고유값과 고유벡터

(고유벡터에서 넘어옴)


정의[편집 | 원본 편집]

F 위의 벡터공간 V가 주어지고 [math]\displaystyle{ \mathbf{v}\in V }[/math]를 영이 아닌 임의의 벡터라고 하자. 선형연산자 [math]\displaystyle{ L:V\to V }[/math]에 대해

[math]\displaystyle{ L(\mathbf{v})=\lambda \mathbf{v} }[/math]

스칼라 λ가 존재하면 [math]\displaystyle{ \mathbf{v} }[/math]L고유벡터(Eigenvector)라고 하고, λ를 고유값(Eigenvalue)이라고 한다.

선형연산자 L을 나타내는 행렬A라고 하자. 그러면 방정식

[math]\displaystyle{ A\mathbf{v}=\lambda \mathbf{v} }[/math]

가 된다. 즉,

[math]\displaystyle{ (A-\lambda I)\mathbf{v}=O }[/math]

이고 이 방정식이 영이 아닌 근을 가지므로

[math]\displaystyle{ \det(A-\lambda I)=0 }[/math]

이다.

x를 체 F의 원소라고 하자. [math]\displaystyle{ A-xI }[/math]A특성행렬(characteristic matrix), [math]\displaystyle{ \det(A-xI) }[/math]A특성다항식(characteristic polynomial), 방정식 [math]\displaystyle{ \det(A-xI)=0 }[/math]A특성방정식(characteristic equation)이라고 한다. 특성방정식의 근의 집합은 A스펙트럼(spectrum)이라 한다.

고유값의 존재성[편집 | 원본 편집]

선형연산자 [math]\displaystyle{ T:\mathbb{R}^2\to \mathbb{R}^2 }[/math]가 다음과 같이 정의되었다고 하자.

[math]\displaystyle{ T(x,y)=(x-y,x+y) }[/math]

이때 [math]\displaystyle{ T }[/math]의 고윳값이 존재한다고 가정하고 [math]\displaystyle{ \lambda }[/math]로 쓰자. 그러면

[math]\displaystyle{ (x-y,x+y)=(\lambda x,\lambda y) }[/math]

이고 따라서 [math]\displaystyle{ (1-\lambda)x=y,(1-\lambda)y=-x }[/math]이다. [math]\displaystyle{ \lambda=1 }[/math]이면 [math]\displaystyle{ x=y=0 }[/math]이 되므로 불가능하다. 따라서 [math]\displaystyle{ \lambda\ne 1 }[/math]이다. 그러면

[math]\displaystyle{ x= -(1-\lambda)y=-(1-\lambda)^2 x }[/math]

이므로

[math]\displaystyle{ (1+(1-\lambda)^2)x=0 }[/math]

이다. 따라서 [math]\displaystyle{ T }[/math]의 고유값은 존재하지 않는다. 그러나 [math]\displaystyle{ \mathbb{R} }[/math][math]\displaystyle{ \mathbb{C} }[/math]로 바꾸면 [math]\displaystyle{ T }[/math]의 고유값은 [math]\displaystyle{ \lambda_1=1+i }[/math], [math]\displaystyle{ \lambda_2=1-i }[/math]임을 알 수 있다.

일반적으로 선형연산자 [math]\displaystyle{ T:\mathbb{C}^n\to \mathbb{C}^n }[/math]의 고유값은 존재한다.[1][2] [math]\displaystyle{ \mathbf{x} }[/math]를 영이 아닌 임의의 벡터라고 하자. 집합 [math]\displaystyle{ S }[/math]

[math]\displaystyle{ S=\{\mathbf{x},T\mathbf{x},T^2\mathbf{x},\cdots,T^n\mathbf{x}\} }[/math]

로 정의하면 [math]\displaystyle{ S }[/math]의 원소의 수는 [math]\displaystyle{ n+1 }[/math]개이므로 [math]\displaystyle{ S }[/math]일차종속이다. 따라서

[math]\displaystyle{ a_0 \mathbf{x}+a_1 T\mathbf{x}+ a_2 T^2\mathbf{x}+\cdots + a_n T^n \mathbf{x}=\mathbf{0} }[/math]

를 만족하는 [math]\displaystyle{ a_0,a_1,\cdots, a_n \in \mathbb{C} }[/math]가 존재한다. 이때 [math]\displaystyle{ a_1,a_2,\cdots,a_n }[/math] 중 하나는 반드시 영이 아닌데, [math]\displaystyle{ a_1,a_2,\cdots,a_n }[/math]이 모두 영이라면 [math]\displaystyle{ a_0 \mathbf{x}=0 }[/math]이 되어 모순이기 때문이다. [math]\displaystyle{ a_i\ne 0 }[/math]을 만족하는 [math]\displaystyle{ i\in \{0,1,\cdots, n\} }[/math] 중 가장 큰 값을 [math]\displaystyle{ m }[/math]이라 하자. 다항식 [math]\displaystyle{ p(x) }[/math]

[math]\displaystyle{ p(x)=a_0 + a_1 x + a_2 x^2 +\cdots + a_m x^m }[/math]

