引理
X
∈
R
m
×
n
,
Y
∈
R
n
×
m
\mathbf{X}\in\mathbb{R}^{m\times n},\mathbf{Y}\in\mathbb{R}^{n\times m}
X∈Rm×n,Y∈Rn×m,则
X
Y
,
Y
X
\mathbf{XY},\mathbf{YX}
XY,YX具有相同的非零特征值（重数也一样）
证明：
(
I
m
X
0
I
n
)
−
1
(
X
Y
0
Y
0
)
(
I
m
X
0
I
n
)
=
(
0
0
Y
Y
X
)
\begin{pmatrix} \mathbf{I}_m &\mathbf{X}\\ 0&\mathbf{I}_n \end{pmatrix}^{-1}\begin{pmatrix} \mathbf{XY}&0\\ \mathbf{Y}&0 \end{pmatrix}\begin{pmatrix} \mathbf{I}_m &\mathbf{X}\\ 0&\mathbf{I}_n \end{pmatrix}=\begin{pmatrix} 0&0\\ \mathbf{Y}&\mathbf{YX} \end{pmatrix}
(Im0XIn)−1(XYY00)(Im0XIn)=(0Y0YX)
因为相似，所以具有相同的非零特征值（重数也一样）
Sherman–Morrison formula
设
A
∈
R
n
×
n
\mathbf{A}\in\mathbb{R}^{n\times n}
A∈Rn×n非奇异，
u
,
v
∈
R
n
\mathbf{u},\mathbf{v}\in\mathbb{R}^n
u,v∈Rn,则
A
+
u
v
T
\mathbf{A}+\mathbf{u}\mathbf{v}^T
A+uvT可逆当且仅当
1
+
v
T
A
−
1
u
≠
0
1+\mathbf{v}^T\mathbf{A}^{-1}\mathbf{u}\neq 0
1+vTA−1u=0,且
(
A
+
u
v
T
)
−
1
=
A
−
1
−
A
−
1
u
v
T
A
−
1
1
+
v
T
A
−
1
u
\left(\mathbf{A}+\mathbf{u}\mathbf{v}^{T}\right)^{-1}=\mathbf{A}^{-1}-\frac{\mathbf{A}^{-1}\mathbf{u}\mathbf{v}^T\mathbf{A}^{-1}}{1+\mathbf{v}^{T}\mathbf{A}^{-1}\mathbf{u}}
(A+uvT)−1=A−1−1+vTA−1uA−1uvTA−1
证明：
假设
1
+
v
T
A
−
1
u
≠
0
1+\mathbf{v}^T\mathbf{A}^{-1}\mathbf{u}\neq 0
1+vTA−1u=0
(
A
+
u
v
T
)
(
A
−
1
−
A
−
1
u
v
T
A
−
1
1
+
v
T
A
−
1
u
)
=
I
+
u
v
T
A
−
1
−
u
v
T
A
−
1
+
u
v
T
A
−
1
u
v
T
A
−
1
1
+
v
T
A
−
1
u
=
I
+
u
v
T
A
−
1
−
u
v
T
(
1
+
v
T
A
−
1
u
)
A
−
1
1
+
v
T
A
−
1
u
=
I
+
u
v
T
A
−
1
−
u
v
T
A
−
1
=
I
\begin{aligned} &\quad\left(\mathbf{A}+\mathbf{u}\mathbf{v}^{T}\right)\left(\mathbf{A}^{-1}-\frac{\mathbf{A}^{-1}\mathbf{u}\mathbf{v}^T\mathbf{A}^{-1}}{1+\mathbf{v}^{T}\mathbf{A}^{-1}\mathbf{u}}\right)\\ &=\mathbf{I}+\mathbf{u}\mathbf{v}^T\mathbf{A}^{-1}-\frac{\mathbf{u}\mathbf{v}^T\mathbf{A}^{-1}+\mathbf{u}\mathbf{v}^T\mathbf{A}^{-1}\mathbf{u}\mathbf{v}^T\mathbf{A}^{-1}}{1+\mathbf{v}^{T}\mathbf{A}^{-1}\mathbf{u}}\\ &=\mathbf{I}+\mathbf{u}\mathbf{v}^T\mathbf{A}^{-1}-\frac{\mathbf{u}\mathbf{v}^T\left(1+\mathbf{v}^{T}\mathbf{A}^{-1}\mathbf{u}\right)\mathbf{A}^{-1}}{1+\mathbf{v}^{T}\mathbf{A}^{-1}\mathbf{u}}\\ &=\mathbf{I}+\mathbf{u}\mathbf{v}^T\mathbf{A}^{-1}-\mathbf{u}\mathbf{v}^T\mathbf{A}^{-1}\\ &=\mathbf{I} \end{aligned}
