18.

06 Professor Edelman Final Exam December 15, 2010

Grading

1
Your PRINTED name is:
2

4
Please circle your recitation: 5

7
R01 T9 2-132 S. Kleiman 2-278 3-4996 kleiman
8
R02 T 10 2-132 S. Kleiman 2-278 3-4996 kleiman
9
R03 T 11 2-132 S. Sam 2-487 3-7826 ssam
R04 T 12 2-132 Y. Zhang 2-487 3-7826 yanzhang
R05 T1 2-132 V. Vertesi 2-233 3-2689 18.06
R06 T2 2-131 V. Vertesi 2-233 3-2689 18.06
1 (16 pts.)

1 0 e

a. (8 pts) Give bases for each of the four fundamental subspaces of A = 0 1 1 0



0 0 0 0
Clearly the rst two columns are independant and generate the column space.
The left null space is the orthogonal of the column space. A basis is given by the vector :

0


0

1
A possible basis for the nullspace is :

e


1 0
,
1 0


0 1

The row space has dimension 2. The two rst rows clearly generate it, therefore they form
a basis.

2
b. (8 pts) Give bases for each of the four fundamental subspaces of

1 1 1 1 1 1 0 0 0

1 1 1

1 2 0 1 0 0
A=

1 1 1 1 3 0 0 1 1


1 1 1 1 0 0 0 1 1

(Each of the three matrices in the above product has orthogonal columns.)
The matrices on both side are non singular. Therefore the rank of the product is 3.
To compute the column space, we can restrict to the product of the rst two matrices. The
third is invertible and will not change the column space. Clearly the column space of the
product of the rst two matrices is spanned by the rst three columns of the leftmost matrix.
Since these vectors are independant they form a basis.
To compute the row space we can restrict to the last two matrices of the product for the
same reason as above. We nd that the row space has as basis the rst three rows of the
rightmost matrix.
Since the nullspace is the orthogonal of the row space, and we know that the rows of the
rightmost matrix are orthogonal, the last row of the right most matrix is a basis for the
nullspace.
Similarly, the last column of the leftmost matrix is a basis for the left nullspace.
Remark. Another way to see it is to notice that this is almost the SVD of A (we just need
to normalize the columns of the leftmost and rightmost matrix). Our answer is exactly the
usual way to nd a basis of the four fundamental subspaces when we have found the SVD.

3
This page intentionally blank.

4
 2 (14 pts.)

Let P1 be the projection matrix onto the line through (1, 1, 0) and P2 is the projection onto
the line through (0, 1, 1).

(a) (4 pts) What are the eigenvalues of P1 ?

The eigenvalue of a projection matrix are 1 and 0. The question remain of the multi-
plicity of each eigenvalue. The multiplicity of 0 is the dimension of the nullspace which
is 3 the rank. Here P1 is a projection on a line, therefore the columnspace is made
of vector on that line and has dimension 1. Therefore 0 has multiplicity 2 and 1 has
multiplicity 1.

(a2)(bonus) (This question is from an earlier version of the exam.)

Find an eigenvalue and an eigenvector of P1 + P2 .
The problem asks only for one eigenvalue and one eigenvector, but since you're taking
this exam for practice, you may as well nd all three.
For simplicity of notation, let ~a1 := (1, 1, 0) and ~a2 := (0, 1, 1). Note that the vectors

~a1 and ~a2 have the same length, 2. Since ~a1 ~a2 = 1 = 2 cos(60 ), we know that the
angle between ~a1 and ~a2 is 60 . It is possible to see the following facts geometrically:

First, ~v1 := ~a1 ~a2 = (1, 1, 1) is an eigenvector of P1 + P2 , with eigenvalue

1 := 0. Indeed, since v1 is perpendicular to both (1, 1, 0) and (0, 1, 1), we know
that v1 lies in the nullspace of both P1 and P2 , and hence in the nullspace of
P1 + P2 as well.

Second, ~v2 := ~a1 + ~a2 = (1, 2, 1) is an eigenvector of P1 + P2 , with eigenvalue

2 := 2 cos2 (60 /2) = 1 + cos(60 ) = 3/2.

Third, ~v3 := ~a1 ~a2 = (1, 0, 1) is an eigenvector of P1 + P2 , with eigenvalue

3 := 2 sin2 (60 /2) = 1 cos(60 ) = 1/2.

5
 The second and third facts can be derived with a bit of trigonometry, but if you don't
want to get into that, you can just do the usual linear-algebra calculation. Note that

1 1 1 0
1 h i 1
P1 = 2 1 1 1 0 = 1 1 0 ,

1 + 12 + 0 2 2
0 0 0 0

and similarly
0 0 0
1
P2 = 0 1 1 ,

2
0 1 1
so that
1 1 0
1
P1 + P 2 = .

1 2 1
2
0 1 1
You may solve for the roots 1 , 2 , 3 of det(P1 + P2 I) = 0 and nd the nullspaces
of P1 + P2 i I (for i = 1, 2, 3) as usual, and get the results that we described above.

(b) (10 pts) Compute P = P2 P1 . (Careful, the answer is not 0)

Let's just do it directly:

0 0 0 1 1 0 0 0 0
1 1
P = 0 1 1 = 1 1 0 .

1 1 0
4 4
0 1 1 0 0 0 1 1 0

6
This page intentionally blank.

7
 3 (10 pts.)

The nullspace of the matrix A is exactly the multiples of (1, 1, 1, 1, 1).

(a) (2 pts.) How many columns are in A?

Five columns. Otherwise it doesn't make sense to multiply A with the given vector.

(b) (3 pts.) What is the rank of A?

We know that the sum of the rank of A and the dimension of the nullspace is the
number of columns. Since the nullspace has dimension 1, the rank must be 4.

(c) (5 pts.) Construct a 5 5 matrix A with exactly this nullspace.

We need to construct a matrix whose sum of columns is 0 and with rank 4.
The following works :

1 0 0 0 1

0 1 0 0 1


1

0 0 1 0

1

0 0 0 1

0 0 0 0 0

8
This page intentionally blank.

9
4 (15 pts.)

Find the solution to

du 1 2
= u
dt 1 2

3
starting with u(0) = .
0
(Note the minus sign.)
The general formula for the solution to this dierential equation is entirely analogous to the
formula you learned in 18.01:
1 2
t


1 2

~u(t) = e ~u(0).

Of course, you should actually rewrite this in simplest form.

To facilitate taking the matrix exponential, let's diagonalize

1 2
.
1 2

The eigenvalues are easily computed to be 0 (because the matrix is obviously singular) and
3 (because the trace is 3), and the corresponding eigenvectors are (2, 1) and (1, 1), so
1
1 2 2 1 0 0 2 1
= .
1 2 1 1 0 3 1 1

It follows that

1 2 0 0
1 1
t t


0t

1 2

2 1
0 3

2 1 2 1 e 0 2 1
e = e = .
1 1 1 1 1 1 0 e3t 1 1

Now 1
2 1 1 1 1
= ,
1 1 3 1 2

10
so

0t
1 2 1 e 0 1 1
~u(t) = ~u(0)
3 1 1 0 e 3t
1 2

1 2 1 1 0 1 1 3
=
3 1 1 0 e3t 1 2 0

2 1 1 0 1 1 1
=
1 1 0 e3t 1 2 0

2 1 1 0 1
=
1 1 0 e3t 1

2 1 1
=
1 1 e3t

3t
e +2
= .
3t
e 1

11
This page intentionally blank.

12
 5 (10 pts.)

The 3 3 matrix A satises det(tI A) = (t 2)3 .

(a) (2pts) What is the determinant of A?

If we take t = 0, we nd det(A) = 8. Since A is 3 3, det(A) = det(A).
Therefore det(A) = 8.

(b) (8pts) Describe all possible Jordan normal forms for A.

The eigenvalues of A are 2 with multiplicity 3. The possible Jordan form of A are :

2 1 or 0 0

1 or

0 2 0

0 0 2

There are 4 possibilities.

13
This page intentionally blank.

14
 6 (7 pts.)

1 0
The matrix A =
C 1

(a) (2 pts) What are the eigenvalues of A?

This is a triangular matrix, therefore the eigenvalues are the diagonal entries 1 and 1.

(b) (5 pts) Suppose 1 and 2 are the two singular values of A. What is 12 + 22 ?
12 and 22 are the eigenvalues of AT A. The question is asking for the trace of AT A.

2
1 + C ?
AT A =
? 1

(The question marks are entries that we don't care about)

The trace is 2 + C 2 .

15
This page intentionally blank.

16
 7 (8 pts.)

For each transformation below, say whether it is linear or nonlinear, and briey explain why.

(a) (2 pts) T (v) = v/kvk

Not linear. For a positive real number c, T (cv) = T (v) and if T was linear we would
have T (cv) = cT (v).

(b) (2 pts) T (v) = v1 + v2 + v3

Linear. We have T (0) = 0, T (v + w) = T (v) + T (w) and T (cv) = cT (v).

(c) (2 pts) T (v) = smallest component of v

1
Not linear. Let's say that v is 2-dimensional. Take v = , then T (v) = 1 6=
0
T (v) = 0.

(d) (2 pts) T (v) = 0

Linear. Clearly satises all the requirement.

17
This page intentionally blank.

18
 8 (10 pts.)

V is the vector space of (at most) quadratic polynomials with basis v1 = 1,v2 = (x 1),v3 =
(x 1)2 . W is the same vector space, but we will use the basis w1 , w2 , w3 = 1,x,x2 .

(a) (5 pts) Suppose T (p(x)) = p(x + 1). What is the 3 3 matrix for T from V to W in the
indicated bases?
We have to compute T (v1 ), T (v2 ), T (v3 ) and write them in the basis w1 , w2 , w3 . We
have T (v1 ) = 1 = w1 , T (v2 ) = (x + 1 1) = x = w2 and T (v3 ) = (x 1 + 1)2 = w3 .
Therefore the matrix of T in these basis is the identity matrix :

1 0 0


0 1 0

0 0 1

Note that since the two basis are dierent, this does not imply that T is the identity.

(b) (5 pts) Suppose T (p(x)) = p(x). What is the 3 3 matrix for T from V to W in the indicated
bases?
We do the same thing, T (v1 ) = w1 , T (v2 ) = x 1 = w2 w1 , T (v3 ) = x2 2x + 1 =
w3 2w2 + w1 . Therefore the matrix of T is :

1 1 1

2

0 1

0 0 1

19
This page intentionally blank.

20
 9 (10 pts.)

In all of the following we are looking for a real 2 2 matrix or a simple and clear reason that
one can not exist.
Please remember we are asking for a real 2 2 matrix.

(a) (2 pts) A with determinant 1 and singular values 1 and 1.

1 0
Take A = . Clearly |A| = 1 and AT A = I therefore, the singular values are
0 1
1 and 1.

(b) (2 pts) A with eigenvalues 1 and 1 and singular values 1 and 0.

This is impossible. If the eigenvalues are 1 and 1, the matrix is non-singular, therefore
it has two non-zero singular values (in general the number of non-zero singular values
is the rank of the matrix).

(c) (2 pts) A with eigenvalues 0 and 0 and singular values 0 and 1

0 1
Take A = .
0 0

(d) (2 pts) A with rank r = 1 and determinant 1

Impossible. If the rank is 1, the matrix is singular and has determinant 0.

(e) (2 pts) A with complex eigenvalues and determinant 1

The determinant is the product of the eigenvalues, hence we need to nd complex
numbers whose product is 1. One possibility is i and i. We need to nd a real matrix
with i and i as eigenvalues. The following matrix works :

0 1
A=
1 0

21
 Bonus problem (From an earlier version of the exam)
A basis for the nullspace of the matrix A consists of the three vectors :

1 3 6

2 2 5


1 , 0 , 0



0 1 0

0 0 1
A basis for the column space of the matrix A consists of the two vectors :

1 2

1 , 1


1 0
(a) (2 pts.) How many rows and how many columns are in A?
If A is an m n matrix, then the nullspace N (A) is a subspace of Rn , while the column
space C(A) is a subspace of Rm . In our case, this tells us n = 5 and m = 3. So A is a
3 5 matrix; A has 3 rows and 5 columns.

