CONCEPTUAL PROBLEMS SOLUTIONS, AUTUMN 2020

Contents

1. CP 1 2
1.1. Ch 1, #1 2
1.2. Ch 1, #4 3
1.3. Ch 1, #7 4
1.4. Ch 1, #9 5
2. CP 2 5
2.1. Ch 2, #1 5
2.2. Ch 2, #3 6
2.3. Ch 2, #8 6
2.4. Ch 2, #9 7
3. CP 3 9
3.1. Ch 2, #5 9
3.2. Ch 2, #7 10
3.3. Ch 3, #1 11
3.4. Ch 3, #2 12
4. CP 4 13
4.1. Ch 3, #3 13
4.2. Ch 3, #5 14
4.3. Ch 3, #6 16
4.4. Ch 3, #7 17
5. CP 5 18
5.1. Ch 3, #8 18
5.2. Ch 3, #10 19
5.3. Ch 3, #12 19
5.4. Ch 3, #14 20
6. CP 6 21
6.1. Ch 4, #1 21
6.2. Ch 4, #2 22
6.3. Ch 4, #4 23
6.4. Ch 4, #5 24

Date: Last updated: 2020/12/05, at 8:34pm.

1
2 CONCEPTUAL PROBLEMS SOLUTIONS, AUTUMN 2020

7. CP 7 25
7.1. Ch. 4, #7 25
7.2. Ch. 4, #8 25
7.3. Ch. 4, #9 26
7.4. Ch. 4, #10 27
8. CP 8 27
8.1. Ch 5, #2 27
8.2. Ch 5, #3 28
8.3. Ch 5, #8 30
8.4. Ch 6, #2 30
9. CP 9 30
9.1. Ch 6, #3 30
9.2. Ch 6, #6 32
9.3. Ch 6, #7 32
9.4. Ch 5, #9 33

1. CP 1
1.1. Ch 1, #1. (1.1: Modeling) Matt is a software engineer writing a script involving 6 tasks. Each must
be done one after the other. Let ti be the time needed to complete the ith task. These times have a certain
structure:
ˆThe total time needed to complete any 3 adjacent tasks is half the total time required to complete the
next two tasks.
ˆThe second task takes 1 second.
ˆThe fourth task takes 10 seconds.

(a) Write an augmented matrix for the system of equations describing the length of each task.
(b) Reduce this augmented matrix to reduced echelon form.
(c) Suppose he knows additionally that the sixth task will take 20 seconds and the first three tasks together
will take 50 seconds. Write the extra rows that you would add to your answer in (b) to take account of this
new information.
(d) Solve the system of equations in (c).

Solution. The equations are

t1 + t2 + t3 = 21 t4 + 12 t5
t2 + t3 + t4 = 12 t5 + 12 t6
t2 = 1
t4 = 10.
 CONCEPTUAL PROBLEMS SOLUTIONS, AUTUMN 2020 3

(a)–(b) The augmented matrix, and its reduced echelon form, are:
1 1 1 − 12 − 12 1
   
0 0 1 0 0 0 0 2 15
 0 1 1 1 −1 −1 0   0 1 0 0 0 0 1 
 2 2 ∼ 
 0 1 0 0 0 0 1   0 0 1 0 − 12 − 12 −11 
0 0 0 1 0 0 10 0 0 0 1 0 0 10
The row operations used were (in order):

• R2 ↔ R3 , then R3 − R2 → R3 (to reach echelon form), then

• R3 − R4 → R3 and R1 + 12 R4 → R1 (to clear out the fourth column), then
• R1 − R3 → R3 (to clear out the third column), then
• R1 − R2 → R1 (to clear out the second column).
 
1 1 1 0 0 0 50
(c) The additional rows would be:
0 0 0 0 0 1 20
(d) To solve the new system, add the new rows to the echelon form in (b) as additional rows and then reduce
to echelon form. You can do this since the solution set of the original system and the equations from the
echelon form is the same. The unique solution of the new system is (t1 , . . . , t6 ) = (5, 1, 44, 10, 90, 20). 

1.2. Ch 1, #4. (1.1: Geometry) Suppose we want to express the point (2, 3) in R2 as the solution space of
a system of linear equations.

(a) What is the smallest number of equations you would need? Write down such a system.
(b) Can you add one more equation to the system in (a) so that the new system still has the unique solution
(2, 3)?
(c) What is the maximum number of distinct equations you can add to your system in (a) to still maintain
the unique solution (2, 3)?
(d) Is there a general form for the equations in (c)?

Solutions. Many answers are possible.

(a) Two equations are needed. The simplest two equations are {x = 2, y = 3}. Other equations include
x + y = 5, or x − y = −1, or 2x = 4 (you need to select two equations in total). Basically any pair
of distinct equations satisfied by (x, y) = (2, 3).
(b) Yes: for instance, {x = 2, y = 3, x + y = 5}.
(c) Infinitely-many different equations are possible! For instance {x + 2y = 8, x + 3y = 11, x + 4y = 14,
. . . , 5x − y = 7}, and so on.
(d) Let’s figure this out. We’re considering all the equations
c1 x + c2 y = b
that are satisfied by (x, y) = (2, 3). In other words, c1 · 2 + c2 · 3 should equal b.
Let’s treat this as an equation telling us what b should be in terms of c1 , c2 : we have b = 2c1 + 3c2 .
So the general form of the equations is
c1 x + c2 y = 2c1 + 3c2 , where c1 , c2 are any real numbers.
Comment: Can you figure out what choices of c1 , c2 lead to each of the equations we had above?


4 CONCEPTUAL PROBLEMS SOLUTIONS, AUTUMN 2020

1.3. Ch 1, #7. (1.2: Solving linear equations) (a) Use Gauss-Jordan elimination to find the general solution
for the following system of linear equations:

z2 + 3z3 − z4 = 0
−z1 − z2 − z3 + z4 = 0
−2z1 − 4z2 + 4z3 − 2z4 = 0

(b) Give an example of a non-zero solution to the previous system of linear equations.
(c) The points (1, 0, 3), (1, 1, 1), and (−2, −1, 2) lie on a unique plane a1 x1 + a2 x2 + a3 x3 = b. Using your
previous answers, find an equation for this plane. (Hint: think about the relationship between the previous
system and the one you would need to solve in this question.)

Solution. (a) We’ll use row operations, starting from the augmented matrix. Remember to work left-to-right
(during Gaussian elimination, until reaching echelon form), then right-to-left (during the “Jordan” part, to
reach reduced echelon form).
     
0 1 3 −1 0 −1 −1 −1 1 0 −1 −1 −1 1 0
−1 −1 −1 1 0 ∼R1 ↔R2  0 1 3 −1 0 ∼R3 −2R1 →R3  0 1 3 −1 0
−2 −4 4 −2 0 −2 −4 4 −2 0 0 −2 6 −4 0
   
−1 −1 −1 1 0 1 1 1 −1 0
∼R3 +2R2 →R3  0 1 3 −1 0 (echelon form!) ∼ −R1 → R1 0 1 3 −1 0
0 0 12 −6 0 1 R → R
12 2 2 0 0 1 − 12 0
1 1 0 − 12 0
     
1 1 1 −1 0 1 0 0 −1 0
1 1 1
∼R2 −3R3 →R2 0 1 0 2 0 ∼R1 −R3 →R1 0 1 0 2 0 ∼R1 −R2 →R1 0 1 0 2 0.
1
0 0 1 −2 0 0 0 1 −2 0 1
0 0 1 − 12 0

Whew. So, the final equations are z1 − z4 = 0, z2 + 21 z4 = 0, z3 − 12 z4 = 0. The leading variables

are z1 , z2 , z3 ; the free variable is z4 , so we set z4 = t, a free parameter. The general solution is:
z = (z1 , z2 , z3 , z4 ) = (t, − 12 t, 12 t, t).

(b) Since the system is homogeneous, we know immediately that there’s the trivial solution: (0, 0, 0, 0).
For a nontrivial solution, we could set t = 2 to get z = (2, −1, 1, 2).

(c) We can plug in the three points to get equations in terms of a1 , a2 , a3 , b:

a1 + 3a3 − b = 0
a1 + a2 + a3 − b = 0
−2a1 − a2 + 2a3 − b = 0

These are almost the same equations! We have to swap the first two variables: z1 = a2 and z2 = a1 (and
z3 = a3 , z4 = b). Then each equation will be proportional to the corresponding z equation above – so the
solution sets are the same. So, our solution z = (2, −1, 1, 2) tells us (a1 , a2 , a3 , b) = (−1, 2, 1, 2), giving the
plane:
−x1 + 2x2 + x3 = 2.
Notice that we needed to have a nontrivial solution to part (b). We could not use (a1 , a2 , a3 , b) = (0, 0, 0, 0)
– that doesn’t give an equation of a plane.

(Comment: implicitly, swapping the first two variables was a column operation, and it changed the
solution vector. By contrast, row operations do not change the set of solutions.) 
 CONCEPTUAL PROBLEMS SOLUTIONS, AUTUMN 2020 5

1.4. Ch 1, #9. (1.2: Solving linear equations) Consider the following linear system with a and b unknown
non-zero constants.
x1 − 3x2 + x3 = 4
2x1 − 8x3 = −2
−6x1 + 6x2 + ax3 = b
(1) For what values of a and b does the system have infinitely many solutions?
(2) Given an example of a and b where the system has exactly one solution.
(3) Give an example of a and b for which the system has no solutions.

Solution. The echelon form of the system is

 
1 −3 1 4
 0 3 −5 −5  .
0 0 a − 14 b+4

(a) The system has infinitely many solutions when the last row is all zero. Therefore, a = 14 and b = −4.

(b) For one solution it suffices to take a 6= 14, say a = 15. Then for a given b, the unique solution is
x3 = b + 4, x2 = −5+5(b+4)
3 = 15+5b
3 and x1 = 4 − (b + 4) + (−5 + 5(b + 4)) = 4b + 15. If we choose b = 0 for
example, then the unique solution with a = 15 and b = 0 is (15, 5, 4).

(c) The system has no solution if a = 14 but b 6= −4. 

2. CP 2
2.1. Ch 2, #1. When Jake works from home, he typically spends 40 minutes of each hour on research, and
10 on teaching, and drinks half a cup of coffee. (The remaining time is spent on the internet.) For each hour
he works in the math department, he spends around 20 minutes on research and 30 on teaching, and doesn’t
drink any coffee. Lastly, if he works at a coffeeshop for an hour, he spends 25 minutes each on research and
teaching, and drinks a cup of coffee.
(Note: be careful about units of minutes versus hours.)