(A+uvT)(A−1−1+vTA−1uA−1uvTA−1)=I+uvTA−1−1+vTA−1uuvTA−1+uvTA−1uvTA−1=I+uvTA−1−1+vTA−1uuvT(1+vTA−1u)A−1=I+uvTA−1−uvTA−1=I
假设
A
+
u
v
T
\mathbf{A}+\mathbf{u}\mathbf{v}^T
A+uvT可逆
根据矩阵行列式引理，
∣
A
+
u
v
T
∣
=
(
1
+
v
T
A
−
1
u
)
∣
A
∣
\left|\mathbf{A}+\mathbf{u}\mathbf{v}^T\right|=\left(1+\mathbf{v}^T\mathbf{A}^{-1}\mathbf{u}\right)\left|\mathbf{A}\right|
∣∣A+uvT∣∣=(1+vTA−1u)∣A∣
于是
1
+
v
T
A
−
1
u
≠
0
1+\mathbf{v}^T\mathbf{A}^{-1}\mathbf{u}\neq 0
1+vTA−1u=0
Sherman–Morrison-Woodbury formula
设
A
∈
R
n
×
n
\mathbf{A}\in\mathbb{R}^{n\times n}
A∈Rn×n非奇异，
U
,
V
∈
R
n
×
k
\mathbf{U},\mathbf{V}\in\mathbb{R}^{n\times k}
U,V∈Rn×k,如果
A
+
U
V
T
\mathbf{A}+\mathbf{U}\mathbf{V}^T
A+UVT非奇异，则
(
A
+
U
V
T
)
−
1
=
A
−
1
−
A
−
1
U
(
I
k
+
V
T
A
−
1
U
)
−
1
V
T
A
−
1
\left(\mathbf{A}+\mathbf{U}\mathbf{V}^{T}\right)^{-1}=\mathbf{A}^{-1}-\mathbf{A}^{-1}\mathbf{U}\left(\mathbf{I}_k+\mathbf{V}^T\mathbf{A}^{-1}\mathbf{U}\right)^{-1}\mathbf{V}^T\mathbf{A}^{-1}
(A+UVT)−1=A−1−A−1U(Ik+VTA−1U)−1VTA−1
(
n
=
1
n=1
n=1时就是Sherman–Morrison formula)
证明：
因为
A
+
U
V
T
\mathbf{A}+\mathbf{U}\mathbf{V}^T
A+UVT非奇异，
A
+
U
V
T
=
A
(
I
n
+
A
−
1
U
V
T
)
\mathbf{A}+\mathbf{U}\mathbf{V}^T=\mathbf{A}\left(\mathbf{I}_n+\mathbf{A}^{-1}\mathbf{U}\mathbf{V}^T\right)
A+UVT=A(In+A−1UVT)
所以
I
n
+
A
−
1
U
V
T
\mathbf{I}_n+\mathbf{A}^{-1}\mathbf{U}\mathbf{V}^T
In+A−1UVT
根据引理，
(
A
−
1
U
)
V
T
\left(\mathbf{A}^{-1}\mathbf{U}\right)\mathbf{V}^T
(A−1U)VT和
V
T
(
A
−
1
U
)
\mathbf{V}^{T}\left(\mathbf{A}^{-1}\mathbf{U}\right)
VT(A−1U)具有相同的非零特征值
所以
I
k
+
V
T
A
−
1
U
\mathbf{I}_k+\mathbf{V}^T\mathbf{A}^{-1}\mathbf{U}
Ik+VTA−1U可逆
( A + U V T ) ( A − 1 − A − 1 U ( I k + V T A − 1 U ) − 1 V T A − 1 ) = I + U V T A − 1 − ( U + U V T A − 1 U ) ( I k + V T A − 1 U ) − 1 V T A − 1 = I + U V T A − 1 − U ( I + V T A − 1 U ) ( I k + V T A − 1 U ) − 1 V T A − 1 = I + U V T A − 1 − U V T A − 1 = I \begin{aligned} &\quad \left(\mathbf{A}+\mathbf{U}\mathbf{V}^{T}\right)\left(\mathbf{A}^{-1}-\mathbf{A}^{-1}\mathbf{U}\left(\mathbf{I}_k+\mathbf{V}^T\mathbf{A}^{-1}\mathbf{U}\right)^{-1}\mathbf{V}^T\mathbf{A}^{-1}\right)\\ &=\mathbf{I}+\mathbf{U}\mathbf{V}^T\mathbf{A}^{-1}-\left(\mathbf{U}+\mathbf{U}\mathbf{V}^T\mathbf{A}^{-1}\mathbf{U}\right)\left(\mathbf{I}_k+\mathbf{V}^T\mathbf{A}^{-1}\mathbf{U}\right)^{-1}\mathbf{V}^T\mathbf{A}^{-1}\\ &=\mathbf{I}+\mathbf{U}\mathbf{V}^T\mathbf{A}^{-1}-\mathbf{U}\left(\mathbf{I}+\mathbf{V}^T\mathbf{A}^{-1}\mathbf{U}\right)\left(\mathbf{I}_k+\mathbf{V}^T\mathbf{A}^{-1}\mathbf{U}\right)^{-1}\mathbf{V}^T\mathbf{A}^{-1}\\ &=\mathbf{I}+\mathbf{U}\mathbf{V}^T\mathbf{A}^{-1}-\mathbf{U}\mathbf{V}^T\mathbf{A}^{-1}\\ &=\mathbf{I} \end{aligned} (A+UVT)(A−1−A−1U(Ik+VTA−1U)−1VTA−1)=I+UVTA−1−(U+UVTA−1U)(Ik+VTA−1U)−1VTA−1=I+UVTA−1−U(I+VTA−1U)(Ik+VTA−1U)−1VTA−1=I+UVTA−1−UVTA−1=I
Woodbury matrix identity
设
A
∈
R
n
×
n
\mathbf{A}\in\mathbb{R}^{n\times n}
A∈Rn×n非奇异,
C
∈
R
k
×
k
,
U
∈
R
n
×
k
,
V
∈
R
k
×
n
\mathbf{C}\in\mathbb{R}^{k\times k},\mathbf{U}\in\mathbb{R}^{n\times k},\mathbf{V}\in\mathbb{R}^{k\times n}
C∈Rk×k,U∈Rn×k,V∈Rk×n
如果
C
,
A
+
U
C
V
\mathbf{C},\mathbf{A}+\mathbf{U}\mathbf{C}\mathbf{V}
C,A+UCV可逆，则
(
A
+
U
C
V
)
−
1
=
A
−
1
−
A
−
1
U
(
C
−
1
+
V
A
−
1
U
)
V
A
−
1
\left(\mathbf{A}+\mathbf{U}\mathbf{C}\mathbf{V}\right)^{-1}=\mathbf{A}^{-1}-\mathbf{A}^{-1}\mathbf{U}\left(\mathbf{C}^{-1}+\mathbf{V}\mathbf{A}^{-1}\mathbf{U}\right)\mathbf{V}\mathbf{A}^{-1}
(A+UCV)−1=A−1−A−1U(C−1+VA−1U)VA−1
证明：
根据矩阵行列式引理
∣
A
+
U
C
V
∣
=
∣
C
−
1
+
V
A
−
1
U
∣
∣
C
∣
∣
A
∣
\left|\mathbf{A}+\mathbf{U}\mathbf{C}\mathbf{V}\right|=\left|\mathbf{C}^{-1}+\mathbf{V}\mathbf{A}^{-1}\mathbf{U}\right|\left|\mathbf{C}\right|\left|\mathbf{A}\right|
∣A+UCV∣=∣∣C−1+VA−1U∣∣∣C∣∣A∣
所以
C
−
1
+
V
A
−
1
U
\mathbf{C}^{-1}+\mathbf{V}\mathbf{A}^{-1}\mathbf{U}
C−1+VA−1U可逆
