本文为《Linear algebra and its applications》的读书笔记
目录
The Row Space
行空间
- The set of all linear combinations of the row vectors is called the row space of A A A and is denoted by R o w A Row\ A Row A. Since the rows of A A A are identified with the columns of A T A^T AT, we could also write C o l A T Col\ A^T Col AT in place of R o w A Row\ A Row A.
- The main result of this section involves the three spaces: R o w A Row\ A Row A, C o l A Col\ A Col A, and N u l A Nul\ A Nul A. The following example shows how one sequence of row operations on A A A leads to bases for all three spaces.
EXAMPLE 2
- Find bases for the row space, the column space, and the null space of the matrix
A = [ − 2 − 5 8 0 − 17 1 3 − 5 1 5 3 11 − 19 7 1 1 7 − 13 5 − 3 ] A=\left[\begin{array}{rrrrr} -2 & -5 & 8 & 0 & -17 \\ 1 & 3 & -5 & 1 & 5 \\ 3 & 11 & -19 & 7 & 1 \\ 1 & 7 & -13 & 5 & -3 \end{array}\right] A=⎣⎢⎢⎡−2131−531178−5−19−130175−1751−3⎦⎥⎥⎤
SOLUTION
- Row reduce
A
A
A to an echelon form:
A ∼ B = [ 1 3 − 5 1 5 0 1 − 2 2 − 7 0 0 0 − 4 20 0 0 0 0 0 ] A \sim B=\left[\begin{array}{rrrrr} 1 & 3 & -5 & 1 & 5 \\ 0 & 1 & -2 & 2 & -7 \\ 0 & 0 & 0 & -4 & 20 \\ 0 & 0 & 0 & 0 & 0 \end{array}\right] A∼B=⎣⎢⎢⎡10003100−5−20012−405−7200⎦⎥⎥⎤ Basis for Row A : { ( 1 , 3 , − 5 , 1 , 5 ) , ( 0 , 1 , − 2 , 2 , − 7 ) , ( 0 , 0 , 0 , − 4 , 20 ) } Basis for Col A : { [ − 2 1 3 1 ] , [ − 5 3 11 7 ] , [ 0 1 7 5 ] } \begin{array}{l} \text { Basis for Row } A:\{(1,3,-5,1,5),(0,1,-2,2,-7),(0,0,0,-4,20)\}\\ \text { Basis for } \operatorname{Col} A:\left\{\left[\begin{array}{r} -2 \\ 1 \\ 3 \\ 1 \end{array}\right],\left[\begin{array}{r} -5 \\ 3 \\ 11 \\ 7 \end{array}\right],\left[\begin{array}{c} 0 \\ 1 \\ 7 \\ 5 \end{array}\right]\right\} \end{array} Basis for Row A:{(1,3,−5,1,5),(0,1,−2,2,−7),(0,0,0,−4,20)} Basis for ColA:⎩⎪⎪⎨⎪⎪⎧⎣⎢⎢⎡−2131⎦⎥⎥⎤,⎣⎢⎢⎡−53117⎦⎥⎥⎤,⎣⎢⎢⎡0175⎦⎥⎥⎤⎭⎪⎪⎬⎪⎪⎫ - For N u l A Nul\ A Nul A, we need the reduced echelon form. Further row operations on B B B yield A ∼ B ∼ C = [ 1 0 1 0 1 0 1 − 2 0 3 0 0 0 1 − 5 0 0 0 0 0 ] x 1 + x 3 + x 5 = 0 x 2 − 2 x 3 + 3 x 5 = 0 x 4 − 5 x 5 = 0 Basis for Nul A : { [ − 1 2 1 0 0 ] , [ − 1 − 3 0 5 1 ] } \begin{array}{l} \begin{array}{r} A \sim B \sim C=\left[\begin{array}{rrrrr} 1 & 0 & 1 & 0 & 1 \\ 0 & 1 & -2 & 0 & 3 \\ 0 & 0 & 0 & 1 & -5 \\ 0 & 0 & 0 & 0 & 0 \end{array}\right] \\ x_{1}+\quad x_{3} \quad+x_{5}=0 \\ x_{2}-2 x_{3} \quad+3 x_{5}=0 \\ x_{4}-5 x_{5}=0 \end{array}\\ \text { Basis for Nul } A:\left\{\left[\begin{array}{r} -1 \\ 2 \\ 1 \\ 0 \\ 0 \end{array}\right],\left[\begin{array}{r} -1 \\ -3 \\ 0 \\ 5 \\ 1 \end{array}\right]\right\} \end{array} A∼B∼C=⎣⎢⎢⎡100001001−200001013−50⎦⎥⎥⎤x1+x3+x5=0x2−2x3+3x5=0x4−5x5=0 Basis for Nul A:⎩⎪⎪⎪⎪⎨⎪⎪⎪⎪⎧⎣⎢⎢⎢⎢⎡−12100⎦⎥⎥⎥⎥⎤,⎣⎢⎢⎢⎢⎡−1−3051⎦⎥⎥⎥⎥⎤⎭⎪⎪⎪⎪⎬⎪⎪⎪⎪⎫
