MATH10202 Linear Algebra A 2019-20

Exercise Sheet 5 (with solutions)

These exercises should be done during Week 5. Please hand in your solutions to Exercises 5.1
5.6(ii), 5.10(i) and 5.12 in advance of the Week 6 Supervision Class, as directed by your
group leader, and be prepared to discuss the other exercises in class.

5.1. Which of the following subsets U of R3 are subspaces of R3 ?

(i) {(a, 0, 0) | a ∈ R} (ii) {(a, 1, 1) | a ∈ R}
(iii) {(a, b, c) | a, b, c ∈ R, b = a + c} (iv) {(a, b, c) | a, b, c ∈ Z}

Solution to 5.1. Recall the definition of a subspace from Section 5.4.

(i) U is not empty, because for example (0, 0, 0) ∈ U . If u, v ∈ U and λ ∈ R then we can write
u = (a, 0, 0) and v = (b, 0, 0) for some a, b ∈ R whereupon

u + v = (a, 0, 0) + (b, 0, 0) = (a + b, 0, 0) ∈ U and λu = λ(a, 0, 0) = (λa, 0, 0) ∈ U.

Thus, U satisfies all the conditions to be a subspace of R3 .

(ii) (1, 1, 1) ∈ U but (1, 1, 1) + (1, 1, 1) = (2, 2, 2) ∈

/ U , so U doesn’t satisfy the conditions and is not
a subspace of R .3

(iii) U is non-empty, since it contains (0, 0, 0). Suppose u, v ∈ U and λ ∈ R. Then u = (a, b, c) and
v = (a0 , b0 , c0 ) where b = a + c and b0 = a0 + c0 . Now

u + v = (a, b, c) + (a0 , b0 , c0 ) = (a + a0 , b + b0 , c + c0 )

where b + b0 = (a + c) + (a0 + c0 ) = (a + a0 ) + (c + c0 ), so that u + v ∈ U . Similarly,

λu = λ(a, b, c) = (λa, λb, λc)

where λb = λ(a + c) = (λa) + (λc) so that λu ∈ U . Thus, U is a subspace of R3 .

(iv) (1, 1, 1) ∈ U but 21 (1, 1, 1) = ( 21 , 21 , 12 ) ∈

/ U , so U cannot be a subspace.

5.2. Let A be an n × n matrix over a field K and λ ∈ K. Define

Vλ = {v | v is a column n-vector and Av = λv}.

Prove that Vλ is a subspace of K n . [See the solution for some remarks, even if you can do this one.]

Solution to 5.2. One perfectly good approach is to show that Vλ satisfies the conditions in the
definition of a subspace (see Section 5.4). Alternatively, notice that λv = λ(In v) = (λIn )v, so we have

Av = λv ⇐⇒ Av = (λIn )v ⇐⇒ Av − (λIn )v = 0n×1 ⇐⇒ (A − λIn )v = 0n×1 .

This means that Vλ is the nullspace (see Section 5.4) of the matrix A − λIn , and it follows from
Theorem 5.4 that Vλ is a subspace of K n .

Remark. If λ happens to be an eigenvalue of A then Vλ is the set comprising all the eigenvectors of
A corresponding to λ, plus the zero vector (which by definition is not a eigenvector but still satisfies
the equation Av = λv). We call Vλ the eigenspace of A corresponding to λ. (What is the space Vλ
when λ is not an eigenvalue?)

5.3 (More Challenging). Let S = {w1 , . . . , wt } ⊆ Rn . Recall that by Theorem 5.3, span(S) is a
subspace of Rn . Let
κ = {T ⊆ Rn | T is a subspace of Rn and S ⊆ T }.
and let \
κ = {v ∈ Rn | v ∈ T for all T ∈ κ},
T
be the intersection of all sets in κ. Prove that κ = span(S).

Solution to 5.3. First

T notice that span(S) is Titself a subspace of Rn containing S, so span(S) ∈ κ.
This means that v ∈ κ =⇒ v ∈ span(S), so κ ⊆ span(S).
Conversely, suppose v ∈ span(S). By definition this means that
v = λ1 w1 + · · · + λt wt
for some λ1 , . . . , λt ∈ R. Now for any T ∈ κ we have
w1 , . . . , w t ∈ S ⊆ T
and since T is aTsubspace, it follows thatTv = λ1 w1 + · · · + λt wt ∈ T . Thus, v ∈ T for all T ∈ κ, which
means that v ∈ κ. Hence, span(S) T ⊆ Tκ. T
We have shown that span(S) ⊆ κ and κ ⊆ span(S), so it follows that κ = span(S).

Remark. Question 5.3 shows that span(S) is, in a certain sense, the “smallest” subspace of Rn which
contains the set S as a subset. Intuitively, this can sometimes be a more helpful way to think about
the span than just using the formal definition.

