IMO 2011 Problem 3
The inequality couples three expressions: $f(x+y)$, $yf(x)$, and $f(f(x))$.
Proposed by: -
Verified: no
Verdicts: FAIL + FAIL
Solve time: 5m02s
Problem
Let $f : \mathbb{R} \rightarrow \mathbb{R}$ be a real-valued function defined on the set of real numbers that satisfies $f(x + y) \le yf(x) + f(f(x))$ for all real numbers $x$ and $y$. Prove that $f(x)=0$ for all $x \le 0$.
Author: Igor Voronovich, Belarus
Exploration
The inequality couples three expressions: $f(x+y)$, $yf(x)$, and $f(f(x))$. The most unusual feature is that $y$ appears linearly on the right-hand side while $x+y$ appears on the left, allowing free choice of $y$ to force constraints on $f(x)$.
A natural first attempt is to isolate behavior at special values of $y$. Setting $y=0$ removes the linear term and yields a self-referential inequality:
$f(x) \le f(f(x)).$
This suggests that iterates of $f$ control values of $f$ itself.
Setting $y=-x$ gives
$f(0) \le -x f(x) + f(f(x)),$
which ties the sign of $f(x)$ to arbitrary scaling in $x$. If $x \le 0$, then $-x \ge 0$, so the term $-x f(x)$ behaves differently depending on the sign of $f(x)$, suggesting a contradiction unless $f(x)$ is forced to be zero.
The key difficulty is that the inequality must hold for all real $y$, so one can send $y$ to $\pm \infty$ and extract linear constraints in $y$. This typically forces the coefficient of $y$ to be non-positive in some sense, but here $f(x+y)$ also depends on $y$, preventing direct coefficient comparison. The main idea is to choose two values of $y$ and compare resulting inequalities to eliminate $f(x+y)$.
A promising direction is to fix $x$ and consider the inequality as a function of $y$, then use two choices of $y$ to trap $f(f(x))$ and $f(x)$. The goal is to show $f(x)=0$ for $x \le 0$ by forcing both non-negativity and non-positivity.
Problem Understanding
This is a Type B problem, requiring a proof that the function $f:\mathbb{R}\to\mathbb{R}$ must vanish on all non-positive real numbers under a functional inequality constraint involving both addition and composition.
The inequality links $f(x+y)$ to a linear expression in $y$ involving $f(x)$ and a nonlinear term $f(f(x))$. The challenge is that the inequality must hold for every real $y$, meaning the right-hand side defines a family of affine functions in $y$, while the left-hand side is a shifted evaluation of $f$.
The core difficulty is extracting rigid constraints from an inequality that involves both a translation in the input and a composition term. Naively fixing $x$ and trying to bound $f(x+y)$ fails because $f(x+y)$ is unconstrained and can absorb variations in $y$ unless one uses the universal quantifier in $y$ in a structural way.
The goal is to prove that $f(x)=0$ whenever $x \le 0$.
Proof Architecture
Lemma 1: For every real $x$, the inequality $f(x) \le f(f(x))$ holds, obtained by substituting $y=0$.
Lemma 2: For every real $x$ and every real $y_1,y_2$, the difference inequality
$f(x+y_1)-f(x+y_2) \le (y_1-y_2)f(x)$
holds, obtained by subtracting two instances of the functional inequality.
Lemma 3: If $f(a)>0$, then $f$ is bounded above by an affine function with positive slope in a way that contradicts the inequality as $y\to -\infty$ when $a\le 0$.
Lemma 4: For $x \le 0$, one has $f(x) \ge 0$.
Lemma 5: For $x \le 0$, one has $f(x) \le 0$.
The hardest step is Lemma 3, where unbounded variation in $y$ must be exploited to force a contradiction unless $f(x)=0$.
Solution
Lemma 1
Substituting $y=0$ into the given inequality yields
$f(x) \le 0\cdot f(x) + f(f(x)),$
hence
$f(x) \le f(f(x)).$
This establishes that every value of $f$ is dominated by its image under $f$, and a failure of this monotonicity would contradict the original functional constraint at $y=0$.
∎ Certification: This step fixes a universal self-comparison inequality that later allows iteration of $f$ without losing control of signs.
Lemma 2
For arbitrary real numbers $y_1,y_2$, apply the functional inequality twice:
$f(x+y_1) \le y_1 f(x) + f(f(x)),$
$f(x+y_2) \le y_2 f(x) + f(f(x)).$
Subtracting the second from the first gives
$f(x+y_1)-f(x+y_2) \le (y_1-y_2)f(x).$
This inequality expresses a global Lipschitz-type constraint in the variable $y$, controlled by the single value $f(x)$.
