设 a i j ( λ ) ( i = 1... m , j = 1... n ) a_{ij}(\lambda)(i=1...m,j=1...n) aij(λ)(i=1...m,j=1...n)为数域 F F F上多项式,则称
A ( λ ) = [ a 11 ( λ ) a 12 ( λ ) ⋯ a 1 n ( λ ) a 21 ( λ ) a 22 ( λ ) ⋯ a 2 n ( λ ) ⋮ ⋮ ⋱ ⋮ a m 1 ( λ ) a m 2 ( λ ) ⋯ a m n ( λ ) ] A(\lambda)= \begin{bmatrix} a_{11}(\lambda) & a_{12}(\lambda) & \cdots & a_{1n}(\lambda)\\ a_{21}(\lambda) & a_{22}(\lambda) & \cdots & a_{2n}(\lambda) \\ \vdots & \vdots & \ddots & \vdots\\ a_{m1}(\lambda) & a_{m2}(\lambda) & \cdots & a_{mn}(\lambda) \end{bmatrix} A(λ)=⎣⎢⎢⎢⎡a11(λ)a21(λ)⋮am1(λ)a12(λ)a22(λ)⋮am2(λ)⋯⋯⋱⋯a1n(λ)a2n(λ)⋮amn(λ)⎦⎥⎥⎥⎤为多项式矩阵或 λ \lambda λ矩阵
其中 a i j ( λ ) ( i = 1... m , j = 1... n ) a_{ij}(\lambda)(i=1...m,j=1...n) aij(λ)(i=1...m,j=1...n)中最高次数为 A ( λ ) A(\lambda) A(λ)次数
例如数字矩阵,特征矩阵 λ E − A \lambda E-A λE−A
若 λ \lambda λ矩阵 A ( λ ) A(\lambda) A(λ)中有 r ( r ≥ 1 ) r ( r\ge 1 ) r(r≥1)阶子式不为零,而所有 r + 1 r+1 r+1阶子式(若有的话)全为零,则称 A ( λ ) A(\lambda) A(λ)的秩为 r r r,记为 r a n k A ( λ ) = r rankA(\lambda)=r rankA(λ)=r,零矩阵的秩为 0 0 0
其中倍数为 ϕ ( λ ) \phi(\lambda) ϕ(λ),为 λ \lambda λ的一个多项式
行(列)变换等价于左(右)乘相应初等矩阵
若 A ( λ ) A(\lambda) A(λ)经过有限次初等变换后变成 B ( λ ) B(\lambda) B(λ)则称 A ( λ ) A(\lambda) A(λ)与 B ( λ ) B(\lambda) B(λ)等价,记为 A ( λ ) ≃ B ( λ ) A(\lambda)\simeq B(\lambda) A(λ)≃B(λ)
等价关系满足
对任意非零 λ \lambda λ矩阵 A ( λ ) A(\lambda) A(λ)都等价于一个“对角矩阵”
A ( λ ) ≃ [ d 1 ( λ ) ⋱ d r ( λ ) 0 ⋱ 0 ] A(\lambda)\simeq \begin{bmatrix} d_{1}(\lambda) & & & & & \\ & \ddots & & & & \\ & & d_{r}(\lambda) & & & \\ & & & 0 & & \\ & & & & \ddots & \\ & & & & &0 \end{bmatrix} A(λ)≃⎣⎢⎢⎢⎢⎢⎢⎡d1(λ)⋱dr(λ)0⋱0⎦⎥⎥⎥⎥⎥⎥⎤
其中 r ≥ 1 r \ge 1 r≥1, d i ( λ ) d_{i}(\lambda) di(λ)首项系数为 1 1 1,且 d i ( λ ) ∣ d i + 1 ( λ ) d_{i}(\lambda)|d_{i+1}(\lambda) di(λ)∣di+1(λ)
这种形式的 λ \lambda λ矩阵称Smith标准形
d 1 ( λ ) , . . . , d r ( λ ) d_{1}(\lambda),...,d_{r}(\lambda) d1(λ),...,dr(λ)称 A ( λ ) A(\lambda) A(λ)不变因子
