Problem Sheet 9 with Solutions

GRA 6035 Mathematics

BI Norwegian Business School

2

Problems

1. Solve the optimization problem

max f (x, y) = x2 + y2 + y − 1 subject to x2 + y2 ≤ 1

What is the maximum value?

2. Solve max(1 − x2 − y2 ) subject to x ≥ 2 and y ≥ 3 by a direct argument and then

see what the Kuhn-Tucker conditions have to say about the problem.

3. Solve the following problem (assuming it has a solution):

(
2 2 x2 + y ≥ 2
min 4 ln(x + 2) + y subject to
x≥1

4. Mock Final Exam in GRA6035 12/2010, 4

We consider the following optimization problem: Maximize f (x, y, z) = xy + yz − xz
subject to the constraint x2 + y2 + z2 ≤ 1.
a) Write down the first order conditions for this problem, and solve the first order
conditions for x, y, z using matrix methods.
b) Solve the optimization problem. Make sure that you check the non-degenerate
constraint qualification, and also make sure that you show that the problem has a
solution.

5. Final Exam in GRA6035 10/12/2010, 4

We consider the function f (x, y, z) = xyz.
a) The function g is defined on the set D = {(x, y, z) : x > 0, y > 0, z > 0}, and it is
given by
1 1
g(x, y, z) = =
f (x, y, z) xyz
Is g a convex or concave function on D?
b) Maximize f (x, y, z) subject to x2 + y2 + z2 ≤ 1.

6. Final Exam in GRA6035 30/05/2011, Problem 4

We consider the function f , given by f (x, y) = xy ex+y , with domain of definition
D f = {(x, y) : (x + 1)2 + (y + 1)2 ≤ 1}.
a) Compute the Hessian of f . Is f a convex function? Is f a concave function?
b) Find the maximum and minimum values of f .
 3

Solutions

1 The Lagrangian is L (x, y, z, λ ) = x2 + y2 + y − 1 − λ (x2 + y2 ), and we write down

and solve the Kuhn-Tucker conditions: The first order conditions

Lx0 = 2x − λ · 2x = 0
Ly0 = 2y + 1 − λ · 2y = 0

together with the constraint x2 + y2 ≤ 1 and the complementary slackness conditions

λ ≥0 and λ = 0 if x2 + y2 < 1

From the first of the first order conditions, we see that x = 0 or λ = 1. Since λ = 1
contradicts the second of the first order conditions, we see that x = 0. From the
second of the first order conditions we have
1
2y(λ − 1) = 1 ⇒ y=
2(λ − 1)

This implies that

1
x 2 + y2 = ≤1
4(λ − 1)2
We first consider the case x2 + y2 = 1. In this case, we have 1 = 4(λ − 1)2 , or that
λ = ±1/2 + 1. Both solutions satisfy λ ≥ 0, and this gives two solution

3 1
(x, y; λ ) = (0, 1; ), (0, −1; )
2 2
of the Kuhn-Tucker conditions in case x2 + y2 = 1. Secondly, we consider the case
when x2 + y2 < 1, which gives λ = 0 by the complementary slackness condition.
Since x2 + y2 = 1/4 < 1, this gives the solution

1
(x, y; λ ) = (0, − ; 0)
2
of the Kuhn-Tucker conditions in case x2 +y2 < 1. The solutions of the Kuhn-Tucker
conditions give f (0, 1) = 1, f (0, −1) = −1 and f (0, −1/2) = −5/4, so the solu-
tion (x∗ , y∗ ; λ ∗ ) = (0, 1; 3/2) is the best candidate for maximum. The Lagrangian
L (x, y; λ ∗ ) = −x2 /2 − y2 /2 + y − 1 is clearly concave, so (x∗ , y∗ ) = (0, 1) is the
maximum point, with maximal value f ∗ = f (x∗ , y∗ ) = 1.
2 We see that f (x, y) = 1 − x2 − y2 will decrease as x increases and as y increases,
since x and y are positive. Therefore the maximum will occur at x = 2, y = 3, which
are the smallest possible values of x and y, and the maximum value is

f (2, 3) = 1 − 4 − 9 = −12
4

To write down the Kuhn-Tucker conditions, we write the problem in standard form
(
−x ≤ −2
max 1 − x2 − y2 subject to
−y ≤ −3

