Final Examination

Mathematical Programming with Applications to Economics

Total Score: 100

Note: All matrix multiplications are done by taking appropriate transposes, which is not
shown in notations. You may draw the figures in pencil but use a pen otherwise.

Part A
This part contains multiple choice questions. Each question may contain multiple correct
answers. Mark all the correct choices. Each question carries 4 marks.

1. Two linear programs (P1) and (P2) are given with the same set of variables and the
same set of constraints except that (P1) has all variables non-negative and (P2) has
all variables free. The dual of (P2) is unbounded. Then, which of the following are
possible?

a) The dual of (P1) is infeasible.

b) The dual of (P1) is unbounded. (Answer)
c) The dual of (P1) has a feasible solution. (Answer)
d) The dual of (P1) has an optimal solution.

2. If the simplex method cycles then which of the following are true for two consecutive
degenerate iterations?

a) The objective function value does not change. (Answer)

b) The values of variables do not change. (Answer)
c) The values of variables may change but the objective function value does not
change.
d) The set of basic variables change. (Answer)

3. The first phase of the two-phase simplex method terminated with an objective function
value z 6= 0 for a linear program (P). Then,

a) (P) is unbounded.
b) (P) is infeasible. (Answer)
c) (P) has an optimal solution.
d) Dual of (P) is either unbounded or infeasible. (Answer)

1
4. Consider the polyhedron P = {x ∈ Rn : Ax ≤ b}, where A is an m × n matrix and b
is a m × 1 matrix. Suppose P has an extreme point z. Define Az to be the set of rows
of A such that ai z = bi for every row i of Az . Then, which of the following are true.

a) m ≤ n.
b) m ≥ n. (Answer)
c) rank(Az ) = n. (Answer)
d) rank(Az ) = m.

5. Which of the following are true about a minimum cost spanning tree (MCST) of a
graph?

a) The minimum cost edge always belongs to an MCST. (Answer)

b) An MCST contains no cycles. (Answer)
c) There may be multiple paths between a pair of vertices in an MCST.
d) The number of edges in an MCST will change if the weights of edges are changed.

6. A relaxation of an integer program (in maximization form) must satisfy which of the
following requirements.

a) It must have the same objective function as the integer program.

b) Its feasible region must include the feasible region of the integer program. (An-
swer)
c) For every point in the feasible region of the integer program, the objective function
of the relaxed problem must have higher value than the objective function of the
integer program. (Answer - I will give full marks if you have not chosen this)
d) The optimal solution of the relaxed problem must coincide with the optimal so-
lution of the integer program.

7. Suppose A is an m × n matrix. Which of the following are convex sets?

a) Cone of A. (Answer)
b) Convex hull of columns of A. (Answer)
c) Union of cone of A and {x ∈ Rm : kxk ≤ 1}.
d) Intersection of cone of A and {x ∈ Rm : kxk ≤ 1}. (Answer)

8. A linear program (P) is solved using the two-phase simplex method (with smallest
subscript rule for choosing leaving and entering variables). Which of the following are
NOT possible in any iteration in the second phase of the simplex method?

2
 a) The coeffcients of the non-basic variables in the final row (corresponding to the
objective function) are all negative.
b) An entering variable can be increased by an arbitrarily large amount.
c) The second phase never terminates, i.e., cycles. (Answer)
d) We conclude that (P) is infeasible. (Answer)

9. In a weighted directed graph the shortest path from any node i to a node j is denoted
as s(i, j). Suppose s(i, j) + s(j, i) < 0 for some nodes i and j. Then which of the
following are true.

a) A potential exists in this graph.

b) No potential exists in this graph. (Answer)
c) A unique potential exists in this graph.
d) The set of potentials form a 45-degree line in this graph.

10. Let (IP) be an integer program (in maximization form) and (LP) be its linear program-
ming relaxation. Suppose (LP) has an integral solution. Then which of the following
are always true?

a) The optimal solution value of (LP) is strictly higher than the optimal solution
value of (IP).
b) The optimal solution value of (LP) is equal to the optimal solution value of (IP).
(Answer)
c) The optimal solution value of dual of (LP) is equal to the optimal solution value
of (IP). (Answer)
d) The dual of (LP) may be unbounded.

3
 Part B
1. Let G = (N, E) be a directed graph with weight w : E → Z, i.e., integer weights.
Assume that G contains no cycle of negative weight. Consider the following inequalities,
defining the non-negative potentials of G.

p(j) − p(i) ≤ w(i, j) ∀ (i, j) ∈ E (P)

p(i) ≥ 0 ∀ i ∈ N.

Show that every extreme point of (P) is integral. In other words, if weights of a directed
graph is integral, then extreme non-negative potentials are integral. (15 marks)
Answer: The polyhedra described by the potential inequalities can be written as
P = {p ∈ R#N : p(j) − p(i) ≤ w(i, j) ∀ (i, j) ∈ E, p(i) ≥ 0 ∀ i ∈ N }. In matrix form,
we can write this as P = {p ∈ R#N : Ap ≤ w, p ≥ 0}, where A is the constraint matrix
and w is a column of integers. We notice that:

• Every entry of A is 0, −1, or 1.

• Each row containts exactly two non-zero entries and both entries are of opposite
sign.

Hence, A is totally unimoduar by the result we proved in the class.

Now any extreme point z of P can be written as Az z = wz , where Az corresponds to
the rows of A that are tight. We know that rank(Az ) = n. Hence, Az is invertible.
So, z = wz A−1
z . Since A is TU, Az is integral. Since wz is integral, we get that z is
−1

integral.

