Real Analysis in Functional Equations

Konstantinos Konstantinidis
arXiv:2311.10878v1 [math.HO] 17 Nov 2023

[email protected]
11 November 2023

Acknowledgements
I express my sincere gratitude to my teacher, Athanasios Kontogeorgis, for his meticulous review of this paper and
to Minos Margaritis for his invaluable suggestions.

Contents
1. Introduction 2
2. Squeeze theorem and Limits of Sequences3
3. Supremum and Infimum 6
4. Limits Chasing 14
5. Differentiability 19
6. Problems for Practice 23
7. References 24

1
 Abstract
In this article, we will showcase some analytical concepts that can be used to tackle Functional Equations (FE) in the
positive real numbers domain. Such concepts and related techniques have occasionally appeared in recent High School
Math Olympiads, and they are often accompanied neatly by other known techniques. In each section, we develop a
theoretical background; next, we briefly mention methods that employ the theory and conclude the article by providing
unsolved problems that the reader can try independently.

2
1 Introduction
Essentially, in this article we intrinsically employ two powerful methods: Bounding and Optimizing. In rough terms,
in every example the goal is to find two ”almost equal” functions fn and gn wrapping the function we are looking for, i.e,

fn (x) ≤ f (x) ≤ gn (x) ∀n ∈ N

Then, by letting n → ∞ we can derive the closed form of f .
In order to ”wrap” the function f , we try to ”optimize” the bounds found on the n − 1 step by bounding a side (either
the left or the right) of the given functional by imposing the inequality of the n − 1 step to each term. We simplify and
hope that fn would be a better approximation for f than fn−1 .
Usually, the simply stated ”simplification” can be cumbersome and would be too technical to perform. This is when
the supremum and infimum come in handy. More specifically, rather than following a ”bottom-up” approach, i.e., starting
from f0 , g0 and generating fn , gn , we follow a ”top-down” approach, i.e., we assume that we have already found the func-
tions sup f (x) = fn (x) 1 and inf f (x) = gn (x) that best approximate f and proceed by contradiction. In other words,
any other approximation for f should be worse than the one given by sup f (x), inf f (x).
Furthermore, instead of ”globally” applying this method, i.e., searching for fn , gn such that all x obey the inequality,
we can apply it ”locally”. In other words, we can search for fn , gn such that:

fn (x0 ) < f (x0 ) < gn (x0 )

for a particular x0 and derive the limit l = limx→x0 f (x) and use it to find f . In particular, when x0 is a boundary point
(e.g., x0 = 0 for positive real-valued functions), the monotonicity of f can connect the limit l with the sup or the inf of f .
f (x)
Usually, the solution we expect is of the form f (x) = cx, therefore, it is advisable to apply this method to and
x
f (x) f (x)
constants fn (x), gn (x) to deduce that x or to calculate the limit limx→x0 x , i.e, the derivative of f at x0 . Though
at first glance the limit limx→x0 f (x)
x might not be as useful as the limx→x0 f (x) we have included in this article some
examples that illustrate methods to harness the existence of the derivative to solve the problem, such as the derivation of
a differential equation.
The concepts briefly mentioned previously and further explained in the following sections provide a broad toolbox for
solving even intricate functional equations problems. I hope, therefore, that this article will not only communicate to the
reader this toolbox but also bestow valuable insights for solving such problems.

1
only for the purposes of the introduction, by abuse of notation we assumed that ”sup f (x)” (6= sup f (x)) is a function rather than number. We
mean that” sup f (x)” is the function such that sup f (x) > f (x) and ∀n ∈ N, fn > sup f (x)

3
2 Squeeze Theorem and Limits of Sequences

Definition 2.1: We call x the limit of the sequence xn if the following condition holds: For each real number
ε > 0, there exists a natural number N such that, for every natural number n ≥ N we have:

|xn − x| < ε

Some properties of limits of real sequences that we will be using extensively later are the following (provided that the
limits on the right exist):
Properties of limits

• The limit of a sequence is unique.

• lim (an ± bn ) = lim an ± lim bn
n→∞ n→∞ n→∞

• lim can = c · lim an

n→∞ n→∞

an
 lim an
n→∞
• lim (an · bn ) = ( lim an ) · ( lim bn ) lim = provided limn→∞ bn 6= 0
n→∞ n→∞ n→∞ n→∞ bn limn→∞ bn
h ip
• lim apn = lim an
n→∞ n→∞

• If an ≤ bn for all n greater than some N , then lim an ≤ lim bn .

n→∞ n→∞

• Squeeze Theorem: If an ≤ cn ≤ bn for all n > N and lim an = lim bn = L, then lim cn = L.
n→∞ n→∞ n→∞

• If a sequence is bounded and monotonic, it is convergent.

• A sequence is convergent if and only if every subsequence is convergent.
• If every subsequence of a sequence has its own subsequence that converges to the same point, then the original
sequence converges to that point.
• Bolzano-Weirstrass Theorem: Every bounded sequence has at least one convergent subsequence.
• If lim an → l, f is a continuous function, and an+1 = f (an ), then f (l) = l
n→∞

Squeezing Technique

When we have guessed that the function f we are looking for has the form f (x) = cx, we often try to find two
sequences (an ), (bn ) converging to the same limit such that an x ≥ f (x) ≥ bn x for all n and an arbitrary but fixed
x. By letting n → ∞ yields f (x) = cx where c is the common limit of (an ), (bn ).

Now let’s see an example:

4
 Problem 2.1

Find all functions f : R+ → R+ such that: f (x) + f (f (x)) = 2x for all x ∈ R+ .

Though this problem is prone to another (more standard) method by recurrence relations (see Example 11, Continu-
ation 1, [3]), we give another equivalent approach as a first sight to this technique.

