矩阵分析
矩阵序列
矩阵序列收敛==矩阵分量序列收敛
设有
C
m
×
n
C^{m \times n}
Cm×n中的矩阵序列
{
A
(
k
)
}
\{\pmb A^{(k)}\}
{A(k)},其中
A
(
k
)
=
(
a
i
j
(
k
)
)
m
×
n
\pmb A^{(k)} = (a_{ij}^{(k)})_{m \times n}
A(k)=(aij(k))m×n,若
lim
k
→
+
∞
a
i
j
(
k
)
=
a
i
j
\lim_{k \rightarrow +\infty} a_{ij}^{(k)} = a_{ij}
limk→+∞aij(k)=aij,则称矩阵序列
{
A
(
k
)
}
\{\pmb A^{(k)}\}
{A(k)}收敛于
A
=
(
a
i
j
)
m
×
n
\pmb A = (a_{ij})_{m \times n}
A=(aij)m×n,或称
A
\pmb A
A为矩阵序列
{
A
(
k
)
}
\{\pmb A^{(k)}\}
{A(k)}的极限,记为
lim
k
→
+
∞
A
(
k
)
=
A
o
r
A
(
k
)
→
A
\lim_{k \rightarrow +\infty} \pmb A^{(k)} = \pmb A \ \ \ or \ \ \ \pmb A^{(k)} \rightarrow A
k→+∞limA(k)=A or A(k)→A
不收敛的矩阵序列称为发散
矩阵序列收敛==矩阵范数序列收敛
设 A ( k ) , A ∈ C m × n \pmb A^{(k)},\pmb A \in \pmb C^{m \times n} A(k),A∈Cm×n,则 lim k → + ∞ A ( k ) = A \lim_{k \rightarrow +\infty} \pmb A^{(k)} = \pmb A limk→+∞A(k)=A的充分必要条件是 lim k → + ∞ ∥ A ( k ) − A ∥ = 0 \lim_{k \rightarrow + \infty} \parallel \pmb A^{(k)} - \pmb A \parallel =0 limk→+∞∥A(k)−A∥=0,其中 ∥ ⋅ ∥ \parallel \cdot \parallel ∥⋅∥是 C m × n C^{m \times n} Cm×n上的任一矩阵范数
- lim k → + ∞ ( α A ( k ) + β B ( k ) ) = α A + β B \lim_{k \rightarrow + \infty} (\alpha \pmb A^{(k)} + \beta \pmb B^{(k)}) = \alpha \pmb A + \beta \pmb B limk→+∞(αA(k)+βB(k))=αA+βB
- lim k → + ∞ A ( k ) B ( k ) = A B \lim_{k \rightarrow +\infty} \pmb A^{(k)} \pmb B^{(k)} = \pmb A \pmb B limk→+∞A(k)B(k)=AB
- 当 A ( k ) , A \pmb A^{(k)},\pmb A A(k),A均可逆时, lim k → + ∞ ( A ( k ) ) − 1 = A − 1 \lim_{k \rightarrow + \infty} (\pmb A^{(k)})^{-1} = \pmb A^{-1} limk→+∞(A(k))−1=A−1
收敛矩阵
- 设 A ∈ C m × n \pmb A \in \pmb C^{m \times n} A∈Cm×n,若 lim k → + ∞ A k = O \lim_{k \rightarrow + \infty} \pmb A^k = \pmb O limk→+∞Ak=O,则 A \pmb A A为收敛矩阵
- 设 A ∈ C m × n \pmb A \in \pmb C^{m \times n} A∈Cm×n,则 A \pmb A A为收敛矩阵的充分必要条件是 ρ ( A ) < 1 \rho(\pmb A) <1 ρ(A)<1
- 设 A ∈ C m × n \pmb A \in \pmb C^{m \times n} A∈Cm×n,若对 C m × n \pmb C^{m \times n} Cm×n上的某一矩阵范数 ∥ ⋅ ∥ \parallel \cdot \parallel ∥⋅∥有 ∥ A ∥ < 1 \parallel \pmb A \parallel <1 ∥A∥<1,则 A \pmb A A为收敛矩阵
矩阵级数
由
C
m
×
n
