5
$\begingroup$

Let $f(x,y)$ be a polynomial with integer coefficients, and let $\alpha=(\alpha_1,\alpha_2)\in \mathbb{C}^2$ be a complex point. I want to show that $f$ cannot vanish at $\alpha$ to high order unless $\alpha$ is ``simple''. More precisely, suppose that

  • f(x,y) vanishes to order at least $0.99\cdot \deg f$ at $(\alpha_1,\alpha_2)$, and
  • f(1,1)=1

Does it imply existence of an integer polynomial $g(x,y)$ satisfying $g(\alpha_1,\alpha_2)=0$ and $g(1,1)=1$ that is linear, i.e., of degree $1$?

``Vanishing to order $m$'' at $\alpha$ means that all possible partial derivatives of order at most $m-1$ vanish at point $\alpha$.

Note that the single-variable version is a simple consequence of Gauss's lemma.

I previously asked an analogous question over $\mathbb{C}$, but the answer turned out to be contrary to what I expected. This is the original question that I am interested in.

I do not know how to answer this question even if one replaces $\mathbb{Z}$ by $\mathbb{Q}$ throughout.

$\endgroup$
14
  • 1
    $\begingroup$ In the one-variable case, the condition that $f(\alpha)$ vanishes implies that $\alpha$ is algebraic over $\mathbb Q$. In your situation you're only assuming that $\alpha_1$ and $\alpha_2$ are algebraically dependent over $\mathbb Q$. If you're willing to assume that they are both algebraic over $\mathbb Q$, it might be easier to prove what you want. Or do you really need to allow the $\alpha_i$'s to be transcendental? $\endgroup$ Jun 22, 2015 at 2:51
  • $\begingroup$ Maybe I am confused, but there is a general way involving multiplier ideals to do something like this. Suppose that there is a polynomial of degree $d$ in $\mathbb P ^n$ vanishing at points $Z=\{z_1,\ldots , z_m\}$, then there is a polynomial of degree $\leq nd/k-1$ (see A on page 76 of the paper by Esnault-Viehweg mi.fu-berlin.de/users/esnault/preprints/ec/minoration.pdf). For $n=2$ (this case is due to Chudnoski) and $k=.99d$ this bound is $[2/.99-1]=2$. So for vanishing $\geq deg f$ it would have worked. There may be way around this using LTC's. If important I can look into it. $\endgroup$
    – Hacon
    Jun 22, 2015 at 6:22
  • $\begingroup$ Likely I am missing something, but are you asking for which $\alpha_1,\alpha_2$ exists linear g(x,y) with the desired properties? This appears system of linear equations in two variables. $\endgroup$
    – joro
    Jun 22, 2015 at 12:00
  • $\begingroup$ Isn't $\alpha_1=\alpha_2=1,f(x,y)=1/2\,{x}^{2}-1/2\,{y}^{2}-x+y$ counterexample? $\endgroup$
    – joro
    Jun 22, 2015 at 12:32
  • $\begingroup$ @JoeSilverman The case of transcendental $\alpha_i$'s is actually the easier case. Indeed, then the ideal of polynomial relations that $\alpha_i$'s satisfy is principal, say $I=(h)$, and $f$ vanishes to order $m$ iff $f$ is divisible by $h^m$. $\endgroup$
    – Boris Bukh
    Jun 22, 2015 at 13:03

2 Answers 2

3
$\begingroup$

Over $\mathbf{Q}$ you can argue as follows.

Case I. $a$ is the unique point where $f$ vanishes to high order. Then $a$ is invariant under $\text{Aut}(\mathbf{C})$ and hence defined over $\mathbf{Q}$ and hence there is a line defined over $\mathbf{Q}$ passing through $a$.

Case II. There are exactly $2$ points $a, b$ where $f$ vanishes to high order. Then the line $ab$ is defined over $\mathbf{Q}$.

Case III. There are finitely many $> 2$ points where $f$ vanishes to high degree. This is impossible because then every conic passing through these points must be contained in $f = 0$. (Use Bezout.)

