矩阵的秩
定义1.1:对于 m × n m\times n m×n阶矩阵 A \bold A A,将不为零的子式的最大阶数 k k k称作矩阵 A \bold A A的秩,记作 r ( A ) = k r(\bold A)=k r(A)=k。
定理1.1:初等变换不改变矩阵的秩.
证明:
\quad
先证明
A
\bold A
A经过一次初等行变换为
B
\bold B
B,则
r
(
A
)
≤
r
(
B
)
r(\bold A)\le r(\bold B)
r(A)≤r(B).
\qquad
\quad
设
r
(
A
)
=
k
r(\bold A)=k
r(A)=k,则
A
\bold A
A的一个
k
k
k阶子式
D
≠
0
\bold D\ne0
D=0.
\qquad
\quad
(1)若进行
A
\bold A
A交换两行的初等变换为
B
\bold B
B,显然能找到子式
D
1
=
−
D
\bold {D_1}=-\bold D
D1=−D or
D
≠
0
\bold D\ne 0
D=0,故
r
(
B
)
≥
k
r(\bold B)\ge k
r(B)≥k.
\qquad
\quad
(2)若进行
A
\bold A
A某行数乘的初等变换为
B
\bold B
B,显然能找到子式
D
1
=
k
D
\bold {D_1}=k\bold D
D1=kD or
D
≠
0
\bold D\ne 0
D=0,故
r
(
B
)
≥
k
r(\bold B)\ge k
r(B)≥k.
\qquad
\quad
(3)若将第
i
i
i行的
k
k
k倍加至第
j
j
j行变换为
B
\bold B
B,分如下情况考虑:
\qquad
\quad
\quad
(3.1)如
A
\bold A
A的k阶子式
D
\bold D
D不包含第
j
j
j行,则
B
\bold B
B同样包含子式
D
\bold D
D,故
r
(
B
)
≥
k
r(\bold B)\ge k
r(B)≥k.
\qquad
\quad
\quad
(3.2)如
A
\bold A
A的k阶子式
D
\bold D
D包含第
j
j
j行,则
B
\bold B
B对应(
D
1
与
D
\bold {D_1}与\bold D
D1与D行对应)子式有:
\qquad
\quad
\qquad
\quad
\qquad
D
1
=
[
α
p
⋮
α
l
⋮
α
j
+
k
α
i
⋮
α
q
]
=
[
α
p
⋮
α
l
⋮
α
j
⋮
α
q
]
+
k
[
α
p
⋮
α
l
⋮
α
i
⋮
α
q
]
=
D
+
k
D
2
\bold {D_1}=\begin{bmatrix} \alpha_p\\ \vdots \\ \alpha_l \\ \vdots \\ \alpha_j + k\alpha_i \\ \vdots \\ \alpha_q\end{bmatrix}=\begin{bmatrix} \alpha_p\\ \vdots \\ \alpha_l \\ \vdots \\ \alpha_j \\ \vdots \\ \alpha_q\end{bmatrix}+k\begin{bmatrix} \alpha_p\\ \vdots \\ \alpha_l \\ \vdots \\ \alpha_i \\ \vdots \\ \alpha_q\end{bmatrix}=\bold {D}+k\bold {D_2}
D1=⎣
⎡αp⋮αl⋮αj+kαi⋮αq⎦
⎤=⎣
⎡αp⋮αl⋮αj⋮αq⎦
⎤+k⎣
⎡αp⋮αl⋮αi⋮αq⎦
⎤=D+kD2
\qquad
\quad
\qquad
\quad
若
D
1
\bold {D_1}
D1包含
D
\bold D
D的第
i
i
i行,即
α
l
=
α
i
\alpha_l=\alpha_i
αl=αi,则
D
2
=
0
\bold {D_2}=0
D2=0,此时
D
1
≠
0
\bold {D_1}\ne0
D1=0.
\qquad
\quad
\qquad
\quad
若
D
1
\bold {D_1}
D1不包含
D
\bold D
D的第
i
i
i行,则
D
1
−
k
D
2
=
D
≠
0
\bold {D_1}-k\bold {D_2}=\bold {D}\ne0
D1−kD2=D=0,那么
D
1
\bold {D_1}
D1与
D
2
\bold {D_2}
D2不同时为零
\qquad
\quad
综上,总可以寻找到
B
\bold B
B的
k
k
k阶子式不为零,即
r
(
B
)
≥
k
=
r
(
A
)
r(\bold B)\ge k=r(\bold A)
r(B)≥k=r(A).
\qquad
\quad
由传递性,进行初等行变换将不影响矩阵的秩。
\qquad
\quad
对
A
T
\bold A^T
AT进行初等行变换为
B
T
\bold B^T
BT即相当于对
A
\bold A
A进行列变换为
B
\bold B
B,又
r
(
A
T
)
=
r
(
B
T
)
r(\bold A^T)=r(\bold B^T)
r(AT)=r(BT),且转置不
\qquad
\quad
影响矩阵的秩,则
r
(
A
)
=
r
(
B
)
r(\bold A)=r(\bold B)
r(A)=r(B),即初等列变换也不影响矩阵的秩。
定理1.2:设有 m × n m\times n m×n阶矩阵 A \bold A A与可逆矩阵 P \bold P P( m m m阶), Q \bold Q Q( n n n阶),则有 r ( A ) = r ( P A ) = r ( A Q ) = r ( P A Q ) r(\bold A)=r(\bold {PA})=r(\bold {AQ})=r(\bold {PAQ}) r(A)=r(PA)=r(AQ)=r(PAQ),即左(右)乘可逆矩阵不改变矩阵的秩。
证明该定理需先理解定理1.3与定理1.4:
定理1.3:对 A \bold A A进行初等行变换相当于左乘一系列初等矩阵( R i j \bold R_{ij} Rij or R i ( k ) \bold R_{i(k)} Ri(k)or R i + j ( k ) \bold R_{i+j(k)} Ri+j(k)),对 A \bold A A进行初等列变换相当于右乘一系列初等矩阵( C i j \bold C_{ij} Cij or C i ( k ) \bold C_{i(k)} Ci(k)or C i + j ( k ) \bold C_{i+j(k)} Ci+j(k)).
定义1.2:对单位矩阵进行一次初等变换得到的矩阵称为初等矩阵,包括:
(1)替换任意两行(或者两列)
R
i
j
\bold {R_{ij}}
Rij(或
C
i
j
\bold {C_{ij}}
Cij).