\pmb C^{m \times n}
Cm×n中的矩阵序列
{
A
(
k
)
}
\{\pmb A^{(k)}\}
{A(k)}构成的无穷和
A
(
0
)
+
A
(
1
)
+
⋯
+
A
(
k
)
+
⋯
\pmb A^{(0)} + \pmb A^{(1)} + \cdots + \pmb A^{(k)} + \cdots
A(0)+A(1)+⋯+A(k)+⋯
称为矩阵级数,记为
∑
k
=
0
+
∞
A
(
k
)
\sum_{k=0}^{+\infty}\pmb A^{(k)}
∑k=0+∞A(k)。对任一正整数
N
N
N,称
S
(
N
)
=
∑
k
=
0
N
A
(
k
)
\pmb S^{(N)} = \sum_{k=0}^N \pmb A^{(k)}
S(N)=∑k=0NA(k)为矩阵级数的部分和。如果由部分和构成的矩阵序列
{
S
(
N
)
}
\{\pmb S^{(N)}\}
{S(N)}收敛,且有极限
S
\pmb S
S,即
lim
N
→
+
∞
S
(
N
)
=
S
\lim_{N \rightarrow + \infty} \pmb S^{(N)} = \pmb S
limN→+∞S(N)=S,则称矩阵级数
∑
k
=
0
+
∞
A
(
k
)
\sum_{k=0}^{+\infty}\pmb A^{(k)}
∑k=0+∞A(k)收敛,其和为
S
\pmb S
S
S
=
∑
k
=
0
+
∞
A
(
k
)
\pmb S = \sum_{k=0}^{+\infty} \pmb A^{(k)}
S=k=0∑+∞A(k)
不收敛的矩阵级数称为发散的。
设
A
(
k
)
=
(
a
i
j
(
k
)
)
m
×
n
∈
C
m
×
n
\pmb A^{(k)} = (a_{ij}^{(k)})_{m \times n} \in \pmb C^{ m \times n}
A(k)=(aij(k))m×n∈Cm×n。如果
m
n
mn
mn个项数级数
∑
k
=
0
+
∞
a
i
j
(
k
)
\sum_{k=0}^{+\infty}a_{ij}^{(k)}
k=0∑+∞aij(k)
都绝对收敛,即
∑
k
=
0
+
∞
∣
a
i
j
(
k
)
∣
\sum_{k=0}^{+\infty} \mid a_{ij}^{(k)} \mid
∑k=0+∞∣aij(k)∣都收敛,则称矩阵级数
∑
k
=
0
+
∞
A
(
k
)
\sum_{k=0}^{+\infty} \pmb A^{(k)}
∑k=0+∞A(k)绝对收敛。
-
设 A ( k ) = ( a i j ( k ) ) m × n ∈ C m × n \pmb A^{(k)} = (a_{ij}^{(k)})_{m \times n} \in \pmb C^{ m \times n} A(k)=(aij(k))m×n∈Cm×n,则矩阵级数 ∑ k = 0 + ∞ A ( k ) \sum_{k=0}^{+\infty}\pmb A^{(k)} ∑k=0+∞A(k)绝对收敛的充分必要条件是正项级数 ∑ k = 0 + ∞ ∥ A ( k ) ∥ \sum_{k=0}^{+\infty} \parallel \pmb A^{(k)} \parallel ∑k=0+∞∥A(k)∥收敛,其中 ∥ ⋅ ∥ \parallel \cdot \parallel ∥⋅∥是 C m × n \pmb C^{m \times n} Cm×n上任一矩阵范数
-
有 ∑ k = 0 + ∞ A ( k ) , ∑ k = 0 + ∞ B ( k ) \sum_{k=0}^{+\infty}\pmb A^{(k)},\sum_{k=0}^{+\infty}\pmb B^{(k)} ∑k=0+∞A(k),∑k=0+∞B(k)
-
∑ k = 0 + ∞ ( A ( k ) + B ( k ) ) = A + B \sum_{k=0}^{+\infty}(\pmb A^{(k)} + \pmb B^{(k)}) = \pmb A + \pmb B ∑k=0+∞(A(k)+B(k))=A+B
-
∀ λ ∈ C , ∑ k = 0 + ∞ λ A ( k ) = λ A \forall \lambda \in \pmb C,\sum_{k=0}^{+\infty}\lambda \pmb A^{(k)} = \lambda \pmb A ∀λ∈C,∑k=0+∞λA(k)=λA
-
绝对收敛的矩阵级数必收敛,并且任意调换其项的顺序所得的矩阵级数仍收敛,且其和不变
-