Case IV. There is an infinite nr of points where $f$ vanishes to high degree. The locus of these points has to be an algebraic curve $g = 0$ and $g$ has to have degee $1$ because a high power of $g$ divides $f$.

Since $\mathbf{Z} \subset \mathbf{Q}$ this also proves something over $\mathbf{Z}$ and with a variant of Gauss's lemma you can get a suitable statement, but I haven't thought it through. Maybe Case I is the most interesting one from this point of view.

$\endgroup$
2
  • $\begingroup$ This is very nice and helpful! This shows a direction to get what I want. $\endgroup$
    – Boris Bukh
    Jun 22, 2015 at 13:21
  • $\begingroup$ How should I acknowledge this answer in a paper? You can e-mail me, or otherwise I can refer to you by the penname "Dracula". $\endgroup$
    – Boris Bukh
    Jul 28, 2015 at 9:28
1
$\begingroup$

The solution to this question appeared in Section 4 of my paper on the ranks of matrices with few distinct entries. The solution borrows an idea from the Dracula's answer. Please upvote his answer. Below is a sketch of the solution.

Step 1) Let $V$ be the set of points where $f$ vanishes to order greater than $\tfrac{2}{3}\deg f$. If $l$ is any secant of $V$, then $f$ vanishes to the order exceeding $\tfrac{1}{3}\deg f$ on each point of $l$. Hence, $f$ vanishes to the order $\mathord{\geq} \tfrac{1}{3}\deg f$ on the secant variety $\operatorname{Sec}_1(V)$. Similarly $f$ vanishes on any line passing through a point of $V$ and a point of $\operatorname{Sec}_1(V)$. So, $f$ vanishes on the $2$'nd secant variety $\operatorname{Sec}_2(V)$. It is clear that $\operatorname{Sec}_2(V)$ is an affine subspace defined over $\mathbb{Q}$.

Step 2) If there is no linear polynomial $g$ that vanishes on $\operatorname{Sec}_2(V)$ and satisfies $g(1,1)=1$, then in fact there is a rational point $v\in \operatorname{Sec}_2(V)$ for which no such polynomial vanishing at $v$ exists. This is proved by using a theorem of van der Waerden characterizing when a system of linear equations in integers admits no solution.

Step 3) One then shows that there is no integer polynomial $g$ satisfying $g(1,1)=1$ vanishing at $v\in \mathbb{Q}^2$ if and only if $v\equiv (1,1)\pmod p$ for some prime $p$. In that case, however, $f(v)\equiv f(1,1)\equiv 1 \pmod p$, contradicting the fact that $f$ vanishes on $\operatorname{Sec}_2(V)$.

Furthermore The bound $\tfrac{2}{3}\deg f$ is sharp. Given a cubic Galois extension $F/\mathbb{Q}$, one can find three points $p_1,p_2,p_3\in F^2$ and three linear polynomials $f_1,f_2,f_3$ with the following properties:

  • The points $p_1,p_2,p_3$ are Galois conjugates of one another,
  • The linear polynomials $f_1,f_2,f_3$ are Galois conjugates of one another,
  • The linear polynomial $f_1$ vanishes on the line $p_2p_3$, the $f_2$ vanishes on $p_1p_3$, and $f_3$ --- on $p_1p_2$,
  • The coefficients of $f_1,f_2,f_3$ are algebraic integers, and $f_1(1,1)=f_2(1,1)=f_3(1,1)=1$.

Then the polynomial $f=f_1f_2f_3$ has integer coefficients, and vanishes to order $2$ at each of $p_1,p_2,p_3$. It also satisfies $f(1,1)=1$. However, no integer linear polynomial $g$ vanishes at $p_1$, say.

A concrete example that one obtains using the splitting field of $x^3-3x+1$ consists of the polynomial $f(x,y)=x^3+3 x^2 y-6 x^2-3 x y+3 x-y^3+3 y+1$ and points $p_i=(\rho_i^2-2,3-\rho_i^2-\rho_i)$ where $\rho_1,\rho_2,\rho_3$ are the roots of $x^3-3x+1$.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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