4️⃣

Lecutre 04

Intro

Linear Combination

𝑐1 𝐯𝟏 + 𝑐2 𝐯𝟐 + ⋯ + 𝑐𝑝 𝐯𝒑
𝑓𝑜𝑟 𝑠𝑐𝑎𝑙𝑎𝑟𝑠 𝑐𝑖 𝑖 = 1,2, … p
이걸 matrix form 으로 적으면

Subspace

어떤 집합 내의 원소들의 선형결합도 집합안에 있는 경우. → 1v +(-1)v = 0 도 포함해야하니까 영벡터 포함도 필수임.

n-dim 에서 subspace의 종류

2차원에서
zero space
line through the origin
ℝ^2
3차원에서
zero space
line through the origin
plane through the origin
ℝ^3

Linearly (in)dependence

Linearly dependent (rank < #(var) )하면 Ax=0이 무수히 많은 해를 가짐.
Linearly independent (rank = #(var))하면 Ax=0이 x=0(trivial solution)으로 유일한 해를 가짐.
why? 선형종속의 정의 𝑐1𝐯1 + 𝑐2 𝐯2 + ⋯ + 𝑐𝑝 𝐯p = 𝟎 를 생각하면 당연함.

Row Operation은 column space를 보존하지 않는다.

기저 찾기, 선형 독립성 판별, Column space 표현하기 할때 유의사항
REF의 pivot column index를 고르되, 실제로는 원래 행렬의 해당 column을 선택해야한다.
Pivot은 위치만, Column은 원본에서

span되는 벡터인지 판별과 동치인 명제

어떤 벡터 u 가 u_1, u_2에 의해 span 되는지 판별하는 문제와 동치인 문제들.

\exists x \ \text{s.t.} \ A x = \mathbf{u}

\text{where } A = [\mathbf{u}_1 \ \mathbf{u}_2]
Ax=u 의 해 x가 존재한다.

\Longleftrightarrow \ \mathbf{u} \in \mathcal{C}(A)
A의 column인 u_1 u_2의 선형결합으로 생기는 공간인 C(A)에 u가 포함된다.

\Longleftrightarrow \ \operatorname{rank}(A) \stackrel{?}{=} \operatorname{rank}([A \mid \mathbf{u}])
A 옆에 u 컬럼을 이어 붙여 추가한 증강행렬의 rank = dim(ColSpace) 가 동일하다.
새로운 컬럼이 다른 컬럼들의 선형결합으로 표현된다.

Null Space / Column space

kernel, image

파란선은 null space = kernel by A → 영벡터로 간다.
검정선은 임의의 x=[ a b]^T를 잡았을 때, [a-2b a-2b]^T 즉 y=x 직선으로 가는걸 나타냄
문제풀이
어떤 직선에 대하여 linear transformation 한 image를 구하는 방법.
직선을 벡터표기(direction vector t + passing vector)로 바꾼다.
matrix multiplication을 진행한다.
다시 방정식 표기로 바꾼다.

determinant의 기하적 의미

det의 크기가 행렬의 column들로 구성된 도형의 넓이, 부피의 변화의 scale factor임

Orthogonal matrix

definition

QQ^T=I 인 n x n matrix

properties

1.
orthonormal columns
a_i \cdot a_j가 Kronecker delta 성질을 가지고 있어야 곱했을 때 Identity Matrix가 된다. 다시말하면 자신끼리 dot product를 했을 때는 1이 나와야하고 = norm 이 1이다.
다른 column 과 dot product을 했을 때는 0이 나와야한다= 두 컬럼은 직교한다.
1.
det(A) = +-1
det(AA^T)= det(A)det(A^T) = det(A)^2 = det(I) = 1
determinant의 성질에 의해 다음과 같이 된다.
1.
dot product preserved 된다. ↔ norm, angle, distance가 preserved 된다.
u \cdot v = u^T v 을 사용
dot product preserve 되는 점











\begin{aligned}
(Q\mathbf{x}) \cdot (Q\mathbf{y})
&= (Q\mathbf{x})^T (Q\mathbf{y}) \\
&= \mathbf{x}^T Q^T Q \mathbf{y} \\
&= \mathbf{x}^T I \mathbf{y} \\
&= \mathbf{x}^T \mathbf{y} \\
&= \mathbf{x} \cdot \mathbf{y}
\end{aligned}

norm이 보존된다.








