Mathematical Foundations

-1-

Convexity and concavity

Convex set Convex function Strictly Convex function Concave and strictly concave functions Quasi-concave function Supporting hyperplane

John Riley

October 16, 2013

Mathematical Foundations Convex combination

-2-

Convexity and concavity

x is a convex combination of x 0 , x1 n , if for some (0,1)

x = (1 ) x 0 + x1
Convex Set (1 ) x 0 + x1 S S n is convex if, for each x 0 , x1 S every convex combination x = Strictly convex set The function f is strictly convex if, for each x 0 , x1 S every convex combination x lies in the interior of S ( x int S )

John Riley

October 16, 2013

Mathematical Foundations

-3-

Convexity and concavity

We consider functions f defined over a convex set S n

C1:

f is a convex function on S if, for any x 0 , x1 S and convex combination x , 0 < < 1,

f ( x ) (1 ) f ( x 0 ) + f ( x1 )

SC1:

f is a strictly convex function on S if, for any x 0 , x1 S and convex combination x , 0 < < 1,

f ( x ) < (1 ) f ( x 0 ) + f ( x1 )

John Riley

October 16, 2013

Mathematical Foundations

-4-

Convexity and concavity

What is the relationship between a convex set and a convex function? Definition: Graph of a function f : S n G = {( x, y ) n+1 | x S , y = f ( x)} A function is convex if and only if the set
A = {( x, y ) | y f ( x), x S } is a convex set
Points on or above the graph of f y A

Exercise: Prove that if A = {( x, y ) | y f ( x), x S } is a convex set then f is a convex function. HINT: Suppose ( x 0 , y 0 ) and ( x1 , y1 ) are both in A. Then y 0 f ( x 0 ) and y 0 f ( x 0 ) . Then appeal to the convexity of f. Exercise: Prove that if f is convex then A = {( x, y ) | y f ( x), x S } is a convex set. HINT: Consider ( x 0 , y 0 ) = ( x 0 , f ( x 0 )) and ( x1 , y1 ) = ( x1 , f ( x1 )) . By C2 the convex combination

( x , y ) = ( x ,(1 ) f ( x 0 ) + f ( x1 )) A

John Riley

October 16, 2013

Mathematical Foundations Alternative definition of a convex function

-5-

Convexity and concavity

Key idea: For and x 0 , x1 S we can write the linear combinations where the weights sum to 1, as follows: x( ) = (1 ) x 0 + x1 We the consider the properties of the mapping g ( ) f ( x( )) .

John Riley

October 16, 2013

Mathematical Foundations

-6-

Convexity and concavity

John Riley

October 16, 2013

Mathematical Foundations

-7-

Convexity and concavity

C2: For any x 0 , x1 S define x( ) = (1 ) x 0 + x1 and g ( ) = f ( x( )) . Then

x( 0 ), x( 1 ) S g ((1 ) 0 + 1 ) (1 ) g ( 0 ) + g ( 1 )

CS2: For any x 0 , x1 S define x( ) = (1 ) x 0 + x1 and g ( ) = f ( x( )) . Then

x( 0 ), x( 1 ) S g ((1 ) 0 + 1 ) > (1 ) g ( 0 ) + g ( 1 )

John Riley

October 16, 2013

Mathematical Foundations Proposition: C1 C 2 and CS1 CS 2 Proof: C 2 C1:

-8-

Convexity and concavity

If C 2 holds, x(0) = x 0 and x(1) = x1 and so g ( ) (1 ) g (0) + g (1) , that is f ( x ) f ( x( )) (1 ) f ( x 0 ) + f ( x1 ) . C1 C 2 : (1 ) x( 0 ) + x( 1 ) . x( ) is linear therefore x( ) x((1 ) 0 + 1 ) = Since f is convex, g ( ) f ( x( )) = f ((1 ) x( 0 ) + x( 1 )) (1 ) f ( x( 0 )) + f ( x( 1 )) = (1 ) g ( 0 ) + g ( 1 ) . Therefore C1 C 2 . The proof that SC1 SC 2 is almost identical. QED

John Riley

October 16, 2013

Mathematical Foundations

-9-

Convexity and concavity

C3: x( 0 ), x( 1 ) S g ( 0 )( 1 0 ) g ( 1 ) g ( 0 )

