Writing finite sets of positive integers as sets of positive values of a polynomial

71 Views Asked by At

This question came to me after answering this question about Diophantine sets over at MathOverflow. (It may strike the reader as frivolous, but there you go.) Prompted by this question I started wondering if there were interesting Diophantine subsets $A$ of $\mathbb{N}_{>0}$ for which one could write down relatively simple polynomials $f \in \mathbb{Z}[X]$ such that the set of positive values of $f$ equals $A$ (i.e. $f(\mathbb{Z}) \cap \mathbb{N}_{>0} = A$). It occurred to me that I should start by asking about finite Diophantine sets.

I then found that yes, it is easy to do this: for finite sets $A$ of positive integers, the following polynomial $$ f = X \left( 1 - X \cdot \prod_{a \in A} (X - a)^2 \right) $$ takes the value $x$ at $X=x$ if $x \in A$, and is negative if $x \notin A$. (Proof. If $x>0$ and we let $X=x$ in the expression for $f$, then clearly the second factor in the expression is non-positive, except if $x \in A$, when it equals unity. If $x \leq 0$ on the other hand, then it is clearly positive. Combining the conclusions from both cases, the product comes out negative or zero if $x \notin A$, and as $x$ when $x \in A$.)

So if $A$ has $n$ elements, this gives us a degree $2n+2$ polynomial $f$ such that the set of positive values of $f$ equals $A$. Since I like to be economical in the construction of my polynomials (parsimony is a virtue after all), I wonder whether this number is in fact the best we can do.

For each $n$, do we need a degree $2n+2$ polynomial in general to represent an $n$-element set $A$ of positive integers in the above manner? Put more precisely, for each $n$ does there exist an $n$-element set $A$ of positive integers such that $A$ is not the set of positive values of $f$ for any polynomial $f$ of degree $\leq 2n+1$?