4 Images, Kernels, and Subspaces

In our study of linear transformations we’ve examined some of the conditions under which a
transformation is invertible. Now we’re ready to investigate some ideas similar to invertibility.
Namely, we would like to measure the ways in which a transformation that is not invertible
fails to have an inverse.

4.1 The Image and Kernel of a Linear Transformation

Definition. The image of a function consists of all the values the function assumes. If
f : X → Y is a function from X to Y , then

im(f ) = {f (x) : x ∈ X}.

Notice that im(f ) is a subset of Y .

Definition. The kernel of a function whose range is Rn consists of all the values in its
domain at which the function assumes the value 0. If f : X → Rn is a function from X to
Rn , then
ker(f ) = {x ∈ X : f (x) = 0}.
Notice that ker(f ) is a subset of X. Also, if T (x) = Ax is a linear transformation from Rm
to Rn , then ker(T ) (also denoted ker(A)) is the set of solutions to the equation Ax = 0.

The kernel gives us some new ways to characterize invertible matrices.

Theorem 1. Let A be an n × n matrix. Then the following statements are equivalent.
1. A is invertible.

2. The linear system Ax = b has a unique solution x for every b ∈ Rn .

3. rref(A) = In .

4. rank(A) = n.

5. im(A) = Rn .

6. ker(A) = {0}.
Example 13. (§3.1, Exercise 39 of [1]) Consider an n × p matrix A and a p × m matrix B.
(a) What is the relationship between ker(AB) and ker(B)? Are they always equal? Is one
of them always contained in the other?

(b) What is the relationship between im(A) and im(AB)?

(Solution)

22
 (a) Recall that ker(AB) is the set of vectors x ∈ Rm for which ABx = 0, and similarly
that ker(B) is the set of vectors x ∈ Rm for which Bx = 0. Now if x is in ker(B), then
Bx = 0, so ABx = 0. This means that x is in ker(AB), so we see that ker(B) must
always be contained in ker(AB). On the other hand, ker(AB) might not be a subset
of ker(B). For instance, suppose that
   
0 0 1 0
A= and B = .
0 0 0 1

Then B is the identity matrix, so ker(B) = {0}. But every vector has image zero under
AB, so ker(AB) = R2 . Certainly ker(B) does not contain ker(AB) in this case.

(b) Suppose y is in the image of AB. Then y = ABx for some x ∈ Rm . That is,

y = ABx = A(Bx),

so y is the image of Bx under multiplication by A, and is thus in the image of A. So

im(A) contains im(AB). On the other hand, consider
   
1 0 0 0
A= and B = .
0 0 0 1

Then 

0 0
AB = ,
0 0
 
1
so im(AB) = {0}, but the image of A is the span of the vector . So im(AB) does
0
not necessarily contain im(A).
♦

Example 14. (§3.1, Exercise 48 of [1]) Consider a 2 × 2 matrix A with A2 = A.

(a) If w is in the image of A, what is the relationship between w and Aw?

(b) What can you say about A if rank(A) = 2? What if rank(A) = 0?

(c) If rank(A) = 1, show that the linear transformation T (x) = Ax is the projection onto
im(A) along ker(A).

(Solution)
(a) If w is in the image of A, then w = Av for some v ∈ R2 . Then

Aw = A(Av) = A2 v = Av = w,

since A2 = A. So Aw = w.

23
 (b) If rank(A) = 2, then A is invertible. Since A2 = A, we see that

A = I2 A = (A−1 A)A = A−1 A2 = A−1 A = I2 .

So the only rank 2 2 × 2 matrix with the property that A2 = A is the identity matrix.
On the other hand, if rank(A) = 0 then A must be the zero matrix.