(1) Last week, Jake spent 10 hours working from home, 15 hours working in his office in Padelford Hall,
and 2 hours working at Cafe Allegro. Compute what was accomplished, and express the result as a
vector equation.
Solution. Let h denote the vector whose components record the fraction of time Jake spends on
research, fraction of time spent on teaching and the amount of coffee he drinks while at home.
Similarly, d and c recordthis  information
 1  atthe department and coffeeshop.
 Then we have
2 25

3 3 60 12.5 hours research
10h + 15d + 2c = 10  16  + 15  12  + 2  25
60
 =  10 hours teaching  . 
1
2 0 1 7 cups of coffee
(2) This week, Jake has 15 hours of research to work on and 10 hours of work related to teaching. He
also wants 11 cups of coffee, because... of... very important reasons. How much time should he
spend working from home, from his office, and from the coffeeshop?
 
15
Solution. We’re trying to solve the equation x1 h + x2 d + x3 c = 10. The augmented matrix, and
11
its reduced echelon form, are:

[ccc|c] 32 13 60
25
   
15 [ccc|c]1 0 0 14
 1 1 25
10 ∼  0 1 0 12 .
6 2 60
1
2 0 1 11 0 0 1 4
6 CONCEPTUAL PROBLEMS SOLUTIONS, AUTUMN 2020

So Jake should spend 14 hours working from home, 12 hours working in the office, and 4 hours at
the coffeeshop. 
(3) Describe the situation in part (b) as a vector equation and a matrix equation At = w. What do the
vectors t and w mean in this context? For which other vectors w does the equation At = w have a
solution?
Solution. As a vector and matrix equation:
   2 1 25     
15 3 3 60 t1 w1
t1 h + t2 d + t3 c = 10  1 1 25  · t2  = w2  .
6 2 60
1
11 2 0 1 t3 w3
The entries of t are the amounts of time spent in each location. The entries of w are the amounts
of work done, in hours (or coffee consumed, in cups).
Since the echelon form has no zero rows, the system is consistent for any choice of w. This means
that it’s possible – mathematically at least – to find a combination of times at home, at the office,
and at the coffeeshop, to accomplish any combination of work and coffee. 1 
(4) Jake tries working in the math department lounge for an hour, and gets 30 minutes of research and
20 minutes of teaching work done, while having time to drink 31 of a cup of coffee. Not bad. But
Jake’s colleague Vasu claims that there’s no need to work in the lounge – the other options already
give enough flexibility. Is he right? Explain mathematically.

Solution. Vasu is right: based on the previous question, we know that l ∈ span(h, d, c). Explicitly (if
you solve the equations): l = 10 3 2
21 h + 7 d + 21 c. Thus, there is no extra flexibility gained by including
l as an option. 
     
2 1 −4
2.2. Ch 2, #3. (after 2.2) Find all values z1 and z2 such that −1, 2, and  z1  do not span R3 .
3 2 z2

Solution. The vectors will not span R3 if and only if they are linearly dependent. This is the same as saying
that the echelon form of the matrix whose columns are the three vectors will not have a pivot in every row
which means there is a row of zeros in the echelon form. So, row-reduce:

       
2 1 −4 2 1 −4 2 1 −4 2 1 −4
−1 2 z1  ∼ 0 2.5 z1 − 2 ∼ 0 2.5 z1 − 2 ∼ 0 2.5 z1 − 2 .
3 2 z2 3 2 z2 0 0.5 z2 + 6 0 0 z2 + 6 − 15 (z1 − 2)

The vectors will fail to span if and only if z2 +6− 51 z1 + 25 = 0. In other words, if and only if z1 −5z2 = 32. 

2.3. Ch 2, #8. (after 2.3) In each of the following cases, either find an example that contradicts the
statement showing that it is false, or explain why the statement is always true.

(a) If {u1 , u2 , u3 } is a spanning set for Rn , then {u1 , u2 , u3 , u4 } is also a spanning set for Rn . What are
all possible values of n for which three vectors u1 , u2 , u3 can span Rn ?
Solution. This is true. We are given that every b ∈ Rn is a linear combination of {u1 , u2 , u3 }. If
we bring in a new vector u4 then b is still a linear combination of the first three vectors and hence
a linear combination of all four vectors. Three vectors can span R1 , R2 or R3 , so the possible values
of n are 1, 2, 3. 
(b) If {u1 , u2 , u3 } is a spanning set for Rn , then {u1 , u1 + u2 , u1 − u3 } also spans Rn .

1But in practice there’s an important caveat: negative amounts of time don’t make sense! So we would also need impose
the inequalities t1 , t2 , t3 ≥ 0. We haven’t learned how to do this, but the techniques exist (called linear programming) and are
important for optimization.
 CONCEPTUAL PROBLEMS SOLUTIONS, AUTUMN 2020 7

Solution. If {u1 , u2 , u3 } is a spanning set for Rn then any b ∈ Rn can be written as u1 c1 +u2 c2 +u3 c3 .
If {u1 , u1 + u2 , u1 − u3 } should span Rn , then for any b ∈ Rn there must be a solution to
b = u1 λ1 + (u1 + u2 )λ2 + (u1 − u3 )λ3 .
Rewriting this we see that there must be λ1 , λ2 , λ3 such that
b = u1 (λ1 + λ2 + λ3 ) + u2 λ2 + u3 (−λ3 ) .
On the other hand we know that there are c1 , c2 , c3 such that
b = u1 c1 + u2 c2 + u3 c3 .
Setting
c1 = λ1 + λ2 + λ3 , c2 = λ2 and c3 = −λ3
we see that we just need to take
λ3 = −c3 , λ2 = c2 and λ1 = c1 − c2 + c3 .
Therefore the statement is true. 
(c) If u1 , u2 , u3 are linearly independent, then u1 , u2 , u3 , u4 are also linearly independent.
Solution. This is false. We could take u4 to be 0 and then u1 , u2 , u3 , u4 would be linearly dependent.

(d) If u1 , u2 , u3 are linearly independent, then u1 , u1 + u2 , u1 − u3 are also linearly independent.
Solution. Consider the equation
u1 x1 + (u1 + u2 )x2 + (u1 − u3 )x3 = 0.
Rewriting we have,
u1 (x1 + x2 + x3 ) + u2 x2 + u3 (−x3 ) = 0.
Since u1 , u2 , u3 are linearly independent, the above equation only has the trivial solution. Therefore,
x3 = 0, x2 = 0, x1 = 0. This means that u1 , u1 +u2 , u1 −u3 are also linearly independent. Therefore,
the statement is true. 
3 3
(e) If u1 , u2 , u3 do not span R , then there is a plane P in R that contain all of them. (Bonus: how
can we find this plane? Does the plane go through the origin?)

Solution. If the vectors don’t span R3 , they either span {0}, a line in R3 or a plane in R3 which
are all the possible types of spans in R3 except for all of R3 . Spans of vectors always contain the
origin, since the origin is always a linear combination of any set of vectors by using 0’s as the scalar
multiples. The span is {0} if and only if all the vectors are 0. The span is a line if and only if at
least one vector is non-zero and the others are multiples of this one. If neither of these happen then
the span is a plane. In all cases, we can certainly find a plane that contains the span. Therefore, the
statement is true. 

2.4. Ch 2, #9. (after 2.3) Recall that if we have m vectors u1 , u2 , . . . , um in Rn , then we can form the
matrix A whose columns are u1 , . . . , um . Let B be the echelon form of A. All the questions below are based
on such a matrix B. Most questions have a yes/no answer. Give full reasons for all answers.
Suppose we are given the following matrix B:
 
3 0 −1 5
0 0 2 −1
0 0 0 0
Note that the columns of B are not u1 , . . . , um .

(a) What is n?
Solution. n = 3. This is the size of each ui . 
(b) What is m?
Solution. m = 4. This is the number of vectors ui . 
(c) Are u1 , . . . , um linearly independent?
8 CONCEPTUAL PROBLEMS SOLUTIONS, AUTUMN 2020

Solution. No, they are not since B does not have a pivot in every column. So there would be
free variables when you solve Ax = 0 or equivalently, Bx = 0 allowing for non-zero solutions to
Ax = 0. 
n
(d) Does {u1 , . . . , um } span R ?
Solution. No, since B does not have a pivot in every row. If you want to solve Ax = b, then the
echelon form of the augmented matrix [A | b] would be [B | b0 ] for some vector b0 . Since the last row
of B is all zeros, you would need the last entry of b0 to also be zero for Ax = b to have a solution.
So not all b in R3 is in the span of u1 , u2 , u3 , u4 . 
(e) Looking at B can you write down a subset of the original set {u1 , . . . , um } that would be guaranteed
to be linearly independent?
Solution. {u1 , u3 } are linearly independent. This is because the echelon form of [u1 , u3 ] consists of
the first and third column of B which has a pivot in every column. 
n
(f) Is there a subset of the original set {u1 , . . . , um } that would be guaranteed to span R ?
Solution. No. Recall that the span of a collection of vectors is the set of all linear combinations of
the vectors. The original set {u1 , . . . , u4 } already did not span R3 , so a subset cannot either. 
(g) Write down a b ∈ Rn for which Bx = b does not have a solution.
 
0
Solution. Pick a b with non-zero last coordinate. Example: 0. Then the last row of [B | b] says
1
0x1 + 0x2 + 0x3 + 0x4 = 1 which does not have a solution. 
(h) Write down a b ∈ Rn for which Bx = b has a solution.
 
1
Solution. Pick a b with zero last coordinate. Example: 1. 
0
(i) Write down a b ∈ Rn for which Bx = b has a unique solution.
Solution. There is no such b since there are free variables in the solution of Bx = b when there is
a solution. 
n
(j) Is there a new vector w ∈ R that you could add to the set {u1 , . . . , um } to guarantee that
{u1 , . . . , um , w} will span Rn ?
 
