MATH 2121 — Linear algebra (Fall 2020) Lecture 9

TLDR
Quick summary of today’s notes. Lecture starts on next page.
• If A and B are n × n matrices with AB = In , then BA = In and A−1 = B.
• A subspace H of Rn is a subset of Rn containing the zero vector that is closed under linear combi-
nations. This means that 0 ∈ H and if u, v ∈ H and c ∈ R then u + v ∈ H and cv ∈ H.
The zero subspace of Rn is the set {0} containing just 0 ∈ Rn .
• Let A be an m × n matrix.
The column space of A is the span of the columns of A. Denoted Col A. This is a subspace of Rm .
          
1 0 0 
 1 0 0  
 a 

 0 1 2 

0   1   2   b 

Col   = R-span  , ,  =   : a, b ∈ R ⊆ R4
 1 0 0    1   0   0   a  
 
  
0 0 0 0 0 0 0
  

The null space of A is the set of vectors Nul A = {v ∈ Rn : Av = 0}. This is a subspace of Rn .
 
1 0 0      
 0 1  x 0
2 
  
 y  ∈ R3 : x = y + 2z = 0 =  −2z  : z ∈ R ⊆ R3 .

Nul  1 0 =
0  
z z
  
0 0 0

• A basis for a subspace H ⊆ Rn is a linearly independent set of vectors v1 , v2 , . . . , vp ∈ Rn such that

H = R-span{v1 , v2 , . . . , vp }.

The standard basis of Rn is e1 , e2 , . . . , en where ei ∈ Rn is the vector with 1 in row i and 0 in all
other rows. Any subspace of Rn has a basis with at most n vectors.
• The pivot columns of an m × n matrix A form a basis for Col A.
• Both A and RREF(A) always have the same null space. Usually Col A 6= Col RREF(A).
To find a basis for Nul A, determine the indices i1 , i2 , . . . , ip of the non-pivot columns of A.
Then there are unique vectors v1 , v2 , . . . , vp ∈ Rn such that any
 
x1
 x2 
x =  .  ∈ Rn with RREF(A)x = 0
 
 .. 
xn

can be written as x = xi1 v1 + xi2 v2 + · · · + xip vp . The vectors v1 , v2 , . . . , vp are a basis for Nul A.
 
1 2 0 4 −1
For example, if RREF(A) = then any x ∈ R5 with RREF(A)x = 0 has
0 0 1 0 2
         
x1 −2x2 − 4x4 + x5 −2 −4 1

 x2  
  x2 


 1 


 0 


 0 

x=
 x3 =
  −2x5  = x2 
  0  + x4 
  0  + x5 
  −2 .

 x4   x4   0   1   0 
x5 x5 0 0 1

The three vectors on the right are a basis for Nul A = Nul RREF(A).

0
MATH 2121 — Linear algebra (Fall 2020) Lecture 9

1 Last time: inverses

The following all mean the same thing for a function f : X → Y :
1. f is invertible.
2. f is one-to-one and onto.
3. For each b ∈ Y there is exactly one a ∈ X with f (a) = b.
4. There is a unique function f −1 : Y → X, called the inverse of f , such that

f −1 (f (a)) = a and f (f −1 (b)) = b for all a ∈ X and b ∈ Y .

Proposition. If T : Rn → Rm is linear and invertible then m = n and T −1 is linear and invertible.

The following all mean the same thing for an n × n matrix A:

1. A is invertible.
2. A is the standard matrix of an invertible linear function T : Rn → Rn .
3. There is a unique n × n matrix A−1 , called the inverse of A, such that
 
1
 1 
A−1 A = AA−1 = In where we define In =  .
 
..
 . 
1

4. For each b ∈ Rn the equation Ax = b has a unique solution.

5. RREF(A) = In
6. The columns of A are linearly independent and their span is Rn .
 
a b
Proposition. Let A = be a 2 × 2 matrix.
c d
(1) If ad − bc = 0 then A is not invertible.
 
