Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: daixieit

Math 170A, FA 2022

Midterm review

Q1.  (10pts) Consider the following matrix

A =  1

1

≠ 1

1.(、) .

(a) perform the following operations on the matrix A:

● Step 1: update row 2: add 3 ↓ row 1 to row 2.

● Step 2: swap row 3 and row 2.

Write down two matrices (L, P) perform these operations, so that PLA realizes this transformation. (2pts)

Your Answer:

L =  3(1)   0

(

0(0)

1 ,

P =  0(1)   0(0)

(0   1

1(0)

0 ,

Write down the resulting matrix PLA. (1pt)

Your Answer:

PLA =  0(2)

(

 1

 1

 1


2(0) 

 1 .

(b) Determine if the matrix A is positive definite and symmetric. (4pts)

Your Answer:

Proof. First, it is clear that A is symmetric.

To show positive definite, we try to do the Cholesky decomposition on A:

Assume RT R = A where

R =  T0(11)

(

 

T12

T22

0

T13  

First, T 11(2)  = 2, thus T11  = ^2 and T12  =   , T13  = 0.

(c)  Compute #A#F , #A#1  and #A#o . (3pts)

Your Answer:

#A#F  = ìùi,jAij(2)  =^3 ì 22 + 4 ì ( 1)2  = ^16 = 4

3

#A#1  = m x ù 怅Aj(a) i,j = max(3, 4, 3) = 4

i=1

3

#A#o  = m x ù 怅Ai(a) i,j  = max(3, 4, 3) = 4

j=1

Q2.  (5pts) Vector norm

(a)  Given the Cauchy-Schwarz inequality: for all x, y ↓ Rn ,

 xiyj    xi(2)\1/2   yi(2)\1/2 ,

prove that

怅怅x怅怅2  =   xi(2)\1/2

is a vector norm. (4pts)

Proof. To show # ì #2  is a norm, we need to prove that it satisfies the 3 properties of vector norms.

i. For x  0, at least one of xi   0, which means ( xi(2)  > 0, thus #x#2  > 0. If x = 0, then #x#2  = (( 02 )1/2  = 0.

ii. For a ↓ R,

n                                          n                                    n

#ax#2  = (ù a2 x )i(2) 1/2  = (a2 ù x )i(2) 1/2  = a(ù xi(2))1/2  = a#x#2

i=1                                     i=1                                i=1

iii. Lastly, we want to show the triangle inequality:

n                                    n                                      n

#x + y#2(2)  =ù (xi + yi )2  =ù (xi(2) + y ) + 2 ùi(2) xiyi


i=1 n

n

i=1

n


i=1

n


 ù xi(2) +ù yi(2) + 2(ù x )i(2) 1/2(ùyi(2))1/2

i=1               i=1                   i=1                  i=1

n                             n

= ((ù x )i(2) 1/2 + (ù yi(2))1/2)2  = (#x#2 + #y#2 )2

i=1                         i=1

Take the square root of both side of the above inequality, we have #x + y#2  女 #x#2 + #y#2 . 

(b)  Given

x =  1.(、) ,

Compute 怅怅x怅怅1 , 怅怅x怅怅o . (1pt)

#x#1  =ù 怅xi  = 2 + 1 + 0 = 3

i

#x#o  = max辛怅2  10怅含 = 2

Q3.  (10pts) An experiment has produced the following four points of 2-dimensional data:

(0, 1), (1, 3), (2, 1), and (4, 2).

a)  Set up a least squares problem for the best line t for this set of points, explicitly writing out the matrix A and the vector y . (3pts)

Your Answer:   Suppose the line is given by P (t) = a0 + a1t. Then the residual

T is given by T = y Ax:

 =  3(1)  

2 │(│) a(a)1(0)

The least square problem is given by

min #y Ax#2(2)

.

b)  Set up a least squares problem for the best quadratic t for this set of points,

explicitly writing out the matrix A and the vector y . (3pts)

Your Answer:

Suppose the line is given by P (t) = a0 + a1t + a2t2 . Then the residual T is given

 =  3(1)     

The least square problem is given by

1(0) 

4   

16

a(a)1(0)

(a2

min #y Ax#2(2)

.

c) Based on your answer of b), use the  classical  Gram-Schmidt  process to

calculate the reduced QR decomposation of A. (4pts)

Your Answer:

Let v1  = [1 1 1 1]T , v2  = [0 1 2 4]T , and v3  = [0 1 4 16]T . Then

1            1           

q1  = #v1 #2 v1  = 2 v1  =  .(.)    │(│) ,

q˜3  = v3  < v3 , q1  > q1  < v3 , q2  > q2  = ( 10.5( 12.2547(0..075(2(1)..47(9)

q3  = q˜3  = q˜3  =

Therefore, the QR decomposition is given by:

        =     0(0)..05(6)   .005.7(3)

0.5641

( .

2.00(0)00 (    0

3.5000

2.9580

0

 3.5456

(Bonus) Based on your reduced QR decomposation of Q3. c), compute the minimizer

x that solve the least squares problem. (1pt) (Hint: understand how reduced QR can help solve the least square problems.)

Your Answer:   x = [≠ 1.6818 0.3409 0.1591]T