\qquad
\qquad
\qquad
\qquad
\qquad
\qquad
\qquad
i
\qquad i
i列
\qquad
j
\ \ \ j
j列
\qquad
\quad
\qquad
\quad
R
i
j
=
C
i
j
=
[
1
0
…
0
0
0
…
0
0
0
…
0
0
1
…
0
0
0
…
0
0
0
…
0
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
0
0
…
1
0
0
…
0
0
0
…
0
0
0
…
0
0
0
…
0
1
0
…
0
0
0
…
0
0
1
…
0
0
0
…
0
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
0
0
…
0
0
0
…
1
0
0
…
0
0
0
…
0
1
0
…
0
0
0
…
0
0
0
…
0
0
0
…
0
0
1
…
0
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
0
0
…
0
0
0
…
0
0
0
…
1
]
\bold {R_{ij}}=\bold {C_{ij}}=\begin{bmatrix} 1 \ 0 \ \dots \ 0 \ 0\ 0 \ \dots\ 0\ 0\ 0\ \dots\ 0\\ 0 \ 1 \ \dots \ 0 \ 0\ 0 \ \dots\ 0\ 0\ 0\ \dots\ 0\\ \ \vdots \ \ \vdots \qquad \vdots \ \ \vdots\ \ \vdots \qquad \vdots \ \ \vdots\ \ \vdots \qquad \ \vdots\\ 0 \ 0\ \dots \ 1\ 0 \ 0 \ \dots\ 0\ 0\ 0\ \dots\ 0\\ 0 \ 0\ \dots \ 0\ 0 \ 0 \ \dots\ 0\ 1\ 0\ \dots\ 0\\ 0 \ 0\ \dots \ 0\ 0 \ 1 \ \dots\ 0\ 0\ 0\ \dots\ 0\\ \ \vdots \ \ \vdots \qquad \vdots \ \ \vdots\ \ \vdots \qquad \vdots \ \ \vdots\ \ \vdots \qquad \ \vdots\\ 0 \ 0\ \dots \ 0\ 0 \ 0 \ \dots\ 1\ 0\ 0\ \dots\ 0\\ 0 \ 0\ \dots \ 0\ 1 \ 0 \ \dots\ 0\ 0\ 0\ \dots\ 0\\ 0 \ 0\ \dots \ 0\ 0 \ 0 \ \dots\ 0\ 0\ 1\ \dots\ 0\\ \ \vdots \ \ \vdots \qquad \vdots \ \ \vdots\ \ \vdots \qquad \vdots \ \ \vdots\ \ \vdots \qquad \ \vdots\\ 0 \ 0\ \dots \ 0\ 0 \ 0 \ \dots\ 0\ 0\ 0\ \dots\ 1\\ \end{bmatrix}
Rij=Cij=⎣
⎡1 0 … 0 0 0 … 0 0 0 … 00 1 … 0 0 0 … 0 0 0 … 0 ⋮ ⋮⋮ ⋮ ⋮⋮ ⋮ ⋮ ⋮0 0 … 1 0 0 … 0 0 0 … 00 0 … 0 0 0 … 0 1 0 … 00 0 … 0 0 1 … 0 0 0 … 0 ⋮ ⋮⋮ ⋮ ⋮⋮ ⋮ ⋮ ⋮0 0 … 0 0 0 … 1 0 0 … 00 0 … 0 1 0 … 0 0 0 … 00 0 … 0 0 0 … 0 0 1 … 0 ⋮ ⋮⋮ ⋮ ⋮⋮ ⋮ ⋮ ⋮0 0 … 0 0 0 … 0 0 0 … 1⎦
⎤
i
行
j
行
\begin{matrix} \\ \\ \\ \\ i行 \\ \\ \\ \\ j行\\ \\ \\ \\ \end{matrix}
i行j行
(2)非零数
k
k
k乘任意行(或列)
R
i
(
k
)
\bold {R_{i(k)}}
Ri(k)(或
C
i
(
k
)
\bold {C_{i(k)}}
Ci(k)).
\qquad
\qquad
\qquad
\qquad
\qquad
\qquad
\qquad
i
\qquad i
i列
\qquad
\qquad
\quad
\qquad
R
i
(
k
)
=
C
i
(
k
)
=
[
1
0
…
0
0
0
…
0
0
0
…
0
0
1
…
0
0
0
…
0
0
0
…
0
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
0
0
…
1
0
0
…
0
0
0
…
0
0
0
…
0
k
0
…
0
0
0
…
0
0
0
…
0
0
1
…
0
0
0
…
0
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
0
0
…
0
0
0
…
1
0
0
…
0
0
0
…
0
0
0
…
0
1
0
…
0
0
0
…
0
0
0
…
0
0
1
…
0
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
0
0
…
0
0
0
…
0
0
0
…
1
]
\bold {R_{i(k)}}=\bold {C_{i(k)}}=\begin{bmatrix} 1 \ 0 \ \dots \ 0 \ 0\ 0 \ \dots\ 0\ 0\ 0\ \dots\ 0\\ 0 \ 1 \ \dots \ 0 \ 0\ 0 \ \dots\ 0\ 0\ 0\ \dots\ 0\\ \ \vdots \ \ \vdots \qquad \vdots \ \ \vdots\ \ \vdots \qquad \vdots \ \ \vdots\ \ \vdots \qquad \ \vdots\\ 0 \ 0\ \dots \ 1\ 0 \ 0 \ \dots\ 0\ 0\ 0\ \dots\ 0\\ 0 \ 0\ \dots \ 0\ k \ 0 \ \dots\ 0\ 0\ 0\ \dots\ 0\\ 0 \ 0\ \dots \ 0\ 0 \ 1 \ \dots\ 0\ 0\ 0\ \dots\ 0\\ \ \vdots \ \ \vdots \qquad \vdots \ \ \vdots\ \ \vdots \qquad \vdots \ \ \vdots\ \ \vdots \qquad \ \vdots\\ 0 \ 0\ \dots \ 0\ 0 \ 0 \ \dots\ 1\ 0\ 0\ \dots\ 0\\ 0 \ 0\ \dots \ 0\ 0 \ 0 \ \dots\ 0\ 1\ 0\ \dots\ 0\\ 0 \ 0\ \dots \ 0\ 0 \ 0 \ \dots\ 0\ 0\ 1\ \dots\ 0\\ \ \vdots \ \ \vdots \qquad \vdots \ \ \vdots\ \ \vdots \qquad \vdots \ \ \vdots\ \ \vdots \qquad \ \vdots\\ 0 \ 0\ \dots \ 0\ 0 \ 0 \ \dots\ 0\ 0\ 0\ \dots\ 1\\ \end{bmatrix}
Ri(k)=Ci(k)=⎣
⎡1 0 … 0 0 0 … 0 0 0 … 00 1 … 0 0 0 … 0 0 0 … 0 ⋮ ⋮⋮ ⋮ ⋮⋮ ⋮ ⋮ ⋮0 0 … 1 0 0 … 0 0 0 … 00 0 … 0 k 0 … 0 0 0 … 00 0 … 0 0 1 … 0 0 0 … 0 ⋮ ⋮⋮ ⋮ ⋮⋮ ⋮ ⋮ ⋮0 0 … 0 0 0 … 1 0 0 … 00 0 … 0 0 0 … 0 1 0 … 00 0 … 0 0 0 … 0 0 1 … 0 ⋮ ⋮⋮ ⋮ ⋮⋮ ⋮ ⋮ ⋮0 0 … 0 0 0 … 0 0 0 … 1⎦
⎤
i
行
\begin{matrix} \\ \\ \\ \\ i行 \\ \\ \\ \\ \\ \\ \\ \\ \end{matrix}
i行
(3)某行(或列)的倍数加至另一行(或列)上
R
i
+
j
(
k
)
\bold {R_{i+j(k)}}
Ri+j(k)(或
C
i
+
j
(
k
)
\bold {C_{i+j(k)}}
Ci+j(k)).
