1. 方阵函数
1.1 方阵函数的定义
首先我们定义一些符号方便使用。
定义 1.1.1 特征多项式
χ
A
(
λ
)
=
∣
λ
E
−
A
∣
\chi_A(\lambda)=|\lambda E-A|
χA(λ)=∣λE−A∣ 的根系(连重数)称为方阵
A
A
A 的 谱
S
p
e
c
A
\mathrm {Spec}A
SpecA.
m
A
(
λ
)
m_A(\lambda)
mA(λ) 的根系(连重数)称为方阵
A
A
A 的 简谱
R
s
p
e
c
A
\mathrm{Rspec}A
RspecA.
定义 1.1.2 设
f
f
f 是任一函数 ( 有后面所需的各处各阶导数 ) , 则
{
f
(
j
)
(
λ
i
)
∣
j
=
0
,
1
,
⋯
,
m
i
−
1
,
i
=
1
,
⋯
,
k
}
\{f^{(j)}(\lambda_i)|j=0,1,\cdots,m_{i-1},i=1,\cdots,k\}
{f(j)(λi)∣j=0,1,⋯,mi−1,i=1,⋯,k} 为
f
f
f 在
A
A
A 的简谱
{
λ
1
,
⋯
,
λ
1
,
⋯
,
λ
k
,
⋯
,
λ
k
}
\{\lambda_1 ,\cdots ,\lambda_1 ,\cdots,\lambda_k,\cdots,\lambda_k\}
{λ1,⋯,λ1,⋯,λk,⋯,λk} 处的值 , 记作
f
(
R
s
p
e
c
A
)
f(\mathrm{Rspec}A)
f(RspecA) 或
f
(
Λ
A
)
f(\Lambda_A)
f(ΛA) .
根据多项式函数
f
(
λ
)
=
a
0
λ
l
+
a
1
λ
l
−
1
+
⋯
+
a
l
f(\lambda)=a_0\lambda^l+a_1\lambda^{l-1}+\cdots+a_l
f(λ)=a0λl+a1λl−1+⋯+al , 定义方阵函数
f
(
A
)
=
a
0
A
l
+
a
1
A
l
−
1
+
⋯
+
a
l
I
n
.
f(\boldsymbol{A})=a_0\boldsymbol{A}^l+a_1\boldsymbol{A}^{l-1}+\cdots+a_l\boldsymbol{I}_n.
f(A)=a0Al+a1Al−1+⋯+alIn. 这时,若有另一多项式
g
(
λ
)
g(\lambda)
g(λ), 则易知
f
(
A
)
=
g
(
A
)
⇔
m
A
(
λ
)
∣
(
f
(
λ
)
−
g
(
λ
)
)
,
f(A)=g(A)\Leftrightarrow m_A(\lambda)\:|\:(f(\lambda)-g(\lambda))\:,
f(A)=g(A)⇔mA(λ)∣(f(λ)−g(λ)),
m
A
(
λ
)
m_{A}(\lambda)
mA(λ) 是
A
A
A 的最小多项式. 若记
m
A
(
λ
)
=
(
λ
−
λ
1
)
m
1
⋯
(
λ
−
λ
k
)
m
k
m_A(\lambda)=(\lambda-\lambda_1)^{m_1}\cdots(\lambda-\lambda_k)^{m_k}
mA(λ)=(λ−λ1)m1⋯(λ−λk)mk,
d
e
g
m
A
(
λ
)
=
m
1
+
⋯
+
m
k
=
m
\mathrm{deg}~m_A(\lambda)=m_1+\cdots+m_k=m
deg mA(λ)=m1+⋯+mk=m ,则
f
(
λ
)
−
g
(
λ
)
f(\lambda)-g(\lambda)
f(λ)−g(λ) 在
λ
i
\lambda_i
λi 处至少有
m
i
m_i
mi 重根,即
m
A
(
λ
)
∣
(
f
(
λ
)
−
g
(
λ
)
)
⇔
f
(
j
)
(
λ
i
)
=
g
(
j
)
(
λ
i
)
(
i
=
1
,
⋯
,
k
;
j
=
0
,
1
,
⋯
,
m
i
−
1
)
m_A(\lambda)\mid(f(\lambda)-g(\lambda))\Leftrightarrow f^{(j)}(\lambda_i)=g^{(j)}(\lambda_i)( i = 1, \cdots, k~; j = 0, 1, \cdots, m_{i} - 1 )
mA(λ)∣(f(λ)−g(λ))⇔f(j)(λi)=g(j)(λi)(i=1,⋯,k ;j=0,1,⋯,mi−1)这里
f
(
j
)
(
λ
i
)
f^{(j)}(\lambda_{i})
f(j)(λi) 是多项式函数
f
(
λ
)
f(\lambda)
f(λ) 的
j
j
j 次导数在
λ
=
λ
i
\lambda=\lambda_i
λ=λi 时的值 .
