ECON7250

Problem Set 5 Solutions

September 6, 2014

Problem 17.2

(a)
F (x, y, z) = x2 + 6xy + y 2 3yz + 4z 2 10x 5y 21z

Let Fx denote
0

F
x

, then
0

Fx
0

Fy
0

Fz

2x + 6y 10

6x + 2y 3z 5

= 3y + 8z 21.

Suppose (x , y , z ) is a critical point. Then it must be the case that Fx (x , y , z ) = Fy (x , y , z ) =

0
Fz (x , y , z ) = 0 and we get the following system of equations
0

2x + 6y

10

6x + 2y 3z

21,

3y + 8z

or equivalently

6
0

2
3

10

3 y = 5 .
8
z
21

We can solve the above using Gaussian elimination and show that (x , y , z ) = (2, 1, 3) is the only possible
solution and hence the unique critical point.
The Hessian matrix is

Fxx

Fxy

Fxz

D2 F = Fxy

Fxz

Fyy

=
Fyz
6

Fzz
0

Fyz

3 .
8

The the 2nd order LPM is 32 < 0. Since we have an even-order principal minor which is negative, the
Hessian is neither positive nor negative semidenite and hence must be indenate. Hence by Theorem 17.2,
1

the critical point (2, 1, 3) is neither a local min or max and hence must be a saddle point.
(b)


2
2
2
F (x, y, z) = x2 + 2y 2 + 3z 2 e(x +y +z ) .


2
2
2
Let A = e(x +y +z ) and B = x2 + 2y 2 + 3z 2 , such that F = AB.
0
0
0
0
Recalling that a ef (a) = f (a) ef (a) , we have Ax = 2xA, Ay = 2yA, and Az = 2zA. We also have

Bx = 2x, By = 4y and Bz = 6z . Then,

0

Fx
0

Fy
0

Fz

Since e > 0 such that e(x

+y 2 +z 2 )

= Ax B + ABx = 2xA (1 B)
= Ay B + ABy = 2yA (2 B)
= Az B + ABz = 2zA (3 B)
> 0,

= 2x 1 x2 + 2y 2 + 3z 2

=0

= 4y 2 x2 + 2y 2 + 3z 2

=0

2

=0

Fx = 0
Fy = 0
0

Fy = 0

= 6z 3 x2 + 2y 2 + 3z

= (x = 0) or x2 + 2y 2 + 3z 2 = 1


= (y = 0) or x2 + 2y 2 + 3z 2 = 2


= (z = 0) or x2 + 2y 2 + 3z 2 = 3

Since we need to satisfy one of 2 possible statements across 3 dierent rst order necessary condi



tions, there are 8 possible scenarios. However noticing that x2 + 2y 2 + 3z 2 = 1 , x2 + 2y 2 + 3z 2 = 2 and


x2 + 2y 2 + 3z 2 = 3 are mutually exclusive, we eliminate the 4 scenarios in which x2 + 2y 2 + 3z 2 = i = j
where i 6= j . We are therefore left with 4 dierent scenarios to consider for the critical points:
Scenario 1: x = y = z = 0. Hence (0, 0, 0) is one critical point.
Scenario 2: x = 0, y = 0 and x2 + 2y 2 + 3z 2 = 3. This implies that z 2 = 1 or equivalently, z = 1 or
z = 1. Hence (0, 0, 1) and (0, 0, 1) are also critical points.
Scenario 3: x = 0, x2 + 2y 2 + 3z 2 = 2 and z = 0. As above this implies that y = 1 or y = 1, hence
(0, 1, 0) and (0, 1, 0) are critical points.
Scenario 4: x2 + 2y 2 + 3z 2 = 2, y = 0 and z = 0, and (1, 0, 0) and (1, 0, 0) are critical points.
Let's consider whether each of the critical points are min, max or neither.

Hessian matrix1 would be given as



2A (1 B) 2x2 (2 B)

4xyA [3 B]

4xyA [3 B]



2A (2 B) 2y 2 (4 B)

4xzA [4 B]

4yzA [5 B]

D2 F =

4xzA [4 B]

4yzA [5 B]
.


2
2A (3 B) 2z (6 B)

Since A (0, 0, 0) = 1 and B (0, 0, 0) = 6, D2 F (0, 0, 0) =

0 4 0 is positive denite hence (0, 0, 0) is
0

local min.

Since A (1, 0, 0) = e1 > 0 and B (1, 0, 0) = 1, the Hessian is

4e1

2e1

4e1

and is

indenite hence (1, 0, 0) is a local saddle.

Repeating the procedure, we nd that (0, 1, 0) are saddles and (0, 0, 1) are local max.
2

Problem 17.5

The prot function is

= pQ (wx + ry) ,

where
= xa y b

Q (x, y)

is the production function and

wx + ry

is the cost of production.

