New York University

Department of Economics
V31.0006 C. Wilson
Mathematics for Economists September 15, 2011
Concave and Quasi-Concave Functions
A set X R
n
is convex if x, y X implies x + (1 ) y X for all [0, 1] .
Geometrically, if x, y R
n
, then {z R
n
: z = x + (1 ) y for [0, 1]} constitutes the straight
line connecting x and y. So a convex set is any set that contains the entire line segment between
any two vectors in the set.
The intersection of two convex sets is convex. Can you prove this?
The union of two convex sets is not necessarily convex. Why not?
A vector z R
n
is a convex combination of x
1
, ..., x
m
R
n
if
z =
m
X
j=1

j
x
j
for some
1
, ...,
m
0 with
m
X
j=1

j
= 1.
In the gure below:
x
12
=
1
2
x
1
+
1
2
x
2
, x
13
=
1
2
x
1
+
1
2
x
3
, x
23
=
1
2
x
2
+
1
2
x
3
x
123
=
2
3
x
12
+
1
3
x
3
=
2
3
x
13
+
1
3
x
2
=
2
3
x
23
+
1
3
x
1
=
1
3
x
1
+
1
3
x
2
+
1
3
x
3
http://homepages.nyu.edu/ caw1/
V31.0006: Concave and Quasi-Concave Functions September 15, 2011 Page 2
Theorem 1: A set X R
n
is convex if and only if it contains any convex combination of any
vectors x
1
, ..., x
m
X.
Proof. (if) If X contains any convex combination of its vectors, then as a special case, x +
(1 ) y X for all x, y X and [0, 1] ..
(only if) The proof is by mathematical induction on m. For m = 1, the only convex combination of
vector x is x itself. So the basis statement for m = 1 is true. The induction step is to suppose that
the proposition is true for m1 > 0 vectors, and then to show that this implies the proposition is
true for m vectors. So consider any convex combination
P
m
j=1

j
x
j
of m vectors contained in X.
Since m 2 and each
j
0 with
P
m
j=1

j
= 1, we may suppose WLOG that
m
< 1. Then since
m1
X
j=1

j
1
m
=
P
m1
j=1

j
1
m
=
1
m
1
m
= 1,
the induction hypothesis implies that y
P
m1
j=1

j
1
m

x
j
X. Then the denition of a convex
set implies
m
X
j=1

j
x
j
=
m1
X
j=1

j
x
j
+
m
x
m
= (1
m
)
m1
X
j=1

j
1
m

x
j
+
m
x
m
= (1
m
) y +
m
x
m
X.
The proposition then follows from mathematical induction.
Given any set X R
n
, the convex hull Co(X) is the intersection of all convex sets that contain X.
Since the intersection of any two convex sets is convex, it follows that the convex hull is the
smallest convex set that contains X.
If X is convex, then Co(X) = X. Why?
Theorem 2: Suppose X R
n
. Then Co(X) is the set of all convex combinations of vectors in X.
Proof. See Appendix.
http://homepages.nyu.edu/ caw1/
V31.0006: Concave and Quasi-Concave Functions September 15, 2011 Page 3
Concave Functions
For the remainder of these notes, we suppose that X R
n
is a convex set.
f : X R is concave if for any x, z X, we have, for all (0, 1) ,
f(z + (1 ) x) f (z) + (1 ) f(x).
f : X R is strictly concave if for any x, z X with x 6= z, we have, for all (0, 1) ,
f(z + (1 ) x) > f (z) + (1 ) f(x).
concave not concave
A constant function is concave. Why?
A linear function is concave. Why?
f : X R is concave if and only if f((z x) +x) (f(z) f(x)) +f(x) for all x, z X and
(0, 1) . Why?
f : X R is concave if and only if f(x + x) (f(x +x) f(x)) + f(x) for all
x, (x +x) X and (0, 1) . Why?
Theorem 3: (Jensens Inequality) Suppose that f : X R is concave. If x
1
, . . . , x
m
X and

1
, . . . ,
m
R
+
with
P
m
i=1

i
= 1, then f(
P
m
i=1

i
x
i
)
P
m
i=1

i
f(x
i
).
Proof. Left as an exercise [Hint: use the denition of a concave function and mathematical induc-
tion.]
Linear Combinations of Concave Functions
Consider a list of functions f
i
: X R for i = 1, ..., n, and a list of numbers
1
, ...,
n
. The function
f
P
n
i=1

i
f
i
is called a linear combination of f
1
, ..., f
n
. If each of the weights
i
0, then f is a
nonnegative linear combination of f
1
, ..., f
n
.
The next proposition establishes that any nonnegative linear combination of concave functions is
also a concave function.
http://homepages.nyu.edu/ caw1/
V31.0006: Concave and Quasi-Concave Functions September 15, 2011 Page 4
Theorem 4: Suppose f
1
, ..., f
n
are concave functions. Then for any
1
, ...,
n
, for which each