所以,对多项式
f
(
λ
)
f(\lambda)
f(λ) 来说,
f
(
A
)
f(A)
f(A) 有总共
m
=
m
1
+
⋯
+
m
k
m=m_1+\cdots+m_k
m=m1+⋯+mk 个值
f
(
λ
i
)
,
f
(
1
)
(
λ
i
)
,
⋯
,
f
(
m
i
−
1
)
(
λ
i
)
(
i
=
1
,
⋯
,
k
)
f(\lambda_i),f^{(1)}(\lambda_i),\cdots,f^{(m_i-1)}(\lambda_i)\quad(i=1,\cdots,k)
f(λi),f(1)(λi),⋯,f(mi−1)(λi)(i=1,⋯,k)是唯一确定的. 如果
m
m
m 个等式
f
(
j
)
(
λ
i
)
=
g
(
j
)
(
λ
i
)
(
i
=
1
,
⋯
,
k
;
j
=
0
,
1
,
⋯
,
m
i
−
1
)
f^{(j)}\left(\lambda_i\right)\:=\:g^{(j)}\left(\lambda_i\right)\quad(i=1,\cdots,k\:;j=0,1,\cdots,m_i-1)
f(j)(λi)=g(j)(λi)(i=1,⋯,k;j=0,1,⋯,mi−1) 成立,这
m
m
m 个等式可简记为
f
(
Λ
A
)
=
g
(
Λ
A
)
f(\Lambda_A)=g(\Lambda_A)
f(ΛA)=g(ΛA) , 则对多项式
f
(
λ
)
f(\lambda)
f(λ) 和
g
(
λ
)
g(\lambda)
g(λ), 有
f
(
A
)
=
g
(
A
)
⇔
f
(
Λ
A
)
=
g
(
Λ
A
)
.
f(A)\:=\:g(A)\Leftrightarrow f(\Lambda_A)=g(\Lambda_A)\:.
f(A)=g(A)⇔f(ΛA)=g(ΛA).现在来对一般的
f
(
λ
)
f(\lambda)
f(λ) 定义
f
(
A
)
.
f(\boldsymbol{A}).
f(A).
我们总 假设 f ( λ ) f(\lambda) f(λ) 在各 λ i \lambda_i λi 处有直到 m i − 1 m_i-1 mi−1 阶的导数——简称为 f ( λ ) f(\lambda) f(λ) 在 Λ A \Lambda_\mathrm{A} ΛA 上有意义.
定义 1.1.3 (矩阵函数的第一种定义) 对函数 f ( λ ) f(\lambda) f(λ) , 如果有多项式 p ( λ ) p(\lambda) p(λ) 使 p ( Λ A ) = f ( Λ A ) p(\Lambda_{A})=f(\Lambda_{A}) p(ΛA)=f(ΛA), 则定义 f ( A ) = p ( A ) f(\boldsymbol{A})=p(\boldsymbol{A}) f(A)=p(A), 并称 p ( λ ) p(\lambda) p(λ) 是 f ( A ) f(\boldsymbol{A}) f(A) 的 定义多项式。
即 p ( λ ) p(\lambda) p(λ) 是与 f ( λ ) f(\lambda) f(λ) 在 A \boldsymbol{A} A 的简谱处的值全部相等的多项式, p ( j ) ( λ i ) = f ( j ) ( λ i ) p^{(j)}(\lambda_{i})=f^{(j)}(\lambda_{i}) p(j)(λi)=f(j)(λi) ( j = 0 , 1 , ⋯ , m i − 1 , i = 1 , ⋯ , k ) (j=0,1,\cdots,m_{i}-1,i=1,\cdots,k) (j=0,1,⋯,mi−1,i=1,⋯,k)。定义多项式不唯一,两定义多项式之差是 m A ( λ ) m_{A}(\lambda) mA(λ) 的倍数,次数小于 m A ( λ ) m_{A}(\lambda) mA(λ) 的定义多项式唯一。
例 1.1.4 设 A = ( 1 4 3 2 ) A=\begin{pmatrix}1&4\\3&2\end{pmatrix} A=(1342),求 e A e^{A} eA .
解 f A ( λ ) = ∣ λ − 1 − 4 − 3 λ − 2 ∣ = ( λ − 5 ) ( λ + 2 ) . f_{A}(\lambda)=\left|\begin{matrix}\lambda-1&-4\\-3&\lambda-2\end{matrix}\right|=(\lambda-5)(\lambda+2). fA(λ)= λ−1−3−4λ−2 =(λ−5)(λ+2). 则最小多项式 m A ( λ ) m_{A}(\lambda) mA(λ) 也是 ( λ − 5 ) ( λ + 2 ) (\lambda-5)(\lambda+2) (λ−5)(λ+2) .