\qquad
\qquad
\qquad
\qquad
\qquad
\qquad
\qquad
i
\qquad i
i列
\qquad
j
\ \ \ j
j列
\qquad
\quad
R
i
+
j
(
k
)
=
C
j
+
i
(
k
)
=
[
1
0
…
0
0
0
…
0
0
0
…
0
0
1
…
0
0
0
…
0
0
0
…
0
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
0
0
…
1
0
0
…
0
0
0
…
0
0
0
…
0
1
0
…
0
k
0
…
0
0
0
…
0
0
1
…
0
0
0
…
0
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
0
0
…
0
0
0
…
1
0
0
…
0
0
0
…
0
0
0
…
0
1
0
…
0
0
0
…
0
0
0
…
0
0
1
…
0
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
⋮
0
0
…
0
0
0
…
0
0
0
…
1
]
\bold {R_{i+j(k)}}=\bold {C_{j+i(k)}}=\begin{bmatrix} 1 \ 0 \ \dots \ 0 \ 0\ 0 \ \dots\ 0\ 0\ 0\ \dots\ 0\\ 0 \ 1 \ \dots \ 0 \ 0\ 0 \ \dots\ 0\ 0\ 0\ \dots\ 0\\ \ \vdots \ \ \vdots \qquad \vdots \ \ \vdots\ \ \vdots \qquad \vdots \ \ \vdots\ \ \vdots \qquad \ \vdots\\ 0 \ 0\ \dots \ 1\ 0 \ 0 \ \dots\ 0\ 0\ 0\ \dots\ 0\\ 0 \ 0\ \dots \ 0\ 1 \ 0 \ \dots\ 0\ k\ 0\ \dots\ 0\\ 0 \ 0\ \dots \ 0\ 0 \ 1 \ \dots\ 0\ 0\ 0\ \dots\ 0\\ \ \vdots \ \ \vdots \qquad \vdots \ \ \vdots\ \ \vdots \qquad \vdots \ \ \vdots\ \ \vdots \qquad \ \vdots\\ 0 \ 0\ \dots \ 0\ 0 \ 0 \ \dots\ 1\ 0\ 0\ \dots\ 0\\ 0 \ 0\ \dots \ 0\ 0 \ 0 \ \dots\ 0\ 1\ 0\ \dots\ 0\\ 0 \ 0\ \dots \ 0\ 0 \ 0 \ \dots\ 0\ 0\ 1\ \dots\ 0\\ \ \vdots \ \ \vdots \qquad \vdots \ \ \vdots\ \ \vdots \qquad \vdots \ \ \vdots\ \ \vdots \qquad \ \vdots\\ 0 \ 0\ \dots \ 0\ 0 \ 0 \ \dots\ 0\ 0\ 0\ \dots\ 1\\ \end{bmatrix}
Ri+j(k)=Cj+i(k)=⎣
⎡1 0 … 0 0 0 … 0 0 0 … 00 1 … 0 0 0 … 0 0 0 … 0 ⋮ ⋮⋮ ⋮ ⋮⋮ ⋮ ⋮ ⋮0 0 … 1 0 0 … 0 0 0 … 00 0 … 0 1 0 … 0 k 0 … 00 0 … 0 0 1 … 0 0 0 … 0 ⋮ ⋮⋮ ⋮ ⋮⋮ ⋮ ⋮ ⋮0 0 … 0 0 0 … 1 0 0 … 00 0 … 0 0 0 … 0 1 0 … 00 0 … 0 0 0 … 0 0 1 … 0 ⋮ ⋮⋮ ⋮ ⋮⋮ ⋮ ⋮ ⋮0 0 … 0 0 0 … 0 0 0 … 1⎦
⎤
i
行
j
行
\begin{matrix} \\ \\ \\ \\ i行 \\ \\ \\ \\ j行\\ \\ \\ \\ \end{matrix}
i行j行
由行列式的性质可知初等矩阵的行列式均不为零,则初等矩阵为可逆矩阵,易验证其逆分别为:
\qquad\qquad\qquad
R
i
j
−
1
=
R
i
j
\bold {R_{ij}^{-1}}=\bold {R_{ij}}
Rij−1=Rij;
\qquad
R
i
(
k
)
−
1
=
R
i
(
1
k
)
\bold {R_{i(k)}^{-1}}=\bold {R_{i(\frac{1}{k})}}
Ri(k)−1=Ri(k1);
\qquad
R
i
+
j
(
k
)
−
1
=
R
i
−
j
(
k
)
\bold {R_{i+j(k)}^{-1}}=\bold {R_{i-j(k)}}
Ri+j(k)−1=Ri−j(k);
\qquad\qquad\qquad
C
i
j
−
1
=
C
i
j
\bold {C_{ij}^{-1}}=\bold {C_{ij}}
Cij−1=Cij;
\qquad
C
i
(
k
)
−
1
=
C
i
(
1
k
)
\bold {C_{i(k)}^{-1}}=\bold {C_{i(\frac{1}{k})}}
Ci(k)−1=Ci(k1);
\qquad
C
i
+
j
(
k
)
−
1
=
C
i
−
j
(
k
)
\bold {C_{i+j(k)}^{-1}}=\bold {C_{i-j(k)}}
Ci+j(k)−1=Ci−j(k);
现对定理1.3进行证明:
证明: \quad 现有 m × n m\times n m×n阶矩阵 A = [ A 1 r A 2 r ⋮ A m r ] \bold A=\begin{bmatrix}A_1^r \\ A_2^r\\ \vdots \\A_m^r \end{bmatrix} A=⎣ ⎡A1rA2r⋮Amr⎦ ⎤(按行分块)与 m m m阶方阵 B \bold B B.
