0% found this document useful (0 votes)
74 views

Quiz3 Solutions PDF

This document contains a quiz with multiple choice and free response questions about eigenvectors and eigenvalues of a parameter-dependent matrix A(t). Key points: - The matrix A(t) has the eigenvalue 1 with an eigenvector that does not depend on t. - The other eigenvalue is t-1 which depends on t, as does its corresponding eigenvector. - A(t) is not diagonalizable when t=2. - A recursion formula involving A(t) is given, which converges to a finite limit as k goes to infinity if 0<t<2, due to the eigenvalues of A(t). - The limit is calculated to be 1/(2-t)

Uploaded by

double K
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
74 views

Quiz3 Solutions PDF

This document contains a quiz with multiple choice and free response questions about eigenvectors and eigenvalues of a parameter-dependent matrix A(t). Key points: - The matrix A(t) has the eigenvalue 1 with an eigenvector that does not depend on t. - The other eigenvalue is t-1 which depends on t, as does its corresponding eigenvector. - A(t) is not diagonalizable when t=2. - A recursion formula involving A(t) is given, which converges to a finite limit as k goes to infinity if 0<t<2, due to the eigenvalues of A(t). - The limit is calculated to be 1/(2-t)

Uploaded by

double K
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 14

18.

06 Professor Edelman Quiz 3 May 4, 2018

Your PRINTED name is:

Please circle your recitation:


Grading
(1) T 10 26-328 D. Kubrak
(2) T 11 26-328 D. Kubrak
1
(3) T 12 4-159 P.B. Alvarez
(7) T 12 4-153 E. Belmont
2
(4) T1 4-149 P.B. Alvarez
(5) T2 4-149 E. Belmont
3
(6) T3 4-261 J. Wang

Note: We are not planning to use gradescope for this exam. Total:
Your Initials:

1 (30 pts.) Consider the matrices,


   
0 1 1 −1
A(t) =   + t .
1 0 0 0

a. (5 pts) Is it possible to find a vector v and a scalar λ that does not depend
on t that serves as an eigenvector/eigenvalue for A(t) for all t?

SOLUTION:
 
1
A(t) has eigenvalue λ1 = 1 with eigenvector v1 =   for all t. One way to
1
 
1 −1
see this is that v1 is in the null space of t   (so A(t)v1 does not depend
0 0
 
0 1
on t), and it also happens to be an eigenvector of  . Alternatively, you
1 0
could compute and factor the characteristic polynomial as we do in b.

b. (5 pts.) Find both an eigenvector and an eigenvalue of A(t) that does


depend on t.

SOLUTION:
 
t 1−t
The characteristic polynomial of A(t) =   is
1 0

det(A(t) − λI) = (t − λ)(−λ) − (1 − t)

= λ2 − λt + (t − 1)

= (λ − 1)(λ − (t − 1)).

2
(To help you factor this, you could use the fact from (a) that 1 is a root of the
characteristic polynomial.) So λ2 = t − 1, and
 
1 1−t
A(t) − λ2 I = A(t) − (t − 1)I =  
1 1−t
 
t−1
has null space spanned by v2 =  , so that is the corresponding eigenvec-
1
tor.

c. (5 pts.) For which t, if any, is the matrix A(t) not diagonalizable. Explain
briefly.

SOLUTION:

If λ1 6= λ2 , then A(t) is diagonalizable. So the only case in which non-


diagonalizability is possible is when 1 = t − 1, i.e., t = 2. In this case, the
characteristic polynomial is (λ − 1)2 , andA(2) isdiagonalizable if and only if
1 −1
dim N (A(2) − I) = 2. But A(2) − I =   which has rank 1 and null
1 −1
space of dimension 1. So it is not diagonalizable at t = 2.

3
Your Initials:

d. (5 pts.) Consider the sequence x0 = 0, x1 = 1, xk+2 = t ∗ xk+1 + (1 − t) ∗ xk .


You can assume 0 < t < 2. Why does xk converge to a finite number as
k → ∞? Explain briefly.

SOLUTION: Briefly, this happens because A(t) has eigenvalues 1 and t − 1 and
since 0 < t < 2 we have |t − 1| < 1.