Solution: Fix a real number x. We obviously have that 2x = f (x) + f (f (x)) > f (x), hence

f (x) < 2x (1)

2x
Plugging f (x) in (1) we obtain: f (f (x)) < 2f (x). Hence: f (x) + f (f (x)) = 2x < f (x) + 2f (x) = 3f (x) hence f (x) > 3
(3)
2x
f (x) > (2)
3
2f (x) 2f (x) 5f (x)
Iterating (2), we deduce that f (f (x)) > , and, using (1), we deduce: 2x = f (x) + f (f (x)) > f (x) + =
3 3 3
hence
6x
f (x) < (3)
5
We obviously see a pattern, and now we need to formalize it. We need to find those two sequences (an ), (bn ) such that
bn x < f (x) < an x. To this end, we employ the previous logic with an instead of fixed numbers (by the previous discussion
we can observe that a1 = 2), to specify those sequences.
If f (x) < an x then f (f (x)) < an f (x) hence:

f (f (x)) + f (x) = 2x < an f (x) + f (x)

This yields
2x
f (x) > ∀x
an + 1
2
We set bn = (4)
an + 1
Therefore,
f (f (x)) + f (x) = 2x > bn f (x) + f (x)
f (x) 2 2 2(an + 1) 4
hence it holds that: < . Set an+1 = or (by (4)) an+1 = . Note that: an+1 = 2 − (5).
x bn + 1 bn + 1 an + 3 an + 3
4
Now, using (5), by induction, one can prove that an ≥ 1 for all n. Now, one can easily check that an+1 = 2− ≤ an
an + 3
hence (an ) is (possibly weakly) decreasing. Since (an ) is a bounded decreasing sequence of real numbers, it should con-
verge.
4
Hence, (an ) has a limit. To calculate it, we use the last property of limits, i.e., we solve l = 2 − which yields
l+3
l = 1 or l = −2, with the last case being vacuous, since f (x) ≥ 0 for all x. Hence, since f (x) ≤ an x for all n, tending

5
n → ∞ we obtain f (x) ≤ x.
We can also work the same way with bn , but we may omit this step using the following shortcut:
f (x) ≤ x hence f (f (x)) ≤ f (x), so f (f (x)) + f (x) = 2x ≤ 2f (x) so f (x) ≥ x, hence f (x) = x as desired.

Problem 2.2 - ”A problem that solves itself”

Find all functions f : [0, ∞) → [0, ∞) satisfying the functional equation f (f (x) − x) = 2x∀x ∈ [0, ∞)

Solution: In order for LHS to be defined, we get f (x) ≥ x ∀x. Hence, f (f (x) − x) ≥ f (x) − x ∀x ⇐⇒ f (x) ≤ 3x,
thus x ≤ f (x) ≤ 3x
To construct the sequences mentioned in the previous example, consider an x ≤ f (x) ≤ bn x, to get an (f (x) − x) ≤
2x ≤ bn (f (x) − x) and so:
bn + 2 an + 2
x ≤ f (x) ≤ x
bn an
Hence, the sequences we are looking for are:
bn + 2 an + 2
a1 = 1, b1 = 3, an+1 = (1), bn+1 = (2)
bn an

We see that a1 = 1 < a2 = 35 . Assume that for all i ≤ k the relation ai < ai+1 holds.
By (2) we have that bi < bi−1 for all i ≤ k, hence, by (1), ak+1 > ak . Thus, by induction, an is strictly increasing, and
bn is a strictly decreasing sequence of positive real numbers.
We can manually check that the limit of (an ) is 2, and the limit of (bn ) is 2 too.
And so f (x) = 2x, which indeed satisfies the given relation.

6
3 Supremum and Infimum

Warm-Up Problem 3.1

Does there exist a function f : N → N such that: f (n) = f (f (n − 1)) + f (f (n + 1)) for all n ∈ N?

Solution: On the RHS we have two terms of f adding up to the one term on the LHS. This leads us to suspect that the
range of f should ”flood” in N. We formalize this insight by considering the k = min f (x), and let n = argmaxf (x) ≥ 2.
x>1,x∈N x

Note that k ≥ 2, otherwise 1 = f (n) = f (f (n − 1)) + f (f (n + 1)) ≥ 2, absurd. Now, f (n + 1) ≥ k ≥ 2, hence

k ≤ f (f (n + 1)) = f (n) − f (f (n − 1)) < k, a contradiction. Hence no such function exists.

When the range of a function is discrete and lower/upper bounded, of course, we can consider its minimum/maximum
element, respectively. Now, let us define the notion of supremum and infimum, which generalizes this strategy/construc-
tion to the real numbers.

Definition 2.2: The infimum of a subset S of a set T is the greatest element in T that is less than or equal to all
elements of S, if such an element exists.
Definition 2.3: The supremum of a subset S of a set T is the least element in T that is greater than or equal to
all elements of S, if such an element exists.

Note that an upper bounded set of real numbers has a supremum, and a lower bounded set of real numbers has an
infimum.
We also prove a lemma that we will use extensively later in this section:

”Lemma”

Assume that a function f defined on an (non-trivial) interval A of R has supremum S. Then there is a sequence
(aN ) ⊂ A such that limn→∞ f (an ) = S.

Proof: Define the sequence ǫ1 = 1, ǫn = ǫn−1 10 . There is a real number an ∈ A such that f (an ) > S − ǫn (otherwise,
S − ǫn < S and f (x) ≤ S − ǫn for all x ∈ A contrary to the definition of the supremum) or ǫn > S − f (an ) ≥ 0 (f (x) ≤ S
for all x ∈ A).
As n increases, ǫn → 0, hence, for all ǫ there is a N such that ǫn < ǫ for all n ≥ N . By the construction of (an ) it
follows that 0 ≤ S − f (an ) ≤ ǫn ≤ ǫ for all n ≥ N , hence, by the definition of the limit, lim f (an ) = S.
n→∞
Similarly, we can also prove that the property holds for the infimum.

When considering the supremum and infimum, it is often wise to prove inf f = sup f in some range, to conclude
that it is constant.

7
 Problem 3.2
 
f sup f
Let ∅ 6= A ⊂ R. Find all bounded functions g : A → [1, +∞) such that sup = for all functions
g inf g
f : A → [1, +∞) with f 6= g.

Proof: Let m = inf g, M = sup g. If g is not constant, picking f = g 2 we obtain that mM = sup g 2 = M 2 (which is
self-explained). Thus, m = M a contradiction, since we assume that g is not constant. Hence g is constant.

Problem 3.3

Find all functions f : R+ → R+ such that: f (xf (y) + y) = f (xy) + f (y) for all x, y ∈ R+ .

Solution: Denote by P(x,y) the given relation. If there is a such that f (a) < a, then by P ( a−fa(a),a ) yields f (a) = 0, a
contradiction. Hence f (x) ≥ x for all x.
x xf (y) xf (y)
Now, P ( , y) =⇒ f ( + y) = f (x) + f (y) ≥ +y (1)
y y y
by the previous property.
The relation (1) rewrites as:
f (y) y f (y) f (x)
− ≥ −
x x y x
(2)
The pith of the problem lies in proving that f (1) = 1. Assume the opposite, i.e. that C = f (1) 6= 1. Since f (1) ≥ 1,
we deduce that C = f (1) > 1, or C − 1 > 0.
P (1, y) =⇒ f (f (y) + y) = 2f (y).