i
0, f
P
n
i=1

i
f
i
is also a concave function. If, in addition, at least one f
j
is strictly concave
and
j
> 0, then f is strictly concave.
Proof. Consider any x, y X and (0, 1) . If each f
i
is concave, we have
f
i
(x + (1 ) y) f
i
(x) + (1 ) f
i
(y)
Therefore,
f(x + (1 ) y)
n
X
i=1

i
f
i
(x + (1 ) y)
n
X
i=1

i
(f
i
(x) + (1 ) f
i
(y))
=
n
X
i=1

i
f
i
(x) + (1 )
n
X
i=1

i
f
i
(y) f(x) + (1 ) f(y).
This establishes that f is concave. If some f
i
is strictly concave and
i
> 0, then the inequality is
strict.
Since a constant function is concave, Theorem 4 implies:
If f is concave, then any ane transformation f + with 0 is also concave.
If f is strictly concave, then any ane transformation f + with > 0 is also strictly concave.
Theorem 5: Let I R be an interval and let X = I
n
. Suppose f : X R is dened by
f(x) =
P
n
i=1

i
(x
i
), where each
i
: I R.
(a) If each
i
is concave, then f is concave.
(b) If each
i
is strictly concave, then f is strictly concave.
Proof. Left as an exercise.
A function f : X R
n
of the form f(x) =
P
n
i=1

i
(x) is sometimes called a linearly separable
function.
Why does Theorem 5 require that each
i
be strictly concave to ensure that f is strictly concave,
while Theorem 4 requires only one f
i
be strictly concave to ensure that f is is strictly concave?
Concave Functions of an Ane Function
Theorem 6: Let X R
n
be convex and f : X R is a concave function.
(a) Let g : R
m
R
n
be dened by g(y) = Ay + b, where A is an n m matrix and suppose
g[Y ] X. Then h = (f g) : Y R is a concave function.
(b) If f is strictly concave and g is 1-1, then h is strictly concave.
Proof. Left an exercise.
http://homepages.nyu.edu/ caw1/
V31.0006: Concave and Quasi-Concave Functions September 15, 2011 Page 5
Quasi-Concave Functions
f : X R is quasi-concave if for any x, z X, we have f(z + (1 ) x) min{f (x) , f(z)} for
all (0, 1) .
f is strictly quasi-concave if for any x, z X and x 6= z, we have f(z+(1 ) x) > min{f (x) , f(z)}
for all (0, 1) .
Theorem 7: A (strictly) concave function is (strictly) quasi-concave.
Proof. The theorem follows immediately from the observation that if f is concave, then for all
x, z X, we have
f(z + (1 ) x) f(z) + (1 ) f(x) min{f (x) , f(z)} for all (0, 1) .
If f is strictly concave, we have for all x, z X and x 6= z,
f(z + (1 ) x) > f(z) + (1 ) f(x) min{f (x) , f(z)} for all (0, 1) .
f is quasi-concave f is not quasi-concave
If X R, then f : X R is quasi-concave if and only if it is either monotonic or rst
nondecreasing and then nonincreasing. Why?
Our next theorem states that any monotone nondecreasing transformation of a quasi-concave
function is quasi-concave.
Theorem 8: Suppose f : X R is quasi-concave and : f(X) R is nondecreasing. Then
f : X R is quasi-concave. If f is strictly quasi-concave and is strictly increasing, then f
is strictly quasi-concave.
Proof. Consider any x, y X. If f is quasi-concave, then f (x + (1 ) y) min{f(x), f(y)} .
Therefore, nondecreasing implies
(f (x + (1 ) y)) (min{f(x), f(y)}) = min{(f(x)), (f(y))} .
If f is strictly quasi-concave, then for x 6= y, we have f(x + (1 ) y) > min{f(x), f(y)}.
Therefore, if is strictly increasing, we have
(f (x + (1 ) y)) > (min{f(x), f(y)}) = min{(f(x)), (f(y))} .
http://homepages.nyu.edu/ caw1/
V31.0006: Concave and Quasi-Concave Functions September 15, 2011 Page 6
Recall that for any x X, P(x) {z X : f(z) f(x)} is called the better set of x.
Theorem 9: f : X R is quasi-concave if and only if P(x) is a convex set for each x X.
Proof. (only if) Suppose f is quasi-concave. Choose an arbitrary x
0
X. To show that P(x
0
) is
convex, consider any x, y P(x
0
). Then, f(x), f(y) f(x
0
) and the quasi-concavity of f imply
that
f(x + (1 ) y) min{f(x), f(y)} f(x
0
)
which implies that x + (1 ) y P(x
0
) for any (0, 1) .
(if) Suppose P(x
0
) is convex for each x
0
X. Now consider any x, y X. WLOG, suppose that
f(x) f(y). Then letting x = x
0
, we have that x, y P(x
0
) and therefore y + (1 ) x P(x
0
)
for any (0, 1) . It then follows from the denition of P(x
0
) that
f(y + (1 ) x) f(x) = min{f(x), f(y)} .
Theorem 9 is illustrated below for an increasing function f : R
2
+
R. Notice that all convex
combinations of vectors in P(x
1
) are also elements of P(x
1
). Also notice that if f is strictly quasi-
concave, then the level set can contain no straight line segments.
f is strictly quasi-concave P(x
1
) is convex
Corollary 1: Suppose f : X R attains a maximum on X. (a) If f is quasi-concave, then the
set of maximizers is convex. (b) If f is strictly quasi-concave, then the maximizer of f is unique.
Proof. (a) Let x
0
be a maximizer of f. Then f(x) f(x
0
) for all x X implies that P(x
0
) is the
set of maximizers of f. If f is quasi-concave, then Theorem 7 implies that P(x
0
) is convex.
(b) If f is strictly quasi-concave, suppose x, y are both maximizers of f. Then x 6= y implies
f(
1
2
x +
1
2
y) > min{f(x), f(y)} = f(x) which implies that x is not a maximizer of f.
http://homepages.nyu.edu/ caw1/
V31.0006: Concave and Quasi-Concave Functions September 15, 2011 Page 7
Example: Let I > 0 be the income of some household and let p = (p
1
, ..., p
n
) R
n
+
denote the
vector of prices of the n goods. Then
B(p, I)