2. Here is a dictionary of a linear program in the simplex method.

x2 = 2 − x 1 + x 4
x3 = 4x1 − 3x4
3
Z = 5 − 2x1 − x4 .
2
• Identify the basic and non-basic variables in this dictionary. (5 Marks)
Answer: x1 and x4 are non-basic variables. x2 and x3 are basic variables.
• Does this dictionary correspond to the optimal solution of the linear program?
Explain your answer. (5 marks)
Answer: Since the coefficients of non-basic variables in the objective function
row are negative, this corresponds to an optimal dictionary.

4
 • Find the optimal solution (values of variables and objective function) of this linear
program and its dual (x3 and x4 are slack variables). (10 marks)
Answer: The value of the optimal solution is z = 5, x1 = 0, x2 = 2, x3 = 0, x4 =
0. The dual will contain a variable corresponding to each slack variable. Let
these be y1 (corresponding to x4 ) and y2 (corresponding to x3 ). We can read
the optimal solution of the dual from the objective function row of the primal
optimal dictionary. More precisely, negative of coefficients of slack variables equal
the optimal value of dual variables. Hence, y1 = 32 , y2 = 0 is the optimal dual
solution. The optimal dual objective function value is equal to 5 by strong duality.
To get the same solution in detail, we can write the problem in terms of original
variables x1 and x2 as
3 7 3
Z = max 5 − 2x1 − [x2 + x1 − 2] = max 8 − x1 − x2
2 2 2
s.t. − x1 − x2 ≤ −2
−x1 + 3x2 ≤ 6
x1 , x2 ≥ 0.

Note that the slack variable x4 corresponds to the first constraint and x3 corre-
sponds to the second constraint. The dual of this linear program is

8 + min −2y1 + 6y2

7
s.t. − y 1 − y2 ≥ −
2
3
−y1 + 3y2 ≥ −
2
y1 , y2 ≥ 0.

Set y1 = 23 and y2 = 0. We see that this is a feasible solution. The objective

function value of dual for this feasible solution is 5. Hence, by strong duality, this
is also an optimal solution.

3. Write down the dual of the following linear programs: (10 marks)

max 2x1 − 3x2

s.t. x1 + 3x2 ≥ 6
−3x1 + x2 ≤ 2
2x1 + 4x2 = 3
x1 ≥ 0.

5
 Answer: After writing the primal in standard form, one can write the dual as
min −6y1 + 2y2 + 3y3
s.t. − y1 − 3y2 + 2y3 ≥ 2
−3y1 + y2 + 4y3 = −3
y1 , y2 ≥ 0.

min 2x1 − 3x2

s.t. x1 + 3x2 ≥ 6
−3x1 + x2 ≤ 2
2x1 + 4x2 = 3
x1 , x2 ≥ 0.

Answer: After writing the primal in standard form, one can write the dual as
max 6y1 − 2y2 + 3y3
s.t. y1 + 3y2 + 2y3 ≤ 2
3y1 − y2 + 4y3 ≤ −3
y1 , y2 ≥ 0.

4. Give an example of a linear program which is infeasible and whose dual is also infeasible.
(5 marks)
Answer: Here is a linear program which is infeasible (infeasible since adding the two
constraints give us 0 ≤ −1),
max 2x1 − x2
s.t.x1 − x2 ≤ −2
x2 − x 1 ≤ 1
x1 , x2 ≥ 0.

Dual of this linear program is:

min −2y1 + y2
s.t. y1 − y2 ≥ 2
−y1 + y2 ≥ −1
y1 , y2 ≥ 0.
This dual is infeasible, since adding the two constraints give us 0 ≥ 1.

6
5. Consider the following integer program (IP):

X
5
min c j xj
j=1
7 2 5 5 19 8
s.t. x1 − x2 + x3 − x4 + x5 = (IP)
4 3 2 12 6 3
xi ∈ Z + ∀ i ∈ {1, 2, 3, 4, 5}.

Show that the problem (IP-1) is a relaxation of (IP). (10 marks)

X
5
min c j xj
j=1
3 1 1 7 1 2
s.t. x1 + x2 + x3 + x4 + x5 = + w (IP-1)
4 3 2 12 6 3
w ∈ Z+
xi ∈ Z + ∀ i ∈ {1, 2, 3, 4, 5}.

Answer: Since objective functions of both the problems are same, we only need to
show that every feasible solution of (IP) is also a feasible solution of (IP-1). Let
x1 , . . . , x5 be a feasible solution of (IP). Then, we can write
7 2 5 5 19 8
x1 − x2 + x3 − x4 + x5 =
4 3 2 12 6 3
3 1 1 7 1 2
⇔ ( x1 + x2 + x3 + x4 + x5 ) + (x1 − x2 + 2x3 − x4 + 3x5 ) = 2 + .
4 3 2 12 6 3
Let w = 2 − (x1 − x2 + 2x3 − x4 + 3x5 ). Let w̄ = 43 x1 + 31 x2 + 12 x3 + 12
7
x4 + 16 x5 . We
first argue that w ≥ 0. Assume for contradiction w < 0. Since x1 , . . . , x5 are integers,
then 2 − (x1 − x2 + 2x3 − x4 + 3x5 ) ≤ −1. Hence x1 − x2 + 2x3 − x4 + 3x5 ≥ 3.
But x1 − x2 + 2x3 − x4 + 3x5 = 38 − w̄. Hence, w̄ ≤ − 31 . This is impossible since
x1 , . . . , x5 ≥ 0 implies that w̄ ≥ 0.
Now, since x1 , . . . , x5 are integers w is also a non-negative integer. Hence x1 , . . . , x5 , w ∈
Z+ is a feasible solution of (IP-1).