线性代数 | 秩—零化度定理

注:本文为 “线性代数 · 秩—零化度定理” 相关合辑。
英文引文,机翻未校。
中文引文,略作重排。
如有内容异常,请看原文。


Rank and Nullity

秩与零化度

Rank and Nullity are two essential concepts related to matrices in Linear Algebra. The nullity of a matrix is determined by the difference between the order and rank of the matrix. The rank of a matrix is the number of linearly independent row or column vectors of a matrix. If n n n is the order of the square matrix A A A, then the nullity of A A A is given by n – r n – r nr. Thus, the rank of a matrix is the number of linearly independent or non-zero vectors of a matrix, whereas nullity is the number of zero vectors of a matrix.
在线性代数中,秩(Rank)与零化度(Nullity)是与矩阵相关的两个核心概念。矩阵的零化度由其阶数与秩的差值确定,矩阵的秩则是矩阵中线性无关的行向量或列向量的数量。若 n n n 为方阵 A A A 的阶数,则 A A A 的零化度可表示为 n – r n – r nr(其中 r r r 为矩阵的秩)。因此,矩阵的秩是矩阵中线性无关向量(或非零向量)的数量,而零化度是矩阵中零向量的数量。

The rank of matrix A A A is denoted as ρ ( A ) \rho(A) ρ(A), and the nullity is denoted as N ( A ) N(A) N(A). Evidently, if the rank of the matrix is equal to the order of the matrix, then the nullity of the matrix is zero.
矩阵 A A A 的秩记为 ρ ( A ) \rho(A) ρ(A),零化度记为 N ( A ) N(A) N(A)。显然,若矩阵的秩等于其阶数,则该矩阵的零化度为 0。

Rank and Nullity Theorem for Matrix

矩阵的秩-零化度定理

The rank and nullity theorem for matrices is one of the important theorems in linear algebra and a requirement to derive many more results. Before discussing the theorem, we must know the concept of null spaces.
矩阵的秩-零化度定理是线性代数中的重要定理之一,也是推导更多结论的基础。在讨论该定理之前,我们需要先了解零空间(Null Space)的概念。

Nullspace

零空间

Let A A A be a real matrix of order m × n m \times n m×n, the set of the solutions associated with the system of homogeneous equation A X = 0 \boldsymbol{AX = 0} AX=0 is said to be the null space of A A A.
A A A m × n m \times n m×n 阶实矩阵,齐次方程组 A X = 0 \boldsymbol{AX = 0} AX=0 的所有解构成的集合,称为矩阵 A A A 的零空间。

Nullspace of A A A = { x ∈ R n ∣ A x = 0 } \{ x \in \boldsymbol{R}^n \mid Ax = 0 \} {xRnAx=0}. Then the nullity of A A A will be the dimension of the Nullspace of A A A.
A A A 的零空间可表示为 { x ∈ R n ∣ A x = 0 } \{ x \in \boldsymbol{R}^n \mid Ax = 0 \} {xRnAx=0},而 A A A 的零化度就是其零空间的维数。

If the rank of A A A is r r r, there are r r r leading variables in row-reduced echelon form of A A A and n – r n – r nr free variables, which are solutions of the homogeneous system of equation A X = 0 AX = 0 AX=0. Thus, n – r n – r nr is the dimension of the null space of A A A. This fact motivates the rank and nullity theorem for matrices.
A A A 的秩为 r r r,则其行简化阶梯形矩阵中含有 r r r 个主变量(leading variable)和 n – r n – r nr 个自由变量(free variable),这些自由变量对应齐次方程组 A X = 0 AX = 0 AX=0 的解。因此, A A A 的零空间维数为 n – r n – r nr,这一结论为矩阵秩-零化度定理的提出提供了依据。

Row-reduced Echelon Form of A A A
矩阵 A A A 的行简化阶梯形

Rank and Nullity

Rank and Nullity Theorem

秩-零化度定理

If A A A is a matrix of order m × n m \times n m×n, then Rank of A A A + Nullity of A A A = Number of columns in A A A = n n n
A A A m × n m \times n m×n 阶矩阵,则 A A A 的秩 + A A A 的零化度 = A A A 的列数 = n n n

Proof: We already have a result, “Let A A A be a matrix of order m × n m \times n m×n, then the rank of A A A is equal to the number of leading columns of row-reduced echelon form of A A A.”
证明:已知结论“设 A A A m × n m \times n m×n 阶矩阵,则 A A A 的秩等于其行简化阶梯形矩阵中主列(leading column)的数量”。

Let r r r be the rank of A A A, and then we have two cases as follows:
r r r A A A 的秩,分以下两种情况讨论:

Case I: If A A A is a non-singular matrix, then r = n r = n r=n as the row-reduced echelon form will have no zero rows. Then, x = 0 x = 0 x=0 will be the trivial solution of A X = 0 AX = 0 AX=0; that is, the nullity of A A A will be zero.
情况一: A A A 为非奇异矩阵(可逆矩阵),则其行简化阶梯形矩阵不含零行,因此 r = n r = n r=n。此时,齐次方程组 A X = 0 AX = 0 AX=0 仅有平凡解 x = 0 x = 0 x=0,即 A A A 的零化度为 0。

Hence, rank of A A A + nullity of A A A = n + 0 = n n + 0 = n n+0=n = Number of columns in A A A
因此, A A A 的秩 + A A A 的零化度 = n + 0 = n n + 0 = n n+0=n = A A A 的列数

Case II: If A A A is singular, then rank of A A A will be less than the order of A A A, that is, r < n r < n r<n. Therefore, there are r r r non-zero rows or r r r leading columns of the row-reduced echelon form. Consequently, there will be n – r n – r nr zero rows, which contributes to the solution of A X = 0 AX = 0 AX=0, which means the nullity of A A A is n – r n – r nr.
情况二: A A A 为奇异矩阵(不可逆矩阵),则 A A A 的秩小于其阶数,即 r < n r < n r<n。此时,其行简化阶梯形矩阵含有 r r r 个非零行(或 r r r 个主列),因此会有 n – r n – r nr 个零行,这些零行对应齐次方程组 A X = 0 AX = 0 AX=0 的解,即 A A A 的零化度为 n – r n – r nr

Hence, rank of A A A + nullity of A A A = r + n – r = n r + n – r = n r+nr=n = Number of columns in A A A.
因此, A A A 的秩 + A A A 的零化度 = r + ( n – r ) = n r + (n – r) = n r+(nr)=n = A A A 的列数。

In both cases, we get the same result which proves our claim.
两种情况均得到相同结论,定理得证。

Important Facts on Rank and Nullity

关于秩与零化度的重要结论

  • The rank of an invertible matrix is equal to the order of the matrix, and its nullity is equal to zero.
    可逆矩阵的秩等于其阶数,零化度为 0。

  • Rank is the number of leading columns or non-zero row vectors of the row-reduced echelon form of the given matrix, and the number of zero rows is the nullity.
    矩阵的秩等于其行简化阶梯形矩阵中主列或非零行向量的数量,零化度等于其行简化阶梯形矩阵中零行的数量。

  • The nullity of a matrix is the dimension of the null space of A A A, also called the kernel of A A A.
    矩阵的零化度是其零空间的维数,零空间也称为核(kernel)。

  • If A A A is an invertible matrix, then null space ( A A A) = { 0 } \{0\} {0}.
    A A A 为可逆矩阵,则其零空间 null space ( A A A) = { 0 } \{0\} {0}(仅含零向量)。

  • The rank of a matrix is the number of non-zero eigenvalues of the matrix, and the number of zero eigenvalues determines the nullity of the matrix.
    矩阵的秩等于其非零特征值的数量,零特征值的数量决定了矩阵的零化度。

Solved Examples on Rank and Nullity

秩与零化度的求解示例

Example 1

示例 1

Verify the rank and nullity theorem for the matrix
验证下述矩阵是否满足秩-零化度定理:

A = [ 1 1 2 3 3 4 − 1 2 − 1 − 2 5 4 ] A=\begin{bmatrix}1 & 1 & 2 & 3 \\3 & 4 & -1 & 2 \\-1 & -2 & 5 & 4 \\\end{bmatrix} A= 131142215324

Solution

解答

Given the matrix:
已知矩阵:

A = [ 1 1 2 3 3 4 − 1 2 − 1 − 2 5 4 ] A=\begin{bmatrix}1 & 1 & 2 & 3 \\3 & 4 & -1 & 2 \\-1 & -2 & 5 & 4 \\\end{bmatrix} A= 131142215324

We reduce A A A to its row-reduced echelon form through elementary row and column operations:
通过初等行变换与列变换,将 A A A 化为行简化阶梯形:

  1. Apply R 2 → R 2 + ( − 3 ) R 1 R_2 \to R_2 + (-3)R_1 R2R2+(3)R1 and R 3 → R 3 + R 1 R_3 \to R_3 + R_1 R3R3+R1
    执行行变换 R 2 → R 2 + ( − 3 ) R 1 R_2 \to R_2 + (-3)R_1 R2R2+(3)R1 R 3 → R 3 + R 1 R_3 \to R_3 + R_1 R3R3+R1
    A ∼ [ 1 1 2 3 0 1 − 7 − 7 0 − 1 7 7 ] A\sim\begin{bmatrix}1 & 1 & 2 & 3 \\0 & 1 & -7 & -7 \\0 & -1 & 7 & 7 \\\end{bmatrix} A 100111277377

  2. Apply R 3 → R 3 + R 2 R_3 \to R_3 + R_2 R3R3+R2
    执行行变换 R 3 → R 3 + R 2 R_3 \to R_3 + R_2 R3R3+R2
    A ∼ [ 1 1 2 3 0 1 − 7 − 7 0 0 0 0 ] A\sim\begin{bmatrix}1 & 1 & 2 & 3 \\0 & 1 & -7 & -7 \\0 & 0 & 0 & 0 \\\end{bmatrix} A 100110270370

  3. Apply C 2 → C 2 + C 1 C_2 \to C_2 + C_1 C2C2+C1, C 3 → C 3 + ( − 2 ) C 1 C_3 \to C_3 + (-2)C_1 C3C3+(2)C1 and C 4 → C 4 + ( − 3 ) C 1 C_4 \to C_4 + (-3)C_1 C4C4+(3)C1
    执行列变换 C 2 → C 2 + C 1 C_2 \to C_2 + C_1 C2C2+C1 C 3 → C 3 + ( − 2 ) C 1 C_3 \to C_3 + (-2)C_1 C3C3+(2)C1 C 4 → C 4 + ( − 3 ) C 1 C_4 \to C_4 + (-3)C_1 C4C4+(3)C1
    A ∼ [ 1 0 0 0 0 1 − 7 − 7 0 0 0 0 ] A\sim\begin{bmatrix}1 & 0 & 0 & 0 \\0 & 1 & -7 & -7 \\0 & 0 & 0 & 0 \\\end{bmatrix} A 100010070070

  4. Apply C 3 → C 3 + 7 C 2 C_3 \to C_3 + 7C_2 C3C3+7C2 and C 4 → C 4 + 7 C 2 C_4 \to C_4 + 7C_2 C4C4+7C2
    执行列变换 C 3 → C 3 + 7 C 2 C_3 \to C_3 + 7C_2 C3C3+7C2 C 4 → C 4 + 7 C 2 C_4 \to C_4 + 7C_2 C4C4+7C2
    A ∼ [ 1 0 0 0 0 1 0 0 0 0 0 0 ] A\sim\begin{bmatrix}1 & 0 & 0 & 0 \\0 & 1 & 0 & 0 \\0 & 0 & 0 & 0 \\\end{bmatrix} A 100010000000

Clearly, rank ( A ) = 2 \text{rank}(A) = 2 rank(A)=2 and nullity ( A ) = 2 \text{nullity}(A) = 2 nullity(A)=2.
显然, rank ( A ) = 2 \text{rank}(A) = 2 rank(A)=2 nullity ( A ) = 2 \text{nullity}(A) = 2 nullity(A)=2

Therefore, rank ( A ) + nullity ( A ) = 2 + 2 = 4 \text{rank}(A) + \text{nullity}(A) = 2 + 2 = 4 rank(A)+nullity(A)=2+2=4, which equals the number of columns of A A A.
因此, rank ( A ) + nullity ( A ) = 2 + 2 = 4 \text{rank}(A) + \text{nullity}(A) = 2 + 2 = 4 rank(A)+nullity(A)=2+2=4,等于矩阵 A A A 的列数。

Example 2

示例 2

Find the nullity of the matrix
求下述矩阵的零化度:

A = [ 1 3 4 3 3 9 12 9 1 3 4 1 ] A=\begin{bmatrix}1 & 3 & 4 & 3 \\3 & 9 & 12 & 9 \\1 & 3 & 4 & 1 \\\end{bmatrix} A= 1313934124391

Solution

解答

Given the matrix:
已知矩阵:

A = [ 1 3 4 3 3 9 12 9 1 3 4 1 ] A=\begin{bmatrix}1 & 3 & 4 & 3 \\3 & 9 & 12 & 9 \\1 & 3 & 4 & 1 \\\end{bmatrix} A= 1313934124391

We reduce A A A to its row-reduced echelon form through elementary operations:
通过初等变换,将 A A A 化为行简化阶梯形:

  1. Apply elementary row operations R 21 ( − 3 ) R_{21}(-3) R21(3) (i.e., R 2 = R 2 − 3 R 1 R_2 = R_2 - 3R_1 R2=R23R1) and R 31 ( − 1 ) R_{31}(-1) R31(1) (i.e., R 3 = R 3 − R 1 R_3 = R_3 - R_1 R3=R3R1)
    执行初等行变换 R 21 ( − 3 ) R_{21}(-3) R21(3)(即 R 2 = R 2 − 3 R 1 R_2 = R_2 - 3R_1 R2=R23R1)和 R 31 ( − 1 ) R_{31}(-1) R31(1)(即 R 3 = R 3 − R 1 R_3 = R_3 - R_1 R3=R3R1):
    A ∼ [ 1 3 4 3 0 0 0 0 0 0 0 − 2 ] A\sim \begin{bmatrix}1 & 3 & 4 & 3 \\ 0 & 0 & 0 & 0 \\0 & 0 & 0 & -2 \\\end{bmatrix} A 100300400302

  2. Apply elementary row operation R 3 ( − 1 2 ) R_3(-\frac{1}{2}) R3(21) (i.e., R 3 = − 1 2 R 3 R_3 = -\frac{1}{2}R_3 R3=21R3)
    执行初等行变换 R 3 ( − 1 2 ) R_3(-\frac{1}{2}) R3(21)(即 R 3 = − 1 2 R 3 R_3 = -\frac{1}{2}R_3 R3=21R3):
    A ∼ [ 1 3 4 3 0 0 0 0 0 0 0 1 ] A\sim \begin{bmatrix}1 & 3 & 4 & 3 \\ 0 & 0 & 0 & 0 \\0 & 0 & 0 & 1 \\\end{bmatrix} A 100300400301

  3. Apply elementary column operation C 24 C_{24} C24 (i.e., swap Column 2 and Column 4)
    执行初等列变换 C 24 C_{24} C24(即交换第 2 列与第 4 列):
    A ∼ [ 1 3 4 3 0 1 0 0 0 0 0 0 ] A\sim \begin{bmatrix}1 & 3 & 4 & 3 \\ 0 & 1 & 0 & 0 \\0 & 0 & 0 & 0 \\\end{bmatrix} A 100310400300

