Let $D$ be the set of polynomials of integer coefficients $f\in\mathbb{Z}[x]$ such that $f(x)\ge 0$ at $x\in[-2,2]$, where the zero polynomial $f=0$ is excluded. Can I find a finite "minimal" set $F\subseteq D$ on $[-2,2]$? That is, $F$ is finite and for each $f\in D$, there exists $g\in F$ such that $f(x)\ge g(x)$ for every $x\in[-2,2]$. If not, at least can I find a finite minimal set of each degree? I don't know even how I approach this problem, except for the degree 1 case..

  • 1
    $\begingroup$ Why not $F:=\{0\}$? $\endgroup$ – Guest May 17 '15 at 5:35
  • $\begingroup$ Ah, it's my mistake. We must trivially exclude the zero. $\endgroup$ – D. Lee May 17 '15 at 5:37

Let $n$ be a natural number greater than the degrees of every polynomial in $F$, and consider $x^n$. We must have $0 \le g(x) \le x^n$ for all $x\in [-2,2]$. Note that in particular we have $g(0) = 0$, and since $g(x) \ge 0$ in a neighborhood, the nonzero coefficient of lowest degree must be positive. However, $x^n < |g(x)|$ for $x$ suitably close to $0$ for such polynomials $g$, since that lowest coefficient must necessarily be in degree less than $n$. This contradiction shows that no such $F$ can exist.

If you want counterexamples in a specified degree, note that each such polynomial intersects the $x$-axis finitely often, and therefore the set of those $x$ coordinates such that $g(x) = 0$ is finite. You can choose a rational $p/q$ that is not in this set, and then set $f(x) = (qx-p)^2h(x)$, where $h(x)$ is any polynomial that is positive on $[-2,2]$. Then $f(p/q) = 0 < g(p/q)$ for all $g$ in your finite set $F$. This covers every case except the case of linear functions.

In that case choose a point $(r_1,r_2)$ lying below the graphs of every element of $F$ (such a point exists if you take $r_1 = p/q$ as above, and then set $r_2$ less than all values of $g(p/q)$). Then the line between the points $(r_1, r_2)$ and $(-2, 0)$ will provide the counterexample.

  • $\begingroup$ Thank you very much! you're a genius! $\endgroup$ – D. Lee May 17 '15 at 5:55

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.