1 Concave and Convex Functions

Lecture notes based on
Foundations of Mathematical Economics
c 2001 Michael Carter
All rights reserved
Concave and convex functions

1 Concave and convex functions
1.1 Denition
: is concave if for every x1 , x2 in
(x1 + (1 )x2 ) (x1 ) + (1 ) (x2 ) for every 0 1
It is strictly concave if the inequality is strict, and convex if the
inequality is reversed.
Equivalently, a function : is concave if and only if
hypo = {(x, ) : (x), x } is convex (Proposition 3.7, Exercise 3.125)
(x) (x0 ) + (x x0 ) for every x, x0 (Exercise 4.67)
: is locally concave at x0 if there exists a convex neighborhood of x0 such that for every x1 , x2
(x1 + (1 )x2 ) (x1 ) + (1 ) (x2 ) for every 0 1
is concave if and only if it is locally concave at every x .
1.2 Examples
2 , are convex.
2 , log are concave.
Cobb-Douglas (x) = 11 22 . . . is concave if
CES
1.
is concave if 1 and convex otherwise
The prot function (p) = maxy is convex.
The cost function (w, ) = minx () is concave in w

(
)1/
(x) = 1 1 + 2 2 + . . .

All rights reserved
1.3 Properties
If and are concave then
is convex (Exercise 3.124)
1/ is convex if > 0 (Exercise 3.135)
1/ is concave if < 0 (Exercise 3.135)
1 is convex (Example 3.45)
+ is concave (Exercise 3.131)
is concave for every 0 (Exercise 3.131)
is concave if is increasing (Exercise 3.133)
log concave (Example 3.51)
is continuous on the interior of its domain (Corollary 3.8.1)
is dierentiable almost everywhere (Remark 4.14)
1.4 Identication
Plotting the function (e.g. Mathematica)
Apply properties to combinations of known functions (Examples 3.73
and 3.74)
Hessian matrix (Proposition 4.1)
{
}
{
}
convex
nonnegative
at x (x) is
denite
concave
nonpositive
{
}
{
}
positive
convex
denite.
is strictly locally
at x if (x) is
negative
concave
is locally
The deniteness of the Hessian can be assessed by

eigenvalues (Exercise 3.96)
determinantal tests (Simon & Blume 381-386, Varian 475-477)

All rights reserved
2 Quasiconcavity
2.1 Denition
is quasiconcave if
(x1 +(1)x2 ) min{ (x1 ), (x2 )} for every x1 , x2 and 0 1
is quasiconvex if
(x1 +(1)x2 ) max{ (x1 ), (x2 )} for every x1 , x2 and 0 1
Equivalently, is quasiconcave if and only if every upper contour set
is convex, that is () = { x : (x) } is convex for every
. Similarly, is quasiconvex if and only if every lower contour
set is convex, that is () = { x : (x) } is convex for every
.
2.2 Examples
Every concave function is quasiconcave.
3 is quasiconcave and quasiconvex.
Cobb-Douglas (x) = 11 22 . . . is quasiconcave
CES (1 1 + 2 2 + . . . )
1/
is quasiconcave if 1
The utility function (x) is quasiconcave if preferences are convex.

The indirect utility function (p, ) = maxx (x) is quasiconvex in p.
2.3 Properties
If is quasiconcave then
is quasiconvex (Exercise 3.143)
is quasiconcave if is increasing (Exercise 3.148)
is quasiconcave if is quasiconcave (Exercise 3.153)
+ is ????? (Example 3.57)
is concave if > 0 and homogenous of degree 1 (Proposition
3.12)
3
All rights reserved

2.4 Identication
Plot contours
Apply properties to combinations of known functions (Exercises 3.148

and 3.151 to 3.153)
Bordered Hessian matrix (Simon & Blume 386-393)
3 Inner product
Given two vectors a, x , the angle between them is given by
a
cos =
a x
a x
where
a x = 1 1 + 2 2 + + =
and x =
a x
=1
The orthogonal hyperplane a x = 0 divides the space into two halfspaces, containing respectively those vectors that make acute and obtuse
angles with a.
a
a x > 0
a x < 0
a x = 0
The gradient of a functional : at x

(x ) = (
,
,...,
)
1 2

4
All rights reserved

points in the direction of steepest ascent. For any x, the inner product

(x )( )
(x ) (x x ) =
=1

measures the angle between (x ) and (x x ), and the orthogonal

hyperplane divides the space accordingly.
(x ) (x x ) > 0
(x ) (x x ) = 0
4 Pseudoconcavity
4.1 Denition
is quasiconcave if and only if
(x) (x ) = (x ) (x x ) 0
is pseudoconcave if
(x) > (x ) = (x ) (x x ) > 0
x x
All rights reserved

