7
$\begingroup$

I met with a very elementary question on linear algebra which turns out to be quite resistant when I work on it. Any comments or references on either solutions or the possible value of the question will be greatly appreciated.

Initial problem: Given $n$ homogeneous polynomials $ F_1,\dots, F_n $ in $R[x_1,\dots, x_m]$ such that $\deg(F_1)= \dots= \deg(F_n)=d$. For a fixed positive integer $k < n$, set $$P_k=\{F_{i_1}F_{i_2} \cdots F_{i_k} \,\vert\, 1\leq i_1 < i_2 < \cdots < i_k \leq n \}$$ If $ F_1,\dots, F_n $ are linearly independent, that is, $P_1$ is a linearly independent set, then is $P_k$ a linearly independent set for $k > 1$?

The question is trivial for $d=1$. In this case monomials in $F_1,\dots,F_n$ are always linearly independent as we can assume $n \le m$ and $F_i = x_i$ up to a coordinate change. On the other hand, the question has a negative answer when $d > 1$ if we do not impose any additional conditions. For instance, set $$F_1=x_1^2, F_2=x_1x_2,F_3=x_1x_3,F_4=x_2x_3,$$ then $F_1,F_2,F_3,F_4$ are independent. However, $P_2$ is dependent since $F_1F_4=F_2F_3$.

Now the question becomes:

Question 1. Assume $d > 1$. Under what condition will $P_k$ be linearly independent?

A first thought is that we restrict to homogeneous polynomials that are powers of linear polynomials. But this seems to be complicated as well. For instance, the question for $P_2$ and $d = 2$ goes as follows:

Question 2. Let $F_1,\dots, F_n \in R[x_1,\dots, x_m]$ be homogeneous polynomials of degree one such that $F_1^2, F_2^2,\dots, F_n^2$ are linearly independent ($F_1,\dots, F_n$ might be linearly dependent in this case). Under what condition will $F_1^2F_2^2, F_1^2F_3^2,\cdots, F_{n -1}^2F_n^2$ be linearly independent?

$\endgroup$
4
  • $\begingroup$ There is a sufficient condition for $P_{n - 1}$ to be a linearly independent set over $R$, assuming $R$ is a field. Let $A$ be an $R$-algebra which is also a unique factorisation domain. For $a \in A$, denote by $\pi(a)$ the set of prime elements of $A$ dividing $a$. Let $E = \{a_1, \dots, a_n\}$ be a subset of $A$ such that $\pi(a_i)$ is not contained in $\bigcup_{j \neq i} \pi(a_j)$ for every $i$ (e.g., the sets $\pi(a_i)$ are pairwise disjoint). Then the set $P_{n - 1}$ consisting of the products of $n - 1$ distinct elements of $E$ is linearly independent over $R$. $\endgroup$
    – Luc Guyot
    Commented Jul 4, 2023 at 6:08
  • $\begingroup$ Could you please clarify what $R$ is? It seems likely that your ring of coefficients is the field of real numbers $\mathbb{R}$, but I cannot exclude that you meant something different. $\endgroup$
    – Luc Guyot
    Commented Jul 4, 2023 at 6:34
  • $\begingroup$ @luc guyot: Just let R be either real or complex numbers, and thanks for the editing and comments. The sufficient condition looks nice to work on the context of Question 2 where prime factors are just linear polynomials. Are there any references on why $P_{n-1}$ is special? $\endgroup$
    – LichenSDU
    Commented Jul 4, 2023 at 8:36
  • $\begingroup$ The case $P_{n - 1}$ is special in the sense that it relates linear dependence over $K$ to divisibility in the ring of polynomials in an obvious way. Similar criteria can be established for $P_k$ with $k < n - 1$, see my answer below, which is a more general version of my comment. However, this is only a remark with limited bearing and it doesn't refer to homogeneous polynomials specifically. $\endgroup$
    – Luc Guyot
    Commented Jul 4, 2023 at 15:43

1 Answer 1

2
$\begingroup$

This is only a long comment with limited bearing. It consists of two ideas of a similar flavour. Each can be used to address specific instances of the problem.

Claim 1. Let $K$ be a field and let $A$ be $K$-algebra with no zero divisors. Let $E$ be subset of $A$ with $n > 1$ elements and let $k < n$ be such that for every subset $S \subset E$ with $n - k$ elements, there is a valuation $\nu = \nu_S$ defined on the field of fractions of $A$, and non-negative on $A$, such that $$\min_{a \in S} \nu(a) > \sum_{a \in E \setminus S} \nu(a).$$ Then the products of $k$ distinct elements of $E$ are linearly independent over $K$.

Example 1. Let $p$ be a prime element of an integral domain $A$ which satisfies the ascending chain conditions on principal ideals. The largest power $\nu_p(a)$ of $p$ which divides $a \in A$ defines a non-negative valuation on $A$ which extends to a discrete valuation on the field of fractions of $A$.

Example 2. Let $K$ be an algebraically closed field and let $\mathfrak{m}$ be a maximal ideal of $A = K[x_1, \dots, x_m]$. Then $\nu_{\mathfrak{m}}(a) := \sup \{n \ge 0 \, \vert \, a \in \mathfrak{m}^n \}$ defines a non-negative valuation of $A$ which extends to a discrete valuation of $K(x_1, \dots, x_m)$.

Proof of Claim 1. A non-trivial $K$-linear dependence relation between $k$-fold products yields a relation of the form $$a_1 \cdots a_k \in Aa_{k + 1} + \cdots + A a_n$$ for some suitable labelling $E = \{a_1, \dots, a_n\}$. Let $S = \{a_{k + 1}, \dots, a_n\}$ and let $\nu = \nu_S$ be the valuation afforded by hypothesis. As $\nu(\sum_{a \in S}Aa) \ge \min_{a \in S} \nu(a)$, we obtain a contradiction.

Claim 2. Let $K$ be field, $A = K[x_1, \dots, x_m]$ and let $\prec$ be a monomial order on $A$. For a polynomial $a \in A$, denote by $\operatorname{LT}(a)$ the leading term with respect to $\prec$ and by $\deg(M) \in \mathbb{N}^m$ the multi-degree of a monomial $M$ of $A$. Let $a_1, \dots, a_n$ be elements of $A$ such that the sums $\deg(\operatorname{LT}(a_i)) + \deg(\operatorname{LT}(a_j))$, with $i < j$, are distinct. Then the products $a_i a_j$, with $i < j$, are linearly independent over $K$.

Proof of Claim 2. By hypothesis, the leading terms $\operatorname{LT}(a_i a_i)$, with $i < j$, are linearly independent over $K$. This ensures that the products $a_i a_j$, with $i < j$, are linearly independent over $K$.

$\endgroup$

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.