\qquad \ \ \quad 则, B A = [ b 11 A 1 r + b 12 A 2 r + ⋯ + b 1 m A m r b 21 A 1 r + b 22 A 2 r + ⋯ + b 2 m A m r ⋮ b m 1 A 1 r + b m 2 A 2 r + ⋯ + b m m A m r ] \bold {BA}=\begin{bmatrix} b_{11}\bold {A_1^r}+b_{12}\bold {A_2^r}+\dots+ b_{1m}\bold {A_m^r}\\ b_{21}\bold {A_1^r}+b_{22}\bold {A_2^r}+\dots+ b_{2m}\bold {A_m^r}\\ \vdots \\ b_{m1}\bold {A_1^r}+b_{m2}\bold {A_2^r}+\dots+ b_{mm}\bold {A_m^r}\\ \end{bmatrix} BA=⎣ ⎡b11A1r+b12A2r+⋯+b1mAmrb21A1r+b22A2r+⋯+b2mAmr⋮bm1A1r+bm2A2r+⋯+bmmAmr⎦ ⎤
\qquad
\ \ \quad
(1)若
B
=
R
i
j
\bold B=\bold {R_{ij}}
B=Rij,则
B
A
=
[
A
1
r
A
2
r
⋮
A
j
r
⋮
A
i
r
⋮
A
m
r
]
\bold {BA}=\begin{bmatrix}\bold {A_1^r}\\ \bold {A_2^r}\\ \vdots \\ \bold {A_j^r}\\ \vdots\\ \bold {A_i^r}\\ \vdots\\ \bold {A_m^r}\end{bmatrix}
BA=⎣
⎡A1rA2r⋮Ajr⋮Air⋮Amr⎦
⎤;
\qquad
\ \ \quad
(2)若
B
=
R
i
(
k
)
\bold B=\bold {R_{i(k)}}
B=Ri(k),则
B
A
=
[
A
1
r
A
2
r
⋮
k
A
i
r
⋮
A
m
r
]
\bold {BA}=\begin{bmatrix}\bold {A_1^r}\\ \bold {A_2^r}\\ \vdots \\ k\bold {A_i^r}\\ \vdots\\ \bold {A_m^r}\end{bmatrix}
BA=⎣
⎡A1rA2r⋮kAir⋮Amr⎦
⎤;
\qquad
\ \ \quad
(3)若
B
=
R
i
+
j
(
k
)
\bold B=\bold {R_{i+j(k)}}
B=Ri+j(k),则
B
A
=
[
A
1
r
A
2
r
⋮
A
i
r
+
k
A
j
r
⋮
A
j
r
⋮
A
m
r
]
\bold {BA}=\begin{bmatrix}\bold {A_1^r}\\ \bold {A_2^r}\\ \vdots \\ \bold {A_i^r}+k \bold {A_j^r}\\ \vdots\\ \bold {A_j^r}\\ \vdots\\ \bold {A_m^r}\end{bmatrix}
BA=⎣
⎡A1rA2r⋮Air+kAjr⋮Ajr⋮Amr⎦
⎤;
\qquad
\ \ \quad
显然,左乘一个初等矩阵完成了一次相应的初等行变换。
\qquad
\ \ \quad
同样,对
A
=
[
A
1
c
A
2
c
…
A
n
c
]
\bold A=\begin{bmatrix}A_1^c \ A_2^c\dots A_n^c \end{bmatrix}
A=[A1c A2c…Anc](按列分块)与
n
n
n阶方阵
C
\bold C
C有:
\qquad \ \ \quad A C = [ ( c 11 A 1 c + c 21 A 2 c + ⋯ + c n 1 A n c ) T ( c 12 A 1 c + c 22 A 2 c + ⋯ + c n 2 A n c ) T ⋮ ( c 1 n A 1 c + c 2 n A 2 c + ⋯ + c n n A n c ) T ] T \bold {AC}=\begin{bmatrix} (c_{11}\bold {A_1^c}+c_{21}\bold {A_2^c}+\dots+ c_{n1}\bold {A_n^c})^T\\ (c_{12}\bold {A_1^c}+c_{22}\bold {A_2^c}+\dots+ c_{n2}\bold {A_n^c})^T\\ \vdots \\ (c_{1n}\bold {A_1^c}+c_{2n}\bold {A_2^c}+\dots+ c_{nn}\bold {A_n^c})^T\\ \end{bmatrix}^T AC=⎣ ⎡(c11A1c+c21A2c+⋯+cn1Anc)T(c12A1c+c22A2c+⋯+cn2Anc)T⋮(c1nA1c+c2nA2c+⋯+cnnAnc)T⎦ ⎤T
\qquad
\ \ \quad
(1)若
C
=
C
i
j
\bold C=\bold {C_{ij}}
C=Cij,则
A
C
=
[
A
1
c
,
A
2
c
,
…
,
A
j
c
i
,
…
,
A
i
c
j
,
…
,
A
m
c
]
\bold {AC}=[\bold {A_1^c},\bold {A_2^c},\dots,\underset{i}{\bold {A_j^c}},\dots,\underset{j}{\bold {A_i^c}},\dots,\bold {A_m^c}]
AC=[A1c,A2c,…,iAjc,…,jAic,…,Amc];
\qquad
\ \ \quad
(2)若
C
=
C
i
(
k
)
\bold C=\bold {C_{i(k)}}
C=Ci(k),则
A
C
=
[
A
1
c
,
A
2
c
,
…
,
k
A
i
c
i
,
…
,
A
m
c
]
T
\bold {AC}=[\bold {A_1^c},\bold {A_2^c},\dots,\underset{i}{k\bold {A_i^c}},\dots,\bold {A_m^c}]^T
AC=[A1c,A2c,…,ikAic,…,Amc]T;
\qquad
\ \ \quad
(3)若
C
=
C
i
+
j
(
k
)
\bold C=\bold {C_{i+j(k)}}
C=Ci+j(k),则
A
C
=
[
A
1
c
,
A
2
c
,
…
,
A
i
c
i
,
…
,
A
j
c
+
k
A
i
c
j
,
…
,
A
m
c
]
T
\bold {AC}=[\bold {A_1^c},\bold {A_2^c},\dots,\underset{i}{\bold {A_i^c}},\dots,\underset{j}{\bold {A_j^c}+k\bold {A_i^c}},\dots,\bold {A_m^c}]^T
AC=[A1c,A2c,…,iAic,…,jAjc+kAic,…,Amc]T.
\qquad
\ \ \quad
显然,右乘一个初等矩阵完成了一次相应的初等列变换。(证毕)
定理1.4:可逆矩阵可表示为有限个初等矩阵的乘积。
证明:
\quad
设有
n
n
n阶可逆矩阵
A
\bold A
A,由定理1.3可知存在一系列初等矩阵使得:
\qquad
\ \quad \quad\quad
R
p
R
p
−
1
…
R
2
R
1
A
C
1
C
2
…
C
q
−
1
C
q
=
B
=
[
I
r
0
0
0
]
\bold {R_pR_{p-1}\dots R_2R_1AC_1C_2\dots C_{q-1}C_q}=\bold B=\begin{bmatrix} \bold {I_r}\ 0\\ 0 \ \ 0\end{bmatrix}
RpRp−1…R2R1AC1C2…Cq−1Cq=B=[Ir 00 0] 其中,
r
(
A
)
=
r
r(\bold A)=r
r(A)=r.
\qquad
\ \ \quad
由于
A
\bold A
A与初等矩阵可逆(行列式非零),则
B
\bold B
B中不含零行或零列,则
r
=
n
r=n
r=n(满秩矩阵),即
\qquad
\ \quad \quad\quad
R
p
R
p
−
1
…
R
2
R
1
A
C
1
C
2
…
C
q
−
1
C
q
=
I
n
\bold {R_pR_{p-1}\dots R_2R_1AC_1C_2\dots C_{q-1}C_q}=\bold {I_n}
RpRp−1…R2R1AC1C2…Cq−1Cq=In
\qquad \ \ \quad ⟹ \Longrightarrow ⟹ A = R 1 − 1 R 2 − 1 … R p − 1 − 1 R p − 1 C q − 1 C q − 1 − 1 … C 2 − 1 C 1 − 1 \bold A=\bold {R_1^{-1}R_2^{-1}\dots R_{p-1}^{-1}R_p^{-1}C_q^{-1}C_{q-1}^{-1}\dots C_2^{-1}C_1^{-1} } A=R1−1R2−1…Rp−1−1Rp−1Cq−1Cq−1−1…C2−1C1−1(证毕)
Remark: 考虑到行初等矩阵与列初等矩阵的对应关系,可逆矩阵可表示为仅含行(列)初等矩阵的乘积。
结合 定理1.1、定理1.3、定理1.4 定理1.2显然可以得证。
定义1.3:若 A \bold A A经过行(列)初等变换可化为矩阵 B \bold B B,则称 A \bold A A与 B \bold B B行(列)等价,若 A \bold A A经过初等变换可化为矩阵 B \bold B B,则称 A \bold A A与 B \bold B B等价。
结合矩阵等价的定义,以及定理1.3可知:
定理1.5:
m
×
n
m\times n
m×n阶矩阵
A
\bold A
A与
B
\bold B
B等价
⟺
\Longleftrightarrow
⟺
∃
\exists
∃ 可逆方阵
R
m
\bold R_m
Rm与
C
n
\bold C_n
Cn,
s
.
t
.