(c) If rank(A) = 1, then A is not invertible, so ker(A) 6= {0}. But we also know that A
is not the zero matrix, so ker(A) 6= R2 . We conclude that ker(A) must be a line in
R2 . Next, suppose we have w ∈ ker(A) ∩ im(A). Then Aw = 0 and, according to part
(a), Aw = w. So w is the zero vector, meaning that ker(A) ∩ im(A) = {0}. Since
im(A) is neither 0 nor all of R2 , it also must be a line in R2 . So ker(A) and im(A)
are non-parallel lines in R2 . Now choose x ∈ R2 and let w = x − Ax. Notice that
Aw = Ax − A2 x = 0, so w ∈ ker(A). Then we may write x as the sum of an element
of im(A) and an element of ker(A):

x = Ax + w.

According to Exercise 2.2.33, the map T (x) = Ax is then the projection onto im(A)
along ker(A).
♦

Example 15. (§3.1, Exercise 50 of [1]) Consider a square matrix A with ker(A2 ) = ker(A3 ).
Is ker(A3 ) = ker(A4 )? Justify your answer.

(Solution) Suppose x ∈ ker(A3 ). Then A3 x = 0, so

A4 x = A(A3 x) = A0 = 0,

meaning that x ∈ ker(A4 ). So ker(A3 ) is contained in ker(A4 ). On the other hand, suppose
x ∈ ker(A4 ). Then A4 x = 0, so A3 (Ax) = 0. This means that Ax is in the kernel of A3 , and
thus in ker(A2 ). So
A3 x = A2 (Ax) = 0,
meaning that x ∈ ker(A3 ). So ker(A4 ) is contained in ker(A3 ). Since each set contains the
other, the two are equal: ker(A3 ) = ker(A4 ). ♦

4.2 Subspaces
Definition. A subset W of the vector space Rn is called a subspace of Rn if it
(i) contains the zero vector;

(ii) is closed under vector addition;

(iii) is closed under scalar multiplication.

24
 One important observation we can immediately make is that for any n × m matrix A,
ker(A) is a subspace of Rm and im(A) is a subspace of Rn .

Definition. Suppose we have vectors v1 , . . . , vm in Rn . We say that a vector vi is re-

dundant if vi is a linear combination of the preceding vectors v1 , . . . , vi−1 . We say that
the set of vectors v1 , . . . , vm is linearly independent if none of them is redundant, and
linearly dependent otherwise. If the vectors v1 , . . . , vm are linearly independent and span
a subspace V of Rn , we say that v1 , . . . , vm form a basis of V .

Example 16. (§3.2, Exercise 26 of [1]) Find a redundant column vector of the following
matrix and write it as a linear combination of the preceding columns. Use this representation
to write a nontrivial relation among the columns, and thus find a nonzero vector in the kernel
of A.  
1 3 6
A = 1 2 5 .
1 1 4

(Solution) First we notice that

     
1 3 6
3 1 + 2 = 5 ,
    
1 1 4

meaning that the third vector of A is redundant. This allows us to write a nontrivial relation
       
1 3 6 0
3 1 + 1 2 − 1 5 = 0
1 1 4 0

among the vectors. Finally, these coefficients give us a nonzero element of ker(A), since
          
1 3 6 3 1 3 6 0
1 2 5  1  = 3 1 + 1 2 − 1 5 = 0 .
1 1 4 −1 1 1 4 0

Example 17. (§3.2, Exercise 53 of [1]) Consider a subspace V of Rn . We define the orthog-
onal complement V ⊥ of V as the set of those vectors w in Rn that are perpendicular to all
vectors in V ; that is, w · v = 0, for all v in V . Show that V ⊥ is a subspace of Rn .

(Solution) We have three properties to check: that V ⊥ contains the zero vector, that it is
closed under addition, and that it is closed under scalar multiplication. Certainly 0 · v = 0
for every v ∈ V , so 0 ∈ V ⊥ . Next, suppose we have vectors w1 and w2 in V ⊥ . Then

(w1 + w2 ) · v = w1 · v + w2 · v = 0 + 0,

25
since w1 · v = 0 and w2 · v = 0. So w1 + w2 is in V ⊥ , meaning that V ⊥ is closed under
addition. Finally, suppose we have w in V ⊥ and a scalar k. Then

(kw) · v = k(w · v) = 0,

so kw ∈ V ⊥ . So V ⊥ is closed under scalar addition, and is thus a subspace of Rn . ♦

 
1
Example 18. (§3.2, Exercise 54 of [1]) Consider the line L spanned by 2 in R3 . Find a

3
⊥
basis of L . See Exercise 53.