0
Solution. Yes, pick 0. Then the echelon form of A (assuming that you did not switch any rows
1
while doing Gaussian elimination) will still be this vector. Now there is a pivot in every row of the
echelon form and the five vectors together will span R3 . 
(k) Is there a column of B that is in the span of the rest? If so, find it.
Solution. Yes. The columns of B are linearly dependent since not all columns of B have a pivot.
Solving Bx = 0 we see that x2 and x4 are free variables. Set x4 = 2 and x2 = 1. Then we get the
solution x1 = −3 and x3 = 1. This means that −3B1 + B2 + B3 + 2B4 = 0 where Bi is the ith
column of B. So any column is in the span of the others. 
(l) Looking at B do you see a ui that is in the span of the others? How can you identify it?
Solution. Remember that Ax = 0 and Bx = 0 have the same solutions. So the above solution
(−3, 1, 1, 2) is also a solution of Ax = 0. By the same logic as above, every column of A is the span
of the others. 
(m) Put B into reduced echelon form.
Solution.  
1 0 0 3/2
0 0 1 −1/2
0 0 0 0

(n) Write down a non-zero solution of Ax = 0 if you can.
 CONCEPTUAL PROBLEMS SOLUTIONS, AUTUMN 2020 9

Solution. We already saw one in (12). From the reduced echelon form you can get the solutions
x3 = 12 x4 and x1 = − 32 x4 . Choose x4 = 1 say and you get the solution (−3/2, 0, 1/2, 1). 
(o) How many free variables are there in the set of solutions to Ax = b when there is a solution?
Solution. Two (x2 and x4 ). 
(p) If you erased the last row of zeros in B then would the columns of the resulting matrix be linearly
independent?
Solution. No. Some columns are still missing a pivot. 
(q) Can you add rows to B to make the columns of the new matrix linearly independent? If yes, give
an example of the new matrix you would construct.
Solution. Yes. You need to add enough rows to get a pivot in every column. For example:
 
3 0 −1 5
0 0 2 −1
 
0 0 0 0
 
 
0 1 0 0
0 0 0 1


3. CP 3
3.1. Ch 2, #5. (after 2.3)
     
[r]1 [r]0 [r]t
(1) Let a1 =  1 , a2 =  2 , and a3 =  −3 . Find all values of t for which there will be a unique
−1 3 −7
solution to a1 x1 + a2 x2 + a3 x3 = b for every vector b in R3 . Explain your answer.
Solution. Since we have 3 vectors in R3 , the “unique solution for every b” property is equivalent to
the three vectors spanning R3 or that they are linearly independent. Basically, we want the echelon
form of the matrix to have three distinct pivots (no free variables or rows of zeroes). So, row-reduce:
     
1 0 t 1 0 t 1 0 t
 1 2 −3 ∼ 0 2 −3 − t ∼ 0 2 −3 − t 
3
−1 3 −7 0 3 −7 + t 0 0 −7 + t − 2 (−3 − t)
So, we want the bottom-right entry, which simplifies to 2.5t − 2.5, to be nonzero. In other words,
t 6= 1. 
(2) Are the vectors a1 and a2 from part (a) linearly independent? Explain your answer.
Solution. Yes, they are linearly independent. This is because it’s easy to tell when a set of two
vectors is linearly (in)dependent: they’re dependent if and only if they are proportional (i.e., one is
a scalar multiple of the other). In this case, they clearly are not proportional. 
 
[r]1
(3) Let a1 , a2 and a3 be as in (a). Let a4 =  4 . Without doing any further calculations, find all
−5
values of t for which there will be a unique solutions to a1 y1 + a2 y2 + a3 y3 + a4 y4 = c for every
vector c in R3 . Explain your answer.
Solution. The property we’re considering:
“for every c ∈ R3 , there’s a unique solution to a1 y1 + a2 y2 + a3 y3 + a4 y4 = c”
would mean that the vectors {a1 , a2 , a3 , a4 } are linearly independent and span R3 . The spanning
part is possible, but the linear independence part is impossible regardless of the value of t we choose,
because four vectors in R3 are always linearly dependent. 
10 CONCEPTUAL PROBLEMS SOLUTIONS, AUTUMN 2020

3.2. Ch 2, #7. For each of the situations described below, give an example (if it’s possible) or explain
why it’s not possible.

(a) A set of vectors that does not span R3 . After adding one more vector, the set does span R3 .
   
1 0
Solution. The simplest example is to start with 0 , 1, which doesn’t span R3 . Then we add in
0 0
 
0
the vector 0, and now the vectors do span R3 . 
1
(b) A set of vectors that are linearly dependent. After adding one more vector, the set becomes linearly
independent.
Solution. This is impossible. Suppose our original set of vectors has a nontrivial linear combination
equalling zero, say x1 v1 + · · · + xm vm = 0. Well, this still counts as a nontrivial linear combination
when we add in the a new vector w. In other words, we can write
x1 v1 + · · · + xm vm + 0w = 0.
This is nontrivial because one of the x1 , . . . , xm coefficients is nonzero (therefore it’s fine for the last
coefficient to be zero). 
3
(c) A set of vectors in R with the following properties (four possibilities):
spans R3 , spans R3 ,
linearly independent linearly dependent
doesn’t span R3 , doesn’t span R3 ,
linearly independent linearly dependent
For each case that is possible, how many vectors could be in the set? (State any constraints, as
in “there must be at least...” or “at most...”)
Solution. All four cases are possible. Here are some relatively-simple examples:
spans R3 , spans R3 ,
linearly independent linearly dependent
             
1 0 0 1 0 0 2
             
0 , 1 , 0 0 , 1 , 0 , 0
             
0 0 1 0 0 1 0
3 3
doesn’t span R , doesn’t span R ,
linearly independent linearly dependent
         
1 0 1 2 3
         
0 , 1 0 , 0 , 0 , . . .
         
0 0 1 2 3

Also, there are some constraints on the number of vectors:

spans R3 , spans R3 ,
linearly independent linearly dependent
exactly 3 vectors at least 4 vectors
doesn’t span R3 , doesn’t span R3 ,
linearly independent linearly dependent
at most 2 vectors any number of vectors

(d) A system of equations with a unique solution. After adding another equation to the system, the new
system has infinitely-many solutions.
 CONCEPTUAL PROBLEMS SOLUTIONS, AUTUMN 2020 11

Solution. This is impossible. Adding another equation to the system can only reduce the size of
the solution set (or, if we’re lucky, leave it unchanged). 
(e) A system of equations without any solutions. After deleting an equation, the system has infinitely-
many solutions.
Solution. This is possible. Let’s take a system that would have had infinitely-many solutions, then
add in an inconsistent equation. For instance:

x + y + z = 1

y+z =5

y + z = 3.


There are no solutions, but if we delete y + z = 3, we then have infinitely-many. 

3.3. Ch 3, #1. (after 3.1)

(1) Find a linear transformation T : R2 → R3 such that

     
  1   −2   −1
1 0 1
T = 0 , T =  2  , and T =  2 ,
1 2 3
2 0 2
or if it’s impossible, explain why.  
  −1
1
(2) How does your answer change if the third condition changes to T =  2 ?
3
1

Solution. (1) First, note that T is sending a vector in R2 to a vector in R3 . Therefore, T is defined by
a 3 × 2 matrix A such that T (x) = Ax. Our job is to find this matrix A. Let us start by setting
 
a b
A = c d .
e f
Then the conditions given to us say that
a + b = 1, c + d = 0, e + f = 2, 2b = −2, 2d = 2, 2f = 0.
Solving this we get that b = −1, d = 1, f = 0, a = 2, c = −1, e = 2. Therefore
 
2 −1
A = −1 1  .
2 0
Check that    
  2 −1   −1
1 1
T = −1 1 =  2 .
3 3
2 0 2
(2) Recall that for a linear transformation T , we need to have that T (u + v) = T (u) + T (v). Therefore,
since      
1 1 0
= + ,
3 1 2
we must get that
     
      1 −2 −1
1 1 0
T =T +T = 0 +  2  =  2  .
3 1 2
2 0 2
However, this is not what we are given, and hence there cannot be a linear transformation as in (b).

12 CONCEPTUAL PROBLEMS SOLUTIONS, AUTUMN 2020

3.4. Ch 3, #2. (after 3.1) For each of the regions below, find a linear transformation T : R2 → R2 that
takes the unit square
U = {(x, y) ∈ R2 : 0 ≤ x ≤ 1 and 0 ≤ y ≤ 1}
to the given region, or explain why it’s impossible to do so.

(−1, 1) (1, 1)

(a)
Solution. To describe a linear transformation T , we have to do is determine where the standard basis
vectors go. One option is:     
x 1 −1 x
T = .
y 1 1 y
Check that (1, 1) goes to (0, 2) under T and that (0, 0) goes to (0, 0). Thus the corners of the unit
square U go to the corners of a the diamond in the above picture and it seems the diamond is T (U).
In Math 308 this is probably enough, but if you want a serious argument, you have to show that
T (U) is exactly the above diamond shape. For this, we use the fact that any point in U is of the
form        
0 1 0 1
α1 + α2 + α3 + α4
0 0 1 1
where αi ≥ 0 for i = 1, . . . , 4 and α1 + α2 + α3 + α4 = 1. This is a linear combination with
some conditions on the coefficients α1 , . . . , α4 , called a convex combination of the four corners of U.
Similarly, any point in the diamond is a convex combination of its corners and is therefore of the
form,        
0 1 −1 0
α1 + α2 + α3 + α4
0 1 1 2
where αi ≥ 0 for i = 1, . . . , 4 and α1 + α2 + α3 + α4 = 1.
If we know this then applying T to a point in U we get
                
0 1 0 1 0 1 0 1
T α1 + α2 + α3 + α4 =α1 T + α2 T + α3 T + α4 T
0 0 1 1 0 0 1 1
       
0 1 −1 0
=α1 + α2 + α3 + α4
0 1 1 2
which is how any point in the diamond looks. Thus we conclude that we have the right T . (The
other option for T is to swap the columns in the matrix defining T .) 
(1, 2)

(1, 1) (2, 1)
(b)
Solution. This is impossible because if there was a T then T (0) 6= 0 which is not possible in a linear
transformation. 
(1, 0)

(0, −1)
(c)
 CONCEPTUAL PROBLEMS SOLUTIONS, AUTUMN 2020 13

Solution. Same as (a). One option is:

    
x 1 0 x
T = .
y 0 −1 y
(Only other option: Swap the columns.) 

(2, 1)

(d)
Solution. If there is a linear transformation T in this case, then it is mapping (squishing) U onto the
line segment connecting 0 and (2, 1). A point on this line segment looks like (2t, t) where 0 ≤ t ≤ 1.
Suppose we send the basis element (1, 0) to (2t, t). Then we should send the other basis element
(0, 1) to (2s, s) so that when you sum the two images we get the point where (1, 1) gets sent to,
which we see has to be (2, 1). This means that we need
     
2t 2s 2
+ =
t s 1
which tells us that s = (1 − t). Since neither basis element should be sent to the origin (why?), we
get that the general solution is
    
x 2t 2(1 − t) x
T =
y t (1 − t) y
for any 0 < t < 1. So there are infinitely many possibilities here for T . 

4. CP 4
4.1. Ch 3, #3. (after 3.1) Assume T : Rm → Rn is a linear transformation.

(1) Suppose there is a nonzero vector x ∈ Rm such that T (x) = 0. Is it possible that T is one-to-one?
Give an example, or explain why it’s not possible.
Solution. This is impossible. Since T is a linear transformation, we know that T (0) = 0. If there’s a
different vector x ∈ Rm such that T (x) = 0, then T cannot be one-to-one since two different vectors
are being sent to 0 by T . 
m
(2) Suppose there is a nonzero vector x ∈ R such that T (x) = 0. Is it possible that T is onto? Give
an example, or explain why it’s not possible.
Solution. Yes, though we do need m > n (why?). For example, take
 
1 0 0
T : R3 → R2 such that T (x) = Ax = x.
0 1 0
 
0  
0
The matrix A has a pivot in every row, so T is onto, but T 0 = . 
0
1
(3) Suppose that u and v are linearly dependent vectors in Rm . Show that T (u) and T (v) are also
linearly dependent.
Solution. Two vectors u and v are linearly dependent if there exist some a, b not both equal to zero
so that
au + bv = 0
Let’s apply T to both sides of the above equation. We get
T (au + bv) = T (0)
aT (u) + bT(v) = 0
so T (u) and T (v) are also linearly dependent. 
14 CONCEPTUAL PROBLEMS SOLUTIONS, AUTUMN 2020

(4) Suppose that u and v are linearly independent vectors in Rm . Is it guaranteed that T (u) and T (v)
are also linearly independent? If yes, explain why. If no, give an example where this is not the case.
Solution. This is not always the case. Consider m = n = 2 and
 
0 0
T : R2 → R2 such that T (x) = Ax = x.
0 0
This is called the zero transformation as it sends all x ∈ R2 to 0. In particular, regardless of what
u and v are, {T (u), T (v)} is linearly dependent since any set containing the zero vector is always
linearly dependent. 