R
m
A
C
n
=
B
s.t.\ \bold {R_m AC_n}=\bold B
s.t. RmACn=B
由此可知: 推论1.1:等价矩阵的秩相等。
反过来有: 定理1.6:同阶的秩等矩阵必等价。
证明:设
r
(
A
m
×
n
)
=
r
(
B
m
×
n
)
=
r
r(\bold A_{m\times n})=r(\bold B_{m\times n})=r
r(Am×n)=r(Bm×n)=r.
\quad\ \ \quad
则存在可逆矩阵
R
m
A
,
R
m
B
,
C
n
A
,
C
n
B
\bold R_m^A,\bold R_m^B,\bold C_n^A,\bold C_n^B
RmA,RmB,CnA,CnB 使得
R
m
A
A
C
n
A
=
R
m
B
B
C
n
B
=
[
I
r
0
0
0
]
\bold {R_m^AA\bold C_n^A}=\bold {R_m^BB\bold C_n^B}=\begin{bmatrix}\bold{I_r}\ 0\\ 0\ \ 0\end{bmatrix}
RmAACnA=RmBBCnB=[Ir 00 0]
\quad\ \ \quad
那么
(
R
m
B
−
1
R
m
A
)
A
(
C
n
A
C
n
B
−
1
)
=
R
A
C
=
B
\bold{({R_m^B}^{-1}R_m^A)A(\bold C_n^A{C_n^B}^{-1})}=\bold{RAC=B}
(RmB−1RmA)A(CnACnB−1)=RAC=B,即
A
,
B
\bold{A,B}
A,B等价。(证毕)
定理1.7:矩阵秩的不等式:
(1)
r
(
A
)
+
r
(
B
)
−
q
≤
r
(
A
B
)
⏟
S
y
l
v
e
s
t
e
r
公式
≤
m
i
n
{
r
(
A
)
,
r
(
B
)
}
\quad \underbrace{r(\bold A)+r(\bold B)-q\le r(\bold {AB})}_{Sylvester公式}\le min\{r(\bold A),r(\bold B)\}
Sylvester公式
r(A)+r(B)−q≤r(AB)≤min{r(A),r(B)} 其中,
q
q
q为
A
\bold A
A的列数或
B
\bold B
B的行数;
(2)
∣
r
(
A
)
−
r
(
B
)
∣
≤
r
(
A
±
B
)
≤
r
(
[
A
B
]
)
≤
r
(
A
)
+
r
(
B
)
\quad |r(\bold A)-r(\bold B)|\le r(\bold A\pm \bold B)\le r([\bold {A\ B}])\le r(\bold A)+r(\bold B)
∣r(A)−r(B)∣≤r(A±B)≤r([A B])≤r(A)+r(B).
证明:
(
1
)
(1)\quad
(1)设
r
(
A
)
=
r
A
r(\bold A)=r_A
r(A)=rA,
r
(
B
)
=
r
B
r(\bold B)=r_B
r(B)=rB.
\qquad
\ \ \ \quad\quad
则
R
A
A
C
A
=
[
I
r
A
0
0
0
]
\bold {R_AAC_A}=\begin{bmatrix}\bold {I_{r_A}} \ 0\\0 \ \ \ \ \ 0\end{bmatrix}
RAACA=[IrA 00 0],其中
R
A
\bold {R_A}
RA与
C
A
\bold {C_A}
CA可逆.
\qquad
\quad \quad\ \ \ \quad
R
B
B
C
B
=
[
I
r
B
0
0
0
]
\bold {R_BBC_B}=\begin{bmatrix}\bold {I_{r_B}} \ 0\\0 \ \ \ \ \ 0\end{bmatrix}
RBBCB=[IrB 00 0],其中
R
B
\bold {R_B}
RB与
C
B
\bold {C_B}
CB可逆.
\qquad
\ \ \ \quad\quad
那么
A
(
p
×
q
)
B
(
q
×
l
)
=
R
A
−
1
(
p
)
[
I
r
A
0
0
0
]
(
p
×
q
)
C
A
−
1
(
q
)
B
=
R
A
−
1
[
I
r
A
0
0
0
]
[
M
1
(
r
A
×
l
)
M
2
(
(
q
−
r
A
)
×
l
)
]
(
矩阵行分块
)
=
R
A
−
1
[
M
1
0
]
\bold {A_{(p\times q)}B_{(q\times l)}}=\bold {R_A^{-1}}_{(\bold p)}\begin{bmatrix}\bold {I_{r_A}} \ 0\\0 \ \ \ \ \ 0\end{bmatrix}_{(p\times q)}\bold {C_A^{-1}}_{\bold (q)}\bold B \\ \quad\\ \qquad\qquad \qquad \ \qquad\qquad\qquad=\bold {R_A^{-1}}\begin{bmatrix}\bold {I_{r_A}} \ 0\\0 \ \ \ \ \ 0\end{bmatrix} \begin{bmatrix}\bold {M_1}_{(r_A\times l)} \\ \bold {M_2}_{((q-r_A)\times l)}\end{bmatrix} (矩阵行分块)\\ \quad \\\qquad\qquad \qquad \ \qquad\qquad\qquad=\bold {R_A^{-1}}\begin{bmatrix}\bold {M_1} \\ \bold {0}\end{bmatrix}
A(p×q)B(q×l)=RA−1(p)[IrA 00 0](p×q)CA−1(q)B =RA−1[IrA 00 0][M1(rA×l)M2((q−rA)×l)](矩阵行分块) =RA−1[M10]
\qquad \ \ \ \quad\quad 则 r ( A B ) = r ( R A − 1 [ M 1 0 ] ) = r ( [ M 1 0 ] ) = r ( M 1 ) ≤ r A = r ( A ) r(\bold{AB})=r(\bold {R_A^{-1}}\begin{bmatrix}\bold {M_1} \\ \bold {0}\end{bmatrix})=r(\begin{bmatrix}\bold {M_1} \\ \bold {0}\end{bmatrix})=r(\bold {M_1})\le r_A=r(\bold A) r(AB)=r(RA−1[M10])=r([M10])=r(M1)≤rA=r(A)
\qquad \ \ \ \quad\quad 同理 A ( p × q ) B ( q × l ) = A R B − 1 ( q ) [ I r B 0 0 0 ] ( q × l ) C B − 1 ( l ) = [ N 1 ( p × r B ) N 2 ( p × ( q − r B ) ) ] [ I r B 0 0 0 ] C B − 1 ( 矩阵列分块 ) = [ N 1 0 ] C B − 1 \bold {A_{(p\times q)}B_{(q\times l)}}=\bold A\bold {R_B^{-1}}_{(\bold q)}\begin{bmatrix}\bold {I_{r_B}} \ 0\\0 \ \ \ \ \ 0\end{bmatrix}_{(q\times l)}\bold {C_B^{-1}}_{\bold (l)} \\ \quad\\ \qquad\qquad \qquad \ \qquad\qquad\qquad=\begin{bmatrix}\bold {N_1}_{(p\times r_B)}\quad \bold {N_2}_{(p\times (q-r_B))}\end{bmatrix}\begin{bmatrix}\bold {I_{r_B}} \ 0\\0 \ \ \ \ \ 0\end{bmatrix} \bold {C_B^{-1}} (矩阵列分块)\\ \quad \\\qquad\qquad \qquad \ \qquad\qquad\qquad=\begin{bmatrix}\bold {N_1} \quad \bold {0}\end{bmatrix}\bold {C_B^{-1}} A(p×q)B(q×l)=ARB−1(q)[IrB 00 0](q×l)CB−1(l) =[N1(p×rB)N2(p×(q−rB))][IrB 00 0]CB−1(矩阵列分块) =[N10]CB−1
