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=QVQT(orthonormal eigenvector column)
why realeigenvalue?
A=QVQT=la1 q1q1T+la2 q2q2T+… ------> a comb of perp projection matrices
A=AT
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
xTAx>0
Lecture 27
complex matrix
fast Fourier transform ----FFT(in book)
z barTz=zHz(hermite matrix) conjugate
perpendicular
Lecture 28
A=[a b; c d]
calculus: MATRIX OF 2nd DERIVS IS POS DEF
【fxx fxy; fyx fyy】
Lecture 29
positive definite means
xTAx>0(except for x=0)
ATA---->square,symmetric, pos defc
check:xTATAX=||Ax||2>=n(becau rank(A)=0,so Ax~= 0)—>in least square,(ATA)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=QVQT
Rn in row space--------->v1,v2 in rowspace R2(orthonormal)
Rm in column space---->u1,u2 in col space R2(orthonormal)
sigma1>0, sigma2>0
AV=U Msigma
ATA=VMsigmaUTUMsigmaVT=Vdiad(sigma12,sigma22,…) VT . sigmai,vi are eigenvalue,eigenvector of ATA
U is same with AAT
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(AT)
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 45o
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
AMinputcoords=Moutputcoords(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—>Cbar(many zeros)–>xbar=MsigmaCibarvi
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)