特征值问题
定义5.1:设方阵 A n \bold A_n An,若存在数 λ \lambda λ与非零向量 X ⃗ \vec{X} X,使得 A X ⃗ = λ X ⃗ A\vec{X}=\lambda\vec{X} AX=λX则称 λ \lambda λ为方阵 A \bold A A的 特征值, X ⃗ \vec{X} X为方阵 A \bold A A对应于 λ \lambda λ的 特征向量。
Remark:不可逆矩阵必有0特征值.
证明:若 A A A不可逆,则 A X ⃗ = 0 A\vec{X}=0 AX=0存在非零解 X ⃗ 0 \vec{X}_0 X0,故 A X ⃗ 0 = 0 = 0 X ⃗ 0 A\vec{X}_0=0=0\vec{X}_0 AX0=0=0X0
特征值与特征向量的计算:
A
X
⃗
=
λ
X
⃗
⟺
(
A
−
λ
E
)
X
⃗
=
0
(
1
)
A\vec{X}=\lambda\vec{X}\Longleftrightarrow (A-\lambda E)\vec{X}=0\qquad (1)
AX=λX⟺(A−λE)X=0(1)
若(1)有非零解,则
f
(
λ
)
=
d
e
t
(
A
−
λ
E
)
=
0
f(\lambda)=det(A-\lambda E)=0
f(λ)=det(A−λE)=0,将
f
(
λ
)
f(\lambda)
f(λ)称作 A的特征多项式,求解出复数域的
n
n
n个特征值后则可反代入
(
1
)
(1)
(1)求解得到对应的特征向量。
关于特征多项式的展开:
f ( λ ) = d e t ( A − λ E ) = ∣ a 11 − λ a 12 a 13 a 14 … a 1 n a 21 a 22 − λ a 23 a 24 … a 1 n a 31 a 32 a 33 − λ a 34 … a 1 n a 41 a 42 a 43 a 44 − λ … a 1 n ⋮ ⋮ ⋮ ⋮ ⋮ a n 1 a n 2 a n 3 a n 4 … a n n − λ ∣ = ∣ a 11 a 12 a 13 a 14 … a 1 n a 21 a 22 − λ a 23 a 24 … a 1 n a 31 a 32 a 33 − λ a 34 … a 1 n a 41 a 42 a 43 a 44 − λ … a 1 n ⋮ ⋮ ⋮ ⋮ ⋮ a n 1 a n 2 a n 3 a n 4 … a n n − λ ∣ + ∣ − λ a 12 a 13 a 14 … a 1 n 0 a 22 − λ a 23 a 24 … a 1 n 0 a 32 a 33 − λ a 34 … a 1 n 0 a 42 a 43 a 44 − λ … a 1 n ⋮ ⋮ ⋮ ⋮ ⋮ 0 a n 2 a n 3 a n 4 … a n n − λ ∣ = ∣ a 11 a 12 a 13 a 14 … a 1 n a 21 a 22 a 23 a 24 … a 1 n a 31 a 32 a 33 − λ a 34 … a 1 n a 41 a 42 a 43 a 44 − λ … a 1 n ⋮ ⋮ ⋮ ⋮ ⋮ a n 1 a n 2 a n 3 a n 4 … a n n − λ ∣ + ∣ a 11 0 a 13 a 14 … a 1 n a 21 − λ a 23 a 24 … a 1 n a 31 0 a 33 − λ a 34 … a 1 n a 41 0 a 43 a 44 − λ … a 1 n ⋮ ⋮ ⋮ ⋮ ⋮ a n 1 0 a n 3 a n 4 … a n n − λ ∣ + ∣ − λ a 12 a 13 a 14 … a 1 n 0 a 22 a 23 a 24 … a 1 n 0 a 32 a 33 − λ a 34 … a 1 n 0 a 42 a 43 a 44 − λ … a 1 n ⋮ ⋮ ⋮ ⋮ ⋮ 0 a n 2 a n 3 a n 4 … a n n − λ ∣ + ∣ − λ 0 a 13 a 14 … a 1 n 0 − λ a 23 a 24 … a 1 n 0 0 a 33 − λ a 34 … a 1 n 0 0 a 43 a 44 − λ … a 1 n ⋮ ⋮ ⋮ ⋮ ⋮ 0 0 a n 3 a n 4 … a n n − λ ∣ 记符号 A ( n ) a i ↔ e i m 表示将 n 阶方阵的任意 m 个列向量 a i 替换为对应的单位向量 e i 后得到的 C n m 个矩阵,如 A ( 2 ) a i ↔ e i 1 = { [ 1 a 1 2 0 a 22 ] 、 [ a 11 0 a 21 1 ] } , 而 d e t ( A ( n ) a i ↔ e i m ) 表示这一组方阵行列式之和,那么: f ( λ ) = ∑ m = 0 n d e t ( A ( n ) a i ↔ λ e i m ) = ∑ m = 0 n ∑ l = 1 C n m ( − λ ) m P ( n − m ) l = ∑ m = 0 n ( − λ ) m ∑ l = 1 C n m P ( n − m ) l ( 说明:求解 d e t ( A ( n ) a i ↔ λ e i m ) 时按照 λ e i 所在的列展开, P ( n − m ) l 为一个 n − m 阶主子式 ) f(\lambda)=det(A-\lambda E)= \begin{vmatrix} %1 a_{11}-\lambda&a_{12}&a_{13}&a_{14}&\dots&a_{1n}\\ a_{21}&a_{22}-\lambda&a_{23}&a_{24}&\dots&a_{1n}\\ a_{31}&a_{32}&a_{33}-\lambda&a_{34}&\dots&a_{1n}\\ a_{41}&a_{42}&a_{43}&a_{44}-\lambda&\dots&a_{1n}\\ \vdots&\vdots&\vdots&\vdots&&\vdots\\ a_{n1}&a_{n2}&a_{n3}&a_{n4}&\dots&a_{nn}-\lambda\\ \end{vmatrix}= \begin{vmatrix} %2.1 a_{11}&a_{12}&a_{13}&a_{14}&\dots&a_{1n}\\ a_{21}&a_{22}-\lambda&a_{23}&a_{24}&\dots&a_{1n}\\ a_{31}&a_{32}&a_{33}-\lambda&a_{34}&\dots&a_{1n}\\ a_{41}&a_{42}&a_{43}&a_{44}-\lambda&\dots&a_{1n}\\ \vdots&\vdots&\vdots&\vdots&&\vdots\\ a_{n1}&a_{n2}&a_{n3}&a_{n4}&\dots&a_{nn}-\lambda\\ \end{vmatrix}+ \begin{vmatrix}%2.2 -\lambda&a_{12}&a_{13}&a_{14}&\dots&a_{1n}\\ 0&a_{22}-\lambda&a_{23}&a_{24}&\dots&a_{1n}\\ 0&a_{32}&a_{33}-\lambda&a_{34}&\dots&a_{1n}\\ 0&a_{42}&a_{43}&a_{44}-\lambda&\dots&a_{1n}\\ \vdots&\vdots&\vdots&\vdots&&\vdots\\ 0&a_{n2}&a_{n3}&a_{n4}&\dots&a_{nn}-\lambda\\ \end{vmatrix}\\ \qquad\\= \begin{vmatrix}%3.1 a_{11}&a_{12}&a_{13}&a_{14}&\dots&a_{1n}\\ a_{21}&a_{22}&a_{23}&a_{24}&\dots&a_{1n}\\ a_{31}&a_{32}&a_{33}-\lambda&a_{34}&\dots&a_{1n}\\ a_{41}&a_{42}&a_{43}&a_{44}-\lambda&\dots&a_{1n}\\ \vdots&\vdots&\vdots&\vdots&&\vdots\\ a_{n1}&a_{n2}&a_{n3}&a_{n4}&\dots&a_{nn}-\lambda\\ \end{vmatrix}+ \begin{vmatrix}%3.2 a_{11}&0&a_{13}&a_{14}&\dots&a_{1n}\\ a_{21}&-\lambda&a_{23}&a_{24}&\dots&a_{1n}\\ a_{31}&0&a_{33}-\lambda&a_{34}&\dots&a_{1n}\\ a_{41}&0&a_{43}&a_{44}-\lambda&\dots&a_{1n}\\ \vdots&\vdots&\vdots&\vdots&&\vdots\\ a_{n1}&0&a_{n3}&a_{n4}&\dots&a_{nn}-\lambda \end{vmatrix}+ \begin{vmatrix}%3.3 -\lambda&a_{12}&a_{13}&a_{14}&\dots&a_{1n}\\ 0&a_{22}&a_{23}&a_{24}&\dots&a_{1n}\\ 0&a_{32}&a_{33}-\lambda&a_{34}&\dots&a_{1n}\\ 0&a_{42}&a_{43}&a_{44}-\lambda&\dots&a_{1n}\\ \vdots&\vdots&\vdots&\vdots&&\vdots\\ 0&a_{n2}&a_{n3}&a_{n4}&\dots&a_{nn}-\lambda\\ \end{vmatrix}+ \begin{vmatrix}%3.4 -\lambda&0&a_{13}&a_{14}&\dots&a_{1n}\\ 0&-\lambda&a_{23}&a_{24}&\dots&a_{1n}\\ 0&0&a_{33}-\lambda&a_{34}&\dots&a_{1n}\\ 0&0&a_{43}&a_{44}-\lambda&\dots&a_{1n}\\ \vdots&\vdots&\vdots&\vdots&&\vdots\\ 0&0&a_{n3}&a_{n4}&\dots&a_{nn}-\lambda\\ \end{vmatrix}\\ \quad\\ 记符号A_{(n)a_i\leftrightarrow e_i}^{m}表示将n阶方阵的任意m个列向量a_i替换为对应的单位向量e_i后得到的C_n^m个矩阵,如A_{(2)a_i\leftrightarrow e_i}^1=\{\begin{bmatrix}1&a_12\\0&a_{22}\end{bmatrix}、\begin{bmatrix}a_{11}&0\\a_{21}&1\end{bmatrix}\},而det(A_{(n)a_i\leftrightarrow e_i}^m)表示这一组方阵行列式之和,那么:\\\quad\\ f(\lambda)=\sum_{m=0}^ndet(A_{(n)a_i\leftrightarrow \lambda e_i}^m)=\sum_{m=0}^n\sum_{l=1}^{C_n^m}(-\lambda)^mP^l_{(n-m)}=\sum_{m=0}^n(-\lambda)^m\sum_{l=1}^{C_n^m}P^l_{(n-m)}(说明:求解det(A_{(n)a_i\leftrightarrow \lambda e_i}^m)时按照\lambda e_i所在的列展开,P^l_{(n-m)}为一个n-m阶主子式)\\ \quad f(λ)=det(A−λE)=∣ ∣a11−λa21a31a41⋮an1a12a22−λa32a42⋮an2a13a23a33−λa43⋮an3a14a24a34a44−λ⋮an4……………a1na1na1na1n⋮ann−λ∣ ∣=∣ ∣a11a21a31a41⋮an1a12a22−λa32a42⋮an2a13a23a33−λa43⋮an3a14a24a34a44−λ⋮an4……………a1na1na1na1n⋮ann−λ∣ ∣+∣ ∣−λ000⋮0a12a22−λa32a42⋮an2a13a23a33−λa43⋮an3a14a24a34a44−λ⋮an4……………a1na1na1na1n⋮ann−λ∣ ∣=∣ ∣a11a21a31a41⋮an1a12a22a32a42⋮an2a13a23a33−λa43⋮an3a14a24a34a44−λ⋮an4……………a1na1na1na1n⋮ann−λ∣ ∣+∣ ∣a11a21a31a41⋮an10−λ00⋮0a13a23a33−λa43⋮an3a14a24a34a44−λ⋮an4……………a1na1na1na1n⋮ann−λ∣ ∣+∣ ∣−λ000⋮0a12a22a32a42⋮an2a13a23a33−λa43⋮an3a14a24a34a44−λ⋮an4……………a1na1na1na1n⋮ann−λ∣ ∣+∣ ∣−λ000⋮00−λ00⋮0a13a23a33−λa43⋮an3a14a24a34a44−λ⋮an4……………a1na1na1na1n⋮ann−λ∣ ∣记符号A(n)ai↔eim表示将n阶方阵的任意m个列向量ai替换为对应的单位向量ei后得到的Cnm个矩阵,如A(2)ai↔ei1={[10a12a22]、[a11a2101]},而det(A(n)ai↔eim)表示这一组方阵行列式之和,那么:f(λ)=m=0∑ndet(A(n)ai↔λeim)=m=0∑nl=1∑Cnm(−λ)mP(n−m)l=m=0∑n(−λ)ml=1∑CnmP(n−m)l(说明:求解det(A(n)ai↔λeim)时按照λei所在的列展开,P(n−m)l为一个n−m阶主子式)
上式说明特征多项式 ( − λ ) m (-\lambda)^m (−λ)m项的系数为 n n n阶方阵 A A A所有 ( n − m ) (n-m) (n−m)阶主子式的和
定理5.1:设
n
n
n阶方阵的
n
n
n个特征值为:
λ
i
(
i
=
1
,
2
,
…
,
n
)
\lambda_i\quad(i=1,2,\dots,n)
λi(i=1,2,…,n),则:
(1)
d
e
t
(
A
)
=
λ
1
λ
2
∙
⋯
∙
λ
n
det(A)=\lambda_1\lambda_2\bullet\dots\bullet\lambda_n
det(A)=λ1λ2∙⋯∙λn
(2)
λ
1
+
λ
2
+
⋯
+
λ
n
=
a
11
+
a
22
+
⋯
+
a
n
n
=
t
r
(
A
)
\lambda_1+\lambda_2+\dots+\lambda_n=a_{11}+a_{22}+\dots+a_{nn}=tr(A)
λ1+λ2+⋯+λn=a11+a22+⋯+ann=tr(A)
(1)证明:
d
e
t
(
A
−
λ
E
)
=
(
−
λ
+
λ
1
)
(
−
λ
+
λ
2
)
∙
⋯
∙
(
−
λ
+
λ
n
)
=
(
−
1
)
n
λ
n
+
(
−
1
)
n
−
1
(
λ
1
+
λ
2
+
⋯
+
λ
n
)
λ
n
−
1
+
⋯
+
λ
1
λ
2
∙
⋯
∙
λ
n
det(A-\lambda E)=(-\lambda+\lambda_1)(-\lambda+\lambda_2)\bullet\dots\bullet(-\lambda+\lambda_n)=(-1)^n\lambda^n+(-1)^{n-1}(\lambda_1+\lambda_2+\dots+\lambda_n)\lambda^{n-1}+\dots+\lambda_1\lambda_2\bullet\dots\bullet\lambda_n
det(A−λE)=(−λ+λ1)(−λ+λ2)∙⋯∙(−λ+λn)=(−1)nλn+(−1)n−1(λ1+λ2+⋯+λn)λn−1+⋯+λ1λ2∙⋯∙λn
令
λ
=
0
\lambda=0
λ=0,则
d
e
t
(
A
)
=
λ
1
λ
2
∙
⋯
∙
λ
n
det(A)=\lambda_1\lambda_2\bullet\dots\bullet\lambda_n
det(A)=λ1λ2∙⋯∙λn(证毕)
(2)证明:
由(1)知特征值之和为
(
−
λ
)
n
−
1
(-\lambda)^{n-1}
(−λ)n−1的系数,应等于所有一阶主子式之和,即对角元素之和,那么:
λ
1
+
λ
2
+
⋯
+
λ
n
=
a
11
+
a
22
+
⋯
+
a
n
n
(证毕)
\lambda_1+\lambda_2+\dots+\lambda_n=a_{11}+a_{22}+\dots+a_{nn}(证毕)
λ1+λ2+⋯+λn=a11+a22+⋯+ann(证毕)
推论5.1:方阵 A n A_n An可逆 ⟺ \Longleftrightarrow ⟺其特征值 λ i ≠ 0 ( i = 1 , 2 , … , n ) \lambda_i\ne0\quad(i=1,2,\dots,n) λi=0(i=1,2,…,n)
定理5.2:方阵
A
A
A的特征值为
λ
\lambda