\qquad \ \ \ \quad\quad 则 r ( A B ) = r ( [ N 1 0 ] C B − 1 ) = r ( [ N 1 0 ] ) = r ( N 1 ) ≤ r B = r ( B ) r(\bold{AB})=r(\begin{bmatrix}\bold {N_1} \quad \bold {0}\end{bmatrix}\bold {C_B^{-1}})=r(\begin{bmatrix}\bold {N_1} \quad \bold {0}\end{bmatrix})=r(\bold {N_1})\le r_B=r(\bold B) r(AB)=r([N10]CB−1)=r([N10])=r(N1)≤rB=r(B)
\qquad \ \ \ \quad\quad 综上, r ( A B ) ≤ m i n { r ( A ) , r ( B ) } r(\bold{AB})\le min\{r(\bold A),r(\bold B)\} r(AB)≤min{r(A),r(B)}
\qquad \ \ \ \quad\quad 由于, [ I q 0 − A I p ] \begin{bmatrix} \bold {I_q} \ \ 0 \\ -\bold A \ \bold {I_p} \end{bmatrix} [Iq 0−A Ip] [ I q B A 0 ] \begin{bmatrix} \bold {I_q} \ \bold B \\ \bold A \ \ 0 \end{bmatrix} [Iq BA 0] [ I q − B 0 I l ] = [ I q 0 0 − A B ] \begin{bmatrix} \bold {I_q} \ -\bold B \\ 0 \ \ \ \ \ \ \bold {I_l} \end{bmatrix}=\begin{bmatrix} \bold {I_q} \ \quad \bold 0\quad \\ \bold 0 \ -\bold {AB} \end{bmatrix} [Iq −B0 Il]=[Iq 00 −AB]
\qquad \ \ \ \quad\quad 则 r ( [ I q B A 0 ] ) = r ( [ I q 0 0 − A B ] ) = q + r ( A B ) = r ( [ B I q 0 A ] ) ≥ r ( A ) + r ( B ) r(\begin{bmatrix} \bold {I_q} \ \bold B \\ \bold A \ \ 0 \end{bmatrix})=r(\begin{bmatrix} \bold {I_q} \ \quad \bold 0\quad \\ \bold 0 \ -\bold {AB} \end{bmatrix})=q+r(\bold{AB})=r(\begin{bmatrix} \bold B \ \bold {I_q} \\ 0 \ \ \bold A \end{bmatrix})\ge r(\bold A)+r(\bold B) r([Iq BA 0])=r([Iq 00 −AB])=q+r(AB)=r([B Iq0 A])≥r(A)+r(B)
\qquad \ \ \ \quad\quad 则有 r ( A B ) ≥ r ( A ) + r ( B ) − q r(\bold{AB})\ge r(\bold A)+r(\bold B)-q r(AB)≥r(A)+r(B)−q(证毕)
Remark:(1)准对角矩阵的秩等于对角块秩之和。
\quad
证明:首先证明: r ( [ A 0 0 B ] ) = r ( [ 0 A B 0 ] ) = r ( A ) + r ( B ) r(\begin{bmatrix}\bold A \ 0 \\0 \ \ \bold B\end{bmatrix})=r(\begin{bmatrix}0 \ \ \bold A \\\bold B \ \ 0\end{bmatrix})=r(\bold A)+r(\bold B) r([A 00 B])=r([0 AB 0])=r(A)+r(B)
\ \ \qquad 对于 A m × n , B p × q \bold{A_{m\times n},B_{p\times q}} Am×n,Bp×q分别存在可逆矩阵 { R m 1 , C n 1 } , { R p 2 , C q 2 } \{\bold{R^1_m},\bold{C^1_n}\},\{\bold{R^2_p},\bold{C^2_q}\} {Rm1,Cn1},{Rp2,Cq2}使得:
\quad
\ \ \qquad R m 1 A C n 1 = [ I r ( A ) 0 0 0 ] \bold{R^1_mAC^1_n}=\begin{bmatrix}\bold{I_{r(\bold A)}}\ 0\\ 0\ \ \ \ \ \ \ 0\end{bmatrix} Rm1ACn1=[Ir(A) 00 0]; R p 2 B C q 2 = [ I r ( B ) 0 0 0 ] \bold{R^2_pBC^2_q}=\begin{bmatrix}\bold{I_{r(\bold B)}}\ 0\\ 0\ \ \ \ \ \ \ 0\end{bmatrix} Rp2BCq2=[Ir(B) 00 0]
\quad
\ \ \qquad 又 r ( [ R m 1 0 0 R p 2 ] r(\begin{bmatrix}\bold{R^1_m}\ 0\\ 0\ \ \bold{R^2_p}\end{bmatrix} r([Rm1 00 Rp2] [ A 0 0 B ] \begin{bmatrix}\bold{A}\ 0\\ 0\ \bold{B}\end{bmatrix} [A 00 B] [ C n 1 0 0 C q 2 ] ) = r ( [ A 0 0 B ] ) = \begin{bmatrix}\bold{C^1_n}\ 0\\ 0\ \ \bold{C^2_q}\end{bmatrix})=r(\begin{bmatrix}\bold{A}\ 0\\ 0\ \bold{B}\end{bmatrix})= [Cn1 00 Cq2])=r([A 00 B])= r ( [ R m 1 A C n 1 0 0 R p 2 B C q 2 ] ) r(\begin{bmatrix}\bold{R^1_mAC^1_n}\qquad\qquad 0\\ 0\qquad\qquad \bold{R^2_pBC^2_q}\end{bmatrix}) r([Rm1ACn100Rp2BCq2])
\quad
= r ( [ ( I r ( A ) 0 0 0 ) 0 0 ( I r ( B ) 0 0 0 ) ] ) = r ( A ) + r ( B ) \\ \ \qquad =r(\begin{bmatrix}\begin{pmatrix}\bold{I_{r(\bold A)}}\ 0\\ 0\ \ \ \ \ \ \ 0\end{pmatrix}\qquad\qquad 0\\ 0\qquad\qquad \begin{pmatrix}\bold{I_{r(\bold B)}}\ 0\\ 0\ \ \ \ \ \ \ 0\end{pmatrix}\end{bmatrix})=r(A)+r(B) =r(⎣ ⎡(Ir(A) 00 0)00(Ir(B) 00 0)⎦ ⎤)=r(A)+r(B)