1 Notice

that Bx = Ax/A, By = 2Ay/A and By = 3Az/A. This implies that for any number n,

A (N B)
x

A (N B)
y

A (N B)
z

Ax (n B) ABx = Ax ((n + 1) B) = 2xA ((n + 1) B)

Ay (n B) ABy = Ay ((n + 2) B) = 2yA ((n + 2) B)

Az (n B) ABz = Az ((n + 3) B) = 2zA ((n + 3) B) ,

or equivalently,

A (n B) = 2vA ([n + Nv ] B) ,
v
where Nv = 1, 2, 3 respectively when v = x, y, z .
0
Notice then that since Fv = 2vA (Nv B) we have

Fvv
= 2A (Nv B) + 2v




A (Nv B) = 2A (Nv B) 4v 2 A (2Nv B) = 2A (Nv B) 2v 2 (2Nv B)

v

and for w 6= v {x, y, z},

Fvw
= 2v

A (Nv B) = 2v (2wA ([Nv + Nw ] B)) = 4vwA ([Nv + Nw ] B) .

w

This allows us to ll in the Hessian matrix with minimal possibility of error

Notice that the rst order partial derivatives of the production functions are
a
Q
x
b
= Q.
y

= a xa1 y b =

= b xa y b1

Qx
Qy

The rst order necessary conditions for a critical point (x , y ) requires that
0

x (x , y )
=

= pQx (x , y ) w = 0

a
pQ (x , y )
wx

and
0

x (x , y )
=

= pQy (x , y ) r = 0

b
pQ (x , y )
ry

1.

Therefore
a
pQ (x , y )
wx

b
pQ (x , y )
ry
 a 1  b 
=
x .
w
r
=

= y

1=

To get the optimal input bundle, we substitute the above expression for y into x = 0 to get
0

a+b1

p (x )

 a 1b  b b
w

r
= x

  
 1
w 1b  r b a+b1
p
.
a
b

Substitute the above expression for x back into y =

a 1
w

b
r

x and we get

    
 1 1
w a r 1a a+b
y = p
.
a
b

We have found our optimal input bundle (x , y ) =

h

w 1b
a

r b
b

1
i a+b
1 h
, p

w a
a

r 1a
b


1
i a+b1

We now want to nd the values of the parameters p, r, w, a, b for which the critical point (x , y ) is a
global max.
Recall from Theorem 17.8 from Simon & Blume that if is a concave function on R2+ 2 and (x , y ) is
a critical point, then (x , y ) is the global max of on R2 . So we want to nd values of the parameters for
which is a concave function.3
2 R2 := (x, y) R2 : x 0, y 0 .
+

This is the appropriate domain for the production function (in this case) since from
an economic point of view the production function (and hence prot function) cannot take negative inputs into the production
process.
3 A shortcut would be to prove that Q is a concave function since the sum of 2 concave functions is concave and a linear
function (i.e. wx + ry is a linear function of x and y ) is concave and convex, such that is concave whenever Q is concave...
but anyway...

Further recall from Theorem 17.8 that a function is concave if and only if its Hessian matrix is negative
semidenite for all possible points on the domain , so what we precisely need to do is to nd the values of
the parameters for which the Hessian matrix is negative semidenite for all (x, y) . We rst obtain the 2nd
order partial derivatives of :
0

x
0

a
pQ r
x
0
a
= pQy r = pQ r
x
0

= pQx w =

xx

yy

xy

pQ
a 0
a
pQ pQ = a (a 1) 2
x x x2
x
pQ
b
b 0
pQ pQ = b (b 1) 2
y y y2
y
0
a
ab
pQ =
pQ,
x y
xy

and we can then write the Hessian matrix of the prot function as
"
H=

a (a 1) pQ
x2

ab pQ
xy

ab pQ
xy

b (b 1) pQ
y2

In order for the Hessian matrix to be negative semidenite, we require the rst order principal minors to be
non-positive and the 2nd order principal minor to be non-negative, for all possible values of (x, y) (0, 0)
and Q (x, y) 0 i.e.
pQ
x2
pQ
b (b 1) 2
y
2


(pQ)
ab (a 1) (b 1) a2 b2
x2 y 2
a (a 1)

0
0
0.

Notice that situations in which Q (x, y) = 0 or equivalently situations in which x = 0 or y = 0 imposes no

conditions at all on the possible values of the parameters. But since H must be negative semidenite for all
possible values of (x, y) R2+ , we must also consider situtations in which (x, y)  (0, 0) and p > 0.4 As an
aside, we note that if p = 0, then H is a zero matrix and hence is negative-semidenite by denition.
If p > 0 and(x, y)  0, then pQ > 0 and
5
a (a 1) pQ
x2 0 if and only if a (a 1) 0, and a (a 1) 0 if and only if 0 a 1. Hence

a (a 1) pQ
x2 0 if and only if 0 a 1.

Similarly, b (b 1) pQ
y 2 0 if and only if 0 b 1.
4 (x, y)  (0, 0) means that x > 0 and y > 0.
5 Claim: a (a 1) 0 if and only if 0 a 1.

Proof:

0 a 1 = a (a 1) 0 is obvious. We consider a (a 1) 0 = 0 a 1.
a (a 1) 0 implies that either a 0 or (a 1) 0 but not both. Suppose a 0 and (a 1) 0. But this implies that
1 a 0 which is impossible, hence the alternative that a 0 and a 1 is the only possible conclusion.