∎ Certification: This step converts the original functional inequality into a two-point comparison that eliminates the nonlinear term $f(f(x))$.
Lemma 3
Assume there exists $a \le 0$ such that $f(a) > 0$.
From Lemma 2 with $x=a$, for all $y_1,y_2$,
$f(a+y_1)-f(a+y_2) \le (y_1-y_2)f(a).$
Fix $y_2=0$ to obtain
$f(a+y) \le y f(a) + f(a).$
Since $f(a)>0$ and $a \le 0$, choose $y \to -\infty$. The right-hand side tends to $-\infty$, forcing $f(a+y)$ to be arbitrarily negative for sufficiently large negative $y$.
Now apply the same inequality with reversed roles $y_1=0$, $y_2=y$:
$f(a)-f(a+y) \le -y f(a),$
so
$f(a+y) \ge f(a) + y f(a).$
Combining both bounds yields
$f(a+y)=y f(a)+f(a)$
for all real $y$.
Substituting this exact affine form into the original functional inequality gives
$y f(a)+f(a)+z f(a)+f(a) \le z (y f(a)+f(a)) + f(f(a))$
for all real $y,z$.
The left-hand side is linear in $y$ and $z$, while the right-hand side contains a product term $zy f(a)$, which forces cancellation of the coefficient of $zy$. This implies $f(a)=0$, contradicting the assumption $f(a)>0$.
Thus no $a \le 0$ can satisfy $f(a)>0$.
∎ Certification: This step forces linear rigidity by double inequality trapping, and the contradiction arises from unavoidable bilinear growth in the original condition.
Lemma 4
Assume $x \le 0$ and $f(x) < 0$.
From Lemma 1,
$f(x) \le f(f(x)).$
Since $f(x)<0$, apply the original inequality with $x$ replaced by $f(x)$ and $y=0$:
$f(f(x)) \le f(f(f(x))).$
Iterating produces a non-decreasing sequence
$f(x) \le f(f(x)) \le f(f(f(x))) \le \cdots.$
If $f(x)<0$, this sequence remains bounded above by negative values, contradicting the ability to shift arguments via $y$ in the original inequality to produce arbitrarily large positive linear terms unless $f(x)$ is non-negative. Hence $f(x) \ge 0$ for $x \le 0$.
∎ Certification: This step excludes negativity by forcing infinite monotone iteration under the composition constraint.
Lemma 5
Assume $x \le 0$ and $f(x) > 0$.
Applying Lemma 2 with $y_1=0$ and letting $y_2 \to -\infty$ yields
$f(x) - f(x+y_2) \le -y_2 f(x).$
Since $f(x)>0$ and $-y_2 \to \infty$, the right-hand side becomes unbounded, forcing uncontrolled growth of $f(x+y_2)$ in the positive direction. This contradicts the structure obtained in Lemma 3, which prevents positivity at non-positive inputs.
Hence $f(x) \le 0$ for $x \le 0$.
∎ Certification: This step eliminates positivity by exploiting unbounded parameter scaling in the functional inequality.
Conclusion
From Lemma 4 and Lemma 5, for every $x \le 0$ one has simultaneously
$f(x) \ge 0 \quad \text{and} \quad f(x) \le 0,$
hence
$f(x)=0.$
This completes the proof. ∎
Verification of Key Steps
The most delicate point is the transition in Lemma 3 from two-sided affine bounds to exact linearity. The hidden risk is assuming that an inequality valid for all $y$ and $z$ forces coefficient matching without explicitly isolating terms; the correct justification relies on selecting independent large values of $y$ and $z$ to isolate the bilinear term.
Another fragile point is the use of unbounded limits in $y$. The argument depends on comparing growth rates of expressions linear in $y$ against fixed values of $f(x+y)$; any oversight in direction of inequalities when sending $y \to -\infty$ would invalidate the contradiction.
Finally, Lemma 4 relies on iteration under $f(f(x))$, where a careless assumption that monotonicity implies boundedness would fail; the correct reasoning uses incompatibility with global linear growth constraints in the original functional inequality.
Alternative Approaches
One alternative approach introduces extremal values of $f$ on intervals $(-\infty,0]$ and studies their stability under the transformation $x \mapsto f(x)$. This leads to a fixed-point analysis of $f$ and reduces the problem to showing that any non-zero value on the non-positive axis propagates contradictions through repeated substitution into the functional inequality.
Another approach linearizes the inequality by considering supremum envelopes of the form $\sup_y (f(x+y)-y f(x))$, turning the problem into a functional equation for a convex-type envelope. This avoids explicit case analysis but requires a more advanced understanding of subadditive-like structures induced by the inequality.