으로 정의하면, 대수학의 기본 정리에 의해 [math]\displaystyle{ p(x)=c(x-\lambda_1)(x-\lambda_2)\cdots (x-\lambda_m) }[/math][math]\displaystyle{ c,\lambda_1,\lambda_2,\cdots,\lambda_m\in \mathbb{C} }[/math]가 존재한다. 따라서

[math]\displaystyle{ \begin{align} \mathbf{0}&=a_0 \mathbf{x}+a_1 T\mathbf{x}+ a_2 T^2\mathbf{x}+\cdots + a_n T^n \mathbf{x}\\ &=(a_0 + a_1 T +a_2 T^2 +\cdots + a_n T^n )\mathbf{x}\\ &=c(T-\lambda_1 I)(T-\lambda_2 I)\cdots (T-\lambda_m I)\mathbf{x} \end{align} }[/math]

이고, 따라서 [math]\displaystyle{ T-\lambda_i I }[/math] 중 하나는 일대일 함수가 아니다. 따라서 [math]\displaystyle{ T }[/math]는 고유값을 가진다.

더욱이, 대수적으로 닫힌 체 [math]\displaystyle{ F }[/math] 위에서 정의된 임의의 자기준동형사상 [math]\displaystyle{ T:F^n \to F^n }[/math]의 고유값은 존재한다.

예시[편집 | 원본 편집]

선형연산자 L을 나타내는 행렬 A가 다음과 같이 주어졌다고 하자.

[math]\displaystyle{ A=\begin{bmatrix} 8 & 5 & 6 & 0\\ 0 & -2 & 0 & 0\\ -10 & -5 & -8 & 0\\ 2 & 1 & 1 & 2 \end{bmatrix} }[/math]

그러면 A의 특성다항식은

[math]\displaystyle{ \det(A-xI)=\begin{vmatrix} 8-x & 5 & 6 & 0\\ 0 & -2-x & 0 & 0\\ -10 & -5 & -8-x & 0\\ 2 & 1 & 1 & 2-x \end{vmatrix}=x^4-8x^2+16 }[/math]

이므로 특성방정식 [math]\displaystyle{ \det(A-xI)=0 }[/math]를 만족하는 x[math]\displaystyle{ x=\pm 2 }[/math]이다. 따라서 A의 고유값은 2와 -2이다. 한편

[math]\displaystyle{ A-2I=\begin{bmatrix} 6 & 5 & 6 & 0\\ 0 & -4 & 0 & 0\\ -10 & -5 & -10 & 0\\ 2 & 1 & 1 & 0 \end{bmatrix} }[/math]
[math]\displaystyle{ A+2I=\begin{bmatrix} 10 & 5 & 6 & 0\\ 0 & 0 & 0 & 0\\ -10 & -5 & -6 & 0\\ 2 & 1 & 1 & 4 \end{bmatrix} }[/math]

이고 [math]\displaystyle{ A-2I,A+2I }[/math]기본행연산(elementary row operation)을 거쳐 기약행사다리꼴(reduced row echelon form)로 만들면 각각

[math]\displaystyle{ \begin{bmatrix} 1 & 0 & 0 & 0\\ 0 & 1 & 0 & 0\\ 0 & 0 & 1 & 0\\ 0 & 0 & 0 & 0\\ \end{bmatrix},\;\begin{bmatrix} 1 & \frac{1}{2} & 0 & 12\\ 0 & 0 & 1 & -20\\ 0 & 0 & 0 & 0\\ 0 & 0 & 0 & 0 \end{bmatrix} }[/math]

이 되므로 각 방정식의 해, 즉 고유벡터는 [math]\displaystyle{ (A-2I)\mathbf{x}=\mathbf{0} }[/math]의 경우

[math]\displaystyle{ \mathbf{x}=c\begin{bmatrix} 0\\ 0\\ 0\\ 1 \end{bmatrix} }[/math]

이고 [math]\displaystyle{ (A+2I)\mathbf{x}=\mathbf{0} }[/math]의 경우

[math]\displaystyle{ \mathbf{x}=\begin{bmatrix} -\frac{1}{2}a-12b\\ a\\ 20b\\ b \end{bmatrix}=a\begin{bmatrix}-\frac{1}{2} \\ 1 \\ 0 \\ 0\end{bmatrix}+b\begin{bmatrix}-12\\ 0 \\ 20 \\ 1\end{bmatrix} }[/math]

이다. 이때 고유벡터의 정의에 의해 [math]\displaystyle{ abc\ne 0 }[/math]이다.