λ,特征向量为
X
⃗
\vec{X}
X,则
(1)
A
−
1
A^{-1}
A−1的特征值为
λ
−
1
\lambda^{-1}
λ−1,特征向量为
X
⃗
\vec{X}
X;
(2)矩阵多项式
g
(
A
)
=
a
k
A
k
+
a
k
−
1
A
k
−
1
+
⋯
+
a
1
A
+
a
0
E
g(A)=a_kA^k+a_{k-1}A^{k-1}+\dots+a_1A+a_0E
g(A)=akAk+ak−1Ak−1+⋯+a1A+a0E 的特征值为
g
(
λ
)
=
a
k
λ
k
+
a
k
−
1
λ
k
−
1
+
⋯
+
a
1
λ
+
a
0
g(\lambda)=a_k\lambda^k+a_{k-1}\lambda^{k-1}+\dots+a_1\lambda+a_0
g(λ)=akλk+ak−1λk−1+⋯+a1λ+a0,特征向量为
X
⃗
\vec{X}
X
(1)证明:
A X ⃗ = λ X ⃗ ⟹ X ⃗ = A − 1 λ X ⃗ ⟹ A − 1 X ⃗ = λ − 1 X ⃗ \qquad A\vec{X}=\lambda\vec{X}\Longrightarrow \vec{X}=A^{-1}\lambda\vec{X}\Longrightarrow A^{-1}\vec{X}=\lambda^{-1}\vec{X} AX=λX⟹X=A−1λX⟹A−1X=λ−1X(证毕)
(2)证明:
g ( A ) X ⃗ = a k A k X ⃗ + a k − 1 A k − 1 X ⃗ + ⋯ + a 1 A X ⃗ + a 0 X ⃗ \qquad g(A)\vec{X}=a_kA^k\vec{X}+a_{k-1}A^{k-1}\vec{X}+\dots+a_1A\vec{X}+a_0\vec{X} g(A)X=akAkX+ak−1Ak−1X+⋯+a1AX+a0X
= a k λ k X ⃗ + a k − 1 λ k − 1 X ⃗ + ⋯ + a 1 λ X ⃗ + a 0 X ⃗ = g ( λ ) X ⃗ \qquad =a_k\lambda^k\vec{X}+a_{k-1}\lambda^{k-1}\vec{X}+\dots+a_1\lambda\vec{X}+a_0\vec{X}=g(\lambda)\vec{X} =akλkX+ak−1λk−1X+⋯+a1λX+a0X=g(λ)X(证毕)
由于 λ i \lambda_i λi对应的特征向量集合是 ( A − λ i E ) X ⃗ = 0 (A-\lambda_i E)\vec{X}=0 (A−λiE)X=0的解空间,故 λ i \lambda_i λi对应的特征向量的线性组合仍然是 λ i \lambda_i λi对应的特征向量。
定理5.3:不同特征值对应的特征向量线性无关。
证明:采用数学归纳法:
\ \qquad
X
⃗
1
\vec{X}_1
X1与
X
⃗
2
\vec{X}_2
X2分别为
λ
1
\lambda_1
λ1与
λ
2
\lambda_2
λ2对应的特征向量
(
λ
1
≠
λ
2
)
(\lambda_1\ne \lambda_2)
(λ1=λ2)。
\ \qquad
令
k
1
X
⃗
1
+
k
2
X
⃗
2
=
0
(
1
)
k_1\vec{X}_1+k_2\vec{X}_2=0\qquad (1)
k1X1+k2X2=0(1)
\ \qquad
则
A
(
k
1
X
⃗
1
+
k
2
X
⃗
2
)
=
k
1
λ
1
X
⃗
1
+
k
2
λ
2
X
⃗
2
=
0
(
2
)
A(k_1\vec{X}_1+k_2\vec{X}_2)=k_1\lambda_1\vec{X}_1+k_2\lambda_2\vec{X}_2=0\qquad (2)
A(k1X1+k2X2)=k1λ1X1+k2λ2X2=0(2)
\ \qquad
那么,
(
1
)
×
λ
1
−
(
2
)
:
k
2
(
λ
1
−
λ
2
)
X
⃗
2
=
0
⟹
k
2
=
0
(1)\times\lambda_1-(2):k_2(\lambda_1-\lambda_2)\vec{X}_2=0\Longrightarrow k_2=0
(1)×λ1−(2):k2(λ1−λ2)X2=0⟹k2=0
\ \qquad\ \ \qquad
(
1
)
×
λ
2
−
(
2
)
:
k
1
(
λ
2
−
λ
1
)
X
⃗
2
=
0
⟹
k
1
=
0
(1)\times\lambda_2-(2):k_1(\lambda_2-\lambda_1)\vec{X}_2=0\Longrightarrow k_1=0
(1)×λ2−(2):k1(λ2−λ1)X2=0⟹k1=0
\ \qquad
即,
X
⃗
1
\vec{X}_1
X1与
X
⃗
2
\vec{X}_2
X2线性无关。
\ \qquad
假设
λ
1
,
λ
2
,
…
,
λ
r
−
1
\lambda_1,\lambda_2,\dots,\lambda_{r-1}
λ1,λ2,…,λr−1对应的特征向量
X
⃗
1
,
X
⃗
2
,
…
,
X
⃗
r
−
1
\vec{X}_1,\vec{X}_2,\dots,\vec{X}_{r-1}
X1,X2,…,Xr−1线性无关。
\ \qquad
讨论
X
⃗
1
,
X
⃗
2
,
…
,
X
⃗
r
−
1
,
X
⃗
r
\vec{X}_1,\vec{X}_2,\dots,\vec{X}_{r-1},\vec{X}_r
X1,X2,…,Xr−1,Xr的线性相关性。
\ \qquad
令
k
1
X
⃗
1
+
k
2
X
⃗
2
+
⋯
+
k
r
X
⃗
r
=
0
(
3
)