若矩阵级数 ∑ k = 0 + ∞ A ( k ) \sum_{k=0}^{+\infty} \pmb A^{(k)} ∑k=0+∞A(k)收敛(或绝对收敛),则矩阵级数 ∑ k = 0 + ∞ P A ( k ) Q \sum_{k=0}^{+\infty} \pmb P \pmb A^{(k)} \pmb Q ∑k=0+∞PA(k)Q也收敛(或绝对收敛),并且有
∑ k = 0 + ∞ P A ( k ) Q = P ( ∑ k = 0 + ∞ A ( k ) ) Q \sum_{k=0}^{+\infty} \pmb P \pmb A^{(k)} \pmb Q = \pmb P \bigg(\sum_{k=0}^{+\infty} \pmb A^{(k)} \bigg) \pmb Q k=0∑+∞PA(k)Q=P(k=0∑+∞A(k))Q -
若 ∑ k = 0 + ∞ A ( k ) , ∑ k = 0 + ∞ B ( k ) \sum_{k=0}^{+\infty} \pmb A^{(k)},\sum_{k=0}^{+\infty} \pmb B^{(k)} ∑k=0+∞A(k),∑k=0+∞B(k)均绝对收敛,则它们按项相乘所得的矩阵级数
A ( 0 ) B ( 0 ) + ( A ( 0 ) B ( 1 ) + A ( 1 ) B ( 0 ) ) + ⋯ + ( A ( 0 ) B ( k ) + A ( 1 ) B ( k − 1 ) + ⋯ + A ( k ) B ( 0 ) ) + ⋯ \pmb A^{(0)} \pmb B^{(0)} + (\pmb A^{(0)} \pmb B^{(1)} + \pmb A^{(1)} \pmb B^{(0)} ) + \cdots + (\pmb A^{(0)} \pmb B^{(k)} + \pmb A^{(1)} \pmb B^{(k-1)} + \cdots + \pmb A^{(k)} \pmb B^{(0)} ) + \cdots A(0)B(0)+(A(0)B(1)+A(1)B(0))+⋯+(A(0)B(k)+A(1)B(k−1)+⋯+A(k)B(0))+⋯
也绝对收敛,且其和为 A B \pmb A \pmb B AB
-
幂级数
设
A
(
k
)
∈
C
m
×
n
,
a
k
∈
C
\pmb A^{(k)} \in \pmb C^{ m \times n},a_k \in \pmb C
A(k)∈Cm×n,ak∈C,称矩阵级数
∑
k
=
0
+
∞
a
k
A
k
\sum_{k=0}^{+\infty} a_k \pmb A^k
k=0∑+∞akAk
为矩阵
A
\pmb A
A的幂级数
设幂级数 ∑ k = 0 + ∞ a k z k \sum_{k=0}^{+\infty} a_k z^k ∑k=0+∞akzk的收敛半径为 r , A ∈ C n × n r,\pmb A \in \pmb C^{n \times n} r,A∈Cn×n,则
- ρ ( A ) < r \rho(\pmb A) <r ρ(A)<r时,矩阵幂级数 ∑ k = 0 + ∞ a k A k \sum_{k=0}^{+\infty} a_k \pmb A^k ∑k=0+∞akAk绝对收敛
- ρ ( A ) > r \rho(\pmb A) >r ρ(A)>r时,矩阵幂级数 ∑ k = 0 + ∞ a k A k \sum_{k=0}^{+\infty} a_k \pmb A^k ∑k=0+∞akAk发散
- 若存在某一范数 ∥ A ∥ < r \parallel \pmb A \parallel <r ∥A∥<r,则矩阵幂级数 ∑ k = 0 + ∞ a k A k \sum_{k=0}^{+\infty} a_k \pmb A^k ∑k=0+∞akAk绝对收敛
- 矩阵幂级数 ∑ k = 0 + ∞ A k \sum_{k=0}^{+\infty} \pmb A^k ∑k=0+∞Ak收敛的充分必要条件是 ρ ( A ) < 1 \rho(\pmb A) <1 ρ(A)<1,且收敛于 ( I − A ) − 1 (\pmb I -\pmb A)^{-1} (I−A)−1
矩阵函数
定义
设幂级数
∑
k
=
0
+
∞
a
k
z
k
\sum_{k=0}^{+\infty} a_k z^k
∑k=0+∞akzk的收敛半径为
r
r
r,且当
∣
z
∣
<
r
\mid z \mid <r
∣z∣<r时,幂级数收敛于函数
f
(
z
)
f(z)
f(z),即
f
(
z
)
=
∑
k
=
0
+
∞
a
k
z
k
(
∣
z
∣
<
r
)
f(z) = \sum_{k=0}^{+\infty} a_k z^k (\mid z \mid <r)