( A + U C V ) [ A − 1 − A − 1 U ( C − 1 + V A − 1 U ) − 1 V A − 1 ] = { I − U ( C − 1 + V A − 1 U ) − 1 V A − 1 } + { U C V A − 1 − U C V A − 1 U ( C − 1 + V A − 1 U ) − 1 V A − 1 } = { I + U C V A − 1 } − { U ( C − 1 + V A − 1 U ) − 1 V A − 1 + U C V A − 1 U ( C − 1 + V A − 1 U ) − 1 V A − 1 } = I + U C V A − 1 − ( U + U C V A − 1 U ) ( C − 1 + V A − 1 U ) − 1 V A − 1 = I + U C V A − 1 − U C ( C − 1 + V A − 1 U ) ( C − 1 + V A − 1 U ) − 1 V A − 1 = I + U C V A − 1 − U C V A − 1 = I \begin{aligned} &(\mathbf{A}+\mathbf{U} \mathbf{C} \mathbf{V})\left[\mathbf{A}^{-1}-\mathbf{A}^{-1} \mathbf{U}\left(\mathbf{C}^{-1}+\mathbf{V} \mathbf{A}^{-1} \mathbf{U}\right)^{-1} \mathbf{V} \mathbf{A}^{-1}\right] \\ =&\left\{\mathbf{I}-\mathbf{U}\left(\mathbf{C}^{-1}+\mathbf{V} \mathbf{A}^{-1} \mathbf{U}\right)^{-1} \mathbf{V} \mathbf{A}^{-1}\right\}+\left\{\mathbf{U} \mathbf{C} \mathbf{V} \mathbf{A}^{-1}-\mathbf{U} \mathbf{C} \mathbf{V} \mathbf{A}^{-1} \mathbf{U}\left(\mathbf{C}^{-1}+\mathbf{V} \mathbf{A}^{-1} \mathbf{U}\right)^{-1} \mathbf{V} \mathbf{A}^{-1}\right\} \\ =&\left\{\mathbf{I}+\mathbf{U} \mathbf{C} \mathbf{V} \mathbf{A}^{-1}\right\}-\left\{\mathbf{U}\left(\mathbf{C}^{-1}+\mathbf{V} \mathbf{A}^{-1} \mathbf{U}\right)^{-1} \mathbf{V} \mathbf{A}^{-1}+\mathbf{U} \mathbf{C} \mathbf{V} \mathbf{A}^{-1} \mathbf{U}\left(\mathbf{C}^{-1}+\mathbf{V} \mathbf{A}^{-1} \mathbf{U}\right)^{-1} \mathbf{V} \mathbf{A}^{-1}\right\} \\ =& \mathbf{I}+\mathbf{U} \mathbf{C} \mathbf{V} \mathbf{A}^{-1}-\left(\mathbf{U}+\mathbf{U} \mathbf{C} \mathbf{V} \mathbf{A}^{-1} \mathbf{U}\right)\left(\mathbf{C}^{-1}+\mathbf{V} \mathbf{A}^{-1} \mathbf{U}\right)^{-1} \mathbf{V} \mathbf{A}^{-1} \\ =& \mathbf{I}+\mathbf{U} \mathbf{C} \mathbf{V} \mathbf{A}^{-1}-\mathbf{U} \mathbf{C}\left(\mathbf{C}^{-1}+\mathbf{V} \mathbf{A}^{-1} \mathbf{U}\right)\left(\mathbf{C}^{-1}+\mathbf{V} \mathbf{A}^{-1} \mathbf{U}\right)^{-1} \mathbf{V} \mathbf{A}^{-1} \\ =& \mathbf{I}+\mathbf{U} \mathbf{C} \mathbf{V} \mathbf{A}^{-1}-\mathbf{U} \mathbf{C} \mathbf{V} \mathbf{A}^{-1} \\ =& \mathbf{I} \end{aligned} ======(A+UCV)[A−1−A−1U(C−1+VA−1U)−1VA−1]{I−U(C−1+VA−1U)−1VA−1}+{UCVA−1−UCVA−1U(C−1+VA−1U)−1VA−1}{I+UCVA−1}−{U(C−1+VA−1U)−1VA−1+UCVA−1U(C−1+VA−1U)−1VA−1}I+UCVA−1−(U+UCVA−1U)(C−1+VA−1U)−1VA−1I+UCVA−1−UC(C−1+VA−1U)(C−1+VA−1U)−1VA−1I+UCVA−1−UCVA−1I
3万+
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