4.2. Ch 3, #5. (Geometry Question, after 3.1) Suppose we are given the unit square A in the plane with
corners (0, 0), (1, 0), (1, 1) and (0, 1).

(1) Find a linear transformation T that sends A to the parallelogram B with corners (0, 0), (1, 2), (2, 2)
and (1, 0).
Solution. We want T to map the edges of A to the edges of B:
2 2

1 1

T
-2 -1 1 2
−−−−→ -2 -1 1 2

-1 -1

-2 -2

The most direct way to do this is to make T send the left and bottom edges of A to the left and
bottom edges of B:        
1 1 0 1
T = and T = .
0 0 1 2
 
1 1
This means T (x) = · x. (Check: T sends the rest of A to the rest of B.) 
0 2
(2) Is the linear transformation T unique? Why or why not?
Solution. A linear transformation always has T (0) = 0, so T must map the edges of A touching the
origin to the edges of B touching the origin. However, you have two choices in doing this: send the
bottom edge of A to the bottom edge of B and the left edge of A to the left edge of B as above OR
switch the edges.
Switching the two edges attached at the origin, or sending the left edge of A to the bottom edge
of B and the bottom edge of A to the left edge of B mathematically can be written as:
       
1 1 0 1
S = and S = .
0 2 1 0
 
1 1
This means S(x) = · x.
2 0
This transformation maps A to B but also applies a ‘reflection’. 
0
(3) What linear transformation T , other than the identity, would send A to itself?
Solution. As in part (c), we have two choices: send the bottom edge of A to the bottom edge of A
(and left to left), or send the bottom edge of A to the left edge (and left to bottom). Mathematically,
this gives us two options:
 CONCEPTUAL PROBLEMS SOLUTIONS, AUTUMN 2020 15

Bottom to bottom and left to left:

       
1 1 0 0
T0 = and T 0 = .
0 0 1 1
 
1 0
This means T 0 (x) = · x.
0 1
Bottom to left and left to bottom:
       
1 0 0 1
S0 = and S 0 = .
0 1 1 0
 
0 1
This means S 0 (x) = · x.
1 0
T 0 is the identity matrix, so sends every point in A to itself. S 0 is instead a reflection, so sends
A to A, but each point in A is not necessarily mapped to itself (it is mapped to its reflection across
the line y = x). 
(4) Is there a linear transformation whose overall effect would be to send A to B and then translate B
in the horizontal direction by one unit? Explain.
Solution. We want to map A to the parallelogram
2 3

T 00
-2 -1 1 2
−−−−−→ -3 -2 -1 1 2 3

-1

-1

-2

-2 -3

00 00
 We can’t do this with a linear transformation because we don’t want T (0) = 0 (we want T (0) =
1
instead.) But a linear transformation always sends 0 to 0.
0
Instead, we could first map A to B like in part (a), then add a vector to shift the whole thing
to the right by one unit. In other words, let’s make T be a function of the form T 00 (x) = M x + v
(and figure out what M and v should be) so T 00 (x) is obtained by first applying the transformation
corresponding to M , then shifting
 the result by +v.
1 1
So, we can take M = from part (a), so M x is on the unshifted parallelogram. Then we
  0 2
1
can set v = . So in words, M x + v is just M x shifted to the right by one unit. So, our desired
0
transformation (which is an ‘affine transformation’, not a linear transformation) is
   
1 1 1
T 00 (x) = x+ .
0 2 0

(5) Find a linear transformation that will send A to a parallelogram C of area 4 while still keeping (0, 0)
and (1, 0) as two of its corners.
Solution. First note that there are many possible parallelograms C with area 4 and corners (0, 0)
and (1, 0). Let’s say another side of the parallelogram is the line from (0, 0) to the
 point (a, b) for
1
some a and b. If we know one side of the parallelogram is given by the vector and the other is
0
16 CONCEPTUAL PROBLEMS SOLUTIONS, AUTUMN 2020

 
a
given by , we know the formula for the linear transformation. If it sends the bottom edge of A to
b
the line segment from (0, 0) to (1, 0) and the left edge of A to the line segment from (0, 0) to (a, b),
then we know the linear transformation that does this:
       
1 1 0 a
T1 = and T1 = .
0 0 1 b
 
1 a
This means T1 (x) = · x.
0 b
As in part (c), we could also switch the edges, giving us the linear transformation T2 where
       
1 a 0 1
T2 = and T1 = .
0 b 1 0
 
a 1
so T2 (x) = · x.
b 0

In either case, we know the formula for the linear transformation in terms of the point (a, b).
So, we need to figure out what points (a, b) give us a parallelogram of area 4. The area of the
parallelogram C is given by A(C) = lh, where l is the length of base of the parallelogram and h is
the height. The base has length 1 and the height is h = b, so as long as b = 4, the area of C is 4.
Therefore, for any number a, the linear transformations
 
1 a
T1 (x) = ·x
0 4
and  
a 1
T2 (x) = ·x
4 0
map A to a parallelogram of area 4. 
(6) What is the general formula for the linear transformation that sends A to a parallelogram of area k
while still keeping (0, 0) and (1, 0) as two of its corners?
Solution. Using the same argument as in part (h), to get a parallelogram of area k, the other side
of the parallelogram must be the line segment from (0, 0) to (a, k) for some choice of a. Therefore,
the general formula for the linear transformations are:
 
1 a
T1 (x) = ·x
0 k
and  
a 1
T2 (x) = · x.
k 0


4.3. Ch 3, #6. (Geometry Question, after 3.1) Consider the triangle ∆ in R3 with corners (1, 0, 0), (0, 1, 0),
(0, 0, 1).

(1) Find the image of ∆ under the projection that sends (x, y, z) ∈ R3 7→ (x, y) ∈ R2 . Is this a linear
transformation?
Solution. The three corners of ∆ project to (1, 0), (0, 1) and (0, 0). The image of the projection T is
the triangle with these three points as corners. This is a linear transformation since
  
x    
x x
T α y  = α = αT , and
y y
z
   0 
x x
x + x0
     0 
0  x x
T y + y = 0 =T +T .
y0
   
y + y y
z z0
 CONCEPTUAL PROBLEMS SOLUTIONS, AUTUMN 2020 17


(2) What the area of the image of ∆ after the projection?
Solution. Using the (base × height) formula for the area of a triangle, we get that the area of the
projected triangle is 1/2. 
(3) Is there a linear transformation that will send ∆ to the xy-plane so that the image has the same
area as ∆, and one of the corners of the image is (0, 0)?
Solution. There are multiple answers. The first thing to notice is that in order to map the triangle
with corners (1, 0, 0), (0, 1, 0), (0, 0, 1) to the plane, we will need to map points in R3 to points in
R2 . To represent this transformation via a matrix, it will have to be a 2 × 3 matrix. In order for
one of the corners to be at (0, 0), we want to send one of the corners in R3 to (0, 0) ∈ R2 , so lets say
(0, 0, 1) goes to (0, 0). A simple example of a matrix that does this looks like
 
a 0 0
0 b 0
since applying the formula for multiplying a matrix by a vector gives
 
  0  
a 0 0   0
0 =
0 b 0 0
1
This matrix certainly sends the triangle to the plane with a corner where we want it, but how can
we ensure that it preserves area?
In order to answer that question, we first need to find the area of the original triangle. Recall that
given two vectors x, y, the area of the parallelogram spanned by the two vectors is given by |x × y|,
and the area of the corresponding triangle is 12 |x × y|. Denoting the corners of the triangle by the
vectors whose tips lie at the corner points, let
     
1 0 0
e1 = 0 e2 = 1 e3 = 0
0 0 1
√
Then the area of the triangle is given by 12 |(e1 − e2 ) × (e1 − e3 )| = 23 . This means that the triangle
√
we map to has to have this same area. All we need now is to take a = 3, b = 1, which turns our
matrix into
√ 
3 0 0
0 1 0
We can now see that
   
√  1 √  √  0  
3 0 0   3 3 0 0   0
0 = and 1 =
0 1 0 0 0 1 0 1
0 0
√
So this transformation gives us a new triangle with area 23 . Note there are several other matrices
that get this done. What happens when we swap columns of the above matrix? Does it still send
the same corner to (0, 0)? 

4.4. Ch 3, #7. (after 3.1) Say we have linear transformations T : R3 → R2 and S : R2 → R4 . Let
T S
S ◦ T : R3 → R4 be the composition (that is, R3 −−→ R2 −−→ R4 ).
Can S ◦ T be one-to-one?
(Hint: Start by thinking about all x such that T (x) = 0.)

Solution. Note that T is not one-to-one since its corresponding matrix A has more columns than rows (there
are 2 rows and 3 columns). Thus there is a nonzero vector x ∈ R3 such that T (x) = 0. This means
S(T (x)) = 0. Thus S ◦ T is not one-to-one. 
18 CONCEPTUAL PROBLEMS SOLUTIONS, AUTUMN 2020

5. CP 5
5.1. Ch 3, #8. (after 3.2) Let  
1 1 1
A= .
2 1 4
Find a 3 × 2 matrix B with AB = I2 . Is there more than one matrix B with this property? Justify your
answer.

Solution. We want to find a, b, c, d, e, f such that

 
  a b  
1 1 1  1 0
· c d = .
2 1 4 0 1
e f
Multiplying out the left side of the equation, we need to get
a+c+e=1
b+d+f =0
2a + c + 4e = 0
2b + d + 4f = 1.

This is a system of linear equations (surprise!) in six variables a, b, c, d, e, f , so we can solve it by Gaussian
elimination. The augmented matrix, and reduced echelon form, are:
   
1 0 1 0 1 0 1 1 0 0 0 3 0 −1
 0 1 0 1 0 1 0   0 1 0 0 0 3 1 
 2 0 1 0 4 0 0  ∼  0 0 1 0 −2 0
   
2 
0 2 0 1 0 4 1 0 0 0 1 0 −2 −1
Note: This matrix is not directly related to either of the matrices A or B. It is just the augmented matrix
for the four equations above in terms of a, b, c, d, e, f .
To describe the solutions, let e = s1 and f = s2 (free variables), and solve for the others, to get
   
a −1 − 3s1
 b   1 − 3s2 
   
 c   2 + 2s1 
 =
 d  −1 + 2s2  .

   
e  s1 
f s2

So, putting a, b, c, d, e, f back into the original matrix equation, we can see that, as we hoped,
 
  −1 − 3s1 1 − 3s2  
1 1 1  1 0
· 2 + 2s1 −1 + 2s2 =  .
2 1 4 0 1
s1 s2
 
−1 1
The simplest such matrix would be (by setting s1 = s2 = 0) B =  2 −1, but there are infinitely-many
0 0
possible matrices B that work. 