4.2 Examples
() = 3 is strictly quasiconcave but not pseudoconcave

(1) > (0) but (0)(1 0) = 0
Cobb-Douglas is pseudoconcave
The utility function (x) is pseudoconcave if preferences are convex
and nonsatiated
4.3 Properties
concave = pseudoconcave = quasiconcave
Every regular quasiconcave function is pseudoconcave
A pseudoconcave function is locally concave at every stationary point
5 Linear and quadratic approximation (Taylors theorem)

For smooth functions
(x0 + x) = (x0 ) + (x0 ) x
+ (x) x ,
(x0 + x) = (x0 ) + (x0 ) x + 12 x (x0 )x

where
0
(x0 ) x = =1 [x0 ] = =1 [x ]
x (x0 )x =
2 [x0 ] =
(x) 0 as x 0
(
)
+ 2 (x) x2
2 [x0 ]

See Section 4.4, especially Theorem 4.3, Corollary 4.3.1. and Example 4.33.
Note: [x0 ] denotes the derivative of at x0 (a linear function), [x0 ](x)
denotes its value at x. [x0 ] denotes the partial derivative of with
respect to (= / ) and 2 [x0 ] denotes the second partial derivative
evaluated at x0 . (See Remark 4.3, p.8 and Remark 4.7. p.15).
6 Quadratic forms
Given a square, symmetrix matrix , the function
(
)( )
11 12
1
(x) = x x = (1 , 2 )
= 11 21 + 212 1 2 + 22 22
12 22
2

All rights reserved
is called a quadratic form.

{
}
{
}
positive
(x) > 0
is
denite if
for every x = 0
negative
(x) < 0
Similarly, we say that the matrix is positive (negative) denitive if its
quadratic form is positive (negative) denite.
Completing the square, can be rewritten as
(x) = x x = 11 (1 +
12 2
11 22 212 2
2 ) + (
)2
11
11
From this, we can deduce that

}
{
}
{
positive
11 > 0
and 11 22 > 212
is
denite if
11 < 0
negative
Similarly
is
nonnegative
nonpositive
}
11 , 22 0
denite if
and 11 22 212
11 , 22 0
The Hessian of a 2 function is a square, symmetric matrix

(
)
11 12
(x) =
12 22
is nonpositive denitive if and only if
11 0,
2
22 0, and 11 22 12
For the Cobb-Douglas function

= 11 22
11 = 1 (1 1)11 2 22
22 = 2 (2 1)11 22 2
12 = 1 2 11 1 22 1
Given 1 , 2 0
11 0
22 0
if and only if 0 1 1
if and only if 0 2 1
Under these preceding conditions

2
11 22 12
= (1 2 (1 1)(2 1) (1 2 )2 )121 2 222 2
= (1 2 (1 2 1 2 + 1 1 2 ))121 2 222 2
= 1 2 (1 1 2 )121 2 222 2
0 provided 1 + 2 1
7

All rights reserved
7 Homework
1. Show that the cost function (w, ) of a competitive rm (Example
2.31) is concave in input prices w.
2. Show that the indirect utility function is quasiconvex in p. [Hint:
Show that the lower contour sets () = { p : (p, ) } are
convex for every . ]
3. Is the CES function (x) = (1 1 + 2 2 + . . . )1/ pseudoconcave?
4. Show
(a) Every dierentiable concave function is pseudoconcave.
(b) Every pseudoconcave function is quasiconcave
(c) Every regular quasiconcave function is pseudoconcave.

All rights reserved
Solutions: Concave and convex functions

1 Suppose that x1 minimizes the cost of producing at input prices w1 while
be the weighted average
x2 minimizes cost at w2 . For some [0, 1], let w
price, that is
= w1 + (1 )w2
w
minimizes cost at w.
Then
and suppose that x
) = w
x
(w,
x
= (w1 + (1 )w2 )
+ (1 )w2 x
= w1 x
But since x1 and x2 minimize cost at w1 and w2 respectively
w1x1 = (w1, )
w1 x
(1 )w2 x2 = (1 )(w2, )
(1 )w2 x
so that
+(1)w2x
(w1 , )+(1)w2, )
) = (w1+(1)w2 , ) = w1 x
(w,
This establishes that the cost function is concave in w.
2 For given and , choose any p1 and p2 in (). For any 0 1, let
= p1 + (1 )p2 . The key step is to show that any commodity bundle
p
is also aordable at either p1 or p2 . Assume that
x which is aordable at p
, that is x is in the budget set
x is aordable at p
x }
x (
p, ) = { x : p
To show that x is aordable at either p1 or p2 , that is
x (p1 , ) or x (p2 , )
assume to the contrary that
/ (p2 , )
x
/ (p1 , ) and x
This implies that
so that
p1 x > and p2 x >

p1 x > and (1 )p2 > (1 )
1

All rights reserved
Summing these two inequalities

x = (p1 + (1 )p2 )x >
p
contradicting the assumption that x (
p, ). We conclude that
(
p, ) (p1 , ) (p2 , )
Now
(
, ) = sup{ (x) : x (
p, ) }
sup{ (x) : x (p1 , ) (p2 , ) }
() for every 0 1. Thus, () is convex and so is