고유공간[편집 | 원본 편집]

F 위의 벡터공간 V에서 정의된 선형연산자 L의 고윳값 λ가 주어졌을 때, 모든 고유벡터와 영벡터의 집합 [math]\displaystyle{ \mathbf{V}_\lambda }[/math]벡터공간을 이룬다. 왜냐 하면 임의의 [math]\displaystyle{ \mathbf{v}_1,\mathbf{v}_2\in\mathbf{V}_\lambda }[/math][math]\displaystyle{ c\in F }[/math]에 대해

[math]\displaystyle{ L(\mathbf{v}_1+\mathbf{v}_2)=L(\mathbf{v}_1)+L(\mathbf{v}_2)=\lambda\mathbf{v}_1+\lambda\mathbf{v}_2=\lambda(\mathbf{v}_1+\mathbf{v}_2) }[/math]
[math]\displaystyle{ L(c\mathbf{v}_1)=cL(\mathbf{v}_1)=c\lambda \mathbf{v}_1=\lambda(c\mathbf{v}_1) }[/math]

이므로 V의 부분공간(subspace)이기 때문이다. 이 벡터공간을 고윳값 λ이 연관된 L의 고유공간(eigenspace)이라고 한다. [math]\displaystyle{ \mathbf{V}_\lambda }[/math]차원(dimension)은 λ의 기하중복도(geometric multiplicity)라고 한다.

[math]\displaystyle{ \lambda_1,\lambda_2,\cdots,\lambda_r }[/math]가 선형연산자 L의 서로 다른 고유값이고, [math]\displaystyle{ \mathbf{v}_1,\mathbf{v}_2,\cdots,\mathbf{v}_r }[/math][math]\displaystyle{ \lambda_1,\lambda_2,\cdots,\lambda_r }[/math]와 연관된 고유벡터라고 하자. 이때 [math]\displaystyle{ \mathbf{v}_1,\mathbf{v}_2,\cdots,\mathbf{v}_r }[/math]는 선형독립이다.

[math]\displaystyle{ \mathbf{v}_1\ne \mathbf{0} }[/math]이므로 [math]\displaystyle{ \{\mathbf{v}_1\} }[/math]선형독립인 집합이다. 이제 [math]\displaystyle{ 1\le k\le r }[/math]인 정수 k에 대해 [math]\displaystyle{ \{\mathbf{v}_1,\mathbf{v}_2,\cdots,\mathbf{v}_k\} }[/math]가 선형독립이라고 가정하자. 만약 [math]\displaystyle{ k=r }[/math]이면 증명이 끝나므로 [math]\displaystyle{ k\lt r }[/math]라고 가정하자. 방정식

[math]\displaystyle{ c_1\mathbf{v}_1+c_2\mathbf{v}_2+\cdots+c_{k+1}\mathbf{v}_{k+1}=\mathbf{0} }[/math]

에서 L에 의한 연산을 거치면

[math]\displaystyle{ c_1\lambda_1\mathbf{v}_1+c_2\lambda_2\mathbf{v}_2+\cdots+c_{k+1}\lambda_{k+1}\mathbf{v}_{k+1}=\mathbf{0} }[/math]

이고 처음 식에 [math]\displaystyle{ \lambda_{k+1} }[/math]을 곱하면

[math]\displaystyle{ c_1\lambda_{k+1}\mathbf{v}_1+c_2\lambda_{k+1}\mathbf{v}_2+\cdots+c_{k+1}\lambda_{k+1}\mathbf{v}_{k+1}=\mathbf{0} }[/math]

이다. 그러면

[math]\displaystyle{ c_1(\lambda_1-\lambda_{k+1})\mathbf{v}_1+c_2(\lambda_2-\lambda_{k+1})\mathbf{v}_2+\cdots+c_k(\lambda_k-\lambda_{k+1})\mathbf{v}_k=0 }[/math]

이고 (귀납법) 가정에 의해 각 고유값이 다르고 [math]\displaystyle{ \{\mathbf{v}_1,\mathbf{v}_2,\cdots,\mathbf{v}_k\} }[/math]이 선형독립이므로

[math]\displaystyle{ c_1=c_2=\cdots=c_k=0 }[/math]

이다. 따라서

[math]\displaystyle{ c_{k+1}\mathbf{v}_{k+1}=\mathbf{0} }[/math]

이고 [math]\displaystyle{ \mathbf{v}_{k+1}\ne\mathbf{0} }[/math]이므로 [math]\displaystyle{ c_{k+1}=0 }[/math]이다. 따라서 [math]\displaystyle{ \{\mathbf{v}_1,\mathbf{v}_2,\cdots,\mathbf{v}_{k+1}\} }[/math]은 선형독립이다. 결국 수학적 귀납법에 의해 원하는 결론을 얻는다.

대각화[편집 | 원본 편집]

어떤 [math]\displaystyle{ n }[/math]차 정사각행렬 [math]\displaystyle{ A }[/math]가 대각화 가능할 필요충분조건은 [math]\displaystyle{ A }[/math]의 선형독립인 고유벡터가 [math]\displaystyle{ n }[/math]개 존재하는 것이다.

각주

  1. Sheldon Axler. 〈Chapter 5. Eigenvalues, Eigenvectors, and Invariant Subspaces〉, 《Linear Algebra Done Right》, 3rd edition, Springer. ISBN 0387982582. 2016년 5월 14일에 확인.
  2. Robert A. Beezer. 〈Chapter E. Eigenvalues〉, 《A First Course in Linear Algebra》, Version 3.50. 2016년 5월 14일에 확인.