1. ∃ ϵ ∈ V n ( F ) 且 λ ∈ F , ϵ ≠ 0 , 使 T ( ϵ ) = λ ϵ \exist \epsilon \in V_n(F)且\lambda \in F,\epsilon \neq 0,使T(\epsilon) = \lambda \epsilon ∃ϵ∈Vn(F)且λ∈F,ϵ=0,使T(ϵ)=λϵ,则 λ \lambda λ是T的特征值,向量 ϵ \epsilon ϵ是线性变换T对应于 λ \lambda λ的特征向量
证明:
ϵ = ( α 1 α 2 ⋯ α n ) X \epsilon = (\alpha_1 \quad \alpha_2 \quad \cdots \alpha_n) X ϵ=(α1α2⋯αn)X
T ( ϵ ) = ( α 1 α 2 ⋯ α n ) A X T(\epsilon) = (\alpha_1 \quad \alpha_2 \quad \cdots \alpha_n) AX T(ϵ)=(α1α2⋯αn)AX
因此 T ( ϵ ) = λ ϵ ⟺ A X = λ X T(\epsilon) = \lambda \epsilon \iff AX = \lambda X T(ϵ)=λϵ⟺AX=λX
2. 线性变换T在基 { α 1 α 2 ⋯ α n } \{\alpha_1 \quad \alpha_2 \quad \cdots \alpha_n\} {α1α2⋯αn}下矩阵为A,则A的特征值是T的特征值;若X是A的特征向量,则 ϵ = ( α 1 α 2 ⋯ α n ) X \epsilon = (\alpha_1 \quad \alpha_2 \quad \cdots \alpha_n) X ϵ=(α1α2⋯αn)X是T的特征向量
T的特征值由T决定,和基与变换矩阵的选择无关
不同特征值对应的特征向量必线性无关
λ \lambda λ是线性变换T的特征值, ϵ 1 , ϵ 2 , ⋯ , ϵ t \epsilon_1,\ \epsilon_2,\ \cdots,\ \epsilon_t ϵ1, ϵ2, ⋯, ϵt是T对应于 λ \lambda λ的特征向量的极大线性无关组,则称子空间 V λ = L { ϵ 1 , ϵ 2 , ⋯ , ϵ t } V_\lambda = L\{\epsilon_1,\ \epsilon_2,\ \cdots,\ \epsilon_t\} Vλ=L{ϵ1, ϵ2, ⋯, ϵt}为T关于 λ \lambda λ的特征子空间
∀ α ∈ V λ , α ≠ 0 , α \forall \alpha \in V_\lambda,\ \alpha \neq 0,\ \alpha ∀α∈Vλ, α=0, α就是T关于 λ \lambda λ的特征向量, V λ = N ( T − λ I ) V_\lambda=N(T - \lambda I) Vλ=N(T−λI)
若T有s个互异的特征值,则它们就对应s个特征子空间
奇异矩阵: 该矩阵的秩不是满秩
λ 1 , λ 2 , ⋯ , λ s 是 V n ( F ) \lambda_1,\ \lambda_2,\ \cdots,\ \lambda_s是V_n(F) λ1, λ2, ⋯, λs是Vn(F)上线性变换T的s个互异的特征值, V λ i V_{\lambda_i} Vλi是 λ i \lambda_i λi的特征子空间,则
V λ i \quad V_{\lambda_i} Vλi是T的不变子空间
λ i ≠ λ j 时 , V λ i ∩ V λ j = { 0 } \quad \lambda_i \neq \lambda_j时,V_{\lambda_i} \cap V_{\lambda_j} = \{0\} λi=λj时,Vλi∩Vλj={0}
若 λ i \quad 若\lambda_i 若λi是T的k重特征值,则 d i m V λ i ≤ k dimV_{\lambda_i} \leq k dimVλi≤k