5.4. Which of the following sets of vectors (in R2 or R3 as appropriate) are linearly independent, and
which are linearly dependent?
(i) {(2, 1), (3, 0)} (ii) {(3, 9), (−4, −12)} (iii) {(1, −1, 2), (1, 1, 2), (0, 0, 1)}
(iv) {(1, −1, 2), (−1, 1, 2), (0, 0, 1)} (v) {(−1, 3, 4), (0, 0, 0), (2, 1, 4)}

Solution to 5.4. (i) and (iii) are linearly independent; (ii), (iv) and (v) are linearly dependent.

5.5 (Discussion Point). Suppose 0n ∈ S ⊆ Rn . Can S be linearly independent?

Solution to 5.5. No, it can’t. If S = {0n , w1 , . . . , wt } then we can write (for example)
0n = 670n + 0w1 + · · · + 0wt
as a linear combination of elements from S where not all of the coefficients are 0, so S must be linearly
dependent.

5.6. Determine whether or not each of the following sets of vectors is a basis for the given space:
(i) {(−1, 1), (1, 2)} for R2 (ii) {(−1, 3, 4), (1, 5, −1), (1, 13, 2)} for R3
(iii) {(2, 1, 0, 2), (2, −3, 1, 0), (3, 2, 0, 0), (5, 0, 0, 0)} for R4

Solution to 5.6. In each case the size of the given set is the dimension of the containing space, so
Lemma 5.7 tells us that it will be a basis if (and only if ) it is linearly independent. So we just need to
check linear independence.
 (i) The vectors will be linearly independent exactly if the only solution to

02 = λ1 (−1, 1) + λ2 (1, 2)

is λ1 = λ2 = 0. Unpacking this vector equation as a (homogeneous) SLE it becomes

−λ1 + λ2 = 0, λ1 + 2λ2 = 0.

Solving (e.g. by Gaussian elimination) we see that λ1 = λ2 = 0 is indeed the only solution, so
the set is linearly independent, and hence is a basis for R2 .

(ii) Similarly, the question is whether

03 = λ1 (−1, 3, 4) + λ2 (1, 5, −1) + λ3 (1, 13, 2)

has a non-trivial solution. Turning this vector equation into an SLE and then into an augmented
matrix we get:  
−1 1 1 0
 3 5 13 0  .
4 −1 2 0
Notice how the columns of this matrix just come from the original vectors: can you see why this
has happened? Now reducing to row echelon form we get
 
1 −1 −1 0
A= 0 1 2 0 .
0 0 0 0

We could proceed to describe the solution set, but it should be obvious from the form of this
matrix that the solution set will have a parameter, and hence will be infinite. So λ1 = λ2 = λ3
cannot be the only solution, which means the set is linearly dependent and cannot be a basis for
R3 (or for anything else, for that matter!)

(iii) This time we end up with the matrix

 
2 2 3 5 0
 1 −3 2 0 0 
 .
 0 1 0 0 0 
2 0 0 0 0

Reducing to row echelon form we find that there is a unique solution, that is, a unique way of
writing 04 as a linear combination of the given vectors, so the set is linearly independent and is
a basis for R4 .

5.7. Let u1 , . . . , uk ∈ K n where K is a field. Prove that the following statements are equivalent:
(i) {u1 , . . . , uk } is a linearly dependent set;

(ii) there exists i ∈ {1, . . . , k} such that ui is a linear combination of the

other vectors u1 , . . . , ui−1 , ui+1 , . . . , uk .

Solution to 5.7. Suppose (i) holds. By the definition of linear dependence (see Section 5.5), there
exist scalars λ1 , . . . , λk such that

0 = λ1 u1 + λ2 u2 + · · · + λk uk
where the λi s are not all 0. Choose i such that λi 6= 0. Then scaling both sides by − λ1i and adding ui
to both sides, we obtain
       
λ1 λi−1 λi+1 λk
ui = − u1 + · · · + − ui−1 + − ui+1 + · · · + − uk
λi λi λi λi
which mean that (ii) holds.
Conversely, if (ii) holds then there is an i such that we can write

ui = ρ1 u1 + · · · + ρi−1 ui−1 + ρi+1 ui+1 + · · · + ρk uk

for some scalars ρ1 , . . . , ρi−1 , ρi+1 , . . . , ρk ∈ R. Now subtracting ui from both sides we have

0 = ρ1 u1 + · · · + ρi−1 ui−1 + (−1)ui + ρi+1 ui+1 + · · · + ρk uk .

Setting λi = −1 (so in particular λi 6= 0) and λj = ρj for j 6= i, we see that the vectors u1 , . . . , uk are
linearly dependent, so (i) holds.
   