  4. Apply elementary column operations C 21 ( − 3 ) C_{21}(-3) C21(3) (i.e., C 2 = C 2 − 3 C 1 C_2 = C_2 - 3C_1 C2=C23C1), C 31 ( − 4 ) C_{31}(-4) C31(4) (i.e., C 3 = C 3 − 4 C 1 C_3 = C_3 - 4C_1 C3=C34C1) and C 41 ( − 3 ) C_{41}(-3) C41(3) (i.e., C 4 = C 4 − 3 C 1 C_4 = C_4 - 3C_1 C4=C43C1)
    执行初等列变换 C 21 ( − 3 ) C_{21}(-3) C21(3)(即 C 2 = C 2 − 3 C 1 C_2 = C_2 - 3C_1 C2=C23C1)、 C 31 ( − 4 ) C_{31}(-4) C31(4)(即 C 3 = C 3 − 4 C 1 C_3 = C_3 - 4C_1 C3=C34C1)和 C 41 ( − 3 ) C_{41}(-3) C41(3)(即 C 4 = C 4 − 3 C 1 C_4 = C_4 - 3C_1 C4=C43C1):
    A ∼ [ 1 0 0 0 0 1 0 0 0 0 0 0 ] A\sim \begin{bmatrix}1 & 0 & 0 & 0 \\ 0 & 1 & 0 & 0 \\0 & 0 & 0 & 0 \\\end{bmatrix} A 100010000000

Thus, Rank ( A ) = 2 \text{Rank}(A) = 2 Rank(A)=2, and Nullity ( A ) = Number of columns − Rank = 4 − 2 = 2 \text{Nullity}(A) = \text{Number of columns} - \text{Rank} = 4 - 2 = 2 Nullity(A)=Number of columnsRank=42=2.
因此, Rank ( A ) = 2 \text{Rank}(A) = 2 Rank(A)=2 Nullity ( A ) = 列数 − 秩 = 4 − 2 = 2 \text{Nullity}(A) = \text{列数} - \text{秩} = 4 - 2 = 2 Nullity(A)=列数=42=2

Frequently Asked Questions on Rank and Nullity

关于秩与零化度的常见问题

Q1: What is the rank of the matrix?

问题 1:矩阵的秩是什么?

The number of linearly independent row or column vectors of a matrix is the rank of the matrix.
矩阵中线性无关的行向量或列向量的数量,称为矩阵的秩。

Q2: What is the nullity of the matrix?

问题 2:矩阵的零化度是什么?

The dimension of the nullspace or kernel of the given matrix is the nullity of the matrix.
给定矩阵的零空间(或核)的维数,称为矩阵的零化度。

Q3: What is the rank and nullity theorem for matrices?

问题 3:矩阵的秩-零化度定理是什么?

For any matrix A A A of order m × n m \times n m×n, rank ( A ) + nullity ( A ) = n \text{rank}(A) + \text{nullity}(A) = n rank(A)+nullity(A)=n, where n n n is the number of columns in A A A.
对于任意 m × n m \times n m×n 阶矩阵 A A A,均有 rank ( A ) + nullity ( A ) = n \text{rank}(A) + \text{nullity}(A) = n rank(A)+nullity(A)=n(其中 n n n A A A 的列数)。

Q4: What is the nullity of an invertible matrix?

问题 4:可逆矩阵的零化度是多少?
The nullity of an invertible matrix is zero.
可逆矩阵的零化度为 0。


The Rank-Nullity Theorem

秩-零化度定理

节选自 普渡大学数学系 2010 年春季学期的微分方程课程(MA26200)
Differential Equations and Linear Algebra by Stephen Goode and Scott A. Annin, Third Edition.Chapter 4.9

In Section 4.3, we defined the null space of a real m × n m \times n m×n matrix A A A to be the set of all real solutions to the associated homogeneous linear system A x = 0 Ax = 0 Ax=0. Thus,
在第 4.3 节中,我们定义了实数 m × n m \times n m×n 矩阵 A A A 的零空间为与齐次线性方程组 A x = 0 Ax = 0 Ax=0 相关的所有实数解的集合。因此,
nullspace ( A ) = { x ∈ R n : A x = 0 } . \text{nullspace}(A) = \{ x \in \mathbb{R}^n : Ax = 0 \}. nullspace(A)={xRn:Ax=0}.

The dimension of nullspace( A A A) is referred to as the nullity of A A A and is denoted nullity ( A ) \text{nullity}(A) nullity(A).
零空间 ( A A A) 的维度被称为 A A A 的零化度,并记作 nullity ( A ) \text{nullity}(A) nullity(A)

In order to find nullity ( A ) \text{nullity}(A) nullity(A), we need to determine a basis for nullspace ( A ) \text{nullspace}(A) nullspace(A). Recall that if rank ( A ) = r \text{rank}(A) = r rank(A)=r, then any row-echelon form of A A A contains r r r leading ones, which correspond to the bound variables in the linear system. Thus, there are n − r n - r nr columns without leading ones, which correspond to free variables in the solution of the system A x = 0 Ax = 0 Ax=0. Hence, there are n − r n - r nr free variables in the solution of the system A x = 0 Ax = 0 Ax=0. We might therefore suspect that nullity ( A ) = n − r \text{nullity}(A) = n - r nullity(A)=nr. Our next theorem, often referred to as the Rank-Nullity Theorem, establishes that this is indeed the case.
为了找到 nullity ( A ) \text{nullity}(A) nullity(A),我们需要确定 nullspace ( A ) \text{nullspace}(A) nullspace(A) 的一个基。回忆一下,如果 rank ( A ) = r \text{rank}(A) = r rank(A)=r,那么 A A A 的任何行阶梯形式都包含 r r r 个主元,这些主元对应于线性系统中的绑定变量。因此,有 n − r n - r nr 列没有主元,这些对应于系统 A x = 0 Ax = 0 Ax=0 解中的自由变量。因此,在系统 A x = 0 Ax = 0 Ax=0 的解中有 n − r n - r nr 个自由变量。我们因此可能怀疑 nullity ( A ) = n − r \text{nullity}(A) = n - r nullity(A)=nr。我们的下一个定理,通常称为秩-零化度定理,确立了这确实是这种情况。

Theorem 4.9.1 (Rank-Nullity Theorem)
定理 4.9.1(秩-零化度定理)

For any m × n m \times n m×n matrix A A A,
对于任何 m × n m \times n m×n 矩阵 A A A

rank ( A ) + nullity ( A ) = n . \text{rank}(A) + \text{nullity}(A) = n. rank(A)+nullity(A)=n.

(4.9.1)

Proof If rank ( A ) = n \text{rank}(A) = n rank(A)=n, then by the Invertible Matrix Theorem, the only solution to A x = 0 Ax = 0 Ax=0 is the trivial solution x = 0 x = 0 x=0. Hence, in this case, nullspace ( A ) = { 0 } \text{nullspace}(A) = \{0\} nullspace(A)={0}, so nullity ( A ) = 0 \text{nullity}(A) = 0 nullity(A)=0 and Equation (4.9.1) holds.
证明 如果 rank ( A ) = n \text{rank}(A) = n rank(A)=n,那么根据可逆矩阵定理, A x = 0 Ax = 0 Ax=0 的唯一解是平凡解 x = 0 x = 0 x=0。因此,在这种情况下, nullspace ( A ) = { 0 } \text{nullspace}(A) = \{0\} nullspace(A)={0},所以 nullity ( A ) = 0 \text{nullity}(A) = 0 nullity(A)=0,方程 (4.9.1) 成立。

