Linear Algebra and ItsApplications 3rd Edition
——DavidC.Lay
Preface:
此书由浅入深,登堂入室,于是难免定理累赘,不够深入计算机图形层面。
(本书中指的向量的起点都是原点)
个人想法:
对xi的操作 (操作对象) (操作后的结果)
xn [] [] = []
行数限制最大维度,列向量确定实际维度
Chatpter 1 : Systems of LinearEquations
Definition:
多个 形如 A1x1+ A2x2 + … + Anxn =b 的组合
·解的情况:
Inconsistent-> 1、无解, n = 0
/ 2、唯一解,n = 1
Consistent \ 3、无数解,n >= 2
MatrixNotation:
对于Ax = b,系数矩阵coefficientmatrix [A]
增广矩阵augmented matrix [A b]
ElementaryRow Operations
1.Replacement,(倍加)把某一行换成它本身与另一行倍数的和
2.Interchange,(换行)把两行交换
3.Scaling,(数乘)某行的所有元素乘于非零常数
FundamentalProblem:Existence and Uniqueness
基本概念:
阶梯型,简化阶梯型,先导元素,主元位置,主元列,主元,自由变量,基本变量
Theorem1:
Each matrix is row equivalent to one and onlyone reduced echelon matrix.
行化简算法Row Reduction Algorithm:
Forward phase, backward phase
结果:显式的通解
解集的参数向量表示【我们约定用自由变量作参数】
Theorem2:
线性方程组相容 <=> 增广矩阵最右列不是主元列
若方程组相容,则 (i)自由变量数 = 0,唯一解
(ii)自由变量数 > 0,无穷多解
列向量:仅含一列的矩阵,简称向量。
(注:(a, b)和[]只是不同的表示方法)
若向量相等,则各分量相等
线性组合:
y是v1,…,vn以c1,…,cp 为权的线性组合
Span{v1,…,vn}:
表示v1,…,vn的所有线性组合,称为v1,…,vn生成的Rn的子集
Theorem 3:
Let A be an m x n matrix,then the following statements are logically equivalent.
a. For each b in, Ax = b has a solution
b. Each b in Rm is alinear combination of the columns ofA
c. The columns of A span Rm
d. A has a pivot position inevery row.
HomogeneousLinear Systems齐次线性方程组:
Ax= 0
trival solution零解(平凡解)
非齐次线性方程组:
解集 = 特解 + 对应的线性方程组的任意一个通解
Linearindependence:
has the only trival solution
反之,则是线性相关(linear dependence)
Theorem 4:
Any set {v1,…,vp} in Rn is linearly independent if p>n
Theorem5:
If a set contains the zero vector, then theset is linearly dependent.
T:Rm->Rn, 则Rn为余定义域(domain)
所有T(v)的集合为值域(range)
Matrix Transformation:
对基底进行变换,就对整个图形变换。
Linear transformation:
1. cT(u) = T(cu)标量乘法
2. T(u + v) = T(u) + T(v)加法
Superposition principle:(叠加原理)
T(c1v1 + … + cpvp)= c1T(v1) + … + cpT(vp)
RotationTransformation:
[] 旋转 α
单射:余定义域中的像只有一个原像与之对应,
即Ax = b 仅有一个解,即Ax = 0 只有平凡解di
满射:余定义域中每个元素都是像(余定义域=值域)
Chapter 3 : Determinant(square matrix)
Definition(略)
Cofactorexpansion 余因子展开式 Cij=(-1)i+jdetAij
Theorem1:
If A is a triangular matrix, then det A isthe product of the entries on the main diagonal of A.
三角形矩阵的行列式为主对角线元素之积
Theorem2: Row operation
Let A |-> B
1. Replacement, detB = detA
2. Interchange, detB = -detA
3.Scaling. detB = k·detA
Theorem3:
A is invertible, iff det A ≠ 0
Theorem4:
det AT = det A
Theorem5:
det(AB) = detA·detB
Theorem6:
设
则 xi = det Ai(b) / det A
Theorem7:
其中 Cij为余因子式
Theorem8:
若A为2阶方阵,则|detA|为A中的列确定的平行四边形的面积
若A为3阶方阵,则|detA|为A中的列确定的平行六面体的体积
Theorem9:
令 T: R2->R2是由2阶方阵确定的线性变换,则
T{area of T(S)} = |detA|·{area of S}
同理,三阶方阵确定的T: R3->R3
Chapter 4 : Vector Space
Definition :
A nonempty set V of vectors,on which are defined two operations—addtion and multiplication by scalars,satisfy closed.
