Solution to Practice Questions Midterm Exam 2 ORI 391: Linear Programming

1. Consider the linear programming model shown below.

Maximize zP = 2x1 + 0x2 + 1x3

subject to x1 + 2x2 + 3x3 ≥ 6
3x1 + 2x2 + x3 ≥ 6
x1 + x2 + x3 ≤ 4
x1 ≥ 0 x2 ≥ 0 x3 ≥ 0

a. Write the dual for this problem.

Minimize zD= 6p1 + 6p2 + 4p3

subject to +p1 + 3P2 + p3 ≥ 2
+ 2p1 + 2p2 + p3 ≥ 0
+ 3p1 + p2 + p3 ≥ 1
p1 ≤ 0 p2 ≤ 0 p3 ≥ 0

Note: There are other solutions if the primal is rewritten in another form.

b. The optimal solution to this problem has x2 and x3 basic and the first constraint is loose. Referring to
the dual problem described in part a, what dual variables are basic and which dual constraints are
loose or binding. To answer this question, you need to consider the dual slacks.
Solution
x2 basic implies 2nd dual constraint is tight.
x3 basic implies 3rd dual constraint is tight.
The first primal constraint is loose. This implies that p1 = 0.
The first dual constraint is loose (unless there is degeneracy).
The dual basic variables must be p2, p3 and slack variable for first dual constraint (ps1).

c. For the above problem, x1 = 1.5, x2 = 0, and x3 = 1.5 is a feasible solution. What does this indicate
about the objective function value for the optimal solution for the dual problem?
Solution
For this solution, the primal objective function value is zP = 1.5. By weak
duality, the objective function value for the optimal dual solution must be greater
than or equal to 1.5.

d. What is the complementary dual solution associated with the primal basic solution corresponding to
x1 = 1.5, x2 = 0, and x3 = 1.5?
Solution
With respect to the dual problem, the first and third dual constraints must be
tight. Since the last primal constraint is loose, we have p3 = 0.
The first and third constraints in terms of p1 and p2 are
p1 + 3p2 = 2
3p1+ p2 = 1
Solving gives p1 =5/8, p2= -7/8.

-1-
2. This problem relates to the problem shown below with variables x1,…,x6. This is the same as
problem 1 but the BFS is different.

c = [2, 0, 1, 0, 0, 1]

1 2 3 1 0 0  é6 ù
  ê ú
A = 3 2 1 0 1 0  b = ê6 ú
1 1 1 0 0 1  ê4ú
ë û

For the basis defined by the index set {1, 5, 3}, the basis inverse is

 1 / 2 0 3 / 2 
 4 
B-1 =  1 1
 1 / 2 0 1 / 2 
Note, the order of the basic variables is important.

a. Find the primal and dual solutions associated with the given basis. Identify each solution as feasible
or infeasible. If a solution is feasible, determine if it is optimal or nonoptimal.
Solution

 1 / 2 0 3 / 2  é6 ù é 3ù
 ê ú ê ú
Primal: xB = (x1, x5, x3) = B-1b =  1 1 4  ê6 ú = ê4 ú
 1 / 2 0 1 / 2  ê4ú ê1ú
ë û ë û
This is feasible but not optimal (must compute the reduced costs for x2, x4, x6).

 1 / 2 0 3 / 2 
 4 
Dual: p = cBTB-1= [2, 0, 1]  1 1
 1 / 2 0 1 / 2 

[p1 p2,p3] = [3/2, 0, 7/2]. This is not feasible.

b. A tableau for this problem appears below with a different basis. Find the primal and dual solutions
shown in and implied by the tableau, respectively. Identify each solution as feasible or infeasible. If
a solution is feasible, determine if it is optimal or nonoptimal.

-2-
  0 1 1
The current basis inverse is: B =  0 1 2 
1

 1 1 4 

Solution
xB = [x2, x3, x4]T = [2, 2, 4] from the right-hand side.
p= cBTB1=[ 0, 1, 2], and pTb = 2.
Because cBTxB = 2 (from tableau), both solutions are optimal (or from the tableau
we see that all the reduced costs are nonnegative).

4. Indicate whether the following questions are true or false. Justify your answer.

a. Consider the LP: min{cTx : Ax = b, x  0}. If the m  n matrix A has rank m and
the optimal BFS is nondegenerate, then the optimal dual solution is unique.
Solution
a. True. Let B be the optimal m  m basis. Complementary slackness tells us that out of thedualn
inequality constraints, the m constraints associated with the basic variables xB are binding at
optimality. This leads to the following systems of equations: pTB = cBT, which has a unique
solution, p*.

b. Assume n > m and x>=0 in the standard LP. Suppose we form and solve the phase-I problem
and further suppose that the optimal solution has no artificial variables in the basis. Then the
corresponding dual solution to the Phase-I model is degenerate.
Hint: Consider the dual of the phase-I problem.

Solution
b. True. If phase -I problem has no artificial variables in the optimal basis then, the optimal value of
phase-I is zero and also the corresponding cB=0. Therefore, the dual variables p=cBTB-1 = 0 which implies
that the dual constraints corresponding to primal variables x, pTA <= 0T, are all active. There are n such
constraints. So, at this point there are 'n' active constraints with 'm' variables. Hence this point is
degenerate.

-3-
c. There is no real advantage of the revised simplex algorithm over the tableau form of the
algorithm unless an implicit form of the basis inverse, like the product form of the inverse
or LU decomposition, is used to perform the calculations.

Solution
c. False. The tableau form of the simplex algorithm requires the full updating of the m  n A
matrix at each iteration to get B1. Even when using the revised simplex algorithm with
the explicit form of the inverse, only the m  m basis matrix B has to be updated at each
iteration. This is much less work.

-4-