定义
对矩阵
A
∈
C
n
→
C
m
\boldsymbol A \in \mathbb C^n \rightarrow \mathbb C^m
A∈Cn→Cm,
矩阵的核(kernel):
ker
A
=
Set of all
x
∈
C
n
such that
A
x
=
0
\text {ker} \boldsymbol A = \text{Set of all } \boldsymbol x \in \mathbb C^n \text{ such that } \boldsymbol {Ax}= \boldsymbol 0
kerA=Set of all x∈Cn such that Ax=0
矩阵的像(image/range):
range
A
=
Set of all vectors in
C
m
which are
A
x
for
x
∈
C
n
\text {range} \boldsymbol A = \text{Set of all vectors in } \mathbb C^m \text{ which are } \boldsymbol {Ax} \text{ for} \boldsymbol x \in \mathbb C^n
rangeA=Set of all vectors in Cm which are Ax forx∈Cn
性质
(1)
(
range
A
)
⊥
=
ker
A
H
{\left ( \text {range} \boldsymbol A \right )}^{\bot} = \text{ker} \boldsymbol A^H
(rangeA)⊥=kerAH
证明:
x
∈
(
range
A
)
⊥
⟺
<
x
,
A
y
>
=
0
⟺
<
A
H
x
,
y
>
=
0
∀
y
⟺
A
H
x
=
0
\boldsymbol x \in (\text {range } \boldsymbol A )^{\bot} \Longleftrightarrow <\boldsymbol x,\boldsymbol A \boldsymbol y>=0 \Longleftrightarrow <\boldsymbol A^H \boldsymbol x,\boldsymbol y>=0 \ \forall \boldsymbol y \Longleftrightarrow \boldsymbol A^H \boldsymbol x=0
x∈(range A)⊥⟺<x,Ay>=0⟺<AHx,y>=0 ∀y⟺AHx=0
(2)
(
ker
A
)
⊥
=
range
A
H
{\left ( \text {ker} \boldsymbol A \right )}^{\bot} = \text{range} \boldsymbol A^H
(kerA)⊥=rangeAH
证明:
(
range
A
H
)
⊥
=
ker
A
⟹
(
range
A
H
)
⊥
⊥
=
(
ker
A
)
⊥
⟺
(
range
A
H
)
=
(
ker
A
)
⊥
{\left ( \text {range} \boldsymbol A^H \right )}^{\bot} = \text{ker} \boldsymbol A \Longrightarrow ( \text {range} \boldsymbol A^H)^{\bot \bot} = { (\text {ker} \boldsymbol A) }^{\bot} \Longleftrightarrow ( \text {range} \boldsymbol A^H)^{} = { (\text {ker} \boldsymbol A) }^{\bot}
(rangeAH)⊥=kerA⟹(rangeAH)⊥⊥=(kerA)⊥⟺(rangeAH)=(kerA)⊥
(3) 若矩阵
A
∈
R
m
×
n
\boldsymbol A \in \mathbb R^{m \times n}
A∈Rm×n列满秩(即线性映射:
R
n
→
R
m
\mathbb R^{n} \rightarrow \mathbb R^{m}
Rn→Rm为单射),则
range
(
A
)
=
range
(
A
(
A
T
A
)
−
1
A
T
)
\text {range} (\boldsymbol A) = \text{range} \left( {\boldsymbol A (\boldsymbol A^T \boldsymbol A)^{-1} \boldsymbol A^T} \right )
range(A)=range(A(ATA)−1AT)
证明:要证等式成立,只需证
(
A
T
A
)
−
1
A
T
(\boldsymbol A^T \boldsymbol A)^{-1} \boldsymbol A^T
(ATA)−1AT的线性映射:
R
m
→
R
n
\mathbb R^{m} \rightarrow \mathbb R^{n}
Rm→Rn为满射。因为
A
x
=
y
∈
range
(
A
)
,
∀
x
⟹
x
=
(
A
T
A
)
−
1
A
T
y
,
∀
x
\boldsymbol A \boldsymbol x = \boldsymbol y \in \text{range}(\boldsymbol A), \ \forall \boldsymbol x \\ \Longrightarrow \boldsymbol x = (\boldsymbol A^T \boldsymbol A)^{-1} \boldsymbol A^T \boldsymbol y, \forall \boldsymbol x
Ax=y∈range(A), ∀x⟹x=(ATA)−1ATy,∀x
因此 ( A T A ) − 1 A T (\boldsymbol A^T \boldsymbol A)^{-1} \boldsymbol A^T (ATA)−1AT的线性映射为满射,因此 range ( A ) = range ( A ( A T A ) − 1 A T ) \text {range} (\boldsymbol A) = \text{range} \left( {\boldsymbol A (\boldsymbol A^T \boldsymbol A)^{-1} \boldsymbol A^T} \right ) range(A)=range(A(ATA)−1AT)
(4) 对一般的矩阵
A
∈
R
m
×
n
\boldsymbol A \in \mathbb R^{m \times n}
A∈Rm×n,
rank
(
A
)
=
r
\text{rank}(\boldsymbol A)=r
rank(A)=r,
A
\boldsymbol A
A的奇异值分解为:
A
=
U
Σ
V
T
\boldsymbol A=\boldsymbol U \boldsymbol {\Sigma} \boldsymbol V^T
A=UΣVT,其中,
U
∈
R
m
×
r
,
Σ
∈
R
r
×
r
,
V
∈
R
n
×
r
\boldsymbol U \in \mathbb R^{m \times r}, \ \boldsymbol {\Sigma} \in \mathbb R^{r \times r}, \ \boldsymbol V \in \mathbb R^{n \times r}
U∈Rm×r, Σ∈Rr×r, V∈Rn×r,则
range
(
A
)
=
range
(
U
)
=
range
(
U
U
T
)
\text{range}(\boldsymbol A)=\text{range}(\boldsymbol U)=\text{range}(\boldsymbol U \boldsymbol U^T)
range(A)=range(U)=range(UUT)。
证明:
A
=
U
Σ
V
T
⟹
A
T
=
V
Σ
U
T
⟹
ker
(
A
T
)
=
range
(
U
)
⊥
=
ker
(
U
T
)
=
ker
(
U
U
T
)
⟹
range
(
A
)
=
range
(
U
)
=
range
(
U
U
T
)
\boldsymbol A=\boldsymbol U \boldsymbol {\Sigma} \boldsymbol V^T \\ \Longrightarrow \boldsymbol A^T=\boldsymbol V \boldsymbol {\Sigma} \boldsymbol U^T \\ \Longrightarrow \text{ker}(\boldsymbol A^T)=\text{range}(\boldsymbol U)^{\bot}=\text{ker}(\boldsymbol U^T)=\text{ker}(\boldsymbol U \boldsymbol U^T) \\ \Longrightarrow \text{range}(\boldsymbol A) = \text{range}(\boldsymbol U) = \text{range}(\boldsymbol U \boldsymbol U^T)
A=UΣVT⟹AT=VΣUT⟹ker(AT)=range(U)⊥=ker(UT)=ker(UUT)⟹range(A)=range(U)=range(UUT)
补充: U U T \boldsymbol U \boldsymbol U^T UUT把奇异对角阵理解为单位阵,因此其秩也为 r r r。