(b) (6 pts.) Provide a basis for N (AT ) and C(AT ).

The left-nullspace of A, i.e., N (AT ), is the orthogonal complement of C(A) in R3 . Let's
take the cross product of the two vectors in the basis for C(A):

(1, 1, 1) (2, 1, 0) = (1, 2, 1).

The vector (1, 2, 1) is perpendicular to the plane C(A), so it's a basis for the line that
is orthogonal to this plane, i.e., the line N (AT ). But suppose you didn't think of using
the cross product; what could you have done instead? Well, N (AT ) is the space of
solutions (x1 , x2 , x3 ) to the equation

x1
1 1 1 0
= .
x2
2 1 0 0
x3

22
 Let's put the 2 3 matrix in reduced row-echelon form:

1 1 1 1 1 1 1 0 1
.
2 1 0 0 1 2 0 1 2

Now x3 is a free variable; set it to 1 and use back substitution to nd x1 and x2 .
You get exactly (x1 , x2 , x3 ) = (1, 2, 1) as above; this special solution forms a basis for
N (AT ).

Similarly, the row space of A, i.e., C(AT ), is the orthogonal complement of N (A), so
it consists of solutions (x1 , . . . , x5 ) to the equation

x
1
1 2 1 0 0 x2 0


3 2 0 1 0 x3 = 0 .



6 5 0 0 1 x4 0

x5

One could solve this by Gaussian elimination, but it's probably easiest to observe
that, if you reversed the order of the columns , the 3 5 matrix would already be in
reduced row-echelon form, so the special solutions can be found by setting (x1 , x2 ) to
(1, 0) or (0, 1), and solving for (x3 , x4 , x5 ). In this way, we get the special solutions
(1, 0, 1, 3, 6) and (0, 1, 2, 2, 5); these two vectors form a basis for C(AT ).

(c) (6 pts.) Write down an example of such a matrix A.

Thanks to our answer to (b), we know that A has the same column space as

1 0


0 1

1 2

(you could also just say

1 2

1 1


1 0

23
here, but our method makes the subsequent computations a bit easier). Also from our
answer to (b), we know that A has the same row space as

1 0 1 3 6
.
0 1 2 2 5

So A is any matrix of the form

1 0
1 0 1 3 6
B ,

0 1
0 1 2 2 5
1 2

where B is an invertible 2 2 matrix (if you don't see why, please ask). For example,
if we take B to be the identity, we get

1 0 1 3 6

A= 0 2 5 .

1 2

1 2 3 1 16

24
(d) If the vector b is the sum of the four columns of A, write down the complete solution to
Ax = b.

Answer:

1 2 3
1
+ x2 1 + x4 2

x=
1 0 0
1 0 1
2. (11 points) This problem finds the curve y = C + D 2 t which gives the best least squares fit
to the points (t, y) = (0, 6), (1, 4), (2, 0).

(a) Write down the 3 equations that would be satisfied if the curve went through all 3 points.

Answer:

C + 1D = 6
C + 2D = 4
C + 4D = 0

(b) Find the coefficients C and D of the best curve y = C + D2 t .

Answer:

" # 1 1 " #
T
1 1 1 3 7
A A= 1
=
2
1 2 4 7 21

1 4

" # 6 " #
1 1 1 10
AT b =
=
4
1 2 4 14

0
Solve AT Ax = AT b :
           
3 7 C 10 C 1 21 7 10 8
= gives = = .
7 21 D 14 D 14 7 3 14 2

(c) What values should y have at times t = 0, 1, 2 so that the best curve is y = 0?

Answer:
The projection is p = (0, 0, 0) if AT b = 0. In this case, b = values of y = c(2, 3, 1).
3. (11 points) Suppose Avi = bi for the vectors v1 , . . . , vn and b1 , . . . , bn in Rn . Put the vs into
the columns of V and put the bs into the columns of B.

(a) Write those equations Avi = bi in matrix form. What condition on which vectors allows A
to be determined uniquely? Assuming this condition, find A from V and B.

Answer:
A [v1 vn ] = [b1 bn ] or AV = B. Then A = BV 1 if the v 0 s are independent.

(b) Describe the column space of that matrix A in terms of the given vectors.

Answer:
The column space of A consists of all linear combinations of b1 , , bn .

(c) What additional condition on which vectors makes A an invertible matrix? Assuming this,
find A1 from V and B.

Answer:
If the b0 s are independent, then B is invertible and A1 = V B 1 .
4. (11 points)
(a) Suppose xk is the fraction of MIT students who prefer calculus to linear algebra at year k.
The remaining fraction yk = 1 xk prefers linear algebra.

At year k + 1, 1/5 of those who prefer calculus change their mind (possibly after taking
18.03). Also at year k + 1, 1/10 of those who prefer linear algebra change their mind
(possibly because of this exam).
" # " # " #
xk+1 xk k
1
Create the matrix A to give =A and find the limit of A as k .
yk+1 yk 0
Answer:
 
.8 .1
A= .
.2 .9
 
1/3
The eigenvector with = 1 is .
2/3
 
1
This is the steady state starting from .
0
2
of all students prefer linear algebra! I agree.
3

(b) Solve these differential equations, starting from x(0) = 1, y(0) = 0 :

dx dy
= 3x 4y = 2x 3y .
dt dt

Answer:
 
3 4
A= .
2 3

has eigenvalues 1 = 1 and 2 = 1 with eigenvectors x1 = (2, 1) and x2 = (1, 1).

The initial vector (x(0), y(0)) = (1, 0) is x1 x2 .
So the solution is (x(t), y(t)) = et (2, 1) + et (1, 1).
 " # " #
x(0) x(t)
(c) For what initial conditions does the solution to this differential equation
y(0) y(t)

lie on a single straight line in R 2 for all t?

Answer:
If the initial conditions are a multiple of either eigenvector (2, 1) or (1, 1), the solution
is at all times a multiple of that eigenvector.
5. (11 points)

(a) Consider a 120 rotation around the axis x = y = z. Show that the vector i = (1, 0, 0) is
rotated to the vector j = (0, 1, 0). (Similarly j is rotated to k = (0, 0, 1) and k is rotated to
i.) How is j i related to the vector (1, 1, 1) along the axis?

Answer:

1
ji= 1
0

1
is orthogonal to the axis vector 1 .
1
So are k j and i k. By symmetry the rotation takes i to j, j to k, k to i.

(b) Find the matrix A that produces this rotation (so Av is the rotation of v). Explain why
A3 = I. What are the eigenvalues of A?

Answer:

A3 = I because this is three 120 rotations (so 360 ). The eigenvalues satisfy 3 = 1
so = 1, e2i/3 , e2i/3 = e4i/3 .

(c) If a 3 by 3 matrix P projects every vector onto the plane x+2y+z = 0, find three eigenvalues
and three independent eigenvectors of P. No need to compute P .

Answer: The plane is perpendicular to the vector (1, 2, 1). This is an eigenvector of
P with = 0. The vectors (2, 1, 0) and (1, 1, 1) are eigenvectors with = 0.
 6. (11 points) This problem is about the matrix

1 2
A = 2 4 .
3 6

(a) Find the eigenvalues of AT A and also of AAT . For both matrices find a complete set of
orthonormal eigenvectors.

Answer:

 1 2  
T 1 2 3 14 28
A A= 2 4 =
2 4 6 28 56
3 6
   
1 1 1 2
has 1 = 70 and 2 = 0 with eigenvectors x1 = and x2 = .
5 2 5 1

1 2   5 10 15
T 1 2 3
AA = 2 4
= 10
20 30 has 1 = 70, 2 = 0, 3 = 0 with
2 4 6
3 6 15 30 45

1 2 3
1 1 and x3 = 1 6 .
x1 = 2 and x2 = 1
14 3 5 0 70 5

(b) If you apply the Gram-Schmidt process (orthonormalization) to the columns of this matrix A, what
is the resulting output?

Answer:
Gram-Schmidt will find the unit vector

1
1
q1 = 2 .
14 3

But the construction of q2 fails because column 2 = 2 (column 1).

(c) If A is any m by n matrix with m > n, tell me why AA T cannot be positive definite. Is AT A
always positive definite? (If not, what is the test on A?)

Answer

AAT is m by m but its rank is not greater than n (all columns of AAT are combinations of
columns of A). Since n < m, AAT is singular.
AT A is positive definite if A has full colum rank n. (Not always true, A can even be a zero
matrix.)
7. (11 points) This problem is to find the determinants of

1 1 1 1 0 1 1 1 x 1 1 1
1 1 1 1 1 1 1 1 1 1 1 1
A= 1 1 1 0 B= 1 1 1 0 C= 1 1

1 0
1 1 0 0 1 1 0 0 1 1 0 0

(a) Find det A and give a reason.

Answer:
det A = 0 because two rows are equal.

(b) Find the cofactor C11 and then find det B. This is the volume of what region in R 4 ?

Answer:
The cofactor C11 = 1. Then det B = det A C11 = 1. This is the volume of a box
in R4 with edges = rows of B.

(c) Find det C for any value of x. You could use linearity in row 1.

Answer:
det C = xC11 + det B = x + 1. Check this answer (zero), for x = 1 when C = A.
8. (11 points)

(a) When A is similar to B = M 1 AM, prove this statement:

If Ak 0 when k , then also B k 0.

Answer:
A and B have the same eigenvalues. If Ak 0 then all || < 1. Therefore B k 0.

(b) Suppose S is a fixed invertible 3 by 3 matrix.

This question is about all the matrices A that are diagonalized by S, so that
S 1 AS is diagonal. Show that these matrices A form a subspace of
3 by 3 matrix space. (Test the requirements for a subspace.)

Answer:
If A1 and A2 are in the space, they are diagonalized by S. Then S 1 (cA1 + dA2 )S is
diagonal + diagonal = diagonal.

(c) Give a basis for the space of 3 by 3 diagonal matrices. Find a basis for the space in part (b)
all the matrices A that are diagonalized by S.

Answer:
A basis for the diagonal matrices is

1 0 0
D1 = 0 D2 = 1 D3 = 0
0 0 1

Then SD1 S 1 , SD2 S 1 , SD3 S 1 are all diagonalized by S: a basis for the subspace.
(b) Compute the matrix AT A. What is its rank? What is its nullspace?

Answer:

3 1 1 1

1 3 1 1
AT A =
1 1 3 1
1 1 1 3
has rank 3 like A. The nullspace is the line through (1, 1, 1, 1).

(c) Suppose v1 = 1 and v4 = 0. If each edge contains a unit resistor, the currents (w 1 , w2 , w3 , w4 , w5 , w6 )
on the 6 edges will be w = Av by Ohms Law. Then Kirchhoffs Current Law (flow in =
flow out at every node) gives AT w = 0 which means AT Av = 0. Solve AT Av = 0 for the
unknown voltages v2 and v3 . Find all 6 currents w1 to w6 . How much current enters node
4?

Answer:

Note: As stated there is no solution (my apologies!). All solutions to AT Av = 0 are

multiples of (1, 1, 1, 1) which rules out v1 = 1 and v4 = 0.
Intended problem: I meant to solve the reduced equations using KCL only at nodes
2 and 3. In fact symmetry gives v2 = v3 = 12 . Then the currents are w1 = w2 = w5 =
1 1
w6 = 2
around the sides and w3 = 1 and w4 = 0 (symmetry). So w3 + w5 + w6 = 2
is
the total current into node 4.
18.06 Professor Edelman Final Exam December 22, 2011

Grading

3
Your PRINTED name is: 4

Please circle your recitation:

1 T 9 2-132 Kestutis Cesnavicius 2-089 2-1195 kestutis

2 T 10 2-132 Niels Moeller 2-588 3-4110 moller
3 T 10 2-146 Kestutis Cesnavicius 2-089 2-1195 kestutis
4 T 11 2-132 Niels Moeller 2-588 3-4110 moller
5 T 12 2-132 Yan Zhang 2-487 3-4083 yanzhang
6 T 1 2-132 Taedong Yun 2-342 3-7578 tedyun
1 (13 pts.)