k_1\vec{X}_1+k_2\vec{X}_2+\dots+k_r\vec{X}_r=0\qquad (3)
k1X1+k2X2+⋯+krXr=0(3)
\ \qquad
则
A
(
k
1
X
⃗
1
+
k
2
X
⃗
2
+
⋯
+
k
r
X
⃗
r
)
=
k
1
λ
1
X
⃗
1
+
k
2
λ
2
X
⃗
2
+
⋯
+
k
r
λ
r
X
⃗
r
=
0
(
4
)
A(k_1\vec{X}_1+k_2\vec{X}_2+\dots+k_r\vec{X}_r)=k_1\lambda_1\vec{X}_1+k_2\lambda_2\vec{X}_2+\dots+k_r\lambda_r\vec{X}_r=0\qquad (4)
A(k1X1+k2X2+⋯+krXr)=k1λ1X1+k2λ2X2+⋯+krλrXr=0(4)
\ \qquad
那么,
(
3
)
×
λ
r
−
(
4
)
:
(3)\times\lambda_r-(4):
(3)×λr−(4):
k
1
(
λ
r
−
λ
1
)
X
⃗
1
+
k
2
(
λ
r
−
λ
2
)
X
⃗
2
+
⋯
+
k
r
−
1
(
λ
r
−
λ
r
−
1
)
X
⃗
r
−
1
=
0
k_1(\lambda_r-\lambda_1)\vec{X}_1+k_2(\lambda_r-\lambda_2)\vec{X}_2+\dots+k_{r-1}(\lambda_r-\lambda_{r-1})\vec{X}_{r-1}=0
k1(λr−λ1)X1+k2(λr−λ2)X2+⋯+kr−1(λr−λr−1)Xr−1=0
\ \qquad
由于
X
⃗
1
,
X
⃗
2
,
…
X
⃗
r
−
1
\vec{X}_1,\vec{X}_2,\dots\vec{X}_{r-1}
X1,X2,…Xr−1线性无关
\ \qquad
则
k
i
(
λ
r
−
λ
i
)
=
0
(
i
=
1
,
…
,
r
−
1
)
⟹
k
i
=
0
(
i
=
1
,
…
,
r
−
1
)
k_{i}(\lambda_r-\lambda_{i})=0\ (i=1,\dots,r-1)\Longrightarrow k_i=0\ (i=1,\dots,r-1)
ki(λr−λi)=0 (i=1,…,r−1)⟹ki=0 (i=1,…,r−1)
\ \qquad
带入式(3),
k
r
=
0
k_r=0
kr=0
\ \qquad
故,
X
⃗
1
,
X
⃗
2
,
…
,
X
⃗
r
−
1
,
X
⃗
r
\vec{X}_1,\vec{X}_2,\dots,\vec{X}_{r-1},\vec{X}_r
X1,X2,…,Xr−1,Xr线性无关。
定理5.4:将不同特征值对应的线性无关特征向量组进行组合得到的向量组依然线性无关。即,
设
λ
1
,
λ
2
,
…
,
λ
r
\lambda_1,\lambda_2,\dots,\lambda_{r}
λ1,λ2,…,λr为
A
A
A的不同特征值且
X
⃗
1
i
,
X
⃗
2
i
,
…
,
X
⃗
s
i
,
i
\vec{X}_{1i},\vec{X}_{2i},\dots,\vec{X}_{s_i,i}
X1i,X2i,…,Xsi,i为特征值
λ
i
\lambda_i
λi对应的
s
i
s_i
si个线性无关的特征向量,则
X
⃗
11
,
X
⃗
21
,
…
,
X
⃗
s
1
,
1
,
X
⃗
12
,
X
⃗
22
,
…
,
X
⃗
s
2
,
2
,
X
⃗
1
r
,
X
⃗
2
r
,
…
,
X
⃗
s
r
,
r
\vec{X}_{11},\vec{X}_{21},\dots,\vec{X}_{s_1,1},\vec{X}_{12},\vec{X}_{22},\dots,\vec{X}_{s_2,2},\vec{X}_{1r},\vec{X}_{2r},\dots,\vec{X}_{s_r,r}
X11,X21,…,Xs1,1,X12,X22,…,Xs2,2,X1r,X2r,…,Xsr,r向量组依然线性无关。
证明:令
∑
j
=
1
s
i
k
j
i
X
⃗
j
i
=
∑
j
=
1
s
1
k
j
1
X
⃗
j
1
+
∑
j
=
1
s
2
k
j
2
X
⃗
j
2
+
⋯
+
∑
j
=
1
s
r
k
j
r
X
⃗
j
r
=
Z
⃗
1
+
Z
⃗
2
+
⋯
+
Z
⃗
r
\sum_{j=1}^{s_i}k_{ji}\vec{X}_{ji}=\sum_{j=1}^{s_1}k_{j1}\vec{X}_{j1}+\sum_{j=1}^{s_2}k_{j2}\vec{X}_{j2}+\dots+\sum_{j=1}^{s_r}k_{jr}\vec{X}_{jr}=\vec{Z}_{1}+\vec{Z}_{2}+\dots+\vec{Z}_{r}
j=1∑sikjiXji=j=1∑s1kj1Xj1+j=1∑s2kj2Xj2+⋯+j=1∑srkjrXjr=Z1+Z2+⋯+Zr
由于
λ
i
\lambda_i
λi对应的特征向量的线性组合
Z
⃗
i
\vec{Z}_{i}
Zi仍然是
λ
i
\lambda_i
λi对应的特征向量,则
Z
⃗
1
+
Z
⃗
2
+
⋯
+
Z
⃗
r
=
0
\vec{Z}_{1}+\vec{Z}_{2}+\dots+\vec{Z}_{r}=0
Z1+Z2+⋯+Zr=0成立当且仅当
Z
⃗
i
=
0
(
i
=
1
,
…
,
r
)
\vec{Z}_{i}=0\ (i=1,\dots,r)
Zi=0 (i=1,…,r)。
又,选取的是同一特征值对应的线性无关特征向量,则
k
j
i
=
0
k_{ji}=0
kji=0.