V λ 1 + V λ 2 + ⋯ + V λ s = V λ 1 ⨁ V λ 2 ⨁ ⋯ ⨁ V λ s ⊂ V n ( F ) V_{\lambda_1} + V_{\lambda_2} + \cdots + V_{\lambda_s} = V_{\lambda_1} \bigoplus V_{\lambda_2} \bigoplus \cdots \bigoplus V_{\lambda_s} \subset V_n(F) Vλ1+Vλ2+⋯+Vλs=Vλ1⨁Vλ2⨁⋯⨁Vλs⊂Vn(F)
∑ i = 1 s d i m V λ i ≤ d i m V n ( F ) = n \sum^s_{i = 1} dimV_{\lambda_i} \leq dimV_n(F) = n ∑i=1sdimVλi≤dimVn(F)=n
∑ i = 1 s d i m V λ i = n ⇒ V λ 1 ⨁ V λ 2 ⨁ ⋯ ⨁ V λ s = V n ( F ) \sum^s_{i = 1} dimV_{\lambda_i} = n \Rightarrow V_{\lambda_1} \bigoplus V_{\lambda_2} \bigoplus \cdots \bigoplus V_{\lambda_s} = V_n(F) ∑i=1sdimVλi=n⇒Vλ1⨁Vλ2⨁⋯⨁Vλs=Vn(F)
线性变换T有对角矩阵 ⟺ \iff ⟺ T有n个线性无关的特征向量 ⟺ \iff ⟺ V λ 1 ⨁ V λ 2 ⨁ ⋯ ⨁ V λ s = V n ( F ) V_{\lambda_1} \bigoplus V_{\lambda_2} \bigoplus \cdots \bigoplus V_{\lambda_s} = V_n(F) Vλ1⨁Vλ2⨁⋯⨁Vλs=Vn(F) ⟺ \iff ⟺ m ( λ ) = ( λ − λ 1 ) ( λ − λ 2 ) ⋯ ( λ − λ s ) m(\lambda) = (\lambda - \lambda_1)(\lambda - \lambda_2) \cdots (\lambda - \lambda_s) m(λ)=(λ−λ1)(λ−λ2)⋯(λ−λs)
V λ 1 ⨁ V λ 2 ⨁ ⋯ ⨁ V λ s = V n ( F ) V_{\lambda_1} \bigoplus V_{\lambda_2} \bigoplus \cdots \bigoplus V_{\lambda_s} = V_n(F) Vλ1⨁Vλ2⨁⋯⨁Vλs=Vn(F) ⟺ \iff ⟺ λ i 是 T 的 k i 重 特 征 值 , 必 有 d i m V λ i = k i , i = 1 , 2 , ⋯ , n \lambda_i是T的k_i重特征值,必有dimV_{\lambda_i} = k_i,\ i = 1,\ 2,\ \cdots,\ n λi是T的ki重特征值,必有dimVλi=ki, i=1, 2, ⋯, n
若 λ i \lambda_i λi为T的单特征值,则必有 d i m V λ i = 1 dimV_{\lambda_i} = 1 dimVλi=1
线性变换T有n个互异的特征值 λ 1 , λ 2 , ⋯ , λ n \lambda_1,\ \lambda_2,\ \cdots,\ \lambda_n λ1, λ2, ⋯, λn ⇒ \Rightarrow ⇒ V λ 1 ⨁ V λ 2 ⨁ ⋯ ⨁ V λ s = V n ( F ) V_{\lambda_1} \bigoplus V_{\lambda_2} \bigoplus \cdots \bigoplus V_{\lambda_s} = V_n(F) Vλ1⨁Vλ2⨁⋯⨁Vλs=Vn(F)
V n ( F ) V_n(F) Vn(F)上线性变换有对角矩阵表示 ⟺ \iff ⟺ V n ( F ) V_n(F) Vn(F)可分解为T的一维不变子空间的直和
J ( λ ) = [ λ 1 λ 1 ⋱ 1 λ ] J(\lambda) = \begin{bmatrix} \lambda & 1 & & \\ & \lambda & 1 \\ & & \ddots &1 \\ & & & \lambda \end{bmatrix} J(λ)=⎣⎢⎢⎡λ1λ1⋱1λ⎦⎥⎥⎤