Suppose the matrix A is the product

1 0 0 1 0 0 5

A= 2 1 0

0 1 4 0

3 1 1 0 0 0 0

(a) (3 pts.) What is the rank of A?

A has rank 2. (Since the rst matrix is non-singular, it does not aect

the rank.)

(b) (5 pts.) Give a basis for the nullspace of A.

0 5

4

0
, Columns 1 and 2 are pivot columns. The other two

1 0

0 1
are free. We assign 1,0 and 0,1 to the free variables.

1

(c) (5 pts.) For what values of t (if any) are there solutions to Ax = 1 ?


t

1 0 1 1 0 1

t = 2. Elimination on yields 1 .

2 1 1 0 1

3 1 t 0 0 t2

2
2 (12 pts.)

1 4 7

Let A = 2 5 8 .


3 6 9
(a) (3 pts.) Find a basis for the column space of A.

1 4

2 , 5 . This matrix is familar from class. The rst two


3 6
columns are pivot columns, the third is free.

(b) (3 pts.) Find a basis for the column space of where A = U V T is the svd of A.

is diagonal with rst two

diagonal
elements positive. Hence a basis

1 0

for the column space is 0 , 1 .


0 0

(c) (3 pts.) Find a basis for the column space of the matrix exponential eA

The matrix exponential has full rank, so the three columns of the

identity or any linearly independent set of three vectors will do.

d
(d) (3 pts.) Find a non-zero constant solution (meaning no dependence on t) to
dt
u(t) =
Au(t).

1

d
u(t) = 0 = Au = u(t) = 2 , the eigenvector corresonding

dt

1
to 0.

3
3 (12 pts.)

(a) ( 3 pts.) Give an example of a nondiagonalizable matrix A which satises det(tI A) =

(4 t)4

4 1


4 1


4 1

4
is a Jordan block hence is non-diagonalizable.

(b) (3 pts.) Give an example of two dierent matrices that are similar and both satisfy

det(tI A) = (1 t)(2 t)(3 t)(4 t).

1 4


2 3
and

3 2

4 1

(c) (3 pts.) Give an example if possible of two matrices that are not similar and both satisfy

det(tI A) = (1 t)(2 t)(3 t)(4 t).

All matrices with distinct eigenvalues 1,2,3,4 are similar, so this is

impossible.

(d) (3 pts) Give an example of two dierent 4x4 matrices that have singular values 1,2,3,4.

1 1

2

2
and
3

3

4 4

4
4 (16 pts.)

1 1 1 1

1 i 1

i
The matrix G= .
1 1 1 1


1 i 1 i
(a) (3 pts.) This matrix has two eigenvalues = 2, and one eigenvalue = 2. Given that,

nd the fourth eigenvalue.

The trace is 2 2i = 2 + 2 2+? so the fourth eigenvalue is 2i.

(b) (3 pts.) Find a real eigenvector and show that it is indeed an eigenvector.

1 2


0 0
=
G .
One can write down G 2I =
1 2

0 0

1 1 1 1

1 2 i 1

i
and notice that columns 1 and
1 1 1

1

1 i 1 2 i
columns 3 add to 0.

5

1 1 1 1

1 i 1

i
(Problem 4 continued.) The matrix G=
.
1 1 1 1


1 i 1 i

(c) (4 pts.) Is G a Hermitian matrix? Why or why not. (Remember Hermitian means that

Hjk = Hkj where the bar indicates complex conjugate.)

No, the diagonals are not real.

(d) (4 pts.) Give an example of a real non-diagonal matrix X for which GH XG is Hermitian.
Any real symmetrix non-diagonal
X will do, for example X ==
1 1 1 1


1 1 1 1
.

1 1 1 1

1 1 1 1

6
 5 (16 pts.)

The following operators apply to dierentiable functions f (x) transforming them to another
function g(x). For each one state clearly whether it is linear or not, (expalnations not

needed). (2 pts each problem)

d
(a) g(x) = dx
f (x) linear (for all linear cases check cf (x) goes to cg(x) and f1 (x) + f2 (x)
goes to g1 (x)+g2 (x)
d
(b) g(x) = dx
f (x) + 2 not linear (zero does not go to 0)
d
(c) g(x) = dx
f (2x) linear

(d) g(x) = f (x + 2) linear

(e) g(x) = f (x)2 not linear (the function cf (x) should go to cg(x) but it goes to c2 g(x).)

(f ) g(x) = f (x2 ) linear

(g) g(x) = 0 linear

(h) g(x) = f (x)+f (2) linear (don't be fooled, this one is indeed linear)

7
6 (20 pts.)

1 1 1 1

Let A = I3 cE3 = c 1 1 1 .

1

1 1 1 1
(a) (4 pts.) There are two values of c that make A a projection matrix. Find them by

guessing, calculating, or understanding projection matrices. Check that A is a projection

matrix for these two c.

A = A2 = I 2cE 2 2
+ 3c E = 3c = c so c = 0 or c = 1/3. Thus
2 1 1

1
A = I or A = 3 1 2 1 which upon squaring is itself.


1 1 2

(b) (4 pts.) There are two values of c that make A an orthogonal matrix. Find them and

check that A is orthogonal for these two c.

I = AT A = A2 = I 2c + 3c2 E = 3c2 = 2c so c = 0 or c = 2/3.
1 2 2

1
Thus A = I or A =
3 2 1 2 which upon squaring is the


2 2 1
identity.

(c) (4 pts.) For which values of c is A diagonalizable?

The matrix is symmetric, so all values of c make A diagonalizable.

8

1 1 1 1

(Problem 6 Continued) Let A = I3 cE3 = c .

1 1 1 1

1 1 1 1
(d) (4 pts.) Find the eigenvalues of A1 (if it exists) in terms of c. (Hint: nd the eigenvalues

of E3 rst.)

E3 is rank 1 and trace 3 so the eigenvalues are 3,0,0. Then A has

eigenvalues 1-3c,1,1. Finally A1 has eigenvalues

1
13c
, 1, 1.

(e) (4 pts.) For which values of c is A positive denite?

1
13c
>0 so c < 1/3.

9
7 (11 pts.)

The general equation of a circle in the plane has the form x2 + y 2 + Cx + Dy + E = 0.

Suppose you are trying to t n 3 distinct points (xi , yi ), i = 1, ..., n to obtain a best least

squares circle, it is reasonable to write a generally unsolvable equation Ax = b.

(a) (7 pts.) Describe A and b clearly, indicating the number of rows and columns of A and

the number of elements in

b.
x y 1 x21 y12
1 1 C
x22 y22

x2 y 2 1

.. .
D
. =
.
. The matrix A has n
. . .

. . .
.
E
xn y n 1 x2n yn2
rows and 3 columns, while b has n elements.

(b) (4 pts.) When n=3 it is possible to describe when the equation is and is not solvable.

You can use your geometric intuition, or a determinant area formula to describe when A is

singular. Give a simple geometrical description. (We are looking for a specic word  so

only a short answer will be accepted.)

A circle is determined by three points as long as they are not colinear.

The matrix A is the area matrix for a triangle, when n=3, so the

interpretation is that we can solve the equation, when the area of the

triangle is not-zero, i.e. the triangle does not collapse to a line.

10
18.06 Final Exam Professor Strang May 18, 2011

Your PRINTED name is 1.

Your Recitation Instructor (and time) is 2.
Instructors: (Hezari)(Pires)(Sheridan)(Yoo) 3.
4.
5.
6.
7.
8.
9.

Please show enough work so we can see your method and give due credit.

1. (a) For this matrix A, find the usual P (permutation) and L and U

so that P A = LU.
1 1 2 1

A=
2 2 4 2 .

3 4 7 3

(b) Find a basis for the nullspace of A.

(c) The vector (b1 , b2 , b3 ) is in the column space of A provided it is orthogonal to

(give a numerical answer).
2. (a) Compute the 4 by 4 matrix P that projects every vector in R4 onto the column space
of A:
1 1

1
1
A= .
1 1

1 1

(b) What are the four eigenvalues of P ? Explain your reasoning.

(c) Find a unit vector u (length 1) that is as far away as possible from the column space
of A.
3. Suppose A is an m by n matrix and its pivot columns (not free columns) are c1 , c2 , . . . , cr .
Put these columns into a matrix C.

(a) Every column of A is a of the columns of C. How would you produce

from this a matrix R so that A = CR? Explain how to construct R.

(b) Using C from part (a) factor the following matrix A into CR, where C has indepen-
dent columns and R has independent rows.

1 2 1

A= 2 4 2
.

1 2 5
4. (a) Find the cofactor matrix C for this matrix A. (The i, j entry of C is the cofactor
including sign of the i, j position in A.)

1 3 0

A= 0 1 4 .

0 0 1

(b) If a square matrix B is invertible, how do you know that its cofactor matrix is invert-
ible?

(c) True or false with a reason, if B is invertible with cofactor matrix C:

determinant of C
determinant of B 1 =
determinant of B
5. (a) Find the eigenvalues 1 , 2 , 3 and a full set of independent eigenvectors x1 , x2 , x3
(if possible) for
2 1 1

A=
1 2 1 .

1 1 2

(b) Suppose u0 = 3x1 + 7x2 + 5x3 is a combination of your eigenvectors of A:

kAk+1 u0 k
Find Ak u0 . If kvk is the length of v, find the limit of kAk u0 k
as k .
6. (a) For this directed graph, write down the 5 by 4 incidence matrix A. Describe the
nullspace of A.
2

Edge 1
2

3
Node 1 3

4 5

(b) Find the matrix G = AT A. Is this matrix G positive definite? (Explain why or why
not.) The first entry is G11 = 3 because the graph has .

(c) What is the sum of the squares of the singular values of A? Hint: Remember that
those numbers 2 are .
7. Suppose A is a positive definite symmetric matrix with n different eigenvalues: Axi = i xi .
(a) What are the properties of those s and xs? How would you find an orthogonal
matrix Q so that A = QQT with = diag (1 , . . . , n )?

(b) I am looking for a symmetric positive definite matrix B with B 2 = A (a square root
of A). What will be the eigenvectors and eigenvalues of B? Can you find a formula
for B using Q and from part (a)?

(c) What are the eigenvalues and eigenvectors of the matrix eA ? Is this matrix also
positive definite and why?
8. Suppose the 2 by 3 matrix A has Av1 = 3u1 and Av2 = 5u2 with orthonormal v1 , v2 in R3
and orthonormal u1 , u2 in R2 .

(a) Describe the nullspace of A.

(b) Find the eigenvalues of AT A.

(c) Find the eigenvalues and eigenvectors of AAT .

9. (a) The index of a matrix A is the dimension of its nullspace minus the dimension of
the nullspace of AT . If A is a 9 by 7 matrix of rank r, what is its index?

(b) Suppose M is the vector space consisting of all 2 by 2 matrices. (So those matrices
are the vectors in M.) Write down a basis for this vector space M: linearly
independent and spanning the space M.

" #
1 0
(c) S = is a specific matrix in M. For every 2 by 2 matrix A, the transformation
1 1
T produces T (A) = S 1 AS. Is this a linear transformation? What tests do you have
to check?
18.06 Professor Edelman Final Exam December 20, 2012

Grading

Your PRINTED name is: 3

6
Please circle your recitation:
7

1 T9 2-132 Andrey Grinshpun 2-349 3-7578 agrinshp

2 T 10 2-132 Rosalie Belanger-Rioux 2-331 3-5029 robr
3 T 10 2-146 Andrey Grinshpun 2-349 3-7578 agrinshp
4 T 11 2-132 Rosalie Belanger-Rioux 2-331 3-5029 robr
5 T 12 2-132 Georoy Horel 2-490 3-4094 ghorel
6 T1 2-132 Tiankai Liu 2-491 3-4091 tiankai
7 T2 2-132 Tiankai Liu 2-491 3-4091 tiankai
1 (10 pts.)