CS3:

x( 0 ), x( 1 ) S g ( 0 )( 1 0 ) < g ( 1 ) g ( 0 )

Proposition: If f is differentiable on S then C1 C 3 Proof: C1 C 3 : C1 C 2 therefore g ( ) (1 ) g ( 0 ) + g ( 0 ) for all (0,1) . Rearranging the inequality:

g ( ) g ( 0 ) ( g ( 1 ) g ( 0 ))
and so g ( ) g ( 0 ) 0 ( ) g ( 1 ) g ( 0 ) . 0 Since 0= ( 1 0 ) it follows that g ( ) g ( 0 ) 1 ( 0 ) g ( 1 ) g ( 0 ) . 0

John Riley

October 16, 2013

Mathematical Foundations Taking the limit,

-10-

Convexity and concavity

g ( 0 )( 1 0 ) g ( 1 ) g ( 0 ) .
C 3 C1 If x( 0 ), x( 1 ) S then by the convexity of S x( ) S . Appealing to C3, g ( )( 0 ) g ( 0 ) g ( ) and g ( )( 1 ) g ( 1 ) g ( ) Therefore g ( )((1 ) 0 (1 ) ) (1 ) g ( 0 ) (1 ) g ( ) and g ( )( 1 ) g ( 1 ) g ( ) . Adding the two inequalities yields 0 (1 ) g ( 0 ) + g ( 1 ) g ( ) . QED

John Riley

October 16, 2013

Mathematical Foundations

-11-

Convexity and concavity

Proposition: if f is differentiable on S then SC1 SC 3 Proof: SC 3 SC1: The proof is essentially the same as the proof that C 3 C1. SC1 SC 3 : By SC1 C1 C 3 . Therefore g ( 0 )( 0 ) g ( ) g ( 0 ) . But 0= ( 1 0 ) and by SC3, g ( ) g ( 0 ) < ( g ( 1 ) g ( 0 ) . Therefore

g ( 0 )( 1 0 ) < ( g ( 1 ) g ( 0 ))
QED

John Riley

October 16, 2013

Mathematical Foundations

-12-

Convexity and concavity

C4: For f 1 and for any x 0 , x1 S , define g ( ) = f ((1 ) x 0 + x1 ) x( ), x( ) S ( )[ g ( ) g ( )] 0 Graphically, C4 is the statement that the slope of g ( ) is increasing. Proposition: If f is differentiable then C 4 C 3 Proof:

C 4 C3
f ( x1 ) f ( x 0 ) = g (1) g (0)
= g ( )d g (0)d = g (0) d = g (0)
0 0 0 1 1 1

But g ( ) =

f ( x ).( x1 x 0 ) . x f 0 ( x ).( x1 x 0 ) x

(0) Therefore f ( x1 ) f ( x 0 ) g =

John Riley

October 16, 2013

Mathematical Foundations

-13-

Convexity and concavity

C3 C4
Proof: Since f is convex, g is convex. Appealing to C3,

g ( ) g ( ) + g ( )( ) and g ( ) g ( ) + g ( )( )
Rewriting the second inequality,

g ( ) g ( ) g ( )( )
QED An equivalent way of writing this is that for any and

g ( ) g ( ) 0 (*)
That is, the second derivative of g ( ) must be positive. Also if the second derivative is positive then (*) holds. Therefore if f 2 C4 and C5 are also equivalent. C5: For f 2 and any x 0 , x1 S
g (0) 0 where g ( ) = f ( x( )) = f ((1 ) x 0 + x1 ))

John Riley

October 16, 2013

Mathematical Foundations Restatement of C5 Suppose that f is twice differentiable on n Define g ( ) = f ( x ) = f ( x 0 + ( x1 x 0 )) .

-14-

Convexity and concavity

n n 2 f 1 2 f d 2g 1 0 1 0 1 0 0 0 = = (0) ( ) ( ) ( ) ( ) x x x x x x x (x x ) 0 i i j j 2 d xi x j = i 1= i 1 xi x j

Then we can rewrite C5 as follows: C5: For f 2 defined on S and any x 0 , x1 S

2 f ( x x ) H ( x )( x x ) 0 where H ( x ) = [ ( x 0 )] xi x j
1 0 0 1 0 0

If S = n then C5 is the statement that the Hessian Matrix H ( x 0 ) must be positive semi-definite for all x 0 n

John Riley

October 16, 2013

Mathematical Foundations

-15-

Convexity and concavity

SC4: For f 1 and for any x 0 , x1 S , define g ( ) = f ((1 ) x 0 + x1 ) x( ), x( ) S ( )[ g ( ) g ( )] > 0

Exercise: Prove that SC4 SC3 HINT: Proof follows the proof that C4 C3 Exercise: Prove that SC3 SC4 HINT: Appeal to the definition first with 0 = then with 0 = .