Now suppose rank ( A ) = r < n \text{rank}(A) = r < n rank(A)=r<n. In this case, there are n − r > 0 n - r > 0 nr>0 free variables in the solution to A x = 0 Ax = 0 Ax=0. Let t 1 , t 2 , … , t n − r t_1, t_2, \ldots, t_{n-r} t1,t2,,tnr denote these free variables (chosen as those variables not attached to a leading one in any row-echelon form of A A A), and let x 1 , x 2 , … , x n − r x_1, x_2, \ldots, x_{n-r} x1,x2,,xnr denote the solutions obtained by sequentially setting each free variable to 1 and the remaining free variables to zero. Note that { x 1 , x 2 , … , x n − r } \{x_1, x_2, \ldots, x_{n-r}\} {x1,x2,,xnr} is linearly independent. Moreover, every solution to A x = 0 Ax = 0 Ax=0 is a linear combination of x 1 , x 2 , … , x n − r x_1, x_2, \ldots, x_{n-r} x1,x2,,xnr:
现在假设 rank ( A ) = r < n \text{rank}(A) = r < n rank(A)=r<n。在这种情况下, A x = 0 Ax = 0 Ax=0 的解中有 n − r > 0 n - r > 0 nr>0 个自由变量。令 t 1 , t 2 , … , t n − r t_1, t_2, \ldots, t_{n-r} t1,t2,,tnr 表示这些自由变量(选择为在 A A A 的任何行阶梯形式中不与主元相连的变量),并令 x 1 , x 2 , … , x n − r x_1, x_2, \ldots, x_{n-r} x1,x2,,xnr 表示通过依次将每个自由变量设为 1 并将其余自由变量设为 0 而得到的解。注意 { x 1 , x 2 , … , x n − r } \{x_1, x_2, \ldots, x_{n-r}\} {x1,x2,,xnr} 是线性无关的。此外, A x = 0 Ax = 0 Ax=0 的每个解都是 x 1 , x 2 , … , x n − r x_1, x_2, \ldots, x_{n-r} x1,x2,,xnr 的线性组合:

x = t 1 x 1 + t 2 x 2 + ⋯ + t n − r x n − r , x = t_1x_1 + t_2x_2 + \cdots + t_{n-r}x_{n-r}, x=t1x1+t2x2++tnrxnr,

which shows that { x 1 , x 2 , … , x n − r } \{x_1, x_2, \ldots, x_{n-r}\} {x1,x2,,xnr} spans nullspace ( A ) \text{nullspace}(A) nullspace(A). Thus, { x 1 , x 2 , … , x n − r } \{x_1, x_2, \ldots, x_{n-r}\} {x1,x2,,xnr} is a basis for nullspace ( A ) \text{nullspace}(A) nullspace(A), and nullity ( A ) = n − r \text{nullity}(A) = n - r nullity(A)=nr.
这表明 { x 1 , x 2 , … , x n − r } \{x_1, x_2, \ldots, x_{n-r}\} {x1,x2,,xnr} 跨越了 nullspace ( A ) \text{nullspace}(A) nullspace(A)。因此, { x 1 , x 2 , … , x n − r } \{x_1, x_2, \ldots, x_{n-r}\} {x1,x2,,xnr} nullspace ( A ) \text{nullspace}(A) nullspace(A) 的一个基,并且 nullity ( A ) = n − r \text{nullity}(A) = n - r nullity(A)=nr

Example 4.9.2
例 4.9.2

If
如果

A = [ 1 1 2 3 3 4 − 1 2 − 1 − 2 5 4 ] , A = \begin{bmatrix} 1 & 1 & 2 & 3 \\ 3 & 4 & -1 & 2 \\ -1 & -2 & 5 & 4 \end{bmatrix}, A= 131142215324 ,

find a basis for nullspace ( A ) \text{nullspace}(A) nullspace(A) and verify Theorem 4.9.1.
找到 nullspace ( A ) \text{nullspace}(A) nullspace(A) 的一个基,并验证定理 4.9.1。

Solution We must find all solutions to A x = 0 Ax = 0 Ax=0. Reducing the augmented matrix of this system yields
我们必须找到 A x = 0 Ax = 0 Ax=0 的所有解。将该系统的增广矩阵化简得到

A # ∼ [ 1 1 2 3 0 0 1 − 7 − 7 0 0 − 1 7 7 0 ] ∼ [ 1 1 2 3 0 0 1 − 7 − 7 0 0 0 0 0 0 ] . A^{\#} \sim \begin{bmatrix} 1 & 1 & 2 & 3 & 0 \\ 0 & 1 & -7 & -7 & 0 \\ 0 & -1 & 7 & 7 & 0 \end{bmatrix} \sim \begin{bmatrix} 1 & 1 & 2 & 3 & 0 \\ 0 & 1 & -7 & -7 & 0 \\ 0 & 0 & 0 & 0 & 0 \end{bmatrix}. A# 100111277377000 100110270370000 .

  1. A 12 ( − 3 ) A_{12}(-3) A12(3), A 13 ( 1 ) A_{13}(1) A13(1) 2. A 23 ( 1 ) A_{23}(1) A23(1)

Consequently, there are two free variables, x 3 = t 1 x_3 = t_1 x3=t1 and x 4 = t 2 x_4 = t_2 x4=t2, so that
因此,有两个自由变量, x 3 = t 1 x_3 = t_1 x3=t1 x 4 = t 2 x_4 = t_2 x4=t2,使得

x 2 = 7 t 1 + 7 t 2 , x 1 = − 9 t 1 − 10 t 2 . x_2 = 7t_1 + 7t_2, \quad x_1 = -9t_1 - 10t_2. x2=7t1+7t2,x1=9t110t2.

Hence,
因此,

n u l l s p a c e ( A ) = { ( − 9 t 1 − 10 t 2 , 7 t 1 + 7 t 2 , t 1 , t 2 ) : t 1 , t 2 ∈ } = { t 1 ( − 9 , 7 , 1 , 0 ) + t 2 ( − 10 , 7 , 0 , 1 ) : t 1 , t 2 ∈ } = s p a n { ( − 9 , 7 , 1 , 0 ) , ( − 10 , 7 , 0 , 1 ) } . \begin{array}{lll} {\rm{nullspace}}(A) & = \{ ( - 9{t_1} - 10{t_2},7{t_1} + 7{t_2},{t_1},{t_2}):{t_1},{t_2} \in \} \\ & = \{ {t_1}( - 9,7,1,0) + {t_2}( - 10,7,0,1):{t_1},{t_2} \in \} \\ & = {\rm{span}}\{ ( - 9,7,1,0),( - 10,7,0,1)\} . \end{array} nullspace(A)={(9t110t2,7t1+7t2,t1,t2):t1,t2}={t1(9,7,1,0)+t2(10,7,0,1):t1,t2}=span{(9,7,1,0),(10,7,0,1)}.

Since the two vectors in this spanning set are not proportional, they are linearly independent. Consequently, a basis for nullspace ( A ) \text{nullspace}(A) nullspace(A) is { ( − 9 , 7 , 1 , 0 ) , ( − 10 , 7 , 0 , 1 ) } \{(-9, 7, 1, 0), (-10, 7, 0, 1)\} {(9,7,1,0),(10,7,0,1)}, so that nullity ( A ) = 2 \text{nullity}(A) = 2 nullity(A)=2. In this problem, A A A is a 3 × 4 3 \times 4 3×4 matrix, and so, in the Rank-Nullity Theorem, n = 4 n = 4 n=4. Further, from the foregoing row-echelon form of the augmented matrix of the system A x = 0 Ax = 0 Ax=0, we see that rank ( A ) = 2 \text{rank}(A) = 2 rank(A)=2. Hence,
由于该生成集中的两个向量不成比例,它们是线性无关的。因此, nullspace ( A ) \text{nullspace}(A) nullspace(A) 的一个基是 { ( − 9 , 7 , 1 , 0 ) , ( − 10 , 7 , 0 , 1 ) } \{(-9, 7, 1, 0), (-10, 7, 0, 1)\} {(9,7,1,0),(10,7,0,1)},所以 nullity ( A ) = 2 \text{nullity}(A) = 2 nullity(A)=2。在这个问题中, A A A 是一个 3 × 4 3 \times 4 3×4 矩阵,因此,在秩-零化度定理中, n = 4 n = 4 n=4。此外,从系统 A x = 0 Ax = 0 Ax=0 的增广矩阵的行阶梯形式中,我们看到 rank ( A ) = 2 \text{rank}(A) = 2 rank(A)=2。因此,
rank ( A ) + nullity ( A ) = 2 + 2 = 4 = n , \text{rank}(A) + \text{nullity}(A) = 2 + 2 = 4 = n, rank(A)+nullity(A)=2+2=4=n,

and the Rank-Nullity Theorem is verified.
并且秩-零化度定理得到了验证。


秩-零化度定理(Rank-Nullity Theorem)