 Finally,
ab (a 1) (b 1) a2 b2

(pQ)2

x2 y 2

ab ((a 1) (b 1) ab)

(1 (a + b))

a+b1

(pQ)
> 0)
x2 y 2
0 (Since a 0 and b 0)

0 (Since

So if a + b 1, a [0, 1] and b [0, 1], then the prot function is concave and (x , y ) is the global
maximiser of .
3

Problem 17.9

(a) a2 + b2 2ab = (a b)2 0 for any possible values of a and b, since the square of any number must be
non-negative. Alternatively, show that since
a2 + b2 2ab =
"

and

"

#"

is positive-semidenite. Notice that a2 + b2 > 2ab whenever a 6= b.

(b) We simply have to demonstrate that

X

2xi xj (n 1)

n
X

x2i .

i=1

i<j

From (a), we found that 2xi xj x2i + x2j . Hence

X

2xi xj

i<j

x2i + x2j .

i<j

We read i<j x2i + x2j as the sum of all possible terms of x2i + x2j for which i < j. Notice that this implies
that j > 1 as for j = 1, there does not exist any i = 1, 2, ..., n such that i < j. Similarly, i < n as for i = n,
there does not exist j n such that j > i. Then,
P

2xi xj

i<j

x2i + x2j

i<j

x2i +

i<j

n1
X

x2j

i<j
n
X

x2i +

i=1 j=i+1

n1
X

j1
n X
X
j=2 i=1

(n i) x2i +

i=1

x2j

n
X

(j 1) x2j

j=2

(n 1) x21 +

n1
X

(n i) x2i +

i=2

n
X

n1
X

(j 1) x2i + (n 1) x2n

j=2

(n 1) x2i .

i=1

Notice that the inequality is strict if xi 6= xj for at least one i 6= j . That is, the inequality is strict so long
as the x0i s are not identical.
(c) From equation (12), we were given that
S
m
S
b

implying that

2S
m2

=2

Pn

i=1

x2i ,

2S
mb

=2

H=

n
X

mx2i + bxi xi yi

i=1

n
X

mx2i + b yi ,

i=1

Pn

i=1

xi and

Pn

x2i
Pi=1
n
2 i=1 xi

2S
b2

= 2n. The Hessian is given as

Pn

i=1 xi

2n

!
.
 P

The rst order LPM is |H1 | = 2 ni=1 x2i 0. The 2nd order LPM is |H| = 4 n i x2i ( i xi )2 0 by
our ndings from part (b), hence H is positive semidenite and S is a concave function. If the x0i s are not
all 0's and not all identical, then |H1 | > 0 and |H| > 0 and the Hessian is positive denite, in which case S
is strictly concave. In either case, (m , b ) is a global min of S , though in the latter case, it would be the
unique global minimiser.
P

Problem 18.9

The problem is
max

w.r.t. x,y,z

f (x, y, z) = x2 y 2 z 2 , s.t. h (x, y, z) := x2 + y 2 + z 2 = c2 .

The associated Lagrangian is

L (x, y, z, ) = f (x, y, z) h (x, y, z) c2 .

The gradient vector of the constraint function is

2x

5h = 2y
2z

and the constraint h (x, y, z) = x2 + y 2 + z 2 = c2 requires x > 0 or y > 0 or z > 0, there are no critical points
for h in the constraint set, and the constraint qualication is satised.
Setting the partial derivatives to 0, we obtain
L
= 2xy 2 z 2 2x = 0
x
L
= 2yx2 z 2 2y = 0
y
L
= 2zy 2 z 2 2z = 0
y
L
= h (x, y, z) c2 = 0

y2 z2 =

x2 z 2 =

x2 y 2 =

x2 + y 2 + z 2 = c2 .

From y 2 z 2 = = x2 z 2 , we obtain y 2 = x2 . From x2 z 2 = = x2 y 2 , we obtain z2 =y 2 . Hence it must be

2
2
2
the case that the set of points (x, y, z, ) such that x2 = y 2 = z 2 = c3 , and = c3 constitutes a set of
critical points for the Lagrangian.
However, it is important to note that this is not the only set of possible solutions.
Consider the set of points in the constraint set for which x = 0 or y = 0 or z = 0, and = 0. Notice that
at such points the partial derivatives of the Lagrangian are also zero. Hence the points (x, y, z) for which
at least one of the inputs equal 0 constitutes another set of possible solutions to our maximisation problem.
However, it should be obvious that at such points, the objective function is 0 hence these are the solutions
to the constrained minimisation problem.
q
2
Hence the points (x , y , z ) for which |x | = |y | = |z | = c3 are solutions to the maximisation
problem.


The maximum value of the objective function on the constraint x R3 : ||x|| = c set is obviously

f (x , y , z ) =

c2
3

3


=

x2 + y 2 + z 2
3

3
.

Furthermore, this results holds for all possible values of c2 (i.e. over all constrained maximisation problems
with the same objective functions and contraint functions, but with dierent values of c), this result must
hold for all (x, y, z) R3 .