Solution. (Alternate solution) The task is to find a right inverse for A. We write A, I2 next to each other
and convert the resulting 2 × 5 (augmented) matrix into REF:
     
1 1 1 1 0 R2 −2R1 →R2 1 1 1 1 0 R1 +R2 →R1 1 0 3 −1 1
→ →
2 1 4 0 1 0 −1 2 −2 1 0 −1 2 −2 1
 
−R2 →R2 1 0 3 −1 1
→
0 1 −2 2 −1
 CONCEPTUAL PROBLEMS SOLUTIONS, AUTUMN 2020 19

We can view this as the augmented matrix corresponding to two linear systems (one involving a, c, e and one
involving b, d, f in Solution
 1). Then
 e (resp.
 f ) is a free
 variable;
 setting
 e = s1 (resp. f = s2 ) and solving
a −3s1 − 1 b −3s2 + 1
for a, c (resp. b, d) gives  c  =  2s1 + 2  (resp.  d  =  2s2 − 1 ). 
e s1 f s2

5.2. Ch 3, #10. (after 3.2) Let  

1 z
B= .
4 3
Find all values of z such that the linear transformation T induced by B fixes no line in R2 . (By “fixing a
line” we mean that T (v) = v for every point v on the line.)

 
v
Solution. We will try to find v = 1 such that T (v) = v, that is,
v2
     
1 z v v
· 1 = 1 .
4 3 v2 v2
In equations, this says v1 + zv2 = v1 and 4v1 + 3v2 = v2 , or after simplifying,
zv2 = 0
4v1 + 2v2 = 0.
If z 6= 0, the only solution is v2 = v1 = 0, so v = 0 so no lines are fixed by
 thislinear transformation. But,
−1/2 −1/2
if z = 0, the solutions are v = t · . So, the entire line spanned by is fixed by T if z = 0. 
1 1

Solution. (Alternate solution) Another way to write the equation Bv = v is Bv = Iv, where I is the
identity matrix. This seems unnecessary since Iv = v, but we can use it cleverly, to convert the equation to
something familiar:
Bv = Iv Bv − Iv = 0 (B − I)v = 0.
So this says v ∈ null(B − I), the nullspace of this new matrix,
     
1 z 1 0 0 z
B−I = − = .
4 3 0 1 4 2
Now we can solve for v the way we find nullspaces, with the augmented matrix
   
0 z 0 4 2 0
∼
4 2 0 0 z 0
If z 6= 0, this is in echelon
   and we see null(B − I) = {0}. But if z = 0, there are nontrivial solutions:
form,
−1/2
null(B − I) = span . 
1

5.3. Ch 3, #12. (after 3.3)

(1) Find a linear transformation T : R2 → R2 such that T (x) = Ax that reflects a vector (x1 , x2 ) about
the x2 -axis.
 
−1 0
Solution. Such a T would send (1, 0) to (−1, 0) and (0, 1) to itself. So A = . 
0 1
(2) Find a linear transformation S : R2 → R2 such that T (x) = Bx that rotates a vector (x1 , x2 )
counterclockwise by 135 degrees.
Solution. Use Pythagoras’ Theorem√to calculate
√ where the basis vectors would
√ √go after this rotation.
We’ll get that (1, 0) would go 21 (− 2, 2) and (0, 1) would go to 21 (− 2, − 2). Therefore, B =
 √ √ 
1 − √ 2 − √2 .
2 
2 − 2
20 CONCEPTUAL PROBLEMS SOLUTIONS, AUTUMN 2020

(3) Find a linear transformation U (with domain and codomain) that has the effect of first reflecting
as in (a) and then rotating as in (b). Give the matrix of this transformation explicitly. How is this
transformation related to T and S?
Solution. U (x) = BAx. Computing explicitly we get that U x = Cx where
 √ √   √ √ 
1 − 2 − 2 −1 0 1 2 −√ 2
C = BA = √ √ = √ .
2 2 − 2 0 1 2 − 2 − 2
U is the composition of T followed by S. 
(4) Find the matrix representing the linear transformation that first rotates as in (b), then reflects as in
(a), and then rotates backwards (i.e., clockwise by 135 degrees).
Solution. Rotating clockwise by 135 degrees is the inverse of the linear transformation S. Therefore,
it’s matrix is  √ √ 
1 − 2
B −1 = √ √2 .
2 − 2 − 2
The linear transformation that first rotates as in (b), then reflects as in (a), and then rotates
backwards is given by the matrix C = B −1 AB. 
(5) What matrix do you get if you repeat the sequence in part (d) ten times? Write this matrix in terms
of A and B. Can you write this matrix explicitly?
Solution. If you repeat the previous operation 10 times we have
(B −1 AB)(B −1 AB)(B −1 AB) · · · (10 times) = B −1 A10 B
which is equal to
 √ √   10  √ √ 
1 − 2 −1 0 −√ 2 −√2
√ √2
4 − 2 − 2 0 1 2 − 2
 √ √   √ √ 
1 − 2 2 1 0 −√ 2 −√2
= √ √ = I2 .
4 − 2 − 2 0 1 2 − 2
This makes sense because (B −1 AB)(B −1 AB) = I2 . Think this through geometrically to make sure
you agree. Thus if we repeat the transformation in part (d) an even number of times, we will just
come back to where we were at the beginning. 

5.4. Ch 3, #14. (after 3.3) Find all values of x for which the following matrix does not have an inverse.
 
1 0 3
−2 x 1
4 −1 2

Solution. Let us call this matrix “A”. By a theorem from lecture, A is invertible if and only if an echelon
form of A has a pivot in each row (since A is square, this is equivalent to having a pivot in each column).
Thus we convert A to echelon form:
       
1 0 3 R2 +2R1 →R2 1 0 3 1 0 3 1 0 3
−2 x 1 R3 −4R →R
→1 3 0 x
R +xR →R
7  2 →3 2 0 0 7 − 10x 2→ 3 0 −1
R ↔R
−10 
4 −1 2 0 −1 −10 0 −1 −10 0 0 7 − 10x
There is a pivot in the last row if and only if 7 − 10x 6= 0. Thus A does not have an inverse exactly when
x = 7/10. 

Solution. (Alternate solution) We compute the inverse of this matrix by appending an identity at the end
and doing Gaussian elimination.

   
1 0 3 1 0 0 1 0 3 1 0 0
−2 x 1 0 1 0 → 0 x 7 2 1 0
4 −1 2 0 0 1 0 −1 −10 −4 0 1
 CONCEPTUAL PROBLEMS SOLUTIONS, AUTUMN 2020 21

Switching the second and third rows, we have

     
1 0 3 1 0 0 1 0 3 1 0 0 1 0 3 1 0 0
0 −1 −10 −4 0 1 → 0 1 10 4 0 −1 → 0 1 10 4 0 −1
0 x 7 2 1 0 0 x 7 2 1 0 0 0 7 − 10x 2 − 4x 1 x
If 7 − 10x = 0, i.e., x = 7/10 then the original matrix cannot be inverted. Otherwise, we continue the
Gaussian elimination:
   
1 0 3 1 0 0 1 0 3 1 0 0
−10
0 1 10 4 0 −1  → 0 1 0 4 − 20−40x 7−10x 7−10x
10x 
−1 − 7−10x
2−4x 1 x 2−4x 1 x
0 0 1 7−10x 7−10x 7−10x 0 0 1 7−10x 7−10x 7−10x
6−12x 3 3x
1 0 0 1 − 7−10x − 7−10x − 7−10x
 
−10
→ 0 1 0 4 − 20−40x 7−10x 7−10x −1 10x 
− 7−10x
2−4x 1 x
0 0 1 7−10x 7−10x 7−10x
Therefore the inverse of the original matrix is
1 − 6−12x 3
− 7−10x 3x
− 7−10x
 
7−10x
4 − 20−40x −10 10x 
7−10x 7−10x −1 − 7−10x
2−4x 1 x
7−10x 7−10x 7−10x
7
This inverse exists as long as x 6= 10 . 

6. CP 6
6.1. Ch 4, #1. Let S be a plane in R3 passing through the origin, so that S is a two-dimensional subspace
of R3 . Say that a linear transformation T : R3 → R3 is a reflection about S if T (v) = v for any vector v in
S and T (n) = −n whenever n is perpendicular to S.

Shown: Solid arrows v ∈ S, w a generic vector, n perpendicular to S, and the reflections of each vector as
dashed arrows. (Note that T (v) = v and T (n) = −n.)

Let T be the linear transformation given by T (x) = Ax, where A is the matrix
 
−1 −2 2
1
−2 2 1 .
3
2 1 2
This linear transformation is the reflection about a plane S. Find a basis for S.
22 CONCEPTUAL PROBLEMS SOLUTIONS, AUTUMN 2020

Solution. From the setup, S is the set of vectors v satisfying the equation Av = v. We can write Av = Iv
as (A − I)v = 0 where I is the 3 × 3 identity matrix. This shows that for every vector v in S, (A − I)v = 0,
i.e. S is the null-space of the matrix

1 21 − 21
     
−4 −2 2 R1 −2R2 →R1 0 0 0
1 R3 +R2 →R3
A−I = −2 −1 1  → −2 −1 1 ∼ 0 0 0 
3
2 1 −1 0 0 0 0 0 0
From this reduced echelon form, we see that S is the set of solutions of 2x1 + x2 − x3 = 0. To obtain a basis
we solve the equation. For example, by setting x2 = s, x3 = t we find that
 1
− 2 s + 12 t
  1 1
−2 2
v= s  = s  1  + t 0 .
t 0 1
Thus, a basis for S = null(A − I) is the pair of vectors
 1   1 
 −2 2 
 1  , 0 .
0 1
 

6.2. Ch 4, #2. Define T : R3 → R3 by

    
x 0 3 −2 x
T y  = 1 −2 2  y  .
z 2 −6 5 z

Let V be the set of all vectors that are fixed by T , which means that V = v ∈ R3 : T (v) = v .

(a) Show, using the definition of subspace, that V is a subspace of R3 .

Solution. We check the three properties for a set in R3 to be a subspace to show that V is a
subspace. You will notice that we don’t need anything about the particular linear transformation
for these checks to go through. We will just need the property that V consists of all vectors that are
fixed by a linear transformation.
(a) Suppose v and w are in V. Then T (v) = v and T (w) = w. Therefore,
T (v + w) = T (v) + T (w) (since T is a linear transformation)
= v + w (by the definition of V)
Therefore, if v and w are in V then so is their sum.
(b) For c ∈ R and v ∈ V, T (cv) = cT (v) = cv. The first equality is since T is a linear transformation
and the second from the definition of V. Therefore, all scalar multiples of a vector in V is again
in V.
(c) T (0) = 0. Therefore, 0 is in V.