\quad
\ \ \qquad 同理, r ( [ 0 R p 2 R m 1 0 ] r(\begin{bmatrix}0\ \bold{R^2_p}\\ \bold{R^1_m}\ \ 0\end{bmatrix} r([0 Rp2Rm1 0] [ 0 A B 0 ] \begin{bmatrix} 0\ \bold{A}\\ \bold{B}\ 0\end{bmatrix} [0 AB 0] [ C q 2 0 0 C n 1 ] ) = r ( [ 0 A B 0 ] ) = \begin{bmatrix}\bold{C^2_q}\ 0 \\ 0\ \ \bold{C^1_n}\end{bmatrix})=r(\begin{bmatrix}0\ \bold{A}\\ \bold{B}\ 0\end{bmatrix})= [Cq2 00 Cn1])=r([0 AB 0])= r ( [ R p 2 B C q 2 0 0 R m 1 A C n 1 ] ) r(\begin{bmatrix}\bold{R^2_pBC^2_q}\qquad\qquad 0\\ 0\qquad\qquad \bold{R^1_mAC^1_n}\end{bmatrix}) r([Rp2BCq200Rm1ACn1])
\quad
= r ( [ ( I r ( B ) 0 0 0 ) 0 0 ( I r ( A ) 0 0 0 ) ] ) = r ( A ) + r ( B ) \\ \ \qquad =r(\begin{bmatrix}\begin{pmatrix}\bold{I_{r(\bold B)}}\ 0\\ 0\ \ \ \ \ \ \ 0\end{pmatrix}\qquad\qquad 0\\ 0\qquad\qquad \begin{pmatrix}\bold{I_{r(\bold A)}}\ 0\\ 0\ \ \ \ \ \ \ 0\end{pmatrix}\end{bmatrix})=r(A)+r(B) =r(⎣ ⎡(Ir(B) 00 0)00(Ir(A) 00 0)⎦ ⎤)=r(A)+r(B)
\ \ \qquad 进一步可推知: r ( [ A 1 A 2 A 3 ⋱ A n − 1 A n ] ) = r ( [ [ A 1 A 2 A 3 ⋱ A n − 1 ] A n ] ) = r ( [ A 1 A 2 A 3 ⋱ A n − 1 ] ) + r ( A n ) = ⋯ = ∑ i = 1 n r ( A i ) r(\begin{bmatrix}A_1\\&A_2\\&&A_3\\&&&\ddots\\&&&&A_{n-1}\\&&&&&A_{n}\end{bmatrix})=r(\begin{bmatrix}\begin{bmatrix}A_1\\&A_2\\&&A_3\\&&&\ddots\\&&&&A_{n-1}\end{bmatrix}\\&A_n\end{bmatrix})=r(\begin{bmatrix}A_1\\&A_2\\&&A_3\\&&&\ddots\\&&&&A_{n-1}\end{bmatrix})+r(A_n)=\dots=\sum_{i=1}^nr(A_i) r(⎣ ⎡A1A2A3⋱An−1An⎦ ⎤)=r(⎣ ⎡⎣ ⎡A1A2A3⋱An−1⎦ ⎤An⎦ ⎤)=r(⎣ ⎡A1A2A3⋱An−1⎦ ⎤)+r(An)=⋯=i=1∑nr(Ai)
\quad (证毕)
\quad
(2)准三角矩阵的秩等于对角块的秩之和
\ \ \qquad
证明:首先证明: r ( [ A B 0 C ] ) = r ( [ A 0 B C ] ) = r ( [ A 0 0 C ] ) = r ( A ) + r ( C ) r(\begin{bmatrix}A&B\\0&C\end{bmatrix})=r(\begin{bmatrix}A&0\\B&C\end{bmatrix})=r(\begin{bmatrix}A&0\\0&C\end{bmatrix})=r(A)+r(C) r([A0BC])=r([AB0C])=r([A00C])=r(A)+r(C)
\ \ \qquad 对于准上(下)三角阵可通过简单的行变换变化为准对角矩阵,即: r ( [ ( ∗ ∗ … ∗ ∗ ∗ … ∗ ⋮ ⋮ ⋮ ∗ ∗ … ∗ ) n a × n a ( ∗ ∗ … ∗ ∗ ∗ … ∗ ⋮ ⋮ ⋮ ∗ ∗ … ∗ ) n a × n b 0 ( ∗ ∗ … ∗ ∗ ∗ … ∗ ⋮ ⋮ ⋮ ∗ ∗ … ∗ ) n b × n b ] ) = r ( [ ( ∗ ∗ … ∗ 0 ∗ … ∗ ⋮ ⋮ ⋮ 0 0 … ∗ ) n a × n a ( ∗ ∗ … ∗ ∗ ∗ … ∗ ⋮ ⋮ ⋮ ∗ ∗ … ∗ ) n a × n b 0 ( ∗ ∗ … ∗ 0 ∗ … ∗ ⋮ ⋮ ⋮ 0 0 … ∗ ) n b × n b ] ) = r ( [ ( ∗ ∗ … ∗ 0 ∗ … ∗ ⋮ ⋮ ⋮ 0 0 … ∗ ) n a × n a ( 0 0 … 0 0 0 … 0 ⋮ ⋮ ⋮ 0 0 … 0 ) n a × n b 0 ( ∗ 0 … 0 0 ∗ … 0 ⋮ ⋮ ⋮ 0 0 … ∗ ) n b × n b ] ) = r ( [ A ′ B ′ ] ) = r ( A ′ ) + r ( B ′ ) = r ( A ) + r ( B ) r(\begin{bmatrix}\begin{pmatrix}*&*&\dots&*\\ *&*&\dots&*\\ \vdots&\vdots&&\vdots\\ *&*&\dots&*\\\end{pmatrix}_{n_a\times n_a}& \begin{pmatrix}*&*&\dots&*\\ *&*&\dots&*\\ \vdots&\vdots&&\vdots\\ *&*&\dots&*\\\end{pmatrix}_{n_a\times n_b}\\0& \begin{pmatrix}*&*&\dots&*\\ *&*&\dots&*\\ \vdots&\vdots&&\vdots\\ *&*&\dots&*\\\end{pmatrix}_{n_b\times n_b}\end{bmatrix}) =r(\begin{bmatrix}\begin{pmatrix}*&*&\dots&*\\ 0&*&\dots&*\\ \vdots&\vdots&&\vdots\\ 0&0&\dots&*\\\end{pmatrix}_{n_a\times n_a}& \begin{pmatrix}*&*&\dots&*\\ *&*&\dots&*\\ \vdots&\vdots&&\vdots\\ *&*&\dots&*\\\end{pmatrix}_{n_a\times n_b}\\0& \begin{pmatrix}*&*&\dots&*\\ 0&*&\dots&*\\ \vdots&\vdots&&\vdots\\ 0&0&\dots&*\\\end{pmatrix}_{n_b\times n_b}\end{bmatrix}) =r(\begin{bmatrix}\begin{pmatrix}*&*&\dots&*\\ 0&*&\dots&*\\ \vdots&\vdots&&\vdots\\ 0&0&\dots&*\\\end{pmatrix}_{n_a\times n_a}& \begin{pmatrix}0&0&\dots&0\\ 0&0&\dots&0\\ \vdots&\vdots&&\vdots\\ 0&0&\dots&0\\\end{pmatrix}_{n_a\times n_b}\\0& \begin{pmatrix}*&0&\dots&0\\ 0&*&\dots&0\\ \vdots&\vdots&&\vdots\\ 0&0&\dots&*\\\end{pmatrix}_{n_b\times n_b}\end{bmatrix})=r(\begin{bmatrix}A'\\&B'\end{bmatrix})=r(A')+r(B')=r(A)+r(B) r(⎣ ⎡⎝ ⎛∗∗⋮∗∗∗⋮∗………∗∗⋮∗⎠ ⎞na×na0⎝ ⎛∗∗⋮∗∗∗⋮∗………∗∗⋮∗⎠ ⎞na×nb⎝ ⎛∗∗⋮∗∗∗⋮∗………∗∗⋮∗⎠ ⎞nb×nb⎦ ⎤)=r(⎣ ⎡⎝ ⎛∗0⋮0∗∗⋮0………∗∗⋮∗⎠ ⎞na×na0⎝ ⎛∗∗⋮∗∗∗⋮∗………∗∗⋮∗⎠ ⎞na×nb⎝ ⎛∗0⋮0∗∗⋮0………∗∗⋮∗⎠ ⎞nb×nb⎦ ⎤)=r(⎣ ⎡⎝ ⎛∗0⋮0∗∗⋮0………∗∗⋮∗⎠ ⎞na×na0⎝ ⎛00⋮000⋮0………00⋮0⎠ ⎞na×nb⎝ ⎛∗0⋮00∗⋮0………00⋮∗⎠ ⎞nb×nb⎦ ⎤)=r([A′B′])=r(A′)+r(B′)=r(A)+r(B)