What condition on b makes the equation below solvable? Find the complete solution to x in
the case it is solvable.

x
1 3 1 2 1 1

x2

2 6 4 8

= 3

.


x3
0 0 2 4 b
x4

Solution:
Let's use Gaussian elimination. Starting from

1 3 1 2 1

x = 3 ,

2 6 4 8

0 0 2 4 b

1 0 0
multiply both sides by the elementary matrix on the left, which has the eect

2 1 0


0 0 1
of subtracting twice the rst row from the second:

1 3 1 2 1

x = 1 .

0 0 2 4

0 0 2 4 b

1 0 0
Then multiply both sides by the elementary matrix on the left, which has the


0 1 0

0 1 1
eect of subtracting the second row from the third:

1 3 1 2 1

x = .

0 0 2 4 1

0 0 0 0 b1

2
Comparing the third row on both sides, we nd that 0 = b 1. The rst and second rows
of the matrix on the left-hand side both have pivots, so there are no other restrictions. The
equation is solvable precisely when b = 1. Let us solve it in this case.
The free variables are x and x . To nd a
2 4 to the equation at hand, set
particular solution

both free variables to zero, and solve for the pivot variables; we get x = ( , 0, , 0). To nd
1
2
1
2

the complete solution, we must solve the homogeneous equation

1 3 1 2

x = 0.

0 0 2 4

0 0 0 0

The two special solutions to this homogeneous equation are found by setting one of the free
variables to 1, the other to 0: we get (3, 1, 0, 0) and (0, 0, 2, 1). Therefore, the complete
solution to the original equation when b = 1 is

1 1
2
3x2 2
3 0


x2 0 1 0
x= = + x2 + x4 .
1
2x4 12 2

2
0

x4 0 0 1

3
2 (6 pts.)

Let C be the cofactor matrix of an n n matrix A. Recall that C satises AC = (det A)I .
T
n

Write a formula for det C in terms of det A and n.

Solution:
Since AC = (det A)I , we get det(AC ) = det((det A)I ). The left hand side simplies to
T
n
T
n

det A det C and the right hand side is equal (det A) . This gives det C = (det A) .
n n1

The conscientious may object that we have divided both sides of the equation det A det C =
(det A) by det A, which is invalid if det A = 0. So we still have to prove that, if det A = 0,
n

then C must also be singular. Well, assume for the sake of contradiction that det A = 0
but C is invertible. Then C is also invertible, and we may multiply the original equation
T

AC = (det A)I by (C ) :
T
n
T 1

A = (det A)(C T )1 = 0(C T )1 = 0.

So A is the zero matrix, but in this case obviously so is its cofactor matrix C . This contra-
diction shows that indeed det C = (det A) in all cases.
n1

4
3 (25 pts.)

5 2 1
The matrix satises A

2
A= = 6A.

2 2 2

1 2 5

(a) (4 pts.) The eigenvalues of A are =______, =______, and =______.

1 2 3

(b) (5 pts) Find a basis for the nullspace of A and the column space of A.

5
(c) (16 pts.) Circle all that apply. The matrix M = A is
1
6

1.orthogonal 3. a projection 5. singular 7. a Fourier matrix

2. symmetric 4. a permutation 6. Markov 8. positive denite

6
4 (12 pts.)

1 1 1 1

The matrix 1 2 i 1
, where i = 1.

i
G=
1 1 1

1

1 i 1 2 i
(a) (6 pts) Use elimination or otherwise to nd the rank of G.

(b) (6 pts) Find a real nonzero solution to d

dt
x(t) = Gx(t).

7
5 (12 pts.)

Given a vector x in R , we can obtain a new vector y = cumsum(x), the cumulative sum,
n

by the following recipe:

y 1 = x1
yj = yj1 + xj , for j = 2, ..., n.
(a) (7 pts) What is the Jordan form of the matrix of this linear transformation?
Solution:
Let's rst nd the matrix A representing the linear transformation cumsum, and then worry
about nding the Jordan form of A. Note that cumsum maps (1, 0, . . . , 0) to (1, 1, . . . , 1),
so the rst column of A should be (1, 1, . . . , 1). The other columns of A can be found in a
similar way, and
1 0 0 0 0


. . . .. ..
1 1 0 0 0

.. .. .. . . .

1 1 1
A=

.

0 0



1 1 1 1 0

1 1 1 1 1

If n = 1, then A is already in Jordan form. So, for the rest of this solution, let's assume

n > 2.

To nd the Jordan form of A, let's rst nd its eigenvalues. We see that A is a lower
triangular matrix, so its eigenvalues are just its diagonal entries, which are all equal to 1.
Thus, 1 is the only eigenvalue of A, and it occurs with arithmetic multiplicity n. This fact
alone is not enough to determine the Jordan form of A, however. In fact, there are p(n)
non-similar n n matrices whose only eigenvalue is 1, where p(n) denotes the number of
partitions of n  the number of distinct ways of writing n as a sum of positive integers, if
order is irrelevant. Of the p(n) possible distinct Jordan forms of such matrices, which one is
actually the Jordan form of A?
8
One possibility that we can immediately eliminate is the identity matrix I . The only matrix
n

similar to I is I itself: M I M = I for any invertible matrix M . Since A isn't the identity
n n n
1
n

matrix, it isn't similar to I either, and its Jordan form is not I .

n n

The key is to consider A I : this matrix has rank n 1, because (for example) its transpose
n

is in row-echelon form with n 1 pivots. Thus, A I has a 1-dimensional kernel, which is

n

to say A has a 1-dimensional eigenspace for the eigenvalue 1. This means that the Jordan
form of A consists of a single Jordan block, which must therefore be

1 1

1 1

...



1
...



1


1 1

1

(the empty entries are zeroes).

(b) (5 pts) For every n,nd an eigenvector of cumsum.
Solution:
An eigenvector for A is the same thing as a vector in the nullspace of

0 0 0 0 0

1 0 0 0 0



.. .. . . . . . . .. ..

1 1 0 0 0
A In =

.



1 1

1 0 0

1 1 1 1 0

Row operations or mere inspection quickly leads to the conclusion that only the nth column
is free, while all other columns have pivots. So all eigenvectors for A are scalar multiples
of (0, . . . , 0, 1). It is easy to check that (0, . . . , 0, 1) is unchanged by the transformation
cumsum, so this makes sense.
9
6 (20 pts.)

This problem concerns matrices whose entries

are taken from the
values +1 and -1. In other
1 1 ... 1

words, the general form of this matrix is 1 1 ... 1

.. .. . . . We will call these matrices


.



1 1 ... 1
1 matrices.
1 1 1
One 3x3 example of such a matrix is 1

1 1 .


1 1 1

a) (5 pts.) Find a two by two example of a 1 matrix with eigenvalues 1 =0 and 2 =0

or prove it is impossible.

b) (5 pts.) Suppose A is a 10 10 example of a 1 matrix. Compute 2

1 + 22 + . . . + 10
2

10
c) (5 pts.) The big determinant formula for a 5 5 A has exactly ________ terms. A
computer package for matrices computes that the determinant of a 1 matrix that is 5 5
is an odd integer. If this is possible exhibit such a 1 matrix, if not argue clearly why the
package must not be giving the right answer for this 5 5 matrix.

d) (5 pts.) For every n, construct a 1matrix A with (n 1) eigenvalues exactly equal to

n

2. (Hint: Think about A 2I.)

n

11
7 (15 pts.)

Let V be the six dimensional vector space of functions f (x, y) of the form ax + bxy +
2

cy + dx + ey + f. Let W be the three dimensional vector space of (at most) second degree
2

quadratics in x.
a) (6 pts.) Write down a basis for V and a basis for W.
Solution:
A basis for V is 1, x, y, xy, x , y . A basis for W is 1, x, x .
2 2 2

b (9 pts.) In your chosen basis, what is the matrix of the linear transformation from V to
W that takes f (x, y) to g(x) = f (x, x)?

Recall that the ith column of the matrix simply describes the image of the ith basis vector
of V as a linear combination of the basis vectors of W . Therefore, the transformation is
represented by

1 0 0 0 0 0


0 1 1 0 0 0

0 0 0 1 1 1

12
18.06 Professor Strang Final Solutions May 23rd, 2012
Grading
1
2
Your PRINTED name is: 3
4
5
6
7
8

r01 T 11 4-159 Ailsa Keating ailsa

r02 T 11 36-153 Rune Haugseng haugseng
r03 T 12 4-159 Jennifer Park jmypark
r04 T 12 36-153 Rune Haugseng haugseng
r05 T 1 4-153 Dimiter Ostrev ostrev
r06 T 1 4-159 Uhi Rinn Suh ursuh
r07 T 1 66-144 Ailsa Keating ailsa
r08 T 2 66-144 Niels Martin Moller moller
r09 T 2 4-153 Dimiter Ostrev ostrev
r10 ESG Gabrielle Stoy gstoy
1 (12 pts.)
(a) - Find the eigenvalues and eigenvectors of A.

3 1 4
A = 0 1 5


0 1 5

Solution. The eigenvalues are:

= 0, 3, 6

The corresponding eigenvectors are:

1

=0: v1 = 15


3

1
=3: v2 = 0


0

5

=6: v3 = 3


3


1
(b) - Write the vector v = 1 as a linear combination of eigenvectors of A.


1
10
- Find the vector A v.

Solution. We have that, forming T = [ v1 | v2 | v3 ] (with columns = the three vectors),

1 0

y = T 1 1 = 2/3


1 1/3
Page 2 of 14
 Or in other words:

1 1 2 1 1 5
v = 1 = 0 15 0 + 3

3 3
1 3 0 3

Therefore, we also see:

1 5 100737594
2 1
()

A10 v = 310 0 + 610 3 = 60466176

3 3
0 3 60466176

(*) Required for mental arithmetics wizards only. 

du
(c) If you solve dt
= Au (notice the minus sign), with u(0) a given vector, then as t
the solution u(t) will always approach a multiple of a certain vector w.
- Find this steady-state vector w.

Solution. Since the eigenvalues of A are 0, 3, 6, we see that this steady state is:

1
w = v1 = 15


3

Page 3 of 14
2 (12 pts.)

Suppose A has rank 1, and B has rank 2 (A and B are both 3 3 matrices).

(a) - What are the possible ranks of A + B?

Solution. Of course, 0 rank(A + B) 3. But the only ranks that are possible are:

rank(A + B) = 1, 2, 3.

The reason 0 is not an option is: It implies A + B = 0, i.e. that A = B. But

rank(B) = rank(B), so for that to happen A and B should have had the same rank. 

(b) - Give an example of each possibility you had in (a).

Solution. Here are some simple examples:

Example w/ rank(A + B) = 1: Take e.g.

1 0 0 1 0 0
A = 0 0 0 , and B = 0 1 0


0 0 0 0 0 0

Example w/ rank(A + B) = 2: Take e.g.

1 0 0 1 0 0

A = 0 0 0 , and B = 0 1 0


0 0 0 0 0 0

Example w/ rank(A + B) = 3: Take e.g.

0 0 0 1 0 0

A = 0 0 0 , and B = 0 1 0


0 0 1 0 0 0

(c) - What are the possible ranks of AB?

- Give an example of each possibility.
Page 4 of 14
Solution. As a general rule, recall 0 rank(AB) min(rank(A), rank(B)) = 1. In this
case, both possibilities do happen:

rank(AB) = 0, 1.