f (x)
Claim: There is some M such that > C for all x > M
x

f (x)
Proof. Assume that there is no such M such that > C holds for all x > M . Then, we can construct an increasing, non-
x
f (xi )
bounded, sequence of infinitely many terms (xn ) of positive reals such that ≤ C. By Bolzano Weirstrass Theorem
xi
f (an )
we may consider a sub-sequence of xi , denote it by an such that is convergent.
an
Fix y in (2) and set x = an where n runs through N. This yields:
f (y) y f (an ) f (an ) f (y)
lim ( − + ) = lim > hence
n→∞ an an an n→∞ an y
f (y)
C> for all y
y

f (xf (y) + y) f (xy) + f (y)

Thus, 1 ≤ = < C, where the equality followed from P (x, y). Plugging in y = 1 in the last
xf (y) + y xf (y) + y

8
one, we obtain:
f (x) + C 1 C f (x)
1≤ < C, or C + − < < C. Tending x → ∞, by the squeezing theorem,
xC + 1 x x x
f (x)
lim =C
x→∞ x
Since f (y) + y grows without bound as y → ∞, we have
2f (y)
f (f (y) + y) 2f (y) y 2C
C = lim = lim = lim f (y)
=
y→∞ f (y) + y y→∞ f (y) + y y→∞ C +1
y + 1

hence C = 1 or C = 0, a contradiction in both cases (since C 6= 1).

f (x)
Define D = inf . Notice that D ≥ C > 1. Now, P (1, y), with y > M gives f (f (y) + y) = 2f (y) > D(f (y) + y)
x>M x
or f (y)(2 − D) > Dy. Since Dy > 0 and f (y) > 0, we infer 2 − D > 0 hence

f (y) D
> for all y > M
y 2−D
D
Notice that D < (it boids down to D > 1 which holds), which contradicts the definition of the infimum2 .
2−D
We have now established that C = 1. To finish, we just bash:
P (x.1) =⇒ f (x + 1) = f (x) + 1 (3)
f (x) f (x)
P ( x1 , x) =⇒ f ( + x) = 1 + f (x) ≥ + x hence
x x
(x − 1)(f (x) − x) ≥ 0∀x ∈ R+

(4)
Picking in (4) x < 1 we infer that f (x) ≤ x and since f (x) ≥ x for all x, we infer that f (x) = x for all x ∈ (0, 1) and
(2) implies that f (x) = x for all x, as desired.

9
 Remarks
Remark 1: In the previous proof we used the definition of the supremum or infimum, but, essentially, we established
lower bounds that we made them stronger by ’iterating’ the given functional equation. We could have proven that
C = 1 by more elementary means presented in the previous section as follows:
Now, P (1, y), with y > M gives f (f (y)+y) = 2f (y). Applying the previous claim, 2f (y) = f (f (y)+y) > Cf (y)+Cy
or (2 − C)f (y) > Cy for all y > M .
C
Since Cy > 0 and f (y) > 0, we infer 2 − C > 0 hence f (y) > 2−C . We need to formalize this pattern as explained
in the previous section.
an
Consider the sequence an+1 = , where a1 . By induction on n we infer that f (y) > an y for all n.
2 − an
C
Furthermore, we claim that (an ) is strictly increasing. Indeed, a1 = C > 1 and a2 = and we can check that
2−C
a2 > a1 is equivalent to a1 = C > 1. Assume that an > an−1 for all n < k, hence ak > ... > a1 = C > 1. We can
ak
check that ak < = ak+1 is equivalent to ak > 1 as desired.
2 − ak
Since (an ) is strictly increasing and upper-bounded (an < 2), it converges, and we can calculate the limit of (an )
by the last property of converging sequences. It turns out to be 1.
Hence, an > ... > a1 = C hence 1 = lim an ≥ C, a contradiction.
n→∞
Remark 2: As the reader might have already observed, we can harness the existence of the supremum and/or the
f (x)
infimum on , to surpass the definition of the sequences in the ’Squeezing technique’ explained in the previous
x
section.

Problem 3.4

1
(AOPS) Determine all functions f : R+ → R+ such that f (xy + f (x)) = 2 (f (x) + f (y)), for all x, y ∈ R+ .

Solution: Denote by P (x, y) the given relation.

First of all, we prove that f is upper bounded. Set C = f (1).
f (y) − C
P (1, y) gives f (y + C) − C = (1). Consider a1 = C > u > 0, an = an−1 + C. Obviously, (an ) diverges and by
2
f (an−1 )−C
(1), f (an ) − C = 2 , hence limn→∞ (f (an ) − C) = 0 or lim f (an ) = C.
n→∞
(x)
an − f (x) f (x) + f ( an −f
x ) f (x)
Fix x, and for sufficiently large n, consider P (x, ) to infer that f (an ) = > . Tending
x 2 2
n → ∞, we obtain f (x) ≤ 2C, i.e. f is upper-bounded.

Since 0 < f (x) < 2C, consider N = inf f (x), M = sup f (x). We now bootstrap the sup and the inf:
x>0 x>0

Select two sequences an and bn such that lim f (an ) = M and

n→∞
lim f (bn ) = N .
n→∞
y − f (bn )
Now, P (bn , ) for all y > N + 1, yields:
bn

10
 f ( y−fbn(bn ) ) + f (bn ) M +N
f (y) = lim ≤ ,
n→∞ 2 2
M +N
hence sup f (y) ≤ (1).
y>N +1 2
Similarly, P (an , y−fan(an ) ) for all y > M + 1, yields:

f ( y−fan(an ) ) + f (an ) N +M
f (y) = lim ≥ ,
n→∞ 2 2
M +N
inf f (y) ≥ (2).
y>M+1 2
By (1), (2) we deduce that sup f (x) = inf f (x), hence f is constant in (max(M +1, N +1), ∞),
y>max(M+1,N +1) y>max(M+1,N +1)
say f (x) = D.

Fix x and select y sufficiently large. P (x, y) =⇒ f (x) = D, hence the only functions that satisfy P (x, y) are the
constant ones.

We continue this section with an ”almost-differential” equation that has a marvelous elementary solution that is in
line with the ideas of this section.

Problem 3.5

Determine all functions f : R → R such that f (x) = f ( x2 ) + f ′ (x) x2 ∀x ∈ R.