f(z)=k=0∑+∞akzk(∣z∣<r)
如果
A
∈
C
n
×
n
\pmb A \in \pmb C^{n \times n}
A∈Cn×n满足
ρ
(
A
)
<
r
\rho(\pmb A)<r
ρ(A)<r,则称收敛的矩阵幂级数
∑
k
=
0
+
∞
a
k
A
k
\sum_{k=0}^{+\infty} a_k \pmb A^k
∑k=0+∞akAk的和仍为矩阵函数,记为
f
(
A
)
f(\pmb A)
f(A)
f
(
A
)
=
∑
k
=
0
+
∞
a
k
A
k
f(\pmb A) = \sum_{k=0}^{+\infty} a_k \pmb A^k
f(A)=k=0∑+∞akAk
e A = ∑ k = 0 + ∞ 1 k ! A k sin A = ∑ k = 0 + ∞ ( − 1 ) k ( 2 k + 1 ) ! A 2 k + 1 cos A = ∑ k = 0 + ∞ ( − 1 ) k ( 2 k ) ! A 2 k ( I − A ) − 1 = ∑ k = 0 + ∞ A k , ρ ( A ) < 1 ln ( I + A ) = ∑ k = 0 + ∞ ( − 1 ) k k + 1 A k + 1 , ρ ( A ) < 1 \begin{aligned} &e^{\pmb A} = \sum_{k=0}^{+\infty}\cfrac1{k!} \pmb A^k \\ &\sin \pmb A = \sum_{k=0}^{+\infty} \cfrac{(-1)^k}{(2k+1)!}\pmb A^{2k+1}\\ &\cos \pmb A = \sum_{k=0}^{+\infty} \cfrac{(-1)^k}{(2k)!} \pmb A^{2k} \\ &(\pmb I -\pmb A)^{-1} = \sum_{k=0}^{+\infty}\pmb A^k,\rho(\pmb A)<1\\ &\ln(\pmb I + \pmb A) = \sum_{k=0}^{+\infty} \cfrac{(-1)^k}{k+1}\pmb A^{k+1},\rho(\pmb A)<1 \end{aligned} eA=k=0∑+∞k!1AksinA=k=0∑+∞(2k+1)!(−1)kA2k+1cosA=k=0∑+∞(2k)!(−1)kA2k(I−A)−1=k=0∑+∞Ak,ρ(A)<1ln(I+A)=k=0∑+∞k+1(−1)kAk+1,ρ(A)<1
计算
-
利用 H a m i l t o n − C a y l e y \rm Hamilton- Cayley Hamilton−Cayley定理找出矩阵方幂之间的关系进行化简
-
利用相似对角化
P − 1 A P = d i a g ( λ 1 , λ 2 , ⋯ , λ n ) = Λ ⇒ f ( A ) = P d i a g ( f ( λ 1 ) , f ( λ 2 ) , ⋯ , f ( λ n ) P − 1 \pmb P^{-1} \pmb A \pmb P = diag(\lambda_1,\lambda_2,\cdots,\lambda_n) = \Lambda \Rightarrow f(\pmb A) = \pmb P diag(f(\lambda_1),f(\lambda_2),\cdots,f(\lambda_n) \pmb P^{-1} P−1AP=diag(λ1,λ2,⋯,λn)=Λ⇒f(A)=Pdiag(f(λ1),f(λ2),⋯,f(λn)P−1
-
利用 J o r d a n \rm Jordan Jordan标准形
f ( A t ) = P [ f ( J 1 t ) ⋱ f ( J s t ) ] P − 1 f(\pmb A t) = \pmb P \begin{bmatrix} f(\pmb J_1t) & & \\ &\ddots &\\ && f(\pmb J_st) \end{bmatrix} \pmb P^{-1} f(At)=P f(J1t)⋱f(Jst) P−1f ( J i t ) = [ f ( λ ) t 1 ! f ′ ( λ ) ⋯ t r i − 1 ( r i − 1 ) ! f ( r i − 1 ) ( λ ) f ( λ ) ⋱ ⋮ ⋱ t 1 ! f ′ ( λ ) f ( λ ) ] λ = λ i t f(\pmb J_i t) = \begin{bmatrix} f(\lambda) & \frac{t}{1!}f^\prime(\lambda) &\cdots &\frac{t^{r_i-1}}{(r_i-1)!}f^{(r_i-1)}(\lambda) \\ & f(\lambda) &\ddots &\vdots \\ &&\ddots &\frac{t}{1!}f^\prime(\lambda) \\ &&&f(\lambda) \end{bmatrix}_{\lambda = \lambda_i t} f(Jit)= f(λ)1!tf′(λ)f(λ)⋯⋱⋱(ri−1)!tri−1f(ri−1)(λ)⋮1!tf′(λ)f(λ) λ=λit