Diagonal examples suffice:

Example w/ rank(AB) = 0: Take e.g.

0 0 0 1 0 0
A = 0 0 0 , and B = 0 1 0


0 0 1 0 0 0

Example w/ rank(AB) = 1: Take e.g.

1 0 0 1 0 0

A = 0 0 0 , and B = 0 1 0


0 0 0 0 0 0

Page 5 of 14
3 (12 pts.)
(a) - Find the three pivots and the determinant of A.

1 0 1

A= 0 1 1


1 1 0

Solution. We see that

1 0 1

A 0 1 1


0 0 2

Thus,

The pivots are 1, 1, 2

Since we reduced A without any row switches (permutation P s), or row scalings, we
have:
det A = 1 1 (2) = 2

(b) - The rank of A I is , so that = is an eigenvalue.

- The remaining two eigenvalues of A are = .

- These eigenvalues are all , because AT = A.

Solution. We see that

rank(A I) = 2

So dim N (A I) = 1. Thus,
=1

is an eigenvalue of algebraic and geometric multiplicity one.

The other two eigenvalues of A are:

= 1, 2.
Page 6 of 14
 The eigenvalues are all real values, because A is symmetric.

(c) The unit eigenvectors x1 , x2 , x3 will be orthonormal.

- Prove that:
A = 1 x1 xT1 + 2 x2 xT2 + 3 x3 xT3 .

You may compute the xi s and use numbers. Or, without numbers, you may show that
the right side has the correct eigenvectors x1 , x2 , x3 with eigenvalues 1 , 2 , 3 .

Solution. As suggested, we check that A does the correct thing on the basis {x1 , x2 , x2 }.

1 x1 xT1 + 2 x2 xT2 + 3 x3 xT3 xi = i (xTi xi )xi = i xi = Axi

Having checked this, then by linearity of matrix multiplication, the two expressions agree
always (and hence the matrices are identical).
For the record, the three vectors are:

1
1
x1 = 1

2
0

1
1
x2 = 1

6
2

1
1
x3 = 1

3
1


Page 7 of 14
4 (12 pts.)

This problem is about x + 2y + 2z = 0, which is the equation of a plane through 0 in R3 .

(a) - That plane is the nullspace of what matrix A?

A=

- Find an orthonormal basis for that nullspace (that plane).

Solution.
A = [1 2 2]

We could identify a basis of N (A) as usual, then apply Gram-Schmidt to make it an

orthonormal basis.
But if we can find two orthonormal vectors in N (A), we are done. Here, one can first
easily guess one vector in N (A):


0

v1 = 1 N (A)


1

Then anything of the form [a 1 1] will be orthogonal to v1 , and we pick the one that
is in the null space:


4
v2 = 1 N (A)


1
Then an orthonormal basis is:

0 4
1 1
q1 = 1 , q2 = 1

2 3 2
1 1

(b) That plane is the column space of many matrices B.

- Give two examples of B.
Page 8 of 14
 Solution. We can use the basis vectors from above as columns, and (independent) linear
combinations of them. Or filling in a zero column:

B1 = [v1 v2 ]

B2 = [v1 2v1 + v2 ]

B3 = [v1 v2 0]

Then c(Bi ) = N (A).

(c) - How would you compute the projection matrix P onto that plane? (A formula is
enough)
- What is the rank of P ?

Solution. It can be computed using a matrix B from above (if it has independent columns:
So B1 , B2 but not B3 here), via the usual formula:

P = B(B T B)1 B T

For a projection, c(P ) is always the subpace it projects on, in this case it is the
two-dimensional plane:
rank(P ) = dim c(P ) = 2

Page 9 of 14
5 (12 pts.)
Suppose v is any unit vector in R3 . This question is about the matrix H.

H = I 2vvT .

(a) - Multiply H times H to show that H 2 = I.

Solution.

H 2 = (I 2vvT )2 = I 2 + 4(vvT )2 4vvT = I + 4vvT 4vvT = I

(b) - Show that H passes the tests for being a symmetric matrix and an orthogonal matrix.

Solution. Transpose is linear, I T = I, and anything of the form AAT is symmetric:

(I 2vvT )T = I 2(vT )T vT = I 2vvT

For orthogonality, we use (a) and symmetry:

HH T = H 2 = I

(c) - What are the eigenvalues of H?

You have enough information to answer for any unit vector v, but you can choose one v
and compute the s.

Solution. Note first that (since kvk = 1):

Hv = v 2(vT v)v = v,

so that
= 1

is an eigenvalue (with a one-dimensional eigenspace spanned by v).

Let on the other hand u span{v} be any vector orthogonal to v. Then we have:

Hu = u 2(vT u)v = u,

so that
=1
Page 10 of 14
 is also an eigenvalue.


Since span{v} is two-dimensional, we have found all eigenvalues.


6 (12 pts.)
(a) - Find the closest straight line y = Ct + D to the 5 points:

(t, y) = (2, 0), (1, 0), (0, 1), (1, 1), (2, 1).

Solution. We insert all points into the equation:

2C + D = 0

C + D = 0

0+D =1

1+D =1

2C + D = 1.

Written as a matrix system:

2 1 0

1 1 0

C
Ax =
0 1 = 1 = b
D


1 1 1

2 1 1

We consider instead AT Ax = AT b. We compute:

10 0 3
AT A = , AT b =
0 5 3

and
1/10 0
(AT A)1 = .
0 1/5
Thus finally:
C 3/10
= x = (AT A)1 AT b = .
D 3/5
Page 11 of 14
 So, the closest line to the five points is:
3 3
y= t+ .
10 5


(b) - The word "closest" means that you minimized which quantity to find your line?

Solution. It means that the sum of squares deviation kAx bk2 was minimized.


(c) If AT A is invertible, what do you know about its eigenvalues and eigenvectors? (Technical
point: Assume that the eigenvalues are distinct no eigenvalues are repeated). Since
AT A is symmetric and x (AT Ax) = kAxk2 0 always, it is positive semi-definite. Since
N (AT A) = {0}, zero is not eigenvalue. Hence:
The eigenvalues of AT A are positive, if AT is invertible
By symmetry:
Eigenvectors belonging to different eigenvalues are orthogonal

Page 12 of 14
7 (12 pts.)
This symmetric Hadamard matrix has orthogonal columns:

1 1 1 1

1 1 1 1
H= , and H 2 = 4I.

1
1 1 1
1 1 1 1
(a) What is the determinant of H?

Solution. By row reduction, we get the pivots 1, 2, 2, 4, so:

det H = 16

(b) What are the eigenvalues of H? (Use H 2 = 4I and the trace of H).

Solution. By H 2 = 4I, the eigenvalues are all either 2. They sum up to trH = 0.
Hence:

Two eigenvalues must be +2, and two eigenvalues be 2

Note also that this shows det H = 16 as in (a) 

(c) What are the singular values of H?

The singular values of H are 2, 2, 2, 2

Page 13 of 14
8 (16 pts.)
In this TRUE/FALSE problem, you should circle your answer to each question.
(a) Suppose you have 101 vectors v1 , v2 , . . . , v101 R100 .
- Each vi is a combination of the other 100 vectors:

- Three of the vi s are in the same 2-dimensional plane:

(b) Suppose a matrix A has repeated eigenvalues 7, 7, 7, so det(A I) = (7 )3 .

- Then A certainly cannot be diagonalized (A = SS 1 ):

7 1 0
- The Jordan form of A must be J = 0 7 1:


0 0 7
(c) Suppose A and B are 3 5.

- Then rank(A + B) rank(A) + rank(B):

(d) Suppose A and B are 4 4.

- Then det(A + B) det(A) + det(B):

(e) Suppose u and v are orthonormal, and call the vector b = 3u + v. Take V to be the
line of all multiples of u + v.
- The orthogonal projection of b onto V is 2u + 2v:

Rx
(f) Consider the transformation T (x) = x
f (t)dt, for a fixed function f . The input is x,
the output is T (x).

- Then T is always a linear transformation:

Page 14 of 14
18.06 Professor Edelman Final Exam December 19, 2013
Grading
1
2
Your PRINTED name is: 3
4
5
Please circle your recitation: 6

1 T9 Dan Harris E17-401G 3-7775 dmh

2 T 10 Dan Harris E17-401G 3-7775 dmh
3 T 10 Tanya Khovanova E18-420 4-1459 tanya
4 T 11 Tanya Khovanova E18-420 4-1459 tanya
5 T 12 Saul Glasman E18-301H 3-4091 sglasman
6 T1 Alex Dubbs 32-G580 3-6770 dubbs
7 T2 Alex Dubbs 32-G580 3-6770 dubbs
This page intentionally blank.

2
1 (6 pts.)

Project b onto the column space of A:

1 1 2

(a) (3 pts.) A =
0 1

and b =
3

.


0 0 4

1 1 4

(b) (3 pts.) A = 1 1 and b = 4 .




0 1 6

3
2 (20 pts.)

z z z

The matrix A = z z+c zc .




z zc z+c
a) (4 pts) Under what conditions on z and c would A be positive semidefinite?

b) (4 pts) Under what conditions on z and c would A be Markov?

c) (4 pts) Under what conditions on z and c would the first column of A be a free column?

4
d) (4 pts) Under what conditions on z and c does A have rank r = 2.

e) (4 pts) Assuming A has rank 2, for which b in R3 does the equation Ax = b have a
solution?

5
3 (22 pts.)

3 2
a) (5 pts.) What are the two eigenvalues of A = ?
2 0

b) (5 pts.) What is , the ratio of the maximum and minimum values of kAxk over the unit
circle kxk = 1?

6
c) (6 pts.) What are

the

maximum and minimum value of v T x, for x over the unit circle
2 3
kxk = 1, when v = ? Same question when v = ? (Hint: the minimum is negative)
0 2

d) (6 pts.) When every point on the unit circle is multiplied by A, the result is an ellipse.
Find the dimensions of the tightest rectangle with sides parallel to the coordinate axes that
encloses the ellipse. (Hint: the previous maximum and minimum question is meant to be a
warm-up. Another hint: singular values are not so useful here.)

7
4 (23 pts.)

The 8 8 matrix
1 1 1 1 1 1 1 1

1 1 1 1 1 1 1 1





1 1 1 1 1 1 1 1


1 1 1 1 1 1

1 1
A=


1 1 1 1 1 1 1 1





1 1 1 1 1 1 1 1


1 1 1 1 11 1 1




1 1 1 1 1 1 1 1
is symmetric and satisfies A2 = 8I. The diagonal elements add to 0.

a) (3 pts.) What are the eigenvalues of A and the determinant of A?

b) (2 pts.) What is the condition number (A) = 1 (A)/8 (A)? (Hint 1 always.)

8
c) (3 pts.) Can A have any Jordan blocks of size greater than 1? Explain briefly.

d) (2 pts.) What is the rank of the matrix consisting of the first four columns of A?

e) (3 pts.) The subspace of R8 spanned by the first four columns of A is the

of the last four columns of A. Fill in the blank with the best two words
and explain briefly.

9
f) (5 pts.) Let P project onto the first column of A. P is an 8 8 matrix all of whose 64
entries are the same number. This number is .

g) (5 pts.) Projection onto the last seven columns of A (we are dropping only the first
column) gives an 8 8 projection matrix whose 8 diagonal entries are the same number.
This number is .

10
5 (15 pts.)

Consider the vector space of symmetric 2 2 matrices of the form

x z
S= .

z y
a) (6 pts.) Write down a basis for this vector space.

d 0
b) (6 pts.) If D is the diagonal matrix D = , is T (S) = DSD a linear transforma-
0 e
tion from this space to itself? If so, write down the matrix of this transformation in your
chosen basis.

a b
c) (3 pts.) If A = , we want to know if T (S) = ASA os necessarily a linear
c d
transformation from symmetric 2 2 matrices to symmetric 2 2matrices? If yes, explain
why. If not, explain why not.

11
6 (14 pts.)