John Riley

October 16, 2013

Mathematical Foundations Strictly Concave Function

-16-

Convexity and concavity

f is a strictly concave function if, for any x 0 , x1 S and convex combination x , 0 < < 1,

f ( x ) < (1 ) f ( x 0 ) + f ( x1 )
Concave Function f is a concave function if, for any x 0 , x1 S and convex combination x , 0 < < 1,

f ( x ) (1 ) f ( x 0 ) + f ( x1 )

Reverse all the inequalities in C1 C 5 and SC1 SC 4 to obtain equivalent definitions of a concave and strictly concave function

Exercise: Show that a function f is concave on S if and only if f is convex on S .

John Riley

October 16, 2013

Mathematical Foundations Class exercises (in groups)

-17-

Convexity and concavity

1. Prove that a concave function of a concave function is not necessarily concave. 2. Prove that a strictly concave function of a strictly concave function maybe strictly convex. 3. Show that the sum of concave functions is concave. 4. Show that the sum of strictly concave function is strictly concave. 5. If f1 ( x1 , x2 ) is strictly concave and f1 ( x3 , x4 ) is concave then = h( x) f ( x1 , x2 ) + g ( x3 , x4 ) is strictly concave. Either prove that this statement is true or present a counter example extablishing that the statement is false.

John Riley

October 16, 2013

Mathematical Foundations

-18-

Convexity and concavity

Shortcuts for checking whether a function is convex or concave With one variable, the easiest way to tell if a function is strictly convex is to see if it has a positive second derivative. For functions of more than one variable, the following propositions are often helpful. Proposition: The sum of concave functions is convex. Proposition: Concave function of a function

h( x) = g ( f ( x)) is concave if (i) f is concave & g is increasing and concave, or (ii) f is linear and g is
concave. Proof of (i) Since f is concave, Since g is increasing Define then Since g is convex Appealing to (*)

f ( x ) (1 ) f ( x 0 ) + f ( x1 ) . g ( f ( x )) g ((1 ) f ( x 0 ) + f ( x1 ))
= y 0 f= ( x 0 ) y1 f ( x1 )

g ( f ( x )) g ((1 ) y 0 + y1 )

(*)

g ((1 ) y 0 + y1 ) (1 ) g ( y 0 ) + g ( y1 ) = (1 ) g ( f ( x 0 )) + g ( f ( x1 )) . g ( f ( x )) (1 ) g ( f ( x 0 )) + g ( f ( x1 ))

John Riley

October 16, 2013

Mathematical Foundations Quasi-concave Function

-19-

Convexity and concavity

A function f is quasi-concave on the convex set S n if, for any x 0 , x1 in S, such that f ( x1 ) f ( x 0 ) and any convex combination x = (1 ) x 0 + x1 , 0 < < 1,

f ( x ) f ( x0 ) .
Strictly Quasi-concave Function A function f is quasi-cocave on the convex set S n if, for any x 0 , x1 in S, such that f ( x1 ) f ( x 0 ) and any convex combination x = (1 ) x 0 + x1 , 0 < < 1,

f ( x ) > f ( x0 ) .
Example: U ( x) = x1 x2 is quasi-concave on 2 + and is strictly quasi-concave on
2 = {x | x 2 , x >> 0} ++

But why! Read on.

John Riley

October 16, 2013

Mathematical Foundations

-20-

Convexity and concavity

Proposition: A function is quasi-concave if and only if the upper contour sets of the function are convex Proof of equivalence: Without loss of generality we may assume that f ( x1 ) f ( x 0 ) .

