College of Management, NYCU 2022

1 (10%) NYCU airline is considering the purchase of long-, medium-, and short-
range airplanes. The price would be $335 million for each long-range plane, $250
for each medium-range plane, and $175 million for each short-range plane. The
board has authorized a maximum of $7.5 billion (a billion is a thousand million)
for these purchases. It is estimated that the net annual profit would be $21 million
per long-range plane, $15 million per medium-range plane, and $11.5 million per
short-range plane. It is predicted that enough trained pilots will be available to
crew 30 new airplanes. If only short-range planes were purchased, the
maintenance facilities would be able to handle 40 new planes. However, each
medium-range plane is equivalent to 4/3 short-range planes, and each long-range
plane is equivalent to 5/3 short-range planes in terms of their use of the
maintenance facilities. Management wishes to know how many planes of each
type should be purchased to maximize profit.
(a) Formulate an IP model for this problem. (5%)
(b) Use the binary representation of the variables to reformulate the IP model in
part (a) as a BIP problem. (5%)
Solution:
(a) Let L, M, and S be the number of long-, medium-, and short-range planes to buy.
Max 21L + 15M + 11.5S (1%)
S.T. 33 L + 250M + 175S ≤ 7500 (1%)
L + M + S ≤ 30 (1%)
5/3L + 4/3M + S ≤ 40 (1%)
L, M, S ≥ 0, integers (1%)
!"## %#
(b) 𝐿 ≤ 𝑀𝑖𝑛 ' $$" , 30, "& , = 22.4, 2% ≤ 22.4 ≤ 2" (1%)
$
!"## %#
𝑀 ≤ 𝑀𝑖𝑛 ' '"# , 30, %& , = 30, 2% ≤ 30 ≤ 2"
$
!"## %#
𝑆 ≤ 𝑀𝑖𝑛 2 (!" , 30, ( 3 = 30, 2% ≤ 30 ≤ 2"
Thus, the binary representation of L, M, and S are:
𝐿 = 2# 𝑙# + 2( 𝑙( + 2' 𝑙' + 2$ 𝑙$ + 2% 𝑙% (1%)
𝑀 = 2# 𝑚# + 2( 𝑚( + 2' 𝑚' + 2$ 𝑚$ + 2% 𝑚%
𝑆 = 2# 𝑠# + 2( 𝑠( + 2' 𝑠' + 2$ 𝑠$ + 2% 𝑠%
Substitute L, M, and S with these binary representation, we have
Max 21∑%)*# 2) 𝑙) + 15∑%)*# 2) 𝑚) + 11.5∑%)*# 2) 𝑠) (1%)
S.T. 335∑%)*# 2) 𝑙) + 250∑%)*# 2) 𝑚) + 175∑%)*# 2) 𝑠) ≤ 7500 (1%)
∑%)*# 2) 𝑙) + ∑%)*# 2) 𝑚) + ∑%)*# 2) 𝑠) ≤ 30
5/3∑%)*# 2) 𝑙) + 4/3∑%)*# 2) 𝑚) + ∑%)*# 2) 𝑠) ≤ 40
𝑙) , 𝑚) , 𝑠) 𝑏𝑖𝑛𝑎𝑟𝑦, 𝑖 = 0,1,2,3,4 (1%)

1
College of Management, NYCU 2022

2 (15%) A company considers producing four new products. Managers need to

determine the type and amount of products to be manufactured in order to
maximize total profit. The following table shows the setup costs and marginal
profit of each product.
Product
1 2 3 4
Setup cost $50,000 $40,000 $70,000 $60,000
Marginal Profit $70 $60 $90 $80

Let x1, x2, x3, and x4 represents the production rate of each product. There are three
conditions that need to be satisfied.
(1) At most two products can be produced.
(2) Product 3 can be produced as long as product 1 or 2 is produced.
(3) Either 5x1 + 3x2 + 6x3 + 4x4 £ 6000 or 4x1 + 6x2 + 3x3 + 5x4 £ 6000 is satisfied.
Formulate the above problem as an integer programming problem.
Solution
yi = 1, if product i is made; yi = 0, otherwise (i = 1,…,4) (1%)
y5 = 1, if 5x1 + 3x2 + 6x3 + 4x4 ≤ 6000 is satisfied (1%)
y5 = 0, if 4x1 + 6x2 + 3x3 + 5x4 ≤ 6000 is satisfied (1%)