In more detail, the recursion on xk ’s can be rewritten as


    
x t 1 − t xk+1
 k+2  =   
xk+1 1 0 xk

so we encounter our matrix A(t) and get a formula


         
xk+1 xk x x 1
  = A(t)   = A(t)2  k−1  = . . . = A(t)k  1  = A(t)k  
xk xk−1 xk−2 x0 0
   
1 t−1
Recall that v1 =   and v2 =   are eigenvectors of A(t) with eigenval-
1 1
ues 1 and t − 1. By part c if t 6= 2 the matrix A(t)is diagonalizable and v1 , v2
1
form an eigenbasis. So we can express our vector   as c1 (t)v1 + c2 (t)v2 for
0
some scalars c1 , c2 ∈ R that depend on t. Now we have
 
1
A(t)k   = A(t)k (c1 v1 + c2 v2 ) = c1 (A(t)k v1 ) + c2 (A(t)k v2 ) = c1 v1 + c2 (t − 1)k v2
0

Since |t − 1| < 1 we have limk→∞ c1 v1 + c2 (t − 1)k v2 = c1 v1 , in particular the


limit is a finite vector. Also we get that
   
xk+1 c
lim   = c1 v1 =  1 
k→∞
xk c1

Remark: Note that A(t) is not necessarily Markov since 1 − t can be negative
(if t > 1). To use that was a pretty common mistake on the exam.

4
Your Initials:

e. (10 pts.) (Recommended to do this after completing all other work on the
exam.) Calculate the limit
 of xk from part d as k goes to infinity. (Hint:
xk+1
Consider the vector  .) (Check: If t = 1/2 the limit is 2/3.)
xk

SOLUTION: By what we proved in part d)


   
xk+1 c
lim   = c1 v1 =  1 
k→∞
xk c1
 
1
where c1 v1 + c2 v2 is the decomposition of   in the eigenbasis v1 , v2 . So to find the limit it
0
is enough to find such decomposition. For this note that
         
t−1 1 t−2 1 1
v2 − v1 =  − =  = (t − 2)   , so   = −1 v1 + 1 v2
1 1 0 0 0 t−2 t−2

1 1
We get that c1 = and so limk→∞ xk = . In particular for t = 1/2 we get 2/3.
2−t 2−t
Remark: Another way to do this problem was to diagonilize matrix A(t), since v1 , v2 are
eigenvectors with eigenvalues 1, t − 1 we have
   −1
| |   | |     
 1 0 1 t−1 1 0 1 1−t
 1 
  
A(t) = v1 v1 v2  = 
   
v2    
  0 t−1   1 1 0 t−1 2 − t −1 1
| | | |

and so
   −1
| |   | |     
 1 0 1 t − 1 1 0 1 1 − t
 1 
  
A(t)k = v1 v1 v2  =  =
   
v2   
 0 (t − 1)k  k 2−t
  1 1 0 (t − 1) −1 1
| | | |
  
k+1
1  1 (t − 1) 1 1 − t
=  
2 − t 1 (t − 1)k −1 1

5
When k goes to infinity (t − 1)k dies we get
    
1  1 0 1 1 − t 1 1 − t
lim A(t)k =  = 1  
k→∞ 2−t 1 0 −1 1 2 − t 1 1−t

Finally      
limk→∞ xk+1 1 1/(2 − t)
  = lim A(t)k   =  
k→∞
limk→∞ xk 0 1/(2 − t)
So we get
1
lim xk = .
k→∞ 2−t

6
Your Initials:

2 (30 pts.) In all cases find a two by two matrix which has the given eigenvalues
and the given singular values or explain why it is impossible. Do not use AT A
or AAT in any of your explanations.
a. (5 pts) λ = 0, 1, σ = 1, 1

b. (5 pts) λ = 0, 1 σ = 0, 2
c. (5 pts) λ = 0, 0 σ = 0, 2018
d. (5 pts) λ = i, −i σ = 1, 1
e. (5 pts) λ = 4, 4 σ = 3, 5
q √
f. (5 pts.) λ = −1, 1 σ = (3 ± 5)/2 (You can trust that σ1 σ2 = 1 and σ12 + σ22 = 3)

SOLUTION:
Throughout these solutions, we will use the following facts. If λ1 andλ2 are the eigenvalues
a b
and σ1 and σ2 are the singular values of the 2 × 2 matrix A =  , then
c d

1. σ1 · σ2 = | det(A)| = |λ1 · λ2 |.

2. σ12 + σ22 = a2 + b2 + c2 + d2 .
 
a b
3. An upper triangular matrix   has eigenvalues a and d.
0 d

Now, we have the following.

(a) Since one of the eigenvalues is 0, the matrix A cannot have full rank. But both singular
values are nonzero, so the matrix has rank 2. This is a contradiction, so A does not
exist.

(b) The matrix  


0 1
A= 
0 1

7
satisfies these conditions. It is upper triangular, so by fact 3, the eigenvalues are 0 and
1. The singular values have to satisfy σ1 · σ2 = 0 and σ12 + σ22 = (−1)2 + 12 = 2 by facts
1 and 2. The solution to this pair of equations is unique (up to reordering σ1 and σ2 ),

and so the singular values are 2 and 0.