- Observe that, unlike the basis for
C
o
l
A
Col\ A
Col A, the bases for
R
o
w
A
Row\ A
Row A and
N
u
l
A
Nul\ A
Nul A have no simple connection with the entries in
A
A
A itself.
- It is possible to find a basis for the row space R o w A Row\ A Row A that uses rows of A A A. First form A T A^T AT , and then row reduce until the pivot columns of A T A^T AT are found. These pivot columns of A T A^T AT are rows of A A A, and they form a basis for the row space of A A A.
The Rank Theorem
秩定理
矩阵的秩
- Since R o w A Row\ A Row A is the same as C o l A T Col\ A^T Col AT, the dimension of the row space of A A A is the rank of A T A^T AT.
- The dimension of the null space is sometimes called the nullity (零维) of A A A.
秩定理
- r a n k ( A ) = r a n k ( A T ) rank(A)=rank(A^T) rank(A)=rank(AT)
- r a n k ( A ) + d i m N u l A = n rank(A)+dimNul\ A=n rank(A)+dimNul A=n
Study Tip: If r a n k A = 2 rank A = 2 rankA=2, any two linearly independent columns of A A A form a basis for C o l A Col\ A Col A, and any two linearly independent rows of A A A form a basis for R o w A Row\ A Row A.
EXAMPLE 3
Could a 6 × 9 6 \times 9 6×9 matrix have a two-dimensional null space?
SOLUTION
- No.
可视化零空间、列空间、行空间
- The next example provides a nice way to visualize the subspaces we have been studying. In Chapter 6, we will learn that R o w A Row\ A Row A and N u l A Nul\ A Nul A have only the zero vector in common and are actually “orthogonal” (perpendicular) to each other.
EXAMPLE 4
- Let
A
=
[
3
0
−
1
3
0
−
1
4
0
5
]
A =\begin{bmatrix} 3 &0 &-1\\3 &0 &-1\\4& 0& 5\end{bmatrix}
A=⎣⎡334000−1−15⎦⎤. It is readily checked that
N
u
l
A
Nul\ A
Nul A is the
x
2
x_2
x2-axis,
R
o
w
A
Row\ A
Row A is the
x
1
x
3
x_1x_3
x1x3-plane,
C
o
l
A
Col\ A
Col A is the plane whose equation is
x
1
−
x
2
=
0
x_1 - x_2 = 0
x1−x2=0, and
N
u
l
A
T
Nul\ A^T
Nul AT is the set of all multiples of
(
1
,
−
1
,
0
)
(1,-1, 0)
(1,−1,0). Figure 1 shows
N
u
l
A
Nul\ A
Nul A and
R
o
w
A
Row\ A
Row A in the domain of the linear transformation
x
↦
A
x
\boldsymbol x \mapsto A\boldsymbol x
x↦Ax; the range of this mapping,
C
o
l
A
Col\ A
Col A, is shown in a separate copy of
R
3
\mathbb R^3
R3, along with
N
u
l
A
T
Nul\ A^T
Nul AT.