由小于 m A ( λ ) m_{A}(\lambda) mA(λ) 的定义多项式唯一 ,设定义多项式 r ( a ) = a λ + b r(a)=a\lambda+b r(a)=aλ+b, 则 { r ( 5 ) = 5 a + b = e 5 r ( − 2 ) = − 2 a + b = e − 2 \begin{cases}r(5)=5a+b=e^{5}\\r(-2)=-2a+b=e^{-2}\end{cases} {r(5)=5a+b=e5r(−2)=−2a+b=e−2
⇒ { a = 1 7 ( e 5 − e − 2 ) b = 1 7 ( 2 e 5 + 5 e − 2 ) ⇒ e A = r ( A ) = 1 7 ( e 5 − e − 2 ) A + 1 7 ( 2 e 5 + 5 e − 2 ) I \begin{aligned}&\Rightarrow\begin{cases}a=\frac{1}{7}(e^{5}-e^{-2})\\b=\frac{1}{7}(2e^{5}+5e^{-2})\end{cases}&\Rightarrow e^{A}=r(A)=\frac{1}{7}(e^{5}-e^{-2})A+\frac{1}{7}(2e^{5}+5e^{-2})I\end{aligned} ⇒{a=71(e5−e−2)b=71(2e5+5e−2)⇒eA=r(A)=71(e5−e−2)A+71(2e5+5e−2)I
例 1.1.5 A = ( 0 − 1 4 4 ) A=\begin{pmatrix}0&-1\\4&4\end{pmatrix} A=(04−14) , 求 arcsin A 4 \arcsin\dfrac{A}{4} arcsin4A.
解 f A ( λ ) = ∣ λ 1 − 4 λ − 4 ∣ = ( λ − 2 ) 2 f_{A}(\lambda)=\begin{vmatrix}\lambda&1\\-4&\lambda-4\end{vmatrix}=(\lambda-2)^{2} fA(λ)= λ−41λ−4 =(λ−2)2, D 1 ( x ) = 1 , D 2 ( x ) = ( λ − 2 ) 2 = d 2 ( x ) = m A D_{1}(x)=1,D_{2}(x)=(\lambda-2)^{2}=d_{2}(x)=m_{A} D1(x)=1,D2(x)=(λ−2)2=d2(x)=mA
设 r ( x ) = a λ + b ,则 { r ( 2 ) = 2 a + b = arcsin 2 4 ∗ π 6 r ′ ( 2 ) = a = 1 1 − ( 2 4 ) 2 ∗ 1 4 = 3 6 ⇒ b = π 6 − 2 3 6 arcsin A 4 = r ( A ) = 3 6 A + π − 2 3 6 I \begin{aligned}&\text{设}~r(x)=a\lambda+b,\text{则}\begin{cases}r(2)=2a+b=\arcsin\dfrac{2}{4}*\dfrac{\pi}{6}\\r^{\prime}(2)=a=\dfrac{1}{\sqrt{1-(\frac{2}{4})^{2}}}*\dfrac14=\dfrac{\sqrt{3}}{6}\end{cases} \Rightarrow b=\frac{\pi}{6}-2\dfrac{\sqrt{3}}{6}\\ &\arcsin\frac{A}{4}= r(A)=\frac{\sqrt{3}}{6}A+\frac{\pi-2\sqrt{3}}{6}I\end{aligned} 设 r(x)=aλ+b,则⎩ ⎨ ⎧r(2)=2a+b=arcsin42∗6πr′(2)=a=1−(42)21∗41=63⇒b=6π−263arcsin4A=r(A)=63A+6π−23I
例 1.1.6 设 A A A 可逆, f ( λ ) = 1 λ f(\lambda)=\frac{1}{\lambda} f(λ)=λ1, 则 f ( A ) = A − 1 f(A)=A^{-1} f(A)=A−1。
解 记 f A ( λ ) = a ( λ g ( λ ) − 1 ) f_{A}(\lambda)=a(\lambda g(\lambda)-1) fA(λ)=a(λg(λ)−1),则 λ g ( λ ) \lambda g(\lambda) λg(λ)与 1 1 1 在 R s p e c A \mathrm{RspecA} RspecA 处全相等
⇒ g ( λ ) 与 1 λ 在 R s p e c A 处全相等 ⇒ g ( λ ) = 1 λ ⇒ f ( A ) = g ( A ) = A − 1 \begin{aligned}&\Rightarrow g(\lambda)\text{与}\frac{1}{\lambda}{在~\mathrm{RspecA}~处全相等}\Rightarrow g(\lambda)=\frac{1}{\lambda}\Rightarrow f(A)=g(A)=A^{-1}\end{aligned} ⇒g(λ)与λ1在 RspecA 处全相等⇒g(λ)=λ1⇒f(A)=g(A)=A−1
定义 1.1.7 (矩阵函数的第二种定义) 设复函数
f
(
λ
)
f(\lambda)
f(λ) 在开圆域
∣
λ
−
λ
0
∣
<
r
|\lambda-\lambda_0|<r
∣λ−λ0∣<r 内解析,即
f
(
λ
)
f(\lambda)
f(λ) 在此开圆域内可展开成幂级数
f
(
λ
)
=
∑
k
=
0
∞
a
k
(
λ
−
λ
0
)
k
,
f(\lambda)\:=\:\sum_{k\:=\:0}^\infty a_k(\lambda\:-\:\lambda_0)^k\:,
f(λ)=∑k=0∞ak(λ−λ0)k, 则只要方阵
A
A
A 的所有特征值都在这个开圆域内,就有
f
(
A
)
=
∑
k
=
0
∞
a
k
(
A
−
λ
0
I
)
k
.
