[선형대수] Lecture 28: Similar matrices and Jordan form

이재호·2025년 3월 24일

선형대수

목록 보기
27/31

https://ocw.mit.edu/courses/18-06-linear-algebra-spring-2010/video_galleries/video-lectures/

이번 강의에서는 다음과 같은 내용을 다룬다.

  • Similar Matrices (B=M1AMB=M^{-1}AM)
  • Jordan Form

이전 강의에서 Positive Definite Matrix (xTAx>0 (x0)x^TAx>0 \ (x\ne0)) 에 대해서 배웠다.
그리고 다음과 같이 시작한다.

만약 A,BA,B 모두 positive definte matrix 라면, A+BA+B 는 어떨까?

xTAx>0x^TAx>0
xTBx>0x^TBx>0
xT(A+B)x>0x^T(A+B)x>0

A+BA+B 역시 positive definite matrix 일 것이다.

그렇다면 ATAA^TA 는 어떨까?

A:m×n,rank(A)=nA:m\times n,rank(A)=n
ATA:square(n×n),symmetricA^TA:square(n\times n),symmetric
xT(ATA)x=(Ax)T(Ax)=Ax2:length20x^T(A^TA)x=(Ax)^T(Ax)={||Ax||}^2:length^2\ge 0
xT(ATA)x>0 (x0)ATA is positive definte matrix\therefore x^T(A^TA)x>0 \ (x\ne 0) \to \text{$A^TA$ is positive definte matrix}

이제 다음으로 Similar Matrices 에 대해서 알아보자.

A and B are similar (A,B:n×n)\text{$A$ and $B$ are similar ($A,B:n\times n$)}
B=M1AM  (for some matrix M)\to B=M^{-1}AM ~ \text{ (for some matrix $M$)}

구체적인 예시로 전의 강의에서 다뤘던 S1AS=ΛS^{-1}AS=\Lambda 수식이 있다. 여기서 AAΛ\Lambda 는 similar 하다.
ex.

A=[2112],Λ=[3001]A= \begin{bmatrix} 2 & 1 \\ 1 & 2 \\ \end{bmatrix}, \Lambda = \begin{bmatrix} 3 & 0 \\ 0 & 1 \\ \end{bmatrix}

즉, AAΛ\Lambda가 같은 eigenvalue λ=3,1\lambda=3,1 을 갖는 것을 알 수 있다.
다음 예시도 보자.

M1AMM^{-1}AM
=[1401][2112][1401]=[1401][2916]=[21516]=\begin{bmatrix} 1 & -4 \\ 0 & 1 \\ \end{bmatrix} \begin{bmatrix} 2 & 1 \\ 1 & 2 \\ \end{bmatrix} \begin{bmatrix} 1 & 4 \\ 0 & 1 \\ \end{bmatrix}= \begin{bmatrix} 1 & -4 \\ 0 & 1 \\ \end{bmatrix} \begin{bmatrix} 2 & 9 \\ 1 & 6 \\ \end{bmatrix}= \begin{bmatrix} -2 & -15 \\ 1 & 6 \\ \end{bmatrix}
A=[2112],B=[21516]A=\begin{bmatrix} 2 & 1 \\ 1 & 2 \\ \end{bmatrix}, B=\begin{bmatrix} -2 & -15 \\ 1 & 6 \\ \end{bmatrix}

그리고 BBAA와 마찬가지로 λ=3,1\lambda=3,1 인 것을 알 수 있다.

즉, 다음과 같은 similar matrices 특징 하나를 알았다.

Similar matrices have same λ’s.(1)\text{Similar matrices have same $\lambda$'s.} \tag{1}
Ax=λxAx=\lambda x
AMM1x=λxAMM^{-1}x=\lambda x
M1AMBM1x=λM1x\underbrace{M^{-1}AM}_{B}M^{-1}x=\lambda M^{-1}x
BM1x=λM1xBM^{-1}x=\lambda M^{-1}x
B(M1x)=λ(M1x)B(M^{-1}x)=\lambda (M^{-1}x)

위와 같이 AABB는 같은 eigenvalue 를 갖는다는 것을 증명할 수 있다.


그렇다면 만약 λ1=λ2\lambda_1=\lambda_2 와 같은 경우는 어떨까?
ex.

λ1=λ2=4\lambda_1=\lambda_2=4
A1=[4004],A2=[4104]A_1=\begin{bmatrix} 4 & 0 \\ 0 & 4 \\ \end{bmatrix}, A_2=\begin{bmatrix} 4 & 1 \\ 0 & 4 \\ \end{bmatrix}

위 경우 M1A1MM^{-1}A_1M 의 결과는 A1A_1 그 자체 하나의 결과만 나올 것이다.

