고유값과 고유벡터
미지수가 개인 연립미분방정식
- \displaystyle \begin{cases} a_{11}x_1(t)+a_{12}x_2(t)+\cdots+a_{1n}x_n(t)=x_1'(t) \\a_{21}x_1(t)+a_{22}x_2(t)+\cdots+a_{2n}x_n(t)=x_2'(t) \\ \quad \vdots \\a_{n1}x_1(t)+a_{n2}x_2(t)+\cdots+a_{nn}x_n(t)=x_n'(t) \end{cases}
은 다음과 같은 행렬로 나타낼 수 있다.
- \displaystyle \begin{bmatrix} a_{11} & a_{12} & \cdots & a_{1n} \\ a_{21} & a_{22} & \cdots & a_{2n} \\ \vdots & \vdots & \ddots & \vdots \\ a_{n1} & a_{n2} & \cdots & a_{nn} \end{bmatrix} \begin{bmatrix} x_1(t) \\ x_2(t) \\ \vdots \\ x_n(t)\end{bmatrix}=\begin{bmatrix} x_1'(t) \\ x_2'(t) \\ \vdots \\ x_n'(t)\end{bmatrix}
미지수가 한 개일 때 x'(t)=ax(t)이므로 initial condition x(0)=c를 지정하면[1] x(t)=ce^{at}이다. A는 linear이므로 superposition principle에 따라서[2] Ax(t)_1=x'(t)이고 Ax(t)_2=x'(t)이면 A(cx(t)_1+c_2x(t)_2)=c_1Ax(t)_1+c_2Ax(t)_2=c_1x'(t)_1+c_2x'(t)_2=(c_1x(t)_1+c_2x(t)_2)'이다. 그러므로 a를 고정하여 해의 모든 성분이 x_i(t)=c_ie^{at}라 가정하고 연립미분방정식을 풀면 해가 존재하는 a들을 찾음으로써 그 합으로 0 또는 e^{\lambda t}들로 이루어진 모든 해를 찾을 수 있다. 이러한 a=\lambda들을 행렬 A의 eigenvalue(고유값, 고윳값)라 하고, 해 x_i(t)=c_i e^{\lambda t}에서 상수 c_i들로 이루어진 벡터 c를 A의 eigenvector(고유벡터)라 한다.
이 연립미분방정식의 x(t)는 initial condition t=0을 포함하는 open interval에서 A의 성분이 모두 연속 함수일 때 open interval에 속하는 t마다 독립인 벡터들을 만드는 서로 다른 해들의 linear combination으로 유일하다.[3] x_1(t)=c_ie^{\lambda t}를 단순히 c_i로 대체하면 x_i'(t)=\lambda c_i e^{\lambda t}이므로 연립미분방정식은 Ac=\lambda c로 쓸 수 있다. 따라서 \lambda는 (A-\lambda I)c=O를 만족시킨다. c는 A-\lambda I의 null space의 원소이므로 c\neq 0가 존재하려면 \det(A-\lambda I)=0이어야 한다. 행렬식을 전개하면 \lambda에 대한 다항식이며 이를 A의 characteristic polynomial(특성 다항식)이라고 한다.
- A의 성분이 실수이더라도 \lambda는 복소수일 수 있다. 이제 symmetric matrix와 orthogonal matrix는 conjugate transpose와 Hermitian adjoint를 정의하여 Hermitian matrix와 unitary matrix로 확장할 이유가 생긴다.
- 각 고유값에 대응하는 고유벡터는 eigenspace A-\lambda I의 null space를 구성해야 하므로 하나가 아니다. 즉 \lambda의 고유벡터에 상수를 곱해도 \lambda의 고유벡터이고, 한 고유값에 독립인 여러 개의 고유벡터가 대응할 수도 있다.
- 각 행렬에 대응하는 특성 다항식은 유일하다. 그러나 \lambda_0마다 (\lambda-\lambda_0)^n이 할당되며 이 중근의 개수를 \lambda의 algebraic multiplicity(대수적 중복도)라 하고, \lambda마다 A-\lambda I의 null space가 할당되며 이 차원을 \lambda의 geometric multiplicity(기하적 중복도)라 한다.[4] 대수적 중복도가 1이면 simple eigenvalue이고 두 중복도가 같으면 semisimple eigenvalue이다.
대각화
참고 자료
- ↑ https://math.stackexchange.com/questions/395161/whats-the-difference-between-an-initial-value-problem-and-a-boundary-value-prob
- ↑ https://en.wikipedia.org/wiki/Superposition_principle
- ↑ https://en.wikipedia.org/wiki/Linear_differential_equation
- ↑ https://en.wikipedia.org/wiki/Generalized_eigenvector
- https://people.math.wisc.edu/~aseeger/319/notes2.pdf
- Gilbert Strang. Linear Algebra and Its Applications.