矩阵的逆、广义逆

1. 矩阵的逆

定义:对矩阵 A A A,若存在矩阵 B B B使得 A B = B A = I AB=BA=I AB=BA=I,则 B B B唯一,称之为矩阵的逆,记为 A − 1 A^{-1} A1

矩阵的逆具有如下基本性质:

(i) A − 1 A^{-1} A1存在当且仅当 ∣ A ∣ ≠ 0 |A|\neq0 A=0

证明:若 A − 1 A^{-1} A1存在,对 A A − 1 = E AA^{-1}=E AA1=E两边取行列式,得 ∣ A ∣ ∣ A − 1 ∣ = ∣ E ∣ = 1 |A||A^{-1}|=|E|=1 AA1=E=1,因而 ∣ A ∣ ≠ 0 |A|\neq0 A=0

∣ A ∣ ≠ 0 |A|\neq0 A=0时,由行列式按一行展开的公式知 A A ∗ = A ∗ A = ∣ A ∣ E AA^*=A^*A=|A|E AA=AA=AE,可写 A ( 1 ∣ A ∣ A ∗ ) = ( 1 ∣ A ∣ A ∗ ) A = E A(\frac{1}{|A|}A^*)=(\frac{1}{|A|}A^*)A=E A(A1A)=(A1A)A=E
即得 A − 1 = A ∗ ∣ A ∣ A^{-1}=\frac{A^*}{|A|} A1=AA。% A − 1 A^{-1} A1存在。

(ii) ( A ′ ) − 1 = ( A − 1 ) ′ (A')^{-1}=(A^{-1})' (A)1=(A1)

证明:由 A A − 1 = A − 1 A = E AA^{-1}=A^{-1}A=E AA1=A1A=E两边取转置: ( A − 1 ) ′ A ′ = A ′ ( A − 1 ) ′ = E ′ = E (A^{-1})'A'=A'(A^{-1})'=E'=E (A1)A=A(A1)=E=E于是 ( A ′ ) − 1 = ( A − 1 ) ′ (A')^{-1}=(A^{-1})' (A)1=(A1)

(iii) ( c A ) − 1 = c − 1 A − 1 (cA)^{-1}=c^{-1}A^{-1} (cA)1=c1A1,其中 c c c为非零实数。

(iv) ( A B ) − 1 = B − 1 A − 1 (AB)^{-1}=B^{-1}A^{-1} (AB)1=B1A1,如果 A − 1 A^{-1} A1 B − 1 B^{-1} B1都存在。

证明:由 ( A B ) ( B − 1 A − 1 ) = ( B − 1 A − 1 ) ( A B ) = E (AB)(B^{-1}A^{-1})=(B^{-1}A^{-1})(AB)=E (AB)(B1A1)=(B1A1)(AB)=E即得。

(v) d i a g ( a 1 , . . . , a n ) − 1 = d i a g ( a 1 − 1 , . . . , a n − 1 ) diag(a_1,...,a_n)^{-1}=diag(a_1^{-1},...,a_n^{-1}) diag(a1,...,an)1=diag(a11,...,an1)
更一般地,如果 A A A是分块对角阵, A = d i a g ( A 1 , . . . , A k ) A=diag(A_1,...,A_k) A=diag(A1,...,Ak),其中 A j , 1 ≤ j ≤ k A_j,1\leq j\leq k Aj,1jk非奇异,即 ∣ A j ∣ ≠ 0 , 1 ≤ j ≤ k |A_j|\neq0,1\leq j\leq k Aj=0,1jk,则 A − 1 = d i a g ( A 1 − 1 , . . . , A k − 1 ) A^{-1}=diag(A_1^{-1},...,A_k^{-1}) A1=diag(A11,...,Ak1)

(vi) 对于任意的 n × n n\times n n×n阶非奇异阵 A A A n × q n\times q n×q阶矩阵 U U U q × n q\times n q×n阶矩阵 V V V,有
( A + U V ) − 1 = A − 1 − A − 1 U ( I q + V A − 1 U ) − 1 V A − 1 (A+UV)^{-1}=A^{-1}-A^{-1}U(I_q+VA^{-1}U)^{-1}VA^{-1} (A+UV)1=A1A1U(Iq+VA1U)1VA1
该公式的一个简单应用为:

1 n 1_n 1n n n n维全1向量, J n = 1 n 1 n ′ J_n=1_n1_n' Jn=1n1n,实数 a a a b b b满足 a ≠ 0 a\neq0 a=0 a + n b ≠ 0 a+nb\neq0 a+nb=0,则有:
( a I n + b J n ) − 1 = 1 a ( I n − b a + n b J n ) (aI_n+bJ_n)^{-1}=\frac{1}{a}(I_n-\frac{b}{a+nb}J_n) (aIn+bJn)1=a1(Ina+nbbJn)同时有 ∣ a I n + b J n ∣ = a n − 1 ( a + n b ) |aI_n+bJ_n|=a^{n-1}(a+nb) aIn+bJn=an1(a+nb)

