T1A/T1B, hmcheungae Math 2121 Tutorial 6 October 21, 2018

Mathematics consists in proving the most obvious thing in the least obvious way.——
———————————————————————————————–George Polya

Null Space and Column Space

Nul(A)
1 Nul(A) is a subspace of Rn .
2 Nul(A) is implicitly defined,
i.e., you are only given a condition Ax
that vectors in Nul(A) must satisfy.
3 It takes time to find vectors in Nul(A):
Row operations are required.
4 There is no obvious relation
between Nul(A) and the entries of A.
5 A typical vector v in Nul(A) has the property
that Av = 0.
6 Given a specific vector v, it is easy to tell
if v is in Nul(A). Compute Av.
7 Nul(A)={0} iff
Ax = 0 has only the trivial solution
iff x 7→ Ax is one-to-one.
8 A basis for Nul(A) is obtained by
collecting basic solutions from solving Ax = 0.
Col(A)
Col(A) is a subspace of Rm .
Col(A) is explicitly defined,
i.e., you are told how to bulid vectors in Col(A).
It is easy to find vectors in Col(A):
the columns of A are displayed
and others are formed from them.
There is an obvious relation
between Col(A) and the entries of A,
since each of column of A is in Col(A).
A typical vector v in Col(A) has the property
that Ax = v is consistent.
It takes time to tell whether a given v is in Col(A):
Row operations are required.
Col(A)= Rm iff
Ax = b has a solution for every b ∈ Rm
iff x 7→ Ax is onto.
A basis for Col(A) is obtained by
collecting the pivot columns of A.

Coordinates
Let B = (b1 , · · · , bk ) be the list of basis vectors in some fixed order.
 
c1
c2 
• Given v ∈ H, define [v]B =  .  ∈ Rk as the unique vector with v = c1 b1 + c2 b2 + · · · + ck bk .
 
 .. 
ck
We call [v]B the coordinate vector of v in the basis B or just v in the basis B.
Example. If H = Rn and B = (e1 , e2 , · · · , en ) is the standard basis then [v]B = v.
• φB : Rk → Rk , defined by v 7→ [v]B is an isomorphism, it has an inverse φ−1 k k
B : R → R , which is also
an isomorphism. Note that φB (v) = [v]B
•
 
4
The vector u = has standard coordinates x = 4 and y = 5.
5
If we use the blue coordinate system, whose coordinate axes are labelled x0 and y 0 , the blue coordinates
of u are x0 = 3 and y 0 = 2. The notation is as follows:
           
4 3 1 0 2 −1
[u]S = , [u]B = , where S = , , B= , .
5 2 0 1 1 1
T1A/T1B, hmcheungae Math 2121 Tutorial 6 Page 2 of 7

Figure 1: Two coordinate systems in the plane are displayed.

Dimension of a subspace in Rn
Suppose H is a dimensional subspace of Rn .
• Let H be a subspace of Rn . Then all bases of H have the same number of element, we called that
number of basis to be the dim H.
• If H 0 is a subspace of H and dim H 0 = dim H, then H 0 = H.
• Basis theorem.
If H is a subspace of Rn with dim H = p, then
– Any set of p linearly independent vectors in H is a basis for H.
– Any set of p vectors in H which span H is a basis for H.

Rank of a Matrix

• Let A be an m × n matrix, write

 T
r1
 rT 
 2
A = [a1 a2 · · · an ] =  .  ,
 .. 
rT
m

– The column space of A: Col(A) = Span{a1 , a2 , · · · , an } = {Ax ∈ Rm | x ∈ Rn }.

– The row space of A: Row(A) = Span {r1 , r2 , · · · , rm } = Col(AT ).
• The rank of a matrix A is denoted by rank A := dim(Col A).
– rank A = the number of pivot positions in A.
– rank A = rank AT .
• Characterizations of matrix A by using rank(A) (See example ii (c)):
– If A is an m × n matrix and T (x) = Ax, then
∗ the columns of A are linearly independent iff T is one-to-one iff rank A = n.

