Quasiconcavity,

Quasiconvexity

Quasiconcavity and
Quasiconvexity

If we know the concavity or convexity of the

objective function, no need to check the secondorder condition .
In constrained optimization, it is possible to
dispense with the second-order condition if the
surface or hypersurface has the appropriate type
of configuration.

Quasiconcavity and
Quasiconvexity

The desired configuration is:

quasiconcavity (rather than concavity) for a maximum,

and
quasiconvexity (rather than convexity) for a minimum.

These are weaker conditions than concavity and

convexity.
These can also be strict or non-strict.

GEOMETRIC CHARACTERIZATION:

(a)

(b)

(c)

Let u and v be any two distinct points in the domain (a convex set) of a function f.
Let line segment uv in the domain give rise to arc MN on the graph of the function,
such that point N is higher than or equal in height to point M.
Then function f is said to be quasiconcave (quasiconvex) if all points on arc MN other
than M and N are higher than or equal in height to point M (lower than or equal in
height to point N).
The function f is said to be strictly quasiconcave (strictly quasiconvex) if all the points
on arc MN other than M and N are strictly higher than point M (strictly lower than
point N).

GEOMETRIC CHARACTERIZATION:

(c)

Let u and v be any two distinct points in the domain (a convex set) of a function f.
Let line segment uv in the domain give rise to arc MN on the graph of the function, such
that point N is higher than or equal in height to point M.
Then function f is said to be quasiconcave (quasiconvex) if all points on arc MN other
than M and N are higher than or equal in height to point M (lower than or equal in
height to point N).
The function f is said to be strictly quasiconcave (strictly quasiconvex) if all the points
on arc MN other than M and N are strictly higher than point M (strictly lower than point
N).

a)
Strict quasiconcavity - all the points between M and N on the said arc are strictly
higher than M. Note: N is higher than M.

Function also satisfies the condition for (nonstrict) quasiconcavity

Fails the condition for quasiconvexity, because some points on arc MN are higher
than N, which is forbidden for a quasiconvex function.

b)

All the points on arc M'N' are lower than N (the higher of the two ends), and the
same is true of all arcs that can be drawn.

Thus the function is strictly quasiconvex.

It also satisfies the condition for (nonstrict) quasiconvexity, but fails the condition
for quasiconcavity.

c)

Presence of a horizontal line segment M"N", where all the points have the same
height.

As a result, that line segmentand hence the entire curvecan only meet the
condition for quasiconcavity, but not strict quasiconcavity.

(a)

(b)

Strictly concave.

Not strictly concave

Also strictly quasi-concave.

Strictly quasi-concave: all

points on MN and MN are
higher than

Algebraic Definitions
quasiconcave
A function f is
iff, for any pair of distinct points u and v
quasiconvex

in the (convex-set) domain f , and for 0 1,

f (u )
f (v) f (u ) f ( u (1 )v)

f
(
v

)
strictly quasiconcave
A function f is
iff, for any pair of distinct points u and v
strictly quasiconvex
in the (convex-set) domain f , and for 0 1,
f (u )

f
(
v

f (v) f (u ) f ( u (1 )v)

3 Theorems

Theorem I (negative of a function)

If f(x) is quasiconcave (strictly quasiconcave), then --f(x) is
quasiconvex (strictly quasiconvex).

Theorem II (concavity versus quasiconcavity)

Any concave (convex) function is quasiconcave (quasiconvex), but the
converse is not true. Similarly, any strictly concave (strictly convex)
function is strictly quasiconcave (strictly quasiconvex), but the
converse is not true.

Theorem III (linear function)

If f(x) is a linear function, then it is quasiconcave as well as
quasiconvex.

Theorem I - Proof

Theorem I follows from the fact that multiplying an inequality by -1

reverses the sense of inequality.
Let f(x) be quasiconcave, with f(v) > f(u). Then,
f[ u + (1 - )v] > f (u).
As far as the function -f(x) is concerned, however, we have (after
multiplying the two inequalities through by -1)
-f(u) > -f(v) and -f[ u + (1 - )v] < -f (u).
Interpreting -f(u) as the height of point N, and -f(v) as the height of M, we
see that the function - f(x) satisfies the condition for quasiconvexity.

