MATH 218D

PRACTICE MIDTERM EXAMINATION 3

Name Duke Email @duke.edu

Please read all instructions carefully before beginning.

• You have 180 minutes to complete this exam and upload your work. The exam itself
is meant to take 75 minutes to complete, so hopefully you will have enough time.
• For full credit you must show your work so that your reasoning is clear.
• If you need clarification or think you’ve found a typo, ask a private question on
Piazza. We’ll be monitoring it.
• If you have time, go back and check your work.
• You may use your class notes (not the ones from the website) and the interactive
row reducer during this exam. You may use a calculator for doing arithmetic. All
other materials and aids are strictly prohibited.
• You are not allowed to receive outside help during this exam. Consulting with some-
one else is considered cheating; suspected instances will result in immediate referral
to to the Office of Student Conduct.
• Be sure to tag your answers on Gradescope, and use a scanning app.
• Good luck!

Complete when starting the exam: I will neither give nor receive aid on this exam.

Signed: Time:

Complete after finishing the exam: I have neither given nor received aid on this exam.

Signed: Time:

This is a practice exam. It is meant to be similar in format, length,

and difficulty to the real exam. It is not meant as a comprehensive list
of study problems. I recommend completing the practice exam in 75
minutes, without distractions.
Problem 1. [20 points]

Consider the following matrix and its singular value decomposition A = UΣV T :
 p p p p 
1/p10 1/p15 1/ 2 −1/ 3 3 0

0  p p p T
1/ 3 −1/p6 −1/ 2
 −2/ 10 3/ 15 0 0 0 2 0 p
A= p p p  −1/ 3 −2/p6 0 .
 2/ 10 2/ 15 0 1/  0 0 0 p p
p p p p3 1/ 3 −1/ 6 1/ 2
−1/ 10 −1/ 15 1/ 2 1/ 3 0 0 0

From this you can read off all of the following properties of A without computing A.

a) A is a × matrix of rank r = .

b) Find orthonormal bases of the four fundamental subspaces of A.

c) Compute the matrix PV for orthogonal projection onto V = Col(A) (write it as a
product, without expanding it out).
d) Write the SVD of A in vector form.
e) Find an orthogonal diagonalization AT A = QDQ T .

Solution.
a) The size and rank of A can be read off from Σ: A is a 4 × 3 matrix of rank 2.
b) The columns of U and V give orthonormal bases for the four subspaces.
−1 1 −1
( !) ( ! !)
1 1 1
Nul(A): p 0 Row(A): p −1 , p −2
2 1 3 1 6 −1
         
 1 1   1 −1 
1 −2 1  3 1 0 1 0
   
Col(A): p   , p   Nul(AT ): p   , p   .
  
2 2  0 1 
 10
 15  2
 3
−1 −1 1 1
 

c) We have an orthonormal basis for Col(A) from b); putting these vectors in a matrix
Q, we have PV = QQ T for
 p p 
1/p10 1/p15
 −2/ 10 3/p15 
Q=
 2/ 10
p 
p 2/ p15 
−1/ 10 −1/ 15
d) This is the vector form of the SVD:
   
1 1
1 −2 1
1  3 1

A = 3 · p   · p 1 −1 1 + 2 · p   · p −1 −2 −1
10 2 3 15 2 6
−1 −1
e) We can take
 p p p 
1/p3 −1/p6 −1/ 2 9 0 0
!
Q =  −1/ 3 −2/p6 D= 0 4 0 .
p0

p
1/ 3 −1/ 6 1/ 2 0 0 0
Problem 2. [20 points]

Consider the symmetric matrix

1 0 2
!
S= 0 −1 −2 .
2 −2 0

a) Find an orthogonal matrix Q and a diagonal matrix D such that S = QDQ T .

b) Which of these adjectives describe S?
• positive-definite
• positive-semidefinite
• negative-definite
• negative-semidefinite
• indefinite
c) Write the singular value decomposition of S in matrix form.
d) Find the maximum value of the quadratic form q(x) = x T S x subject to kxk = 1. At
which vectors is this value obtained?