故,将不同特征值对应的线性无关特征向量组进行组合得到的向量组依然线性无关。(证毕)
定理5.4:代数重数(特征多项式根
λ
\lambda
λ的重数)
≥
\ge
≥ 几何重数(
λ
\lambda
λ对应的特征向量构成的向量空间的维数)。
证明:过程参考出处
\ \qquad
设
λ
\lambda
λ为方阵
A
n
A_n
An的一个特征值,并假设其对应的特征向量构成的向量空
\ \qquad
间
V
V
V(
R
n
的子空间
)
R^n的子空间)
Rn的子空间)维数为
r
r
r,基为
{
α
⃗
1
,
α
⃗
2
,
…
,
α
⃗
r
}
\{\vec{\alpha}_1,\vec{\alpha}_2,\dots,\vec{\alpha}_r\}
{α1,α2,…,αr},将其扩充为
R
n
R^n
Rn中
\ \qquad
的一组基
{
α
⃗
1
,
α
⃗
2
,
…
,
α
⃗
r
,
α
⃗
r
+
1
,
.
.
.
,
,
α
⃗
n
}
\{\vec{\alpha}_1,\vec{\alpha}_2,\dots,\vec{\alpha}_r,\vec{\alpha}_{r+1},...,,\vec{\alpha}_n\}
{α1,α2,…,αr,αr+1,...,,αn}。
\ \qquad
则
{
A
α
⃗
i
=
λ
α
⃗
i
(
i
=
1
,
.
.
.
,
r
)
A
α
⃗
i
=
k
1
i
α
⃗
1
+
k
2
i
α
⃗
2
+
.
.
.
+
k
n
i
α
⃗
n
(
i
=
r
+
1
,
.
.
.
,
n
)
\begin{cases}A\vec{\alpha}_i=\lambda\vec{\alpha}_i\qquad (i=1,...,r)\\ A\vec{\alpha}_i=k_{1i}\vec{\alpha}_1+k_{2i}\vec{\alpha}_2+...+k_{ni}\vec{\alpha}_n\ (i=r+1,...,n)\end{cases}
{Aαi=λαi(i=1,...,r)Aαi=k1iα1+k2iα2+...+kniαn (i=r+1,...,n)
\ \qquad
即,
A
[
α
⃗
1
,
α
⃗
2
,
…
,
α
⃗
r
,
α
⃗
r
+
1
,
.
.
.
,
α
⃗
n
]
=
[
α
⃗
1
,
α
⃗
2
,
…
,
α
⃗
r
,
α
⃗
r
+
1
,
…
,
α
⃗
n
]
[
λ
0
…
0
k
1
,
r
+
1
…
k
1
,
n
0
λ
…
0
k
2
,
r
+
1
…
k
2
,
n
⋮
0
0
…
λ
k
r
,
r
+
1
…
k
r
,
n
0
0
…
0
k
r
+
1
,
r
+
1
…
k
r
+
1
,
n
0
0
…
0
k
r
+
2
,
r
+
1
…
k
r
+
2
,
n
⋮
0
0
…
0
k
n
,
r
+
1
…
k
n
,
n
]
A[\vec{\alpha}_1,\vec{\alpha}_2,\dots,\vec{\alpha}_r,\vec{\alpha}_{r+1},...,\vec{\alpha}_n]=[\vec{\alpha}_1,\vec{\alpha}_2,\dots,\vec{\alpha}_r,\vec{\alpha}_{r+1},\dots,\vec{\alpha}_n]\begin{bmatrix} \lambda\ 0\ \dots \ 0\quad k_{1,r+1}\ \dots \quad k_{1,n}\\ 0\ \lambda\ \dots \ 0\quad k_{2,r+1}\ \dots \quad k_{2,n}\\ \vdots \\ 0\ 0\ \dots \ \lambda\quad k_{r,r+1}\ \dots \quad k_{r,n}\\ 0\ 0\ \dots \ 0\ k_{r+1,r+1}\ \dots \ k_{r+1,n}\\ 0\ 0\ \dots \ 0\ k_{r+2,r+1}\ \dots \ k_{r+2,n}\\ \vdots \\ 0\ 0\ \dots \ 0\quad k_{n,r+1}\ \dots \quad k_{n,n} \end{bmatrix}
A[α1,α2,…,αr,αr+1,...,αn]=[α1,α2,…,αr,αr+1,…,αn]⎣
⎡λ 0 … 0k1,r+1 …k1,n0 λ … 0k2,r+1 …k2,n⋮0 0 … λkr,r+1 …kr,n0 0 … 0 kr+1,r+1 … kr+1,n0 0 … 0 kr+2,r+1 … kr+2,n⋮0 0 … 0kn,r+1 …kn,n⎦
⎤
\ \qquad
则
(
A
−
λ
′
E
)
[
α
⃗
1
,
α
⃗
2
,
…
,
α
⃗
r
,
α
⃗
r
+
1
,
.
.
.