(c) We claim that the matrix  


0 2018
A= 
0 0
satisfies these conditions. It is upper triangular, so by fact 3, the eigenvalues are 0 and
0. By facts 1 and 2, we have that σ1 · σ2 = 0 and σ12 + σ22 = 20182 , and so the singular
values are 0 and 2018.

(d) Let  
0 −1
A= .
1 0
This is a rotation counterclockwise by 90 degrees. We can compute its eigenvalues by
calculating det(A − λI) = λ2 + 1 = 0 to get λ = ±i. We can see geometrically that a
rotation has singular values 1 and 1 since it rotates a unit circle in R2 onto the unit
circle again. Alternatively, we can solve for σ1 · σ2 = 1 and σ12 + σ22 = (−1)2 + 12 = 2
to get that the singular values are 1 and 1.

(e) Since λ1 · λ2 = 16 and σ1 · σ2 = 15 6= |16|, by fact 1, a matrix A with these eigenvalues


and singular values does not exist.

(f) Let  
−1 1
A= .
0 1
The matrix is upper triangular and so by fact 3 has eigenvalues −1 and 1. By facts 1
and 2, the singular values satisfy σ1 · σ2 = 1 and σ12 + σ22 = (−1)2 + 12 + 12 = 3, and
q √
so must be the σi = (3 ± 5)/2.

8
Your Initials:
3 (40 pts.) Are the following matrices necessarily positive definite? Explain
why or why not?
a. (5 pts) A = QΛQT where Q is some 4x4 orthogonal matrix and Λ is diagonal
with (1, 2, 3, 4) on the diagonal.
b. (10 pts) A = Q1 ΛQT1 + Q2 ΛQT2 , where Q1 and Q2 are some 4x4 orthogonal
matrices and Λ is diagonal with (1, 2, 3, 4) on the diagonal.
c. (5 pts) A = XΛX T for some matrix X and Λ is as above? (Hint: Be careful.)
d. (5 pts.) P the projection matrix onto (1, 2, 3, 4).
e. (15 pts.) A is the n by n tridiagonal matrix with 2 for each diagonal entry, and 1
for each superdiagonal and subdiagonal entry. n = 1, 2, 3, . . .. (Hint: Probably the
easiest argument involves computing the determinant of T (n) for n = 1, 2, 3, . . ..)
SOLUTION:

(a) First this matrix is clearly symmetric as AT = A. Since Q is orthogonal QT = Q−1 and
so this gives a diagonalization, so A has eigenvalues 1, 2, 3, 4 > 0 and is thus positive
definite.

(b) From part a) we know that each of Ai = Qi ΛQTi are positive definite, so for x 6= 0
xT Ai x > 0. So we get xT Ax = xT A1 x + xT A2 x > 0 and it follows that A is positive
definite as A is clearly symmetric as it is the sum of symmetric matrices.

(c) We have AT = A, so A is symmetric. Also v T Av = (X T v)T Λ(X T v) ≥ 0 as Λ positive


definite and the inequality is strict as long as X T v 6= 0. So we get that A is not positive
definite as long as X T is not full column rank.

(d) This is a projection matrix to a 1 dimensional space, so P has rank 1. It thus has a
non-trivial nullspace and so has a 0 eigenvalue. So not all eigenvalues are positive and
thus is not positive definite.

9
(e)  
2 1 0 0 ··· 0 0
 
1 2 1 0 ··· 0 0
 
 
0 1 2 1 ··· 0 0
 
 
An = 0 2 ··· 0
 
0 1 0
 .. .. .. .. . . .. .. 
 
. . . . . . .
 
0 0 0 ···
 
0 2 1
 
0 0 0 0 ··· 1 2
Lets denote by T (n) the determinant of the nxn matrix An . By expanding the deter-
minant along the first column, we get the formula
 
1 0 0 ··· 0
 
1 2 1 · · · 0
 
 
T (n) = 2T (n − 1) − det(0 1 2 · · · 0) = 2T (n − 1) − T (n − 2)
 
 .. .. .. . . .. 
 
. . . . .
 
0 0 0 ··· 2

Were we expand the second determinant along the first row. Also T (1) = 2 and
T (2) = 3, so we can check T (n) = n + 1 > 0. So we have An is symmetric and all top
left corner determinants are positive so it is positive definite.

10
Your Initials:

11
Your Initials:

Extra Page. Please write problem number and letter if needed.

12
Your Initials:

Extra Page. Please write problem number and letter if needed.

13
Your Initials:

Extra Page. Please write problem number and letter if needed.

14

You might also like