Chapter 4: Solving LP Problems: Post-optimality Analysis

IE 312
Operations
Research II
Second Semester
(2/2020)

1
Post-optimality Analysis

2
Post-optimality Analysis

3
Post-optimality Analysis – Re-optimization
Re-optimization
• LP model in reality are very large
• Many variations of the basic model may be of interest for different scenarios
• So solve again → many time → again and again → applying simplex method from scratch

• Or…re-optimization!!!
• Make change to the final simplex tableau
• Re-use it as the initial tableau and initial basic solution

• If the solution is feasible → then simplex continue normally

• If the solution is infeasible → then use “dual simpolex method”

• Advantage: start with a solution that is normally much closer to the new optimal solution
• Only a few iterations or no iteration if the same optimality is applied

4
Post-optimality Analysis – Shadow Prices
Shadow prices
• LP can be interpreted as allocating resources to activities
• bi is the amounts of the resources to be allocated
• bi can be management’s tentative initial decision
• bi = how much of the resources make available for this activity

• However, bi can be increased with a strong reason

• So the economic interpretation of the model is very useful

• Shadow price (yi*) is the marginal value of the resources, or the marginal increase in the
objective function value with the increase of bi

• Initial bi

5
Post-optimality Analysis – Shadow Prices
Shadow prices
• If bi is changed by 1

• Optimal solution changes to (5/3,13/2) with Z = 37.5 when b2 change from 12 to 13

6
Post-optimality Analysis – Shadow Prices
Shadow prices
• The increase in total profit of 1.5 x $1,000 should be realized?
→ Yes, if there is a product with profit less than $1,500 under production

• b2 increases beyond 18 will not make any difference, the optimal stays at (0,9)
→ The set of basic variables change and the new shadow price is y2* = 0

• y1* = 0: constraint on resource 1 (x1 ≤ 4) is not binding → there is a surplus of this resource
• y2* = 3/2 and y3* = 1: constraints on resource 2 and 3 are binding
→ the limited supplies are used up (b2 = 12 and b3 = 18)

• Resource with positive shadow price is scarce goods

• Resource with zero shadow price is free goods

• If Z represent the total profit → cj is the unit profit of the activity

• yi* is the increase in profit with additional one unit of bi
• If a premium price must be paid in addition to the market value
• yi* is the maximum premium worth paying

7
Post-optimality Analysis – Sensitivity Analysis
Sensitivity analysis
• Certainty assumption on the values of model parameters aij, bi, cj
• Parameter’ values are normally estimated

• Sensitive parameters = those cannot be changed without changing the optimal solution
• They must be estimated with special care to minimize the risk of erroneous optimal solution

Sensitivity analysis - bi
• yi* > 0 then the optimal solution change if bi change
→ bi is sensitive parameters

• yi* = 0 then the optimal solution do not change if bi change (by some amount)
→ bi is not sensitive parameters

• bi must be monitored closely if they have positive shadow prices (especially with large value)

8
Post-optimality Analysis – Sensitivity Analysis
Sensitivity analysis - cj

• c1 = 3 can be between 0
and 7.5 without changing
solution from (2,6)

• c2 =5 can have value no

less than 2

• c1 and c2 are not sensitive

parameter

9
Post-optimality Analysis – Sensitivity Analysis
Sensitivity analysis - aij
• Check if the constraint is binding or not

• Example:
• x1 ≤ 4 is not binding
• Any change in a11 =1 and a12 = 0 will not change the optimal solution
• Not sensitive parameters

• 2x2 ≤ 12 and 3x1 + 2x2 ≤ 18 are binding

• Any change to a21 = 0, a22 = 2, a31 = 3, a32 = 2 are going to change the optimal solution
• Sensitive parameters

10
Post-optimality Analysis – MS Excel

11
Chapter 6: Duality Theory

IE 312
Operations
Research II
Second Semester
(2/2020)

12
The Essence of Duality Theory

• One of the most important discoveries in the early development of linear programming

• Every linear programming problem has associated with it another linear programming problem
called the dual

• Extremely useful in a variety of ways

13
The Essence of Duality Theory

• With the primal in maximization → the dual is in minimization

• The dual problem uses the same parameters as the primal but in different locations

• The coefficients in the objective function of the primal

• → the right hand side of the functional constraints of the dual

• The right hand sides of the functional constraints of the primal

• → the coefficients of the objective function of the dual

• The coefficients of the variable in the functional constraints of the primal

• → the coefficients in a functional constraint of the dual

14
The Essence of Duality Theory
Using matrix notation

15
The Essence of Duality Theory
Example

16
The Essence of Duality Theory
The primal – dual table
The general relationships
between the primal and dual
problems

• The parameters for a

functional constraints in
either problem → the
coefficients of a variable in
the other problem

