A Proof of Brill’s Theorem via Directional Derivatives

Let $K$ be the field of real numbers $\mathbb{R}$ or the field of complex numbers $\mathbb{C}$ . Given a vector $v = (v_1, \ldots, v_n) \in K^n$ . The directional derivative operator with respect to the vector $v$ , denoted by $D_v$ , is a map from the ring $K[x_1, \ldots, x_n]$ to itself, defined by

$\displaystyle D_v(P) = \sum_{i=1}^n v_i \frac{\partial P}{\partial x_i},$

where $\frac{\partial P}{\partial x_i}$ is the formal partial derivative of the polynomial $P$ with respect to the variable $x_i$ .

Theorem 1. Let $P, Q \in K[x_1, \ldots, x_n]$ , constants $a, b \in K$ , and a linear form $L(x) = c_1x_1 + \ldots + c_nx_n$ . The operator $D_v$ has the following properties.

(1) $D_v(aP + bQ) = aD_v(P) + bD_v(Q)$ .

(2) $D_v(PQ) = D_v(P)Q + P D_v(Q)$ .

(3) For any positive integer $k$ , $D_v(P^k) = k P^{k-1} D_v(P)$ .

(4) $D_v(L) = L(v)$ .

Proof. (1) By the definition of the partial derivative of a polynomial, we have

$\displaystyle \frac{\partial (aP + bQ)}{\partial x_i} = a\frac{\partial P}{\partial x_i} + b\frac{\partial Q}{\partial x_i},$

thus we immediately obtain $\displaystyle D_v(aP + bQ) = aD_v(P) + bD_v(Q).$

(2) Since the formal partial derivative obeys the product rule

$\displaystyle \frac{\partial (PQ)}{\partial x_i} = \frac{\partial P}{\partial x_i}Q + P\frac{\partial Q}{\partial x_i}$ we have

$\displaystyle D_v(PQ) = \sum_{i=1}^n v_i \left( \frac{\partial P}{\partial x_i}Q + P\frac{\partial Q}{\partial x_i} \right)$

$\displaystyle = \left( \sum_{i=1}^n v_i \frac{\partial P}{\partial x_i} \right) Q + P \left( \sum_{i=1}^n v_i \frac{\partial Q}{\partial x_i} \right) = D_v(P)Q + P D_v(Q).$

(3) We prove this by mathematical induction on $k$ .

For $k=1$ , the formula becomes $D_v(P^1) = 1 \cdot P^0 \cdot D_v(P)$ , which is trivially true.

Suppose the property holds for $k-1$ , that is

$\displaystyle D_v(P^{k-1}) = (k-1)P^{k-2}D_v(P).$

Applying (2), we obtain

$\displaystyle D_v(P^k) = D_v(P)P^{k-1} + P D_v(P^{k-1})$

$=D_v(P)P^{k-1} + P \big( (k-1)P^{k-2}D_v(P) \big)=k P^{k-1} D_v(P).$

By the principle of mathematical induction, the property holds for all positive integers $k$ .

(4) For a linear form $L(x) = c_1x_1 + \ldots + c_nx_n$ , we have

$\displaystyle D_v(L) = \sum_{i=1}^n v_i \frac{\partial L}{\partial x_i} = \sum_{i=1}^n v_i c_i = L(v).$

The theorem is completely proved. $\Box$

Theorem 2 (Brill). Suppose $K$ is the field of real numbers $\mathbb{R}$ or the field of complex numbers $\mathbb{C}$ . Let $L_1, L_2, \ldots, L_m$ be $m$ non-zero linear forms in $n$ $(n>1)$ variables that are pairwise non-proportional over $K$ . Then, for any integer $k \ge m-1$ , the set of powers $\{L_1^k, L_2^k, \ldots, L_m^k\}$ is linearly independent over $K$ .

Proof. We prove the theorem by mathematical induction on the number of linear forms $m$ . For $m=1$ , we have 1 non-zero linear form $L_1$ . For any integer $k \ge 0$ , the polynomial $L_1^k$ is not the zero polynomial, so the set consisting of only one element $\{L_1^k\}$ is trivially linearly independent. Thus, the assertion holds for $m=1$ .