The Lagrangian is L (x, y, z, λ ) = 1 − x2 − y2 + λ1 x + λ2 y, and we write down and

solve the Kuhn-Tucker conditions: The first order conditions

Lx0 = −2x + λ1 = 0
Ly0 = −2y + λ2 = 0

together with the constraints x ≥ 2 and y ≥ 3 and the complementary slackness

conditions
λ1 , λ2 ≥ 0 and λ1 = 0 if x > 2, λ2 = 0 if y > 3
If x > 2, then λ1 = 0, and this gives x = 0, a contradiction. If y > 3, then λ2 = 0, and
this gives y = 0, a contradiction. The only solution to the Kuhn-Tucker conditions
is therefore x = 2, y = 3, λ1 = 4 and λ2 = 6. This point is the maximum since
L (x, y; 4, 6) = 1 − x2 − y2 + 4x + 6y is concave.
3 We first write the problem in standard form as
(
2 2 −x2 − y ≤ −2
max −4 ln(x + 2) − y subject to
−x ≤ −1

The Lagrangian is L (x, y, z, λ ) = −4 ln(x2 +2)−y2 +λ1 (x2 +y)+λ2 x, and we write
down and solve the Kuhn-Tucker conditions: The first order conditions
−8x
Lx0 = + λ1 · 2x + λ2 = 0
x2 + 2
Ly0 = −2y + λ1 = 0

together with the constraints x2 + y ≥ 2 and x ≥ 1 and the complementary slackness

conditions

λ1 , λ2 ≥ 0 and λ1 = 0 if x2 + y > 2, λ2 = 0 if x > 1

Since there are two constrains, there are four cases to consider. Case 1 is given by
x2 + y = 2, x = 1. This gives x = 1 and y = 1, and by the first order conditions
that λ1 = 2 and λ2 = 8/3 − 4 < 0. This contradicts the complementary slackness
condition, and there are no solutions in this case. Case 2 is the case x2 + y = 2 and
x > 1. In this case, λ2 = 0 and y = 2 − x2 . The first order conditons give λ1 = 2y and
−8x/(x2 + 2) + 4xy = 0. Inserting y = 2 − x2 into the last equation we get

4(2 − x2 )(2 + x2 ) − 8
 
2 8
x 4(2 − x ) − 2 = x· =0
x +2 x2 + 2
 5
√
Since x √> 1, this implies that√4(4 − x4 ) = 8 or x4 = 2, and this gives x = 4 2 and
y = 2 − 2, with λ1 = 2(2 − 2) and λ2 = 0. Since x > 1 and λ1 ≥ 0, this gives the
solution √ √ √
4
(x, y; λ1 , λ2 ) = ( 2, 2 2; 2(2 − 2), 0)
of the Kuhn-Tucker conditions in Case 2. Case 3 is the case x2 + y > 2 and x = 1.
This gives λ1 = 0 by the complementary slackness conditons, and then y = 0 by
the first order conditions. But then x2 + y = 1, and this contradicts x2 + y > 2. Case
4 is the case x2 + y > 2 and x > 1. This gives λ1 = λ2 = 0 by the complementary
slackness conditions, this gives x = y = 0 by the first order conditions. Again, this
contradicts the constraint x2 + y > 2. We conclude that there is only one solution
to the Kuhn-Tucker conditions. We have assumed that the problem √ has
√ a solution.
It follows that this solution must occur at the point (x, y) = ( 4 2, 2 − 2) or at an
admissible point where the NDCQ is not satisfied. In case 1, we see that NDCQ is
given by  
−2x −1
rk =2
−1 0
Since the determinant is −1, the rank is two at all points, and NDCQ holds. Simi-
larly, we see that in case 2 and case 3 that NDCQ holds, since each row contains an
entry that cannot √be zero.√ In case 4, there is no NDCQ. We can therefore conclude
that (x∗ , y∗ ) = ( 4 2, 2 − 2) is the minimum.
4 Mock Final Exam in GRA6035 12/2010, Problem 4
See handwritten solution on the course page of GRA 6035 Mathematics 2010/11.
5 Final Exam in GRA6035 10/12/2010, Problem 4
The Hessian of f is indefinite for all (x, y, z) 6= (0, 0, 0) since it is given by
 
0zy
f 00 =  z 0 x
yx0

and has principal minors −z2 , −y2 , −x2 of order two. Hence f is neither convex nor
concave. We compute the Hessian of g, and find
2 1 1
2 xy xz
1  x1 2 1
g00 = y2
xyz xy yz 

1 1 2
xz yz z2

Hence the leading principal minors are

1 2 1 3 1 4
D1 = > 0, D2 = > 0, D3 = >0
xyz x2 (xyz)2 (xy)2 (xyz)3 (xyz)2

This means that g is convex.