Solution: For x 6= 0 the given relation can be rewritten as:

f (x) − f ( x2 )
f ′ (x) = x
2

(1)
Let x0 > 0 be fixed. For that particular x0 define the set:

A = {a ∈ (0, x0 ] : f ′ (a) = f ′ (x0 )}

Since A ⊆ [0, x0 ), then A is bounded, hence I = inf A ≥ 0 is well-defined, and there exists a sequence (an )∞
n=1 ∈ A
such that lim an = I.
n→∞

Since f ′ (x) is continuous (owing to the continuity of the RHS of (1)), we obtain that f ′ (x0 ) = lim f ′ (an ) =
n→∞
f ′ ( lim an ) = f ′ (I). Hence, I ∈ A.
n→∞

11
 Assume, now, for contradiction that I 6= 0. Then by (1) and the mean-value theorem we conclude: f ′ (x0 ) = f ′ (I) =
f (I) − f ( 2I )
I
= f ′ (θ), where θ ∈ ( 2I , I). Hence, θ ∈ A and θ < I, which is contrary to the definition of the infimum.
2
Therefore I = 0, and as a result f ′ (x0 ) = f ′ (I) = f ′ (0). Following a similar logic for negative values of x0 (with the
slight modification of considering the sup instead of the inf of the respective set) we may conclude that for every x0 ∈ R
the relation f ′ (x0 ) = f ′ (0) holds, i.e. f ′ (x) = f ′ (0) = c ∀x ∈ R, thus f (x) ≡ cx + d∀x ∈ R, as desired. We can check that
all linear functions satisfy the given functional equation.

Problem 3.6

Determine all functions f : R+ → R such that f (x + y) + f (x + z) − f (x)f (y + z) ≥ 1.

Solution: Denote by P (x, y, z) the given relation. P (2t, t, t) =⇒ 2f (3t) ≥ 1 + f (2t)2 ≥ 1, hence f (x) ≥ 21 , thus f is
lower-bounded.
Now, P (t, t, t) =⇒ (2 − f (t))f (2t) ≥ 1 > 0 (1). Since f (x) ≥ 21 > 0, by (1) we infer that 2 − f (x) > 0∀x, hence f is
upper-bounded.
Denote S = sup f (R+ ), I = inf f (R+ ) ≥ 12 > 0. Now, P (x, y, y) yields:
2 2
2f (x + y) ≥ f (x)f (2y) + 1 ≥ I 2 + 1 hence f (x) ≥ I 2+1 . However, by the definition of I, we should have: I ≥ I 2+1 , i.e.
I = 1.
Pick a sequence (an ) such that lim f (an ) = S. Consider P (an , y, z), which yields 2S ≥ f (an + y) + f (an + z) ≥
n→∞
2S−1
f (an )f (y +z)+1, therefore 2S ≥ f (an )f (y +z)+1, hence 2S ≥ lim f (an )f (y +z)+1 = Sf (y +z)+1, hence f (x) ≤ S .
n→∞
However, by the definition of sup, we have S ≤ 2S−1
S hence S = 1.
All in all, S = I = 1, i.e. f (x) = 1∀x ∈ R+ , which works.

Problem 3.7

Find all continuous functions f : R → (0, +∞) such that:

1
ln(f (f (x))) = f (x) (1), f (−x) = (2)∀x ∈ R
f (x)

Solution: In (2), x=0 gives that f (0) = 1 = e0 .

Equation (1) gives f (t) = et ∀t ∈ f (R). If we manage to prove that f (R) = (0, ∞) then f (t) = et ∀t > 0 and
1
f (−t) = f (t) = e1t = e−t ∀t > 0, i.e. f (x) = ex ∀x ∈ R.
Since f is continuous, f (R) is an interval ∆.
If ∆ is upper-bounded, let sup ∆ = b. Then f (b) = lim f (x) = lim ex = eb . It is a fact that f (b) ∈ ∆, that is, hence
x→b− x→b
eb ≤ b, a contradiction.
1
Let inf ∆ = b ≥ 0. Then, since lim f (x) = lim = lim e−x = 0, we have b ≤ 0, hence b = 0.
x→−∞ x→∞ f (x) x→∞
All in all, ∆ = (0, +∞), qed.

12
 We end this section with a very difficult problem that has recently appeared in an online competition. The following
solution is the official solution which fits to the context of this section:

Problem 3.8

(GIMO 2021 - Aops Mock contest) Determine all functions f : R+ → R+ such that: f (x)f (x + 2f (y)) =
xf (x + y) + f (x)f (y), for all x, y ∈ R+ .

Solution: Denote by P (x, y) the given relation. Making arguments equal to cancel each other out, if there is an a such
that f (a) < a2 , P (a − 2f (a), a) yields (a − 2f (a))f (2a − 2f (a)) = 0 hence f (a) = a2 , a contradiction. If f (a) = a2 , P (a, a)
yields a contradiction, hence f (x) > x2 for all x ∈ R+ .
Furthermore, note that (by the last established property) f (x)f (x + 2f (y)) > xf2(x) + f (x)f (y). Thus, P (x, y) implies
that f (x + y) > f (x)
2 , hence whenever x < y =⇒ f (x) < 2f (y).

Claim 1:f (x) < 2x + c∀x ∈ R+ , for some constant c.

Proof. Note that x + y < x + 2f (y) =⇒ f (x)f (x + 2f (y)) < 2xf (x + 2f (y)) + f (x)f (y).
Fix an a, if f (a) ≤ 2a then we are done. Else,

2f (a)f (y) > 2(f (a) − 2a)f (a + 2f (y)) > a(f (a) − 2a) + 2f (y)(f (a) − 2a)

=⇒ 4af (y) > a(f (a) − 2a) =⇒ f (a) − 2a < 4f (y). The claim thus follows.
x
Identity: Let g(x) := 2f (x) and h(x) := f (x) , then for any n ∈ N, we have

n
X
f (x + g n (y)) = h(x)n f (x + y) + h(x)n−i f (g i−1 (y))
i=1

This can be done by induction, where the base case is P (x, y) with both sides divided by f (x).

Claim 2: limx→∞ h(x) = 1.

Pn
Proof. First, note that since h(x) < 2 for all x > 0, there must exist constants kn for all n such that i=1 h(x)n−i f (g i−1 (1)) <
kn for any x and n.
1
Now, fix an ǫ < 1, then we can choose an n ∈ N such that ǫn < 10 . If for some x we have h(x) ≤ ǫ then we get

x + g n (1) 2x + 2 + c
< f (x + g n (1)) < h(x)n f (x + 1) + kn < + kn
2 10
which can be true only for small enough x. Thus, there exists Nǫ such that h(x) > ǫ whenever x > Nǫ .

13
 On the other hand, if we fix an ǫ > 1, then we can choose an n ∈ N such that ǫn > 10. If for some x we have h(x) ≥ ǫ
then we get
2x + 2g n (1) + c > f (x + g n (1)) > h(x)n f (x + 1) > 5x + 5
which, again, can only be true when x is small enough. Thus, there exists Nǫ such that h(x) < ǫ whenever x > Nǫ .
From the two above parts, we can conclude the claim.