\|Qx\| = \sqrt{(Qx) \cdot (Qx)} = \sqrt{x \cdot x} = \|x\|
각도가 보존된다.








\cos\theta = \frac{x \cdot y}{\|x\|\|y\|}

\cos\theta’ = \frac{(Qx) \cdot (Qy)}{\|Qx\|\|Qy\|}
= \frac{x \cdot y}{\|x\|\|y\|} = \cos\theta

Linear Transformation

Reflection

Rotation

constant matrix
det(R_\theta) = 1
orthogonal matrix(컬럼 간 직교, norm 이 1)
scaling factor 가 곱해진 버젼도 있음. 각도 보존, 스케일링.

Reflection about y = tan \theta x

핵심 아이디어 : tan x 선을 x축으로 회전시키고, x축을 기준으로 reflection 한뒤에, 2theta만큼 원복시킨다.








A =
\begin{bmatrix}
\cosθ & -\sinθ \\
\sinθ & \cosθ
\end{bmatrix}
\begin{bmatrix}
1 & 0 \\
0 & -1
\end{bmatrix}
\begin{bmatrix}
\cosθ & \sinθ \\
-\sinθ & \cosθ
\end{bmatrix}








A =
\begin{bmatrix}
\cos^2θ - \sin^2θ & 2\sinθ\cosθ \\
2\sinθ\cosθ & \sin^2θ - \cos^2θ
\end{bmatrix}
det(R_\theta) = -1
orthogonal matrix(컬럼 간 직교, norm 이 1)

Quiz2

PartA

Q1.

A \in \mathbb{R}^{m \times n} : \mathbb{R}^{n} \to \mathbb{R}^{m}이다.
dim(C(A)) = rank(A)
dim(codomain) = m
dim(N(A)) = dim(domain) - rank(A)

Q2. rank(A) = rank(A|b) 인 경우에 Ax=b의 solution이 존재함을 증명하기

rank(A) = rank(A|b) dim(C(A)) = dim(C(A|b)) 와 동치임.
따라서 증강행렬의 추가된 컬럼 b는 이미 Column space에 포함된 거임. b \in C(A)
\Leftrightarrow \exists x_i \in \mathbb{R} \text{ s.t. } b = \sum_{i=1}^{n} x_i a_i 그 얘기는b를 linear combination으로 쓸 수 있다는거고
\Leftrightarrow \exists x \in \mathbb{R}^n \text{ s.t. } Ax = b 행렬곱 형태로 표현하면 Ax=b의 solution이 존재한다는 거임.

PartB

Q2.

T_{A} : \mathbb{R} ^2 \to \mathbb{R}^3, such that \begin{cases}
T\!\left(\begin{bmatrix}1\\[2pt]1\end{bmatrix}\right)=\begin{bmatrix}2\\[2pt]1\\[2pt]-1\end{bmatrix},\\[8pt]
T\!\left(\begin{bmatrix}2\\[2pt]3\end{bmatrix}\right)=\begin{bmatrix}4\\[2pt]0\\[2pt]2\end{bmatrix},
\end{cases} 
evaluate T\!\left(\begin{bmatrix}3\\[2pt]-1\end{bmatrix}\right). 