-
待定系数法
-
求矩阵 A \pmb A A的特征多项式(最高项为 n n n)(如果得到最小多项式则更简单)
-
设 r ( λ ) = b n − 1 λ n − 1 + ⋯ + b 1 λ + b 0 r(\lambda) = b_{n-1}\lambda^{n-1}+ \cdots + b_{1}\lambda + b_0 r(λ)=bn−1λn−1+⋯+b1λ+b0,根据
r ( l ) ( λ i ) = t l f ( l ) ( λ ) ∣ λ = λ i t r^{(l)}(\lambda_i) = t^lf^{(l)}(\lambda)\mid_{\lambda = \lambda_i t} r(l)(λi)=tlf(l)(λ)∣λ=λit
或者
r ( l ) ( λ i ) = f ( l ) ( λ t ) ∣ λ = λ i r^{(l)}(\lambda_i) = f^{(l)}(\lambda t)\mid_{\lambda = \lambda_i} r(l)(λi)=f(l)(λt)∣λ=λi
列方程组求解 b 0 , b 1 , ⋯ , b n − 1 b_0,b_1,\cdots,b_{n-1} b0,b1,⋯,bn−1 -
计算 f ( A t ) = r ( A ) = b n − 1 A n − 1 + ⋯ + b 1 A + b 0 I f(\pmb At) = r(A) = b_{n-1}\pmb A^{n-1}+ \cdots + b_{1}\pmb A + b_0 \pmb I f(At)=r(A)=bn−1An−1+⋯+b1A+b0I
-
性质
-
sin ( A ) = − sin A , cos ( − A ) = cos A \sin(\pmb A) = -\sin \pmb A,\cos (-\pmb A) = \cos \pmb A sin(A)=−sinA,cos(−A)=cosA
-
e i A = cos A + i sin A e^{i \pmb A} = \cos \pmb A + i\sin \pmb A eiA=cosA+isinA
-
cos A = 1 2 ( e i A + e − i A ) , sin A = 1 2 i ( e i A − e − i A ) \cos \pmb A = \cfrac12 (e^{i \pmb A} + e^{-i \pmb A}),\sin \pmb A = \cfrac1{2i}(e^{i\pmb A}- e^{-i\pmb A}) cosA=21(eiA+e−iA),sinA=2i1(eiA−e−iA)
-
e A + B = e A e B = e B e A e^{\pmb A+\pmb B} = e^{\pmb A}e^{\pmb B} = e^{\pmb B}e^{\pmb A} eA+B=eAeB=eBeA
-
sin ( A + B ) = sin A cos B + cos A sin B , ( A B = B A ) \sin(\pmb A + \pmb B) = \sin \pmb A \cos \pmb B + \cos \pmb A \sin \pmb B,(\pmb A\pmb B=\pmb B\pmb A) sin(A+B)=sinAcosB+cosAsinB,(AB=BA)
-
cos ( A + B ) = cos A cos B − sin A sin B , ( A B = B A ) \cos(\pmb A + \pmb B) = \cos \pmb A \cos \pmb B - \sin \pmb A \sin \pmb B,(\pmb A\pmb B=\pmb B\pmb A) cos(A+B)=cosAcosB−sinAsinB,(AB=BA)
-
cos 2 A = cos 2 A − sin 2 A \cos {2\pmb A} = \cos ^2 \pmb A - \sin ^2\pmb A cos2A=cos2A−sin2A
-
sin 2 A = 2 sin A cos A \sin 2\pmb A = 2\sin \pmb A \cos \pmb A sin2A=2sinAcosA