(Solution) Suppose v, with components v1 , v2 , and v3 , is in L⊥ . Then

   
v1 1
0 = v2 · 2 = v1 + 2v2 + 3v3 .
  
v3 3

This is a linear equation in three variables. Its solution set has two free variables – v2 and
v3 – and the remaining variable can be given in terms of these:

v1 = −2v2 − 3v3 .

Consider the vectors    

−2 −3
u1 =  1  and u2 =  0  .
0 1
We can check that u1 and u2 are both in L⊥ , and since neither is a scalar multiple of the
other, these two vectors are linearly independent. Finally, choose any vector
 
−2v2 − 3v3
v= v2 
v3

in L⊥ and notice that

   
−2v2 −3v3
v2 u1 + v3 u2 =  v2  +  0  = v.
0 v3

So the linearly independent vectors u1 and u2 span L⊥ , meaning that they provide a basis
for this space. ♦

26
4.3 The Dimension of a Subspace
Definition. The dimension of a subspace V of Rn is the number of vectors in a basis for
V , and is denoted dim(V ).
We now have a new (and better!) definition for the rank of a matrix which can be verified
to match our previous definition.
Definition. For any matrix A, rank(A) = dim(im(A)).
Example 19. (§3.3, Exercise 78 of [1]) An n × n matrix A is called nilpotent if Am = 0
for some positive integer m. Consider a nilpotent n × n matrix A, and choose the smallest
number m such that Am = 0. Pick a vector v in Rn such that Am−1 v 6= 0. Show that the
vectors v, Av, A2 v, . . . , Am−1 v are linearly independent.
(Solution) Suppose we have coefficients c0 , c1 , . . . , cm−1 so that
c0 v + c1 Av + c2 A2 v + · · · + cm−1 Am−1 v = 0. (7)
Multiplying both sides of this equation by Am−1 gives
c0 Am−1 v + c1 Am v + c2 AAm v + · · · + cm−1 Am−2 Am v = 0,
meaning that c0 Am−1 v = 0. Since Am−1 v 6= 0, this means that c0 = 0. So we may rewrite
Equation 7 as
c1 Av + c2 A2 v + · · · + cm−1 Am−1 v = 0.
We may then multiply both sides of this equation by Am−2 to obtain
c1 Am−1 v + c2 Am v + c3 AAm v + · · · + cm−1 Am−3 Am v = 0.
Similar to before, this simplifies to c1 Am−1 v = 0. This tells us that c1 = 0, so Equation 7
simplifies again to
c2 A2 v + c3 A3 v + · · · + cm−1 Am−1 v = 0.
We may carry on this argument to show that each coefficient ci is zero. This means that
the vectors v, Av, A2 v, . . . , Am−1 v admit only the trivial relation, and are thus linearly
independent. ♦
Example 20. (§3.3, Exercise 79 of [1]) Consider a nilpotent n × n matrix A. Use the result
demonstrated in Exercise 78 to show that An = 0.
(Solution) Let m be the smallest integer so that Am = 0, as in Exercise 78. According to
that exercise, we may choose v so that the vectors
v, Av, A2 v, . . . , Am−1 v
are linearly independent. We know that any collection of more than n vectors in Rn is
linearly dependent, so this collection may have at most n vectors. That is, m ≤ n, so
An = An−m Am = An−m 0 = 0,
as desired. ♦

27
Example 21. (§3.3, Exercise 82 of [1]) If a 3 × 3 matrix A represents the projection onto a
plane in R3 , what is rank(A)?

(Solution) The rank of A is given by the dimension of im(A). Because A represents the
projection onto a plane, the plane onto which we’re projecting is precisely im(A). That is,
im(A) has dimension 2, so rank(A) = 2. ♦

28
References
[1] Otto Bretscher. Linear Algebra with Applications. Pearson Education, Inc., Upper Saddle
River, New Jersey, 2013.

29