Sol1)

transtition matrix A는 \mathbb{R}^{3 \times 2} 의 구조임. codomain의 dimension 이 앞, domain의 dimension이 뒤쪽.
Find c_1, c_2 \in \mathbb{R} s.t. \begin{bmatrix}
3\\-1
\end{bmatrix}
=c_1
\begin{bmatrix}
1\\1
\end{bmatrix}
+c_2
\begin{bmatrix}
2\\3
\end{bmatrix}

.
linear combination으로 [3 -1]^T 를 표현하고 이걸 만족하는 c_1, c_2를 찾는다.
Linear Transformation 의 성질(linearity)에 의해서 T(\begin{bmatrix}
3\\-1
\end{bmatrix})
=c_1
T(\begin{bmatrix}
1\\1
\end{bmatrix})
+c_2
T(\begin{bmatrix}
2\\3
\end{bmatrix})
의 식이 유지됨.
찾은 c_1과 c_2를 적용하여 구하고자 하는 값을 구한다.

Q2.2 A를 구하시오

Transition matrix A를 찾아라.와 같은 새끼 문제가 있다면 이 전이행렬을 먼저 구하고 그 행렬의 곱으로 Linear Transformation을 적용하는 것도 좋음.
A\begin{bmatrix}
1 & 2 \\
1 &3
\end{bmatrix}
=
\begin{bmatrix}
2 &4\\
1 &0 \\
-1 &2
\end{bmatrix}

두 개의 벡터의 곱 결과를 concat 하여 행렬곱을 구성. Square matrix이기에 역행렬로 양변에 곱해서 A를 구할 수 있음.

Q2.3 range(T) 를 구하시오

=Column Space of A = C(A)
rank(A) = 2 이기 때문에 Column space가 원점 지나는 평면이다.
이제는 구체적인 방정식을 구해야한다. normal vector를 구하는 방법. cross product하면 normal vector 나온다.
column space : 2x-8y-4z =0 정리하면 x-4y-2z=0
image 공간이 어디에 그려지는지 표현할 수 있어야함.
이 경우에는 원점을 지나는 평면 x-4y-z=0 위에 있는 타원임. 장축의 길이는 max( Ax ), 단축의 길이는 min( Ax )

Part3

Q1

T(x,y,z) = (2x-y,x+z,y-z)

show that it is linear

행렬에다가 [x,y,z]^T를 곱한거로 표현하면 됨. T(x) = Ax인 행렬 A만 찾으면 선형성이 증명됨. (image가 행렬곱 형태로 표현됨을 보이면 됨.)

is w = [3\ 3\ 0]^T in the range of T?

w\in C(A)인지를 물어보는 문제.

Intro 후반부

Orthogonal matrix

A^TA = I_n인 A
(A^TA)_{ij} = a_i \cdot a_j (where a_i = i th col of A)

property

1.
a_i \cdot a_j = \delta_{ij}= \begin{cases}
0 & \text{if } i \neq j \\
1 & \text{if } i =j
\end{cases}
즉, orthonormal column을 가진다.
2.
det(A^TA) = det(A^T)det(A) = det(A)^2= det(I) = 1 이므로 det(A) = 1 or -1이다.
3.
x_1, x_2 \in \mathbb{R}^n의 이미지 벡터의 내적을 정리하면Ax_1 \cdot Ax_2 = (Ax_1)^T(Ax_2) = x_1A^TAx_2 = x_1 \cdot x_2 원래 값의 내적과 동일하다. dot-product is preserved! ↔norm& angle& distance가 모두 preserved 됨.
4. A^{-1} = A^T

ex1.

show that A가 orthogonal임을 보여라.
각 컬럼의 크기가 1이고, 두 컬럼을 내적하면 0 임. 그래서 orthogonal임을 바로 알 수 있음.
A = \begin{bmatrix}
cos\theta& -sin\theta \\
sin\theta& cos\theta \\
\end{bmatrix}
det(A) = c^2 + s^2 = 1 counter-clockwise,\theta-rotation matrix(angle, norm, distance 유지)
find inverse matrix of A.

ex2.

orthogonal임으로 보여라.
definition으로 푸는 법.A^TA = I_n임을 보인다. → 엄청 오래걸림.
property 1 을 사용하여 column간의 dot product가 0인지, 각 column의 크기가 1인지 보이면 편함.
inverse function 구하는 법. transpose 구하면 됨. by property 4에 의해.
det(A) 구하는 법. 1 아니면 -1임.
A=[v_1 \ v_2 \ v_3] \to det(A) = v_1 \cdot (v_2 \times v_3)
cross product해서 방향이 같으면 양수, 다르면 음수.

