MAT3007 Midterm - Nov 2019 Page 1 of 9

M IDTERM E XAM S OLUTION

MAT 3007
Nov 6, 2019

INSTRUCTIONS

a) Write ALL your answers in this exam paper.

b) One piece of note is allowed. No computer or calculator is allowed.

c) The exam time is 10:30am - 12:00pm.

d) There are 6 questions and 100 points in total. Except the true or false
questions, write down the reasonings for your answers.

In taking this examination, I acknowledge and accept the instructions.

NAME (signed)

NAME (printed)
MAT3007 Midterm - Nov 2019 Page 2 of 9

For grading use. Don’t write in this part

1 (18pts) 2 (20pts) 3 (16pts) 4 (22pts) 5 (14pts) 6 (10pts) Total

Points

Problem 1: True/False (18pts)

State whether each of the following statements is True or False. For each part, only your answer,
which should be one of True or False, will be graded. Explanations will not be read.

(a) For a linear optimization problem in the standard form, suppose it has a finite optimal solution,
then if one increases an objective coefficient (e.g., from 1 to 2), the optimal value will not
decrease. True

(b) In a linear optimization problem, two different basis must correspond to two different vertices.
False

(c) For linear optimization, if a primal problem is feasible, then the dual problem must be feasible
too. False

(d) Let x be an optimal solution to a linear optimization problem. Now we add another constraint.
If x satisfies the additional constraint, then it must still be optimal. True

(e) Let x and y be feasible solutions to a linear optimization problem and its dual, respectively.
Then if they satisfy all the complementarity conditions, they must both be optimal solutions.
True

(f) In the simplex method for linear optimization problem, degeneracy happens when there are
multiple optimal solutions. False
MAT3007 Midterm - Nov 2019 Page 3 of 9

Problem 2: Solve Linear Optimization Problem (20pts)

Use two-phase simplex method to solve the following linear optimization problem:

maximize x1 +2x2 +x3

s.t. x1 +x2 +x3 = 7
2x1 −2x2 +x3 ≥ 10
x1 , x2 , x3 ≥ 0.

Answer. We first convert it to the standard form (2pts):

minimize −x1 −2x2 −x3

s.t. x1 +x2 +x3 = 7
2x1 −2x2 +x3 −x4 = 10
x1 , x2 , x3 , x4 ≥ 0.

Then we construct the phase-one problem (2pts):

minimize x5 +x6
s.t. x1 +x2 +x3 +x5 = 7
2x1 −2x2 +x3 −x4 +x6 = 10
x1 , x2 , x3 , x4 , x5 , x6 ≥ 0.

We write down the initial tableau for the phase-one problem (the top row worths 2 points)

B -3 1 -2 1 0 0 -17
5 1 1 1 0 1 0 7
6 2 -2 1 -1 0 1 10

Then in the first iteration, we choose the number 2 to be the pivot element, the next table becomes
(1pt):

B 0 -2 -1/2 -1/2 0 3/2 -2

5 0 2 1/2 1/2 1 -1/2 2
1 1 -1 1/2 -1/2 0 1/2 5

We choose the second column to be the incoming column, and the number 2 to be the pivot element,
the next table becomes (1pt)

B 0 0 0 0 1 1 0
2 0 1 1/4 1/4 1/2 -1/4 1
1 1 0 3/4 -1/4 1/2 1/4 6

The optimal solution to the phase one problem is x1 = 6, x2 = 1, x3 = x4 = x5 = x6 = 0 and the

optimal value is 0 (4pts).
MAT3007 Midterm - Nov 2019 Page 4 of 9

Now we construct the initial tableau for the original problem (phase two problem). We need to
calculate the reduced cost for x3 and x4 . We have (note that the order of B is (2,1))

c̄3 = c3 − [c2 , c1 ]T A−1 T

B A3 = −1 − [−2, −1] [1/4, 3/4] = 1/4 (2pts)

c̄4 = c4 − [c2 , c1 ]T A−1 T

B A4 = 0 − [−2, −1] [1/4, −1/4] = 1/4 (2pts)