Solution.
a) First we compute the eigenvalues of S. We find the characteristic polynomial by
expanding cofactors along the first column:
1−λ 0 2
!
p(λ) = det 0 −1 − λ −2
2 −2 −λ
−1 − λ −2
 ‹  ‹
0 2
= (1 − λ) det + 2 det
−2 −λ −1 − λ −2
   
= (1 − λ) (−1 − λ)(−λ) − 4 + 2 −2(−1 − λ)
= (1 − λ)(λ2 + λ − 4) − 4(−1 − λ)
= λ2 + λ − 4 − λ3 − λ2 + 4λ + 4 + 4λ
= −λ3 + 9λ = −λ(λ − 3)(λ + 3).
The eigenvalues are 0 and ±3; we compute eigenvectors:
 
−2 0 2 1 0 −1 2
! !
RREF 1 1
λ = 3 : S − 3I = 0 −4 2 0 1
2 v1 = −1
2 −2 −3 0 0 0 3 2
 
1
4 0 2 1 0 −1
! !
RREF 2 1
λ = −3 : S + 3I = 0 2 2  0 1 −1  v2 = 2
2 −2 3 0 0 0 3 2
1 0 2 1 0 2 −2
! ! !
RREF 1
λ = 0 : S − 0I = 0 −1 −2 0 1 2 v3 = −2
2 −2 0 0 0 0 3 1
 Hence S = QDQ T for
2 −1 −2 3 0 0
! !
1
Q= −1 2 −2 D= 0 −3 0 .
3 2 2 1 0 0 0
b) S has a positive and a negative eigenvalue, so it is indefinite.
c) The singular values of S are the absolute values of the nonzero eigenvalues: σ1 =
σ2 = 3. We have
3v1 = Sv1 = σ1 u1 =⇒ u1 = v1
−3v2 = Sv2 = σ2 u2 =⇒ u2 = −v2 .
We can take u3 = v3 as our orthonormal basis of Nul(S) = Nul(S T ), so
!T
2 1 −2 3 0 0 2 −1 −2
! !
1 1
S= −1 −2 −2 0 3 0 −1 2 −2 .
3 2 −2 1 0 0 0 3 2 2 1
d) The maximum value is λ1 = 3. It is achieved at ±v1 = ± 13 (2, −1, 2).
Problem 3. [20 points]

Consider the difference equation vn+1 = Avn for

2 −1
 ‹
A= 3 1 .
2 −2
1


a) Find a closed formula for An v0 for v0 = 2 . What happens when n → ∞?
b) In the diagram, draw and label the eigenspaces of A, and draw the vectors v0 , v1 ,
v2 , v3 , . . . as points. (The grid lines are one unit apart, and the dot is the origin.)
[Hint: you do not have to compute An v0 numerically to do this.]

c) Solve the system of ordinary differential equations

u01 = 2u1 − u2 u1 (0) = 1
u02 = 23 u1 − 12 u2 u2 (0) = 2.

Solution.
a) The characteristic polynomial of A is
3 1 1
 ‹
p(λ) = λ − Tr(A)λ + det(A) = λ − λ + = (λ − 1) λ −
2 2
.
2 2 2
The eigenvalues are λ1 = 1 and λ2 = 21 . We compute eigenvectors:
 ‹  ‹
1 −1 1
A − 1I = w1 =
− − 1
3
1
‹  ‹
−1 2
A− I = 2 w2 =
2 − − 3
Hence A = C DC −1 for
 ‹  ‹
1 2 1 0
C= D= .
1 3 0 12
We eyeball v0 = −w1 + w2 , so
1
An v0 = −w1 + w2 .
2n
 −1