的r阶方阵称为一个r阶Jordan块。由若干个Jordan块 J i ( λ i ) J_i(\lambda_i) Ji(λi)构成的准对角矩阵
J = [ J 1 ( λ 1 ) J 2 ( λ 2 ) ⋱ J m ( λ m ) ] J = \begin{bmatrix} J_1(\lambda_1) & & & \\ & J_2(\lambda_2) & & \\ & & \ddots & \\ & & & J_m(\lambda_m) \end{bmatrix} J=⎣⎢⎢⎡J1(λ1)J2(λ2)⋱Jm(λm)⎦⎥⎥⎤
称为Jordan矩阵
Jordan块都是一阶的Jordan矩阵就是对角矩阵
在复数域C上,每个n阶方阵A都相似于一个Jordan矩阵,即存在可逆矩阵P,使
P − 1 A P = J A = [ J 1 ( λ 1 ) J 2 ( λ 2 ) ⋱ J s ( λ s ) ] P^{-1}AP = J_A = \begin{bmatrix} J_1(\lambda_1) & & & \\ & J_2(\lambda_2) & & \\ & & \ddots & \\ & & & J_s(\lambda_s) \end{bmatrix} P−1AP=JA=⎣⎢⎢⎡J1(λ1)J2(λ2)⋱Js(λs)⎦⎥⎥⎤
其中
J i ( λ i ) = [ J i 1 ( λ i ) J i 2 ( λ i ) ⋱ J i t i ( λ i ) ] ∈ C k i ∗ k i J_i(\lambda_i) = \begin{bmatrix} J_{i_1}(\lambda_i) & & & \\ & J_{i_2}(\lambda_i) & & \\ & & \ddots & \\ & & & J_{i_{t_i}}(\lambda_i) \end{bmatrix} \in C^{k_i * k_i} Ji(λi)=⎣⎢⎢⎡Ji1(λi)Ji2(λi)⋱Jiti(λi)⎦⎥⎥⎤∈Cki∗ki
J i j ( λ i ) 为 n j J_{ij}(\lambda_i)为n_j Jij(λi)为nj阶Jordan块, ∑ j = 1 t i n j = k i \sum^{t_i}_{j = 1} n_j = k_i ∑j=1tinj=ki. J i ( λ i ) 是 k i ∗ k i J_i(\lambda_i)是k_i * k_i Ji(λi)是ki∗ki阶Jordan矩阵, ∑ i = 1 s k i = n \sum^s_{i = 1} k_i = n ∑i=1ski=n
若不计较Jordan的排列次序,则每个方阵的Jordan标准形 J A J_A JA唯一
设A的特征多项式为 ∣ λ I − A ∣ = ( λ − λ 1 ) k 1 ( λ − λ 2 ) k 2 ⋯ ( λ − λ s ) k s |\lambda I - A| = (\lambda - \lambda_1) ^ {k_1} (\lambda - \lambda_2) ^ {k_2} \cdots (\lambda - \lambda_s)^{k_s} ∣λI−A∣=(λ−λ1)k1(λ−λ2)k2⋯(λ−λs)ks其中 λ i 是 A 的 k i 重 特 征 值 , ∑ i = 1 s k i = n , λ 1 , λ 2 , ⋯ , λ s 互 异 \lambda_i是A的k_i重特征值,\sum^s_{i = 1} k_i = n,\ \lambda_1,\ \lambda_2,\ \cdots,\ \lambda_s互异 λi是A的ki重特征值,∑i=1ski=n, λ1, λ2, ⋯, λs互异
⇒ \Rightarrow ⇒
J A = [ J 1 ( λ 1 ) J 2 ( λ 2 ) ⋱ J s ( λ s ) ] ( 2.5 ) J_A = \begin{bmatrix} J_1(\lambda_1) & & & \\ & J_2(\lambda_2) & & \\ & & \ddots & \\ & & & J_s(\lambda_s) \end{bmatrix} \qquad \qquad \qquad (2.5) JA=⎣⎢⎢⎡J1(λ1)J2(λ2)⋱Js(λs)⎦⎥⎥⎤(2.5)