-
det e A = e t r A \det e^{\pmb A} = e^{tr \pmb A} deteA=etrA
-
( e A ) − 1 = e − A (e^{\pmb A})^{-1} = e^{-\pmb A} (eA)−1=e−A
矩阵的微分与积分
函数矩阵的微分与积分
以变量
t
t
t的函数为元素的矩阵
A
(
t
)
=
(
a
i
j
(
t
)
)
m
×
n
\pmb A(t) = (a_{ij}(t))_{m \times n}
A(t)=(aij(t))m×n称为函数矩阵。如果
t
∈
[
a
,
b
]
t\in [a,b]
t∈[a,b],则
A
(
t
)
\pmb A(t)
A(t)的定义域为
[
a
,
b
]
[a,b]
[a,b],如果每个
a
i
j
(
t
)
a_{ij}(t)
aij(t)都在
[
a
,
b
]
[a,b]
[a,b]上连续、可微、可积,则称
A
(
t
)
\pmb A(t)
A(t)在
[
a
,
b
]
[a,b]
[a,b]上是连续、可微、可积。当
A
(
t
)
\pmb A(t)
A(t)可微时,规定其导数为
A
′
(
t
)
=
(
a
i
j
′
(
t
)
)
m
×
n
o
r
d
d
t
A
(
t
)
=
(
d
d
t
a
i
j
(
t
)
)
m
×
n
\pmb A^\prime(t) = (a_{ij}^\prime(t))_{m \times n } \ or \ \frac{d}{dt}\pmb A(t) = \Big(\frac d{dt}a_{ij}(t)\Big)_{m \times n}
A′(t)=(aij′(t))m×n or dtdA(t)=(dtdaij(t))m×n
当
A
(
t
)
\pmb A(t)
A(t)在
[
a
,
b
]
[a,b]
[a,b]上可积时,规定
A
(
t
)
\pmb A(t)
A(t)在
[
a
,
b
]
[a,b]
[a,b]上的积分为
∫
a
b
A
(
t
)
d
t
=
(
∫
a
b
a
i
j
(
t
)
d
t
)
m
×
n
\int_a^b \pmb A(t)dt = \Big(\int_a^ba_{ij}(t)dt\Big)_{m\times n }
∫abA(t)dt=(∫abaij(t)dt)m×n
求导法则
设 A ( t ) \pmb A(t) A(t)与 B ( t ) \pmb B(t) B(t)是适当阶的可微矩阵,则
-
d d t ( A ( t ) + B ( t ) ) = d d t A ( t ) + d d t B ( t ) \cfrac{d}{dt}(\pmb A(t) + \pmb B(t)) = \cfrac{d}{dt}\pmb A(t) + \cfrac{d}{dt}\pmb B(t) dtd(A(t)+B(t))=dtdA(t)+dtdB(t)
-
当 λ ( t ) \lambda(t) λ(t)为可微函数时,有
d d t ( λ ( t ) A ( t ) ) = ( d d t λ ( t ) ) A ( t ) + λ ( t ) d d t A ( t ) \cfrac{d}{dt}(\lambda(t) \pmb A(t)) = \bigg(\cfrac{d}{dt}\lambda(t) \bigg)\pmb A(t) + \lambda(t) \cfrac{d}{dt}\pmb A(t) dtd(λ(t)A(t))=(dtdλ(t))A(t)+λ(t)dtdA(t) -
d d t ( A ( t ) B ( t ) ) = ( d d t A ( t ) ) B ( t ) + A ( t ) d d t B ( t ) \cfrac{d}{dt}(\pmb A(t) \pmb B(t)) = \bigg(\cfrac{d}{dt} \pmb A(t)\bigg)\pmb B(t) + \pmb A(t) \cfrac{d}{dt} \pmb B(t) dtd(A(t)B(t))=(dtdA(t))B(t)+A(t)dtdB(t)
-
当 u = f ( t ) u=f(t) u=f(t)关于 t t t可微时,有
d d t A ( u ) = f ′ ( t ) d d t A ( u ) \cfrac{d}{dt}\pmb A(u) = f^\prime(t) \cfrac{d}{dt}\pmb A(u) dtdA(u)=f′(t)dtdA(u) -
当 A − 1 ( t ) \pmb A^{-1}(t) A−1(t)是可微矩阵时,有