- The four subspaces shown in Figure 1 are called the f u n d a m e n t a l s u b s p a c e s fundamental\ subspaces fundamental subspaces determined by A A A.
一些性质
矩阵乘积的秩 ≤ \leq ≤ 乘积中任何一个矩阵因子的秩
EXERCISE 12
- Show that
r
a
n
k
A
B
rank AB
rankAB cannot exceed
r
a
n
k
A
rank A
rankA or
r
a
n
k
B
rank B
rankB.
- In general, the rank of a product of matrices cannot exceed the rank of any factor in the product.
PROOF
- A B = [ A b 1 . . . A b n ] AB=\begin{bmatrix}A\boldsymbol b_1&...A\boldsymbol b_n \end{bmatrix} AB=[Ab1...Abn]. Each column of A B AB AB is a linear combination of A A A's columns. Thus each column of A B AB AB is in C o l A Col\ A Col A and C o l A B Col\ AB Col AB is clearly a subspace of C o l A Col\ A Col A. So r a n k A B ≤ r a n k A rankAB\leq rankA rankAB≤rankA.
- Using the conclusion above, it can be shown that r a n k A B = r a n k ( A B ) T = r a n k B T A T ≤ r a n k B T = r a n k B . rankAB=rank(AB)^T=rankB^TA^T\leq rankB^T=rankB. rankAB=rank(AB)T=rankBTAT≤rankBT=rankB.
EXAMPLE
Create a 6 × 7 6\times7 6×7 matrix A A A whose rank is at most 4.
- One way to make A A A is to create a random integer-valued 6 × 4 6 \times 4 6×4 matrix J J J and a random integer-valued 4 × 7 4 \times 7 4×7 matrix K K K, and set A = J K A = JK A=JK.
EXAMPLE
Let A A A be an m × n m \times n m×n matrix of rank r > 0 r > 0 r>0 and let U U U be an echelon form of A A A. Use the factorization A = E U A = EU A=EU to write A A A as the sum of r r r rank 1 matrices.
SOLUTION
- It can be proved that an
m
×
n
m \times n
m×n matrix
A
A
A has rank 1 if and only if it is an outer product; that is,
A
=
u
v
T
A = \boldsymbol u\boldsymbol v^T
A=uvT for some
u
\boldsymbol u
u in
R
m
\mathbb R^m
Rm and
v
\boldsymbol v
v in
R
n
\mathbb R^n
Rn.
A = E U = [ c o l 1 ( E ) . . . c o l m ( E ) ] [ r o w 1 ( U ) . . . r o w m ( U ) ] = c o l 1 ( E ) r o w 1 ( U ) + . . . + c o l m ( E ) r o w m ( U ) = c o l 1 ( E ) r o w 1 ( U ) + . . . + c o l r ( E ) r o w r ( U ) ( ∵ r a n k ( A ) = r ∴ r o w i ( U ) = 0 f o r i > r ) \begin{aligned} A&=EU=\begin{bmatrix}col_1(E)& ...& col_m(E)\end{bmatrix}\begin{bmatrix}row_1(U)\\ ...\\ row_m(U)\end{bmatrix}\\&=col_1(E)row_1(U)+...+col_m(E)row_m(U)\\&=col_1(E)row_1(U)+...+col_r(E)row_r(U) \\&(\because rank(A)=r \ \ \ \ \ \ \therefore row_i(U)=\boldsymbol 0\ \ \ \ for\ i>r) \end{aligned} A=EU=[col1(E)...colm(E)]⎣⎡row1(U)...rowm(U)⎦⎤=col1(E)row1(U)+...+colm(E)rowm(U)=col1(E)row1(U)+...+colr(E)rowr(U)(∵rank(A)=r ∴rowi(U)=0 for i>r)
r a n k P A = r a n k A rankPA=rankA rankPA=rankA ( P P P 为可逆矩阵)
EXERCISE 13
Show that if P P P is an invertible m × m m \times m m×m matrix, then r a n k P A = r a n k A rankPA = rankA rankPA=rankA.