Theorem II - Proof

We shall only prove that concavity implies quasiconcavity.

Let f ( x) be concave. Then,
f [ u (1 )v] f (u ) (1 ) f (v)
Now assume that f (v) f (u );
then any weighted average of f (v) and f(u) cannot possibly
be less than f (u ), i.e.,

f (u ) (l - ) f (v) f (u )
Combining these two results, we find that, by transitivity,
f [ u (1 )v] f (u )
for f (v) f (u )
which satisfies the definition of quasiconcavity .
Note, however, that the condition for quasiconcavity cannot
guarantee concavity.

Theorem III - Proof

We already know that a linear function is both

concave and convex, though not strictly so.
In view of Theorem II, a linear function must also be both
quasiconcave and quasiconvex, though not strictly so.

Alternative Definition of quasiconcavity and quasiconvexity:

A function f ( x), where x is a vector of variables,
quasiconcave
is
iff, for any constant k , the set
quasiconvex
S x | f ( x) k

is a convex set.

S x | f ( x) k

The sets S and S are subsets of the domain. A convex function

(or even a concave function) can give rise to a convex set.

The three functions in the all contain concave as well as convex segments
hence are neither convex nor concave.
The function in Fig. 12.5a is quasiconcave, because for any value of k
(only one of which has been illustrated), the set S is convex.
The function in Fig. 12.5b is, on the other hand, quasiconvex since the set S is convex.
The function in Fig. 12.5c - a monotonic function - differs from the other two in that both
S and S are convex sets. Hence that function is quasiconcave as well as quasiconvex.

Example 1:
Check z= x2 (x > 0) for quasiconcavity and quasiconvexity

This function can be verified geometrically to be convex,

but not strictly so. Hence it is quasiconvex.
Interestingly, it is also quasiconcave. For its graphthe
right half of a U-shaped curve, initiating from the point of
origin and increasing at an increasing rate.

First let u and v be any two distinct nonnegative values of x. Then

f (u ) u 2 , f (v) v 2 , and f [ u (1 )v] [ u (1 )v]2
Suppose that f (v) f (u ), that is, v 2 u 2 ; then v u ,
or more specifically, v u (since u and v are distinct).
Inasmuch as the weighted average [ u (1 )v] must lie between u and v
we may write the continuous inequality
v 2 [ u (1 )v]2 u 2 for 0 1
f (v) f [ u (1 )v] f (u )
for 0 1
By (12.20), this result makes the function f both quasiconcave and quasiconvex
- indeed strictly so.

Example 2
Show that z = f(x, y) = xy (with x, y > 0) is quasiconcave
We shall use the criterion in (12.21) and establish that the set
S = {(x, y)| xy > k} is a convex set for any k.
For this purpose, we set xy = k to obtain an isovalue curve for each value of k.
Like x and y, k should be nonnegative.
In case k > 0, the isovalue curve is a rectangular hyperbola in the first
quadrant of the xy plane. The set consisting of all the points on or above a
rectangular hyperbola, is a convex set.
In the other case, with k = 0, the isovalue curve as defined by xy = 0 is Lshaped, with the L coinciding with the nonnegative segments of the x and y
axes.
The set S , consisting this time of the entire nonnegative quadrant, is again a
convex set. Thus, by (12.21), the function z= xy (with x, y > 0) is quasiconcave.

Example 3
Show that z f ( x, y ) ( x a) 2 ( y b) 2 is quasiconvex.
Setting
( x a ) 2 ( y b) 2 k ,
we see that k must be nonnegative. For each k , the isovalue
curve is a circle in the xy plane with its center at (a, b) and
with radius k .
Since S = {( x, y ) | ( x a) 2 ( y b) 2 k}
is the set of all points on or inside a circle, it constitutes a
convex set. This is true even when k 0 -- when the circle
degenerates into a single point, (a, b) -- since by convention
a single point is considered as a convex set. Thus the given
function is quasiconvex.