MatNoble 编辑于 2020-02-09 08:16

引言

线性代数(高等代数)微积分(数学分析) 存在差异,线性代数是一门持续发展的学科。其在实际应用中产生的新问题会反哺教学,而教学的进步又能进一步推动实际应用的发展。

秩-零化度定理(Rank-Nullity Theorem)

如图 1 所示,线性变换 T T T 从有限维向量空间 V \mathcal{V} V(定义域)映射到有限维向量空间 W \mathcal{W} W(值域),记为 T : V → W T:\mathcal{V}\to\mathcal{W} T:VW

img
图 1 线性变换 T : V → W T:\mathcal{V}\to\mathcal{W} T:VW 的示意图

在该映射关系中,存在两个重要的子空间,分别定义如下:

1. 核空间(Kernel Space)

向量空间 V \mathcal{V} V 中所有经线性变换 T T T 映射后得到零元素的元素所构成的集合,称为 T T T 的核(子)空间,记为 ker ⁡ ( T ) \ker (T) ker(T)
核空间的维数(dimension)称为零化度(nullity),记为 dim ⁡ ker ⁡ ( T ) \dim \ker (T) dimker(T),用于度量核空间的规模。

2. 值域(Range)

向量空间 V \mathcal{V} V 中所有元素经线性变换 T T T 映射后所构成的集合,称为 T T T 的值域,记为 r a n ( T ) {\rm ran} (T) ran(T) R ( T ) R(T) R(T)
值域的维数(dimension)称为(rank),记为 r a n k   T {\rm rank}\, T rankT dim ⁡ r a n ( T ) \dim {\rm ran} (T) dimran(T)

秩-零化度定理的表述

线性变换 T : V → W T:\mathcal{V}\to\mathcal{W} T:VW 的定义域 V \mathcal{V} V 的维数,等于其核空间 ker ⁡ ( T ) \ker (T) ker(T) 的维数与值域 r a n ( T ) {\rm ran} (T) ran(T) 的维数之和,即:
dim ⁡ V = dim ⁡ ker ⁡ ( T ) + r a n k   T \dim \mathcal{V} = \dim \ker (T) + {\rm rank}\, T dimV=dimker(T)+rankT

秩-零化度定理的证明

秩-零化度定理可从矩阵角度线性变换角度分别证明,两种证明思路各有侧重,共同支撑定理的严谨性。

1. 矩阵角度的证明

线性变换可通过矩阵具象化表示,因此可借助矩阵的零空间与列空间证明定理。

步骤 1:设定基本条件

设线性变换 T : V → W T: \mathcal{V} \to \mathcal{W} T:VW m × n m \times n m×n 阶矩阵 A A A 表示,其中:

  • n = dim ⁡ V n = \dim \mathcal{V} n=dimV(定义域 V \mathcal{V} V 的维数);
  • m = dim ⁡ W m = \dim \mathcal{W} m=dimW(值域 W \mathcal{W} W 的维数);
  • 矩阵 A A A 的零空间 N ( A ) N(A) N(A) 对应线性变换 T T T 的核空间 ker ⁡ ( T ) \ker (T) ker(T)
  • 矩阵 A A A 的列空间 C ( A ) C(A) C(A) 对应线性变换 T T T 的值域 r a n ( T ) {\rm ran} (T) ran(T)

需证明的结论转化为:
n = dim ⁡ N ( A ) + r a n k A n = \dim N(A) + {\rm rank} A n=dimN(A)+rankA

步骤 2:矩阵的初等行变换化简

对矩阵 A A A 进行初等行变换,可将其化为如下分块矩阵形式:
R = [ E r F 0 0 ] R = \begin{bmatrix} E_r & F \\ 0 & 0 \end{bmatrix} R=[Er0F0]
其中:

  • E r E_r Er r r r 阶单位矩阵;
  • F F F r × ( n − r ) r \times (n-r) r×(nr) 阶矩阵;
  • 矩阵 R R R 的秩为 r r r,即 r a n k R = r {\rm rank} R = r rankR=r

由于初等行变换不改变矩阵的秩与零空间,因此有:

  • r a n k A = r a n k R = r {\rm rank} A = {\rm rank} R = r rankA=rankR=r
  • N ( A ) = N ( R ) N(A) = N(R) N(A)=N(R)(矩阵 A A A R R R 的零空间相同)。
步骤 3:构造零空间矩阵并验证

观察分块矩阵 R R R,构造其 n × ( n − r ) n \times (n-r) n×(nr) 阶零空间矩阵 P P P
P = [ − F E n − r ] P = \begin{bmatrix} -F \\ E_{n-r} \end{bmatrix} P=[FEnr]
其中 E n − r E_{n-r} Enr ( n − r ) (n-r) (nr) 阶单位矩阵。

验证 P P P 满足零空间定义(即 R P = 0 RP = 0 RP=0):
R P = [ E r F 0 0 ] [ − F E n − r ] = [ − F + F 0 ] = 0 RP = \begin{bmatrix} E_r & F \\ 0 & 0 \end{bmatrix}\begin{bmatrix} -F \\ E_{n-r} \end{bmatrix} = \begin{bmatrix} -F + F \\ 0 \end{bmatrix} = 0 RP=[Er0F0][FEnr]=[F+F0]=0

步骤 4:证明 C ( P ) = N ( R ) C(P) = N(R) C(P)=N(R)
  • 首先,矩阵 P P P 的秩 r a n k   P = n − r {\rm rank}\, P = n-r rankP=nr,说明其列向量线性无关;
  • 其次,任取 x ∈ N ( R ) x \in N(R) xN(R)(即 R x = 0 Rx = 0 Rx=0),将 x x x 按分块形式表示为 x = [ x 1 x 2 ] x = \begin{bmatrix} x_1 \\ x_2 \end{bmatrix} x=[x1x2],其中 x 1 x_1 x1 r r r 维向量, x 2 x_2 x2 ( n − r ) (n-r) (nr) 维向量。代入 R x = 0 Rx = 0 Rx=0 得:
    R x = [ E r F 0 0 ] [ x 1 x 2 ] = [ x 1 + F x 2 0 ] = 0 Rx = \begin{bmatrix} E_r & F \\ 0 & 0 \end{bmatrix}\begin{bmatrix} x_1 \\ x_2 \end{bmatrix} = \begin{bmatrix} x_1 + Fx_2 \\ 0 \end{bmatrix} = 0 Rx=[Er0F0][x1x2]=[x1+Fx20]=0
    由此推出 x 1 = − F x 2 x_1 = -Fx_2 x1=Fx2,进而有:
    x = [ x 1 x 2 ] = [ − F x 2 x 2 ] = [ − F E n − r ] x 2 = P x 2 x = \begin{bmatrix} x_1 \\ x_2 \end{bmatrix} = \begin{bmatrix} -Fx_2 \\ x_2 \end{bmatrix} = \begin{bmatrix} -F \\ E_{n-r} \end{bmatrix}x_2 = Px_2 x=[x1x2]=[Fx2x2]=[FEnr]x2=Px2
    x x x 可由 P P P 的列向量线性表出。

综上, C ( P ) = N ( R ) C(P) = N(R) C(P)=N(R),因此 dim ⁡ N ( A ) = dim ⁡ N ( R ) = r a n k P = n − r \dim N(A) = \dim N(R) = {\rm rank} P = n-r dimN(A)=dimN(R)=rankP=nr。代入 r a n k A = r {\rm rank} A = r rankA=r,得:
n = dim ⁡ N ( A ) + r a n k A n = \dim N(A) + {\rm rank} A n=dimN(A)+rankA
矩阵角度的证明完成。