PROOF
- One way to prove is to transform A A A into P − 1 P A P^{-1}PA P−1PA and use the conclusion in exercise 12.
- Another way is to show that P A PA PA and A A A have the same null space.
- 也可以这么想: P = E 1 . . . E n P=E_1...E_n P=E1...En ( E i E_i Ei is any elementary matrix)，因此 P A PA PA 就相当于对 A A A 作一系列的行变换，并不会改变它的行空间，因此 d i m R o w P A = d i m A dim\ RowPA=dim\ A dim RowPA=dim A， r a n k A = r a n k P A rank\ A=rank\ PA rank A=rank PA
EXERCISE 15
Let A A A be an m × n m \times n m×n matrix, and let B B B be an n × p n \times p n×p matrix such that A B = 0 AB = 0 AB=0. Show that r a n k A + r a n k B ≤ n rankA + rankB \leq n rankA+rankB≤n.
PROOF
- [Hint: C o l B ColB ColB is contained in N u l A NulA NulA.]
推论: 初等变换不改变矩阵的秩
秩分解
Rank Factorization
- Let A A A be an m × n m\times n m×n matrix. Construct a matrix C C C whose columns are the pivot columns of A A A, and construct a matrix R R R whose rows are the nonzero rows of the reduced echelon form of A A A. Show that A = C R A=CR A=CR.
Proof
- 下面只需要证明 a i = C r i ( 1 ≤ i ≤ n ) \boldsymbol a_i=C\boldsymbol r_i(1\leq i\leq n) ai=Cri(1≤i≤n)
-
C
C
C 实际上是
C
o
l
A
ColA
ColA 的一组由
A
A
A 中主元列组成的基，
C
C
C 的每一列
c
i
\boldsymbol c_i
ci 均为
A
A
A 中的一个主元列；同时由于行变换是不改变矩阵列之间的线性相关关系的，因此简化阶梯型
R
R
R 中的每一列实际上就表示了
A
A
A 中每一列相对于主元列的线性相关关系，因此得证
- 例如在 EXAMPLE 2 中，
C = [ 2 − 5 0 1 3 1 3 11 7 1 7 5 ] , R = [ 1 0 1 0 1 0 1 − 2 0 3 0 0 0 1 − 5 ] C=\left[\begin{array}{rrrrr} 2 & -5 & 0 \\ 1&3& 1 \\ 3& 11& 7\\ 1& 7& 5 \end{array}\right],R=\left[\begin{array}{rrrrr} 1&0 & 1&0&1\\ 0&1& -2&0&3\\ 0&0 & 0&1&-5 \end{array}\right] C=⎣⎢⎢⎡2131−531170175⎦⎥⎥⎤,R=⎣⎡1000101−2000113−5⎦⎤
- 例如在 EXAMPLE 2 中，
补充性质
- r a n k ( k A ) = r a n k ( A ) rank(kA)=rank(A) rank(kA)=rank(A) ( k ≠ 0 k\neq 0 k=0)
- r a n k ( [ A 0 0 B ] ) = r a n k ( A ) + r a n k ( B ) rank(\begin{bmatrix}A&0\\0&B\end{bmatrix})=rank(A)+rank(B) rank([A00B])=rank(A)+rank(B)
- r a n k ( [ A C 0 B ] ) ≥ r a n k ( A ) + r a n k ( B ) rank(\begin{bmatrix}A&C\\0&B\end{bmatrix})\geq rank(A)+rank(B) rank([A0CB])≥rank(A)+rank(B)
-
r
a
n
k
(
[
A
B
]
)
≤
r
a
n
k
(
A
)
+
r
a
n
k
(
B
)
rank(\begin{bmatrix}A&B\end{bmatrix})\leq rank(A)+rank(B)
rank([AB])≤rank(A)+rank(B)
- 证法1： [ A B ] \begin{bmatrix}A&B\end{bmatrix} [AB] 中所有列向量都能由 A A A 和 B B B 的基的线性组合来表示，因此得证