(b) Come up with
  an equation that also defines V. (In other words, find a linear equation ax+by +cz = d
x
such that y  ∈ V ⇔ ax + by + cz = d.)
z
Solution. This is similar to the previous exercise. T (v) = Av = v is the same as saying (A−I)v = 0.
Therefore V is the kernel of the linear transformation defined by A − I which is
   
−1 3 −2 −1 3 −2
 1 −3 2  ∼  0 0 0  .
2 −6 4 0 0 0
Therefore, V = {(x, y, z) : x − 3y + 2z = 0}. 
(c) Geometrically, what kind of object is V (point/line/plane etc)?
 CONCEPTUAL PROBLEMS SOLUTIONS, AUTUMN 2020 23

Solution. Since V is the solution set of a single equation, it is a plane in R3 . You can also see this by
computing a basis of V as we will do next and noting that this basis consists of 2 linearly independent
vectors. 
(d) Find a basis for V.
Solution. We need to solve x − 3y + 2z = 0. Setting y = s and z = t we see that the solutions are
all vectors of the form (3s − 2t, s, t) as s, t vary over R. A basis of V consists of {(3, 1, 0), (−2, 0, 1)}
obtained by taking s = 1, t = 0 and s = 0, t = 1. 

6.3. Ch 4, #4. Let V be the subspace of R4 defined as

V = {(w, x, y, z) ∈ R4 : w + x + y + z = 0}.
   
−1 0
0 −1
Check that the vectors 1

 0  can be part of a basis for V . Then expand the set consisting of
and  

0 1
these two vectors to a basis of V.
   
−1 0
0 −1
Solution 1. Let v1 =  1  and v2 =  0 . The vectors v1 , v2 are contained in V since (−1) + (0) + (1) +
  

0 1
(0) = 0 and (0) + (−1) + (0) + (1) = 0. Moreover, {v1 , v2 } is linearly independent (for example because they
are not scalar multiples of each other). Thus {v1 , v2 } can be part of a basis for V .
We first find a set of vectors w1 , . . . , wn that spans V , then remove
 some wi from the set {v1 , v2 , w1 , . . . , wn }
to get a basis for V . Note V = nul(A) where A = 1 1 1 1 ; thus we apply the procedure for finding a
basis for the null space of a matrix. We solve for the general solution to Ax = 0. The matrix A is already
in echelon form, and x2 , x3 , x4 are free variables and x1 = −x2 − x3 − x4 so the general solution is
         
x1 −x2 − x3 − x4 −1 −1 −1
x2   x 2
 1 0 0
 =  = x2   + x3   + x4   .
x3   x3  0 1 0
x4 x4 0 0 1
Let’s call these vectors w1 , w2 , w3 . We know that the set {v1 , v2 , w1 , w2 , w3 } spans V (since {w1 , w2 , w3 }
spans V ) so we can apply the procedure to find a linearly independent subset of {v1 , v2 , w1 , w2 , w3 } which
also spans V :
     
−1 0 −1 −1 −1 0 0 −1 0 −1
v1 v2 w1 w2 w3   0 −1 1 0 0 →R 0 −1 1 0 0
 =  R3 +R→1 1  
  1 0 0 1 0 1 0 0 1 0
0 1 0 0 1 0 1 0 0 1
     
0 0 −1 0 −1 0 0 0 0 0 1 0 0 1 0
0 0 1 0 1  R2 +R
R4 +R2 →R2  →R 0 0 1 0 1  switch 0 1 0 0 1
  
→ 1 0 0 1 0  →1 1  1 0 0 1 0 → 0 0 1 0 1
0 1 0 0 1 0 1 0 0 1 0 0 0 0 0
Here the pivots are columns 1,2,3 so we have that {v1 , v2 , w1 } is a basis for V . 

Solution 2. The subspace S is a 3-dimensional plane in R4 (the system of equations defining S is automati-
cally in echelon form since it only has 1 equation; we see right away that it has 3 free variables). So, a basis
for S must consist of three vectors.
Now check that the two given vectors lie in V . Check also that these two vectors are linearly independent.
Therefore we just need one more vector in S that isn’t in the span of the first two. In this example, we could
24 CONCEPTUAL PROBLEMS SOLUTIONS, AUTUMN 2020

try some simple things to get the solution. Like the vectors given, any vector in R4 with a 1 and −1 and
two 0s will lie in V . So let’s find a few more like that such as
   
w −1
x  1 
 = 
y   0 
z 0

Take the collection of three vectors and check that they are linearly independent:
   
−1 0 −1 −1 0 −1
 0 −1 1   0 −1 1 
 ∼ .
1 0 0 0 0 −1
0 1 0 0 0 0

Since every column of the echelon form has a leading term, the three original vectors are linearly independent.
Therefore (since we know S has dimension 3) they are a basis for S. 

6.4. Ch 4, #5. Find a 3 × 4 matrix A with nullity 2 and with column space
     
 1 4 3 
col(A) = Span  0  ,  −3  ,  −2  ,
1 7 5
 

or explain why such a matrix cannot exist.

Solution. By the rank-nullity theorem, A will have to have rank 2. The rank of the matrix is also the
dimension of col(A), so for that dimension to be 2, the three vectors must be linearly dependent and span a
2-dimensional plane. Since there are three vectors given, this seems unlikely to be possible. Let’s check:
     
1 4 3 1 4 3 1 4 3
0 −3 −2 ∼ 0 −3 −2 ∼ 0 −3 −2 .
1 7 5 0 3 2 0 0 0

This means that the first two vectors are linearly independent, and the third one is a linear combination of
the first two.
So, to find A, we should pick four columns in the subspace we’ve described (so that the resulting columns
still span this subspace and nothing more). We should make sure to include two linearly independent
columns, so that col(A) is not smaller than desired. One easy option is
 
1 4 3 0
A = 0 −3 −2 0 .
1 7 5 0

An even simpler one is

 
1 4 0 0
A = 0 −3 0 0 .
1 7 0 0
A more complicated solution would be to take the fourth column to be another linear combination of the
two basis vectors we found. For instance, their sum:
 
1 4 3 5
A = 0 −3 −2 −3 .
1 7 5 8


 CONCEPTUAL PROBLEMS SOLUTIONS, AUTUMN 2020 25

7. CP 7
7.1. Ch. 4, #7. Give an example of a linear transformation from T : R2 → R3 with the following two
properties:

(1) T is not one-to-one, and

(2)
  
 x 
range(T ) = y  ∈ R3 : x − y + 2z = 0 ;
z
 

or explain why this is not possible. If you give an example, you must include an explanation for why
your linear transformation has the desired properties.

Solution. The matrix A representing T (which means T (x) = Ax) must have size 3 × 2 since T : R2 → R3 .
If T is not one-to-one, A it must have a nullspace of dimension at least 1. Check that the set of solutions to
x − y + 2z = 0 is a 2-dimensional plane. So part (b) says that the column span of A is 2-dimensional, and
thus the rank of A must be 2. This means that the two columns of A must be linearly independent. But
then the nullspace of A can only have 0 in it. Therefore, no such A exists. 

7.2. Ch. 4, #8. Consider the following row equivalent matrices:

 
  3 0 −1 5  
A = a1 a2 a3 a4 ∼ 0 0 2 −1 = B = b1 b2 b3 b4
0 0 0 0
Let col(A) be the column space of A. Answer the following with reasons.

(1) Is {a1 , a3 } a basis for col(A)?

(2) Is {a1 + a2 , a3 + a4 } a basis for col(A)?
(3) Is {a1 − 31 a3 , a4 } a basis for col(A)?
(4) Is {a1 + a3 , a4 } a basis for col(A)?

Solution. The key to this exercise is to remember that if A is a matrix and B is obtained by doing row
operations on A, then the solutions to Ax = 0 and Bx = 0 are exactly the same. Also recall that Ax is the
linear combination of the columns of A by x1 , x2 , . . .. Together this means that if a linear combination of
the columns of A is zero, then the same linear combination of the corresponding columns of B is also zero.
In fact, a set of columns of A is linearly independent (dependent) if and only if the same columns in B are
linearly independent (dependent). Using all this let’s answer the questions.

(1) Is {a1 , a3 } a basis for col(A)?

Solution. The first and third columns in the second matrix are linearly independent. Therefore, the
first and third columns of the first matrix are also linearly independent. The rank of the second
matrix is 2 since it has two linearly independent rows (pivot rows). Therefore the rank of the first
matrix is also 2. This means that the dimension of the column space of the first matrix is 2 and so
{a1 , a3 } is a basis for col(A). 
(2) Is {a1 + a2 , a3 + a4 } a basis for col(A)?
Solution. We need to check if a1 + a2 and a3 + a4 are linearly independent. If they were linearly
dependent, there is are c, d ∈ R such that c(a1 + a2 ) + d(a3 + a4 ) = 0, or equivalently (c, c, d, d) lies
in the nullspace of A and hence also in the nullspace of B. This means that {b1 + b2 , b3 + b4 } are
linearly dependent. However, this is not true since b1 + b2 is not a multiple of b3 + b4 . 
(3) Is {a1 − 31 a3 , a4 } a basis for col(A)?
26 CONCEPTUAL PROBLEMS SOLUTIONS, AUTUMN 2020

Solution. As in the previous part, the answer is “yes” if and only if b1 − 31 b3 and b4 are linearly
independent. Check that b1 − 13 b3 = ( 10 2
3 , − 3 , 0) while b4 = (5, −1, 0). These two vectors are
1
multiples of each other and so b1 − 3 b3 and b4 are linearly dependent. Thus {a1 − 31 a3 , a4 } do not
form a basis for col(A). 
(4) Is {a1 + a3 , a4 } a basis for col(A)?
Solution. We have b1 + b3 = (2, 2, 0) and b4 = (5, −1, 0) which are not multiples of each other.
Therefore, {a1 + a3 , a4 } is a basis for col(A). 

7.3. Ch. 4, #9. Find, if possible, an example of a matrix A such that:

 
1
(1) A is 2 × 3, col(A) = R2 , and null(A) = Span 1.
1
Solution. col(A) = R2 means that rank(A) = 2, so nullity of A is 1 which is what is prescribed.
 
1 0 −1
A= .
0 1 −1

    
1 0
(2) A is 2 × 3, col(A) = R2 , and null(A) = Span −1 ,  1 .
0 −1
Solution. This is not possible since by the same argument as in the previous case, if rank(A) = 2
then nullity of A must be 1, but we are given that the nullspace is 2-dimensional. 
   
1 3
(3) A is 2 × 2, row(A) = Span , and null(A) = Span .
1 −3
Solution.  
1 1
A= .
2 2

    
1 3
(4) A is 3 × 2, row(A) = Span , and null(A) = Span .
1 −3
Solution.  
1 1
A = 2 2 .
3 3

    
1 3
(5) A is 2 × 2, row(A) = Span , and null(A) = Span .
2 1
Solution. This is not possible since if (1, 2) lies in the row span of A and (3, 1) in the nullspace of
A, then we must have that 1 × 3 + 2 × 1 = 0 which is not true. 
 