Online class

Orthogonal projection about line y=(tan \theta )x

v는 norm이 1인 unit vector.
projection의 공식 \text{proj}_{\mathbf{v}}(\mathbf{x}) = \frac{\mathbf{x} \cdot \mathbf{v}}{\mathbf{v} \cdot \mathbf{v}} \, \mathbf{v} 인데, v가 unit vector여서 T(x) = (x \cdot v)v임. 여기서 앞에 dot product로 되어있는 부분이 스칼라고, v가 방향을 나타냄.
projection 공식 유도
벡터 \mathbf{x}\mathbf{v} 방향으로 투영한 벡터는 \mathbf{v}와 같은 방향을 가지되 크기만 다른 형태 c\mathbf{v}로 표현되며, 수직 조건 (\mathbf{x} - c\mathbf{v}) \cdot \mathbf{v} = 0으로부터 c = \frac{\mathbf{x}\cdot \mathbf{v}}{\mathbf{v}\cdot \mathbf{v}}가 된다. 따라서 투영 벡터는 \text{proj}{\mathbf{v}}(\mathbf{x}) = \frac{\mathbf{x}\cdot \mathbf{v}}{\mathbf{v}\cdot \mathbf{v}}\mathbf{v} 로 주어진다. 이는 \mathbf{x} \mathbf{v} 방향으로 얼마나 겹치는지를 나타내는 내적 \mathbf{x}\cdot\mathbf{v}를 \mathbf{v}의 크기 제곱으로 정규화한 뒤, 그만큼 \mathbf{v} 방향으로 늘린 벡터이다. 특히 \mathbf{v}가 단위벡터라면 \mathbf{v}\cdot\mathbf{v}=1이 되어 \text{proj}{\mathbf{v}}(\mathbf{x}) = (\mathbf{x}\cdot\mathbf{v})\mathbf{v}로 단순화된다.
근데 이 dot product를 행렬형태로 표현하면 T(x) =(vv^T)x형태가 되고, P=vv^T를 Projection matrix라고 한다.
→ 결론 : 어떤 normal 벡터 v에 projection 하고 싶으면 Projection Matrix P=vv^T를 원하는 벡터 x에 곱하자.
다른 방법으로도 이 P를 구할 수도 있는데 x를 tan 세타에 대해 reflection한 벡터와 평균을 취해 projection을 표현하는거다. cos/sin 2배각 공식을 써보면 동일함을 알 수 있다.
공간에서의 의미가 헷갈려 정리를 해보자면, projection vector는 x벡터에서 y=tan x 직선에 수선을 내린 직각 삼각형의 밑변 부분에 해당하는 벡터이다. 그리고 P는 x에다가 곱해주기만 한다면 바로 그 projection vector를 구해주는 선형 변환을 해주는 전이행렬이다.
det(P) =0 : 한 점을 압축되니까 det이 0임
rank(P) = 1 : 출력이 projection vector 직선 1차원 공간에 존재함.
C(P) : y = (tan \theta)x : projection vector는 당연히 저 직선 위에 수선을 내리기에 저 위에 존재할 수 밖에 없다.
N(P) : y = (-cot \theta)x : 원점을 지나면서 tan x 에 수직인 이 위의 벡터 x들은 y = (tan \theta)x에 수선을 내리면 원점이다. 그래서 Px=0이 되는 x들이 된다. 이 말은 Null Space 라는 소리.