Now suppose the assertion holds for $m-1$ linear forms, for some integer $m > 1$ . Consider $m$ non-zero pairwise non-proportional linear forms $L_1, \ldots, L_m$ and an exponent $k \ge m-1$ .

Consider the linear relation

$\displaystyle \sum_{i=1}^m c_i L_i^k = 0$ (1)

where $c_i \in K$ . We need to show that $c_i = 0$ for all $i = 1, \ldots, m$ .

Since $L_m$ is not identically $0$ , the set of points $x \in K^n$ such that $L_m(x) = 0$ is an $(n-1)$ -dimensional vector subspace of the space $K^n$ . At the same time, since $L_i$ is not proportional to $L_m$ for all $i < m$ , the intersection of the two spaces $L_i(x) = 0$ and $L_m(x) = 0$ is an $(n-2)$ -dimensional subspace.

The field $K$ has characteristic $0$ and is therefore an infinite field. A vector space over an infinite field cannot be the union of finitely many proper subspaces (see [1]). Therefore, there exists a vector $v = (v_1, \ldots, v_n) \in K^n$ such that $L_m(v) = 0$ and $\displaystyle L_i(v) \neq 0, \quad \forall i = 1, \ldots, m-1.$

Consider the directional derivative operator with respect to the vector $v$ defined as follows

$\displaystyle D_v = \sum_{j=1}^n v_j \frac{\partial}{\partial x_j}.$

When applying $D_v$ to a linear form $\displaystyle L(x) = a_1x_1 + \ldots + a_nx_n,$

we have $\displaystyle D_v(L) = a_1v_1 + \ldots + a_nv_n = L(v).$

When applying this operator to the $k$ -th power of $L$ , we obtain

$\displaystyle D_v(L^k) = k \cdot L^{k-1} \cdot D_v(L) = k \cdot L(v) \cdot L^{k-1}.$

Applying the operator $D_v$ to both sides of equation (1), we have

$\displaystyle \sum_{i=1}^m c_i k L_i(v) L_i^{k-1} = 0.$

Since $L_m(v) = 0$ , the $m$ -th term vanishes completely. Since $m \ge 2$ , we have $k \ge m-1 \ge 1$ , so we can divide both sides by $k$ (which is valid since $K$ has characteristic $0$ ) to obtain

$\displaystyle \sum_{i=1}^{m-1} \big(c_i L_i(v)\big) L_i^{k-1} = 0.$

Since $k \ge m-1$ , we deduce that $k-1 \ge m-2$ . Applying the inductive hypothesis to $m-1$ linear forms with the exponent $k-1$ , the set of polynomials $\{L_1^{k-1}, \ldots, L_{m-1}^{k-1}\}$ is linearly independent. Therefore, all coefficients in the sum above must be zero. That is,

$\displaystyle c_i L_i(v) = 0, \quad \forall i = 1, \ldots, m-1.$ But by the choice of the vector $v$ , we already have $L_i(v) \neq 0$ , so we must have $\displaystyle c_i = 0, \quad \forall i = 1, \ldots, m-1.$ Substituting $c_1 = \ldots = c_{m-1} = 0$ back into (1), we have $c_m L_m^k = 0$ . Since $L_m$ is not zero, $c_m = 0$ .

Thus $c_1 = c_2 = \ldots = c_m = 0$ , and $\{L_1^k, L_2^k, \ldots, L_m^k\}$ is linearly independent. By the principle of mathematical induction, the theorem is proved. $\Box$

References

[1] https://nttuan.org/2010/02/04/finite-unions-of-proper-subspaces-over-an-infinite-field/

Existence of Non-Trivial Solutions in Homogeneous Linear Systems

Định lý. Cho hai số nguyên dương $N>M$ và một bảng các số thực (phức, hữu tỷ) $(a_{i,j})$ có cỡ $M\times N$ . Khi đó, hệ phương trình

$\displaystyle \sum_{j=1}^Na_{i,j}x_j=0,\quad i=1,2,\ldots,M$

luôn có nghiệm thực (tương ứng phức, hữu tỷ) không tầm thường.

Chứng minh. Ta chứng minh đồng thời cho các hệ có hệ số trong mỗi trường $\mathbb K\in\{\mathbb Q,\mathbb R,\mathbb C\}$ bằng quy nạp theo số phương trình $M$ .