\ \ \qquad 准下三角形同理,不多加赘述。
\ \ \qquad 进一步推知: r ( [ A 11 A 12 A 13 … A 1 , n − 1 A n A 22 A 23 … A 2 , n − 1 A 2 , n A 33 … A 3 , n − 1 A 3 , n ⋱ ⋮ ⋮ A n − 1 , n − 1 A n − 1 , n A n , n ] ) = r ( [ ( A 11 A 12 A 13 … A 1 , n − 1 A 22 A 23 … A 2 , n − 1 A 33 … A 3 , n − 1 ⋱ ⋮ A n − 1 , n − 1 ) ( A n A 2 , n A 3 , n ⋮ A n − 1 , n ) 0 A n n ] ) = r ( [ A 11 A 12 A 13 … A 1 , n − 1 A 22 A 23 … A 2 , n − 1 A 33 … A 3 , n − 1 ⋱ ⋮ A n − 1 , n − 1 ] ) + r ( A n n ) = ⋯ = ∑ i = 1 n r ( A i i ) r(\begin{bmatrix}A_{11}&A_{12}&A_{13}&\dots&A_{1,n-1}&A_{n}\\&A_{22}&A_{23}&\dots&A_{2,n-1}&A_{2,n}\\&&A_{33}&\dots&A_{3,n-1}&A_{3,n}\\&&&\ddots&\vdots&\vdots\\&&&&A_{n-1,n-1}&A_{n-1,n}\\&&&&&A_{n,n}\end{bmatrix})=r(\begin{bmatrix}\begin{pmatrix}A_{11}&A_{12}&A_{13}&\dots&A_{1,n-1}\\&A_{22}&A_{23}&\dots&A_{2,n-1}\\&&A_{33}&\dots&A_{3,n-1}\\&&&\ddots&\vdots\\&&&&A_{n-1,n-1}\end{pmatrix}&\begin{pmatrix}A_{n}\\A_{2,n}\\A_{3,n}\\\vdots\\A_{n-1,n}\end{pmatrix}\\0&A_{nn}\end{bmatrix})=r(\begin{bmatrix}A_{11}&A_{12}&A_{13}&\dots&A_{1,n-1}\\&A_{22}&A_{23}&\dots&A_{2,n-1}\\&&A_{33}&\dots&A_{3,n-1}\\&&&\ddots&\vdots\\&&&&A_{n-1,n-1}\end{bmatrix})+r(A_{nn})=\dots=\sum_{i=1}^nr(A_{ii}) r(⎣ ⎡A11A12A22A13A23A33………⋱A1,n−1A2,n−1A3,n−1⋮An−1,n−1AnA2,nA3,n⋮An−1,nAn,n⎦ ⎤)=r(⎣ ⎡⎝ ⎛A11A12A22A13A23A33………⋱A1,n−1A2,n−1A3,n−1⋮An−1,n−1⎠ ⎞0⎝ ⎛AnA2,nA3,n⋮An−1,n⎠ ⎞Ann⎦ ⎤)=r(⎣ ⎡A11A12A22A13A23A33………⋱A1,n−1A2,n−1A3,n−1⋮An−1,n−1⎦ ⎤)+r(Ann)=⋯=i=1∑nr(Aii)
\ \ \qquad (证毕)
(
2
)
\quad\quad\ (2)\quad
(2)
A
m
×
n
±
B
m
×
n
=
[
A
⋮
B
]
m
×
2
n
[
I
n
±
I
n
]
\bold {A_{m\times n}\pm B_{m\times n}=[A\vdots B]_{m\times 2n}\begin{bmatrix}\bold{I_n}\\ \bold{\pm I_n}\end{bmatrix}}
Am×n±Bm×n=[A⋮B]m×2n[In±In]
\quad\quad\ \ \quad \quad
则
r
(
A
±
B
)
≤
r
(
A
⋮
B
)
≤
r
(
[
A
B
A
0
]
)
o
r
r
(
[
A
B
0
B
]
)
r(\bold {A\pm B})\le r(\bold{A\vdots B})\le r(\begin{bmatrix}\bold{A} \ \bold{B} \\ \bold{A}\ \ 0\end{bmatrix})\ or\ r(\begin{bmatrix}\bold{A} \ \bold{B} \\ 0 \ \ \bold{B}\end{bmatrix})
r(A±B)≤r(A⋮B)≤r([A BA 0]) or r([A B0 B])
= r ( [ 0 B A 0 ] ) o r r ( [ A 0 0 B ] ) \quad\quad\ \ \qquad \quad=r(\begin{bmatrix}0 \ \bold{B} \\ \bold{A}\ 0\end{bmatrix})\ or\ r(\begin{bmatrix}\bold{A} \ 0 \\ 0 \ \bold{B}\end{bmatrix}) =r([0 BA 0]) or r([A 00 B]) (初等变换不改变矩阵的秩)
= r ( A ) + r ( B ) \quad\quad\ \ \qquad \quad=r(\bold A)+r(\bold B) =r(A)+r(B)
\quad\quad\ \ \quad \quad r ( A ) = r ( ( A ± B ) ∓ B ) ≤ r ( A ± B ) + r ( ∓ B ) = r ( A ± B ) + r ( B ) r(\bold A)=r(\bold{(A\pm B)\mp B})\le r(\bold{A\pm B})+r(\bold{\mp B})=r(\bold{A\pm B})+r(\bold{B}) r(A)=r((A±B)∓B)≤r(A±B)+r(∓B)=r(A±B)+r(B)
\quad\quad\ \ \quad \quad r ( B ) = r ( ( A ± B ) ∓ A ) ≤ r ( A ± B ) + r ( ∓ A ) = r ( A ± B ) + r ( A ) r(\bold B)=r(\bold{(A\pm B)\mp A})\le r(\bold{A\pm B})+r(\bold{\mp A})=r(\bold{A\pm B})+r(\bold{A}) r(B)=r((A±B)∓A)≤r(A±B)+r(∓A)=r(A±B)+r(A)
\quad\quad\ \ \quad \quad 则, ∣ r ( A ) − r ( B ) ∣ ≤ r ( A ± B ) |r(\bold A)-r(\bold B)|\le r(\bold{A\pm B}) ∣r(A)−r(B)∣≤r(A±B)(证毕)