,
α
⃗
n
]
=
[
α
⃗
1
,
α
⃗
2
,
…
,
α
⃗
r
,
α
⃗
r
+
1
,
…
,
α
⃗
n
]
[
(
λ
−
λ
′
)
0
…
0
k
1
,
r
+
1
…
k
1
,
n
0
(
λ
−
λ
′
)
…
0
k
2
,
r
+
1
…
k
2
,
n
⋮
⋮
⋮
⋮
⋮
⋮
⋮
0
0
…
(
λ
−
λ
′
)
k
r
,
r
+
1
…
k
r
,
n
0
0
…
0
(
k
r
+
1
,
r
+
1
−
λ
′
)
…
k
r
+
1
,
n
0
0
…
0
k
r
+
2
,
r
+
1
…
k
r
+
2
,
n
⋮
⋮
⋮
⋮
⋮
⋮
⋮
0
0
…
0
k
n
,
r
+
1
…
(
k
n
,
n
−
λ
′
)
]
(A-\lambda'E)[\vec{\alpha}_1,\vec{\alpha}_2,\dots,\vec{\alpha}_r,\vec{\alpha}_{r+1},...,\vec{\alpha}_n]=[\vec{\alpha}_1,\vec{\alpha}_2,\dots,\vec{\alpha}_r,\vec{\alpha}_{r+1},\dots,\vec{\alpha}_n]\begin{bmatrix} (\lambda-\lambda') & 0 & \dots & 0 & k_{1,r+1} & \dots & k_{1,n}\\ 0& (\lambda-\lambda')& \dots & 0& k_{2,r+1}& \dots & k_{2,n}\\ \vdots &\vdots &\vdots &\vdots &\vdots &\vdots &\vdots \\ 0& 0& \dots & (\lambda-\lambda')& k_{r,r+1}& \dots& k_{r,n}\\ 0& 0& \dots & 0& (k_{r+1,r+1}-\lambda')& \dots & k_{r+1,n}\\ 0& 0& \dots & 0& \quad k_{r+2,r+1} & \dots & k_{r+2,n}\\ \vdots &\vdots &\vdots &\vdots &\vdots &\vdots &\vdots \\ 0& 0& \dots & 0& k_{n,r+1}&\dots & (k_{n,n}-\lambda') \end{bmatrix}
(A−λ′E)[α1,α2,…,αr,αr+1,...,αn]=[α1,α2,…,αr,αr+1,…,αn]⎣
⎡(λ−λ′)0⋮000⋮00(λ−λ′)⋮000⋮0……⋮………⋮…00⋮(λ−λ′)00⋮0k1,r+1k2,r+1⋮kr,r+1(kr+1,r+1−λ′)kr+2,r+1⋮kn,r+1……⋮………⋮…k1,nk2,n⋮kr,nkr+1,nkr+2,n⋮(kn,n−λ′)⎦
⎤
\ \qquad
因
d
e
t
(
[
α
⃗
1
,
α
⃗
2
,
…
,
α
⃗
r
,
α
⃗
r
+
1
,
.
.
.
,
α
⃗
n
]
)
≠
0
det([\vec{\alpha}_1,\vec{\alpha}_2,\dots,\vec{\alpha}_r,\vec{\alpha}_{r+1},...,\vec{\alpha}_n])\ne0
det([α1,α2,…,αr,αr+1,...,αn])=0进一步有特征多项式:
d
e
t
(
A
−
λ
′
E
)
=
∣
(
λ
−
λ
′
)
0
…
0
k
1
,
r
+
1
…
k
1
,
n
0
(
λ
−
λ
′
)
…
0
k
2
,
r
+
1
…
k
2
,
n
⋮
⋮
⋮
⋮
⋮
⋮
⋮
0
0
…
(
λ
−
λ
′
)
k
r
,
r
+
1
…
k
r
,
n
0
0
…
0
(
k
r
+
1
,
r
+
1
−
λ
′
)
…
k
r
+
1
,
n
0
0
…
0
k
r
+
2
,
r
+
1
…
k
r
+
2
,
n
⋮
⋮
⋮
⋮
⋮
⋮
⋮
0
0
…
0
k
n
,
r
+
1
…
(
k
n
,
n
−
λ
′
)
∣
=
∣
(
λ
−
λ
′
)
0
…
0
0
(
λ
−
λ
′
)
…
0
⋮
⋮
⋮
⋮
0
0
…
(
λ
−
λ
′
)
∣
r
×
r
∣
(
k
r
+
1
,
r
+
1
−
λ
′
)
…
k
r
+
1
,
n
k
r
+
2
,
r
+
1
…
k
r
+
2
,
n
⋮
⋮
⋮
k
n
,
r
+
1
…
(
k
n
,
n
−
λ
′
)
∣
(
n
−
r
)
×
(
n
−
r
)
=
(
λ
−
λ
′
)
r
g
(
λ
′
)
det(A-\lambda'E)=\begin{vmatrix} (\lambda-\lambda') & 0 & \dots & 0 & k_{1,r+1} & \dots & k_{1,n}\\ 0& (\lambda-\lambda')& \dots & 0& k_{2,r+1}& \dots & k_{2,n}\\ \vdots &\vdots &\vdots &\vdots &\vdots &\vdots &\vdots \\ 0& 0& \dots & (\lambda-\lambda')& k_{r,r+1}& \dots& k_{r,n}\\ 0& 0& \dots & 0& (k_{r+1,r+1}-\lambda')& \dots & k_{r+1,n}\\ 0& 0& \dots & 0& k_{r+2,r+1} & \dots & k_{r+2,n}\\ \vdots &\vdots &\vdots &\vdots &\vdots &\vdots &\vdots \\ 0& 0& \dots & 