−1 1 d −b
(2) If ad − bc 6= 0 then A = ad−bc .
−c a

Proposition. Let A and B be n × n matrices.

1. If A is invertible then (A−1 )−1 = A.
2. If A and B are both invertible then AB is invertible and (AB)−1 = B −1 A−1 .
3. If A is invertible then AT is invertible and (AT )−1 = (A−1 )T .

Process to compute A−1

 
Let A be an n × n matrix. Consider the n × 2n matrix A In .
= In A−1 .
 
 
If A is invertible then RREF A In
So to compute A−1 , row reduce A In to reduced echelon form, and then take the last n columns.
 

1
MATH 2121 — Linear algebra (Fall 2020) Lecture 9

2 Stronger characterization of invertible matrices

Remember that a matrix can only be invertible if it has the same number of rows and columns.

Theorem. When A is an n × n matrix, the following are equivalent:

(a) A is invertible.
(b) The columns of A are linearly independent.
(c) The span of the columns of A is Rn

Proof. We already know that (a) implies both (b) and (c).
Assume just (b) holds. Then A has a pivot position in every column, so RREF(A) = In since A has the
same number of rows and columns. But this implies that A is invertible.
Similarly, if (c) holds then A has a pivot position in every row, so RREF(A) = In and A is invertible.

Corollary. Suppose A and B are both n × n matrices. If AB = In then BA = In .

This means that if we want to show that B = A−1 then it is enough to just check that AB = In .

Proof. Assume AB = In . Then the columns of A span Rn since if v ∈ Rn then Au = v for u = Bv ∈ Rn ,

so A is invertible. Therefore B = A−1 AB = A−1 In = A−1 so BA = A−1 A = In .

Important note: this corollary only applies to square matrices.

3 Subspaces of Rn
 
0
 0 
Let n be a positive integer. Write 0 =   ∈ Rn .
 
..
 . 
0
n
Definition. Let H be a subset of R . The subset H is a subspace if these three conditions hold:
1. 0 ∈ H.
2. u + v ∈ H for all u, v ∈ H.
3. cv ∈ H for all c ∈ R and v ∈ H.

Common examples
Rn is a subspace of itself.
The set {0} consisting of just the zero vector is a subspace of Rn .
The empty set ∅ is not a subspace since it does not contain 0.
A subset H ⊆ R2 is a subspace if and only if H = {0} or H = R2 or H = R-span{v} for some v ∈ R2
The span of any set of vectors in Rn is a subspace.
Later, we will see that every subspace is the span of some set of vectors.

2
MATH 2121 — Linear algebra (Fall 2020) Lecture 9

Example. The set    

 v1 
X = v =  v 2  ∈ R3 : v 1 + v 2 + v 3 = 1
v3
 

is not a subspace since 0 ∈

/ X.

Example. The set    

 v1 
H = v =  v 2  ∈ R3 : v 1 + v 2 + v 3 = 0
v3
 

is a subspace since if u, v ∈ H and c ∈ R then

(u1 + v1 ) + (u2 + v2 ) + (u3 + v3 ) = (u1 + u2 + u3 ) + (v1 + v2 + v3 ) = 0 + 0 = 0

and
cv1 + cv2 + cv3 = c(v1 + v2 + v3 ) = 0
so u + v ∈ H and cv ∈ H.

Any matrix A gives rise to two subspaces, called the column space and null space.

Definition. The column space of an m × n matrix A is the subspace

Col A ⊆ Rm

given by the span of the columns of A.

Remark. If T : Rn → Rm is the linear function T (x) = Ax then Col A = range(T ).

Note that Col A = Rm if and only if Ax = b has a solution for each b ∈ Rm .
A vector b ∈ Rm belongs to Col A if and and only if Ax = b has a solution.

Definition. The null space of an m × n matrix A is the subspace

Nul A ⊆ Rn

given by the set of vectors v ∈ Rn with Av = 0.