Claim 3: f (x) ≡ x

Proof. Fix an x and let y → ∞. P (x, y) then gives

x f (x + 2f (y)) − f (y)
= lim
f (x) y→∞ f (x + y)
f (x + 2f (y)) f (y)
= lim − lim
y→∞ f (x + y) y→∞ f (x + y)
2y
h(x + y) x + h(y) h(x + y) y
= lim lim − lim lim
y→∞ h(x + 2f (y)) y→∞ x + y y→∞ h(y) y→∞ x + y
   
2
2 h(y) − 1 x
= 1 × lim  − −1×1
y→∞ h(y) x+y

=1×2−1=1

and thus, we are done.

14
4 Limits Chasing
Limits Chasing Technique

If the function we are looking for has domain and co-domain the R+ and a relation we have found for f has a
variable x ”free” (i.e., it is not wrapped by any ”f”), we can set x → 0 (limit) which can yield something nice. If
f is also continuous (rare), we can do this, even if x is wrapped by some f. Moreover, we may as well replace f (x)
by its limit at some point x0 by tending x → x0 , assuming that we have found it (either with the squeeze theorem
or with the theorem that we will discuss later on in this section).

Problem 4.1

Find all functions f : R+ → R+ such that: f (x2 + y 2 ) + 2f (x)f (y) = (x + y)2 , for all x, y ∈ R+ .

Solution: (by Rafail Tsiamis)

By (x + y)2 ≤ 2(x2 + y 2 ) we get f (x2 + y 2 ) ≤ f (x2 + y 2 ) + 2f (x)f (y) = (x + y)2 ≤ 2(x2 + y 2 ), hence

f (x) < 2x ∀x ∈ R+

(1)
2 2 2 2
Now (1) and the initial equation yield Q(x,y):
√ f (x + y ) = (x + y) − 2f (x)f (y) > (x + y) − 8xy =
= x2 − y(6x − y) for all x, y. Considering Q( x − ǫ, ǫ), for a small ǫ gives f (x) ≥ x − ǫA where A is a function of x.
Tending ǫ → 0 we get f (x) ≥ x for all x.
Using P (x, y) we get (x + y)2 = f (x2 + y 2 ) + 2f (x)f (y) ≥ x2 + y 2 + 2xy = (x + y)2 implying that f (x) = x which
finishes the proof.

Sometimes it is not easy to find the limits to employ the technique mentioned previously. The following theorem comes
in handy in some cases:

Theorem 4.1: Limits and bounds

If f : R+ → R is a monotonic function, then for each x0 ≥ 0 the limits limx→x0 + f (x), limx→x0 − f (x) exist, and it
also holds that:
a)If f is increasing:
f (x0 ) ≤ limx→x0 + f (x) = inf x>x0 f (x) and supx<x0 = limx→x0 − f (x) ≤ f (x0 )
b)If f is decreasing: supx>x0 f (x) ≤ limx→x0 + ≤ f (x0 ) and f (x0 ) ≤ limx→x0 − f (x) = inf x<x0 f (x).

This theorem allows us to consider the (left or right) limits and imposes bounds on them. Therefore, not only may we
calculate the limits at specific points but also exclude some of their possible values that don’t obey the bounds imposed.
On a side note, this theorem is a generalization of the ninth property of the limits of sequences, though here, we deal with
limits in real numbers.

15
 Problem 4.2

Find all functions f : R+ → R+ such that: f (xy + f (x)) = f (x)f (y) + x for all x, y ∈ R+ .

Solution 1: (by Athanasios Kontogeorgis) Denote by P (x, y) the given relation. The proof unveils in 5 steps:
Step 1: f (x) ≥ x for all x > 0.
Proof. If there is an a > 0 such that a > f (a), P (a, a−fa(a) ) yields f (a) = f (a)f ( a−fa(a) ) + a > a, contrary to our
assumption that a > f (a).
Step 2: f (x) < 1 for every x < 1

(a) (a)
Proof. Assume that there is an a < 1 such that f (a) ≥ 1. P (a, f1−a ) yields f ( f1−a )(1 − f (a)) = a ≤ 0, a contradiction.
Step 3: f is (weakly) increasing
Proof. The proof utilizes an important technique:
Assume for contradiction that 0 < x2 < x1 with f (x1 ) < f (x2 ). We consider the (positive) ”ratio of change”:
λ = f (xx21)−f
−x2
(x1 )
or x1 λ + f (x1 ) = x2 λ + f (x2 ). By ”f-ing” the latter, we obtain:
−x2 1
f (x1 λ + f (x1 )) = f (x2 λ + f (x2 )) or f (x1 )f (λ) + x1 = f (x2 f (λ) + x2 or f (λ) = f (xx21)−f (x1 ) = λ . Since f (λ) ≥ λ we have
λ ≤ 1. If λ < 1, by Step 2, we obtain f (λ) < 1 or λ1 < 1 or λ > 1, a contradiction.
Hence λ = 1 or f (1) = 1.
Now, P (1, y) =⇒ f (y + 1) = f (y) + 1 and by iterating y → y + 1 we obtain f (y + n) = f (y) + n(1) for all n ∈ N.
P (x, y + 1) yields Q(x, y) : f (xy + x + f (x)) = f (x)f (y) + f (x) + x = f (x) + f (xy + f (x)). By considering Q(x, y−fx(x) )
for y > f (x) we obtain f (y + x) = f (x) + f (y) for all y > f (x).
Having obtained the property of ”restricted additivity”, it is common to extend this via the relation (1). This is how
we do it:
Fix an arbitrary pair (x, y):
Pick an integer n such that y + n > f (x). Hence, f (y + x) = f (y + n + x) − n = f (y + n) + f (x) − n = f (y) + f (x),
hence f is additive.
Since f (x + y) = f (x) + f (y) > f (x), f is strictly increasing, contrary to our assumption that f is not increasing.
Step 4: lim f (x) = 0
x→0

Proof. Since f is increasing, the limit L = lim f (x) = lim f (x) exists and is finite.
x→0+ x→0
By P (x, y) and Step 1 we deduce that: f (x)f (y) + x ≥ xy + f (x) hence f 2 (x) + x ≥ x2 + f (x). Tending x → 0 we
obtain that L2 ≥ L hence L ≤ 0 or L ≥ 1.
If L ≥ 1, then f (x) ≥ L = inf f (x) ≥ 1, a contradiction from Step 2.
x>0
Hence L ≤ 0. However, f (x) > 0, thus L ≥ 0 which implies L = 0, as desired.
Step 5: f (x) = x for all x > 0
Proof. Tending y → 0 in f (x)f (y) + x ≥ xy + f (x) we obtain that x ≥ f (x). This, in conjunction with Step 1 yields that
f (x) = x for all x ∈ R+ .

