Mathematics for Doctoral Economics I ©2020 Scott Hacker

Concavity, Convexity, Quasiconcavity and Quasicovexity

A. Unconstrained optimization and the shape of the objective function

In the first lecture in this course we used the fact that f ’’(x) < 0 implies concavity in f and that
f ’’(x) > 0 implies convexity in f to help us determine whether there is a relative maximum or a relative
minimum at x0. In the first lecture in this course we also used second-order partial derivatives and
cross-partial derivatives to deal with f(x,z)’s concavity or convexity.

In general what were finding was that if for a function (f(x) or f(x,z)) we find a critical point,1 then in
the neighborhood of that critical point,

 concave then the function has a relative maximum at the critical point.
if the function is 
 convex the n the function has a relative minimum at at the critical point.

But what does it mean really for a function f(x) or a function f(x,z) to be concave or convex?

If for any two points M and N on the surface of the graphed function, the line segment strictly
 below the function' s surface, then the function is concave.
between those points is always on or 
 above the function' s surface, then the function is convex.

[If we remove the words “on or” in the above, then we could put “strictly” in front of the words
“concave” and “convex”.]

In the following graphical examples (here and in part III) when we move past an end-point the curve
or surface continues with a gradients having the same-signed gradient elements as for the gradient at
the end-point, so there are no surprises on the part we cannot see.

1
A critical point refers to the set of the argument value(s) for the function such that all first derivatives or first partial
derivatives of f are zero.
1
Example 1 :
Figure 1 is an example of a concave function, while Figure 2 is an example of a convex function.

f (x)
f (x)

N
M

x x

Figure 1 Figure 2

Example 2:
Figure 11.2(a) (p. 294) and 11.6 (p. 321) in the textbook are three-dimensional examples of a concave
functions.

Example 3:
Figure 11.2(b) (p. 294) is a three-dimensional example a convex function.

B. Convexity of Sets
A set in n-dimensional space is convex if you when you take any two points within that set and draw
a line between them, every point on that line falls within that set.

Examples: In the textbook’s Figure 11.8 (p. 327), the circle, triangle and pentagon figures represent
convex sets. However, the other two figures do not represent convex sets; if you consider either or
those sets you can find a line between two points in that set that have some points on the line outside
that set.

C. Constrained optimization and the shape of the objective function

Suppose we are to optimize z = f(x, y) with respect to x and y given the constraint ax + by = c.
As usual we write down the Lagrangian function
L = f(x,y) + λ(c – ax – by))
and optimize L with respect to x, y, and λ. The first order conditions are
Lx = 0
Ly = 0
L = 0
which when solved simulatenously results in a critical point (x*, y*).

Over the domain of all possible combinations of (x, y),

 strictly quasiconca ve then f(x, y) has a global constraine d maximum at ( x, y) = ( x*, y*)
if f(x,y) is 
 strictly quasiconve x then f(x, y) has a global constraine d minimum at ( x, y) = ( x*, y*)
which is unique.

But what does it mean really for a function f(x) or a function f(x,y) to be strictly quasiconcave or
strictly quasiconvex?

If for any given k, the set of combinations of f ’s argument values

 f  k is convex , then f is quasiconca ve.

that produce 
 f  k is convex , then f is quasiconve x.
2
If f has no plateaus (flat portions) then a quasiconcave function may be referred to as a strictly
quasiconcave function, and a quasiconvex function may be referred to as a strictly quasiconvex
function. Note: the textbook gives alternative definitions of quasiconcavity and quasiconvexity on pp.
364 and 365 of which you should also be aware.

Example 4:
The function shown in the textbook’s Figure 12.5(a) (p. 367) is quasiconcave because if you consider
any value k for the z function and find the set of all x values that produce z  k , then that set is convex.
However, that function is not strictly quasiconcave because of the existence of a flat portion in that
figure. It is not quasiconvex because because there is at least one value k for the z function such that
the set of all x values that produce z  k is not a convex set (e.g., at the shown k).

Example 5:
The function shown in the textbook’s Figure 12.5(b) is quasiconvex because if you consider any value
k for the z function and find the set of all x values that produce z  k , then that set is convex. It is also
strictly quasiconvex because there are no flat portions. It is not quasiconcave because because there
is at least one value k for the z function such that the set of all x values that produce z  k is not a
convex set (e.g., at the shown k).

Example 6:
The function shown in the textbook’s Figure 12.5(c) is both quasiconcave (because if you consider
any value k for the z function and find the set of all x values that produce z  k , then that set is convex)
and quasiconvex (because if you consider any value k for the z function and find the set of all x values
that produce z  k , then that set is convex). That function may also be called strictly quasiconcave
and strictly quasiconvex because there are no flat portions.

Example 7:
Now let’s consider the textbook’s Figures 12.4(a) and Figure 12.4(b). Consider a top-down look of
either function and find where z = k, and you’ll see a shape as in the Figure 3 below. The curve
indicating where z = k could be called an “iso-z” curve (like and isoquant curve or indifference (iso-
utility) curve in economics, with “iso” meaning “equal”). In this figure the set of all x and y values
that produce z  k is also shown, as is the set of the set of all x and y values that produce z  k .
y H H
H H H
L H
H H
H
H
H H
L H H H
H
H H
H H
L H H
L H H
H
H
L L H H
H H
L L H
H
L
L H H
L L
L
L z=k
L L
L L
L

Figure 3

Notably you cannot find a pair of points within the z  k set such that if you draw a line between them
you’ll get a point outside the set. Therefore the z  k set is convex, and this is true regardless of what
k is chosen to be so the z function (for either Figure 12.4(a) or Figure 12.4(b)) is quasiconcave.
Because there are no flat portions in either figure, the z function in both is also strictly quasiconcave.

3
Notably you can find a pair of points within the z  k set such that if you draw a line between them
you’ll get a point outside the set. Therefore the z  k set is not convex, so the z function (for either
Figure12.4(a) or Figure 12.4(b)) is not quasiconvex.

Now let’s consider either Figures 12.4(a) or Figure 12.4(b) turned upside-down. The associated top-
down view showing where z = k, z  k , and z  k is given in Figure 4 below.
y H H
H H H
H H
H H
L H H
H
H H
L L H
H H
L L
L H
L H
L L L H
H H
L
H
L
L H
H
L L
H H
L L
L
L L L
L z=k
L

Figure 4

Since in Figure 4 the area where z  k is a convex set—and that would be true for for any k—the
associated upside-function is quasiconvex (and strictly so since the z function would have no flat
portions) but the area where z  k is not a convex set, so the associated upside-down function is not
quasiconcave.

Now let’s return to the constrained optimization problem at the beginning of this section. Previously
there was provided a rule for determining whether there was a global constrained maximum or global
constrained minimum at a critical point (x*, y*) based on whether the function is strictly quasiconcave
or strictly quasiconvex at that point. If we are interested in whether we have found a relative
constrained maximum or relative constrained minimum (i.e. providing a relative constrained
extremum in the neighborhood of (x*, y*) and not necessarilty a global constrained extremum), then
a weaker conditions are needed:

Within the neighborhood of (x*, y*),

 strictly quasiconca ve then f(x, y) has a relative constraine d maximum at ( x, y) = ( x*, y*).
if f ( x, y) is 
 strictly quasiconve x then f(x, y) has a relative constraine d minimum at ( x, y) = ( x*, y*) .
When f(x,y) is a differentiable function, we can use the bordered Hessian to determine whether a
critical point is resulting in a relative constrained maximum or a relative constrained minimum.