나혼자탐구 : det 값과 기하적인 해석 간의 관계

A=\begin{bmatrix} a & b\\ c & d\end{bmatrix}
의 열벡터를 \mathbf u=(a,c),\ \mathbf v=(b,d)라 하면, 이 두 벡터가 만드는 평행사변형의 넓이
\text{area}(\mathbf u,\mathbf v)=|\,a d - b c\,| = |\det A|.
왜?
벡터 내적 공식







\mathbf{u}\cdot\mathbf{v} = |\mathbf{u}||\mathbf{v}|\cos\theta
→ 식변형







(\sin\theta)^2 = 1 - \cos^2\theta
= 1 - \frac{(\mathbf{u}\cdot\mathbf{v})^2}{(|\mathbf{u}||\mathbf{v}|)^2}
u,v벡터 사이의 면적의 제곱은
A^2 = (|\mathbf{u}||\mathbf{v}|\sin\theta)^2
= (|\mathbf{u}||\mathbf{v}|)^2 (\sin\theta)^2
위에서 (\sin\theta)^2 식을 대입하면:
A^2 = (|\mathbf{u}||\mathbf{v}|)^2
\Big(1 - \frac{(\mathbf{u}\cdot\mathbf{v})^2}{(|\mathbf{u}||\mathbf{v}|)^2}\Big)
분배하면
A^2 = (|\mathbf{u}||\mathbf{v}|)^2 - (\mathbf{u}\cdot\mathbf{v})^2
따라서 이걸 계산하면








|\mathbf{u}|^2 = a^2 + c^2,\quad |\mathbf{v}|^2 = b^2 + d^2,\quad \mathbf{u}\cdot\mathbf{v} = ab + cd








A^2 = (a^2 + c^2)(b^2 + d^2) - (ab + cd)^2 = (ad-bc)^2
단위 정사각형의 넓이는 1인데, image 정사각형의 넓이가 | det A | 이므로, 이게 면적의 배율이다.
n차원으로 확장되어도 det A는 A의 column vector들이 만드는 평행육면체의 부피임.
det A =0 ⇒ A 의 열벡터들은 선형종속이다. ⇒ 차원이 낮아졌다. (부분공간으로 collapse 되었다)⇒ 부피가 0이다.
부호는 basis의 orientation의 변화를 나타냄. det > 0 : orientation preserved , det < 0 : orientation reversed

Horizontal shear transformation with factor k

x축은 고정. k가 양수면 오른쪽으로 이동. 음수면 왼쪽으로 이동.
det(A) = 1이다. 평행사변형의 높이와 밑변이 동일하기에 기하적으로 유추가능.

Contractions and Expansions

방향별 압축 / 확장

Composite Transformation

뒤쪽 변환일 수록 앞쪽에 내적이 곱해짐.
예시. A의 엔트리를 바꿔가며 N자의 모양 변화관찰.
이건 간단한 버젼.

composite examples

y=(tan \theta )x 에 대한 Reflection을 composite transformation으로 표현하기.
그래서 어떻게 행렬을 구성해야하는가? 회전 → 반사 → 역회전 의 순서로 구성된 합성 변환이에요.
R_{-\theta} : 직선 y = (\tan \theta)xx축에 겹치도록 회전
T_x : x축에 대해 반사(reflection about x-axis)
R_{\theta} : 다시 원래 각도로 되돌리는 회전
따라서 A = R_{\theta} \, T_x \, R_{-\theta} 정리하면







A =
\begin{bmatrix}
\cos 2\theta & \sin 2\theta \\
\sin 2\theta & -\cos 2\theta
\end{bmatrix}
임을 알 수 있음.
반시계 2D 회전에서 회전행렬의 합성은 각도의 합과 같아 R_{θ1}R_{θ2}=R_{θ1+θ2}가 된다.

Orthogonal Projection in \mathbb{R}^3

v = unit direction vector of line
Ax = proj_vx = (x \cdot v)v = v(x \cdot v) = v (v^Tx) = (vv^T)x

Reflection과 Projeciton 사이의 관계. 평면의 normal vector 를 사용하여.