  1
1
(6) A is 2 × 3, col(A) = Span , and null(A) = Span 1.
2
1
 
1
Solution. This is also not possible since col(A) = Span means that the rank of A is 1, while
2
 
1
null(A) = Span 1 means that the nullity of A is also 1. This violates the rank-nullity theorem
1
which says that rank and nullity should add up to the number of columns of A which is 3. 
 CONCEPTUAL PROBLEMS SOLUTIONS, AUTUMN 2020 27

   
1 0 0 0
7.4. Ch. 4, #10. Let B = 0 1 0. Note that null(B) = span 0.
0 0 0 1

(1) Find if possible, a 3 × 3 matrix A, where BA has nullity 2. If you find an example, what is the
nullity of the matrix A that you found? Can you find an example with a different nullity?
   
1 0 0 1 0 0
Solution. If A = 1 0 0, then BA = 1 0 0 which has rank 1 and hence nullity 2. The
0 1 0 0 0 0
matrix A has rank 2 and hence nullity 1.
How does one go about finding such an A? If the nullity of BA is to become 2, then the rank of
BA must be 1. Currently B has rank 2. What do we get when we multiply B with the A we chose?
The first column of A says that the first column of BA is the sum of the first two columns of B. The
second column of BA is the third column of B and the last column of BA is the zero column. Thus
BA has only one nonzero column and its rank falls to 1.
We could have
 chosen other A’s. For instance,
 
1 0 0 1 0 0
A = 1 0 0 would also yield BA = 1 0 0. Now A has rank 1 and nullity 2, but BA has
0 0 0 0 0 0
nullity 2 as before.
There is a general lesson here. The rank of BA is at most the minimum of the rank of A and the
rank of B. Why is this? It could be strictly smaller than both like for the first A we chose where
the rank of BA fell to 1 even though both A and B had rank 2. In the second example, the rank of
BA equals the rank of A. 

(2) Find if possible, a 3 × 3 matrix A, where BA has nullity 1. If you find an example, what is the
nullity of the matrix A that you found? Can you find an example with a different nullity?

Solution. If BA must have nullity 1, then the rank of BA must be 2. Our best bet it so multiplying
B by a full rank matrix A such as the identity matrix. In this particular example, any full rank A
will work
 since if  
a b c a b c
A = d e f , then the product BA = d e f  which has rank 2, since the first and second
g h i 0 0 0
rows of A which are also the first and second rows of BA, are linearly independent. This also shows
us that as long as the first and second rows of A are linearly independent, BA would have rank 2
for this particular B. So we could have also chosen an A that has rank 2 with its first and second
rows linearly independent. 

(3) Find if possible, a 3 × 3 matrix A, where BA has nullity 0. If you find an example, what is the
nullity of the matrix A that you found? Can you find an example with a different nullity?

Solution. This is not possible since if the nullity of BA is to become 0 then the rank of BA must be
3. We already mentioned that rank can only stay the same or go down when a matrix is multiplied
by another matrix. 

8. CP 8
8.1. Ch 5, #2. Compute det(5A3 ) for the matrix A below.
28 CONCEPTUAL PROBLEMS SOLUTIONS, AUTUMN 2020

0 2 0 1 0 1 0 1 0 1

0 0 3 0 1 0 1 0 1 0
0 0 0 4 0 1 0 1 0 1
0 0 0 0 5 6 1 0 1 0
 
0 0 0 0 6 5 0 1 0 1
A=
0 0 0 0 0 0 4 0 1

0
0 0 0 0 0 0 0 3 0 1


0 0 0 0 0 0 0 0 2 0
 
0 0 0 0 0 0 0 0 0 1
1 0 1 0 1 0 1 0 1 0

Solution. Since 5A3 = (5I)(A3 ) we have that det(5A3 ) = det(5I) det(A3 ) = 510 (det(A))3 . (Alternatively,
multiplying a matrix by 5 is the same as multiplying every row by 5, and there are 10 rows; then consider
how row operations affect the determinant.) So it remains to compute det(A).
We do cofactor expansion down column 1:
∗
det(A) = a1,1 C1,1 + · · · + a10,1 C10,1 = a10,1 C10,1
where we have the equality ∗ since a1,1 = · · · = a9,1 = 0. We have a10,1 = 1 and
 
2 0 1 0 1 0 1 0 1
0 3 0 1 0 1 0 1 0
 
0 0 4 0 1 0 1 0 1
 
0 0 0 5 6 1 0 1 0
10+1
 
C10,1 = (−1) det(0 0 0 6 5 0 1 0 1 ) .
0 0 0 0 0 4 0 1 0
 
0 0 0 0 0 0 3 0 1
 
0 0 0 0 0 0 0 2 0
0 0 0 0 0 0 0 0 1
Note that this matrix is almost upper-triangular except for the 6 in the (5, 4) position. The result of the
row operation R5 -= 65 R4 (which does not change the determinant) is the matrix
 
2 0 1 0 1 0 1 0 1
0 3 0 1 0 1 0 1 0
 
0 0 4 0 1 0 1 0 1
 
0 0 0 5 6 1 0 1 0
 
0 0 0 0 − 11 − 6 1 − 6 1
 5 5 5 
0 0 0 0 0 4 0 1 0
 
0 0 0 0 0 0 3 0 1
 
0 0 0 0 0 0 0 2 0
0 0 0 0 0 0 0 0 1
whose determinant we can compute by multiplying the diagonal entries: (2)(3)(4)(5)(− 11 5 )(4)(3)(2)(1) =
−11 · (24)2 , thus det(5A3 ) = 510 (1 · (−1) · (−11 · (24)2 ))3 = 510 · 113 · 246 = 2483965440000000000. 

8.2. Ch 5, #3. This exercise uses the notion of cofactors of a matrix defined in Section 5.1. Recall that if
A is an n × n matrix, then Mij denotes the matrix obtained from A by deleting the ith row and jth column
of A.
 
−2 0 2
Let A =  1 1 1.
3 −1 5

(a) Compute all nine cofactors of A, as well as det(A). Let B be the 3×3 matrix containing the cofactors,
with each entry multiplied by the appropriate ± sign. So the ij-entry of B is (−1)i+j det(Mij ).
 CONCEPTUAL PROBLEMS SOLUTIONS, AUTUMN 2020 29

 
1 1
Solution. The first cofactor is (−1)1+1 det = 6, from deleting the first row and column.
−1 5
Computing all cofactors we get
 
6 −2 −4
B = −2 −16 −2
−2 4 −2

To compute det(A), I used a column operation, col1 + col3 → col3 :

   
−2 0 2 −2 0 0  
1 2
det  1 1 1 = det  1 1 2 = −2 det
−1 8
3 −1 5 3 −1 8

(by cofactor-expanding along the top row). This works out to det(A) = −20. 
T
(b) Compute AB . You should get a diagonal matrix with the same number in every diagonal entry, in
other words, a multiple of the identity matrix. What multiple is it (in terms of A)?
Solution. Computing AB T gives
     
−2 0 2 6 −2 −2 −20 0 0
1 1 1 · −2 −16 4  =  0 −20 0 .
3 −1 5 −4 −2 −2 0 0 −20

So, in terms of the identity matrix, this is −20 · I. This appears to be det(A) · I. 
(c) Fill in the blank (with a scalar) to make this equation true:
1
AB T = · I, therefore A−1 =
· BT .


?
?
1
Solution. The scalar is det(A), so AB T = det(A)I, or det(A) AB T = I. (Bear in mind that det(A) is
a number, not a matrix.) If we multiply on the left by A−1 , we get det(A)
1
B T = A−1 . 
(d) Compute the matrix B > when A is the 2 × 2 matrix
 
a b
A= ,
c d

and then use it to compute a formula for A−1 . Does this agree with the formula for A−1 given in
the textbook?
   
d −c d −b
Solution. In this case the matrix B = and hence B > = . The determinant of
−b a −c a
A is ad − bc by cofactor expansion. Therefore, by part (c), we get that
 
1 d −b
A−1 =
ad − bc −c a
which agrees with the formula for the inverse of a 2 × 2 matrix given in the book.
Comment. The formula A−1 = det(A) 1
B T is bad for computational purposes because it requires
computing n2 separate determinants (the entries of B), each of size (n − 1) × (n − 1). And computing
det(A) as well. Computing that many determinants by hand is error-prone and slow, compared to
just row reducing. 2 

2What about with computers? Computing a single determinant really efficiently takes roughly n3 steps, so this method
takes (n − 1)2 · n3 ≈ n5 steps. For a matrix with a million (106 ) rows and columns, this means 1030 steps. By contrast,
computing A−1 directly by row reduction takes roughly n3 steps, total. So, for a matrix with one million rows (106 ), around
1018 steps, which is doable (in 2017, a modern supercomputer can do around 1015 arithmetic operations per second, so it would
take 1000 seconds to do 1018 operations – about 15 minutes. For 1030 operations, it would take 1012 seconds, which is around
30000 years.)
30 CONCEPTUAL PROBLEMS SOLUTIONS, AUTUMN 2020

8.3. Ch 5, #8. Let

f (t) = x1 sin(t) + x2 cos(t) + x3 t sin(t) + x4 t cos(t)
and
f 0 (t) = y1 sin(t) + y2 cos(t) + y3 t sin(t) + y4 t cos(t),
where the xi and yi are coefficients. Find a matrix A expressing the vector y = (y1 , y2 , y3 , y4 ) as y = Ax
where x = (x1 , x2 , x3 , x4 ). What is the interpretation of A2 x and A3 x?

Solution. We first compute the derivative of f (t) to see that

f 0 (t) = (x1 + x4 ) cos(t) + (x3 − x2 ) sin(t) − x4 t sin(t) + x3 t cos(t).
Now comparing this to the f 0 (t) given to us we see that
y1 = x3 − x2 , y2 = x1 + x4 , y3 = −x4 , y4 = x3 .
Writing this with matrix notation we get
    
y1 0 −1 1 0 x1
y2  1 0 0 1 x2 
y= y3  = 0
     = Ax.
0 0 −1 x3 
y4 0 0 1 0 x4
| {z }
A
0
Since f (t) has the same form as f (t), we can calculate higher derivatives of f (t) by using the matrix A. For
example, if f 00 (t) = z1 sin(t) + z2 cos(t) + z3 t sin(t) + z4 t cos(t), then z = Ay = A2 x. Check this by actually
000
computing f 00 (t) and using the method we just used. Similarly, if f (t) = w1 sin(t) + w2 cos(t) + w3 t sin(t) +
w4 t cos(t), then w = Az = A2 y = A3 x. 

