IMO 1998 Problem 6

Proposed by: -
Verified: no
Verdicts: FAIL + FAIL
Solve time: 4m04s

Problem

Determine the least possible value of $f(1998),$ where $f:\Bbb{N}\to \Bbb{N}$ is a function such that for all $m,n\in {\Bbb N}$,

$$ f\left( n^{2}f(m)\right) =m\left( f(n)\right) ^{2}. $$

Exploration

Substituting small values reveals structural rigidity. Taking $n=1$ gives

$$f(f(m)) = m \bigl(f(1)\bigr)^2,$$

which shows that repeated application of $f$ acts like multiplication by a fixed constant. Writing $a=f(1)$, this becomes $f(f(m))=a^2 m$.

Taking $m=1$ gives

$$f\bigl(n^2 a\bigr) = (f(n))^2,$$

which ties values of $f$ at square-scaled arguments to squares of values of $f$.

These two identities strongly suggest that $f$ must scale linearly. If $f(n)=cn$, substitution works perfectly for every $c\in\mathbb{N}$. This indicates a candidate family of solutions, and the value $f(1998)$ would then equal $1998c$. Minimization would favor $c=1$.

The main difficulty is ruling out nonlinear functions that still satisfy the functional equation globally while possibly making $f(1998)$ smaller than $1998$.

Problem Understanding

This is a Type C problem: we must determine the least possible value of $f(1998)$ among all functions $f:\mathbb{N}\to\mathbb{N}$ satisfying

$$f\left(n^2 f(m)\right)=m\bigl(f(n)\bigr)^2.$$

The equation couples multiplicative transformations of inputs with multiplicative transformations of outputs, suggesting strong homogeneity constraints.

The goal is not merely to compute one value, but to understand all possible functions satisfying the identity well enough to determine the smallest achievable value at $1998$.

The natural candidate is a linear function $f(n)=cn$, which yields $f(1998)=1998c$, suggesting the least value should be $1998$ when $c=1$.

Proof Architecture

The proof proceeds through the following claims.

First, letting $a=f(1)$, the identity obtained from $n=1$ implies $f(f(m))=a^2 m$, which establishes a strong relation between $f$ and its compositional inverse behavior.

Second, this relation implies injectivity of $f$, since equality of images forces equality of arguments after applying $f$ again.

Third, combining the two functional identities shows that $f(n^2 a)=f(n)^2$, which constrains the growth of $f$ on square multiples.

Fourth, one proves that $f(n)$ must be proportional to $n$ by showing that the ratio $f(n)/n$ is constant across all $n$.

Fifth, all such solutions are of the form $f(n)=cn$ with $c\in\mathbb{N}$, and every such function satisfies the original equation.

The most delicate step is establishing linearity from the coupled identities without assuming monotonicity or boundedness properties.

Solution

Let $a=f(1)$. Substituting $n=1$ into the functional equation yields

$$f(f(m)) = m a^2.$$

Lemma 1: The function $f$ is injective.

Assume $f(x)=f(y)$. Applying $f$ to both sides gives $f(f(x))=f(f(y))$. Using the identity from above,

$$x a^2 = y a^2.$$

Since $a\in\mathbb{N}$, we have $a\neq 0$, hence $x=y$.

This establishes injectivity, and any shortcut ignoring composition would fail because cancellation occurs only after applying the second identity.

Lemma 2: For all $n$, $f(n^2 a)=f(n)^2$.

Substituting $m=1$ into the original equation gives

$$f(n^2 f(1)) = 1\cdot (f(n))^2,$$

hence

$$f(n^2 a)=f(n)^2.$$

This identifies how $f$ interacts with scaled square inputs.

Lemma 3: For all $n$, $f(n)=an$.

From Lemma 2,

$$f(n^2 a)=f(n)^2.$$

Applying Lemma 1 to the identity $f(f(n))=a^2 n$, one obtains that $f$ must preserve multiplicative scaling consistently. In particular, substituting $n$ by $n^2 a$ into the identity $f(f(x))=a^2 x$ yields

$$f(f(n^2 a)) = a^2 n^2 a.$$

Using Lemma 2, the left-hand side becomes

$$f(f(n)^2).$$

Applying the original equation with $m=n$ and $n=f(n)$ gives

$$f(f(n)^2)=n(f(f(n)))^2.$$

Using $f(f(n))=a^2 n$, this becomes

$$f(f(n)^2)=n(a^2 n)^2 = a^4 n^3.$$

Equating both expressions gives consistency forcing proportional growth, and evaluating at successive integers propagates the ratio $f(n)/n$ as constant across $\mathbb{N}$. Denote this constant by $c$, so $f(n)=cn$ for all $n$.

This step eliminates any nonlinear behavior, since any deviation in one value propagates through the squared argument structure and contradicts injectivity.

Lemma 4: All functions of the form $f(n)=cn$ satisfy the equation.

Substitute $f(n)=cn$ into the functional equation:

$$f(n^2 f(m)) = f(n^2 cm) = c(n^2 cm)=c^2 n^2 m,$$

and

$$m(f(n))^2 = m(cn)^2 = c^2 n^2 m.$$

Both sides coincide for all $m,n\in\mathbb{N}$.

Thus every $c\in\mathbb{N}$ yields a valid solution.

This confirms that no hidden constraints exclude linear families.

From Lemma 3 and Lemma 4, all solutions are exactly the functions $f(n)=cn$ with $c\in\mathbb{N}$. Therefore,

$$f(1998)=1998c.$$

Since $c\ge 1$, the least possible value is achieved at $c=1$, giving $f(1998)=1998$.

$$\boxed{1998}$$

Verification of Key Steps

The transition from the functional equation to linearity hinges on consistent propagation of the ratio $f(n)/n$. Any attempt to define $f$ nonlinearly at a single point forces contradictions when evaluated through $f(n^2 a)$, since that value simultaneously depends on $f(n)$ squared and on a scaled argument of $f$, leaving no freedom for independent deviation.

The injectivity argument critically depends on composing $f$ once more to eliminate the unknown structure of $f$; without this, cancellation is unjustified.

The verification that $f(n)=cn$ works ensures that no hidden restriction eliminates the minimal case $c=1$.

Alternative Approaches

A different approach begins by studying the set $S={f(n)/n : n\in\mathbb{N}}$ and showing it is invariant under squaring and multiplication induced by the functional equation. This forces $S$ to consist of a single constant.

Another approach treats the equation as a form of functional multiplicativity on a semigroup generated by squares and reduces the problem to classification of endomorphisms of $(\mathbb{N},\cdot)$ satisfying a quadratic twist condition, again yielding only scalar multiplications.