2. 线性变换角度的证明

线性变换的核心是“保持线性运算”,可通过基底的映射关系证明定理。

步骤 1:设定基本条件与基底扩充

设向量空间 V \mathcal{V} V 的维数 dim ⁡ V = n \dim \mathcal{V} = n dimV=n,线性变换 T T T 的核空间维数 dim ⁡ ker ⁡ ( T ) = p \dim \ker(T) = p dimker(T)=p(显然 p ≤ n p \leq n pn)。
ker ⁡ ( T ) \ker(T) ker(T) 的一组基底 { u 1 , u 2 , … , u p } \{u_1, u_2, \dots, u_p\} {u1,u2,,up},将其扩充为 V \mathcal{V} V 的一组基底:
{ u 1 , u 2 , … , u p , w 1 , w 2 , … , w r } \{u_1, u_2, \dots, u_p, w_1, w_2, \dots, w_r\} {u1,u2,,up,w1,w2,,wr}
其中 n = p + r n = p + r n=p+r。需证明的结论转化为:
r a n k   T = r {\rm rank}\, T = r rankT=r

步骤 2:分析任意向量的像

V \mathcal{V} V 中任意向量 v v v,由基底的线性表示性质,可将其表示为:
v = a 1 u 1 + a 2 u 2 + ⋯ + a p u p + b 1 w 1 + b 2 w 2 + ⋯ + b r w r v = a_1u_1 + a_2u_2 + \cdots + a_pu_p + b_1w_1 + b_2w_2 + \cdots + b_rw_r v=a1u1+a2u2++apup+b1w1+b2w2++brwr
其中 a 1 , … , a p , b 1 , … , b r a_1, \dots, a_p, b_1, \dots, b_r a1,,ap,b1,,br 为常数。

v v v 施加线性变换 T T T,根据线性变换的性质(可加性、齐次性),其像 T ( v ) T(v) T(v) 满足:
T ( v ) = T ( a 1 u 1 + ⋯ + a p u p + b 1 w 1 + ⋯ + b r w r ) = a 1 T ( u 1 ) + ⋯ + a p T ( u p ) + b 1 T ( w 1 ) + ⋯ + b r T ( w r ) \begin{aligned} T(v) &= T\left(a_1u_1 + \cdots + a_pu_p + b_1w_1 + \cdots + b_rw_r\right) \\ &= a_1T(u_1) + \cdots + a_pT(u_p) + b_1T(w_1) + \cdots + b_rT(w_r) \end{aligned} T(v)=T(a1u1++apup+b1w1++brwr)=a1T(u1)++apT(up)+b1T(w1)++brT(wr)
由于 u 1 , … , u p ∈ ker ⁡ ( T ) u_1, \dots, u_p \in \ker(T) u1,,upker(T),故 T ( u 1 ) = ⋯ = T ( u p ) = 0 T(u_1) = \cdots = T(u_p) = 0 T(u1)==T(up)=0,因此:
T ( v ) = b 1 T ( w 1 ) + b 2 T ( w 2 ) + ⋯ + b r T ( w r ) T(v) = b_1T(w_1) + b_2T(w_2) + \cdots + b_rT(w_r) T(v)=b1T(w1)+b2T(w2)++brT(wr)
这表明 T ( v ) T(v) T(v) 可由 { T ( w 1 ) , … , T ( w r ) } \{T(w_1), \dots, T(w_r)\} {T(w1),,T(wr)} 线性表示,即 r a n   T {\rm ran}\, T ranT 可由该向量组张成。

步骤 3:证明基底的线性无关性

只需证明 { T ( w 1 ) , T ( w 2 ) , … , T ( w r ) } \{T(w_1), T(w_2), \dots, T(w_r)\} {T(w1),T(w2),,T(wr)} 线性无关,即可确认其为 r a n   T {\rm ran}\, T ranT 的一组基底(维数为 r r r)。

假设存在常数 c 1 , c 2 , … , c r c_1, c_2, \dots, c_r c1,c2,,cr,使得:
c 1 T ( w 1 ) + c 2 T ( w 2 ) + ⋯ + c r T ( w r ) = 0 c_1T(w_1) + c_2T(w_2) + \cdots + c_rT(w_r) = 0 c1T(w1)+c2T(w2)++crT(wr)=0
根据线性变换的性质,上式可改写为:
T ( c 1 w 1 + c 2 w 2 + ⋯ + c r w r ) = 0 T\left(c_1w_1 + c_2w_2 + \cdots + c_rw_r\right) = 0 T(c1w1+c2w2++crwr)=0
这表明 c 1 w 1 + ⋯ + c r w r ∈ ker ⁡ ( T ) c_1w_1 + \cdots + c_rw_r \in \ker(T) c1w1++crwrker(T)。由于 { u 1 , … , u p } \{u_1, \dots, u_p\} {u1,,up} ker ⁡ ( T ) \ker(T) ker(T) 的基底,故存在常数 d 1 , … , d p d_1, \dots, d_p d1,,dp,使得:
c 1 w 1 + ⋯ + c r w r = d 1 u 1 + ⋯ + d p u p c_1w_1 + \cdots + c_rw_r = d_1u_1 + \cdots + d_pu_p c1w1++crwr=d1u1++dpup
整理得:
d 1 u 1 + ⋯ + d p u p − c 1 w 1 − ⋯ − c r w r = 0 d_1u_1 + \cdots + d_pu_p - c_1w_1 - \cdots - c_rw_r = 0 d1u1++dpupc1w1crwr=0
由于 { u 1 , … , u p , w 1 , … , w r } \{u_1, \dots, u_p, w_1, \dots, w_r\} {u1,,up,w1,,wr} V \mathcal{V} V 的基底(线性无关),故所有系数均为零,即 c 1 = ⋯ = c r = 0 c_1 = \cdots = c_r = 0 c1==cr=0

因此, { T ( w 1 ) , … , T ( w r ) } \{T(w_1), \dots, T(w_r)\} {T(w1),,T(wr)} 线性无关,即 r a n k   T = r {\rm rank}\, T = r rankT=r。结合 n = p + r n = p + r n=p+r p = dim ⁡ ker ⁡ ( T ) p = \dim \ker(T) p=dimker(T),得:
dim ⁡ V = dim ⁡ ker ⁡ ( T ) + r a n k   T \dim \mathcal{V} = \dim \ker(T) + {\rm rank}\, T dimV=dimker(T)+rankT
线性变换角度的证明完成。

秩-零化度定理的推论

基于秩-零化度定理,可推导出关于线性变换“非零核”与“非满射”的重要结论,也可转化为矩阵语言表述。

推论 1:定义域维数大于值域维数时,存在非零核

dim ⁡ V > dim ⁡ W \dim \mathcal{V} > \dim \mathcal{W} dimV>dimW,由秩-零化度定理 dim ⁡ ker ⁡ ( T ) = dim ⁡ V − dim ⁡ r a n ( T ) \dim \ker(T) = \dim \mathcal{V} - \dim {\rm ran}(T) dimker(T)=dimVdimran(T),结合 dim ⁡ r a n ( T ) ≤ dim ⁡ W \dim {\rm ran}(T) \leq \dim \mathcal{W} dimran(T)dimW,得:
dim ⁡ ker ⁡ ( T ) ≥ dim ⁡ V − dim ⁡ W > 0 \dim \ker(T) \geq \dim \mathcal{V} - \dim \mathcal{W} > 0 dimker(T)dimVdimW>0
这表明 ker ⁡ ( T ) \ker(T) ker(T) 中存在非零向量,即存在 x ∈ V \mathbf{x} \in \mathcal{V} xV x ≠ 0 \mathbf{x} \neq 0 x=0),使得 T ( x ) = 0 T(\mathbf{x}) = 0 T(x)=0