, consider vectors x 0 and x1 which lie in the upper contour Suppose f is quasi-concave and for any x
) of this function. That is f ( x 0 ) f ( x ) and f ( x1 ) f ( x ) . From the definition of quasiset C U ( x
concavity, for any convex combination,

(1 ) x 0 + x1 , f ( x ) f ( x 0 ) . x = ) , it follows that for all convex combinations f ( x ) f ( x ) . Hence all convex Since f ( x 0 ) f ( x
) is convex. combinations lie in the upper contour set. Thus CU ( x

Conversely, if the upper contour sets are convex, then CU ( x 0 ) is convex. Since f ( x1 ) f ( x 0 ) . Then

x 0 and x1 are both in CU ( x 0 ) therefore, by the convexity of CU ( x 0 ) , all convex combination lie in this
set. That is f ( x ) f ( x 0 ) for all , 0 < < 1 . Q.E.D.

John Riley

October 16, 2013

Mathematical Foundations

-21-

Convexity and concavity

Proposition: If there is some strictly increasing function g such that h( x) g ( f ( x)) is concave then f(x) is quasi-concave. Proof: We need to establish that if f ( x1 ) f ( x 0 ) then f ( x ) f ( x 0 ), 0 < < 1 By concavity h( x ) (1 )h( x 0 ) + h( x1 ) , that is g ( f ( x )) (1 ) g ( f ( x 0 )) + g ( f ( x1 )) . Since g is increasing and f ( x1 ) f ( x 0 ) , g ( f ( x1 )) g ( f ( x 0 )) . Then substituting into (*), (*)

g ( f ( x )) g ( f ( x 0 )) .
By hypothesis, g is a strictly increasing function. Then f ( x ) f ( x 0 ) . Q.E.D.

Remark: It is in this last line of the proof that we require that g is strictly increasing. For otherwise g could be constant over some neighborhood N ( x 0 , ) and so we could have f ( x1 ) < f ( x 0 ) and

g ( f ( x1 )) = g ( f ( x 0 )) .

John Riley

October 16, 2013

Mathematical Foundations

-22-

Convexity and concavity

Proposition B.2-9: If f : n + is positive, homogeneous of degree 1 (i.e. for all > 0

f ( z ) = f ( z ) ) and quasi-concave then f is concave.

Proof: Because f is homogeneous of degree 1 it follows that for any x 0 , x1 and (0,1)

(1 ) f ( x 0 ) + f ( x1 ) =f ((1 ) x 0 ) + f ( x1 ) ,
Since f > 0 , there exists > 0 such that

f ((1 ) x 0 ) = f ( x1 ) .
Then

(1)

(1 ) f ( x 0 ) + f ( x1 ) = (1 + ) f ( x1 ) .

(2)

Because f is homogeneous of degree 1 it follows from (1) that f ((1 ) x 0 ) = f ( x1 ) . Note that

1+

((1 ) x 0 + x1 )=

1+

(1 ) x 0 +

1 x1 . 1+

Therefore

1+

((1 ) x 0 + x1 ) is a convex combination of (1 ) x 0 and x1 .

John Riley

October 16, 2013

Mathematical Foundations We showed that

-23-

Convexity and concavity

1+

((1 ) x 0 + x1 ) is a convex combination of (1 ) x 0 and x1 .

Therefore, by the quasi-concavity of f

f( ((1 ) x 0 + x 0 )) f ( x1 ) . 1+

(3)

Because f is homogeneous of degree 1 it follows that

1+

f ((1 ) x 0 + x 0 ) f ( x1 )

and hence that

f ((1 ) x 0 + x 0 ) (1 + ) f ( x1 ) .
Appealing to (2),

f ((1 ) x 0 + x 0 ) (1 ) f ( x 0 ) + f ( x1 ) .
QED

John Riley

October 16, 2013

Mathematical Foundations Class examples:

-24-

Convexity and concavity

In each case is f (i) quasi-concave (ii) strictly quasi-concave (iii) concave (a) f ( x) = (1 + x1 )(2 + x2 ), x 2 +

(b) = f ( x) x1 x2 , x 2 + (c) = f ( x) x1 x2 , x 2 ++ (d) (e) (f) (g) f ( x) = x1 x2 + x1 x3 , x 3 ++ f ( x) = x11/2 + x21/2 + x31/2 f ( x) = ( x11/2 + x21/2 + x31/2 ) 2 f ( x) = ( x11/2 + x21/2 + x31/2 )3/2

John Riley

October 16, 2013

Mathematical Foundations

-25-

Convexity and concavity

Class Exercise: Are the following statements true or false? 1. For any convex set S n if the quasi-concave function f : S has a local maximum at x* , then f takes on its global maximum at x* . 2. For any set S n if the strictly quasi-concave function f : S has a local maximum at x* , then f takes on its global maximum at x* .