Therefore p
quasiconvex (Exercise 3.146).
3 The CES function is quasiconcave provided 1 (Exercise 3.58). Since
(x) > 0 for all x + +, the CES function with 1 is pseudoconcave
on ++ .
4
(a) If [] is concave (and dierentiable)

(x) (x0 ) + (x0 ) (x x0 )
for every x, x0 . Therefore
(x) > (x0 ) = (x0 ) (x x0 ) > 0
is pseudoconcave.
(b) Assume to the contrary that is pseudoconcave but not quasiconcave.
= x1 + (1 )x2 , x1 , x2
Then, there exists points x1 , x2 and x
such that
(
x) < min{ (x1 ), (x2 )}
(1)
Pseudoconcavity implies
) > 0
(
x) (x1 x
(2)
that is x1 lies on the positive side of the orthogonal hyperplane (

x) (x
) = 0. Therefore, x2 must lie on the negative side, that is
x
) < 0
(
x) (x2 x
contradicting the pseudconcavity of .
2
All rights reserved

(
x)
x2
x
x1
(
x) (x x ) = 0
(1 )x2
More precisely, since x1 = x
) = (1 )(
(x1 x
x x2 )
and therefore
=
x1 x
1
)
(x2 x
Substituting in (2) gives

1
) < 0
(
x) (x2 x
which by pseudoconcavity implies (x2 ) (

x) contradicting our assumption (1) .
(c) Suppose to the contrary that is regular and quasiconcave but not
pseudoconcave, so that there exists x, x0 such that
(x) > (x0 ) and (x0 ) (x x0 ) 0
Since is regular, (x0 ) = 0 and we can move a small distance from
x in the direction opposite to (x0 ) to nd a point x at which
(x ) > (x0 ) and (x0 ) (x x0 ) < 0
contradicting the assumed quasiconcavity of .
All rights reserved

(x0 )
x
x
x0
(x0 )
(x0 ) (x x0 ) = 0
To make this precise, for every + , let
x = x (x0 )
Then
(
)
(x0 ) (x x0 ) = (x0 ) x (x0 ) x0
= (x0 ) (x x0 ) (x0 ) (x0 )
(x0 )2 < 0
for every + since is regular. Since (x) > (x0 ) and is
continuous, there exists > 0 such that
(x ) > (x0 ) and (x0 ) (x x0 ) < 0
contradicting the quasiconcavity of .

1 Concave and Convex Functions

Uploaded by

Copyright:

Available Formats

1 Concave and Convex Functions

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

1 Concave and Convex Functions

Uploaded by

Copyright:

Available Formats

Lecture notes based on

Foundations of Mathematical Economics

c 2001 Michael Carter

All rights reserved

Concave and convex functions

is concave if 1 and convex otherwise

The prot function (p) = maxy is convex.

The cost function (w, ) = minx () is concave in w

Lecture notes based on

c 2001 Michael Carter

All rights reserved

The deniteness of the Hessian can be assessed by

Lecture notes based on

c 2001 Michael Carter

All rights reserved

The utility function (x) is quasiconcave if preferences are convex.

c 2001 Michael Carter

All rights reserved

Lecture notes based on

Apply properties to combinations of known functions (Exercises 3.148

The gradient of a functional : at x

c 2001 Michael Carter

All rights reserved

Lecture notes based on

measures the angle between (x ) and (x x ), and the orthogonal

c 2001 Michael Carter

All rights reserved

Lecture notes based on

() = 3 is strictly quasiconcave but not pseudoconcave

5 Linear and quadratic approximation (Taylors theorem)

(x0 + x) = (x0 ) + (x0 ) x + 12 x (x0 )x

Lecture notes based on

c 2001 Michael Carter

All rights reserved

is called a quadratic form.

From this, we can deduce that

The Hessian of a 2 function is a square, symmetric matrix

For the Cobb-Douglas function

Under these preceding conditions

Lecture notes based on

c 2001 Michael Carter

All rights reserved

Lecture notes based on

c 2001 Michael Carter

All rights reserved

Solutions: Concave and convex functions

p1 x > and p2 x >

Lecture notes based on

c 2001 Michael Carter

All rights reserved

Summing these two inequalities

() for every 0 1. Thus, () is convex and so is

(a) If [] is concave (and dierentiable)

that is x1 lies on the positive side of the orthogonal hyperplane (

c 2001 Michael Carter

All rights reserved

Lecture notes based on

Substituting in (2) gives