This approaches −w1 = −1 as n → ∞.
b) The eigenspaces are spanned by w1 and w2 .
1/2
1
v0

w2

−w1

1


c) We need to solve u0 = Au for u(0) = 2 = v0 and A as above. We have u(0) =
−w1 + w2 , so the solution is
u1 = −e t + 2e t/2
u(t) = −e t w1 + e t/2 w2 =⇒
u2 = −e t + 3e t/2
Problem 4. [20 points]

All of the following statements are false. Provide a counterexample to each. You need
not justify your answers.
a) The singular values of a diagonalizable, invertible 2×2 matrix are the absolute values
of the eigenvalues.
b) If S is symmetric, then either S or −S is positive-semidefinite.
−1 2


c) If A = −3 4 and x 6= 0, then kAn xk → ∞ as n → ∞.
d) If λ is an eigenvalue of AAT , then λ is an eigenvalue of AT A.
e) Any invertible matrix is diagonalizable.

Solution.
1 1


a) The matrix A = 0 2 has eigenvalues 1 and 2, so it is invertible and diagonalizable.
However,  ‹ ‹  ‹
1 0 1 1 1 1
A A=
T
=
1 2 0 2 1 5
has characteristic polynomial p(λ) = λ2 − Tr(A)λ + det(A) = λ2 − 6λ + 4; we have
p(12 ) = −1 and p(22 ) = −4, so neither 1 nor 2 is a singular value of A.
b) Any symmetric matrix S with both positive and negative eigenvalues works:
 ‹
1 0
S= .
0 −1
c) The characteristic polynomial if A is p(λ) = λ2 − 3λ + 2 = (λ − 1)(λ − 2). An
1


eigenvector with eigenvalue 1 is x = 1 ; for this vector, we have An x = x, so the
length does not grow.
d) This can only be false for λ = 0. Zero is an eigenvalue of AT A (resp. AAT ) if and only
if A has linearly dependent columns (resp. rows), so we need to find a matrix with
linearly independent columns and linearly dependent rows. For instance:
1 0
!
A= 0 1 .
0 0


e) The shear matrix 10 11 is invertible but not diagonalizable.
Problem 5. [16 points]

All of the following statements are true. Explain why in a sentence or two.
a) If A is a 3 × 3 matrix that has eigenvalues 1 and −1, both of algebraic multiplicity
one, then A is diagonalizable (over the real numbers).
b) Let V be a subspace of Rn and let PV be the matrix for projection onto V . Then PV is
diagonalizable.
c) Any eigenvector of A with nonzero eigenvalue is contained in the column space of
A.
d) A positive definite symmetric matrix has positive numbers on the main diagonal.

Solution.
a) The third eigenvalue is also a real number with multiplicity one.
b) The 1-eigenspace of PV is V , and the 0-eigenspace is V ⊥ . The geometric multiplicities
add to n because dim(V ) + dim(V ⊥ ) = n.
Alternatively, PV is symmetric, so it is diagonalizable by the spectral theorem.
c) If Ax = λx for λ 6= 0 then x = A λ1 x is a linear combination of the columns of A.
d) The (i, i) entry of a matrix S is eiT Sei , which is positive if S is positive definite.
Problem 6. [10 points]

Draw a picture of the ellipse defined by the equation

30x 12 + 35x 22 + 12x 1 x 2 = 1.
Be precise! Label your major and minor axes and radii.

Solution.
First we diagonalize the quadratic

form q(x 1 , x 2 ) = 30x 12 + 35x 22 + 12x 1 x 2 . We have
q(x) = x T S x for S = 30 6
6 35 ; this matrix has orthogonal diagonalization S = QDQ T for
1



Q = p13 23 −2 3
and D = 39 0
0 26 . Hence our ellipse is obtained from the standard ellipse
39 y12 + 26 y22 = 1 by multiplication by Q.

1
p w
39 1

1
 ‹
2
w1 = p
13 3 ‹
1 3
1
p w w2 = p
26 2 13 −2