Convex Functions

Aditya Ghosh
February 28, 2021
Warm-up problem: Suppose that f is twice differentiable in a neighbourhood of c. Show
that the limit
f (c + h) + f (c − h) − 2f (c)
lim
h→0 h2
exists and equals f 00 (c). On the other hand, does the existence of the above limit imply that
f 00 (c) exists and equals the above limit?

Definition 1 (Convex sets). A set S ⊂ Rd (d ≥ 1) is called convex if for every x, y ∈ S

and any t ∈ [0, 1], tx + (1 − t)y ∈ S. In other words, for every two points in the set, the line
segment connecting those points lies completely inside S.

Examples. Any interval in R, any straight line in R2 , any disc in R2 , etc.

Definition 2 (Convex functions). Let X be a convex subset of Rd and let f : X → R be a

function. Then f is called convex if, for all 0 ≤ λ ≤ 1 and x1 , x2 ∈ X, it holds that

f (λx1 + (1 − λ)x2 ) ≤ λf (x1 ) + (1 − λ)f (x2 ).

f is called strictly convex if the above inequality is strict for every x 6= y and any λ ∈ [0, 1].

Definition 3 (Concave functions). Let X be a convex subset of Rd and let f : X → R be a

function. Then f is called concave if, for all 0 ≤ λ ≤ 1 and x1 , x2 ∈ X, it holds that

f (λx1 + (1 − λ)x2 ) ≥ λf (x1 ) + (1 − λ)f (x2 ).

f is called strictly concave if the above inequality is strict for every x 6= y and any λ ∈ [0, 1].

Obviously, f is concave if and only if −f is convex.

In case you are intimidated by the presence of Rd , don’t worry, it was mentioned only
for telling the intuition of line segment joining x and y. Hereafter we shall focus on con-
vex/concave functions of one variable only.
Convince yourself that convex subsets of R are only the intervals (open/closed/semi-
open/semi-closed/bounded/unbounded) or singleton sets (which can also be thought as de-
generate intervals: [a, a] = {a}).
It is not hard to show that |x|, ax + b are convex functions. But proving that a given
function is convex by verifying the above definition is undoubtedly difficult, in general. So
we derive the following tools that would help us for proving convexity.

1
Problem 1. Show that a twice differentiable function defined on an open interval I ⊂ R is
convex if and only if its second derivative is non-negative on I.

The test given by Problem 1 is frequently used for checking convexity. Note that if the
second derivative is positive at all points then the function is strictly convex, but the converse
does not hold — can you give any such example?

Problem 2. Show that a differentiable function defined on an open interval I ⊂ R is convex

if and only if its derivative is non-decreasing on I.

Examples. Convince yourself that the following are true.

• x2 , x4 are convex on R.
• x3 , x5 are convex on [0, ∞) and concave on (−∞, 0]. The point x = 0 where these functions
make the transition from concavity to convexity (or vice-versa) is called an inflection point.
• sin x is concave on [0, π], cos x is concave on [0, π/2].
√
• log x, x are concave on (0, ∞).
• f (x) = 1/x is convex on the interval (0, ∞) and concave on the interval (−∞, 0).
• Define f (x) = 0 if x ∈ (0, 1) and 1 if x = 0, 1. Then f is convex on [0, 1].

Problem 3. Let I ⊂ R be an interval and f : I → R be a function. Show that f is convex

if and only if for all x, y, z ∈ I, say x < y < z, it holds that

f (y) − f (x) f (z) − f (x) f (z) − f (y)

≤ ≤ .
y−x z−x z−y

Problem 4. Show that a convex function f defined on some open interval I ⊂ R must be
continuous on I. If I is closed, then f may fail to be continuous at the endpoints of I — can
you give any such example?

Problem 5. If f : R → R is a convex function and f (0) ≤ 0, show that f (a + b) ≥ f (a) + f (b)

holds for every a, b > 0.

Problem 6. Suppose that f : R+ → R is a convex function, with lim f (x) = 0. Prove that
x→0
g(x) = f (x)/x (defined for x > 0) is increasing.

Problem 7 (Jensen’s inequality). Let f : X → R be convex. Then for any x1 , x2 , . . . , xn ∈ X

and λ1 , λ2 , . . . , λn ∈ [0, 1] such that λ1 + · · · + λn = 1, prove that
n
! n
X X
f λi xi ≤ λi f (xi ).
i=1 i=1

2
Problem 8. As an application of Jensen’s inequality, prove the (simple) AM–GM inequality,
the weighted AM–GM inequality, or more generally, the power mean inequality.

Problem 9. If A, B, C be the angles of a triangle, show that

√
3 3
sin A + sin B + sin C ≤ .
2
Problem 10. If A, B, C be the angles of an acute triangle, show that

1
cos A cos B cos C ≤ .
8
Problem 11. Let f : [a, b] → R be convex. Then show that
  Z b
a+b 1 f (a) + f (b)
f ≤ f (x) dx ≤ .
2 b−a a 2

Problem 12. Let f : R → R be a twice differentiable function such that

Z x+y
1
f (t) dt = f (x)
2y x−y

for all x ∈ R and y > 0. Show that there exist a, b ∈ R such that f (x) = ax + b for all x ∈ R.

Problem 13. Let ai > 0 for i = 1, 2, . . . , n and a1 + a2 + · · · + an = 1. Prove that for any
k ∈ N,       n
k 1 k 1 k 1 k 1
a1 + k a2 + k . . . an + k ≥ n + k .
a1 a2 an n
Problem 14. Let f : (a, b) → R be a continuous function that satisfies
 
x+y f (x) + f (y)
f ≤
2 2

for every x, y ∈ (a, b). Show that f must be a convex function.

Hints: First show that  x + · · · + x  f (x ) + · · · + f (x )
1 k 1 k
f ≤
k k
for all xi ∈ (a, b), and for every k = 2m , m ≥ 1. Next prove by induction that the above for any k
(further hint: consider 2m−1 ≤ k < 2m ). Next, show that f (λx + (1 − λ)y) ≤ λf (x) + (1 − λ)f (y)
for any x, y ∈ (a, b) and λ of the of the form {k/2n : k = 0, 1, . . . , 2n , n ≥ 1}. Numbers in this set
are called dyadic rationals in [0, 1]. Now show that for any real number λ ∈ [0, 1] we can find a
sequence of dyadic rationals that converge to λ. Hence complete the proof.