Proof that Nul A is a subspace. If u, v ∈ Nul A and c ∈ R then A(u + v) = Au + Av = 0 + 0 = 0 and

A(cv) = c(Av) = 0, so u + v ∈ Nul A and cv ∈ Nul A. Thus Nul A is a subspace of Rn .

Remark. If T : Rn → Rm is the linear function T (x) = Ax then Nul A = {x ∈ Rn : T (x) = 0}.

The column space is a subspace of Rm where m is the number of rows of A.

The null space is a subspace of Rn where n is the number of columns of A.

Each subspace is completely determined by a finite amount of data. This data will be called a basis.

Definition. Let H be a subspace of Rn . A basis for H is a set of vectors {v1 , v2 , . . . , vk } ⊆ H that are
linearly independent and have span equal to H.
The empty set ∅ = {} is considered to be a basis for the zero subspace {0}.

3
MATH 2121 — Linear algebra (Fall 2020) Lecture 9

   
1 0

 0 


 1 

n
Example. The set {e1 , e2 , . . . , en } ⊆ R where e1 = 
 0 , e2 = 
  0 , and so on, is a basis for Rn .

 ..   .. 
 .   . 
0 0
We call this the standard basis of Rn .

Theorem. Every subspace H of Rn has a basis of size at most n.

Proof. If H = {0} then ∅ is a basis.

Assume H 6= {0}. Let B be a set of linearly independent vectors in H that is as large as possible. The
size of B must be at most n since any n + 1 vectors in Rn are linearly dependent.
Let w1 , w2 , . . . , wk be the elements of B. Since B is as large as possible, if v ∈ H is any vector then
w1 , w2 , . . . , wk , v are linearly dependent so we can write
c1 w1 + c2 w2 + · · · + ck wk + cv = 0
for some numbers c1 , c2 , . . . , ck , c ∈ R which are not all zero.
If c = 0 then this would imply that the vectors in B are linearly dependent. But the vectors in B are
linearly independent, so we must have c 6= 0. Therefore
c1 c2 ck
v= c w1 + c w2 + ··· + c wk .

This means that v is in the span of the vectors in B. Since v ∈ H is an arbitrary vector, we conclude
that the span of the vectors in B is all of H, so B is a basis for H.
 
−3 6 −1 1 −7
Example. Let A =  1 −2 2 3 −1 .
2 −4 5 8 −4
How can we find a basis for Nul A? Well, finding a basis for Nul A is more or less the same task as finding
all solutions to the homogeneous equation Ax = 0. So let’s first try to solve that equation.
 
If we row reduce the 3 × 6 matrix A 0 , we get
 
  1 −2 0 −1 3 0  
A 0 ∼ 0 0 1 2 −2 0  = RREF( A 0 ).
0 0 0 0 0 0
(
x1 − 2x2 − x4 + 3x5 = 0
This tells us that Ax = 0 if and only if
x3 + 2x4 − 2x5 = 0.
Therefore x ∈ Nul A if and only if
         
x1 2x2 + x4 − 3x5 2 1 −3
 x2   x2   1   0   0 
         
x= x
 3  
 =  −2x 4 + 2x 5
 = x2 
  0  + x4 
  −2  + x5 
  2 .

 x4   x4   0   1   0 
x5 x5 0 0 1
The vectors      

 2 1 −3 

1 0 0

 


 
 
 
 



 0 ,
  −2 ,
  2 

0 1 0

     

 

0 0 1
 

4
MATH 2121 — Linear algebra (Fall 2020) Lecture 9

are a basis for Nul A: we just computed that these vectors span the null space, and they are linearly
independent since each has a nonzero entry in a row (namely, either row 2, 4, or 5) where the others have
zeros. (Why does this imply linear independence?)
This example is important: the procedure just described works to construct a basis of Nul A for any
matrix A. The size of this basis will always be equal to the number of free variables in the linear system
Ax = 0. How to find a basis for Nul A is something you should remember at the end of this course.
 