J i ( λ i ) J_i(\lambda_i) Ji(λi)是主对角线元素为 λ i \lambda_i λi的 k i k_i ki阶Jordan矩阵
令 P = ( p 1 , p 2 , ⋯ , p s ) , p i ∈ C n ∗ k i , i = 1 , 2 , ⋯ , s , A P = P J A P = (p_1,\ p_2,\ \cdots,\ p_s),\ p_i \in C ^ {n * k_i},\ i = 1,\ 2,\ \cdots,\ s,\ AP = PJ_A P=(p1, p2, ⋯, ps), pi∈Cn∗ki, i=1, 2, ⋯, s, AP=PJA可表示为:
A ( p 1 p 2 ⋯ p s ) = ( p 1 p 2 ⋯ p s ) [ J 1 ( λ 1 ) J 2 ( λ 2 ) ⋱ J s ( λ s ) ] A(p_1 \quad p_2 \quad \cdots \quad p_s) = (p_1 \quad p_2 \quad \cdots \quad p_s) \begin{bmatrix} J_1(\lambda_1) & & & \\ & J_2(\lambda_2) & & \\ & & \ddots & \\ & & & J_s(\lambda_s) \end{bmatrix} A(p1p2⋯ps)=(p1p2⋯ps)⎣⎢⎢⎡J1(λ1)J2(λ2)⋱Js(λs)⎦⎥⎥⎤
⇒ \Rightarrow ⇒
( A p 1 A p 2 ⋯ A p s ) = ( p 1 J 1 ( λ 1 ) p 2 J 2 ( λ 2 ) ⋯ p s J s ( λ s ) ) (A p_1 \quad A p_2 \quad \cdots \quad A p_s) = (p_1 J_1(\lambda_1) \quad p_2 J_2(\lambda_2) \quad \cdots \quad p_s J_s(\lambda_s)) (Ap1Ap2⋯Aps)=(p1J1(λ1)p2J2(λ2)⋯psJs(λs))
⇒ \Rightarrow ⇒
A p i = p i J i ( λ i ) , i = 1 , 2 , ⋯ , s ( 2.6 ) A p_i = p_i J_i(\lambda_i), \quad i = 1,\ 2,\ \cdots,\ s \qquad \qquad \qquad (2.6) Api=piJi(λi),i=1, 2, ⋯, s(2.6)
取 A p 1 = p 1 J 1 ( λ 1 ) A p_1 = p_1 J_1(\lambda_1) Ap1=p1J1(λ1)为代表来分析,设
J 1 ( λ 1 ) = [ J 11 ( λ 1 ) J 12 ( λ 1 ) ⋱ J 1 t ( λ 1 ) ] J_1(\lambda_1) = \begin{bmatrix} J_{11}(\lambda_1) & & & \\ & J_{12}(\lambda_1) & & \\ & & \ddots & \\ & & & J_{1t}(\lambda_1) \end{bmatrix} J1(λ1)=⎣⎢⎢⎡J11(λ1)J12(λ1)⋱J1t(λ1)⎦⎥⎥⎤
其中 J 1 i 为 n i J_{1i}为n_i J1i为ni阶Jordan块, ∑ i = 1 t n i = k 1 \sum^t_{i =1} n_i = k_1 ∑i=1tni=k1
⇒ \Rightarrow ⇒
J 1 i ( λ 1 ) = [ λ 1 1 λ 1 1 ⋱ ⋱ 1 λ 1 ] n i ∗ n i ( 2.7 ) J_{1i}(\lambda_1) = \begin{bmatrix} \lambda_1 & 1 & & & \\ & \lambda_1 & 1 & & \\ & & & \ddots & \\ & & & \ddots & 1\\ & & & & \lambda_1 \end{bmatrix}_{n_i * n_i} \qquad \qquad \qquad (2.7) J1i(λ1)=⎣⎢⎢⎢⎢⎡λ11λ11⋱⋱1λ1⎦⎥⎥⎥⎥⎤ni∗ni(2.7)