非空向量集V满足加法和数乘封闭。
Subspace :
a. The zero vector of V in H.
b. H is closed under addition
c. H is closed undermultiplication by scalars.
若H是V的子空间, 则 ①H是向量空间 ②dim H <= dim V
Exception: {} zero subspace
Theorem: V中任意向量的组合,可张成V的子空间H
Column Space :
If A = [a1…an], then ColA = Span{ a1…an}
Null Space:
NulA = { :n & A=}
Integration Theorem:
1、 The null space of an m*n matrix A is a subspace of Rn
The column space of m*n matrix A is a subspace of Rm
2、 dim NulA = free variables
dim ColA = pivotcolumns
3、 (Rank Theorem)A is m*n matrix, then
rankA + dim NulA =n (dim ColA = rank A, kernel == nullspace)
Imply ->
dim NulA```free variable```lineardependence
dim ColA ```pivot column```linearindependence
Comment:Elementary row operationsdon’t affect the linear dependence relations among the columns of the matrix.
Basis : ~Column Space
A spanning set as small aspossible
A linearly independent set as large aspossible.
尽可能小的生成集,尽可能大的线性无关集。
Coordinate Map:
Every in V can be represented uniquely with abasis.One-to-one linear transformation,in other words, isomorphism(同构)
行等价即行空间相同,阶梯型矩阵的非零行形成行空间的一组基。
Change of Basis:
P = [ [b1]…[b1] ]
C<-B c c
用新坐标系的基来表示原基,其权重为新坐标
Algorithm(同步求解):
[c1…cn | b1…bn] ~ [I | PC<-B]
Application -> Markov Chain
Chapter 5 : Eigenvalues andEigenvectors (square matrix)
尽管 -> A可能往各个方向移动,但通常会有些特殊向量,A对这些向量的作用是简单的。
Definition :
Aneigenvector of an n*n matrix A is a nonzero vector such A=λ forsome scalar λ, which is called an eigenvalue,
respectively , corresponding to
A=λ存在非平凡解,特征向量所张成的空间为特征空间
Theorem 1 :
Allthe eigenvalues of a triangular matrix are the entries on its main diagonal.
三角矩阵的特征值在主对角线上
Theorem 2 :
If v1…vn areeigenvectorsthat correspond to distinct eigenvalues λ1…λn of an n*n matrix A, then the set { v1…vn} is linearindependence.
不同特征值对应的特征空间不同
IMT(Continued):
Ais invertible iff its eigenvalue don’t include zero
The Characteristic Equation :
det(A-λI) = 0
<=>λ is an eigenvalue of an n*n matrix A
Algebraic Multiplicity:
Itsmultiplicity as a root of the characteristic equation.
Similarity :
If A = PBP-1, then A and B issimilar.
Changing A into B is called a similaritytransformation.
Theorem 3 :
If n*n matrices A andB are similar, then they have the same eigenvalues(with the same multiplicities)
A和B相似,则特征值相同(重数相同)
Theorem 4: The diagonalizationTheorem
If n*n matrix A isdiagonalizable, iff A has n linearly independent eigenvectors.(All is eigenvector basis) In this case, thediagonal entries of D are eigenvalues of A that correspond respectively, to theeigenvectors in P.P must be invertible.
对角化: A 存在n个线性无关的特征向量,构成特征向量基
目标A = PDP-1,
① D是对角矩阵,且主对角线元素是A的特征值
② P中各列是D中各列特征值对应的特征向量,并检验是否线性无关
③ 验证AP ?= DP
Theorem 5: (Combine 2 and 4)
An n*n matrix with n distinct eigenvalues isdiagonalizable.
含有n个不同特征值的n阶方阵,必可对角化
Theorem 6: DiagonalMatrix Representation
Suppose A = PDP-1, where D is adiagonal n*n matrix. If βisthe basis for Rn formed from the columns of P,then D is the β-matrix.
Chapter 6 : Orthogonality and Least Squares
Definition:
u·v = uTv
(定义)If u·v = 0,then u,v areorthogonal
Theorem 1:
(Row A)⊥ =Nul A
正交集:互相正交的向量集
Theorem 2:
Let {u1, …, up} be an orthogonal basis for asubspace W of Rn.For each y in W,
y= c1u1 + … + cpup , where ci
OrthogonalProjection: +()
y在 W 中的投影 projL y
projL y = c1u1 + … + cpup, where ci
Theorem 3:
Orthonormalset U iff(等价) UT·U = I
Property: |Ux| = |x|
当方阵A是标准正交集,则AT = A-1
The Best Approximation Theorem:
Theorem 5: QR Factorization(因式分解)