The set {(x, y, z) : x2 + y2 + z2 ≤ 1} is closed and bounded, so the problem has
solutions by the extreme value theorem. The NDCQ is satisfied, since the rank of
6

2x 2y 2z = 1 when x2 + y2 + z2 = 1. We form the Lagrangian

L = xyz − λ (x2 + y2 + z2 − 1)

and solve the Kuhn-Tucker conditions, consisting of the first order conditions

Lx0 = yz − λ · 2x = 0
Ly0 = xz − λ · 2y = 0
Lz0 = xy − λ · 2z = 0

together with one of the following conditions: i) x2 + y2 + z2 = 1 and λ ≥ 0 or

ii) x2 + y2 + z2 < 1 and λ = 0. We first solve the equations/inequalities in case i):
If x = 0, then we see that y = 0 or z = 0 from the first equation, and we get the
solutions (x, y, z; λ ) = (0, 0, ±1; 0), (0, ±1, 0; 0). If x 6= 0, we get 2λ = yz/x and the
remaining first order conditions give (x2 − y2 )z = 0 and (x2 − z2 )y = 0. If y = 0, we
get the solution (±1, 0, 0; 0). Otherwise, we get x2 = y2 = z2 , hence
 
1 1 1 1
(x, y, z; λ ) = ± √ , ± √ , ± √ ; ± √
3 3 3 2 3
The condition that λ ≥ 0 give that either all three coordinates (x, y, z) are positive, or
that one is positive and two are negative. In total, we obtain four different solutions.
We note that f (x, y, z) = 3√1 3 for each of these four solutions, while f (x, y, z) = 0 for
either of the first three solutions. Finally, we consider case ii), where λ = 0. This
gives xy = xz = yz = 0, and we obtain

(x, y, z; λ ) = (a, 0, 0; 0), (0, a, 0; 0), (0, 0, a; 0)

The condition that x2 +y2 +z2 < 1 give a2 ≤ 1 or a ∈ (−1, 1). For all these solutions,
we get f (x, y, z) = 0. We can therefore
√ conclude
√ that√the solution to the optimiza-
tion problem is (x∗ , y∗ , z∗ ) = (±1/ 3, ±1/ 3, ±1/ 3) with an even number of
variables with negative values. The maximum value is f = 3√1 3 .
6 Final Exam in GRA6035 30/05/2011, Problem 4

a) We compute the Hessian of f , and find

 
00 x+y (x + 2)y (x + 1)(y + 1)
f =e
(x + 1)(y + 1) x(y + 2)

The principal minors are

∆1 = ex+y (x + 2)y, ∆1 = ex+y x(y + 2), D2 = (ex+y )2 (1 − (x + 1)2 − (y + 1)2 )

 7

Since (x + 1)2 + (y + 1)2 ≤ 1, D f is a ball with center in (−1, −1) and radius
r = 1, and it follows that x, y < 0 and x + 2, y + 2 ≥ 0, and therefore ∆1 ≤ 0 and
D2 ≥ 0. This means that f is concave, but not convex.
b) Since D f is closed and bounded, f has maximum and minimum values. We com-
pute the stationary points of f : We have

fx0 = (x + 1)yex+y = 0, fy0 = x(y + 1)ex+y = 0

and (x, y) = (0, 0) and (x, y) = (−1, −1) are the solutions. Hence there is only
one stationary point (x, y) = (−1, −1) in D f , and the f (−1, −1) = e−2 is the
maximum value of f since f is concave. The minimum value most occur for
(x, y) on the boundary of D f . We see that f (x, y) ≥ 0 for all (x, y) ∈ D f while
f (−1, 0) = f (0, −1) = 0. Hence f(−1, 0) = f(0, −1) = 0 is the minimum value
of f .