Chapter 4

Solving Linear Programming Problem: the Simplex Method

4.6

Adapting to other Model Forms

Equality Constraints (Big M Method) Max Z=3x1+5x2 x1 4 2x2 12 3x1+2x2 =18 x 1 0 , x 2 0

B.V.

R.H.S.

Negative Right-Hand Sides x1 - x2 -1 -x1 + x2 1 Functional Constraints in Form (Big M Method) Min Z=0.4x1+0.5x2 0.3x1+0.1x2 2.7 0.5x1+0.5x2 = 6 0.6x1+0.4x2 6 x 1 0 , x 2 0

B.V.

R.H.S.

Functional Constraints in Form (Two Phase Method) Min Z=0.4x1+0.5x2 0.3x1+0.1x2 2.7 0.5x1+0.5x2 = 6 0.6x1+0.4x2 6 x 1 0 , x 2 0 Phase 1: minimize Z = x4 + x6 Phase 2: minimize Z = 0.4x1+0.5x2

(until x4 = 0, x6 = 0) (with x4 = 0, x6 = 0)

B.V.

R.H.S.

No feasible Solution If the original problem has no feasible solutions, then either the Big M method or phase 1 of the two-phase method yields a final solution that has at least one artificial variable greater than zero. Otherwise, they all equal zero. Min Z=0.4x1+0.5x2 0.3x1+0.1x2 1.8 0.5x1+0.5x2 = 6 0.6x1+0.4x2 6 x 1 0 , x 2 0

B.V.

R.H.S.

Variables Allowed to Be Negative Variables with a Bound on the Negative Values Allowed Max Z=3x1+5x2 4 x1 2x2 12 3x1+2x2 18 x1 -10 , x2 0 Variables with a Bound on the Negative Values Allowed Max Z=3x1+5x2 x1 4 2x2 12 3x1+2x2 18 x1 is unrestricted in sign , x2 0
4

4.7 PostOptimality Analysis

Reoptimization In practical situations, LP models are very large with hundreds or thousands of functional constraints and decision variables. Therefore, after having found an optimal solution for one version of a linear programming model, we frequently must solve again (often many times) for the solution of a slightly different version of the model. One approach is simply to reapply the simplex method from scratch for each new version of the model. However, a much more efficient approach is to reoptimize. Reoptimization involves deducing how changes in the model get carried along to the final simplex tableau. The revised tableau and the optimal solution for the prior model are then used as the initial tableau and the initial basic solution for solving the new model. Shadow Prices The shadow price for resource i (denoted by yi*) measures the marginal value of this resource, i.e., the rate at which Z could be increased by (slightly) increasing the amount of this resource (bi) being made available. The simplex method identifies this shadow price by yi* = coefficient of the ith slack variable in row 0 of the final simplex tableau. Max Z=3x1+5x2 B.V. x1 x2 x3 x4 x5 R.H.S. Z 36 0 0 0 3/2 1 x1 4 2 x3 0 0 1 1/3 -1/3 2x2 12 6 0 1 0 1/2 0 x2 3x1+2x2 18 2 1 0 0 -1/3 1/3 x1 x1 0, x2 0 b1= 4, b2 = 12, b3 = 18. * y1 = 0 = shadow price for resource 1 y2* = 3/2 = shadow price for resource 2 y3* = 1 = shadow price for resource 3 b2 = 12 13 Z = 36 ? * y2 = 3/2 is the rate at which Z could be increased by increasing b2 slightly. binding constraints vs. non-binding constraints. Interpretation of Shadow Price: A positive shadow price of yi* for resource i means that the total profit Z can be increased by yi* by purchasing one more unit of this resource at its regular price. yi* : the maximum that would be worth paying theory foundation lies in duality theory
5

Sensitivity Analysis Maximize Z = c x Subject to Ax <= b model parameters : c, A, b (i.e., ci, bi, and aij) generally are just estimates

A main purpose of sensitivity analysis is to identify the sensitive parameters (i.e., those that can not be changed without changing the optimal solution). The sensitivity parameters are the parameters that need to be estimated with special care to minimize the risk of obtaining an erroneous optimal solution. For bi , shadow price If yi* > 0, then the optimal solution changes if bi is changed, so bi is a sensitive parameter. yi* = 0 implies that the optimal solution is not sensitive to at least small changes in bi. Consequently, if the value used for bi is an estimate of the amount of the resource that will be available (rather than a managerial decision), then the bi values that need to be monitored more closely are those with positive shadow price especially those with large show prices. For ci, consider slope of the objective function Z= c1 x1+5x2 the range of c1 is from 0 to 7.5 Z= 3x1+ c2x2 the range of c2 is from 2 to infinite Hence, neither c1 nor c2 is a sensitive parameter.

For aij, consider if the corresponding constraint is binding at the optimal solution. x1 4 is not a binding constraint, any sufficient small change in its coefficients (a11 = 1, a12 = 0) is not going to change the optimal solution, so these are not sensitive parameters. 2x2 12 and 3x1+2x2 18 are binding constraints, so changing any one of their coefficients (a21 = 0, a22 = 2, a31 = 3, a32 = 2) is going to change the optimal solution, and therefore they are sensitive parameters. Greater attention is given to performing sensitivity analysis on the bi and ci than on the aij. The aij values are sometimes called technological coefficient, so there may
6

be relatively little (or no) uncertainty about their final values. Parametric Linear Programming Parametric linear programming involves the systematic study of how the optimal solution changes as many of the parameters change simultaneously over some range. A more important application is the investigation of trade-offs in parameter values. For example, if the cj values represent the unit profits of the respective activities, it may be possible to increase some of the cj values at the expense of decreasing others by an appropriate shifting of personnel and equipment among activities. Similarly, if the bi values represent the amounts of the respective resources being made available, it may be possible to increase some of the bi values by agreeing to accept decreases in some of the others. Will (or will not) discuss later in this course.

4.8 Computer Implementation

Using LINDO.

4.9 The Interior-Point Approach to Solving LP Problems

Skipped

Exercises of Big Method and Two-Phase Method

1. Maximize Z=3x1+2x2+7x3 Subject to: -x1+x2 =10 2x1-x2+x310 x10 x20 x30 Maximize Z=3x1+2x2+4x3 Subject to: 2x1+x2+3x3=60 3x1+3x2+5x3120 x10 x20 x30

2.