推论 2:定义域维数小于值域维数时,变换非满射

dim ⁡ V < dim ⁡ W \dim \mathcal{V} < \dim \mathcal{W} dimV<dimW,由秩-零化度定理 dim ⁡ r a n ( T ) = dim ⁡ V − dim ⁡ ker ⁡ ( T ) \dim {\rm ran}(T) = \dim \mathcal{V} - \dim \ker(T) dimran(T)=dimVdimker(T),结合 dim ⁡ ker ⁡ ( T ) ≥ 0 \dim \ker(T) \geq 0 dimker(T)0,得:
dim ⁡ r a n ( T ) ≤ dim ⁡ V < dim ⁡ W \dim {\rm ran}(T) \leq \dim \mathcal{V} < \dim \mathcal{W} dimran(T)dimV<dimW
这表明 r a n ( T ) {\rm ran}(T) ran(T) W \mathcal{W} W 的真子空间,即存在 y ∈ W \mathbf{y} \in \mathcal{W} yW,使得 y ∉ r a n ( T ) \mathbf{y} \notin {\rm ran}(T) y/ran(T),线性变换 T T T 不是满射。

推论的矩阵语言表述

设矩阵 A A A m × n m \times n m×n 阶矩阵,对应线性变换 T : V → W T: \mathcal{V} \to \mathcal{W} T:VW dim ⁡ V = n \dim \mathcal{V} = n dimV=n dim ⁡ W = m \dim \mathcal{W} = m dimW=m),则推论可转化为:

1. 当 n > m n > m n>m(“矮胖”矩阵)时

dim ⁡ N ( A ) = n − dim ⁡ C ( A ) \dim N(A) = n - \dim C(A) dimN(A)=ndimC(A) dim ⁡ C ( A ) ≤ m \dim C(A) \leq m dimC(A)m,得:
dim ⁡ N ( A ) ≥ n − m > 0 \dim N(A) \geq n - m > 0 dimN(A)nm>0
即矩阵 A A A 的零空间 N ( A ) N(A) N(A) 包含非零向量,齐次线性方程组 A x = 0 A\mathbf{x} = 0 Ax=0 有无限多组解。

img
图 2 “矮胖”矩阵( n > m n > m n>m)的零空间示意图

2. 当 n < m n < m n<m(“瘦高”矩阵)时

dim ⁡ C ( A ) = n − dim ⁡ N ( A ) \dim C(A) = n - \dim N(A) dimC(A)=ndimN(A) dim ⁡ N ( A ) ≥ 0 \dim N(A) \geq 0 dimN(A)0,得:
dim ⁡ C ( A ) ≤ n < m \dim C(A) \leq n < m dimC(A)n<m
即矩阵 A A A 的列空间 C ( A ) C(A) C(A) 无法充满整个 R m \mathbb{R}^m Rm,非齐次线性方程组 A x = b A\mathbf{x} = b Ax=b 可能无解。

img
图 3 “瘦高”矩阵( n < m n < m n<m)的列空间示意图


列空间与零空间,秩—零化度定理

_Equinox 已于 2025-07-27 09:34:02 修改

1. 向量空间的封闭性

任取空间内 A → \overrightarrow{A} A B → \overrightarrow{B} B ,且 A → ≠ k B → \overrightarrow{A} \neq k\overrightarrow{B} A =kB ,满足 l 1 A → + l 2 B → l_1 \overrightarrow{A} + l_2 \overrightarrow{B} l1A +l2B 仍在空间内。

2. 列空间

The short useful word “span” describes all the linear combinations of a set of vectors. So the span of the columns of A (independent or not) is the column space.
“Span” 这个简短而有用的词汇描述了一组向量的所有线性组合。因此,矩阵 A 的列向量的张成(无论是否线性无关)构成了列空间

张成(span):一个向量集合的所有线性组合。矩阵 A A A 的列向量的张成,即矩阵 A A A列空间(column space)

线性方程组的意义在于:对于矩阵 A A A 和向量 B B B,方程 A X = B AX = B AX=B 有解当且仅当 B B B A A A 的列空间内,即 B ∈ C ( A ) B \in C(A) BC(A)

例如,矩阵 A A A
A = [ 1 1 2 2 1 3 3 1 4 4 1 5 ] = [ c 1 c 2 c 3 ] , C ( A ) ⊂ R 4 A = \begin{bmatrix} 1 & 1 & 2 \\ 2 & 1 & 3 \\ 3 & 1 & 4 \\ 4 & 1 & 5 \end{bmatrix} = \begin{bmatrix} c_1 & c_2 & c_3 \end{bmatrix}, \quad C(A) \subset \mathbb{R}^4 A= 123411112345 =[c1c2c3],C(A)R4
对于方程组 A X = B AX = B AX=B,有解:

当且仅当 B = k 1 c 1 + k 2 c 2 + k 3 c 3 B = k_1 c_1 + k_2 c_2 + k_3 c_3 B=k1c1+k2c2+k3c3,即 B B B A A A 的列空间内, B ∈ C ( A ) B \in C(A) BC(A)

因此,方程 A X = B AX = B AX=B 有解 <=> B B B A A A 的列空间内。

3. 零空间

The nullspace N ( A ) N(A) N(A) in R n \mathbb{R}^n Rn contains all solutions x x x to A x = 0 Ax = 0 Ax=0. This includes x = 0 x = 0 x=0.
零空间 N ( A ) N(A) N(A) R n \mathbb{R}^n Rn 中包含所有满足 A x = 0 Ax = 0 Ax=0 的解 x x x。这包括 x = 0 x = 0 x=0

矩阵 A A A 的零空间是所有满足 A X = 0 AX = 0 AX=0 的解向量的集合。显然,零向量一定属于零空间。

4. I F 型

考虑矩阵
[ 1 7 0 8 0 0 1 9 ] x = [ 0 0 ] \begin{bmatrix} 1 & 7 & 0 & 8 \\ 0 & 0 & 1 & 9 \end{bmatrix} x = \begin{bmatrix} 0 \\ 0 \end{bmatrix} [10700189]x=[00]
其解向量为
x = [ − 7 1 0 0 ] , x = [ − 8 0 − 9 1 ] x = \begin{bmatrix} -7 \\ 1 \\ 0 \\ 0 \end{bmatrix}, \quad x = \begin{bmatrix} -8 \\ 0 \\ -9 \\ 1 \end{bmatrix} x= 7100 ,x= 8091
或者可以写成矩阵形式
[ − 7 − 8 1 0 0 − 9 0 1 ] \begin{bmatrix} -7 & -8 \\ 1 & 0 \\ 0 & -9 \\ 0 & 1 \end{bmatrix} 71008091

为了更直观地表示,可以施加一个置换矩阵 P P P,使得
P P T = I PP^T = I PPT=I

[ 1 7 0 8 0 0 1 9 ] P P T [ − 7 − 8 1 0 0 − 9 0 1 ] = [ 1 7 0 8 0 0 1 9 ] [ 1 0 0 0 0 0 1 0 0 1 0 0 0 0 0 1 ] [ 1 0 0 0 0 0 1 0 0 1 0 0 0 0 0 1 ] [ − 7 − 8 1 0 0 − 9 0 1 ] = [ 1 0 7 8 0 1 0 9 ] [ − 7 − 8 1 0 0 − 9 0 1 ] = [ 0 0 0 0 0 0 0 0 ] \begin{aligned} \begin{bmatrix} 1 & 7 & 0 & 8 \\ 0 & 0 & 1 & 9 \end{bmatrix} PP^T \begin{bmatrix} -7 & -8 \\ 1 & 0 \\ 0 & -9 \\ 0 & 1 \end{bmatrix} &= \begin{bmatrix} 1 & 7 & 0 & 8 \\ 0 & 0 & 1 & 9 \end{bmatrix} \begin{bmatrix} 1 & 0 & 0 & 0 \\ 0 & 0 & 1 & 0 \\ 0 & 1 & 0 & 0 \\ 0 & 0 & 0 & 1 \end{bmatrix} \begin{bmatrix} 1 & 0 & 0 & 0 \\ 0 & 0 & 1 & 0 \\ 0 & 1 & 0 & 0 \\ 0 & 0 & 0 & 1 \end{bmatrix} \begin{bmatrix} -7 & -8 \\ 1 & 0 \\ 0 & -9 \\ 0 & 1 \end{bmatrix} \\ &= \begin{bmatrix} 1 & 0 & 7 & 8 \\ 0 & 1 & 0 & 9 \end{bmatrix} \begin{bmatrix} -7 & -8 \\ 1 & 0 \\ 0 & -9 \\ 0 & 1 \end{bmatrix} \\ &= \begin{bmatrix} 0 & 0 \\ 0 & 0 \\ 0 & 0 \\ 0 & 0 \end{bmatrix} \end{aligned} [10700189]PPT 71008091 =[10700189] 1000001001000001 1000001001000001 71008091 =[10017089] 71008091 = 00000000