Với $M=1$ , ta có $N\ge2$ . Nếu mọi hệ số $a_{1,j}$ đều bằng $0$ thì có thể lấy $(x_1,\ldots,x_N)=(1,0,\ldots,0)$ . Nếu không, chọn $k$ sao cho $a_{1,k}\ne0$ và chọn $\ell\ne k$ . Đặt

$\displaystyle x_\ell=1,\qquad x_k=-\frac{a_{1,\ell}}{a_{1,k}},\qquad x_j=0\quad(j\notin\{k,\ell\}).$

Đây là một nghiệm không tầm thường.

Giả sử kết luận đúng với mọi hệ gồm $M-1$ phương trình và có số ẩn lớn hơn $M-1$ . Xét hệ gồm $M$ phương trình với $N>M$ . Nếu phương trình thứ $M$ có mọi hệ số bằng $0$ , bỏ phương trình ấy rồi áp dụng giả thiết quy nạp.

Trong trường hợp còn lại, bằng cách đổi thứ tự các ẩn, ta có thể giả sử $a_{M,N}\ne0$ . Từ phương trình thứ $M$ , suy ra

$\displaystyle x_N=-\sum_{j=1}^{N-1}\frac{a_{M,j}}{a_{M,N}}x_j.$

Thay vào $M-1$ phương trình đầu, ta được hệ thuần nhất

$\displaystyle \sum_{j=1}^{N-1} \left(a_{i,j}-\frac{a_{i,N}a_{M,j}}{a_{M,N}}\right)x_j=0, \qquad i=1,\ldots,M-1.$

Hệ này có $N-1>M-1$ ẩn và các hệ số vẫn thuộc $\mathbb K$ . Theo giả thiết quy nạp, nó có nghiệm $(x_1,\ldots,x_{N-1})$ không tầm thường. Xác định $x_N$ bằng công thức trên, ta thu được một nghiệm của hệ ban đầu. Nghiệm này không tầm thường vì ít nhất một trong các số $x_1,\ldots,x_{N-1}$ khác $0$ .

Vậy kết luận đúng với mọi $M<n$ . $\Box$

Đây là một kết quả hữu ích khác về hệ phương trình tuyến tính nhiều ẩn https://nttuan.org/2007/10/21/siegel/

Finite Unions of Proper Subspaces Over an Infinite Field

A vector space over an infinite field cannot be the union of finitely many proper subspaces.

Proof. Assume for the sake of contradiction that $V$ is a vector space over an infinite field $\mathbb{F}$ and $V = \bigcup_{i=1}^n W_i$ , where $n$ is the smallest positive integer satisfying this property and the $W_i$ are proper subspaces of $V$ .
Since $V$ cannot be a proper subspace of itself, $n$ must be greater than or equal to $2$ . Since $n$ is minimal, $W_n$ is not contained in the union of the remaining subspaces. Thus there exists a vector $u \in W_n$ such that $u \notin W_i$ for all $i < n$ . On the other hand, since $W_n$ is a proper subspace of $V$ , there exists a vector $v \notin W_n$ .
Consider the infinite set of vectors of the form $v + c u$ with $c \in \mathbb{F}$ . Since $V$ is the union of $n$ subspaces and the field $\mathbb{F}$ has infinitely many elements, by the pigeonhole principle there exists a subspace $W_k$ containing at least two distinct vectors from the considered set. Suppose $v + c_1 u \in W_k$ and $v + c_2 u \in W_k$ with $c_1, c_2 \in \mathbb{F}$ and $c_1 \neq c_2$ . Then the difference of these two vectors is $(c_1 - c_2) u \in W_k$ . Since $c_1 \neq c_2$ , the scalar $c_1 - c_2$ is non-zero, hence we have $u \in W_k$ . If $k = n$ , then from $v + c_1 u \in W_n$ and $u \in W_n$ we deduce that $v \in W_n$ , which contradicts the choice of $v$ . If $k < n$ , then $u \in W_k$ , which contradicts the choice of $u$ . All cases lead to a contradiction. Therefore, a vector space over an infinite field cannot be the union of finitely many proper subspaces. $\Box$