f(A)\:=\:\sum_{k\:=\:0}^{\infty}a_{k}\:(A-\lambda_{0}I)^{k}\:.
f(A)=k=0∑∞ak(A−λ0I)k.首先第一种定义的矩阵函数有如下的基本性质:
- A = S B S − 1 ⇒ f ( A= \boldsymbol{SBS}^{- 1}\Rightarrow f( A=SBS−1⇒f( A \boldsymbol{A} A ) = S f ( ) = \boldsymbol{S}f( )=Sf( B \boldsymbol{B} B ) S − 1 ) \boldsymbol{S}^{- 1} )S−1 ;
- f ( A 1 ⊕ ⋯ + ⨁ A l ) = f ( A 1 ) ⊕ ⋯ ⊕ f ( A l ) = p ( Λ B ) . f( \boldsymbol{A}_{1}\oplus \cdots + \bigoplus \boldsymbol{A}_{l}) = f( \boldsymbol{A}_{1}) \oplus \cdots \oplus f( \boldsymbol{A}_{l}) = p( \Lambda _{B}) . f(A1⊕⋯+⨁Al)=f(A1)⊕⋯⊕f(Al)=p(ΛB).
证 (1) 由 A = S B S − 1 A=SBS^{-1} A=SBS−1 可知 m A ( λ ) = m B ( λ ) m_A(\lambda)=m_B(\lambda) mA(λ)=mB(λ) , 从而 Λ A = Λ B , f ( Λ A ) = \Lambda_A=\Lambda_B,f(\Lambda_A)= ΛA=ΛB,f(ΛA)= f ( Λ B ) . f(\Lambda_{B}). f(ΛB).
于是 f ( Λ A ) = p ( Λ A ) ⇒ f(\Lambda_A)=p(\Lambda_A) \Rightarrow f(ΛA)=p(ΛA)⇒ f ( Λ B ) = p ( Λ B ) , f ( A ) = p ( A ) = S p ( B ) S − 1 = f(\Lambda_B)=p(\Lambda_B) ,f(A)=p(A)={S}p({B}){S}^{-1}= f(ΛB)=p(ΛB),f(A)=p(A)=Sp(B)S−1= S f ( Sf( Sf( B B B ) S − 1 S^{- 1} S−1 .
(2) 由于 m A ( λ ) m_A(\lambda) mA(λ) 一定是每个 A i A_i Ai 的零化多项式,所以 m A i ( λ ) ∣ m A ( λ ) ⇒ Λ A i m_{A_i}(\lambda)|m_A(\lambda)\Rightarrow\Lambda_{A_i} mAi(λ)∣mA(λ)⇒ΛAi ⊆ Λ A . \subseteq\Lambda_{\boldsymbol{A}}. ⊆ΛA. 从而对 f ( A ) f(\boldsymbol{A}) f(A) 的任一定义多项式 p ( λ ) p(\lambda) p(λ) 来说 , p ( A 1 ⊕ ⋅ ⋅ ⋅ ⊕ A l ) = p ( A 1 ) ⊕ ⋯ ,p(\boldsymbol{A}_1\oplus\cdotp\cdotp\cdotp\oplus\boldsymbol{A}_l)=p(\boldsymbol{A}_1)\oplus\cdots ,p(A1⊕⋅⋅⋅⊕Al)=p(A1)⊕⋯ + ⊕ p ( A l ) +\oplus p(\boldsymbol{A}_{l}) +⊕p(Al)显然成立,故 2 成立.
- 定理 1.1.8 (若当块的幂级数) 设 f ( z ) = ∑ m = 0 ∞ a m z m f(z)=\sum_{m=0}^{\infty}a_{m}z^{m} f(z)=∑m=0∞amzm 的收敛半径 R R R , J = ( a 1 a ⋱ ⋱ 1 a ) J=\begin{pmatrix}a&1&&\\&a&\ddots&\\&&\ddots&1\\&&&a\end{pmatrix} J= a1a⋱⋱1a , 其中 ∣ a ∣ < R |a|<R ∣a∣<R ,则 f ( J ) = ∑ m = 1 ∞ a m J m = ( f ( a ) f ′ ( a ) 1 ! f ′ ′ ( a ) 2 ! ⋯ f ( p − 1 ) ( a ) ( p − 1 ) ! f ( a ) ⋱ ⋱ ⋱ f ′ ( a ) 1 ! f ( a ) ) f(J)=\left.\sum_{m=1}^{\infty}a_{m}J^{m}=\left(\begin{array}{cccccc}f(a)&\dfrac{f^{\prime}(a)}{1!}&\dfrac{f^{\prime\prime}(a)}{2!}&\cdots&\dfrac{f^{(p-1)}(a)}{(p-1)!}\\&f(a)&&&\\&&\ddots&\ddots&\\&&&\ddots&\dfrac{f^{\prime}(a)}{1!}\\&&&&f(a)\end{array}\right.\right) f(J)=∑m=1∞amJm= f(a)1!f′(a)f(a)2!f′′(a)⋱⋯⋱⋱(p−1)!f(p−1)(a)1!f′(a)f(a)