평면의 unit normal vector n을 사용하여, x의 n에 대한 proj_nx를 구하고, 이 벡터로 평면에 대한 Orthogonal Projection Matrix를 구할 수 있다. 그리고 Reflection은 proj_nx 만큼이 한번 더 빠진거니까 그거도 표현가능하다.
예시1 : 평면 x+2y-2z = 0 의 Orthogonal projection matrix를 구하시오.
평면방정식의 계수 = Normal vector (why? [x,y,z] \cdot [a,b,c] = 0 이 수직임을 나타냄.)
unit normal vector =







\mathbf{u} = \frac{1}{3}
\begin{bmatrix}
1 \\ 2 \\ -2
\end{bmatrix}
Orthogonal projection matrix P = I - uu^T =
\frac{1}{9}
\begin{bmatrix}
8 & -2 & 2 \\
-2 & 5 & 4 \\
2 & 4 & 5
\end{bmatrix}


uu^T쉽게 하기







uu^T =
\begin{bmatrix}
u_1^2 & u_1u_2 & u_1u_3 \\
u_2u_1 & u_2^2 & u_2u_3 \\
u_3u_1 & u_3u_2 & u_3^2
\end{bmatrix}
i행 j열이면, u_i x u_j 쉽다 쉬워.
예시2 : 동일한 Orthogonal projection matrix의 Rank, Nullspace, det
orthogonal projection을 시키면 전부 평면으로 압축됨. 평면이 image 공간임.
rank (P) = dim(C(p)) = 2
Normal vector에서 평면에 Projection 시키면 전부 zero로 감.
N(P) = span \{[1,2,-2]^T\}
orthogonal 평면 위에서 다시 Projection 해도 그 Point가 유지됨. 3차원 공간에 있는 점들이 평면으로 떨어져서 Volume이 없음.
P^n =P \to det(P)=0

conter clockwise \theta-rotation x축 기준

축에 있는건 그대로 감.
standard basis vector 의 image를 column으로 가져다가 구성하면 Rotation matrix 가 된다.
1.
Orthonormal column들임.
2.
Orthogonal Matrix가 됨. ⇒ Inverse = Transpose R_{\theta_x}^{-1} =R_{\theta_x}^T
vector norm 이 유지되기에 거리, 면적 범위 등등이 유지됨.
3.
오른손 법칙을 적용한거기에 det(R_{\theta_x}) = 1








R_x(\theta)

\begin{bmatrix}
1 & 0 & 0 \\
0 & \cos\theta & -\sin\theta \\
0 & \sin\theta & \cos\theta
\end{bmatrix}








R_y(\theta)

\begin{bmatrix}
\cos\theta & 0 & \sin\theta \\
0 & 1 & 0 \\
-\sin\theta & 0 & \cos\theta
\end{bmatrix}
zx 평면에서 회전이 일어나기에 orientation이 반대임.
(그렇다고 det = -1 은 아님. preserve 되니까. -1은 reflection에서만)








R_z(\theta)

\begin{bmatrix}
\cos\theta & -\sin\theta & 0 \\
\sin\theta & \cos\theta & 0 \\
0 & 0 & 1
\end{bmatrix}
ex. 3x3에서 y축 기준으로 30도 회전
별말 없으면 positive 방향 회전임. postitive 방향이라는건 conter clock wise 라는 거.

임의의 axis of rotation

스피리컬 코디네이트는 x축에서 돌아가야함.
이건 z축으로 \alpha만큼 떨어져있고, y축으로 \beta만큼 떨어져 있다고 표시함.
v벡터를 z축으로 만들거임.
1.
z축을 중심으로 \beta만큼 회전시키고,
2.
x축을 중심으로 \alpha만큼 회전시키면 z 축이됨.
3.
여기서 \theta만큼 회전을 시킨 다음에, 다시 Back을 해주면 됨.

\begin{aligned}
A &= R_{-\beta_z} R_{-\alpha_x} R_{\theta_z} R_{\alpha_x} R_{\beta_z} \\[6pt]
&= R_{\beta_z}^T R_{\alpha_x}^T R_{\theta_z} R_{\alpha_x} R_{\beta_z} \\[6pt]
&= (R_{\alpha_x} R_{\beta_z})^T R_{\theta_z} (R_{\alpha_x} R_{\beta_z})
\end{aligned}

