Which Polynomials Map ℚ Onto ℚ? The Complete Answer

Here is a challenge with a one-line statement and an 18-minute fight in it: build a polynomial that eats fractions and produces exactly the fractions. All of them, nothing missing, nothing extra. Squaring looks promising — a fraction squared is a fraction — until you ask it to output 2. That would need $\sqrt{2}$ as an input, and $\sqrt 2$ is famously not a fraction. The fractions crowd in on it forever and keep missing. Staying inside $\mathbb{Q}$ is easy. Reaching all of it is the real fight.

The lines join the club immediately: $P(x) = ax + b$ with $a, b \in \mathbb{Q}$, $a \neq 0$, maps rationals to rationals and solves backwards inside $\mathbb{Q}$. The claim, and the rest of this article, is that the club has no other members. The proof runs three filters, and each one is a tool worth keeping.

Filter zero: the coefficients must be rational

A degree-$n$ polynomial has $n+1$ unknown coefficients, and each evaluation buys one equation. So evaluate at $n+1$ rational points and lean on interpolation: a polynomial of degree at most $n$is pinned completely by its values there, because two candidates that agree would differ by a polynomial with more roots than its degree allows. Lagrange’s formula even hands you the coefficients, built from nothing but the nodes and values, using only subtraction and division. Rational values in, rational coefficients out.

Worth a pause: the same statement is false one floor down, over the integers. $\tfrac{x(x-1)}{2}$ sends every integer to an integer — one of two consecutive integers is even — yet its coefficients are halves. The rational version survives because fractions let you divide. Keep that little polynomial in mind; it returns at the end.

Filter one: shape executes the even degrees

Even degree means both ends of the graph run off in the same direction, so the curve has a global minimum (or maximum, if the leading coefficient is negative). Every value past that extremum is never reached by any real input, let alone a rational one, and that half-line contains infinitely many rationals. Coverage collapses on shape alone. Arithmetic never has to wake up.

Filter two: the denominator fence

Odd degrees shrug at shape: a cubic sweeps every real height. So the last filter is arithmetic. Clear denominators so the polynomial has integer coefficients with leading coefficient $a$, feed it a fraction $p/q$ in lowest terms, and suppose the output is a whole number. Multiply everything out and every term picks up a factor of $q$ except the lone survivor $a\,p^n$. Since $q$ shares no factor with $p$, the conclusion is brutal:

This is the engine inside the rational root theorem, run at full power. The rationals are a dense mist, but the rationals allowed to land on whole numbers form a fence: evenly spaced points with a fixed gap of $1/a$ between neighbors. Rigid, sparse, and it never changes.

The collision

Now aim the two facts at each other. For odd degree at least 3 (leading coefficient positive, after flipping signs if needed), the curve eventually climbs forever, and every large integer height $k$ is crossed exactly once. Coverage forces that crossing point to be rational, and (1) forces it onto the fence. But the mean value theorem says consecutive crossing points are $1/P'(\xi)$ apart for some $\xi$ in between, and the derivative of a degree-3-or-more polynomial grows without bound. The gaps shrink toward zero, eventually below $1/a$ — two distinct fence points forced closer than the fence spacing. Impossible. Odd degrees from three up are gone.

The complete list

Constants die because their image is a point. Irrational coefficients die at filter zero. Even degrees die at filter one. Odd degrees three and up die at filter two. What survives is exactly what we found on day one:

The method matters more than the list. Interpolation turns values into coefficients; the denominator bound turns coverage into a grid; and hard problems fall where analysis (growth) and arithmetic (divisibility) close like a pincer. The same pincer shows up across olympiad number theory, and in disguise in other classification proofs where one stubborn structure survives every filter.

Your turn, same war on the next battlefield: find every polynomial mapping $\mathbb{Z}$ exactly onto $\mathbb{Z}$. Careful — $\tfrac{x(x-1)}{2}$passes the first front without integer coefficients, and its outputs 0, 1, 3, 6, 10 never say 2. Exactly one idea from this article finishes it. The video’s comment section collects the cleanest proofs.