证 记 H = J − a E H=J-aE H=J−aE, 则 J m = ( H + a E ) m = ∑ i = 0 m C m i a m − i H i = ∑ i = 0 ∞ C m i a m − i H i ( 当 i > m y , C m i = 0 ) J^{m}=(H+aE)^{m}=\sum_{i=0}^{m}C_{m}^{i}a^{m-i}H^{i} =\sum_{i=0}^{\infty}C_{m}^{i}a^{m-i}H^{i}(\text{当}i>my,C_{m}^{i}=0) Jm=(H+aE)m=∑i=0mCmiam−iHi=∑i=0∞Cmiam−iHi(当i>my,Cmi=0)
~
∑ m = 0 ∞ a m J m = ∑ m = 0 ∞ a m ∑ i = 0 ∞ C m i a m − i H i = ∑ i = 0 ∞ ( ∑ m = 0 ∞ a m C m i a m − i ) H i = ∑ i = 0 ∞ 1 i ! H i ( ∑ m = 0 ∞ a m m ( m + 1 ) ⋯ ( m − i + 1 ) a m − i ) = ∑ i = 0 ∞ 1 i ! H i f ( i ) ( a ) i ≥ p 时 H i = 0 \qquad\qquad\qquad\qquad \begin{aligned}\sum_{m=0}^{\infty}a_{m}J^{m}&=\sum_{m=0}^{\infty}a_{m}\sum_{i=0}^{\infty}C_{m}^ia^{m-i}H^{i}=\sum_{i=0}^{\infty}(\sum_{m=0}^{\infty}a_{m}C_{m}^{i}a^{m-i})H^{i}\\ &=\sum_{i=0}^{\infty}\frac{1}{i!}H^{i}(\sum_{m=0}^{\infty}a_{m}m(m+1)\cdots(m-i+1)a^{m-i})\\&=\sum_{i=0}^{\infty}\frac{1}{i!}H^{i}f^{(i)}(a)\quad i\geq p~\text{时}~H^{i}=0 \end{aligned} m=0∑∞amJm=m=0∑∞ami=0∑∞Cmiam−iHi=i=0∑∞(m=0∑∞amCmiam−i)Hi=i=0∑∞i!1Hi(m=0∑∞amm(m+1)⋯(m−i+1)am−i)=i=0∑∞i!1Hif(i)(a)i≥p 时 Hi=0
可以发现这与若当块的第一种定义矩阵函数是一样的。所以我们可以得到如下推论。
推论 1.1.9 当函数 f ( z ) f(z) f(z) 和矩阵 A A A 满足第二种定义时 f ( A ) f(A) f(A) 的两种定义一致.
接下来了解一些 矩阵函数的性质
- 若 f ( λ ) = g ( λ ) ± h ( λ ) f(\lambda)=g(\lambda)\pm h(\lambda) f(λ)=g(λ)±h(λ), 则 f ( A ) = g ( A ) ± h ( A ) f(A)=g(A)\pm h(A) f(A)=g(A)±h(A).
- 若 f ( λ ) = g ( λ ) ⋅ h ( λ ) f(\lambda)=g(\lambda) · h(\lambda) f(λ)=g(λ)⋅h(λ), 则 f ( A ) = g ( A ) ⋅ h ( A ) f(A)=g(A)· h(A) f(A)=g(A)⋅h(A).
例 1.1.10 对于任意方阵 A A A, 有 cos 2 A + sin 2 A = E , sin 2 A = 2 sin A cos A \cos^2A+\sin^2A=E , \sin2A=2\sin A\cos A cos2A+sin2A=E,sin2A=2sinAcosA
命题 1.1.11 设以下所有的函数为满足定义所需的各种可导条件, 设 g ( λ ) = h ( f ( λ ) ) g(\lambda)=h(f(\lambda)) g(λ)=h(f(λ)) , 则 g ( A ) = h ( f ( A ) ) g(A)=h(f(A)) g(A)=h(f(A)).
引理 1.1.12 复合函数
f
(
g
(
x
)
)
f(g(x))
f(g(x)) 的高阶导数必为
∑
i
=
1
m
f
(
i
)
(
g
(
x
)
)
r
i
(
x
)
\sum_{i=1}^{m}f^{(i)}(g(x))r_{i(x)}
∑i=1mf(i)(g(x))ri(x) 形式
所有
r
0
(
x
)
,
r
1
(
x
)
,
⋯
,
r
m
(
x
)
r_0(x),r_1(x),\cdots,r_m(x)
r0(x),r1(x),⋯,rm(x) 均为
g
(
x
)
,
g
′
(
x
)
,
⋯
,
g
(
m
)
(
x
)
g(x),g^{\prime}(x),\cdots,g^{(m)}(x)
g(x),g′(x),⋯,g(m)(x) 的多元多项式。
例 1.1.13 对任意方阵
B
B
B , 有
e
B
e
−
B
=
E
e^Be^{-B}=E
eBe−B=E .