5.8. Let u = √1 , √1 , √1 , v = (1, −1, 0), w = 0, √12 , − √12 . Calculate:
3 3 3

(i) hu | vi (ii) hu | wi (iii) hv | wi (iv) hw | vi

(v) ||u|| (vi) ||v|| (vii) ||w||
Which of the vectors u, v, w are unit vectors? Which subsets of {u, v, w} are orthogonal? Which are
orthonormal?

Solution to 5.8. We have

1
hu | vi = hu | wi = 0 and hv | wi = hw | vi = − √
2
(Notice that we must have hv | wi = hw | vi so if you worked this out twice there was no need!) Also
√
||u|| = ||w|| = 1 but ||v|| = 2.

So u and w are unit vectors but v is not.

The subsets {u, v} and {u, w} are orthogonal. Being pedantic (as a mathematician should
be!), so are {u}, {v}, {w} and indeed the empty set ∅. (If you read the definition of an orthogonal
set carefully, you’ll see that a set with 0 or 1 elements is always orthogonal — those flying pigs are at
work again!).
The orthonormal subsets are those which are orthogonal and don’t contain v (the only non-unit
vector), so {u, w}, {u}, {w} and ∅.

5.9. Verify that the vectors

   
3 4 4 3
v1 = − , , 0 , v2 = , ,0 and v3 = (0, 0, 1)
5 5 5 5

form an orthonormal basis for R3 . Now express each of the following as linear combinations of v1 , v2
and v3 :
1 3 5


(i) (1, −1, 2) (ii) (3, −7, 4) (iii) 7 , − ,
7 7

Solution to 5.9. Simple calculation shows that hv1 | v1 i = hv2 | v2 i = hv3 | v3 i = 1, so that all are
unit vectors. Similarly, hv1 | v2 i = hv1 | v3 i = hv2 | v3 i = 0 so {v1 , v2 , v3 } is an orthogonal, and hence
orthonormal, set. Since dim R3 = 3, by Lemma 5.7 this is enough to conclude that {v1 , v2 , v3 } is an
orthonormal basis for R3 .
Lemma 6.5 makes it easy to write any vector as a linear combination of the elements of an
orthonormal basis. Applying it we get

(i) (1, −1, 2) = − 75 v1 + 51 v2 + 2v3 ;

(ii) (3, −7, 4) = − 37 9

5 v1 − 5 v2 + 4v3 ;

(iii) 71 , − 37 , 57 = − 73 v1 − 71 v2 + 75 v3 ;

5.10. Use the Gram-Schmidt process to transform each of the following bases into an orthonormal
basis for R3 :
(i) u1 = (1, 0, 1), u2 = (1, 0, −1), u3 = (0, 3, 4)
(ii) u1 = (1, 0, 1), u2 = (0, 1, 2), u3 = (2, 1, 0)

Solution to 5.10.

(i) We set v1 = u1 = (1, 0, 1), then

hu2 | v1 i 0
v2 = u2 − v1 = (1, 0, −1) − (1, 0, 1) = (1, 0, −1)
hv1 | v1 i 2

Notice how the fact u2 is already orthogonal to v1 means we end up with v2 = u2 : there is no
need for it to be anything different! Next,

hu3 | v2 i hu3 | v1 i −4 4
v3 = u3 − v2 − v1 = (0, 3, 4) − (1, 0, −1) − (1, 0, 1) = (0, 3, 0).
hv2 | v2 i hv1 | v1 i 2 2

Now {v1 , v2 , v3 } is an orthogonal basis, but not yet orthonormal because the vectors aren’t unit
vectors. To turn it into an orthonormal basis we divide each vector out by its norm to get:
    
1 1 1 1
√ , 0, √ , √ , 0, − √ , (0, 1, 0)
2 2 2 2

(ii) The same procedure as above gives an orthonormal basis

     
1 1 1 1 1 1 2 1
√ , 0, √ , −√ , √ , √ , √ , √ , −√
2 2 3 3 3 6 6 6

5.11. Find an orthonormal basis for the subspace of R4 spanned by the vectors

(1, 0, 1, 0), (1, 1, 1, 0) and (1, −1, 0, 1).

What is the dimension of this subspace?

Solution to 5.11. Let u1 , u2 , u3 be the three vectors given, and U = span({u1 , u2 , u3 }) ⊆ R4 . To

find an orthonormal basis for U we would like to use the Gram-Schmidt Process, but the process has
to start from a basis for U . So our first question is: is {u1 , u2 , u3 } a basis for U ? By definition the
vectors span U , so we need only check that they are linearly independent: it turns out they are.
Now applying Gram-Schmidt just as in the previous question gives an orthonormal basis:
   
1 1 1 1 2
√ , 0, √ , 0 , (0, 1, 0, 0) , √ , 0, − √ , √
2 2 6 6 6
The dimension is the size of the basis, which is 3.

