6
$\begingroup$

Question: Let $k$ be a field, let $f \in k[t_1,\ldots,t_N]$ be a nonzero polynomial. Which monomials $m_a = t_1^{a_1} \cdots t_N^{a_n}$ appearing in $f$ are leadable in the sense that they are the leading monomial with respect to some monomial order $\preceq$?

Guess at the Answer: An obvious necessary condition is that there can't be another monomial $m_b = t_1^{b_1} \cdots t_N^{b_N}$ in $f$ with $a_i \leq b_i$ for all $i$ and $\sum_i a_i < \sum_i b_i$. Indeed, if this holds, then $m_a \prec m_b$ for every monomial ordering $\preceq$. My guess (and hope; this would have some favorable consequences in my current work) is that the converse is also true.

Order-Theoretic Translation: Let $\mathbb{N}$ denote the non-negative integers. On the commutative monoid $\mathbb{N}^N$, let $\leq$ be the natural "additive divisibility" partial ordering: $(x_1,\ldots,x_N) \leq (y_1,\ldots,y_N)$ if and only if $x_i \leq y_i$ for all $i$.
Let $S$ be a finite subset of $\mathbb{N}^N$, and let $a \in S$ be a maximal element. Then $\leq$ extends to an admissible (i.e., compatible with the monoid structure) well-ordering $\preceq$ on $\mathbb{N}^N$ with respect to which $a$ is the top element.

Comments:

  • My guess is certainly correct when $N = 1$; this is a trivial case.

  • I finally started learning about Grobner bases this week! Better late than never. So I am very far from being an expert here. I'm aware of a result of Robbiano that parametrizes all monomial orders on $\mathbb{N}^N$ in terms of certain real matrices. So I suppose the idea is to choose the matrix carefully so as to make any $\leq$-maximal monomial the leading monomial. But I haven't even thought carefully about the monomial order associated to a given matrix, so it is not yet clear to me how to proceed. I am however pretty confident that, provided only that my guess is correct, this must be a standard result in the literature, and I would be happy with a reference.

$\endgroup$
6
  • 3
    $\begingroup$ Nice and very natural question! A monomial $m_a = t_1^{a_1} t_2^{a_2} \cdots t_n^{a_n}$ of $f$ is leadable if and only if the row vector $\left(a_1,a_2,\ldots,a_n\right) \in \mathbb{Q}^n$ is a vertex of the convex hull of $M$, where $M$ is set consisting of the zero vector $0 \in \mathbb{Q}^n$ along with all vectors $\left(b_1,b_2,\ldots,b_n\right) \in \mathbb{Z}^n$ for which the monomial $t_1^{b_1} t_2^{b_2} \cdots t_n^{b_n}$ appears in $f$. I would guess this is somewhere in a book like Sturmfels's on Gröbner bases or Maclagan's on toric varieties. $\endgroup$ Commented Jul 1, 2023 at 20:32
  • 3
    $\begingroup$ Brief outline of the proof: First direction: If $m_a$ is leadable, but the vector $v_a := \left(a_1,a_2,\ldots,a_n\right)$ is not a vertex of the convex hull of $M$, then this vector is a convex combination of the vectors in $M$, excluding $v_a$ itself. Thus, for some integer $k>0$, the vector $kv_a$ equals a sum of several vectors in $M$ distinct from $v_a$. Thus, the $k$-th power of the monomial $m_a$ equals a product of $k$ other monomials of $f$. But this is absurd, since $m_a$ is larger than all these latter monomials. $\endgroup$ Commented Jul 1, 2023 at 20:35
  • 3
    $\begingroup$ Now for the other direction: If the vector $v_a := \left(a_1,a_2,\ldots,a_n\right)$ is a vertex of the convex hull of $M$, then there is an (affine) hyperplane that separates it from all the other vectors in $M$. Using this hyperplane, it is easy to construct a weighted lexicographic order on the monomials for which $m_a$ leads $f$ (just use the oriented distance to this hyperplane as the first comparison, and then break ties using the standard lexicographic order). $\endgroup$ Commented Jul 1, 2023 at 20:37
  • 2
    $\begingroup$ Note that your initial guess is not true, since (e.g.) the monomial $x_1x_2$ does not lead the polynomial $x_1^2 + x_1 x_2 + x_2^2$ no matter what monomial order you choose. $\endgroup$ Commented Jul 1, 2023 at 20:38
  • 2
    $\begingroup$ @darij: Thanks very much. Maybe leave it as an answer? $\endgroup$ Commented Jul 1, 2023 at 20:39

0

You must log in to answer this question.

Browse other questions tagged .