Line of Best Fit
Over-determined systems,
e.g., coin flipping:
30 16
60 34
90 51
m
m
m
Number of flips 30 60 90
Number of heads 16 34 51
( m = fraction of heads )
There is no solution since
16 30
34 60
51 90
m m
R
Best fit is
30
60
90
16 30
34 60
51 90
proj m
16 30
34 6030
51 9060
30 3090
60 60
90 90
7110
12600m 0.56
Circulation duration of U.S. currency:
denomination 1 5 10 20 50 100
average life (yrs) 1.5 2 3 5 9 20
1 1
1 5
1 10
1 20
1 50
1 100
A
life = m denom + b
A x = vb
m
x
1.5
2
3
5
9
20
v
1T T A A A v
Best fit is projection of v onto column space of A :
1.05
0.18
bproj
m
A v
Fitting curve a f(x) + b g(x) to a set of data points { m(x1), …, m(xn) }
1 1
n n
f x g x
f x g x
A A x = ma
b
x
1
n
m x
m x
m
1T T A A A v
Best fit is projection of m onto column space of A :
aproj
b
A m
Exercises 3.Topics.1.
1. When the space shuttle Challenger exploded in 1986, one of the criticisms made of NASA’s decision to launch was in the way the analysis of number of O-ring failures versus temperature was made (of course, O-ring failure caused the explosion). Four O-ring failures will cause the rocket to explode. NASA had data from 24 previous flights.
The temperature that day was forecast to be 31ºF.(a) NASA based the decision to launch partially on a chart showing only theflights that had at least one O-ring failure. Find the line that best fits theseseven flights. On the basis of this data, predict the number of O-ring failureswhen the temperature is 31, and when the number of failures will exceed four.(b) Find the line that best fits all 24 flights. On the basis of this extra data,predict the number of O-ring failures when the temperature is 31, and when the number of failures will exceed four.Which do you think is the more accurate method of predicting?
Geometry of Linear Maps
1xf x e 2
2f x x 1 2h x x 2h x x
Linear MapsNon-Linear Maps
Rotation: Linear Projection: Linear
Every H can be written as H = P B Q, where B is a partial identity matrix,and P , Q are products of elementary matrices, Mi (k), Pi , j , and Ci , j (k).
B ~ projection.
Mi (k) ~ dilation
Pi , j ~ reflection
Ci , j (k) ~ skew
Linear map h maps subspaces into subspaces. E.g., line through 0 to line through 0.Dim h(V) cannot be greater than Dim V → a line can’t map onto a plane.
Calculus: Near x0 , f(x) f(x0) + (xx0) f (x0) is a linear map.
Multi-dim: f : n → m , f(x) f(x0) + [ (xx0) · ] f (x0)
Chain rule:
d g fx g f x f x
dx
f dilates ( x ) by a factor of f (x).
g dilates ( f (x) ) by a factor of g (f (x)).
g f dilates ( x ) by a factor of g (f (x)) f (x).
Multi-dim: proj, shear, etc…
Exercises 3.Topics.2.
1. What combination of dilations, flips, skews, and projections produces the map h: 3 → 3 represented with respect to the standard bases by this matrix?
1 2 1
3 6 0
1 2 2
Markov Chains
Player starts with 3 dollars & bets a dollar for each coin toss.
Game is over when he has no money or up to 5 dollars.
6 possible states: s0 , s1 , s2 , s3 , s4 , s5 with game over whenever s0 or s5 is reached.
Let pi (n) be the probability for him to be in state si after n tosses. Then
0 0
1 1
2 2
3 3
4 4
5 5
1 1 .5 0 0 0 0
1 0 0 .5 0 0 0
1 0 .5 0 .5 0 0
1 0 0 .5 0 .5 0
1 0 0 0 .5 0 0
1 0 0 0 0 .5 1
p n p n
p n p n
p n p n
p n p n
p n p n
p n p n
1 11 0.5 0.5i i ip n p n p n
1 21 0.5p n p n 0 0 11 0.5p n p n p n
4 31 0.5p n p n 5 5 41 0.5p n p n p n
i = 2, 3
mi j = probability of state j changing to state i
1i ji
m
Starting with 3 dollars:
Probability for game to be over is .125+.375 = .5 at n = 4,
and .396 + .59676 = .99276 at n = 24.
The coin toss is a Markov process / chain (no memory effect).
States 0 & 5 are absorptive.
M is the transition matrix.
p is the probability vector.
Orthonormal Matrices
Euclidean geometry:
2 figures are the same (congruent) if they have the same size & shape.
f : 2 → 2 is an isometry if it preserves distances.2 figures are congruent if they are related by an isometry.The followings are preserved under isometry:• Collinearity.• Between-ness of points.• Properties of a triangle.• Properties of a circle.
Klein’s Erlanger Program proposes that each kind of geometry — Euclidean, projective, etc.— can be described as the study of the properties that are invariant under some group of transformations.
Characterization of Isometries by Means of Linear Algebra
The only non-linear isometries is translation.1 1
0
x x x
y y y
If f is an isometry that sends 0 to v0 , then v f(v) v0 is linear.
A linear transformation t of the plane is an isometry iff
1 2 1t t e e and 1 2 0t t e e
Proof : From Pythagorean theorem.
Proof : Direct calculation.
Matrix representation of a linear isometry is orthonormal,
i.e., its columns are mutually orthogonal & of length one.
Note: most people call such matrices orthogonal & define it by M MT = I.
Rotation
Reflection
Euclidean study of congruence:(i) a rotation followed by a translation, or (ii) a reflection followed by a translation ( glide reflection).
2 Figures are similar if they are congruent after a change of scale.
i.e., if there exists an orthonormal matrix T s.t. points q & p on them are related by
0k q T p p