Therefore, the initial tableau for the phase two problem is

B 0 0 1/4 1/4 8
2 0 1 1/4 1/4 1
1 1 0 3/4 -1/4 6

The reduced costs are all non-negative, thus the solution is already optimal. Thus the optimal
solution to the original problem is x1 = 6, x2 = 1, x3 = 0, x4 = 0. And the optimal value is 8
(4pts).
MAT3007 Midterm - Nov 2019 Page 5 of 9

Problem 3: Duality and Complementarity Conditions (16pts)

Continue to consider the linear program in the previous question:

maximize x1 +2x2 +x3

s.t. x1 +x2 +x3 = 7
2x1 −2x2 +x3 ≥ 10
x1 , x2 , x3 ≥ 0.

(a) Write down its dual problem. (5pts)

Answer. The dual problem is (each error costs 1 point)

minimize 7y1 +10y2

s.t. y1 +2y2 ≥ 1
y1 −2y2 ≥ 2
y1 +y2 ≥ 1
y1 free, y2 ≤ 0.

(b) Write down the complementarity conditions. (6pts)

Answer. The complementarity conditions include (each wrong equation costs 2 points)

x1 · (y1 + 2y2 − 1) = 0
x2 · (y1 − 2y2 − 2) = 0
x3 · (y1 + y2 − 1) = 0
y2 · (2x1 − 2x2 + x3 − 10) = 0

(c) Use the complementarity conditions and the answer to the previous question to find out the dual
optimal solution. (5pts)

Answer. From the previous question, we know that the optimal solution is x1 = 6, x2 = 1,
x3 = 0. Thus by the complementarity condition, we must have at optimal solution y1 +2y2 −1 =
0 and y1 − 2y2 − 2 = 0. Solve these equations, we have y1 = 3/2, y2 = −1/4. (5pts)
If one solves problem 2 incorrectly but uses the right idea here, then one can get as much
as 3 points.
MAT3007 Midterm - Nov 2019 Page 6 of 9

Problem 4: Sensitivity Analysis (22pts)

Consider the following linear optimization problem:

minimize 2x1 +x2 +2x3 −3x4

subject to 8x1 −4x2 −x3 +3x4 ≤ 10
2x1 +3x2 +x3 −x4 ≤7
−2x2 −x3 +4x4 ≤ 12
x1 , x2 , x3 , x4 ≥0

The following table gives the final simplex tableau when solving the standard form of the above
problem:

B 12/5 0 7/5 0 0 1/5 4/5 11

5 10 0 1/2 0 1 1 -1/2 11
2 4/5 1 3/10 0 0 2/5 1/10 4
4 2/5 0 -1/10 1 0 1/5 3/10 5

(a) What is the optimal solution and the optimal value to the original problem? (4pts)

Answer. The optimal solution is x1 = 0, x2 = 4, x3 = 0, x4 = 5 (2pts), and the optimal value

is −11 (2pts).

(b) What is the optimal solution to the dual problem? (4pts)

Answer. The dual optimal solution is [0, −1/5, −4/5] (4pts).

(c) In what range can we change the second right hand side number 7 so that the current optimal
basis is still the optimal basis? (7pts)

Answer. Consider we change the current b to b + ∆b with ∆b = [0; λ; 0]. We need to ensure
that A−1
B (b + ∆b) ≥ 0. (2pts)
Thus we have
       
11 1 1 −1/2 0 11 1
A−1
B (b+∆b) =
 4 + 0 2/5 1/10 · λ  =  4 +λ  2/5  ≥ 0 (2pts) ,
5 0 1/5 3/10 0 5 1/5

which implies
λ ≥ −11, λ ≥ −10, λ ≥ −25.
Thus overall λ ≥ −10 (1pt), i.e., the range is [−3, ∞) (2pts).
MAT3007 Midterm - Nov 2019 Page 7 of 9