Choose the best choice from must", might", "cant" (and explain briefly.)

a) (2 pts.) A 3x3 matrix M with rank r = 2 have a non-zero solution to

M x = 0.

b) (2 pts.) A 3x3 matrix M with rank r = 3 have a non-zero solution to

M x = 0 with x1 = 0.

c) (2 pts.) A 3x4 matrix M with rank r = 3 have a non-zero solution to

M x = 0 with x1 = 1.

d) (2 pts.) A permutation matrix be singular.

12
e) (2 pts.) A projection matrix be singular.

f) (2 pts.) M I be singular, if is an eigenvalue of M .

g) (2 pts.) We recall the number of columns of an incidence matrix for a graph is the number
of nodes and every row has one entry +1, one entry 1, and the remaining entries 0. Such
an incidence matrix have full column rank.

Thank you for taking linear algebra. We hope you enjoyed it. Linear algebra will serve you
well in the future. Have a happy holiday!

13
18.06 Final Exam Professor Strang May 22, 2013

Please PRINT your name 1.

2.
Please Circle Your Recitation 3.
r1 T 10 36-156 Russell Hewett r7 T 1 36-144 Vinoth Nandakumar 4.
r2 T 11 36-153 Russell Hewett r8 T 1 24-307 Aaron Potechin 5.
r3 T 11 24-407 John Lesieutre r9 T 2 24-307 Aaron Potechin
6.
r4 T 12 36-153 Stephen Curran r10 T 2 36-144 Vinoth Nandakumar
r5 T 12 24-407 John Lesieutre r11 T 3 36-144 Jennifer Park 7.
r6 T 1 36-153 Stephen Curran 8.
9.

(1) (7+7 pts)

(a) Suppose the nullspace of a square matrix A is spanned by the vector v = (4, 2, 2, 0).
Find the reduced echelon form R = rref (A).

(b) Suppose S and T are subspaces of R5 and Y and Z are subspaces of R3 . When can
they be the four fundamental subspaces of a 3 by 5 matrix B? Find any required
conditions to have S = C(B T ), T = N (B), Y = C(B), and Z = N (B T ).

1
(2) (6+6 pts.)

(a) Find bases for all four fundamental subspaces of this R.

1 2 0 4

R=
0 0 1 3

0 0 0 0

(b) Find U, , V in the Singular Value Decomposition A = U V T :

2 2

A=
-1 2
.
2 -1
(3) (5+5 pts.)
Suppose q1 , . . . q5 are orthonormal vectors in R5 .
The 5 by 3 matrix A has columns q1 , q2 , q3 .

b to Ax = b.
(a) If b = q1 + 2q2 + 3q3 + 4q4 + 5q5 , find the best least squares solution x

(b) If terms of q1 , q2 , q3 find the projection matrix P onto the column space of A.
(4) (3+4+3 pts.)
The matrix A is symmetric and also orthogonal.

(a) How is A1 related to A?

(b) What number(s) can be eigenvalues of A and why?

(c) Here is an example of A. What are the eigenvalues of this matrix? I dont recommend
computing with det(A I) ! Find a way to use part (b).

.5 -.5 -.1 -.7

-.5 .5 -.1 -.7

A= .
-.1 -.1 .98 -.14

-.7 -.7 -.14 .02
(5) (4+5+3 pts.)
Suppose the real column vectors q1 and q2 and q3 are orthonormal.

(a) Show that the matrix A = q1 q1T + 2q2 q2T + 5q3 q3T has the eigenvalues = 1, 2, 5.

(b) Solve the differential equation du/dt = Au starting at any vector u(0). Your answer
can involve the matrix Q with columns q1, q2, q3, .

(c) Solve the differential equation du/dt = Au starting from u(0) = q1 q3 .

(6) (4+3+3 pts.)
This graph has m = 12 edges and n = 9 nodes. Its 12 by 9 incidence matrix A has a
single -1 and +1 in every row, to show the start and end nodes of the corresponding
edge in the graph.

(a) Write down the 4 by 4 submatrix SofA that comes from the 4-node graph (a loop)
in the corner. Find a vector x in the nullspace N(S) and a vector y in N(ST).

(b) For the whole matrix A, find a vector Yin N(AT). You won't need to write A or to
know more edge numbers.

(c) The all-ones vector (1, 1, ... , 1) spans N(A) . Find the dimension of the left nullspace
N(AT) (give a number).
(7) (3+3+3+3 pts.)
The equation yn+2 + Byn+1 + Cyn = 0 has the solution yn = n if 2 + B + C = 0. In
most cases this will give two roots 1 , 2 and the complete solution is yn = c1 n1 + c2 n2 .

Now solve the same problem the matrix way (slower). Create this vector unknown and
vector equation un+1 = A un .

" # " # " #" #

yn yn+1 yn
un = and = .
yn+1 yn+2 yn+1

(a) What is the matrix A in that equation?

(b) What equation gives the eigenvalues 1 and 2 of A?

(c) If 1 is an eigenvalue, show directly that

" # " #
1 1
A = 1 so we have the eigenvector.
1 1

(d) If 1 6= 2 , what is now the complete solution un (including constants c1 and c2 ) to

our equation un+1 = A un ? ((Then yn is the first component of un .))
(8) (5+5 pts.)

(a) Find the determinant of this matrix A, using the cofactors of row 1.

1 b 0 0

b 1 b 0

A= .
0 b 1 b

0 0 b 1

(b) Find the determinant of A by the BIG formula with 24 terms. This means to find
all the nonzero terms in that formula with their correct signs.
(9) (6+4 pts.)

(a) Suppose v = (v1 , v2 , v3 ) is a column vector, so A = vvT is a symmetric matrix.

Show that A is positive semidefinite, using one of these tests:

1. The eigenvalue test

2. The determinant test
3. The energy test on xT Ax.

(b) Suppose A is m by n of rank r. What conditions on m and n and r guarantee

that AT A is positive definite? If those conditions fail, prove that AT A will not be
positive definite.
 18.06 Final Exam Professor Strang May 21, 2014

Your PRINTED Name is:

Please circle your section:

Grading 1:

2:
R01 T 10 36-144 Qiang Guang
R02 T 10 35-310 Adrian Vladu 3:
R03 T 11 36-144 Qiang Guang
R04 T 11 4-149 Goncalo Tabuada
R05 T 11 E17-136 Oren Mangoubi
4:
R06 T 12 36-144 Benjamin Iriarte Giraldo
R07 T 12 4-149 Goncalo Tabuada 5:
R08 T 12 36-112 Adrian Vladu
R09 T 1 36-144 Jui-En (Ryan) Chang 6:
R10 T 1 36-153 Benjamin Iriarte Giraldo
R11 T 1 36-155 Tanya Khovanova
7:
R12 T 2 36-144 Jui-En (Ryan) Chang
R13 T 2 36-155 Tanya Khovanova
R14 T 3 36-144 Xuwen Zhu 8:
ESG T 3 Gabrielle Stoy

Thank you for taking 18.06. I hope you have a great summer. You could
look at 18.085 (Computational Science and Engineering) which starts with
applied linear algebra.
This exam has 20 parts, worth 5 points each.
For each problem, explain your answer as much as you can.
 1. (15 points)
(a) If A is a 3 by 4 matrix, what does this tell us about its nullspace?
(b) If we also know that
1
Ax = 1
1
has no solution, what do we know about the rank of A?
(c) If Ax = b and AT y = 0, find y T b by using those equations. This says
that the space of A and the are .

2
 2. (15 points) Suppose Ax = b reduces by the usual row operations to
Ux = c :

x1
2 6 4 8 b1
x2
x3 =
Ux = 0 0 4 4 b2 b1 = c.
0 0 0 0 b3 2b2 + b1
x4

(a) Give a basis for the nullspace of A (that matrix is not shown) and a
basis for the row space of A.
(b) When does Ax = b have a solution? Give a basis for the column space
of A.
(c) Give a basis for the nullspace of AT .

3
 3. (10 points)
(a) Suppose q1 , q2 are orthonormal in R4 , and v is NOT a combination of
q1 and q2 . Find a vector q3 by Gram-Schmidt, so that q1 , q2 , q3 is an
orthonormal basis for the space spanned by q1 , q2 , v.
(b) If p is the projection of b onto the subspace spanned by q1 and q2 and v,
find p as a combination of q1 , q2 , q3 . (You are solving the least squares
problem Ax = b with A = [q1 , q2 , q3 ].)

4
 4. (10 points)
(a) To solve a square system Ax = b when det A 6= 0, Cramers Rule says
that the first component of x is

det B
x1 = with B = [b a2 . . . an ] .
det A
So b goes into the first column of A, replacing a1 . If b = a1 , this formula
det A
gives the right answer x1 = det A = 1.

1. If b = a different column aj , show that this formula gives the right

answer, x1 = .
2. If b is any combination x1 a1 + + xn an , why does this formula give
the right answer x1 ?

(b) Find the determinant of

1 2 0 0
3 4 0 0
A=
5
.
0 6 7
0 0 8 9

5
Scrap Paper

6
 5. (15 points)
(a) Suppose an n by n matrix A has n independent eigenvectors x1 , . . . , xn
with eigenvalues 1 , . . . , n . What matrix equation would you solve for
c1 , . . . , cn to write the vector u0 as a combination u0 = c1 x1 + + cn xn ?
(b) Suppose a sequence of vectors u0 , u1 , u2 , . . . starts from u0 and satisfies
uk+1 = Auk . Find the vector uk as a combination of x1 , . . . , xn .
(c) State the exact requirement on the eigenvalues so that Ak u0 0 as
k for every vector u0 . Prove that your condition must hold.

7
 6. (10 points)
(a) Find the eigenvalues of this matrix A (the numbers in each column add
to zero).
1

1 2 0
A = 1 1 1 .
1
0 2 1

du
(b) If you solve dt = Au, is (1) or (2) or (3) true as t ?
(1) u(t) goes to zero?
(2) u(t) approaches a multiple of (what vector?)
(3) u(t) blows up?

8
 7. (10 points) Every invertible matrix A equals an orthogonal matrix Q
times a positive definite matrix S. This famous fact comes directly from
the SVD for the square matrix A = U V T , by choosing Q = U V T .
(a) How can you prove that Q = U V T is orthogonal?
(b) Substitute Q1 and A to write S = Q1 A in terms of U, V and . How
can you tell that this matrix S is symmetric positive definite?

9
8. (15 points) A 4-node graph has all six possible edges. Its incidence
matrix A and its Laplacian matrix AT A are

1 1 0

0


1 0 1 0
3 1 1 1
1 0 0 1 1 3 1 1
AT A =

A=

0 1 1 0

1 1 3 1
0 1 0 1 1 1 1 3
0 0 1 1

(a) Describe the nullspace of A.

(b) The all-ones matrix B = ones (4) has what eigenvalues? Then what are
the eigenvalues of AT A = 4I B?
(c) For the Singular Value Decomposition A = U V T , can you find the
nonzero entries in the diagonal matrix and one column of the orthog-
onal matrix V ?

10
Scrap Paper

11
 1. (15 points)

(a) If A is a 3 by 4 matrix, what does this tell us about its nullspace?

Solution: dim N (A) 1, since rank(A) 3.

(b) If we also know that

1
Ax = 1
1

has no solution, what do we know about the rank of A?

Solution: C(A) does not span the entire R3 , so rank(A) 2.

(c) If Ax = b and AT y = 0, find y T b by using those equations. This says that the
space of A and the are .
Solution: y T b = y T (Ax) = (AT y)x = 0 . This says that the column space of A and
the null space of AT are orthogonal.
 2. (15 points) Suppose Ax = b reduces by the usual row operations to U x = c :

x1
2 6 4 8 b1
x2

x 3 = b2 b1
Ux = 0 0 4 4 = c.
0 0 0 0 b3 2b2 + b1
x4

(a) Give a basis for the nullspace of A (that matrix is not shown) and a basis for the row
space of A.
Solution: N (A) = N (U ), R(A) = R(U ). Therefore we can read the bases directly from
U:


3 2

1 0

N (A) = span ,


0 1

0 1

2 0

6 0

R(A) = span ,
4 4


8 4

(b) When does Ax = b have a solution? Give a basis for the column space of A.
Solution: b C(A), equivalent to c C(U ).