~
1.2 矩阵函数的初等因子
定义 N S ( A − λ 0 E ) j NS(A-\lambda_0E)^j NS(A−λ0E)j 为 ( A − λ 0 E ) j (A-\lambda_0E)^j (A−λ0E)j 的 n u l l s p a c e \mathrm{null~space} null space (零空间)
引理 1.2.1 设 λ 0 \lambda_0 λ0 是方阵 A A A 的一个特征值 , 记
-
d
j
=
dim
(
N
S
(
A
−
λ
0
E
)
j
)
=
A
d_j=\dim (NS(A-\lambda_0E)^j)=A
dj=dim(NS(A−λ0E)j)=A 的阶数
−
r
a
n
k
(
A
−
λ
0
E
)
j
-\mathrm{rank}\left(A-\lambda_{0}E\right)^{j}
−rank(A−λ0E)j
(
j
=
0
,
1
,
2
)
(j=0,1,2)
(j=0,1,2)
d 0 ≤ d 1 ≤ d 2 ≤ ⋯ ≤ d m = d m + 1 d_{0}\leq d_{1}\leq d_{2}\leq\cdots\leq d_{m}=d_{m+1} d0≤d1≤d2≤⋯≤dm=dm+1, A A A 可逆时 ≤ \leq ≤ 为 < < < ,其中 m m m 是关于 λ 0 \lambda_{0} λ0 的若当块的最大阶数。 - 记 A A A 的关于 λ 0 \lambda_{0} λ0 的 i i i 阶若当块(相当于初等因子 ( λ − λ 0 ) (\lambda-\lambda_{0}) (λ−λ0)) 的个数为 g i ( i = 1 , 2 , ⋯ ) g_i(i=1,2,\cdots) gi(i=1,2,⋯), (此处 g m ≥ 1 , g m + 1 = g m + 2 = ⋯ = 0 g_{m}\geq1,g_{m+1}=g_{m+2}=\cdots=0 gm≥1,gm+1=gm+2=⋯=0 )
则 g j = 2 d j − ( d j − 1 + d j + 1 ) g_{j}=2d_{j}-(d_{j-1}+d_{j+1}) gj=2dj−(dj−1+dj+1)。
证 设 J 0 = ( λ 0 1 ⋱ 1 λ 0 ) J_0=\begin{pmatrix}\lambda_0&1&\\&\ddots&1\\&&\lambda_0\end{pmatrix} J0= λ01⋱1λ0 是 λ 0 \lambda_0 λ0 的一个 i i i 阶若当块,
\qquad\qquad 则 dim ( N S ( J 0 − λ 0 E ) j ) = { i j ≥ i i − ( i − j ) j < i = min { i , j } \dim(NS(J_0-\lambda_0E)^j)=\begin{cases}i&j\geq i\\i-(i-j)&j<i\end{cases}=\min\{i,j\} dim(NS(J0−λ0E)j)={ii−(i−j)j≥ij<i=min{i,j}
于是 d j = dim ( N S ( A − λ 0 E ) j ) = ∑ i = 1 ∞ g i ⋅ min { i , j } = ∑ i = 1 j i g i + j ∑ i = j + 1 m g i \begin{aligned} \text{于是}~d_{j}&=\dim(NS(A-\lambda_{0}E)^{j}) =\sum_{i=1}^{\infty}g_{i}\cdot\min\{i,j\}=\sum_{i=1}^{j}ig_{i}+ j\sum_{i=j+1}^{m}g_{i}\end{aligned} 于是 dj=dim(NS(A−λ0E)j)=i=1∑∞gi⋅min{i,j}=i=1∑jigi+ji=j+1∑mgi即 d 1 = g 1 + g 2 + ⋯ + g m d 2 = g 1 + 2 g 2 + ⋯ + 2 g m d 3 = g 1 + 2 g 2 + 3 g 3 + ⋯ + 3 g m \begin{aligned}\text{即}~&d_{1}=g_{1}+g_{2}+\cdots+g_{m}\\&d_{2}=g_{1}+2g_{2}+\cdots+2g_{m}\\&d_{3}=g_{1}+2g_{2}+3g_{3}+\cdots+3g_{m}\end{aligned} 即 d1=g1+g2+⋯+gmd2=g1+2g2+⋯+2gmd3=g1+2g2+3g3+⋯+3gm
引理 1.2.2 设矩阵 B = ( a 0 a 1 … a p − 1 a 0 ⋱ ⋱ a 1 a 0 ) p × p B=\begin{pmatrix}a_0&a_1&\dots&a_{p-1}\\&a_0&\ddots&\\&&\ddots&a_1\\&&&a_0\end{pmatrix}_{p\times p} B= a0a1a0…⋱⋱ap−1a1a0 p×p 其中 a 1 = ⋯ = a k − 1 = 0 a_1=\dots=a_{k-1}=0 a1=⋯=ak−1=0 而 a k ≠ 0 ( 1 ≤ k ≤ p − 1 ) a_k\neq 0(1\leq k\leq p-1) ak=0(1≤k≤p−1) . 记 p = q k + h ( 1 ≤ h ≤ k − 1 ) p=qk+h(1\leq h\leq k-1) p=qk+h(1≤h≤k−1),则 B B B 的若当标准形由 k − h k-h k−h 个 q q q 阶若当块和 h h h 个 q − 1 q-1 q−1 阶若当块.