M1A1M=M1[4004]M=[4004]=A1M^{-1}A_1M=M^{-1}\begin{bmatrix} 4 & 0 \\ 0 & 4 \\ \end{bmatrix}M=\begin{bmatrix} 4 & 0 \\ 0 & 4 \\ \end{bmatrix}=A_1

반면에 A2A_2 는 2개 이상의 결과가 나올 것이다. 따라서 A1A_1A2A_2는 같은 eigenvalue 를 가졌음에도 simliar 하지 않다.

A2A_2 는 diagonal matrix 가 아니기 때문에 여러 개의 결과가 나올 수 있으며,
이러한 행렬을 " Jordan Form " 이라고 한다.

Jordan Form 행렬을 더 찾으면 아래와 같다.

[4104],[5113],[40174],...,[a??8a]\begin{bmatrix} 4 & 1 \\ 0 & 4 \\ \end{bmatrix}, \begin{bmatrix} 5 & 1 \\ -1 & 3 \\ \end{bmatrix}, \begin{bmatrix} 4 & 0 \\ 17 & 4 \\ \end{bmatrix}, ..., \begin{bmatrix} a & ? \\ ? & 8-a \\ \end{bmatrix}
trace=8,det=16trace=8,det=16

따라서 "(1) Similar matrices have same λ’s.\text{Similar matrices have same $\lambda$'s.}" 에서 다음과 같이 조건이 추가되어야 한다.

Similar matrices have same λ’s and same # egienvectors(2)\text{Similar matrices have same $\lambda$'s and same \# egienvectors} \tag{2}

이어서 다음 예시를 보자.

A1=[0100001000000000]A_1= \begin{bmatrix} 0 & 1 & 0 & 0 \\ 0 & 0 & 1 & 0 \\ 0 & 0 & 0 & 0 \\ 0 & 0 & 0 & 0 \\ \end{bmatrix}

A1A_1에 대해서 정리하면 다음과 같다.

  • λ=0,0,0,0\lambda=0,0,0,0
  • 2 eigenvectors
  • dim(N(A1))=2\dim(N(A_1))=2
  • 2 missing eigenvectors

이어서 A2A_2도 보자.

A2=[0100000000010000]A_2= \begin{bmatrix} 0 & 1 & 0 & 0 \\ 0 & 0 & 0 & 0 \\ 0 & 0 & 0 & 1 \\ 0 & 0 & 0 & 0 \\ \end{bmatrix}
  • λ=0,0,0,0\lambda=0,0,0,0
  • 2 eigenvectors
  • dim(N(A2))=2\dim(N(A_2))=2
  • 2 missing eigenvectors

A1A_1A2A_2 과 similar 한 것처럼 보이지만 그렇지 않다. 이유는 Jordan block 이 다르기 때문이다.

Jordan block J\text{Jordan block J}
Ji=[λi1000λi1000...1000λi]J_i= \begin{bmatrix} \lambda_i & 1 & 0 & 0 \\ 0 & \lambda_i & 1 & 0 \\ 0 & 0 & ... & 1 \\ 0 & 0 & 0 & \lambda_i \\ \end{bmatrix}

즉 위에서 A1,A2A_1,A_2의 Jordan block 은 다음과 같다.

A1=[0100001000000000]A_1= \begin{bmatrix} 0 & 1 & 0 & 0 \\ 0 & 0 & 1 & 0 \\ 0 & 0 & 0 & 0 \\ 0 & 0 & 0 & 0 \\ \end{bmatrix}
[010001000],[0]\to\begin{bmatrix} 0 & 1 & 0 \\ 0 & 0 & 1 \\ 0 & 0 & 0 \\ \end{bmatrix}, \begin{bmatrix} 0\end{bmatrix}
A2=[0100000000010000]A_2= \begin{bmatrix} 0 & 1 & 0 & 0 \\ 0 & 0 & 0 & 0 \\ 0 & 0 & 0 & 1 \\ 0 & 0 & 0 & 0 \\ \end{bmatrix}
[0100],[0100]\to \begin{bmatrix} 0 & 1 \\ 0 & 0 \end{bmatrix},\begin{bmatrix} 0 & 1 \\ 0 & 0 \end{bmatrix}

따라서 "(2) Similar matrices have same λ’s and same # egienvectors\text{Similar matrices have same $\lambda$'s and same \# egienvectors}" 에서 다음과 같이 조건을 수정한다.

Every square matrix A is similar to a Jordan matrix J(3)\text{Every square matrix A is similar to a Jordan matrix $J$} \tag{3}
J=[J10000J20000...0000Jd]J= \begin{bmatrix} J_1 & 0 & 0 & 0 \\ 0 & J_2 & 0 & 0 \\ 0 & 0 & ... & 0 \\ 0 & 0 & 0 & J_d \\ \end{bmatrix}
# blocks=# eigenvectors\text{\# blocks}=\text{\# eigenvectors}

그리고 good case에 해당하는 게 초기에 본 J=ΛJ=\Lambda 인 경우이다.

profile
천천히, 그리고 꾸준히.

0개의 댓글