Differentiable Functions
We can evaluate quasi- concavity/convexity using the first
derivative:
One independent variable:

A differentiable function of one variable, f ( x),

quasiconcave
is
iff, for any pair of distinct points u and v in the domain,
quasiconvex
f '(u )(v u )
f (v) f (u )
0
f
'(
v
)(
v

)
Quasiconcavity and quasiconvexity will be strict, if the weak inequality
on the right is changed to the strict inequality > 0.

Differentiable Functions
Two or more independent variables:

A differentiable function of one variable, f ( x1 ,K , xn ),

quasiconcave
is
iff, for any pair of distinct points u (u1 ,K , un )
quasiconvex
and v (v1 ,K , vn ) in the domain,

j 1

f (v) f (u )

j 1

f j (u )(v j u j )

0
f j (v)(v j u j )

f
where f j
, to be evaluated at u or v as the case may be.
x j

Differentiable Functions
If a function is twice differentiable, quasiconcavity/convexity can be
checked by means of the first and second order partial derivatives,
arranged in a bordered determinant:
If a function z f ( x1 , x2 ,K , xn ) is twice differentiable, form
the bordered determinant
0
f1
B f2
K
fn

f1
f11
f 21
K
f n1

f2
f12
f 22
K
fn2

K
K
K
K
K

fn
f1n
f2n
K
f nn

Leading principal minors: B1 , B2 ,K , Bn

0
B1
f1

f1
,
f11

0
B2 f1
f1

f1
f11
f 21

f
f12 , Bn B
f 22

Differentiable Functions
If a function z f ( x1 , x2 , K , xn ) to be quasiconcave on the nonnegative
orthant, it is necessary that

odd
B1 0,
B2 0,
K , Bn 0 if n is

even
whenever the partial derivatives are evaluated in the nonnegative orthant.
A sufficient condition for/to be strictly quasiconcave
on the nonnegative orthant is that

odd
B1 0,
B2 0,
K , Bn 0 if n is

even
whenever the partial derivatives are evaluated in the nonnegative orthant.

Example: z f ( x1 , x2 ) x1 x2
f1 x2 , f 2 x1 ,
0
B1
x2

f11 f 22 0,

x2
x2 2 0,
0

quasi concave

f12 f 21 1

0
B2 x2

x2
0

x1
1 2 x1 x2 0

x1

Example: z f ( x, y ) x a y b ,
f x ax a 1 y b ,

f y bx a y b 1 ,

f yy b(b 1) x a y b 2 ,
0
B1
fx
0
B2 f x
fy

0 a, b 1
f xx a (a 1) x a 2 y b

f xy f yx abx a 1 y b 1

fx
(ax a 1 y b ) 2 0,
f xx
fx
f xx
f yx

fy
f xy a 2b 2 a (a 1)b 2 a 2b(b 1) x 3a 2 y 3b 2 0
f yy

satisfies condition for strict quasiconcavity

Bordered Determinant vs. Bordered

Hessian
Bordered determinant
0

f1

f2

fn

f1

f11

f12 K

f1n

B f2

f 21

f 22 K

f2n

fn

f n1

fn2 K

f nn

Bordered Hessian
0

g1

g2

gn

g1

Z11

Z12

Z1n

H g2

Z 21

Z 22 K

Z 2n

gn

Z n1

Zn2 K

Z nn

Special case of linear constraint:

g(x1 ,..., xn) = a1x1++ anxn = c
a case frequently encountered in economics.
Z f ( x1 ,K , xn ) (c a1 x1, ,K , an xn )
Z j f j a j and Z ij f ij
g j aj,

Z j f j a j 0

(f.o.c)

f j a j or f j g j
Therefore the border of B is simply the that of H multiplied by
a positive scalar .
B 2 H

Consequently, in the linear-constraint case, the two bordered

determinants always possess the same sign at the stationary point
of Z. By the same token, the leading principal minors
Bi and H i , (i 1,..., n) share the same sign.
It then follows that if the bordered determinant B satisfies the
sufficient condition for strict quasiconcavity , the
bordered Hessian H must then satisfy the second-order sufficient
condition for constrained maximization.