证明:下面证明更一般的Woodbury公式: ( A − U C − 1 V ) − 1 = A − 1 + A − 1 U ( C − V A − 1 U ) − 1 V A − 1 (A-UC^{-1}V)^{-1}=A^{-1}+A^{-1}U(C-VA^{-1}U)^{-1}VA^{-1} (AUC1V)1=A1+A1U(CVA1U)1VA1
其中 C C C可逆。
考虑矩阵 [ A U V C ] \left[\begin{matrix} A&U\\ V&C \end{matrix}\right] [AVUC],由于 A A A可逆,

消除第(2,1)项(行变换):
[ I 0 − V A − 1 I ] [ A U V C ] = [ A U 0 C − V A − 1 U ] \left[\begin{matrix} I&0\\ -VA^{-1}&I \end{matrix}\right]\left[\begin{matrix} A&U\\ V&C \end{matrix}\right]=\left[\begin{matrix} A&U\\ 0&C-VA^{-1}U \end{matrix}\right] [IVA10I][AVUC]=[A0UCVA1U]
消除第(1,2)项(列变换):
[ A U V C ] [ I − A − 1 U 0 I ] = [ A 0 V C − V A − 1 U ] \left[\begin{matrix} A&U\\ V&C \end{matrix}\right]\left[\begin{matrix} I&-A^{-1}U\\ 0&I \end{matrix}\right]=\left[\begin{matrix} A&0\\ V&C-VA^{-1}U \end{matrix}\right] [AVUC][I0A1UI]=[AV0CVA1U]
于是有:
[ I 0 − V A − 1 I ] [ A U V C ] [ I − A − 1 U 0 I ] = [ A 0 0 C − V A − 1 U ] \left[\begin{matrix} I&0\\ -VA^{-1}&I \end{matrix}\right]\left[\begin{matrix} A&U\\ V&C \end{matrix}\right]\left[\begin{matrix} I&-A^{-1}U\\ 0&I \end{matrix}\right]=\left[\begin{matrix} A&0\\ 0&C-VA^{-1}U \end{matrix}\right] [IVA10I][AVUC][I0A1UI]=[A00CVA1U]
进而:
[ A U V C ] = [ I 0 V A − 1 I ] [ A 0 0 C − V A − 1 U ] [ I A − 1 U 0 I ] \left[\begin{matrix} A&U\\ V&C \end{matrix}\right]=\left[\begin{matrix} I&0\\ VA^{-1}&I \end{matrix}\right]\left[\begin{matrix} A&0\\ 0&C-VA^{-1}U \end{matrix}\right]\left[\begin{matrix} I&A^{-1}U\\ 0&I \end{matrix}\right] [AVUC]=[IVA10I][A00CVA1U][I0A1UI]
两边同时取逆:
[ A U V C ] − 1 = [ I A − 1 U 0 I ] − 1 [ A 0 0 C − V A − 1 U ] − 1 [ I 0 V A − 1 I ] − 1 = [ I − A − 1 U 0 I ] [ A − 1 0 0 ( C − V A − 1 U ) − 1 ] [ I 0 − V A − 1 I ] = [ A − 1 + A − 1 U ( C − V A − 1 U ) − 1 V A − 1 − A − 1 U ( C − V A − 1 U ) − 1 − ( C − V A − 1 U ) − 1 V A − 1 ( C − V A − 1 U ) − 1 ] \begin{aligned} \left[\begin{matrix} A&U\\ V&C \end{matrix}\right]^{-1}&=\left[\begin{matrix} I&A^{-1}U\\ 0&I \end{matrix}\right]^{-1}\left[\begin{matrix} A&0\\ 0&C-VA^{-1}U \end{matrix}\right]^{-1}\left[\begin{matrix} I&0\\ VA^{-1}&I \end{matrix}\right]^{-1}\\ &=\left[\begin{matrix} I&-A^{-1}U\\ 0&I \end{matrix}\right]\left[\begin{matrix} A^{-1}&0\\ 0&(C-VA^{-1}U)^{-1} \end{matrix}\right]\left[\begin{matrix} I&0\\ -VA^{-1}&I \end{matrix}\right]\\ &=\left[\begin{matrix} A^{-1}+A^{-1}U(C-VA^{-1}U)^{-1}VA^{-1}&-A^{-1}U(C-VA^{-1}U)^{-1}\\ -(C-VA^{-1}U)^{-1}VA^{-1}&(C-VA^{-1}U)^{-1} \end{matrix} \right] \end{aligned} [AVUC]1=[I0A1UI]1[A00CVA1U]1[IVA10I]1=[I0A1UI][A100(CVA1U)1][IVA10I]=[A1+A1U(CVA1U)1VA1(CVA1U)1VA1A1U(CVA1U)1(CVA1U)1]
同样的,由于 C C C可逆,有:
[ A U V C ] = [ I − U C − 1 0 I ] [ A − U C − 1 V 0 0 C ] [ I 0 − C − 1 V I ] \left[\begin{matrix} A&U\\ V&C \end{matrix}\right]=\left[\begin{matrix} I&-UC^{-1}\\ 0&I \end{matrix}\right]\left[\begin{matrix} A-UC^{-1}V&0\\ 0&C \end{matrix}\right]\left[\begin{matrix} I&0\\ -C^{-1}V&I \end{matrix}\right] [AVUC]=[I0UC1I][AUC1V00C][IC1V0I]
两边同时取逆:
[ A U V C ] − 1 = [ I 0 C − 1 V I ] − 1 [ A − U C − 1 V 0 0 C ] − 1 [ I U C − 1 0 I ] − 1 = [ I 0 − C − 1 V I ] [ ( A − U C − 1 V ) − 1 0 0 C − 1 ] [ I − U C − 1 0 I ] = [ ( A − U C − 1 V ) − 1 − ( A − U C − 1 V ) − 1 U C − 1 − C − 1 V ( A − U C − 1 V ) − 1 C − 1 V ( A − U C − 1 V ) − 1 U C − 1 + C − 1 ] \begin{aligned} \left[\begin{matrix} A&U\\ V&C \end{matrix}\right]^{-1}&=\left[\begin{matrix} I&0\\ C^{-1}V&I \end{matrix}\right]^{-1}\left[\begin{matrix} A-UC^{-1}V&0\\ 0&C \end{matrix}\right]^{-1}\left[\begin{matrix} I&UC^{-1}\\ 0&I \end{matrix}\right]^{-1}\\ &=\left[\begin{matrix} I&0\\ -C^{-1}V&I \end{matrix}\right]\left[\begin{matrix} (A-UC^{-1}V)^{-1}&0\\ 0&C^{-1} \end{matrix}\right]\left[\begin{matrix} I&-UC^{-1}\\ 0&I \end{matrix}\right]\\ &=\left[\begin{matrix} (A-UC^{-1}V)^{-1}&-(A-UC^{-1}V)^{-1}UC^{-1}\\ -C^{-1}V(A-UC^{-1}V)^{-1}&C^{-1}V(A-UC^{-1}V)^{-1}UC^{-1}+C^{-1} \end{matrix} \right] \end{aligned} [AVUC]1=[IC1V0I]1[AUC1V00C]1[I0UC1I]1=[IC1V0I][(AUC1V)100C1][I0UC1I]=[(AUC1V)1C1V(AUC1V)1(AUC1V)1UC1C1V(AUC1V)1UC1+C1]
对比第(1,1)项有:
( A − U C − 1 V ) − 1 = A − 1 + A − 1 U ( C − V A − 1 U ) − 1 V A − 1 (A-UC^{-1}V)^{-1}=A^{-1}+A^{-1}U(C-VA^{-1}U)^{-1}VA^{-1} (AUC1V)1=A1+A1U(CVA1U)1VA1
证毕。