1 0 0
By looking at c as a function of b we can reconstruct A. Let E = 1 1 0. We
1 2 1
1 0 0 2 6 4 8 2 6 4 8
have U = EA, c = Eb. Hence A = E 1 U = 1 1 0 0 0 4 4 = 2 6
8 12
1 2 1 0 0 0 0 2 6 12 16
From here we see that

1 1
C(A) = span 1 , 2 .
1 3

(c) Give a basis for the nullspace of AT .

1
T T
Solution: N (A ) C(A). So N (A ) = span 2 (we only need to find a vector
1

orthogonal to both basis vectors we gave for C(A)).

2
 3. (10 points)

(a) Suppose q1 , q2 are orthonormal in R4 , and v is NOT a combination of q1 and q2 . Find

a vector q3 by Gram-Schmidt, so that q1 , q2 , q3 is an orthonormal basis for the space
spanned by q1 , q2 , v.
Solution:
v (v T q1 )q1 (v T q2 )q2
q3 =
kv (v T q1 )q1 (v T q2 )q2 k

(b) If p is the projection of b onto the subspace spanned by q1 and q2 and v, find p as
a combination of q1 , q2 , q3 . (You are solving the least squares problem Ax = b with
A = [q1 , q2 , q3 ].)
Solution:
p = A(AT A)1 AT b = AAT b = q1 (q1T b) + q2 (q2T b) + q3 (q3T b)

where we used the fact that the columns of A are orthonormal for AT A = I. (It is
easy to see the final result just by thinking that we are actually projecting onto an
orthonormal basis).

3
 4. (10 points)

(a) To solve a square system Ax = b when det A 6= 0, Cramers Rule says that the first
component of x is

det B
x1 = with B = [b a2 . . . an ] .
det A
So b goes into the first column of A, replacing a1 . If b = a1 , this formula gives the right
answer x1 = det A
det A
= 1.

1. If b = a different column aj , show that this formula gives the right answer, x1 =
.
2. If b is any combination x1 a1 + + xn an , why does this formula give the right
answer x1 ?

Solution:

1. x1 = 0, since det(B) = 0. Let us check that it gives the correct answer: the
solution to Ax = aj is x = ej (it is unique, since det(A) 6= 0, so As columns are
linearly independent). Hence x1 = 0.
2. For the same reason as above the first component of x is indeed x1 , the coefficient
of a1 .
Now let us see what Cramers Rule gives. In this case det(B) = det([b a2 . . . an ]) =
det(x1 a1 a2 . . . an ) = x1 det(A). So, indeed, x1 = det(B)/ det(A).

(b) Find the determinant of

1 2 0 0
3 4 0 0
A=
5
.
0 6 7
0 0 8 9

Cofactor expansion by the first row:

4 0 0
C11 = det 0 6 7 = 4 6 9 4 8 7 = 8
0 8 9

3 0 0
C12 = det 5 6 7 = (3 6 9 3 7 8) = 6
0 8 9

So det(A) = 1 (8) + 2 6 = 4

4
 5. (15 points)

(a) Suppose an n by n matrix A has n independent eigenvectors x1 , . . . , xn with eigenvalues

1 , . . . , n . What matrix equation would you solve for c1 , . . . , cn to write the vector u0
as a combination u0 = c1 x1 + + cn xn ?
Solution:
c1
..
[x1 x2 . . . xn ] . = u0
cn

(b) Suppose a sequence of vectors u0 , u1 , u2 , . . . starts from u0 and satisfies uk+1 = Auk .
Find the vector uk as a combination of x1 , . . . , xn .
Solution: Let u0 = c1 x1 + + cn xn . Then uk = Ak u0 = ni=1 (ki ci )xi .
P

(c) State the exact requirement on the eigenvalues so that Ak u0 0 as k for every
vector u0 . Prove that your condition must hold.
Solution: |i | < 1, for all i. Clearly, if this holds, all the coefficients of xi in Ak u0 go
to 0 as k .
For the converse, we require that the coefficient ki ci of xi in Ak u0 to go to 0, for any
choice of u0 . Equivalently, we need ki ci 0 for any ci . Hence |i | < 1.

5
 6. (10 points)

(a) Find the eigenvalues of this matrix A (the numbers in each column add to zero).

1

1 2
0
A = 1 1 1 .
1
0 2
1

Solution: The number in each column add to zero, hence 1 N (AT ), so dim N (AT ) >
0, and thus rank(A) = rank(AT ) < 3. So 1 = 0.
We can easily spot 2 = 1 as another eigenvalue, since subtracting A + I has two
equal columns, and hence det(A + I) = 0.
Looking at the trace we get that 3 = tr(A) 1 2 = 2.
du
(b) If you solve dt
= Au, is (1) or (2) or (3) true as t ?

(1) u(t) goes to zero?

(2) u(t) approaches a multiple of (what vector?)
(3) u(t) blows up?

1
Solution: Approaches a multiple of 2.

1

1
Observe that 2 is an eigenvector for the 0 eigenvalue, and that the only nonzero

1
eigenvalue of eAt as t is e0t = 1, with the same eigenvector.
Also, u(t) = eAt u(0), and that the only nonzero eigenvalue of eAt (as t ) is 1, with
the
same eigenvector. So limt u(t) is a projection of u(0) on the line spanned by
1
2, hence a multiple of it.
1

6
 7. (10 points) Every invertible matrix A equals an orthogonal matrix Q times a positive
definite matrix S. This famous fact comes directly from the SVD for the square matrix
A = U V T , by choosing Q = U V T .

(a) How can you prove that Q = U V T is orthogonal?

Solution: Q is orthogonal if and only if QT Q = I. Notice that since A is invertible, U
and V are both square matrices of full rank.

QT Q = (U V T )T (U V T ) = V U T U V T = V (U T U )V T = V V T = I
We used the fact that U is orthogonal, hence U T U = I, and that V T is orthogonal
because V s columns are eigenvectors of a symmetric matrix AT A, so V T = V 1 .

(b) Substitute Q1 and A to write S = Q1 A in terms of U, V and . How can you tell
that this matrix S is symmetric positive definite?
Solution:

S = Q1 A = (U V T )1 A = (V T )1 U 1 U V T = V V T = (1/2 V T )T (1/2 V T )

7
 8. (15 points) A 4-node graph has all six possible edges. Its incidence matrix A and its
Laplacian matrix AT A are

1 1 0 0

1 0 1 0
3 1 1 1
1 0 0 1 1 3 1 1
A= AT A =

0 1 1 0

1 1 3 1
0 1 0 1 1 1 1 3
0 0 1 1

(a) Describe the nullspace of A.

1

1

Solution: N (A) = span 1 (Recall that applying A to a vector of potentials gives


1

the potential drops along edges, so in order for a vector of potentials to be in the null
space, all the potentials within one connected component must be the same.)

(b) The all-ones matrix B = ones (4) has what eigenvalues? Then what are the eigenvalues
of AT A = 4I B?
Solution: B = ~1 ~1T = 4(~1/2)(~1/2)T , where ~1 is the all-ones vector in R4 . So B has
eigenvalues 4, 0, 0, 0.
I and B diagonalize in the same eigenbasis, so i (4I B) = i (4I) i (B) = 4i (I)
i (B) for all i. So the eigenvalues of AT A are 0, 4, 4, 4.

(c) For the Singular Value Decomposition A = U V T , can you find the nonzero entries in
the diagonal matrix and one column of the orthogonal matrix V ?
p
Solution: i = i (AT A), so the nonzero singular values are 2, 2, 2. We only need to
find one eigenvector of AT A. An obvious one is ~1/2, since all the rows sum up to 0.

8
 18.06 Final Exam Professor Strang May 18, 2015

Your PRINTED Name is:

Please CIRCLE your section:

Grading 1:

2:

3:
R01 T10 26-302 Dmitry Vaintrob
R02 T10 26-322 Francesco Lin
R03 T11 26-302 Dmitry Vaintrob 4:
R04 T11 26-322 Francesco Lin
R05 T11 26-328 Laszlo Lovasz
R06 T12 36-144 Michael Andrews
R07 T12 26-302 Netanel Blaier 5:
R08 T12 26-328 Laszlo Lovasz
R09 T1pm 26-302 Sungyoon Kim
R10 T1pm 36-144 Tanya Khovanova 6:
R11 T1pm 26-322 Jay Shah
R12 T2pm 36-144 Tanya Khovanova
R13 T2pm 26-322 Jay Shah
R14 T3pm 26-322 Carlos Sauer 7:
ESG Gabrielle Stoy

8:

Thank you for taking 18.06! I hope you have a wonderful summer!
EACH PART OF EACH QUESTION IS 5 POINTS.

1. (a) Find the reduced row echelon form R = rref(A) for this matrix A :

1 0 1 2
A=0 0 1 2 .
0 0 1 2

(b) Find a basis for the column space C(A).

(c) Find all solutions (and first tell me the conditions on b1 , b2 , b3 for
solutions to exist!).
b1
Ax = b2 .
b3

2
2. (a) What is the 3 by 3 projection matrix Pa onto the line through
a = (2, 1, 2)?
(b) Suppose Pv is the 3 by 3 projection matrix onto the line through
v = (1, 1, 1). Find a basis for the column space of the matrix A =
Pa Pv (product of 2 projections)

3
3. Suppose I give you an orthonormal basis q1 , . . . , q4 for R4 and an or-
thonormal basis z1 , . . . , z6 for R6 . From these you create the 6 by 4
matrix A = z1 q1T + z2 q2T .

(a) Find a basis for the nullspace of A.

(b) Find a particular solution to Ax = z1 and find the complete solution.
(c) Find AT A and find an eigenvector of AT A with = 1.

4
4. Symmetric positive definite matrices H and orthogonal matrices Q are
the most important. Here is a great theorem: Every square invertible
matrix A can be factored into A = HQ.

(a) Start from A = U V T (the SVD) and choose Q = U V T . Find the

other factor H so that U V T = HQ. Why is your H symmetric and
why is it positive definite?
(b) Factor this 2 by 2 matrix into A = U V T and then into A = HQ :
 
1 3
A= = U V T = HQ
1 3

5
5. (a) Are the vectors (0, 1, 1), (1, 0, 1), (1, 1, 0) independent or dependent?
(b) Suppose T is a linear transformation with input space = output
space = R3 . We have a basis u, v, w for R3 and we know that T (u) =
v + w, T (v) = u + w, T (w) = u + v. Describe the transformation T 2
by finding T 2 (u) and T 2 (v) and T 2 (w).

6
6. Suppose A is a 3 by 3 matrix with eigenvalues = 0, 1, 1 and corre-
sponding eigenvectors x1 , x2 , x3 .

(a) What is the rank of A? Describe all vectors in its column space
C(A).
(b) How would you solve du/dt = Au with u(0) = (1, 1, 1)?
(c) What are the eigenvalues and determinant of eA ?

7
7. (a) Find a 2 by 2 matrix such that
       
1 3 2 3
A = and also A =
2 4 1 4
or say why such a matrix cant exist.
(b) The columns of this matrix H are orthogonal but not orthonormal:

1 1 1 1
1 1 1 1
H=
0 2 1

1
0 0 3 1
Find H 1 by the following procedure. First multiply H by a diagonal
matrix D that makes the columns orthonormal. Then invert. Then
account for the diagonal matrix D to find the 16 entries of H 1 .