更一般地,对于 A X = 0 AX = 0 AX=0,可以对 A X = 0 AX = 0 AX=0 施加高斯消元法,转换成如下形式:
[ I F ] P P T x = 0 \begin{bmatrix} I & F \end{bmatrix} P P^T x = 0 [IF]PPTx=0
其中 I I I 为单位阵, P P P 为置换阵。解向量可以表示为
[ − F I ] \begin{bmatrix} -F \\ I \end{bmatrix} [FI]

事实上,这对应了高斯消元化为阶梯型后的主元和自由元

由于 r ( I ) = r ( A ) r(I) = r(A) r(I)=r(A),可以得到 r ( F ) = n − r ( A ) r(F) = n - r(A) r(F)=nr(A)

矩阵 A A A 的零空间的维度正是 n − r ( A ) n - r(A) nr(A)

5. 秩—零化度定理

秩—零化度定理(Rank–Nullity Theorem) 表示有限维向量空间上线性变换 T : V → W T: V \rightarrow W T:VW 定义空间的维数 dim V \text{dim} V dimV 等于核空间的维数加上像空间的维数。

对于矩阵变换 T ( x ) = A x T(x) = Ax T(x)=Ax 即列数 n n n 等于零空间的维数加上列空间的维数,对于对偶映射 T ( x ) = A T x T(x) = A^T x T(x)=ATx 即行数 m m m 等于左零空间的维数加上行空间的维数,值域维数即矩阵的秩

  • dim V = dim Ker ( T ) + dim Im ( T ) \text{dim} V = \text{dim Ker}(T) + \text{dim Im}(T) dimV=dim Ker(T)+dim Im(T)
  • n = dim Nul A + dim Col A n = \text{dim Nul}A + \text{dim Col}A n=dim NulA+dim ColA
  • m = dim Nul A T + dim Row A m = \text{dim Nul}A^T + \text{dim Row}A m=dim NulAT+dim RowA

对于任意 n × n n \times n n×n 矩阵 A A A,有
r ( A ) + r ( N ( A ) ) = n r(A) + r(N(A)) = n r(A)+r(N(A))=n

根据该定理可以得出一些有关秩的关系式:

  1. 对于 n n n 阶方阵 A A A B B B,若 A B = 0 AB = 0 AB=0,则 r ( A ) + r ( B ) ≤ n r(A) + r(B) \leq n r(A)+r(B)n

    证明:
    A B = 0 ⇒ C ( B ) ⊂ N ( A ) ⇒ r ( B ) ≤ n − r ( A ) ⇒ r ( A ) + r ( B ) ≤ n AB = 0 \Rightarrow C(B) \subset N(A) \Rightarrow r(B) \leq n - r(A) \Rightarrow r(A) + r(B) \leq n AB=0C(B)N(A)r(B)nr(A)r(A)+r(B)n

  2. r ( A T A ) = r ( A ) r(A^T A) = r(A) r(ATA)=r(A)

    证明:
    A T A x = 0 ⇔ x T A T A x = 0 ⇔ ∥ A x ∥ 2 = 0 ⇔ A x = 0 A^T A x = 0 \Leftrightarrow x^T A^T A x = 0 \Leftrightarrow \|Ax\|^2 = 0 \Leftrightarrow Ax = 0 ATAx=0xTATAx=0Ax2=0Ax=0
    因此, N ( A T A ) = N ( A ) N(A^T A) = N(A) N(ATA)=N(A),从而 r ( A T A ) = r ( A ) r(A^T A) = r(A) r(ATA)=r(A)

  3. r ( A ∗ ) r(A^*) r(A) r ( A ) r(A) r(A) 的关系

    • r ( A ) = n r(A) = n r(A)=n 时( A A A 为满秩/可逆矩阵) A ∗ A^* A 也是满秩的,即 r ( A ∗ ) = n r(A^*) = n r(A)=n
    • r ( A ) = n − 1 r(A) = n - 1 r(A)=n1 A ∗ A^* A 的秩为 1,即 r ( A ∗ ) = 1 r(A^*) = 1 r(A)=1
    • r ( A ) < n − 1 r(A) < n - 1 r(A)<n1 A ∗ A^* A 是零矩阵,即 r ( A ∗ ) = 0 r(A^*) = 0 r(A)=0

    证明:

    • r ( A ) = n r(A) = n r(A)=n
      A A ∗ = ∣ A ∣ I A A^* = |A|I AA=AI
      因此, A ∗ A^* A 可逆, r ( A ∗ ) = n r(A^*) = n r(A)=n
    • r ( A ) = n − 1 r(A) = n - 1 r(A)=n1
      A A ∗ = 0 ⇒ C ( A ∗ ) ⊂ N ( A ) ⇒ r ( A ∗ ) ≤ r ( N ( A ) ) = 1 AA^* = 0 \Rightarrow C(A^*) \subset N(A) \Rightarrow r(A^*) \leq r(N(A)) = 1 AA=0C(A)N(A)r(A)r(N(A))=1
      又因为 r ( A ∗ ) ≥ 1 r(A^*) \geq 1 r(A)1,由夹逼定理可知, r ( A ∗ ) = 1 r(A^*) = 1 r(A)=1
    • r ( A ) < n − 1 r(A) < n - 1 r(A)<n1
      根据 A ∗ A^* A 的定义,因为 r ( A ) < n − 1 r(A) < n - 1 r(A)<n1,所以 A A A 的所有 n − 1 n - 1 n1 阶子式均为 0,因此 A ∗ = 0 A^* = 0 A=0
  4. r ( A T A ) = r ( A ) r(A^T A) = r(A) r(ATA)=r(A) C ( A T A ) = C ( A ) C(A^T A) = C(A) C(ATA)=C(A)

    证明:

A x = 0 ⇒ A T A x = 0 ⇒ C ( A ) ⊂ C ( A T A ) \begin{array}{lll} Ax = 0 & \Rightarrow & {A^T}Ax = 0 \\ & \Rightarrow & C(A) \subset C({A^T}A) \end{array} Ax=0ATAx=0C(A)C(ATA)

A T A x = 0 ⇒ x T A T A x = 0 ⇒ ( A x ) T ( A x ) = 0 ⇒ A x = 0 ⇒ C ( A T A ) ⊂ C ( A ) \begin{array}{lll} {A^T}Ax = 0 & \Rightarrow& {x^T}{A^T}Ax = 0\\ & \Rightarrow & {(Ax)^T}(Ax) = 0\\ & \Rightarrow & Ax = 0\\ & \Rightarrow & C({A^T}A) \subset C(A) \end{array} ATAx=0xTATAx=0(Ax)T(Ax)=0Ax=0C(ATA)C(A)

因此, C ( A T A ) = C ( A ) C(A^T A) = C(A) C(ATA)=C(A),从而 r ( A ) = r ( A T A ) r(A) = r(A^T A) r(A)=r(ATA)


via:

评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值