x R
n
+
: px I

denes its budget set the set of all possible nonnegative bundles of goods it may purchase
within its budget. You can verify that y, z B(p, I) implies y + (1 ) z B(p, I) for all
[0, 1]. Therefore, B(p, I) is a convex set.
Now suppose that the household preferences are given by some utility function u : R
n
+
R.
Then Corollary 1 implies that if u is strictly quasi-concave, there is a unique bundle x B(p, I)
that maximizes u : B(p, I) R.
Convex and Quasi-Convex Functions
If we reverse the inequality sign in the denitions of concave and quasi-concave functions we obtain
convex and quasi-convex functions.
f : X R is a convex function if for any x, y X, we have, for all (0, 1) ,
f(x + (1 ) y) f (x) + (1 ) f(y)
f : X R is a strictly convex function if for any x, y X where x 6= y, we have, for all (0, 1) ,
f(x + (1 ) y) < f (x) + (1 ) f(y)
NOTE: A convex set and a convex function are two distinct concepts.
f : X R is quasi-convex if for any x, y X, we have f(x + (1 ) y) max {f (x) , f(y)} for
all (0, 1) .
f is strictly quasi-convex if for any x, y X and x 6= y, we have f(x+(1 ) y) < max {f (x) , f(y)}
for all (0, 1) .
The following proposition is an immediate consequence of the denitions.
http://homepages.nyu.edu/ caw1/
V31.0006: Concave and Quasi-Concave Functions September 15, 2011 Page 8
Theorem 10: (a) f is a (strictly) convex function if and only if f is a (strictly) concave function.
(b) f is a (strictly) quasi-convex function if and only if f is a (strictly) quasi-concave function.
Theorem 6 allows us to easily translate all of our propositions for concave and quasi-concave
functions to the analogues for convex and quasi-convex functions, which are provided here for easy
reference.
Suppose f
1
, ..., f
n
are convex functions. Then for any
1
, ...,
n
, for which each
i
0, then
f
P
n
i=1

i
f
i
is also a convex function. If, in addition, at least one f
j
is strictly convex and

j
> 0, then f is strictly convex.
A linear function is both concave and convex.
A (strictly) convex function is (strictly) quasi-convex.
Suppose f : X R is quasi-convex and : f(X) R is nondecreasing. Then f : X R
is quasi-convex. If f is strictly quasi-convex and is strictly increasing, then f is strictly
quasi-convex.
A function f : X R is quasi-convex if and only if for each x X, W(x) is convex.
Suppose f : X R attains a minimum on X. (a) If f is quasi-convex, then the set of minimizers
is convex. (b) If f is strictly quasi-convex, then the minimizer of f is unique.
Appendix
Theorem 2: Suppose X R
n
. Then Co(X) is the set of all convex combinations of vectors in X.
Proof. Theorem 1 implies that any convex combination of elements x
1
, ..., x
m
X must be con-
tained in Co(X). To show that Co(X) contains only vectors that are convex combinations of some
x
1
, ..., x
m
X, we need to show that Y

x R
n
: x is a convex combination of some x
1
, ..., x
m
X

is a convex set. So consider any y, z Y. Then, by denition, there is a set of vectors y

i
= 1 such that y =
P
m
i=1

i
y
i
. Similarly,
there is a set of vectors z
1
, ..., z
r
X and list of nonnegative numbers
1
, ...,
r
with
P
r
i=1

i
= 1
such that z =
P
r
i=1

i
z
i
. Then for any [0, 1] , we have
y + (1 ) z =
m
X
i=1

i
y
i
+ (1 )
r
X
i=1

i
z
i
=
m
X
i=1

i
y
i
+
r
X
i=1
(1 )
i
z
i
which, since
P
m
i=1

i
+(1 )
P
r
i=1

i
= +(1 ) = 1, implies that y +(1 ) z is a convex
combination of y
1
, ..., y
n
, z
1
, ..., z
r
X.
http://homepages.nyu.edu/ caw1/