Mathematics for Economists-1

Sections 5 (Problem set 4), October 9, 2023

Alexander Tonis, Kirill Savin

1. Let f : Rn → R and g : R → R be continuously differentiable functions. Assume further that g is

strictly increasing. Using the first order characterization of quasi-concavity, show that if f is quasi-
concave, then so is the function g ◦ f : Rn → R defined as g ◦ f (x) = g(f (x)) for every x ∈ Rn .
Solution
According to the first order characterization of quasi-concavity,

f (y) ≥ f (x) ⇒ Df (x)(y − x) ≥ 0.

Hence, since g is strictly increasing,

g(f (y)) ≥ g(f (x)) ⇒ f (y) ≥ f (x) ⇒ Df (x)(y − x) ≥ 0 ⇒

Dg(f (x))(y − x) = g 0 (f (x))Df (x)(y − x) ≥ 0.

Thus, g ◦ f is quasi-concave.
√ √
2. (i) Prove or disprove: v(x, y) = 2( x+ y) is a strictly quasi-concave function of (x, y) ∈ R2++ .
Solution
It is a strictly quasi-concave function as a strictly increasing transformation of a strictly concave
√ √
function u(x, y) = x + y: v(x, y) = h(u(x, y)) where h(z) = 2z is strictly increasing in z.
√ √
The function u(x, y) = x + y is strictly concave because it has a negative definite second
derivative matrix for all x, y > 0:
1
 
− 3/2 0
D2 u(x, y) =  4x 1 .
 
0 − 3/2
4y

Note. A strictly increasing transformation of a (strictly) concave function is (strictly) quasi-

concave but the reverse is not true: there are quasi-concave functions which do not admit this
representation. No strictly increasing transformation of such a function is concave (see the com-
ment after the solution of the last problem).
(ii) Give an example of two strictly increasing utility functions u and v on R2+ that represent the
same preference relation, and such that only u is concave. Explain your answer.
Solution
Let us show that functions u and v from (i) give an appropriate example. Indeed, u is concave
as we have shown while v is not: for example, if we consider f (x) = v(x, x) then
√ √
√ 22 x ln 2 22 x ln 2
 
2 x 0 00 1
f (x) = v(x, x) = 2 ⇒ f (x) = √ ⇒ f (x) = ln 2 − √
x x 2 x

which is positive for high x. Hence, f (which is just v considered on a restricted convex domain)
is not concave ⇒ v is not concave.

3. Graphically, give an example of a function f : R → R that is quasi-concave, but not monotonic or

concave. Is the function f also quasi-convex? Could you construct a similar example which also
features distinct points x∗ and y ∗ that uniquely maximize f on Bε (x∗ ) and Bε (y ∗ ), respectively, for
some ε > 0?
Solution
1
Let f (x) = .
1 + x2
 Upper contour sets are segments ⇒ f is quasi-concave;
2x
f 0 (x) = − ⇒ f 0 (x) > 0 for x < 0, f 0 (x) < 0 for x > 0 ⇒ f is not monotonic;
(1 + x2 )2
2(3x2 − 1) √
f 00 (x) = 2 3
⇒ f 00 (x) > 0 for |x| > 3 ⇒ f is not concave;
(1 + x )
Lower contour sets are complements to intervals ⇒ f is not quasi-convex.
We could not construct a similar example which also features distinct points x∗ and y ∗ that are strict
local argmax of f because if, say, f (x∗ ) ≤ f (y ∗ ) then f (x) ≥ f (x∗ ) for x ∈ (x∗ , y ∗ ), in particular, for
such of these points that are ε-close to x∗ .

3
4. Let f (x) = (x1 )1/2 + (x2 )1/3 for every x = (x1 , x2 ) ∈ R2++ . Prove or disprove the following
statement: For any x, y ∈ R2++ ,

Df (x)(y − x) < 0 ⇒ f (x) > f (y).

Solution The statement to be proved/disproved is just an equivalent reformulation of the first order
characterization of quasi-concavity:

f (y) ≥ f (x) ⇒ Df (x)(y − x) ≥ 0.

So let us check that f (x) is quasi-concave. Indeed, both (x1 )1/2 and (x2 )1/3 are concave ⇒ their sum
is concave ⇒ f (x) is quasi-concave as a strictly increasing transformation of a concave function.

5. Fix an integer n ≥ 2, and let ∆ denote the unit simplex in Rn . So, ∆ := x ∈ Rn+ : ni=1 xi = 1 .
 P
If you think of the elements of ∆ as probability measures (on 1, . . . , n), the so called expected utility
functions on ∆ are characterized by the following property:

f (αx + (1 − α)y) = αf (x) + (1 − α)f (y) ∀x, y ∈ ∆, ∀α ∈ [0, 1]. (1)

(i) Show that f : ∆ → R satisfies (1) iff it is both concave and convex.
Solution
Concavity of f : f (αx + (1 − α)y) ≥ αf (x) + (1 − α)f (y).
Convexity of f : f (αx + (1 − α)y) ≤ αf (x) + (1 − α)f (y).
Both conditions hold iff (1) holds.
Note. Actually an expected utility function is the same as an affine function, i.e. a linear function
(depending on n − 1 variables, say, x1 , . . . , xn−1 ) plus a constant. This is so because
n−1
!
X
f (x1 , . . . , xn−1 ) = f (e1 )x1 + · · · + f (en−1 )xn−1 + f (en ) 1 − xi
i=1

where ei are the unit vectors from the standard basis.

(ii) Graphically, give an example of a function f : ∆ → R that is both quasi-concave and quasi-
convex, but that does not satisfy (1). (Hint. Let n = 3, so that ∆ is a two dimensional triangle.
The level curves of f need to be line segments. But these line segments may or may not be
parallel to each other.)
Solution
x2 + 1
Let n = 3, ∆ = {(x1 , x2 , x3 )> : x1 ≥ 0, x2 ≥ 0, x3 = 1 − x1 − x2 ≥ 0} and f (x1 , x2 ) = .
x1 + 1

Note. The same function f (x1 , x2 ) defined on R2+ is an example of a quasi-concave function which
cannot be a strictly increasing transformation of a concave function. To show this, suppose, on
the contrary, that such a representation is possible, i.e.
 
x2 + 1
g(x1 , x2 ) = h(f (x1 , x2 )) = h
x1 + 1
is concave where h : (0, ∞) → R is strictly increasing. In particular, g(n − 1, n) = h(1 + 1/n)
should be concave in n ∈ [1, ∞). This implies that the sequence
h(2) − h(3/2), h(3/2) − h(4/3), h(4/3) − h(5/4), . . .
should be increasing. Denote a = h(2), b = h(2) − h(3/2) > 0. Then h(1 + 1/n) < a − (n − 1)b,
so h(y) is not defined (is −∞) for y ≤ 1. This contradiction proves the statement.
Graphically, if the level lines f (x1 , x2 ) = n (n = 1, 2, . . .) are fixed in any way, their intersections
with the line x2 = x1 +1 cannot permanently get denser as we move up-right, because the number
of the intersections is finite.