A ( λ ) A(\lambda) A(λ)为 λ \lambda λ矩阵且 r a n k ( A ( λ ) ) = r rank(A(\lambda))=r rank(A(λ))=r对于任意正整数 k , 1 ≤ k ≤ r k , 1 \le k \le r k,1≤k≤r, A ( λ ) A(\lambda) A(λ)必有非零的 k k k阶子式, A ( λ ) A(\lambda) A(λ)的全部 k k k阶子式的首项系数为 1 1 1的最大公因式 D k ( λ ) D_{k}(\lambda) Dk(λ)称为 A ( λ ) A(\lambda) A(λ)的 k k k阶行列式因子
显然,若 r a n k ( A ( λ ) ) = r rank(A(\lambda))=r rank(A(λ))=r,则行列式因子共有 r r r个
等价的 λ \lambda λ矩阵有相同的各阶行列式因子,从而有相同的秩
λ \lambda λ矩阵 A ( λ ) A(\lambda) A(λ)的Smith标准形是唯一的
标准形的各阶行列式因子为
D 1 ( λ ) = d 1 ( λ ) D 2 ( λ ) = d 1 ( λ ) d 2 ( λ ) ⋮ D r ( λ ) = d 1 ( λ ) d 2 ( λ ) ⋯ d r ( λ ) D_{1}(\lambda)=d_{1}(\lambda)\\ D_{2}(\lambda)=d_{1}(\lambda)d_{2}(\lambda)\\ \vdots\\ D_{r}(\lambda)=d_{1}(\lambda)d_{2}(\lambda)\cdots d_{r}(\lambda)\\ D1(λ)=d1(λ)D2(λ)=d1(λ)d2(λ)⋮Dr(λ)=d1(λ)d2(λ)⋯dr(λ)
从而有
d 1 ( λ ) = D 1 ( λ ) d 2 ( λ ) = D 2 ( λ ) D 1 ( λ ) ⋮ d r ( λ ) = D r ( λ ) D r − 1 ( λ ) d_{1}(\lambda)=D_{1}(\lambda)\\ d_{2}(\lambda)=\frac{D_{2}(\lambda)}{D_{1}(\lambda)}\\ \vdots\\ d_{r}(\lambda)=\frac{D_{r}(\lambda)}{D_{r-1}(\lambda)}\\ d1(λ)=D1(λ)d2(λ)=D1(λ)D2(λ)⋮dr(λ)=Dr−1(λ)Dr(λ)
由于 A ( λ ) A(\lambda) A(λ)与上面Smith标准形具有相同各阶行列式因子
所以 A ( λ ) A(\lambda) A(λ)的各阶行列式因子为 D 1 ( λ ) , . . . , D r ( λ ) D_{1}(\lambda),...,D_{r}(\lambda) D1(λ),...,Dr(λ)
给定 A ( λ ) A(\lambda) A(λ)各行列式因子已经确定
从而不变因子
d 1 ( λ ) , . . . , d r ( λ ) d_{1}(\lambda),...,d_{r}(\lambda) d1(λ),...,dr(λ)
由行列式因子唯一确定, A ( λ ) A(\lambda) A(λ)的Smith标准形是唯一的
设 λ \lambda λ矩阵 A ( λ ) A(\lambda) A(λ)不变因子为 d 1 ( λ ) , . . . , d r ( λ ) d_{1}(\lambda),...,d_{r}(\lambda) d1(λ),...,dr(λ),在复数域内将它们分解为一次因式的幂乘积
d 1 ( λ ) = ( λ − λ 1 ) e 11 ( λ − λ 1 ) e 12 . . . ( λ − λ 1 ) e 1 s d 2 ( λ ) = ( λ − λ 1 ) e 21 ( λ − λ 1 ) e 22 . . . ( λ − λ 1 ) e 2 s ⋮ d r ( λ ) = ( λ − λ 1 ) e r 1 ( λ − λ 1 ) e r 2 . . . ( λ − λ 1 ) e r s d_{1}(\lambda)=(\lambda-\lambda_{1})^{e_{11}}(\lambda-\lambda_{1})^{e_{12}}...(\lambda-\lambda_{1})^{e_{1s}}\\ d_{2}(\lambda)=(\lambda-\lambda_{1})^{e_{21}}(\lambda-\lambda_{1})^{e_{22}}...(\lambda-\lambda_{1})^{e_{2s}}\\ \vdots\\ d_{r}(\lambda)=(\lambda-\lambda_{1})^{e_{r1}}(\lambda-\lambda_{1})^{e_{r2}}...(\lambda-\lambda_{1})^{e_{rs}}\\ d1(λ)=(λ−λ1)e11(λ−λ1)e12...(λ−λ1)e1sd2(λ)=(λ−λ1)e21(λ−λ1)e22...(λ−λ1)e2s⋮dr(λ)=(λ−λ1)er1(λ−λ1)er2...(λ−λ1)ers