16
 Solution 2: As previously, f (x) ≥ x∀x.
x
P (x, y) and the previous relation gives: x + f (x)f (y) ≥ xy + f (x) =⇒ f (y) − 1 ≥ (y − 1) (1). Since f (x) ≥
f (x)
x
x =⇒ 1 ≥ (2)
f (x)
x0
Assume that there is x0 such that = m < 1 =⇒ 1 − m > 0. Plugging in (1) x = x0 , yields:
f (x0 )
y y y
f (y) ≥ my − m + 1 > 1 − m =⇒ 0 < < . Now, tending y → 0+ : lim+ = 0. Tending x → 0+ in (1):
f (y) 1−m x→0 f (y)
f (y) ≥ 1, ∀y > 0 =⇒ f ( 21 ) ≥ 1. By P ( 12 , 2f ( 21 )) we have: f (2f ( 21 )) = f (2f ( 21 ))f ( 21 ) + 12 ≥ f (2f ( 21 )) + 12 =⇒ 0 ≥ 1/2, a
contradiction.
x
Hence, f (x) ≥ 1∀x, which, in conjunction with (2), yields f (x) = x, which satisfies P (x, y).

Problem 4.3

(ISL 2020 A8) Determine all functions f : R+ → R+ such that: f (x + f (xy)) + y = f (x)f (y) + 1 for all x,y∈ R+ .

Solution: Denote by P (x, y) the given relation. If there are a, b such that f (a) = f (b), P (1, a) − P (1, b) give a − b = 0
hence a = b, a contradiction. Hence, f is injective.
We claim that x2 > x1 =⇒ f (x2 ) > f (x1 ), i.e. f is strictly increasing. Assume that f (x2 ) ≤ f (x1 ), and since f is
−x1
injective, f (x2 ) < f (x1 ). P ( xy , y) =⇒ Q(x, y) : f ( xy + f (x)) + y = f ( xy )f (y) + 1. Consider λ = f (xx12)−f (x2 ) > 0, or,
x1 x2
equivalently, f (x1 ) + λ = f (x2 ) + λ .
Q(x1 , λ) and Q(x2 , λ) give that f ( xλ1 ) = f ( xλ2 ), a contradiction, since f is injective.
Having proven that f is strictly increasing, the limit lim f (x) = L exists, as well as at every other point, besides 0.
x→0+
On P (x, y), tend x → 0 and notice that x + f (xy) → 0 + L = L, f (x) → L, hence:

lim f (x + f (xy)) + y = lim f (x) + y = C + y

x→0 x→L

(1)
and
lim f (x)f (y) + 1 = Lf (y) + 1
x→0

(2)
By (1), (2) and P (x, y) we obtain that Lf (y) + 1 = C + y for every y ∈ R+ hence f is linear. We can check that only
f (x) = x + 1 works.

17
 Problem 4.4

(Metrix 2020 - Aops Mock contest) Find all functions f : R+ → R+ such that

f (x2 + 2f (xy)) = xf (x + y) + f (xy)

for all x, y ∈ R+

Here is the official solution:

Denote by P (x, y) the given relation.

Claim 1: f (x) > x2 .

Proof. Assume the contrary at x = a.

Consider the equation x2 + 2f (a) = x + xa . As the right side is not smaller than the left side when x = 1, but smaller
when x → ∞, this equation must have a root larger than or equal to 1. Let this root be r. We can see that P (r, ar ) gives
contradiction.

Claim 2: f is non-decreasing.

Proof. Assume the contrary √ that f (b) >

√f (c) for some pair b < c. Let d = 2f (b) − 2f (c).
Consider the equation x + √bx = x + d + √x+d c
. After squaring both sides and clearing some calculation, we can
see that this equation has a positive root k and it satisfies the equation (c − b)k(2k + c + b + 2d) = d(b2 − dk − k 2 ). In
2 2
particular, we
√ haveb dk + k < b . √ √ √ √
Let l = k + k . Comparing P ( k, √bk ) and P ( k + d, √k+d
√ c
) gives ( k + d − k)f (l) = f (b) − f (c) =⇒ 2f (l) =
√ √ √
k + d + k. From claim 1, we would have k + d > √bk or dk + k 2 > b2 , a contradiction.

Claim 2 implies that g(x) = lim+ f (a) exists for all x ≥ 0.

a→x
First, taking x → 0+ in the original equation gives g(2g(0)) = g(0). If g(0) > 0, then g(2g(0)) ≥ f (2g(0)) > g(0), a
contradiction. Thus, g(0) = 0. Taking y → 0+ in the original equation then gives g(x2 ) = xg(x).
f (x + x1 ) f (x)
Now, taking x → ∞ in P ( x1 , x) gives g(f (1)) − f (1) = lim . Let this value be t, then lim =
x→∞ x x→∞ x
1 1
f (x + x ) f (x + x ) x
lim = lim × lim = t.
x→∞ x + 1 x→∞ x x→∞ x + x1
x
f (x + 1) f (x + 1) x+1 g(x)
Since f (x)
x ≤ x ≤
g(x) f (x+1)
x and lim = lim × lim = t, we also have lim = t as well.
x→∞ x x→∞ x + 1 x→∞ x n
x→∞ x
g(x2 ) g(x) g(x2 )
Now, from g(x2 ) = xg(x), we would have g(x) x = x2 , and thus = lim = t for all x > 1. As f (x) ≤ g(x)
n→∞ x2
n
x
and g(x − ǫ) ≤ f (x) for all positive ǫ < x, we must have f (x) = tx for all x > 1 as well.
Finally, P (1, x) implies that f (x) ≡ x. Done.

18
 Problem 4.5

(Danube 2019 P2 Seniors) Find all increasing functions f : R → R such that f (f (x2 ) + y + f (y)) = x2 + 2f (y), for
all x, y ∈ R.

Solution (by Minos Margaritis): Denote by P (x, y) the given relation.

Claim 1: f is injective

Proof. Assume that there are a, b with a < b such that f (a) = f (b) We consider the following cases:
Case 1:. 0 ≤√ a. √
Considering P ( a, y) − P ( b, y) yields what we want.
Case2:. a < 0 < b
Since f is increasing, we have f (x) = f (b)∀x ∈ (0, b), hence, considering a = x0 ∈ (0, b) we reduct this case to the previous
case.
Case 3: a < b ≤ 0
f is obviously not upper bounded (by taking x sufficiently large in P(x,y))
Plugging in x fixed but sufficiently large (such that f (x2 ) + y + f (y) > 0∀y ∈ [a, b]) and varying y in [a,b], where f is
constant, we obtain that f is constant (as y varies) in an interval on the right of zero, which is absurd by the previous
cases. Hence f is 1 − 1.

Claim 2: f is not lower bounded.