把 p 1 p_1 p1按 n 1 列 , n 2 列 , ⋯ , n t 列 分 块 , p 1 = ( p 1 ( 1 ) p 2 ( 1 ) ⋯ p t ( 1 ) ) n_1列,\ n_2列,\ \cdots,\ n_t列分块,\quad p_1 = (p^{(1)}_1 \quad p^{(1)}_2 \quad \cdots p^{(1)}_t) n1列, n2列, ⋯, nt列分块,p1=(p1(1)p2(1)⋯pt(1))
由等式2.6有 A p j ( 1 ) = p 1 ( 1 ) J 1 j ( λ 1 ) , j = 1 , 2 , ⋯ , t ( 2.8 ) A p^{(1)}_j = p^{(1)}_1 J_{1j}(\lambda_1), \quad j = 1,\ 2,\ \cdots,\ t \qquad \qquad \qquad (2.8) Apj(1)=p1(1)J1j(λ1),j=1, 2, ⋯, t(2.8)
设 p j ( 1 ) = ( α 1 β 2 ⋯ β n j ) p^{(1)}_j = (\alpha_1 \quad \beta_2 \quad \cdots \quad \beta_{n_j}) pj(1)=(α1β2⋯βnj),结合等式2.7,等式2.8可化为
A ( α 1 β 2 ⋯ β n j ) = ( α 1 β 2 ⋯ β n j ) [ λ 1 1 λ 1 1 ⋱ ⋱ 1 λ 1 ] A(\alpha_1 \quad \beta_2 \quad \cdots \beta_{n_j}) = (\alpha_1 \quad \beta_2 \cdots \beta_{n_j}) \begin{bmatrix} \lambda_1 & 1 & & & \\ & \lambda_1 & 1 & & \\ & & & \ddots & \\ & & & \ddots & 1 & \\ & & & & \lambda_1 \end{bmatrix} A(α1β2⋯βnj)=(α1β2⋯βnj)⎣⎢⎢⎢⎢⎡λ11λ11⋱⋱1λ1⎦⎥⎥⎥⎥⎤
该矩阵等式等价于
{ ( A − λ 1 I ) α 1 = 0 ( A − λ 1 I ) β 2 = α 1 ( A − λ 1 I ) β 3 = β 2 ⋮ ( A − λ 1 I ) β n j = β n j − 1 ( 2.9 ) \begin{cases} (A - \lambda_1I)\alpha_1 = 0 \\ (A - \lambda_1I)\beta_2 = \alpha_1 \\ (A - \lambda_1I)\beta_3 = \beta_2 \\ \qquad \vdots \\ (A - \lambda_1I)\beta_{n_j} = \beta_{n_{j - 1}} \end{cases} \qquad \qquad \qquad (2.9) ⎩⎪⎪⎪⎪⎪⎪⎨⎪⎪⎪⎪⎪⎪⎧(A−λ1I)α1=0(A−λ1I)β2=α1(A−λ1I)β3=β2⋮(A−λ1I)βnj=βnj−1(2.9)
从等式2.9中求得一组向量 { α 1 , β 2 , ⋯ , β n j } \{\alpha_1,\ \beta_2,\ \cdots,\ \beta_{n_j}\} {α1, β2, ⋯, βnj},称为Jordan链。链中第一个向量 α 1 \alpha_1 α1是A关于 λ 1 \lambda_1 λ1的特征向量, β 2 , ⋯ , β n j \beta_2,\ \cdots,\ \beta_{n_j} β2, ⋯, βnj称为广义特征向量。它的长度 n j n_j nj就是 J 1 j ( λ 1 ) J_{1j}(\lambda_1) J1j(λ1)的阶数
等式2.8 ⇒ n u m ( J 1 j ( λ 1 ) ) = n u m ( J o r d a n 链 ) = n u m ( α ) \Rightarrow num(J_{1j}(\lambda_1)) = num(Jordan链) = num(\alpha) ⇒num(J1j(λ1))=num(Jordan链)=num(α),num表示总数量。用文字可以表述为:
由等式2.8可知, J 1 ( λ 1 ) J_1(\lambda_1) J1(λ1)有多少个Jordan块 J 1 j ( λ 1 ) J_{1j}(\lambda_1) J1j(λ1),就有多少条Jordan链,也就会有多少个线性无关的特征向量
A ∈ F n × n , a i ∈ F , g ( λ ) = a m λ m + a m − 1 λ m − 1 + ⋯ + a 1 λ + a 0 A \in F^{n \times n},\ a_i \in F,\ g(\lambda) = a_m \lambda^m + a_{m - 1} \lambda^{m -1} + \cdots +a_1 \lambda + a_0 A∈Fn×n, ai∈F, g(λ)=amλm+am−1λm−1+⋯+a1λ+a0是一个多项式,则称矩阵 g ( A ) = a m A m + a m − 1 A m − 1 + ⋯ + a 1 A + a 0 I g(A) = a_m A^m + a_{m - 1} A^{m - 1} + \cdots + a_1 A + a_0 I g(A)=amAm+am−1Am−1+⋯+a1A+a0I为A的矩阵多项式。
A ∈ F n × n A \in F ^ {n \times n} A∈Fn×n,g(A)是A的矩阵多项式,则:
方阵g(A)计算
已知方阵A和多项式 g ( λ ) = a m λ m + a m − 1 λ m − 1 + ⋯ + a 1 λ + a 0 g(\lambda) = a_m \lambda^m + a_{m - 1} \lambda ^ {m - 1} + \cdots + a_1 \lambda + a_0 g(λ)=amλm+am−1λm−1+⋯+a1λ+a0
当 m ≥ r m \geq r m≥r时,
g ( J ( λ ) ) = ∑ i = 0 r − 1 g ( i ) ( λ ) i ! U i = [ g ( λ ) g ′ ( λ ) ⋯ g ( r − 1 ) ( λ ) ( r − 1 ) ! g ( λ ) ⋱ ⋮ ⋱ g ′ ( λ ) g ( λ ) ] g(J(\lambda)) = \sum^{r - 1}_{i=0}\frac{g^{(i)}(\lambda)}{i!}U^i = \begin{bmatrix} g(\lambda) & g'(\lambda) & \cdots & \frac{g^{(r - 1)(\lambda)}}{(r - 1)!} \\ & g(\lambda) & \ddots & \vdots \\ & & \ddots & g'(\lambda) \\ & & & g(\lambda) \end{bmatrix} g(J(λ))=i=0∑r−1i!g(i)(λ)Ui=⎣⎢⎢⎢⎢⎡g(λ)g′(λ)g(λ)⋯⋱⋱(r−1)!g(r−1)(λ)⋮g′(λ)g(λ)⎦⎥⎥⎥⎥⎤
当 m < r m < r m<r时,
g ( J ( λ ) ) = [ g ( λ ) g ′ ( λ ) ⋯ g m ( λ ) m ! 0 ⋯ 0 g ( λ ) ⋱ 0 ⋱ g ( m ) ( λ ) m ! ⋱ ⋮ g ′ ( λ ) g ( λ ) ] g(J(\lambda)) = \begin{bmatrix} g(\lambda) & g'(\lambda) & \cdots & \frac{g^m(\lambda)}{m!} & 0 & \cdots & 0 \\ & g(\lambda) & & & \ddots & & 0 \\ & & & \ddots & & & \frac{g^{(m)}(\lambda)}{m!} \\ & & & \ddots & & &\vdots \\ & & & & & & g'(\lambda) \\ & & & & & & g(\lambda) \end{bmatrix} g(J(λ))=⎣⎢⎢⎢⎢⎢⎢⎢⎢⎡g(λ)g′(λ)g(λ)⋯m!gm(λ)⋱⋱0⋱⋯00m!g(m)(λ)⋮g′(λ)g(λ)⎦⎥⎥⎥⎥⎥⎥⎥⎥⎤