其中 λ 1 , . . . , λ s \lambda_{1},...,\lambda_{s} λ1,...,λs为互异复数, e i j e_{ij} eij为非负整数,所有指数大于零的因子 ( λ − λ j ) e i j , e i j > 0 , i = 1... r , j = 1... s (\lambda-\lambda_{j})^{e_{ij}},e_{ij}>0,i=1...r,j=1...s (λ−λj)eij,eij>0,i=1...r,j=1...s称 A ( λ ) A(\lambda) A(λ)初等因子
且有
0 ≤ e 11 ≤ e 21 ≤ . . . ≤ e r 1 0 ≤ e 12 ≤ e 22 ≤ . . . ≤ e r 2 ⋮ 0 ≤ e 1 s ≤ e 2 s ≤ . . . ≤ e r s 0\le e_{11}\le e_{21}\le ...\le e_{r1}\\ 0\le e_{12}\le e_{22}\le ...\le e_{r2}\\ \vdots\\ 0\le e_{1s}\le e_{2s}\le ...\le e_{rs}\\ 0≤e11≤e21≤...≤er10≤e12≤e22≤...≤er2⋮0≤e1s≤e2s≤...≤ers
λ \lambda λ矩阵 A ( λ ) A(\lambda) A(λ)和 B ( λ ) B(\lambda) B(λ)等价的充要条件是相同的秩和初等因子
若 λ \lambda λ矩阵
A ( λ ) = [ A 1 ( λ ) A 1 ( λ ) ⋱ A t ( λ ) ] A(\lambda)= \begin{bmatrix} A_{1}(\lambda) & & & \\ & A_{1}(\lambda) & & \\ & & \ddots & \\ & & & A_{t}(\lambda) \end{bmatrix} A(λ)=⎣⎢⎢⎡A1(λ)A1(λ)⋱At(λ)⎦⎥⎥⎤
则 A 1 ( λ ) , . . . , A t ( λ ) A_{1}(\lambda),...,A_{t}(\lambda) A1(λ),...,At(λ)各个初等因子全体为 A ( λ ) A(\lambda) A(λ)全部初等因子
λ \lambda λ矩阵
A ( λ ) = [ f 1 ( λ ) ⋱ f t ( λ ) 0 ⋱ 0 ] A(\lambda)= \begin{bmatrix} f_{1}(\lambda) & & & & & \\ & \ddots & & & & \\ & & f_{t}(\lambda) & & & \\ & & & 0 & & \\ & & & & \ddots & \\ & & & & & 0 \end{bmatrix} A(λ)=⎣⎢⎢⎢⎢⎢⎢⎡f1(λ)⋱ft(λ)0⋱0⎦⎥⎥⎥⎥⎥⎥⎤
则 f 1 ( λ ) , . . . , f t ( λ ) f_{1}(\lambda),...,f_{t}(\lambda) f1(λ),...,ft(λ)所有一次因式幂全体为 A ( λ ) A(\lambda) A(λ)的全部初等因子
对于数字矩阵 A A A,我们称 λ I − A \lambda I-A λI−A行列式因子 / 不变因子为 A A A行列式因子 / 不变因子,称 λ I − A \lambda I-A λI−A的初等因子为 A A A的初等因子
设 A , B A,B A,B为两个 n n n阶数字矩阵,那么 A A A与 B B B相似的充要条件为它们特征矩阵 λ I − A \lambda I-A λI−A与 λ I − B \lambda I-B λI−B等价
对任意数字矩阵 A A A, ∣ λ I − A ∣ ≠ 0 |\lambda I-A|\neq0 ∣λI−A∣=0则
r a n k ( λ I − A ) = n rank(\lambda I-A)=n rank(λI−A)=n
两个同阶方阵 A , B A,B A,B相似充要条件为它们有相同初等因子