( a I n + b J n ) − 1 = 1 a ( I n + b a J n ) − 1 (aI_n+bJ_n)^{-1}=\frac{1}{a}(I_n+\frac{b}{a}J_n)^{-1} (aIn+bJn)1=a1(In+abJn)1,取 U = b a 1 n U=\frac{b}{a}1_n U=ab1n V = 1 n ′ V=1_n' V=1n即得。

∣ a I n + b J n ∣ = ∣ a + b b . . . b b a + b . . . b . . . . . . . . . . . . b b . . . a + b ∣ = ( a + n b ) ∣ 1 b . . . b 1 a + b . . . b . . . . . . . . . . . . 1 b . . . a + b ∣ = ( a + n b ) ∣ 1 b . . . b 0 a . . . b . . . . . . . . . . . . 0 0 . . . a ∣ = a n − 1 ( a + n b ) \begin{aligned} |aI_n+bJ_n|&=\left|\begin{matrix} a+b&b&...&b\\ b&a+b&...&b\\ ...&...&...&...\\ b&b&...&a+b \end{matrix} \right|\\&=(a+nb)\left|\begin{matrix} 1&b&...&b\\ 1&a+b&...&b\\ ...&...&...&...\\ 1&b&...&a+b \end{matrix} \right|\\&=(a+nb)\left|\begin{matrix} 1&b&...&b\\ 0&a&...&b\\ ...&...&...&...\\ 0&0&...&a \end{matrix} \right|=a^{n-1}(a+nb) \end{aligned} aIn+bJn=a+bb...bba+b...b............bb...a+b=(a+nb)11...1ba+b...b............bb...a+b=(a+nb)10...0ba...0............bb...a=an1(a+nb)