8
8. (a) Factor this symmetric matrix into A = U T U where U is upper
triangular:

1 1 1
A = 1 2 2 .
1 2 3
(b) Show by two different tests that A is symmetric positive definite.
(c) Find and explain an upper bound on the eigenvalues of A. Find and
explain a (positive) lower bound on those eigenvalues if you know
that
2 1 0
A1 = 1 2 1 .
0 1 1

9
Scrap Paper

10
 18.06 Final Exam Professor Strang May 18, 2015
SOLUTIONS

Your PRINTED Name is:

Please CIRCLE your section:

Grading 1:

2:

3:
R01 T10 26-302 Dmitry Vaintrob
R02 T10 26-322 Francesco Lin
R03 T11 26-302 Dmitry Vaintrob 4:
R04 T11 26-322 Francesco Lin
R05 T11 26-328 Laszlo Lovasz
R06 T12 36-144 Michael Andrews
R07 T12 26-302 Netanel Blaier 5:
R08 T12 26-328 Laszlo Lovasz
R09 T1pm 26-302 Sungyoon Kim
R10 T1pm 36-144 Tanya Khovanova 6:
R11 T1pm 26-322 Jay Shah
R12 T2pm 36-144 Tanya Khovanova
R13 T2pm 26-322 Jay Shah
R14 T3pm 26-322 Carlos Sauer 7:
ESG Gabrielle Stoy

8:

Thank you for taking 18.06! I hope you have a wonderful summer!
EACH PART OF EACH QUESTION IS 5 POINTS.

1. (a) Find the reduced row echelon form R = rref(A) for this matrix A :

1 0 1 2
A=0 0 1 2 .
0 0 1 2

Solution. We have

1 0 1 2 1 0 1 2 1 0 0 0
0 0 1 2 0 0 1 2 0 0 1 2
0 0 1 2 0 0 0 0 0 0 0 0

The last matrix is the RREF.

(b) Find a basis for the column space C(A).
Solution. We can see that the pivot columns are columns 1 and 3,
so these columns from the original matrix form a basis,

1 1
0 , 1
0 1

(c) Find all solutions (and first tell me the conditions on b1 , b2 , b3 for
solutions to exist!).
b1
Ax = b2 .
b3

Solution. We can see that we need b2 = b3 . First, let us find a

particular solution. Since x2 , x4 are free variables, we can set them
to 0, and then we can solve to get

b1 b2
0
b2 .

0
Now, we need a basis for the nullspace, the special solutions. Setting
each free variable to 1 and the other to 0, we obtain the special

2
solutions
0 0
1 0
,
0 2
0 1
So, the general solutions are given by vectors

b1 b2 0 0
0
+ c1 1 + c2 0


b2 0 2
0 0 1

3
2. (a) What is the 3 by 3 projection matrix Pa onto the line through
a = (2, 1, 2)?
Solution.
2

1 2 1 2
2 4 2 4
1
Pa = = 2 1 2
2 9

4 2 4
2 1 2 1
2
(b) Suppose Pv is the 3 by 3 projection matrix onto the line through
v = (1, 1, 1). Find a basis for the column space of the matrix A =
Pa Pv (product of 2 projections)
Solution. Pa Pv v = Pa v = 59 a and so a C(Pa Pv ) C(Pa ). Since
C(Pa ) is spanned by a, a basis for C(Pa Pv ) is given by {a}.

4
3. Suppose I give you an orthonormal basis q1 , . . . , q4 for R4 and an or-
thonormal basis z1 , . . . , z6 for R6 . From these you create the 6 by 4
matrix A = z1 q1T + z2 q2T .

(a) Find a basis for the nullspace of A.

Solution. The matrix has SVD ZJQT where J is the 6 by 4 ma-
trix with diagonal entries (1, 1, 0, 0). This means that its nullspace
consists of the qs in columns of Q corresponding to zero singular
values, which is q3 , q4 .
(b) Find a particular solution to Ax = z1 and find the complete solution.
Solution. One particular solution to Ax1 is q1 , since (z1 q1T )q1 =
z1 (q1T z1 ) = z1 (q1 q1 ) = z1 , by and (z2 q2T )q1 = z2 (q2T q1 ) = z1 (q2 q1 ) = 0
by orthonormality of qi . The complete solution is obtained by adding
an element of the nullspace, i.e. a linear combination of basis vectors
of the nullspace: q1 + cq2 + dq4 for scalars c, d.
(c) Find AT A and find an eigenvector of AT A with = 1.
Solution. AT A = (q1 z1T + q2 z2T )(z1 q1T + z2 q2T ). Expanding and
reparenthezising gives AT A = q1 (z1T z1 )q1T +q1 (z1T z2 )q2T +q2 (z2 z1T )q2T +
q2 (z2 z2T )q2T . In every term, the parenthesized scalar in the middle is
a dot product: z1 z2 = 0 for the middle two terms and 1 for the
first and fourth terms, leaving AT A = q1 q1T + q2 q2T . We see that
AT Aq1 = q1 (q1 q1 ) + q2 (q2 q1 ) = q1 and, for the same reason,
AT Aq2 = q2 . So q1 and q2 (or any nonzero linear combination) are
all eigenvectors with eigenvalue 1.

5
4. Symmetric positive definite matrices H and orthogonal matrices Q are
the most important. Here is a great theorem: Every square invertible
matrix A can be factored into A = HQ.
(a) Start from A = U V T (the SVD) and choose Q = U V T . Find the
other factor H so that U V T = HQ. Why is your H symmetric and
why is it positive definite?
Solution. By definition we need U V T = A = HQ = HU V T so we
get by inverting U and V T (which are orthogonal hence invertible)
that H = U U 1 . The last item can also be written as U U T
because U is orthogonal. This matrix is symmetric because H T =
(U U T )T = U T U T = H as is diagonal so it is equal to its own
transpose. To see that it is positive definite we can use the eigenvalue
test: the eigenvalues of H are given by the diagonal elements of ,
i.e. the singular values of A. They are all nonnegative because
they are the eigenvalues of AT A, and they cannot be zero because
A is invertible by assumption. Hence the eigenvalues of H are all
positive.
(b) Factor this 2 by 2 matrix into A = U V T and then into A = HQ :
 
1 3
A= = U V T = HQ
1 3
 
2 0
Solution. We have AT A = so the singular values are
0 18

1 = 18 = 3 2 and 2 = 2 and the corresponding  eigenvectors
0 1
are v1 = (0, 1) and v2 = (1, 0) so that V = . We then have
1 0

u1 = Av1 /1 = (1/ 2, 1/ 2) u2 = Av2 /2 = (1/ 2, 1/ 2),
so the SVD is
   
1/2 1/2 3 2 0 0 1
A= .
1/ 2 1/ 2 0 2 1 0
Finally
   
2 2 2 1/2 1/2
H = U U T = Q = UV T .
2 2 2 1/ 2 1/ 2

6
5. (a) Are the vectors (0, 1, 1), (1, 0, 1), (1, 1, 0) independent or dependent?
Solution. These vectors are independent. One way to see this is
that
0 1 1
det 1 0 1 = 2 6= 0
1 1 0
(b) Suppose T is a linear transformation with input space = output
space = R3 . We have a basis u, v, w for R3 and we know that T (u) =
v + w, T (v) = u + w, T (w) = u + v. Describe the transformation T 2
by finding T 2 (u) and T 2 (v) and T 2 (w).
Solution. We have

T 2 (u) = T (v + w) = T (v) + T (w) = 2u + v + w

T 2 (v) = T (u + w) = T (u) + T (w) = u + 2v + w

T 2 (w) = T (u + v) = T (u) + T (v) = u + v + 2w

Note that this means that in the basis (u, v, w), the matrix of T 2 is

2 1 1
1 2 1
1 1 2

7
6. Suppose A is a 3 by 3 matrix with eigenvalues = 0, 1, 1 and corre-
sponding eigenvectors x1 , x2 , x3 .

(a) What is the rank of A? Describe all vectors in its column space
C(A).
Solution. Vectors x1 , x2 , and x3 are independent. Any vector y in
R3 can be represented as a linear combination of the eigenvectors:
y = ax1 + bx2 + cx3 . Applying A we get Ay = bx2 cx3 . Thus x2
and x3 form a basis in the column space and the rank of A is 2.
(b) How would you solve du/dt = Au with u(0) = (1, 1, 1)?
Solution. By the formula u(t) = c1 e1 t x1 + + cn en t xn , where
i are eigenvalues and xi the corresponding eigenvectors. We are
given i and xi , so we can plug them in to get: u(t) = c1 e0t x1 +
c2 et x2 + c3 et x3 = c1 x1 + c2 et x2 + c3 et x3 . To find the coefficients
c1 , c2 , and c3 , we need to use the initial conditions, that is to solve
the equation: u(0) = (1, 1, 1) = c1 x1 + c2 x2 + c3 x3 .
(c) What are the eigenvalues and determinant of eA ?
Solution. The eigenvalues of eA are the same as the eigenvalues of
e , where is the diagonalization of A. Therefore, the eigenvalues
of eA equal e to the power of the eigenvalues of A: e0 = 1, e1 = e
and e1 = 1/e. The determinant is the product of the eigenvalues
and is equal to 1 e 1/e = 1.

8
7. (a) Find a 2 by 2 matrix such that
       
1 3 2 3
A = and also A =
2 4 1 4
or say why sucha matrix cant
 exist.  
1 1 1
Solution. A = is the 2 by 2 matrix such that A =
4/3 4/3 2
     
3 2 3
and A = . One way to arrive at A is to let B =
4 1 4
   
3 3 1
be the matrix which sends the standard basis vectors
4 4 0
     
0 3 1 2
and both to and let C = be the change of basis
1 4 2 1
   
1 2
matrix which sends the standard basis vectors to and .
2 1
Then A = BC 1 .
(b) The columns of this matrix H are orthogonal but not orthonormal:

1 1 1 1
1 1 1 1
H=
0 2 1

1
0 0 3 1
Find H 1 by the following procedure. First multiply H by a diagonal
matrix D that makes the columns orthonormal. Then invert. Then
account for the diagonal matrix D to find the 16 entries of H 1 .
Solution. To normalize the columns
of H, we let
D be the diagonal
matrix with diagonal entries 1/ 2, 1/ 6, 1/ 12, and 1/2, and
we multiply H by D on the right: H 0 = HD. Because H 0 is an
orthogonal matrix, H 01 = H 0T. Then H 1 = D(HD)1 = DH 0T .
1/2 1/2 0 0
1/6 1/6 1/3 0
Computing, we obtain H 1 = 1/12 1/12 1/12 1/4.

1/4 1/4 1/4 1/4

9
8. (a) Factor this symmetric matrix into A = U T U where U is upper
triangular:

1 1 1
A = 1 2 2 .
1 2 3

Solution. By applying row operations we find the factorization

A = LU
1 0 0 1 1 1
A = 1 1 0 0 1 1
1 1 1 0 0 1
so that L = U T .
(b) Show by two different tests that A is symmetric positive definite.
Solution. Unfortunately it is hard to compute the eigenvalues ex-
plicitly, but nevertheless one can apply one of these tests:
i. A = U T U for U invertible;
ii. the energy test, xT Ax = xU T U x = kU xk2 0 of x 6= 0 because
U is invertible;
iii. the pivots of A are the pivots of U which are all positive;
iv. the upper left determinants of A are all 1 hence positive;
v. the eigenvalues satisfy the equation (3 62 + 5 1) which
cannon be zero for negative by checking the signs in the sum.
(c) Find and explain an upper bound on the eigenvalues of A. Find and
explain a (positive) lower bound on those eigenvalues if you know
that
2 1 0
A1 = 1 2 1 .
0 1 1

Solution. The eigenvalues 1 , 2 , 3 are positive and they sum to

the trace, which is 6, so they can be at most 6. The inverses of the
eigenvalues 1/1 , 1/2 , 1/3 are the eigenvalues of A1 , which has
trace 5, so this tells us that each of the i is at least 1/5.

10
Scrap Paper

11