d d t ( A − 1 ( t ) ) = − A − 1 ( t ) ( d d t A ( t ) ) A − 1 ( t ) \frac d{dt}(\pmb A^{-1}(t)) = -\pmb A^{-1}(t) \bigg(\frac d{dt} \pmb A(t) \bigg) \pmb A^{-1}(t) dtd(A−1(t))=−A−1(t)(dtdA(t))A−1(t)
特定求导公式
- d d t e A t = A e A t = e A t A \cfrac{d}{dt} e^{\pmb A t} =\pmb A e^{\pmb A t} = e^{\pmb A t} \pmb A dtdeAt=AeAt=eAtA
- d d t sin A t = A cos A t = ( cos A t ) A \cfrac{d}{dt} \sin \pmb At = \pmb A \cos \pmb A t = (\cos \pmb At) \pmb A dtdsinAt=AcosAt=(cosAt)A
- d d t cos A t = − A sin A t = − ( sin A t ) A \cfrac{d}{dt} \cos\pmb At = -\pmb A \sin\pmb A t = -(\sin\pmb At) \pmb A dtdcosAt=−AsinAt=−(sinAt)A
积分公式
-
∫ a b ( A ( t ) + B ( t ) ) d t = ∫ a b A ( t ) d t + ∫ a b B ( t ) d t \int_a^b (\pmb A(t) + \pmb B(t))dt = \int_a^b\pmb A(t)dt + \int_a^b\pmb B(t)dt ∫ab(A(t)+B(t))dt=∫abA(t)dt+∫abB(t)dt
-
∫ a b λ A ( t ) = λ ∫ a b A ( t ) d t \int_a^b\lambda\pmb A(t) = \lambda \int_a^b\pmb A(t)dt ∫abλA(t)=λ∫abA(t)dt
-
∫ a b A ( t ) B d t = ( ∫ a b A ( t ) d t ) B , ∫ a b A B ( t ) d t = A ∫ a b B ( t ) d t \int_a^b\pmb A(t)\pmb Bdt = (\int_a^b\pmb A(t)dt) \pmb B,\int_a^b\pmb A\pmb B(t)dt = \pmb A \int_a^b \pmb B(t)dt ∫abA(t)Bdt=(∫abA(t)dt)B,∫abAB(t)dt=A∫abB(t)dt
-
当 A ( t ) \pmb A(t) A(t)在 [ a , b ] [a,b] [a,b]上连续可微时, ∀ t ∈ ( a , b ) \forall t \in (a,b) ∀t∈(a,b),有
d d t ( ∫ a t A ( τ ) d τ ) = A ( t ) \frac d {dt} (\int_a^t \pmb A(\tau) d\tau) = \pmb A(t) dtd(∫atA(τ)dτ)=A(t) -
当 A ( t ) \pmb A(t) A(t)在 [ a , b ] [a,b] [a,b]上连续可微时,有
d d t A ′ ( t ) d t = A ( b ) − A ( a ) \frac d{dt} \pmb A^\prime(t) dt = \pmb A(b) - \pmb A(a) dtdA′(t)dt=A(b)−A(a)
数量函数对矩阵变量的导数
设
f
(
X
)
f(\pmb X)
f(X)是以矩阵
X
=
(
x
i
j
)
m
×
n
\pmb X = (x_{ij})_{m \times n}
X=(xij)m×n为自变量的
m
n
mn
mn元函数,且
∂
f
∂
x
i
j
\cfrac {\partial f}{\partial x_{ij}}
∂xij∂f都存在,规定
f
f
f对矩阵变量
X
\pmb X
X的导数
d
f
d
X
\cfrac{df}{d\pmb X}
dXdf为
d
f
d
X
=
(
∂
f
∂
x
i
j
)
m
×
n
=
[
∂
f
∂
x
11
⋯
∂
f
∂
x
1
n
⋮
⋮
∂
f
∂
x
m
1
⋯
∂
f
∂
x
m
n
]