(d) In what range can we change the second objective coefficient 1 so that the current optimal basis
is still the optimal basis? (7pts)

Answer. Consider we change c to c + ∆c where ∆c = [0, λ, 0, 0] and study the range of λ

such that the reduced costs are still nonnegative. We look at the reduced costs under the new
objective coefficients (2pts). We have (the basis is {5, 2, 4})

c̄new
N = c̄N − ∆cTB A−1
B AN
 
10 1/2 1 −1/2
= [12/5, 7/5, 1/5, 4/5] − [0, λ, 0]T ·  4/5 3/10 2/5 1/10 
2/5 −1/10 1/5 3/10
= [12/5, 7/5, 1/5, 4/5] − λ · [4/5, 3/10, 2/5, 1/10] ≥ 0 (3pts)

Solving from the above, we have λ ≤ 3, λ ≤ 14/3, λ ≤ 1/2, λ ≤ 8. Thus overall λ ≤ 1/2 and
the range for the coefficient is (−∞, 3/2]. (2pts)
MAT3007 Midterm - Nov 2019 Page 8 of 9

Problem 5: Cutting Stock Problem (14pts)

Consider a factory that produces certain lengths of tubes. The raw materials are tubes of W meters
and there are I types of products each with length wi (i = 1, ..., I). The demand of product type i
is di . The objective is to use the minimal number of raw materials to meet the demand.
To solve this question, we consider a set of P cutting patterns (we consider the set of cutting pat-
terns as given). A cutting pattern p can cut a raw material into api number of product i. For
example, if W = 20, w1 = 5, w2 = 7, then there are 3 patterns which are {a11 = 4, a12 = 0},
{a21 = 2, a22 = 1} and {a31 = 1, a32 = 2} (other patterns are not efficient in using materials).
In the following, please write an optimization formulation to determine the best way of producing.
That is, to write an optimization problem to determine how many each patterns to use in order to
use the minimal number of materials to meet the demand.

Answer. Let xp ∈ Z be the decision variable indicating the number of pattern p used in the produc-
tion. (3pts)
The objective is then to minimize x1 + · · · + xP (3pts).
PP
The constraint is that for each type of product, we satisfy the demand, that is p=1 xp api ≥ di
(4pts).
Therefore, the optimization problem is (4pts):
PP
minimize p=1 xp
PP
subject to p=1 xp api ≥ di , ∀i = 1, ..., I
xp ≥ 0, xp ∈ Z

Remark. If one didn’t write the integer constraint, then 2 points will be deducted.
MAT3007 Midterm - Nov 2019 Page 9 of 9

Problem 6: Robust Linear Program (10pts)

Consider the following so-called robust linear program:

maximizex cT x
subject to aT x ≤ b + uT x; ∀u ∈ {v : Av ≤ w}
x≥0

where c ∈ Rn , a ∈ Rn , w ∈ Rk are given column vectors; b is a given scalar; and A ∈ Rk×n is

a given matrix. In this problem, u ∈ Rn is an uncertain state vector and varies in the hypercube
{v : Av ≤ w}. The problem is to find an x ≥ 0 such that aT x ≤ b + uT x for all possible u in the
hypercube and the objective function cT x is maximized. Reformulate the problem as a single linear
optimization problem with a finite number of constraints (currently it has infinite many constraints
indexed by u) and variables and without the state vector u appearing in the problem. (Hint: Apply
duality theory on uT x.)

Answer. For the constraint, basically it means that aT x ≤ b + minv :Av ≤w v T x. (2pts)
Now we apply the dual to the inside optimization problem

minimize xT v
subject to Av ≤ w

Let the dual variable be η. The dual problem is (3pts)

maximize wT η
subject to AT η = x
η≤0

Then the original problem can be written as (5pts)

maximizex,η cT x
subject to aT x ≤ b + wT η
AT η = x
x ≥ 0, η ≤ 0

Solving the above problem will give the optimal solution to the original problem.