MIT学习笔记-Linear Algebra(三)

Teacher：Gilbert Strang

The text of the course: Introduction to Linear Algebra

Website:web.mit.edu/18.06,which got a lot of exercise from the past MATLAB codes, the syllabus for the course

Lecture 26

Symmetric matrix

properties
1、The eigenvalues are REAL
2、The eigenvalues can be chosen perp.
usual:A=SVS^-1
symmetric case:A=QVQ^-1=QVQ^T(orthonormal eigenvector column)
why realeigenvalue?


A=QVQ^T=la₁ q₁q₁^T+la₂ q₂q₂^T+… ------> a comb of perp projection matrices

---

A=A^T
sign of pivots same as sign of lambda’s
#positive pivots=#positive lambda’s
positive definite symmetric matrix
all eigenvalues are positive
all pivots are positive
all sub-determinants are positive
x^TAx>0

Lecture 27

complex　matrix

fast Fourier transform ----FFT(in book)

---

z ^barTz=z^Hz(hermite matrix) conjugate
perpendicular

Lecture 28

A=[a b; c d]
calculus: MATRIX OF 2nd DERIVS IS POS DEF
【f_xx f_xy; f_yx f_yy】

Lecture 29

positive definite means
x^TAx>0(except for x=0)
A^TA---->square,symmetric, pos defc
check:x^TA^TAX=||Ax||²>=n(becau rank(A)=0,so Ax~= 0)—>in least square,(A^TA)invertible

---

similar matrices

A and B are similar , means for some M
B=M^-1AM
similar matrices have some la’s (same # evectors)

---

bad case la₁=la₂, use Jordan block , same or not?
J=diag(la₁,la₁,…) one Jordan block,one ecevtors

Lecture 30

SVD (singular value decomposition)
review:symm pos def
A=QVQ^T
Rⁿ in row space--------->v₁,v₂ in rowspace R₂(orthonormal)
R^m in column space---->u₁,u₂ in col space R²(orthonormal)
sigma₁>0, sigma₂>0
AV=U M_sigma
A^TA=VM_sigmaU^TUM_sigmaV^T=Vdiad(sigma₁²,sigma₂²,…) V^T . sigma_i,v_i are eigenvalue,eigenvector of A^TA
U is same with AA^T

---

v1,v2,…vr are orthonormal basis for rowspace
u1,u2,…ur are orthonormal basis for column space
vr+1,…vn are orthonormal basis for N(A)
ur+1,…um are orthonormal basis for N(A^T)

Lecture 31

LINEAR TRANSFORM ACTIONS

properties:
1、T（v+w）=T(v)+T(w)
2、T(cv)=cT(v); T(0)=0
Example 1 .Projection
Non Example 2: Shift whole plane
Example 2: rotation by 45^o

---

put A=[1 0; 0 -1] in a picture ,what’s the output（沿y轴翻转）

---

Start： T: R³–>R²
Example:T(v)(output in R²)=A(2 by 3)v(input in R³)

---

What’s the information needed to know all T(v) for all inputs? v=c1v1+c2v2+…; T(v)=c1T(v1)+c2T(v2)+…
answer:T(v1),T(v2),… foe any input basis
coordinate come form a basis : v=c1v1+c2v2+…

---

Rule to find A.Given basis v1–vn, w1–wn
AM_inputcoords=M_outputcoords(eigrnvalues basis can simplify calculation, becu these leads to diagonal matrix V).

Lecture 32

Change of Basis

standard basis (columns of I)
better basis—>[ 1 1 1 …];[1 -1 1 -1…]…
Fourier Basis–>[1 1 1 1 1 1 1 1];[1 w w²…]…

---

signal P(input)–>change basis–>coeffs C—>lossy composition—>C^bar(many zeros)–>x^bar=M_sigmaC_i^barvⁱ
wavelets: [1 1 1 1 1 1 1 1];[1 1 1 1 -1 -1 -1 -1];[1 1 -1 -1 0 0 0 0];[0 0 0 0 1 1 -1 -1];[1 -1 0 0 0 0 0 0]…

---

What is good basis?
P=Wc=c1w1+c2w2+…
c=W^-1P
1、FAST FFT;FWT–>orthongonal
2、Few is enough

---

T
with respect to v1,v2,…, it has matrix A
with respect to u1,u2,…, it has matrix B
SIMILAR B=M^-1AM
Eigenvectors basis
T(v1)=lambda vi, What is A?
A=diag(la1, la2, la3,…)

Lecture 35

Pseudo Inverse
最小二乘法的几何意义（空间最近点）
Markov矩阵，趋近于Infi时，人口总数不变（lambda=1时，la*x不改变元素总和）