Proof. Assume that m=inf(f (x)) exists. Since f is strictly increasing: lim f (x) = m, hence if we consider the sequence
x→−∞
xn as xn = n∀n ∈ N and the sequence yn as yn = −f (x2n ) − n, ∀n ∈ N, we deduce, lim (f (x2n ) + yn + f (yn )) = −∞, thus
n→+∞
lim f (f (x2n ) + yn + f (yn )) = m. Moreover, lim f (f (x2n ) + yn + f (yn )) = lim x2n + 2f (yn ) = +∞ which is absurd.
n→+∞ n→+∞ n→+∞
Hence, f is not lower bounded.
P (x,y0 )
Therefore, ∃y0 , f (y0 ) < 0 ⇒ ∃x0 , f (x0 ) = 0. We have P (0, x0 ) : f (f (0) + x0 ) = f (x0 ) and since f is 1-1 we have
f (0) = 0. Now P (x, 0) =⇒ f (f (x2 )) = x2 , ∀x. Since f is strictly increasing and involutive in R+ , f (x) = x, ∀x ≥ 0.
Hence, P (x, y) reduces to f (x2 + y + f (y)) = x2 + 2f (y). For each y there is an x such that x2 + y + f (y) > 0, hence
y + f (y) = 2f (y) or f (y) = y, ∀y which satisfies our equation.

19
5 Differentiability
Often in FE’s, proving that a function is differentiable, even at some point, can be cumbersome. The following theorem
often comes in handy:

Theorem 5.1: Lebesgue Differentiation Theorem [1]

Every strictly monotone function in R+ is differentiable in R+ /A where A is a set of zero measure. In other words,
a strictly monotone function is differentiable almost everywhere.

For our purposes, we don’t have to be pedantic to explain what a measure of a set is. Usually, we need just one point
of differentiability and the previous theorem ensures the existence of it.

Problem 5.1

(ISL 2005 A2) Find all functions f : R+ → R+ : f (x)f (y) = 2f (x + yf (x)) for all x, y ∈ R+ .

Solution: Denote by P (x, y) the given relation. We claim that f is weakly increasing:
If there exists a pair (x2 , x1 ) such that x2 > x1 and f (x1 ) > f (x2 ), consider λ = f (xx12)−f (x2 )
−x1 > 0, or λf (x1 ) + x1 =
λf (x2 ) + x2 (1).
(1), P (x1 , λ) and P (x2 , λ) give f (λ)f (x2 ) = f (λ)f (x1 ) hence f (x1 ) = f (x2 ), contrary to our assumption that f (x1 ) >
f (x2 ).
Hence, x > y =⇒ f (x) ≥ f (y).

Case 1: f is not strictly increasing:

If there is a pair a > b such that f (a) = f (b) then f must be constant in the interval [b, a]. Then, P (b, fa−b
(b) ) gives that
f ( fa−b
(b) ) = 2 hence, there is an A such that f (A) = 2.
Now, P (A, y) gives that f (y) = f (A + yf (A)) hence f is constant in [y, A + yf (A)] or reversely, for all y. Spanning y
in R+ we conclude that f is constant in an interval [M, ∞).
Fix x and pick y sufficiently large. P (x, y) (f (y) = f (x + yf (x)) since f is constant for sufficiently large y) yields
f (x) = 2, hence f (x) = 2 for all x which satisfies the initial functional equation.

Case 2: f is strictly increasing:

f (x + h) − f (x)
By the Lebesgue Monotone Differentiation Theorem, there is a point x0 such that the limit l = lim
h→0 h
exists and is finite.
f (x0 )f (y)
f (yf (x0 ) + x0 ) − f (x0 ) 2 − f (x0 ) f (y) − 2 f (y) − 2
By P (x0 , y) : lim = l = lim = lim , hence lim = l exists.
y→0 yf (x0 ) y→0 yf (x0 ) y→0 2y y→0 2y
Now, by the previous relation, we can see that the right-limit limh→0+ f (x0 +h)−f h
(x0 )
is independent of x0 , hence
f is right differentiable at every point and the right-derivative equals l at every point.
As for the left limit, consider P (x0 − y, f (x0y−y) ) to deduce that f (x0 − y)f ( f (x0y−y) ) = 2f (x0 ) (1). Therefore:
f (x0 ) − f (x0 − y) (1) f ( f (x0y−y) − 2)
lim = lim 2y . Note that:
y→0 y y→0+
f (x −y)
0

20
 y y f (x0 ) − f (x0 − y) f (y) − 2
f (x0 − y) > f ( x20 ) if y < x0
2 hence 0 ≤ lim ≤ lim x0 = 0, thus y→0
lim = lim = l.
y→0 f (x0 − y) y→0 f( 2 ) y y→0 2y
All in all, f ′ (x) = l for all x hence f is linear and we can check that the only possible solution is f (x) = 2.

Problem 5.2

(BMO 2022 P3) Determine all functions f : R+ → R+ such that: f (y(f (x))3 + x) = x3 f (y) + f (x) for all x, y ∈ R+ .

y 3 y
Solution: As usual, denote by P (x, y) the given relation. Considering P (x, f (x) 3 ) =⇒ f (y + x) = x f ( f (x)3 ) + f (x)

hence f (x + y) > f (x), thus f is strictly increasing.

f (x) − f (x0 )
Hence, by Lebesgue Differentiation theorem, there is a point x0 ∈ R+ such that the limit l = lim
x→x0 x − x0
f (yf (x30 ) + x0 ) − f (x0 ) x30 f (y)
exists and is finite. Now, P (y, x0 ) where y → 0 yields that: lim =l= lim , hence
y→0 yf (x0 )3 f (x0 )3 y→0 y
f (y)
lim =L
y→0 y
exists and is finite.
We prove that f is continuous:
f (y)
As proven, lim = L is finite, hence lim f (y) = 0. Thus,
y→0 y y→0

lim f (yf (x30 ) + x0 ) − f (x0 ) = lim f (y) − f (x0 ) = x30 lim f (y) = 0
y→0 y→x0 + y→0

hence lim = f (x0 ), hence f is right-continuous.

x→x0 +
We prove that f is left continuous as well. To that end, fix x0 ∈ R+ , then P (x0 − y, f (x0y−y)3 ) with y < x0
2 yields:
y y
f (x0 ) − f (x0 − y) = (x0 − y)3 f ( ) < (x0 − y)3 f ( x0 3 )
f (x0 − y)3 f( 2 )
, since f is strictly increasing.
y
Hence, tending y → 0, since lim f ( ) = lim f (x) = 0 we obtain that lim f (x) = f (x0 ), hence f is left-continuous
f ( x20 )3
y→0 x→0 x→x0 −
as well, i.e. continuous. Now there are two continuations we can finish:

Continuation 1: We prove that f is differentiable as well.