2. 广义逆

2.1 定义

A A A是任意 m × n m\times n m×n矩阵,称矩阵 A † A^\dagger A A A A的广义逆矩阵,若 A † A^\dagger A满足一下四个条件(常称为Moore-Penrose条件):

(i) A A † A = A AA^\dagger A=A AAA=A;
(ii) A † A A † = A † A^\dagger AA^\dagger=A^\dagger AAA=A;
(iii) A A † AA^\dagger AA为Hermitian矩阵,即 A A † = ( A A † ) H AA^\dagger=(AA^\dagger)^H AA=(AA)H;
(iv) A † A A^\dagger A AA为Hermitian矩阵,即 A † A = ( A † A ) H A^\dagger A=(A^\dagger A)^H AA=(AA)H.

根据满足Moore-Penrose条件的多少,可以对广义逆矩阵进行分类:

(1)满足全部四个条件的矩阵 A † A^\dagger A称为 A A A的Moore-Penrose逆矩阵;
(2)只满足条件(i)和(ii)的矩阵称为 A A A的自反广义逆矩阵;
(3)满足条件(i)(ii)(iii)的矩阵称为 A A A的正规化广义逆矩阵;
(4)满足条件(i)(ii)(iv)的矩阵称为 A A A的弱广义逆矩阵。

2.2 广义逆的存在唯一性

任意 m × n m\times n m×n矩阵 A A A,其Moore-Penrose逆矩阵存在且唯一。

证明:
(存在性)
A A A为零矩阵,显然零矩阵就是 A A A的广义逆。

A ≠ 0 A\neq 0 A=0,则 A A A有奇异值分解,即存在正交矩阵 U U U V V V使得
A = U d i a g { σ 1 , . . , σ r , 0 , . . . , 0 } V H A=Udiag\{\sigma_1,..,\sigma_r,0,...,0\}V^H A=Udiag{σ1,..,σr,0,...,0}VH

A † = V d i a g { σ 1 − 1 , . . , σ r − 1 , 0 , . . . , 0 } U H A^\dagger=Vdiag\{\sigma_1^{-1},..,\sigma_r^{-1},0,...,0\}U^H A=Vdiag{σ11,..,σr1,0,...,0}UH,可以验证 A † A^\dagger A就是 A A A的广义逆。

(唯一性)
B B B A A A的另一个广义逆,即满足Moore-Penrose四个条件。
A † = A † A A † = A † ( A A † ) H = A † ( A † ) H A H = A † ( A † ) H ( A B A ) H = A † ( A † ) H A H B H A H = A † ( A A † ) H ( A B ) H = A † A A † A B = A † A B = A † A ( B A B ) = A † A A H B H B = A H ( A † ) H A H B H B = ( A A † A ) H B H B = A H B H B = ( B A ) H B = B A B = B \begin{aligned} A^\dagger&=A^\dagger AA^\dagger=A^\dagger (AA^\dagger)^H=A^\dagger (A^\dagger)^HA^H=A^\dagger (A^\dagger)^H(ABA)^H=A^\dagger (A^\dagger)^HA^HB^HA^H \\&=A^\dagger (AA^\dagger)^H(AB)^H=A^\dagger AA^\dagger AB=A^\dagger AB=A^\dagger A(BAB)=A^\dagger AA^HB^HB \\&=A^H(A^\dagger)^HA^HB^HB=(AA^\dagger A)^HB^HB=A^HB^HB=(BA)^HB=BAB=B \end{aligned} A=AAA=A(AA)H=A(A)HAH=A(A)H(ABA)H=A(A)HAHBHAH=A(AA)H(AB)H=AAAAB=AAB=AA(BAB)=AAAHBHB=AH(A)HAHBHB=(AAA)HBHB=AHBHB=(BA)HB=BAB=B

其他类型的广义逆矩阵均不是唯一的,关于其唯一性条件可参考:曹东, 杨成. 广义逆矩阵的唯一性[J]. 中南民族大学学报(自然科学版), 1998(1):78-80.

你可能感兴趣的:(矩阵论)