INTERNATIONAL UNIVERSITY (IU) - VIETNAM NATIONAL UNIVERSITY - HCMC

FINAL EXAMINATION
June 2019
Duration: 120 minutes

SUBJECT: OPTIMIZATION 1
Deputy Head of Dept. of Mathematics: Lecturer:

Dr. Tran Vinh Linh Assoc. Prof. Nguyen Ngoc Hai

INSTRUCTIONS: Each student is allowed a scientific calculator and a

maximum of two double-sided sheets of reference material (size A4 or sim-
ilar), stapled together and marked with their name and ID. All other docu-
ments and electronic devices are forbidden.

Question 1 (20 marks) Show that although (0, 0) is a critical point of

f (x1 , x2 ) = x51 − x1 x62 , it is neither a local maximizer nor a local minimizer
of f (x1 , x2 ).

Question 2 (20 marks) Consider the function

f (x, y) = 2x3 − 2y 3 − 4xy + 5, (x, y) ∈ R2 .

Find all stationary points of this function and determine whether they are
local minimizers and maximizers. Does this function have a global minimizer
or a global maximizer?

Question 3 (20 marks) Consider the problem

minimize f (x1 , x2 ) = x21 + 2x22 , (x1 , x2 ) ∈ R2 .

(a) If the starting point is x0 = (2, 1), show that the sequence of points
generated by the steepest-descent algorithm is given by
 
1 2
xk = k
3 (−1)k

if an exact line search is used.

(b) Show that f (xk+1 ) = 19 f (xk ) and find limk→∞ f (xk ).

------continued on next page------

1
Question 4 (20 marks) Let a ∈ Rn , a 6= 0. Solve the problem

maximize aT x
subject to |x| = 1, x ∈ Rn .

Question 5 (20 marks) Given the quadratic problem

minimize (x1 − 2)2 + x22

subject to 2x1 + x2 − 2 ≤ 0.

(a) Determine and solve the dual problem.

(b) Use part (a) to determine the optimal solution of the primal problem.

*** END OF QUESTION PAPER ***

2
 SOLUTIONS
Subject: OPTIMIZATION 1
June 2019
Duration: 120 minutes

Question 1 As
∂f ∂f
= 5x41 − x62 , = −6x1 x52 ,
∂x1 ∂x2
find that ∇f (0, 0) = [0 0] and so (0, 0) is a critical point of f . However, in
any ball with center (0, 0) and radius r > 0, the points (r/3, 0) and (r/3, r/2)
lie in this ball but
   5    
r r r  r 5  r 5 r r
f ,0 = > 0 = f (0, 0) > − =f , .
3 3 3 3 2 3 2
Therefore (0, 0) is the saddle point of f .

Question 2 Solving the system

∂f
= 6x2 − 4y = 0
∂x
∂f
= −4x − 6y 2 = 0
∂y
we obtain two critical points (0, 0) and (−2/3, 2/3). The Hessian of f is
 
12x −4
F (x, y) =
−4 −12y
The matrix  
0 −4
F (0, 0) =
−4 0
has eigenvalues λ1 = −4 < 0, λ2 = 4 > 0. Thus (0, 0) is a saddle point. The
matrix    
2 2 −8 −4
F − , =
3 3 −4 −8
has eigenvalues λ1 = 4 and λ2 = 12. Thus (−2/3, 2/3) is a local max-
imizer. Moreover, since limx→∞ f (x, 0) = limx→∞ (2x3 + 5) = +∞ and
limx→−∞ f (x, 0) = limx→=∞ (2x3 + 5) = −∞. Hence f has no global maxi-
mizer and minimizer.

Question 3 f is a quadratic form since f (x) = 21 xT Qx, where

 
2 2 0
Q=∇ f = .
0 4

3
As eigenvalues of this matrix are λ1 = 2, λ2 = 4, ∇2 f is positive definite
everywhere. Setting
 
T 2x1
g(x) = ∇f (x) = Qx = and gk = g(xk )
4x2

we get the formula

gkT gk 4x21,k + 16x22,k 2x1,k

     
x1,k
xk+1 = xk − gk = − 2
gkT Qgk x2,k 8x1,k + 64x22,k 4x2,k
x21,k + 4x22,k
   
x1,k x1,k
= − 2 , k ∈ N.
x2,k x1,k + 8x22,k 2x2,k
 
1 2
As x0 = (2, 1), the formula xk = holds for k = 0. Suppose that
3k (−1)k  
1 2
the formula holds for k = m, that is xm = . For k = m + 1,
3m (−1)m

x21,m + 4x22,m x1,m

   
x1,m
xm+1 = − 2
x2,m x1,m + 8x22,m 2x2,m
4 · 9−m + 4 · 9−m
   
1 2 2
= m −
3 (−1)m 4 · 9−m + 8 · 9−m 2(−1)m
   
1 2 2 1 2
= m − · m
3 (−1)m 3 3 2(−1)m
   
1 6−4 1 2
= m+1 = m+1 .
3 3(−1)m − 4(−1)m 3 (−1)m+1

Thus the formula holds for k = m+1. By the induction principle, the formula
holds for every k ≥ 0.
(b) We have, for k ∈ N,
2 2
(−1)k
 
2 6
f (xk ) = +2 = .
3k 3k 9k

Hence
6 1 6 1
f (xk+1 ) = = · k = f (xk ) for all k
9k+1 9 9 9
6
and limk→∞ f (xk ) = limk→∞ 9k
= 0.

Question 4 Since the objective function is continuous and the constraint

set is closed and bounded, the given problem has at least a global optimal
solution. The constraint can be rewritten in the form h(x) = 0, where
h(x) = xT x − 1. Since ∇h(x) = 2xT 6= 0, all feasible points are regular.

4
Hence the optimal solution satisfies the KKT conditions. The Lagrangian
function is
L(x, λ) = aT x + λ xT x − 1 .

The KKT conditions are

aT + λ · 2xT = 0
xT x − 1 = 0.
Since a 6= 0, the former equation gives x = −a/2λ. Substituting this into
the latter equation we get λ2 = |a|2 /4, λ = ±|a|/2. Thus the system has two
solutions:
a |a| a |a|
x=− , λ= , and x = , λ=−
|a| 2 |a| 2
Since    
a a
f − = −|a|, f = |a|,
|a| |a|
a
the optimal solution of the given problem is x = |a| and the optimal value is
|a|.
Question 5 The Lagrangian function is
L(x1 , x2 , µ) = (x1 − 2)2 + x22 + µ(2x1 + x2 − 2), (x1 , x2 ) ∈ R2 , µ ≥ 0.
For each µ, the function L(·, µ) is strictly convex since its Hessian
 
2 2 0
∇x L =
0 2
is positive definite for all x = (x1 , x2 ) ∈ R2 . Thus minimum points of L(·, µ)
satisfy the system
2(x1 − 2) + 2µ = 0
2x2 + µ = 0.
Solving gives x1 = 2 − µ, x2 = −µ/2. Hence the dual function is
 µ 2 µ2  5  5
φ(µ) = L 2 − µ, − =µ + + µ 2 − µ = − µ2 + 2µ, µ ≥ 0.
2 4 2 4
We obtain the dual problem
5
maximize φ(µ) = − µ2 + 2µ
4
subject to µ ≥ 0.
φ0 (µ) = − 25 µ + 2 = 0 if and only if µ = 54 > 0. Thus the optimal
solution of the dual problem is µ = 54 and the corresponding optimal value
is max φ = φ(4/5) = 4/5. The solution of the primal problem is therefore
∗
 µ 6 2
x = 2 − µ, − = ,−
2 5 5
and the optimal value is 4/5.