, + 8X) ct to Bx, + 2x, $23 Qxy#12K, $48 X%20 the following problem using the Ford-Fulkerson problem as a weig problem (discussed in Additional Solve the stable marriage git git git 1 Boy 0 Boy 1 Boy 0 Show the stable matching pairs.