其中r表示 J ( λ ) J(\lambda) J(λ)的阶数
化零多项式: n阶方阵A, ∃ 多 项 式 g ( λ ) , 使 g ( A ) = 0 \exists 多项式g(\lambda),使g(A) = 0 ∃多项式g(λ),使g(A)=0
Cayley-Hamilton: f ( λ ) = ∣ λ I − A ∣ = λ n + a n − 1 λ n − 1 + ⋯ + a 1 λ + a 0 ⇒ f ( A ) = 0 f(\lambda) = |\lambda I - A| = \lambda^n + a_{n - 1} \lambda^{n - 1} + \cdots + a_1 \lambda + a_0 \Rightarrow f(A) = 0 f(λ)=∣λI−A∣=λn+an−1λn−1+⋯+a1λ+a0⇒f(A)=0,即方阵A的特征多项式就是A的化零多项式
f ( λ ) f(\lambda) f(λ)为n次多项式, f ( A ) = 0 f(A) = 0 f(A)=0使得 A n A^n An可以表示为 A k ( k < n ) A^k(k < n) Ak(k<n)的幂次项的组合:
A n = − ( a n − 1 A n − 1 + a n − 2 A n − 2 + ⋯ + a 1 A + a 0 I ) A^n = -(a_{n - 1}A^{n - 1} + a_{n - 2}A^{n - 2} + \cdots + a_1A + a_0I) An=−(an−1An−1+an−2An−2+⋯+a1A+a0I)
例题一
例题二
A = [ 1 0 1 − 1 ] , f ( λ ) = ∣ λ I − A ∣ = λ 2 − 1 , g ( λ ) = λ 6 + 2 λ 5 − 4 λ 3 + λ 2 − 5 A = \begin{bmatrix} 1 & 0 \\ 1 & -1 \end{bmatrix},f(\lambda) = |\lambda I - A| = \lambda^2 - 1,g(\lambda) = \lambda^6 + 2 \lambda^5 -4 \lambda^3 + \lambda^2 - 5 A=[110−1],f(λ)=∣λI−A∣=λ2−1,g(λ)=λ6+2λ5−4λ3+λ2−5,求g(A)
解: g ( λ ) = ( λ 2 − 1 ) ( λ 4 + 2 λ 3 + λ 2 − 2 λ + 2 ) + ( − 2 λ − 3 ) ⇒ g ( A ) = − 2 A − 3 g(\lambda) = (\lambda^2 - 1)(\lambda^4 + 2 \lambda^3 + \lambda^2 - 2 \lambda + 2) + (-2 \lambda - 3) \Rightarrow g(A) = -2 A - 3 g(λ)=(λ2−1)(λ4+2λ3+λ2−2λ+2)+(−2λ−3)⇒g(A)=−2A−3
若 A ∈ C n × n A \in C^{n \times n} A∈Cn×n,则A有无数个化零多项式。事实上, f ( λ ) = ∣ λ I − A ∣ , 有 f ( A ) = 0 f(\lambda) = |\lambda I - A|,有f(A) = 0 f(λ)=∣λI−A∣,有f(A)=0
且 ∀ g ( λ ) = h ( λ ) f ( λ ) \forall g(\lambda) = h(\lambda) \ f(\lambda) ∀g(λ)=h(λ) f(λ),均有 g ( A ) = h ( A ) ⋅ f ( A ) g(A) = h(A) \cdot f(A) g(A)=h(A)⋅f(A)
最小多项式: n阶方阵A的所有化零多项式中,次数最低且首系数为1的多项式称为A的最小多项式,记为 m ( λ ) m(\lambda) m(λ)
T的特征多项式 f ( λ ) f(\lambda) f(λ)与最小多项式 m ( λ ) m(\lambda) m(λ)有相同的根(重数不计),即