• The coefficients in the

objective function of either
problem → the right
hand sides for the other
constraints

17
The origin of the dual problem
• At any iteration of the simplex method, the number in row 0 are

• For the coefficient of x1, x2, …, xn

• z = (z1, z2, …, zn)

• The vector that simplex method added to the vector of initial coefficients (-c) in the process
of reaching the current tableau

• y = (y1, y2, …, yn)

• The vector that simplex method added to the vector of initial coefficient xn+1, xn+2, …, xn+m (0
in row 0)

18
The origin of the dual problem
Wyndor example
• First equation

• W = 4y1 + 12y2 +18y3

• Second equation

• z1 = y1 + 3y3
• z2 = 2y2 + 2y3

• Subtract both sides by c1 = 3 and c2 = 5

• (z1 – c1) and (z2 – c2) can be interpreted as surplus variables

• Simplex seeks a set of basic variables, and the BF solution such that all coefficients are
nonnegative

19
The origin of the dual problem
Sense of W?
• Same with primal?

• The only feasible solutions for this new problem

are those that satisfy the condition for
optimality for the primal

• W → minimization problem

20
The origin of the dual problem
• Dual problem can be viewed as a restatement of the goal of the simplex method
→ to reach a solution for the primal problem that satisfies the optimality test

• Before reaching the goal → the corresponding y in row 0 must be infeasible for the dual

• After a goal is reached → y is optimal for the dual

• W is the optimal value of Z

21
The origin of the dual problem

Row 0 is partitioned into:

• the coefficient of the decision variables
• (x1, x2)

• the slack variables (x3, x4, x5)

• the right hand sides (Z)

• the value of dual variables (y1, y2, y3)

• the surplus variables of the dual problem

• (z1 - c1) and (z2 - c2)

22
The origin of the dual problem

• All values are in row 0 without new calculation

• In Iteration 0, dual variables are negative → infeasible

• In Iteration 2, dual variables are non-negative → feasible and optimal

23
The origin of the dual problem
Summary of Primal-Dual Relationship
• Weak duality property
• If x is a feasible solution of the primal problem
• and y is a feasible solution for the dual problem
• then cx ≤ yb

• (x1, x2) = (3, 3) → Z = cx = 24

• (y1, y2, y3) = (1, 1, 2) → W = yb = 52

• Strong duality property

• If x* is an optimal solution for the primal solution for the primal problem
• and y* is an optimal solution for the dual problem
• then cx* = y*b

• The maximum value of Z → the minimum value of W

24
The origin of the dual problem
Summary of Primal-Dual Relationship
• Complementary property
• At each iteration, the simplex method simultaneously identifies a CPF solution x for the
primal problem and identifies a complementary solution y for the dual problem
• where cx = yb

• If x is not optimal for primal → y is not feasible for dual

• (x1, x2) = (0, 6) → Z = cx = 30

• (y1, y2, y3) = (0, 5/2, 0) → W = yb = 30

• Complementary optimal solutions property

• At the final iteration, the simplex method simultaneously identifies an optimal solutions
x* for the primal problem and complementary optimal solution y* for the dual problem
• where cx* = y*b

• y* is the shadow price for the primal problem

• (x1, x2) = (2, 6) → Z = cx = 36

• (y1, y2, y3) = (0, 3/2, 1) → W = yb = 36

25
The origin of the dual problem
Summary of Primal-Dual Relationship
• Symmetry property
• For any primal problem and its dual problem, all relationships between them must be
symmetric because the dual of this dual problem is this primal problem

• Duality theorem
The following are the only possible relationships between the primal and dual problem

• If one problem has feasible solutions and bounded objective function

→ the other is the same
→ weak and strong duality are applicable

• If one problem has feasible solutions and unbounded objective function

→ the other is infeasible

• If one problem has no feasible solutions

→ the other is infeasible or unbounded

26
The origin of the dual problem
Applications
• Dual problem can be solved directly using simplex
• m>n
• Number of primal constraints (m) > number of primal variables (n) = number of dual
constraints (n)
• Substantial reduction in computation effort

• Information on solution quality

• If x is a given feasible solution
• y can be found by inspection of the dual problem

• If y is feasible → cx = yb → optimal

• If not → then find feasible yb (how?)

→ cx < yb
• If yb – cx is small → x is good solution

27
Chapter 8: The Dual Simplex Method

IE 312
Operations
Research II
Second Semester
(2/2020)

28
The Dual Simplex Method
Based on the duality theory
• Observe that:
• Solutions are primal feasible → the primal basic solution is feasible
• Solutions are dual feasible → the complementary dual basic solution is feasible
• The basic solution is optimal → the solutions are both primal feasible and dual feasible

• The dual simplex method = the mirror image of the simplexmethod.