John Riley

October 16, 2013

Mathematical Foundations Supporting hyperplane for a convex set Suppose that for some x 0 , the upper contour set

-26-

Convexity and concavity

= S {x | f ( x) f ( x 0 )}
is convex. (This will be true if f is quasi-concave.) The linear approximation of f at x 0 is
) f ( x0 ) + l(x = f 0 ( x ) ( x x0 ) . x

Consider the upper contour set of l. This is the set

0 = T {x | l ( x) l ( x = )} {x |

f 0 ( x ) ( x x 0 ) 0} x

We will argue that S T .

John Riley

October 16, 2013

Mathematical Foundations Proposition: Supporting hyperplane

-27-

Convexity and concavity

If the set S {x | f ( x) f ( x 0 )} is convex, then S T {x | = The hyperplane hyperplane.

f 0 ( x ) ( x x 0 ) 0}. x

f 0 ( x ) ( x x0 ) = 0 touching the boundary of S at x 0 is called a supporting x

Proof: For x1 S define g ( ) = f ( x( )) f ( x 0 ) = f ( x 0 + ( x1 x 0 )) . Since S is convex, x( ) is in S. Therefore g ( ) 0 . Since g (0) = 0 it follows that But as we have previously shown,
f 0 dg = (0) ( x ) ( x1 x 0 ) . x d dg (0) 0 . d

Then for all x1 S ,

f 0 ( x ) ( x1 x 0 ) 0 . x

Q.E.D.

John Riley

October 16, 2013

Mathematical Foundations Example 1: Consider the contour set of f ( x = ) ( x1 + 2 x21/2 ) 2 through (1,1) . The upper contour set is

-28-

Convexity and concavity

S= {x | ( x1 + 2 x21/2 ) 2 9}
Confirm that f ( x = ) ( x1 + 2 x21/2 ) 2 is quasi-concave. The linear approximation of f at x 0 is
l ( x) = f ( x 0 ) + f 0 ( x ) ( x x 0 ) = 9 + 6( x1 1) + 6( x2 1) . x

Consider the upper contour set of l. This is the set

0 = T {x | l ( x) l ( x 0= )} {x | 6( x1 x10 ) + 6( x2 x2 ) 0}

Example 2: Production set of a firm

= Y {( z , q ) | 3 z q 0} where (0,1] . Consider the tangent line through (1,3)

John Riley

October 16, 2013

Mathematical Foundations

-29-

Convexity and concavity

Application: Market clearing in an exchange economy Suppose all consumers have the same strictly increasing quasi-concave utility function U () 1
h defined on the n + . Consumer h has endowment vector . Finally suppose that

U h ( ) h > 0 x

Appealing to the Supporting hyperplane proposition,

U ( x h ) U ( h ) U h ( ) ( x h h ) 0 x U h ( ) . Since U is strictly increasing p 0 . Then x

Define the price vector p =

U ( x h ) U ( h ) p ( x h h ) 0

(*)

Thus any strictly preferred consumption bundle x1 costs at least as much as 0 . We wish to show that

U ( x h ) > U ( h ) p ( x h h ) > 0 .

(**)

That is, any strictly preferred bundle costs strictly more. Then the consumer can do no better than consume his or her endowment.

John Riley

October 16, 2013

Mathematical Foundations Suppose (**) is false.

-30-

Convexity and concavity

) > U ( h ) and p x = p h . , U (x Then for some x

By hypothesis,
U h ( ) h > 0 . Hence p h > 0 . x

, 0 < < 1 , 0 < p x < p 0 . Then for all x = x

) > U ( h ) and U is continuous, it follows that Since U ( x U ( x ) > U ( h ) for all sufficiently close to 1.
Summarizing our conclusions: For sufficiently close to 1, U ( x ) > U ( h ) and p x < p h . But this contradicts (*).

. Thus there can be no such x

QED We have therefore shown that if the price vector is p, then consumer h will not gain from trading. Then if all consumers have the same endowment, none will gain from trading. Thus p is a market clearing price vector.

John Riley

October 16, 2013