证 记 H = ( 0 1 ⋱ ⋱ 1 0 ) H=\begin{pmatrix}0&1&&\\&&\ddots&\\&&\ddots&1\\&&&0\end{pmatrix} H= 01⋱⋱10 , B = a 0 H 0 + a 1 H 1 + ⋯ + a p − 1 H p − 1 . B=a_{0}H^{0}+a_{1}H^{1}+\cdots+a_{p-1}H^{p-1}. B=a0H0+a1H1+⋯+ap−1Hp−1.
B − a 0 E = a k H k + a k + 1 H k + 1 + ⋯ + a p − 1 H p − 1 . B-a_{0}E=a_{k}H^{k}+a_{k+1}H^{k+1}+\cdots+a_{p-1}H^{p-1}. B−a0E=akHk+ak+1Hk+1+⋯+ap−1Hp−1.
⇒ ( B − a 0 E ) 2 = a k j H k j + { H \Rightarrow(B-a_0E)^2=a_k^jH^{kj}+\{H ⇒(B−a0E)2=akjHkj+{H 的一些高次幂 } \} }
⇒ d j = min { k j , p } = { k j 0 ≤ j ≤ q p j ≥ q + 1 \Rightarrow d_j=\min\left\{k{j},p\right\}=\begin{cases}kj&0\leq j\leq q\\p&j\geq q+1\end{cases} ⇒dj=min{kj,p}={kjp0≤j≤qj≥q+1
( i ) (i)~ (i) 当 j ≠ q , j ≠ q + 1 j\neq q,j\neq q+1 j=q,j=q+1时, d j − 1 , d j , d j + 1 d_{j-1},d_j,d_{j+1} dj−1,dj,dj+1 成等差数列时 g j = 0 g_j=0 gj=0 .
~
( i i ) 当 j = q 时 g j = 2 d q ⋅ ( d q − 1 + d q + 1 ) = 2 k q − [ k ( q − 1 ) + p ] = ( q + 1 ) k − p = ( q + 1 ) k − q k + h = q − h \begin{aligned}\text (ii) 当~j=q~时~~ g_{j}=2d_q\cdot(d_{q-1}+d_{q+1})&=2kq-[k(q-1)+p]\\&=(q+1)k-p=(q+1)k-qk+h\\&=q-h\end{aligned} (ii)当 j=q 时 gj=2dq⋅(dq−1+dq+1)=2kq−[k(q−1)+p]=(q+1)k−p=(q+1)k−qk+h=q−h
( i i i ) (iii) (iii)当 j = q + 1 j=q+1 j=q+1 时, g j = 2 p − ( k q + p ) = p − k q = h g_j=2p-(kq+p)=p-kq=h gj=2p−(kq+p)=p−kq=h
f ′ ( a ) = ⋯ = f ( k − 1 ) ( a ) = 0 , f ( k ) ( a 0 ) ≠ 0 f^{\prime}(a)=\cdots=f^{(k-1)}(a)=0,f^{(k)}(a_0)\neq0 f′(a)=⋯=f(k−1)(a)=0,f(k)(a0)=0 记 p = q k + h p=qk+h p=qk+h , 则矩阵函数 f ( J ) f(J) f(J) 的若当标准形由关于 f ( a ) f(a) f(a) 的 k − h k-h k−h 个 q q q 阶若当块和 h h h 个 q + 1 q+1 q+1 阶若当块组成。
1.3 矩阵分析
1.3.1 矩阵序列
定义 1.3.1 (极限) A m = ( a i j ( m ) ) A_{m}=(a_{ij}^{(m)}) Am=(aij(m)) ,若 lim m → ∞ a i j ( m ) = a i j ( ∀ i , j ) \lim_{m\to\infty}a_{ij}^{(m)}=a_{ij}(\forall i,j) limm→∞aij(m)=aij(∀i,j) , 则 lim m → ∞ A m = ( a i j ) \lim_{m\to\infty}A_{m}=(a_{ij}) limm→∞Am=(aij).
