关于矩阵导数的几个性质的证明
其他
2019-02-23 20:17:18
阅读次数: 0
关于矩阵导数的几个性质的证明
(1)
∇
A
t
r
(
A
B
)
=
B
T
\nabla_Atr(AB)=B^T
∇ A t r ( A B ) = B T
(2)
∇
A
t
r
(
A
B
A
T
C
)
=
C
A
B
+
C
T
A
B
T
\nabla_Atr(ABA^TC)=CAB+C^TAB^T
∇ A t r ( A B A T C ) = C A B + C T A B T
(3)
∇
A
∣
A
∣
=
∣
A
∣
(
A
−
1
)
T
\nabla_A|A|=|A|(A^{-1})^T
∇ A ∣ A ∣ = ∣ A ∣ ( A − 1 ) T Proof of (1) Let
A
=
[
a
1
,
⋯
 
,
a
n
]
,
B
=
[
b
1
′
⋮
b
n
′
]
A=[a_1,\cdots,a_n],B=\begin{bmatrix}b_1'\\\vdots\\b_n'\end{bmatrix}
A = [ a 1 , ⋯ , a n ] , B = ⎣ ⎢ ⎡ b 1 ′ ⋮ b n ′ ⎦ ⎥ ⎤ Then
∇
A
t
r
(
A
B
)
=
∇
A
t
r
(
∑
i
=
1
n
a
i
b
i
′
)
=
∑
i
=
1
n
∇
A
t
r
(
a
i
b
i
′
)
\nabla_Atr(AB)=\nabla_Atr(\sum_{i=1}^{n}a_ib_i')=\sum_{i=1}^{n}\nabla_Atr(a_ib_i')
∇ A t r ( A B ) = ∇ A t r ( ∑ i = 1 n a i b i ′ ) = ∑ i = 1 n ∇ A t r ( a i b i ′ )
∇
A
t
r
(
a
i
b
i
′
)
=
∇
A
t
r
(
[
a
1
i
⋮
a
m
i
]
[
b
i
1
,
⋯
 
,
b
i
m
]
)
=
∇
A
∑
j
=
1
m
a
j
i
b
i
j
=
[
0
,
⋯
 
,
b
i
,
⋯
 
,
0
]
\nabla_Atr(a_ib_i')=\nabla_Atr(\begin{bmatrix}a_{1i}\\\vdots\\a_{mi}\end{bmatrix}[b_{i1},\cdots,b_{im}])=\nabla_A\sum_{j=1}^ma_{ji}b_{ij}=[0,\cdots,b_i,\cdots,0]
∇ A t r ( a i b i ′ ) = ∇ A t r ( ⎣ ⎢ ⎡ a 1 i ⋮ a m i ⎦ ⎥ ⎤ [ b i 1 , ⋯ , b i m ] ) = ∇ A ∑ j = 1 m a j i b i j = [ 0 , ⋯ , b i , ⋯ , 0 ] so
∇
A
t
r
(
A
B
)
=
∑
i
=
1
n
∇
A
t
r
(
a
i
b
i
′
)
=
[
b
1
,
⋯
 
,
b
n
]
=
B
T
\nabla_Atr(AB)=\sum_{i=1}^n\nabla_Atr(a_ib_i')=[b_1,\cdots,b_n]=B^T
∇ A t r ( A B ) = ∑ i = 1 n ∇ A t r ( a i b i ′ ) = [ b 1 , ⋯ , b n ] = B T Proof of 2 这是在知乎上看到的证明https://www.zhihu.com/question/56412790 Proof of (3)
∣
A
∣
=
∑
j
A
i
j
A
i
j
′
|A|=\sum_jA_{ij}A_{ij}'
∣ A ∣ = ∑ j A i j A i j ′ ,
A
i
j
′
A_{ij}'
A i j ′ 是A的ij位置对应的代数余子式 于是
∂
∣
A
∣
∂
A
i
j
=
A
i
j
′
\frac{\partial |A|}{\partial A_{ij}}=A_{ij}'
∂ A i j ∂ ∣ A ∣ = A i j ′ 所以
∇
A
∣
A
∣
=
A
′
=
(
A
∗
)
T
=
(
∣
A
∣
A
−
1
)
T
=
∣
A
∣
(
A
−
1
)
T
\nabla_A|A|=A'=(A^*)^T=(|A|A^{-1})^T=|A|(A^{-1})^T
∇ A ∣ A ∣ = A ′ = ( A ∗ ) T = ( ∣ A ∣ A − 1 ) T = ∣ A ∣ ( A − 1 ) T 第三个等号是因为
A
A
∗
=
∣
A
∣
E
n
AA^*=|A|E_n
A A ∗ = ∣ A ∣ E n
转载自 blog.csdn.net/a1109885671/article/details/81226176