f (yf (u)3 + u) − f (u) u3 f (y) f (y) u3 u3
Let u be an arbitrary point. Then: lim 3
= lim 3
= lim =L = A, i.e. the
y→0 yf (u) y→0 yf (u) x→0 y f (u) f (u)
limit exists, hence f is right-differentiable. We claim that f is left-differentiable as well:
By considering P (u − v, v/f (u − v)3 ) we obtain that:

f (u) − f (u − v) (u − v)3 f ( f (u−v)

v
) v
f ( f (u−v) ) (u − v)3
lim = lim = lim v .
v→0 v v→0 v v→0
f (u−v) f (u − v)

Since f is continuous,
(u − v)3 u3
lim =
v→0 f (u − v) f (u)

21
 v
f ( f (u−v) ) f (y) v f (u) − f (u − v)
and lim v = lim = L ( lim = 0 since lim f (u − v) = f (u)), hence lim =
v→0
f (u−v)
y→0 y v→0 f (u − v) v→0 v→0 v
u3
=L = A.
f (u)
Therefore, the right-hand and left-hand limit of the ratio of change of f around u exist and are equal, hence f is
differentiable.
f (yf (x)3 + x) − f (x) f (y) x3 ′ x3
To finish, by P (x, y) we have: = . Tending y → 0, we obtain that f (x) = L
yf (x)3 y f (x)3 f (x)3
hence f ′ (x)f (x)3 = Lx3 for all x ∈ R+ . For concreteness, we manually solve the latest differential equation:
f (x)4 ′ Lx4 ′ 4 4
f ′ (x)f (x)3 = [ ] = [ ] , hence there is a constant c ∈ R such that f (x)4 = Lx4 + c. Tending x → 0, since
4 4
4 +
lim f (x) = 0 and lim x = 0 yields c = 0, hence f (x) = x for all x ∈ R , as desired.
x→0 x→0

Continuation 2: A more natural ending would be by calculating f (1):

Observe that x > 1 =⇒ f (x) ≥ 1. Indeed, if there is an a > 1 such that f (a) < 1, P (a, λ = 1−fa(a)3 ) implies that
f (λ) = a3 f (λ) + f (a) > a3 f (λ) =⇒ a < 1, a contradiction. Since, f (x) > 1 for x > 1 and lim f (x) = 0 < 1 and f is
x→0
continuous, by the intermediate value property, we obtain that there is a z such that f (z) = 1.
P (y, z) =⇒ f (y + z) = 1 + z 3 > f (y) (since f is strictly increasing), equivalently,
1
z3 > 1 −
f (y)

(1)
Since f is not upper-bounded (by P (x, y)) and strictly increasing, we obtain that lim f (x) = ∞. Tending y → ∞ in
x→∞
3 1 1
(1), we deduce that z ≥ lim 1 − = 1 hence z ≥ 1 (2). If f (1) < 1, P (1, 1−f (1)3 ) yields f (1) = 0, absurd. Hence,
y→∞ f (y)
f (1) ≥ 1 = f (z), from which we infer that 1 ≥ z (3).
(2),(3) yield that z = 1, hence f (1) = 1.
P (1, y) yields that f (y + 1) = f (y) + 1 (4) hence, inductively, f (y + n) = f (y) + n (5) for all n ∈ N.
Now, P (x, y + 1), along with (4), yield f (f (x)3 + y + x) = f (x + y) + x3 , hence f (f (x)3 + y) = f (x + y) + x3 for all
y > x. We further extend this property for all x, y:
Fix an arbitrary pair (x, y) and pick an integer n such that y + n > x. Then:
f (f (x)3 + y + n) = f p(y + n) + x3 hence f (f (x)3 + y) = f p (y) + x3 (6) (by (5) applied on both hands).
3
Plugging
p in x → 3
f (x))
p in (6), we deduce that f (f ( 3 f (x)) p + y) = f (y) + f (x)(7). Swapping x,y, we deduce that
3 3 3
f (f ( f (x))
3
p + y) = f (f ( 3
f (y)) + x) and by
p injectivity, f (f ( 3
f (x)) ) = x + c (8) for all x ∈ R+ . Tending in (8) x → 0,
3 3 3
since lim f (x) = 0 we have that lim f (f ( f (x)) = 0, hence c = 0.
x→0 x→0
Now, (7), (8) yield f (x + y) = f (x) + f (y) for all x, y ∈ R+ , which is Cauchy in x, y ∈ R+ . Since f is lower bounded
(f (x) > 0) it must be linear, and we can check that only f (x) = x satisfies the initial functional equation.

22
 Remarks

Remark 1: We can surpass the calculation of c by using the Cauchy-Type Lemma (CTL) mentioned in pg.6 of
[3]. Indeed, f (x + y + c) = f (x) + f (y) =⇒ f (x) = c(x) + d for all sufficiently large x where c(x) is a solution
to Cauchy’s functional Equation. Note that f (x) > 0 hence c(x) > −d hence c is lower-bounded in some interval
[M, ∞) hence c(x) is linear.
Remark 2: For the record, all Cauchy-Type functional equations in R+ have linear solutions. In fact, a stronger
statement holds:
Let f, g, h : R+ → R+ be functions such that

f (g(x) + y) = h(x) + f (y)

g(x)
holds for all x, y ∈ R+ . Then h(x) is a constant function.
For the proof, have a look here.

6 Problems for Practice

Problem 1: Find all functions f : R+ → R+ such that:

f (x + f (x + y)) = f (2x) + y
for all x, y ∈ R+ .

Problem 2: Solve Problem 3.3 without using supremum/infimum.

Problem 3: Find all f : Z+ → R+ such that f (m) ≥ m and

f (m + n) | f (m) + f (n)
+
for all m, n ∈ Z .

Problem 4: Find all functions f : R+ → R+ such that

f (xf (y)) + f (yf (z)) + f (zf (x)) = xy + yz + zx

for all x, y, z ∈ R+ .

Problem 5: Find all functions f : R → R that satisfy the inequality

   
z−y y−x x+z f (x) + f (z)
f (y) − f (x) + f (z) ≤ f −
z−x z−x 2 2

for all real numbers x < y < z.

23
7 References
1. Pang-Cheng W. (2018). Functional Equation.

2. Chen E. (2016). Introduction to Functional Equations

3. Kontogeorgis A., Tsiamis R. (2019). The CDE Method, a Technique in Functional Equations

4. 169 Functional Equations

5. Andreescu T., Boreico I., Mushkarov O., Nikolov N. (2012). Topics in Functional Equations, 2nd Edition, XYZ
Press, LLC

6. Pavardi A. H. (2018). Functional Equations in Mathematical Olympiads (2017 - 2018): Problems and Solutions
(Vol. I), Kindle Edition

7. Venkatachala B.J. (2002). Functional Equations: A Problem Solving Approach, PRISM BOOKS PVT LTD,
Bangalore

24