0& k_{n,r+1}&\dots & (k_{n,n}-\lambda') \end{vmatrix}=\begin{vmatrix} (\lambda-\lambda') & 0 & \dots & 0 \\ 0& (\lambda-\lambda')& \dots & 0\\ \vdots& \vdots&\vdots&\vdots\\ 0& 0& \dots & (\lambda-\lambda')\\ \end{vmatrix}_{r\times r}\begin{vmatrix} (k_{r+1,r+1}-\lambda')& \dots & k_{r+1,n}\\ k_{r+2,r+1} & \dots & k_{r+2,n}\\ \vdots &\vdots &\vdots \\ k_{n,r+1}&\dots & (k_{n,n}-\lambda') \end{vmatrix}_{(n-r)\times (n-r)}=(\lambda-\lambda')^rg(\lambda')
det(A−λ′E)=∣
∣(λ−λ′)0⋮000⋮00(λ−λ′)⋮000⋮0……⋮………⋮…00⋮(λ−λ′)00⋮0k1,r+1k2,r+1⋮kr,r+1(kr+1,r+1−λ′)kr+2,r+1⋮kn,r+1……⋮………⋮…k1,nk2,n⋮kr,nkr+1,nkr+2,n⋮(kn,n−λ′)∣
∣=∣
∣(λ−λ′)0⋮00(λ−λ′)⋮0……⋮…00⋮(λ−λ′)∣
∣r×r∣
∣(kr+1,r+1−λ′)kr+2,r+1⋮kn,r+1……⋮…kr+1,nkr+2,n⋮(kn,n−λ′)∣
∣(n−r)×(n−r)=(λ−λ′)rg(λ′)
显然,
d
e
t
(
A
−
λ
′
E
)
=
(
λ
−
λ
′
)
r
g
(
λ
′
)
=
0
det(A-\lambda'E)=(\lambda-\lambda')^rg(\lambda')=0
det(A−λ′E)=(λ−λ′)rg(λ′)=0的根
λ
\lambda
λ的重数大于等于
r
r
r。(证毕)
定理5.5:实对称方阵的特征值为实数,且不同特征值对应的特征向量正交。
证明:设
λ
\lambda
λ为实对称矩阵
A
n
×
n
\bold A_{n\times n}
An×n的特征值,
X
⃗
\vec{X}
X为
λ
\lambda
λ对应的
A
\bold A
A的特征向量,则
A
X
⃗
=
λ
X
⃗
(
1
)
\bold{A}\vec{X}=\lambda\vec{X}\qquad(1)
AX=λX(1)
\ \ \qquad
对
(
1
)
(1)
(1)式进行共轭转置并右乘
X
⃗
\vec{X}
X得:
X
⃗
ˉ
T
A
ˉ
T
X
⃗
=
λ
ˉ
X
⃗
ˉ
T
X
⃗
(
2
)
\bar{\vec{X}}^T\bar{\bold{A}}^T\vec{X}=\bar{\lambda}\bar{\vec{X}}^T\vec{X}\qquad(2)
XˉTAˉTX=λˉXˉTX(2)
\ \ \qquad
又由
(
1
)
(1)
(1)得:
X
⃗
ˉ
T
A
X
⃗
=
λ
X
⃗
ˉ
T
X
⃗
(
3
)
\bar{\vec{X}}^T\bold{A}\vec{X}=\lambda\bar{\vec{X}}^T\vec{X}\qquad(3)
XˉTAX=λXˉTX(3)
\ \ \qquad
(
2
)
−
(
3
)
(2)-(3)
(2)−(3)得:
(
λ
ˉ
−
λ
)
X
⃗
ˉ
T
X
⃗
=
0
(\bar{\lambda}-\lambda)\bar{\vec{X}}^T\vec{X}=0
(λˉ−λ)XˉTX=0
\ \ \qquad
由于
X
⃗
≠
0
\vec{X}\ne 0
X=0,故
λ
ˉ
=
λ
\bar{\lambda}=\lambda
λˉ=λ,即特征值
λ
\lambda
λ为实数。
\ \ \qquad
设
X
⃗
1
,
X
⃗
2
\vec{X}_1,\vec{X}_2
X1,X2分别为互异特征值
λ
1
,
λ
2
\lambda_1,\lambda_2
λ1,λ2对应的特征向量,则:
{
A
X
⃗
1
=
λ
1
X
⃗
1
A
X
⃗
2
=
λ
2
X
⃗
2
⟹
(
λ
1
X
⃗
1
,
X
⃗
2
)
=
(
A
X
⃗
1
,
X
⃗
2
)
=
X
⃗
1
T
A
T
X
⃗
2
=
X
⃗
1
T
A
X
⃗
2
=
(
X
⃗
1
,
A
X
⃗
2
)
=
(
X
⃗
1
,
λ
2
X
⃗
2
)
⟹
(
λ
1
−
λ
2
)
(
X
⃗
1
,
X
⃗
2
)
=
0
\begin{cases}A\vec{X}_1=\lambda_1\vec{X}_1\\ A\vec{X}_2=\lambda_2\vec{X}_2\end{cases}\Longrightarrow(\lambda_1\vec{X}_1,\vec{X}_2)=(A\vec{X}_1,\vec{X}_2)=\vec{X}_1^TA^T\vec{X}_2=\vec{X}_1^TA\vec{X}_2=(\vec{X}_1,A\vec{X}_2)=(\vec{X}_1,\lambda_2\vec{X}_2)\Longrightarrow(\lambda_1-\lambda_2)(\vec{X}_1,\vec{X}_2)=0
{AX1=λ1X1AX2=λ2X2⟹(λ1X1,X2)=(AX1,X2)=X1TATX2=X1TAX2=(X1,AX2)=(X1,λ2X2)⟹(λ1−λ2)(X1,X2)=0
\ \ \qquad
则
(
X
⃗
1
,
X
⃗
2
)
=
0
(\vec{X}_1,\vec{X}_2)=0
(X1,X2)=0(证毕)