\qquad 例如, A m = ( ( 1 2 ) m 0 0 ( 1 3 ) m ) , lim m → ∞ A m = ( 0 0 0 0 ) A_{m}=\begin{pmatrix}(\frac{1}{2})^{m}&0\\0&(\frac{1}{3})^{m}\end{pmatrix},\lim_{m\to\infty}A_{m}=\begin{pmatrix}0&0\\0&0\end{pmatrix} Am=((21)m00(31)m),limm→∞Am=(0000)
\qquad
性质 1
(
1
)
lim
n
→
∞
(
A
m
±
B
m
)
=
lim
m
→
∞
A
m
±
lim
m
→
∞
B
m
(
2
)
lim
m
→
∞
(
A
m
⋅
B
m
)
=
lim
m
→
∞
A
m
⋅
lim
m
→
∞
B
m
(
3
)
lim
m
→
∞
(
A
m
⊕
B
m
)
=
lim
m
→
∞
A
m
⊕
lim
m
→
∞
B
m
~~\begin{aligned}(1)&\lim_{n\to\infty}(A_{m}\pm B_{m})=\lim_{m\to\infty}A_{m}\pm\lim_{m\to\infty}B_{m}\\(2)&\lim_{m\to\infty}(A_{m}\cdot B_{m})=\lim_{m\to\infty}A_{m}\cdot\lim_{m\to\infty}B_{m}\\(3)&\lim_{m\to\infty}(A_{m}\oplus B_{m})=\lim_{m\to\infty}A_{m}\oplus\lim_{m\to\infty}B_{m}\end{aligned}
(1)(2)(3)n→∞lim(Am±Bm)=m→∞limAm±m→∞limBmm→∞lim(Am⋅Bm)=m→∞limAm⋅m→∞limBmm→∞lim(Am⊕Bm)=m→∞limAm⊕m→∞limBm
\qquad
性质 2 (同步相似序列的极限) 若有
A
m
=
C
−
1
B
m
C
A_{m}=C^{-1}B_{m}C
Am=C−1BmC ,(
C
C
C 与
m
m
m无关
)
)
),
\qquad
\qquad
\qquad
\qquad
\qquad
\quad
\qquad
则
lim
m
→
∞
A
m
=
C
−
1
(
lim
m
→
∞
B
m
)
C
\lim_{m\to\infty}A_{m}=C^{-1}(\lim_{m\to\infty}B_{m})C
limm→∞Am=C−1(limm→∞Bm)C
定理 1.3.2 (若当块的幂次序列的极限) 设 J = ( a 1 a ⋱ ⋱ 1 a ) p × p ( p ⩾ 2 ) J=\begin{pmatrix}a&1&&\\&a&\ddots&\\&&\ddots&1\\&&&a\end{pmatrix}_{p\times p}(p\geqslant 2) J= a1a⋱⋱1a p×p(p⩾2) , 则 lim m → ∞ J m \lim_{m\to\infty}J^m limm→∞Jm 存在 ⇔ ∣ a ∣ < 1 \Leftrightarrow |a|<1 ⇔∣a∣<1 且极限存在时极限为 0 0 0.
证: J = a E + H ⇒ J m = ∑ i = 0 m C m i a m − i H i J=aE+H\Rightarrow J^m=\sum_{i=0}^mC_m^{i}a^{m-i}H^i J=aE+H⇒Jm=∑i=0mCmiam−iHi
~
= ( a m C m 1 a m − 1 C m 2 a m − 2 ⋯ C m p − 1 a m − p − 1 a m ⋱ ⋱ C m 1 a m − 1 a m − 1 ) =\begin{array}{ccccccc}&\left(\begin{array}{ccccccc} a^m&C_m^1a^{m-1}&C_m^2a^{m-2}&\cdots& C_m^{p-1}a^{m-p-1}\\ &a^m&&&&\\&&&&\\&&\ddots&\ddots&\\ &&&&C_m^1a^{m-1}\\&&&&a^{m-1}\end{array}\right)\end{array} = amCm1am−1amCm2am−2⋱⋯⋱Cmp−1am−p−1Cm1am−1am−1
推论 1.3.3 A A A 是方阵,则 lim m → ∞ A m \lim_{m\to\infty}A^m limm→∞Am 存在的充要条件:
- A A A 的非一特征值的绝对值均小于 1 1 1.
- A A A 的相应于 1 1 1 的若当块(若有的话)都是 1 1 1 阶的,极限存在的时候 lim m → ∞ A m \lim_{m\to\infty}A^m limm→∞Am 必是 C − 1 D C C^{-1}DC C−1DC , D D D 是 ( 0.1 ) (0.1) (0.1) 对角阵
1.3.2 矩阵级数
定义 1.3.4 ∑ m = 0 ∞ A m = lim N → ∞ ( ∑ m = 0 N A m ) \sum_{m=0}^{\infty}A_{m}=\lim_{N\to\infty}(\sum_{m=0}^{N}A_{m}) ∑m=0∞Am=limN→∞(∑m=0NAm) 极限存在时称级数为收敛。
性质 1.3.5 (1) 若
A
m
=
C
−
1
B
m
C
A_{m}=C^{-1}B_{m}C
Am=C−1BmC 对所有
m
m
m 成立(
m
m
m 与
C
C
C 无关),
则
∑
m
=
0
∞
A
m
=
C
−
1
∑
m
=
0
∞
B
m
C
\sum_{m=0}^{\infty}A_{m}=C^{-1}\sum_{m=0}^{\infty}B_{m}C
∑m=0∞Am=C−1∑m=0∞BmC.
(2)若 A m = B m ⊗ C m A_{m}=B_{m}\otimes C_{m} Am=Bm⊗Cm,则 ∑ m = 0 ∞ A m = ∑ m = 0 ∞ B m ⊕ ∑ m = 0 ∞ C m \sum_{m=0}^{\infty}A_{m}=\sum_{m=0}^{\infty}B_{m}\oplus\sum_{m=0}^{\infty}C_{m} ∑m=0∞Am=∑m=0∞Bm⊕∑m=0∞Cm .