- 证法2： r a n k ( [ A B ] ) = r a n k ( [ E E ] [ A 0 0 B ] ) ≤ r a n k ( [ A 0 0 B ] ) = r a n k ( A ) + r a n k ( B ) rank(\begin{bmatrix}A&B\end{bmatrix})=rank(\begin{bmatrix}E&E\end{bmatrix}\begin{bmatrix}A&0\\0&B\end{bmatrix})\leq rank(\begin{bmatrix}A&0\\0&B\end{bmatrix})=rank(A)+rank(B) rank([AB])=rank([EE][A00B])≤rank([A00B])=rank(A)+rank(B)
-
r
a
n
k
(
A
+
B
)
≤
r
a
n
k
(
A
)
+
r
a
n
k
(
B
)
rank(A+B)\leq rank(A)+rank(B)
rank(A+B)≤rank(A)+rank(B)
- 证法1： r a n k ( A + B ) = r a n k ( [ A B ] [ E E ] ) ≤ r a n k ( [ A B ] ) ≤ r a n k ( A ) + r a n k ( B ) rank(A+B)=rank(\begin{bmatrix}A&B\end{bmatrix}\begin{bmatrix}E\\E\end{bmatrix})\leq rank(\begin{bmatrix}A&B\end{bmatrix})\leq rank(A)+rank(B) rank(A+B)=rank([AB][EE])≤rank([AB])≤rank(A)+rank(B)
- 证法2：设
A
A
A,
B
B
B 的列空间的基分别是
A
1
A_1
A1,
B
1
B_1
B1, 则
r a n k ( A + B ) ≤ r a n k ( [ A 1 , B 1 ] ) ≤ r a n k ( A 1 ) + r a n k ( B 1 ) = r a n k ( A ) + r a n k ( B ) rank(A+B)\leq rank([A_1,B_1])\leq rank(A_1)+rank(B_1)=rank(A)+rank(B) rank(A+B)≤rank([A1,B1])≤rank(A1)+rank(B1)=rank(A)+rank(B)(第一个不等式成立是因为 A + B A+B A+B 中的每一列都可以用 A 1 A_1 A1 与 B 1 B_1 B1 进行表示)
-
r
a
n
k
A
+
r
a
n
k
B
−
k
≤
r
a
n
k
(
A
B
)
rankA+rankB-k\leq rank(AB)
rankA+rankB−k≤rank(AB) (
A
A
A 为
m
×
k
m\times k
m×k 矩阵；
B
B
B 为
k
×
n
k\times n
k×n 矩阵)
- 证： r a n k A + r a n k B ≤ r a n k ( [ A 0 I k B ] ) = r a n k ( [ A − A B I k 0 ] ) = r a n k ( [ 0 − A B I k 0 ] ) = r a n k ( − A B ) + r a n k ( I k ) = r a n k ( A B ) + k rankA+rankB\leq rank(\begin{bmatrix}A&0\\I_k&B\end{bmatrix})=rank(\begin{bmatrix}A&-AB\\I_k&0\end{bmatrix})=rank(\begin{bmatrix}0&-AB\\I_k&0\end{bmatrix})=rank(-AB)+rank(I_k)=rank(AB)+k rankA+rankB≤rank([AIk0B])=rank([AIk−AB0])=rank([0Ik−AB0])=rank(−AB)+rank(Ik)=rank(AB)+k
- 矩阵行等价： A ∼ B ⇔ r a n k ( A ) = r a n k ( B ) A\sim B\Leftrightarrow rank(A)=rank(B) A∼B⇔rank(A)=rank(B)
- 向量组等价 (两个向量组能互相线性表示)：
r
a
n
k
(
A
)
=
r
a
n
k
(
B
)
=
r
a
n
k
(
[
A
B
]
)
rank(A)=rank(B)=rank(\begin{bmatrix}A&B\end{bmatrix})
rank(A)=rank(B)=rank([AB])
- 只需证明 A , B , [ A B ] A,B,\begin{bmatrix}A&B\end{bmatrix} A,B,[AB] 的列空间相同即可
Rank and the Invertible Matrix Theorem
3397
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