Max 70x1 + 60x2 + 90x3 + 80x4 – 50000y1 – 40000y2 – 70000y3 – 60000y4 (2%)
S.T. y1 + y2 + y3 + y4 ≤ 2 (1%)
y1 + y2 – y3 ≥ 0 (1%)
5x1 + 3x2 + 6x3 + 4x4 ≤ 6000 + M(1 – y5) (1%)
4x1 + 6x2 + 3x3 + 5x4 ≤ 60000 + My5 (1%)
x1 ≤ My1 (1%); x2 ≤ My2 (1%); x3 ≤ My3 (1%); x4 ≤ My4 (1%)
xi ≥ 0 and integers (i = 1,…,4) (1%)
yi binary (j = 1,…,5) (1%)

2
College of Management, NYCU 2022

3 (10%) Given the following partially finished branch-and-bound tree of a BIP, label
each of the statements as True or False, and then justify your answer. The numbers
associated with each subproblem are the LP relaxation solution. (total: 12%)

(b) The optimal objective value could be 70. (4%)

(a) Can (0, 1, 1, 0) with objective value 60 be a feasible solution to the original
BIP? (3%)
(b) Can (1, 1, 0, 0) with objective value 60 be a feasible solution to the original
BIP? (3%)
Solution
(a) False. Based on the solution of subproblem 2, the upper bound of optimal can’t be greater
than 68.
(b) False. The branch for x1 = 0 leads to subproblem 1. Thus, the upper bound of this branch is
56.
(c) True. The branch for x1 = 1 leads to subproblem 2. Thus, the upper bound of this branch is
68. It is possible to have a feasible solution with objective value equals to 60.

3
College of Management, NYCU 2022

4 (15%) Consider the following problem: (total: 15%)

Max f(x) = x3 + 30x – x6 – 2x4 – 3x2
(a) Given that the optimal solution is within the range of [0.0, 2.0], is the midpoint
(1.0) the optimal solution? (4%)
(b) Based on the concept of the bisection method, please derive the new lower
bound and upper bound of the optimal solution. (5%)
(c) Given the initial solution is 1, apply the Newton’s method for ONE iterations.
That is, identify the next solution. (6%)
Solution
(a) 𝑓 + (𝑥) = 3𝑥 ' + 30 − 6𝑥 " − 8𝑥 $ − 6𝑥 (2%)
𝑓 + (1) = 3 + 30 – 6 – 8 – 6 = 13, which is far away from 0. Thus, 1.0 is not optimal (2%)

(b) 𝑓 + (1) > 0. It means the optimal is greater than 1. (2%)

Thus, the new lower bound =1, and the new upper bound is 2. (3%)

(c) 𝑓 ++ (𝑥) = 6𝑥 − 30𝑥 % − 24𝑥 ' − 6 (2% for each equation)

3𝑥 ' + 30 − 6𝑥 " − 8𝑥 $ − 6𝑥
𝑥),( = 𝑥) −
6𝑥 − 30𝑥 % − 24𝑥 ' − 6
3 + 30 − 6 − 8 − 6 13 67
𝑥' = 1 − =1− =
6 − 30 − 24 − 6 −54 54

4
College of Management, NYCU 2022

5 (15%) Consider the following problem:

Max f(x) = 4x1x2 – 2x12 – x22

We try to solve this problem by using the gradient search algorithm. Given that
(1, 1) is the current (trial) solution, please answer the following questions.
Suppose the error tolerance is 0.01.
(a) What is the suggested searching direction for the next trial solution? (4%)
(b) Suppose the searching direction obtained from (a) is (0, 1). How far can we go
along this searching direction without downgrade the objective value? (4%)
(c) When the gradient method terminates, we obtain the solution (5, 6). Is this the
global optimal solution? Why? (7%)
Solution
(a) ∇𝑓(𝑥) = (4𝑥' − 4𝑥( , 4𝑥( − 2𝑥' ) (2%)
∇𝑓(1,1) = (0, 2) à This is the searching direction. (2%)
(b) (1,1) + 𝑡(0,1) = (1, 1 + 𝑡) (2%)
𝑓(1, 1 + 𝑡) = 4(1 + 𝑡) − 2 − (1 + 𝑡)' = −𝑡 ' + 2𝑡 + 1 (2%)
-.
-/
= −2𝑡 + 2 = 0 à 𝑡 = 1 (2%)
0! . 0! . 0! .
(c) 01"! = −4(1%), 01!!
= −2(1%), 01" 1!
= 4(1%)

(−4)(−2) − (4)' = −8 < 0 (1%)

The function is neither concave nor concave (1%)
Thus, (5,6) is not guarantee to be the global optimal. (2%)

5
College of Management, NYCU 2022

6 (20%) Consider the following nonlinear programming problem:

!!
Max
!" "#
S.T. 𝑥# − 𝑥$ ≤ 2
𝑥# ≥ 0, 𝑥$ ≥ 0
(a) Obtain the KKT conditions for this problem. (7%)
(b) Use the KKT conditions to check whether (𝑥# , 𝑥$ ) = (4, 2) is an optimal
solution. (6%)
(c) Given that 𝑢 ≠ 0 and 𝑥$ = 0, try to identify a feasible solution from these
KKT conditions. (7%)
Solution
(a) The KKT conditions are as following:
(
(1a) 1! ,(
− 𝑢 ≤ 0 (1%)
(
(2a) 𝑥( L1! ,( − 𝑢M = 0 (1%)
1
"
(1b) − (1! ,()! + 𝑢 ≤ 0 (1%)
1"
(2b) 𝑥' N− (1! ,()! + 𝑢O = 0 (1%)
(3) 𝑥( − 𝑥' ≤ 2 (1%)
(4) 𝑢[𝑥( − 𝑥' − 2] = 0 (1%)
𝑥( , 𝑥' , 𝑢 ≥ 0 (1%)
(b) Given (4, 2)
(
From (2a): ',( − 𝑢 = 0 → 𝑢 = 1/3 (2%)
% %
From (2b): − $! + 𝑢 = 0 → 𝑢 = 5 (2%)
The value of u is conflict. Thus, (4, 2) is not the global optimal solution (2%)
(c) Given that 𝑥' = 0, 𝑢 ≠ 0, assume 𝑥( ≠ 0. (1%)
(
From (2a) à #,( − 𝑢 = 0 → 𝑢 = 1 (2%)
From (4) à 𝑥( − 𝑥' = 2 → 𝑥( = 2 (2%)
It can be verified that (𝑥( = 2, 𝑥' = 0, 𝑢 = 1) satisfy all other KKT conditions. (2%)
Thus, (2, 0) is a feasible solution derived from KKT conditions.

6
College of Management, NYCU 2022

7 (15%) Consider the following quadratic programming problem:

Max 2𝑥# − 𝑥#$ + 4𝑥$ − 1.5𝑥$$
S.T.. 𝑥# + 𝑥$ ≤ 2
𝑥# ≥ 0, 𝑥$ ≥ 0.
Suppose that this problem is to be solved by the modified simplex method.
(a) Formulate the linear programming problem that is to be addressed explicitly,
and then identify the additional complementary constraint that is enforced
automatically by the algorithm. (5%)
(b) Apply one iteration of the modified simplex method to the problem as
formulated in part (a). You only need to identify the entering and leaving basic
variables. (5%)
(c) Suppose the complementary constraint is 𝑥# 𝑦# + 𝑥$ 𝑦$ + 𝑢# 𝑣# = 0. Given
that x1, x2, and u1 are the basic variables. Do we allow y1 be the entering basic
variable and x2 be the leaving basic variable in a specific iteration of the
modified simplex method? Why? (5%)
Solution
𝑥( 2 0
Given that 𝐱 = N𝑥' O, c = [2 4], Q = N0 3O, A = [1 1], b = [2]
2 0 𝑥( 1 𝑦( 2
(a) QX + ATu – y = cT à N0 3O N𝑥' O + N1O [𝑢] − N𝑦' O = N4O
𝑥(
Ax + v = b à [1 1] N𝑥' O + [𝑣 ] = [2]
𝑦(
xTy + uTv = 0 à [𝑥( 𝑥' ] N𝑦' O + [𝑢 ][𝑣] = 0
à
2x1 +u – y1 =2
3x2 + u – y2 = 4
x1 + x2 +v =2
x1y1 + x2y2 + uv =0
x1, x2, y1, y2, u, v ≥ 0
Thus, we have the following LP problem with complementary constraint.
Min Z1 + Z2 (1%)
S.T. 2x1 +u – y1 + Z1 = 2 (1%)
3x2 + u – y2 + Z2 = 4 (1%)
x1 + x2 +v = 2 (1%)
x1, x2, y1, y2, u, v, Z1, Z2 ≥ 0
complementary constraint: x1y1 + x2y2 + uv = 0 (1%)

7
College of Management, NYCU 2022

(b) Convert to proper form

Min 6 – 2x1 – 3x2 – 2u + y1 + y2 (1%)
S.T. 2x1 +u – y1 + Z1 =2
3x2 +u – y2 + Z2 = 4
x1 + x2 +v =2
x1, x2, y1, y2, u, v, Z1, Z2 ≥ 0
Pick x2 to enter (%2) and Z2 (2%)to leave
(c) Current basis = { x1, x2, u1}
If y1 enters and x2 leaves, basis becomes { x1, y1, u1}. (2%)
Both x1 and y1 are basic variables, which violates the complementary constraint. (2%)
This choice of entering and leaving basic variables is not allowed. (1%)