f ( λ ) = ∣ λ I − A ∣ = ( λ − λ 1 ) r 1 ( λ − λ 2 ) r 2 ⋯ ( λ − λ s ) r s f(\lambda) = |\lambda I - A| = (\lambda - \lambda_1)^{r_1}(\lambda - \lambda_2)^{r_2} \cdots (\lambda - \lambda_s)^{r_s} f(λ)=∣λI−A∣=(λ−λ1)r1(λ−λ2)r2⋯(λ−λs)rs
⇒ \Rightarrow ⇒ m ( λ ) = ( λ − λ 1 ) t 1 ( λ − λ 2 ) t 2 ⋯ ( λ − λ s ) t s , 1 ≤ t i ≤ r i , i = 1 , 2 , ⋯ , s m(\lambda) = (\lambda - \lambda_1)^{t_1}(\lambda - \lambda_2)^{t_2} \cdots (\lambda - \lambda_s)^{t_s},1 \leq t_i \leq r_i,i = 1,\ 2,\ \cdots,\ s m(λ)=(λ−λ1)t1(λ−λ2)t2⋯(λ−λs)ts,1≤ti≤ri,i=1, 2, ⋯, s
设T的特征多项式为
f ( λ ) = ( λ − λ 1 ) r 1 ( λ − λ 2 ) r 2 ⋯ ( λ − λ s ) r s , f(\lambda) = (\lambda - \lambda_1)^{r_1}(\lambda - \lambda_2)^{r_2} \cdots (\lambda - \lambda_s)^{r_s}, f(λ)=(λ−λ1)r1(λ−λ2)r2⋯(λ−λs)rs,
T的Jordan标准形中关于特征值 λ i \lambda_i λi的Jordan块的最高阶数为 n i ‾ \overline{n_i} ni,则T的最小多项式 m ( λ ) = ( λ − λ 1 ) n 1 ‾ ( λ − λ 2 ) n 2 ‾ ⋯ ( λ − λ s ) n s ‾ m(\lambda) = (\lambda - \lambda_1)^{\overline{n_1}}(\lambda - \lambda_2)^{\overline{n_2}} \cdots (\lambda - \lambda_s)^{\overline{n_s}} m(λ)=(λ−λ1)n1(λ−λ2)n2⋯(λ−λs)ns
线性变换T可对角化 ⟺ \iff ⟺ T有n个线性无关的特征向量 ⟺ \iff ⟺ m ( λ ) = ( λ − λ 1 ) ( λ − λ 2 ) ⋯ ( λ − λ s ) m(\lambda) = (\lambda - \lambda_1)(\lambda - \lambda_2) \cdots (\lambda - \lambda_s) m(λ)=(λ−λ1)(λ−λ2)⋯(λ−λs)
设 A ∈ C n × n A \in C ^ {n \times n} A∈Cn×n,A的最小多项式 m ( λ ) m(\lambda) m(λ),特征多项式 f ( λ ) f(\lambda) f(λ),任一化零多项式 ψ ( λ ) \psi (\lambda) ψ(λ),则
相似矩阵具有相同的最小多项式
∧ = [ λ 1 λ 2 ⋱ λ s ] \wedge = \begin{bmatrix} \lambda_1 & & & \\ & \lambda_2 & & \\ & & \ddots & \\ & & & \lambda_s \end{bmatrix} ∧=⎣⎢⎢⎡λ1λ2⋱λs⎦⎥⎥⎤,若 λ 1 , ⋯ , λ s \lambda1,\ \cdots,\ \lambda_s λ1, ⋯, λs互异,则 m ( λ ) = ( λ − λ 1 ) ⋯ ( λ − λ s ) m(\lambda) = (\lambda - \lambda_1) \cdots (\lambda - \lambda_s) m(λ)=(λ−λ1)⋯(λ−λs)
本博客中所有内容均来自《矩阵论》第二版 华科,加入了一些自己的理解和简化,如有侵权,请联系删除