两个同阶方阵 A , B A,B A,B相似充要条件为它们有相同行列式因子 / 不变因子
称 n i n_{i} ni阶矩阵
J i = [ a i 1 a i 1 ⋱ ⋱ ⋱ 1 a i ] J_{i}= \begin{bmatrix} a_{i} & 1 & & & \\ & a_{i} & 1 & & \\ & & \ddots & \ddots & \\ & & & \ddots & 1\\ & & & & a_{i} \end{bmatrix} Ji=⎣⎢⎢⎢⎢⎡ai1ai1⋱⋱⋱1ai⎦⎥⎥⎥⎥⎤
为Jordan块
( λ I − J i ) (\lambda I-J_{i}) (λI−Ji)行列式因子为
D n i ( λ ) = ( λ − a i ) n i D n i − 1 ( λ ) = . . . = D 1 ( λ ) = 1 D_{n_{i}}(\lambda)=(\lambda-a_{i})^{n_{i}}\\ D_{n_{i}-1}(\lambda)=...=D_{1}(\lambda)=1 Dni(λ)=(λ−ai)niDni−1(λ)=...=D1(λ)=1
所以 J i J_{i} Ji初等因子为 ( λ − a i ) n i (\lambda-a_{i})^{n_{i}} (λ−ai)ni
若 J 1 . . . J s J_{1}...J_{s} J1...Js都为Jordan块,则称准对角矩阵
J = [ J 1 J 2 ⋱ J s ] J= \begin{bmatrix} J_{1} & & & \\ & J_{2} & & \\ & & \ddots & \\ & & & J_{s} \end{bmatrix} J=⎣⎢⎢⎡J1J2⋱Js⎦⎥⎥⎤
为Jordan标准型
J J J的初等因子为 ( λ − a 1 ) n 1 , ( λ − a 2 ) n 2 , . . . , ( λ − a s ) n s (\lambda-a_{1})^{n_{1}},(\lambda-a_{2})^{n_{2}},...,(\lambda-a_{s})^{n_{s}} (λ−a1)n1,(λ−a2)n2,...,(λ−as)ns
设 A A A为 n n n阶方阵,初等因子为
( λ − a 1 ) n 1 , . . . , ( λ − a s ) n s (\lambda-a_{1})^{n_{1}},...,(\lambda-a_{s})^{n_{s}} (λ−a1)n1,...,(λ−as)ns
则 A ∼ J = d i a g ( J 1 , . . . , J S ) A\sim J=diag(J_{1},...,J_{S}) A∼J=diag(J1,...,JS)
Jordan块排序不重要
n n n阶矩阵 A A A可对角化的充要条件为 A A A的初等因子都为一次因式
设 n n n阶方阵 A A A的Jordan标准形为 J J J,则存在可逆矩阵 P P P使得 P − 1 A P = J P^{-1}AP=J P−1AP=J,称 P P P为相似变换矩阵
对 n n n阶矩阵 A A A,若
r a n k ( λ i I − A ) = s 1 , r a n k ( λ i I − A ) 2 = s 2 , ⋮ r a n k ( λ i I − A ) l = s l , r a n k ( λ i I − A ) l + 1 = s l rank(\lambda_{i}I-A)=s_{1},\\ rank(\lambda_{i}I-A)^{2}=s_{2},\\ \vdots\\ rank(\lambda_{i}I-A)^{l}=s_{l},\\ rank(\lambda_{i}I-A)^{l+1}=s_{l} rank(λiI−A)=s1,rank(λiI−A)2=s2,⋮rank(λiI−A)l=sl,rank(λiI−A)l+1=sl
则对于 A A A的特征根 λ = λ i \lambda=\lambda_{i} λ=λi,共有 n − s 1 n-s_{1} n−s1个Jordan块,其中阶数最高为 l l l,阶数 ≥ 2 \ge 2 ≥2的Jordan块有 s 1 − s 2 s_{1}-s_{2} s1−s2个,阶数 ≥ 3 \ge 3 ≥3的Jordan块有 s 2 − s 3 s_{2}-s_{3} s2−s3个,…, l l l阶的有 s l − 1 − s l s_{l-1}-s_{l} sl−1−sl个