• The simplex method deals with basic solutions in the primal problem
• that are primal feasible but not dual feasible.
• Moves to the optimal solution by trying to achieve dual feasibility
• The dual simplex method deals with basic solutions in the primal problem
• that are dual feasible but not primal feasible.
• Moves to the optimal solution by trying to achieve primal feasibility

• The dual simplex method → applied simplex method to the dual problem.

29
The Dual Simplex Method
Benefits
• The dual simplex method is very useful in certain situations.

• Usually easier to find initial basic solutions that are primal feasible than dual feasible.
BUT!!!
• It may require many artificial variables to construct an initial BF solution artificially.
SO!!!
• Possibly easier to begin with a dual feasible basic solution AND use the dual simplex
• May require fewer iterations → since no artificial variables.

• Use in conjunction with sensitivity analysis

• If the optimal solutions is no longer primal feasible
• BUT still satisfies the optimality test
• Apply the dual simplex method with the dual feasible basic solution.
• Lead to the new optimal solution much more quickly

• Useful for huge LP

• usually more efficient than the simplex method
• # rows vs # columns

30
The Dual Simplex Method
Comparison between two algorithms
• The rules are very similar to the simplex method.

• The differences
• the selection of the entering nonbasic variables
• The selection of the leaving basic variables
• The stopping criteria.

• To start the dual simplex method (max LP)

→ Need all nonnegative coefficients in Eq. (0)
→ The basic solution is dual feasible

31
The Dual Simplex Method
Summary of the Dual Simplex Method
Initialization:
• Convert functional constraints in ≥ form to ≤ form (multiply by -1)
• Introduce slack variables
• Find a basic solution such that the coefficients in Eq. (0) are zero for basic variables and
nonnegative for nonbasic variables (so the solution is optimal if it is feasible).
• Go to the feasibility test.

Feasibility test:
• Check if all basic variables are nonnegative.
• If yes, then the solution is feasible and optimal → stop.
• Otherwise, go to iteration.

Iteration:
• the leaving basic variable → the negative basic variable with largest absolute value.
• the entering basic variable → the nonbasic variable with coefficient in Eq. (0) reaches
zero first when the equation with the leaving basic variable is added to Eq. (0).
• Select the nonbasic variables with negative coefficients one & with the smallest
absolute ratio between the coefficients of the Eq. (0) and the equation.
• Obtain the new basic solution → Gaussian elimination
• Go to the feasibility test.
32
The Dual Simplex Method
Summary of the Dual Simplex Method

• Same as applying the simplex method to the complementary basic solutions in the dual
problem

• Determine the leaving basic variable → determine the entering basic variable
• The negative variable with the largest absolute value → the negative coefficient with the
largest absolute value in Eq. (0) of the dual problem

• Determining the entering basic variable → determine the leaving basic variable

• The coefficient in Eq. (0) that reaches zero first → the variable in the dual problem that
reaches zero first

• The stopping criteria are also complementary.

33
The Dual Simplex Method
An Example: Wyndor Glass Co.

• -ve RHS allowed → no artificial variables

• convert the functional constraints to ≤ form & introduce slack variables
• All the coefficients in Eq. (0) are nonnegative

Solution 1 = (y1, y2, y3, y4, y5) = (0, 0, 0, -3, -5) & Z = 0
• Not feasible & leaving basic variable = y5 & entering variable = y2
Solution 2 = (0, 5/2, 0, 0, -3) & Z = 30
• Still not feasible & leaving basic variable = y4 & entering variable = y3
Solution 3 = (0, 3/2, 1. 0. 0) & Z = 36
• feasible & optimal

The optimal of the primal is (x1, x2, x3, x4, x5) = (2, 6, 0, 0, 0)

34
The Dual Simplex Method

35
The Dual Simplex Method
With Sensitivity Analysis
• can quickly re-solve a problem with small changes in the original model.

• So the optimal is no longer primal feasible → one or more RHS are –ve

• But still satisfies the optimality test → no –ve coefficients in Row 0

• Can immediately apply the dual simplex method with the dual feasible basic solution.

36
The Dual Simplex Method
Example:
• Suppose that the problem did not include constraint y1 + 3y3 ≥ 3

• The optimal solution is (y1, y2, y3, y4, y5) = (0, 5, 2, 0, 0) & Z = 30

• Now suppose that sensitivity analysis requires y1 + 3y3 ≥ 3

• The original optimal solution is infeasible

• Find the new optimal solution → by add this constraint with the slack variable and continue to
Eq. 8.1

• Applying the dual simplex method leads to the new optimal solution in just one iteration.

37