8.4. Ch 6, #2. (Practice showing something is a subspace) Suppose λ is an eigenvalue for the matrix A.
Consider the λ-eigenspace of A:
Eλ (A) = {v ∈ Rn : Av = λv},
the set of all vectors v satisfying the equation Av = λv. One reason why Eλ (A) is a subspace is because it
is the nullspace of A − λI. Show that Eλ (A) is a subspace by directly checking the three conditions needed
to be a subspace.

Solution. First, we have to check that 0 ∈ Eλ (A): this is true because A0 = 0 = λ0 (regardless of what the
eigenvalue λ is).

Next, suppose v ∈ Eλ (A) and u ∈ Eλ (A). This means that v, u are vectors such that Av = λv and
Au = λu. Then
A(v + u) = Av + Au = λv + λu = λ(v + u).
Since A(v + u) = λ(v + u), this means v + u ∈ Eλ (A).

Finally, suppose v ∈ Eλ (A) and r ∈ R is a scalar. So we know Av = λv. Then

A(rv) = r(Av) = r(λv) = λ(rv).
Since A(rv) = λ(rv), this means that rv ∈ Eλ (A). Therefore Eλ (A) is a subspace. 

9. CP 9
 
4 1
9.1. Ch 6, #3. Let A = .
3 2

(a) Compute the eigenvalues and eigenvectors of A.

 CONCEPTUAL PROBLEMS SOLUTIONS, AUTUMN 2020 31

Solution. The eigenvalues of A are the roots of the polynomial det(A − λI) = 0.
 
4−λ 1
det = λ2 − 6λ + 5 = (λ − 5)(λ − 1) = 0.
3 2−λ

Therefore, the eigenvalues are 1 and 5.  

3 1
To compute the eigenvectors of λ = 1, we need to find a basis for the nullspace of A − I = .
  3 1
−1
The eigenvector λ = 1 has eigenspace spanned by w = . Similarly, check that the eigenvalue
  3
1
λ = 5 has eigenspace spanned by the eigenvector v = . 
1
(b) If λ is an eigenvalue of A then there is a vector v such that Av = λv. Using this equation, show
that λ2 is an eigenvalue of A2 . What is an eigenvalue of A−1 ? Now compute all eigenvalues and
eigenvectors of A2 and A−1 .
Solution. Av = λv ⇒ A2 v = λAv = λ2 v. Therefore, λ2 is an eigenvalue of A2 with eigenvector
v. Similarly, Av = λv ⇒ A−1 Av = λA−1 v which means that A−1 v = λ1 v. Therefore, λ1 is an
eigenvalue of A−1 with eigenvector v.
Using this information we see that 1 and 25 are eigenvalues of A2 . Since A2 is a 2 × 2 matrix it
has no further eigenvalues. The eigenvectors for each eigenvalue are the same as for A. Similarly,
the eigenvalues of A−1 are 1 and 51 with same eigenvectors as A. 
(c) Find a matrix B that shares an eigenvector with A but has different eigenvalues.
   
a b 1
Solution. Suppose we try to find B = with eigenvector . Then we need
c d 1
      
a b 1 a+b 1
= =µ .
c d 1 c+d 1

This means that we need a + b =µ and c + d = µ. There are infinitely many solutions,
  but we could
3 0 1
for instance take µ = 3 and B = . Then A and B share the eigenvector but have different
0 3 1
eigenvalues. Check that B has only 3 as eigenvalue and the eigenspace of 3 is all of R2 . 
(d) Find an invertible matrix P and a diagonal matrix D so that A = P DP −1 . Then, compute A1000 .
Solution. To diagonalize A, we use the eigenvectors and values we found above. We’ll get
       −1
4 1 1 −1 5 0 1 −1
A= = · · .
3 2 1 3 0 1 1 3
 −1  
1 −1 3 1
Note that = 41
. We have Ak = P DP −1 P DP −1 · · · P DP −1 where the right
1 3 −1 1
hand side is the product of P DP −1 with itself repeated k times. Therefore, Ak = P Dk P −1 . In our
case,
1 1 + 3(5k ) −1 + 5k
   k     
1 −1 5 0 1 3 1
Ak = · ·4 = .
1 3 0 1 −1 1 4 −3 + 3(5k ) 3 + 5k

This is valid for any k, including k = 1000. 

(e) Suppose v is an eigenvector of an arbitrary matrix M , with eigenvalue λ. Show (using matrix
algebra) that v is also an eigenvector of M + I, but with a different eigenvalue. What eigenvalue is
it?
Solution. Proceeding algebraically: (M +I)v = M v+Iv = λv+v = (λ+1)v. So, v is an eigenvector
of M + I with eigenvalue λ + 1. 
32 CONCEPTUAL PROBLEMS SOLUTIONS, AUTUMN 2020

   
1 0
9.2. Ch 6, #6. Find a 3 × 3 matrix A with eigenvectors v1 = 2 with λ = 1, v2 = −1 with λ = 2 and
3 1
 
1
v3 = 1 with λ = 10.
1
(Hint: A must be diagonalizable, A = P DP −1 . Figure out P and D, then compute A directly.)

Solution. Since we’re given the eigenvectors and eigenvalues, we can just set
   
1 0 1 1
P = 2 −1 1 , D= 2 
3 1 1 10
and set A = P DP −1 . After some computation, this works out to:
 
16 −3 −3
 
44
A= 3 − 43 − 10 .
3 
46 11 5
3 −3 −3


9.3. Ch 6, #7. Here is the graph of the characteristic polynomial of a 5 × 5 matrix A:

You may say “not enough information” for any of the following. Justify your answers.
(a) What polynomial has the graph shown above?
Solution. We observe that χA (λ) intersects the x-axis at x = −2, 0, 1, 2, and furthermore it is tangent
to the x-axis at x = 2, hence we have
am(A, −2) ≥ 1 , am(A, 0) ≥ 1 , am(A, 1) ≥ 1 , am(A, 2) ≥ 2
which means (λ−(−2))(λ−0)(λ−1)(λ−2)2 divides χA (λ); here both are polynomials of degree 5 (the
former since 1+1+1+2 = 5 and the latter since in general the degree of the characteristic polynomial
of an n × n matrix is n), hence actually χA (λ) is a scalar multiple of (λ − (−2))(λ − 0)(λ − 1)(λ − 2)2
which implies actually
am(A, −2) = 1 , am(A, 0) = 1 , am(A, 1) = 1 , am(A, 2) = 2 .
The coefficient is in fact −1,3 so that
χA (λ) = −(λ − (−2))(λ − 0)(λ − 1)(λ − 2)2 .

(b) What are the eigenvalues of A?
Solution. The eigenvalues of A are the roots of its characteristic polynomial, namely −2, 0, 1, 2. 
3I won’t prove this, but it has to do with the fact that the leading coefficient of χ (λ) = det(A − λI ) is obtained by
A n
multiplying the coefficients of λ along the main diagonal, which are all −1.
 CONCEPTUAL PROBLEMS SOLUTIONS, AUTUMN 2020 33

(c) Compute nullity(A + 2I).

Solution. Note that nul(A + 2I5 ) = nul(A − (−2)I5 ) is the eigenspace of λ = −2, hence nullity(A +
2I) = gm(A, −2). We have gm(A, λ) ≥ 1 and gm(A, λ) ≤ am(A, λ) for any eigenvalue λ, and in our
case am(A, −2) = 1 by part (a). The chain of (in)equalities 1 ≤ gm(A, −2) ≤ am(A, −2) = 1 implies
that gm(A, −2) = 1. 
(d) What can you say about the value of rank(A − 2I)?
Solution. The nullity of A − 2I is the dimension of the eigenspace of eigenvalue 2. Since λ = 2 is an
eigenvalue of A, we know gm(A, 2) ≥ 1. We have gm(A, λ) ≤ am(A, λ) for any eigenvalue λ, and we
have am(A, 2) = 1 by looking at the factorization of χA (λ) in (a). Thus gm(A, 2) is either 1 or 2.
Since gm(A, 2) = dim(nul(A − 2I5 )), by the rank-nullity theorem, rank(A − 2I) is either 5 − 2 = 3
or 5 − 1 = 4. (Both of these values are possible, see the solution to part (h). 
(e) Compute det(A − λI).
Solution. This is the characteristic polynomial of A, which was computed in part (a) to be −(λ +
2)λ(λ − 1)(λ − 2)2 . 
(f) What can you say about det(A + I)?
Solution. We have det(A + I) = det(A − (−1)I) = χA (−1) = −(−1 + 2)(−1 − 0)(−1 − 1)(−1 − 2)2 =
−18.4 
(g) Do the columns of A span R5 ?
Solution. Since 0 is an eigenvalue of A, we know that nul(A) = nul(A − 0I5 ) 6= {0}. Thus the
columns of A do not span R5 . 
(h) Is A diagonalizable?
Solution. We have that A is diagonalizable if and only if all the roots of χA (λ) are real (which is true
in our case) and am(A, λ) = gm(A, λ) for all λ. From (a), we know that am(A, λ) = gm(A, λ) = 1
for λ = −2, 0, 1 but all we can say about gm(A, 2) is that it is either 1 or 2 (which is am(A, 2)). Here
are examples showing that both are possible:
   
−2 −2

 0  
  0 


 1 
 
 1 

 2 1   2 
2 2

   
−4 3 0 1
9.4. Ch 5, #9. Let A = and B = .
−1 1 −1 3

(1) Compute det(A) and det(B).

Solution. det(A) = −1 and det(B) = 1. 
(2) Using facts about determinants and without computing AB, argue that the columns of AB form a
basis of R2 .
Solution. Since det(AB) = det(A) det(B) = −1, we know that the columns of AB are linearly
independent. Since there are two columns in AB, these columns must be a basis for R2 . 
(3) Compute A−1 and B −1 .
   
−1 3 3 −1
Solution. A−1 = and B −1 = . 
−1 4 1 0
(4) Compute the coordinates of the point (1, 1) in the basis formed by the columns of AB. Express your
computation abstractly in terms of A and B before you use the specific A and B given above.

4We have that λ is an eigenvalue of A if and only if det(A − λI ) = 0. Since −1 is not an eigenvalue of A, we have
n
det(A + I) = det(A − (−1)I) 6= 0.
34 CONCEPTUAL PROBLEMS SOLUTIONS, AUTUMN 2020

Solution. We are looking for (x, y) such that

   
x 1
AB = .
y 1
This means that
              
x −1 1 −1 −1 1 3 −1 −1 3 1 −2 5 1 3
= (AB) =B A = = = .
y 1 1 1 0 −1 4 1 −1 3 1 2

(5) What is the linear transformation T that will undo the change of basis in (d) bringing your answer
in (d) back to (1, 1)? Write T with domain, codomain and matrix. Express your computation
abstractly in terms of A and B before you use the specific A and B given above. Check that it
works.    
x x
Solution. This map must be T : R2 → R2 such that T = AB . Let’s check if this will
y y
   
3 1
send back to .
2 1
          
3 −4 3 0 1 3 −3 5 3 1
AB = = = .
2 −1 1 −1 3 2 −1 2 2 1