5.12. Let A and B be n × n matrices (over a field) such that AB = In . Prove that A is invertible,
and deduce that B = A−1 . [Hint: consider determinants, and use Theorems 3.4 and 4.13. And if you
think what you’re being asked to prove is completely obvious, go back to Section 1.6 and read the
definition of an invertible matrix very carefully!]

Solution to 5.12. The claim is not immediately obvious because, while we are told that AB = In , for
B to satisfy the definition of an inverse of A we also need BA = In . In fact the latter does follow from
the former, but it is surprisingly hard to deduce without the “heavy machinery” of some theorems!
By Theorem 4.13 we have

det(A) det(B) = det(AB) = det(In ) = 1

so in particular det(A) 6= 0. By Theorem 3.4 this means A is invertible. So A has an inverse A−1 ,
and multiplying the known equation AB = In on the left by A−1 gives A−1 AB = A−1 , so In B = A−1 ,
so B = A−1 as required. (And finally, by this convoluted route, we get BA = A−1 A = In .)

5.13 (More Challenging). Recall that if z = a + ib is a complex number then the complex
conjugate of z is z = a − ib. For u = (a1 , . . . an ), v = (b1 , . . . , bn ) ∈ Cn , define a product operation by

hu | vi = a1 b1 + a2 b2 + · · · + an bn .

Show that although hu | vi could be a complex number, hv | vi will always be real.

Now can you generalise Proposition 6.1 to this situation, that is, come up with an analogue
of the proposition which holds here? [Hint: for each part of the proposition, does the proof still work
for this new operation? If not, can you modify it to work? If not, can you modify the statement to
get something which does hold? You might want to use the facts that z + t = z + t and zt = zt for all
z, t ∈ C.]

Solution to 5.13. If z ∈ C, say z = a + bi, then

zz = (a + bi)(a − bi) = a2 + abi − abi − i2 b2 = a2 + b2

which is real. Hence,

hv | vi = v1 v1 + · · · + vn vn
is a sum of real numbers, and so is real.
In particular, this means it makes sense to talk about whether hv | vi ≥ 0. Notice also that
zz = a2 + b2 is non-negative, and is zero if and only if a = b = 0, that is, if and only if z = 0. It now
follows easy that hv | vi ≥ 0, and hv | vi = 0 if and only if v = 0, so that part (iv) of the proposition
still holds here.
Part (ii) holds by a very similar argument to that in the notes.
Part (iii) holds provided λ is a real number, but it would be somehow nicer if we could allow
λ to be a complex number. In this case we still have hλu | vi = λhu | vi by the same argument as in
the notes, but something different happens when the λ is on the other side of the inner product:

hu | λvi = a1 λb1 + · · · + an λbn = λ(a1 b1 + · · · + an bn ) = λhu | vi.

 Part (i) also doesn’t quite work. In general for complex numbers z, t ∈ C, there is no reason
why zt and tz should be equal, so for u, v ∈ Cn there’s no reason why hu | vi should equal hv | ui. So
what should we do? The key thing to spot is that there is a relationship between zt and tz: they are
complex conjugates, in other words, zt = tz (exercise: check this!). From this it is easy to deduce
(exercise again!) that for u, v ∈ Cn we have hu | vi = hv | ui, and it is natural to use this fact in
place of part (i) of the proposition. Putting it all together we have shown:

Proposition. Let u, v ∈ Cn and λ ∈ C. Then

(i) hu | vi = hv | ui;

(ii) hu + v | wi = hu | wi + hv | wi;

(iii) hλu | vi = λhu | vi and hu | λvi = λhu | vi;

(iv) hv | vi ∈ R, hv | vi ≥ 0, and hv | vi = 0 ⇐⇒ v = 0n .

A scalar product of complex vectors with these properties is called a Hermitian product operation.

5.14 (More Challenging). Let A be a symmetric square matrix over R. Suppose λ and µ are
distinct (that is, λ 6= µ) eigenvalues of A with corresponding eigenvectors u and v respectively. Prove
that u and v are orthogonal. [Hint: remember that for column vectors x and y we have hx | yi = xT y,
and consider the relationship between hAu | vi and hu | Avi.]

Solution to 5.14. We have

hAu | vi = (Au)T v = uT AT v = uT Av = uT (Av) = hu | Avi.

But we know Au = λu and Av = µv, so by the basic properties of the inner product this means
hAu | vi = hλu | vi = λhu | vi and hu | Avi = hu | µvi = µhu | vi. Thus, λhu | vi = µhu | vi, which
since λ 6= µ can only mean that hu | vi = 0.