Rodrigues’ rotation formula

임의의 축을 기준으로 공간 Rotation 하는 두번째 방법. Standard basis vector 말고, Orthogonal basis vector를 다시구하기.
회전축을 normal vector로 가지는 평면이 있을 거임.
파란 벡터 : x에서 (x를 v에 projection 시킨 )분홍색 벡터를 빼줌. 평면 위에 있음. v벡터와 수직.
초록 벡터 : 파란벡터를 v축을 기준으로 90도 회전. (평면 위에서 회전이 일어남.)
why? 외적의 기본성질(방향 : u \times v는 u, v 모두에 수직이다. 크기 :








\|\mathbf v\times\mathbf u\| = \|\mathbf v\|\,\|\mathbf u\|\,\sin\theta
그런데 파랑 v가 서로 수직이라 sin\theta가 1이고, v는 unit vector여서 파랑 의 크기와 초록의 크기는 동일함.)
그런 다음에 초록 파랑을 e1, e2로 생각해서 Orthonormal vector를 구성하고, 2D 회전 공식을 쓰면
cos\theta(I-vv^T)x + sin \theta v \times(I-vv^T)x이 식이 나옴.
최종적으로 거기에 높이 성분인 분홍색을 더해주면 끝.
*v x (I-vv^T)계산하는 방법
컬럼 바이 컬럼으로 계산 가능. 그리고 각 컬럼은 e1-av 이런식인데 같은 방향의 cross product는 0되서 av 쪽은 사라짐. 결국 e1, e2, e3 와 v의 Cross product 가 됨.
*vector cross product 까먹음…
determinant 처럼 계산하기.

\mathbf{a}\times\mathbf{b}
=
\begin{vmatrix}
\mathbf{i} & \mathbf{j} & \mathbf{k} \\
a_1 & a_2 & a_3 \\
b_1 & b_2 & b_3
\end{vmatrix}
=
\mathbf{i}(a_2b_3 - a_3b_2)
-\mathbf{j}(a_1b_3 - a_3b_1)
+\mathbf{k}(a_1b_2 - a_2b_1)

Translation

비선형 변환임. 그래서 2x2 행렬로 표시가 안됨.
2D Translation을 3D homogeneous coordinate로 표현:

\begin{bmatrix}
x' \\
y' \\
1
\end{bmatrix}
=
\begin{bmatrix}
1 & 0 & h \\
0 & 1 & k \\
0 & 0 & 1
\end{bmatrix}
\begin{bmatrix}
x \\
y \\
1
\end{bmatrix}
여기서:
(x, y): 원래 좌표
(x', y') = (x+h, y+k): 변환된 좌표
h, k: x, y 방향으로의 이동량
세 번째 성분은 항상 1로 유지됨 (homogeneous coordinate)
이렇게 homogeneous coordinate를 사용하면 비선형 변환인 translation을 선형 변환 형태의 행렬곱으로 표현할 수 있음.
n차원 공간에서의 변환을 (n+1)차원 homogeneous coordinates로 표현하면:

\begin{bmatrix}
\mathbf{x}' \\
1
\end{bmatrix}
=
\begin{bmatrix}
A & \mathbf{t} \\
\mathbf{0}^T & 1
\end{bmatrix}
\begin{bmatrix}
\mathbf{x} \\
1
\end{bmatrix}
여기서:
A \in \mathbb{R}^{n \times n}: 선형 변환 행렬 (회전, 스케일링, 반사 등)
\mathbf{t} \in \mathbb{R}^{n}: translation 벡터
\mathbf{0}^T: n차원 zero row vector
\mathbf{x}, \mathbf{x}' \in \mathbb{R}^{n}: 원래 좌표와 변환된 좌표

\begin{bmatrix}
\mathbf{x}+p \\
1
\end{bmatrix}
=
\begin{bmatrix}
I_2 & p \\
\mathbf{0} & 1
\end{bmatrix}
\begin{bmatrix}
\mathbf{x} \\
1
\end{bmatrix}
합성할때