1 0 −3 5 0
 0 1 2 −1 0 
Example. Let B =   0 0
.
0 0 1 
0 0 0 0 0
This matrix is in reduced echelon form. How to find a basis for Col B?
The columns of B automatically span Col B, but they might not be linearly independent.
The largest linearly independent subset of the columns of B will be a basis for Col B, however.
In our example, the pivot columns 1, 2 and 5 are linearly independent since each has a row with a 1
where the others have 0s. These columns span columns 3 and 4, so a basis for Col B is
     

 1 0 0 
0   1   0 

 , ,  .
 0   0   1 
 
0 0 0
 

This example was special since the matrix B was already in reduced echelon form. To find a basis of the
column space of an arbitrary matrix, we rely on the following observation:

Proposition. Let A be any matrix. The pivot columns of A form a basis for Col A.

Proof sketch. Suppose A is m × n. The reduced echelon form of A is obtained by multiplying A by an

invertible matrix E on the left, so we can write RREF(A) = EA.
 
  Ik
If a1 , a2 , . . . , ak are the pivot columns of A, then E a1 a2 . . . ak is the m × k matrix
0
where the 0 means an (m − k) × n submatrix of zeros. These columns are linearly independent since if
 
a1 a2 . . . ak v = 0
   
k
  Ik v
for v ∈ R then 0 = E a1 a2 . . . ak v = v= which implies that v = 0.
0 0
Suppose w ∈ Rn is a non-pivot column of A. The definition of reduced echelon form implies that the
corresponding column Ew of RREF(A) = EA is in the span of Ea1 , Ea2 , . . . , Eak . (Why?)
If Ew = c1 Ea1 + · · · + ck Eak then multiplying both sides by E −1 gives w = c1 a1 + · · · + ck ak so w is in
the span a1 , a2 , . . . , ak . Therefore the span of the pivot columns of A is equal to Col A.
Since the pivot columns are linearly independent and have span equal to Col A, they form a basis.

Example. The matrix  

1 3 3 2 −9
 −2 −2 2 −8 2 
A=
 2

3 0 7 1 
3 4 −1 11 −8

5
MATH 2121 — Linear algebra (Fall 2020) Lecture 9

is row equivalent to the matrix B in the previous example. Columns 1, 2, and 5 of A have pivots, so
     

 1 3 −9  
 −2  ,  −2  ,  2 
     
  2   3   1 
 
3 4 −8
 

is a basis for Col A.

Next time: we will show that if H is a subspace of Rn then all of its bases have the same size. The
common size of each basis is the dimension of H.

6
MATH 2121 — Linear algebra (Fall 2020) Lecture 9

4 Vocabulary
Keywords from today’s lecture:
1. Subspace of Rn

A subset H ⊆ Rn such that 0 ∈ H; if u, v ∈ H then u + v ∈ H; and if v ∈ H, c ∈ R then cv ∈ H.

Example: Pick any vectors v1 , v2 , . . . , vp ∈ Rn . Then R-span{v1 , v2 , . . . , vp } is a subspace.

2. Column space of an m × n matrix A.

The subspace Col A = {Av : v ∈ Rn } ⊆ Rm . The span of the columns of A.

    
1 0  x 
Example: If A =  0 1  then Col A =  y  ∈ R3 : x, y ∈ R .
0 0 0
 

3. Null space of an m × n matrix A.

The subspace Nul A = {v ∈ Rn : Av = 0} ⊆ Rn .

      
   2x  2 0 
1 −2 0

Example: If A = then Nul A =  x  ∈ R3 : x, y ∈ R = R-span  1  ,  0  .
−1 2 0
y 0 1
   

4. Basis of a subspace H ⊆ Rn

A set of linearly independent vectors in H whose span is H.

      
1 0  v1 
Example: The vectors  −1  ,  1  are a basis for the subspace  v 2  ∈ R3 : v 1 + v 2 + v 3 = 0 .
0 −1 v3
 
     
1 0 0
 0   1   0 
     
n  0  0   . 
The standard basis of R consists of the vectors e1 =  , e2 =  , . . . , en =  .. .

 ..   ..   
 .   .   0 
0 0 1