\frac {df}{d\pmb X} = (\frac{\partial f}{\partial x_{ij}})_{m \times n} = \begin{bmatrix} \cfrac{\partial f}{\partial x_{11}} & \cdots &\cfrac{\partial f}{\partial x_{1n}} \\ \vdots & &\vdots\\ \cfrac{\partial f}{\partial x_{m1}} & \cdots &\cfrac{\partial f}{\partial x_{mn}} \\ \end{bmatrix}
dXdf=(∂xij∂f)m×n=
∂x11∂f⋮∂xm1∂f⋯⋯∂x1n∂f⋮∂xmn∂f
特别的,以
x
=
(
ξ
1
,
ξ
2
,
⋯
,
ξ
n
)
T
\pmb x = (\xi_1,\xi_2,\cdots,\xi_n)^T
x=(ξ1,ξ2,⋯,ξn)T为自变量的函数
f
(
x
)
f(\pmb x)
f(x)的导数
d
f
d
x
=
(
∂
f
∂
ξ
1
,
∂
f
∂
ξ
2
,
⋯
,
∂
f
∂
ξ
n
)
T
\frac{d f}{d \pmb x} = (\frac{\partial f}{\partial \xi_1},\frac{\partial f}{\partial \xi_2},\cdots,\frac{\partial f}{\partial \xi_n})^T
dxdf=(∂ξ1∂f,∂ξ2∂f,⋯,∂ξn∂f)T
称为数量函数对向量变量的导数,即为在数学分析中的
f
f
f的梯度向量,记为
grad
f
\text{grad} \ \ f
grad f
d
(
t
r
(
A
X
)
)
d
X
=
A
T
\frac{d (tr(\pmb A\pmb X))}{d \pmb X} = \pmb A^T
dXd(tr(AX))=AT
d
(
x
T
A
x
)
d
x
=
(
A
+
A
T
)
x
\frac{d(\pmb x^T \pmb A \pmb x)}{d \pmb x} = (\pmb A + \pmb A^T) \pmb x
dxd(xTAx)=(A+AT)x
d
d
X
det
X
=
(
det
X
)
(
X
−
1
)
T
\frac{d}{d \pmb X} \det \pmb X = (\det \pmb X)(\pmb X^{-1})^T
dXddetX=(detX)(X−1)T
矩阵值函数对矩阵变量的导数
矩阵分析应用举例
求解一阶线性常系数微分方程组
KaTeX parse error: Undefined control sequence: \cases at position 2: \̲c̲a̲s̲e̲s̲{ \cfrac{d\pmb …
e A t = b 1 ( t ) A + b 0 ( t ) I , A x = λ x ⟹ e A t x = [ b 1 ( t ) λ + b 0 ( t ) ] x ⟹ e − A t x = [ b 1 ( − t ) λ + b 0 ( − t ) ] x \begin{aligned} &e^{\pmb At} = b_1(t)\pmb A +b_0(t) \pmb I \pmb ,A \pmb x = \lambda \pmb x \\ \Longrightarrow &e^{\pmb At} \pmb x = [b_1(t) \lambda +b_0(t) ] \pmb x \\ \Longrightarrow&e^{-\pmb At} \pmb x = [b_1(-t) \lambda +b_0(-t) ] \pmb x \end{aligned} ⟹⟹eAt=b1(t)A+b0(t)I,Ax=λxeAtx=[b1(t)λ+b0(t)]xe−Atx=[b1(−t)λ+b0(−t)]x
求解矩阵方程
A
X
+
X
B
=
F
AX+XB=F
AX+XB=F
如果
A
,
B
A,B
A,B的所有特征值都具有非负实部,则有唯一解
X
=
−
∫
0
+
∞
e
A
t
F
e
B
t
d
t
X = - \int_0^{+\infty} e^{At}Fe^{Bt}dt
X=−∫0+∞eAtFeBtdt
最小二乘问题
对于矛盾方程组
∥
A
x
0
−
b
∥
2
=
min
x
∈
C
n
∥
A
x
−
b
∥
2
\parallel Ax_0 - b \parallel_2 = \min_{x\in C^n} \parallel Ax-b \parallel_2
∥Ax0−b∥2=x∈Cnmin∥Ax−b∥2
当
A
,
b
A,b
A,b分别为实数矩阵和向量时,解为
A
T
A
x
=
A
T
b
(法方程组)
A^TAx = A^Tb \tag{法方程组}
ATAx=ATb(法方程组)