Cont.
T1A/T1B, hmcheungae Math 2121 Tutorial 6 Page 3 of 7

∗ the columns of A span Rm iff T is onto iff rank A = m.

– If A is an n × n matrix and T (x) = Ax, then A is invertible iff T is invertible iff rank A = n.

The Rank-Nullity Theorem

If A is an m × n matrix, then rank A + dim(Nul(A)) = n.

Example 1
 
1 −1 1 1
−1 1 1 −1
Let A =  .
1 −1 −1 1
−1 1 −1 −1

(a) Find a basis for the Col(A).

(b) Find a basis for the Nul(A).

Answer:
 
1 −1 0 1
0 0 1 0
(a) By Row reducing A, we get rref (A) = 
0
.
0 0 0
0 0 0 0
(i) The first and third columns are linearly independent (have leading 1’s)
   

 1 1 
   
−1 , 1  .

The pivot columns of a matrix A form a basis for Col(A), so a basis for the Col(A) is  1  −1

 
−1 −1
 

Remark: Any non-pivot column is a linear combination of pivot columns on its left hand side.
(b) Solving Ax = 0 from our row reduction above
 
x
y 
 z  with y, w free, z = 0 and x − y + w = 0,
gives  

w
so
        

 y−w 
 
 1 −1 

y  1 0
   
4 4
Nul(A) =    ∈ R : y, w ∈ R = y   + w   ∈ R : y, w ∈ R ,

 0  
  0 0 
  
w 0 1
  
   

 1 −1 
   
1, 0  .

and we can take a basis to be  0  0 

 
0 1
 

Spanning:
   

 1 −1 
   
1 , 0  .

N ul(A) = Span 
0  0 

 
0 1
 

Linearly independence:

Cont.
T1A/T1B, hmcheungae Math 2121 Tutorial 6 Page 4 of 7

         
1 −1 0 y−w 0
1  0  0  y  0
0 + w  0  = 0 =⇒  0  = 0 =⇒ y = w = 0.
y         

0 1 0 w 0

Remark: If the general solution of Ax = 0 is

x = c1 v1 + c2 v2 + · · · + ck vk ,
where c1 , · · · , ck are the free variables. Then {v1 , · · · , vk } form basis of Nul(A).

Example 2
   
−3 6 −1 1 −7 1 −2 0 −1 3
Let A =  1 −2 2 3 −1. Given rref (A) = 0 0 1 2 −2.
2 −4 5 8 −4 0 0 0 0 0
Find a basis for
(a) Col(A).
(b) Nul(A).
Answer:
(a) Column Space: The columns with a leading 1 from the row vectors are 1 and 3.
Hence, a basis for the column space are columns 1 and 3 from the original matrix that is
   
 −3 −1 
 1 , 2  .
2 5
 

Hence the column space has dimension 2.

(b) Null space:



x1 = 2r + s − 3t
x2 = r



The solutions of the system are x3 = −2s + 2t , which can written as

x4 = s





x = t
5
       
x1 2 1 −3
x2  1 0 0
       
x3  = r 0 + s −2 + t  2  .
       
x4  0 1 0
x5 0 0 1
Nul(A) is the subspace spanned by
     

 2 1 −3  
1  0   0 


     

0 , −2 ,  2  ,


    
0  1   0 

 

0 0 1
 

which is clearly linearly independent. Hence it is a basis for Nul(A).

Cont.
T1A/T1B, hmcheungae Math 2121 Tutorial 6 Page 5 of 7

Example 3
 
1 1 1 1
Find the nullity of the matrix A = .
2 3 4 5
Answer:
Elementary row operations do not change the nullspace.
Let us convert A to reduced row echelon form:
     
1 1 1 1 1 1 1 1 1 0 −1 −2
→
− →
−
2 3 4 5 0 1 2 3 0 1 2 3
( (
x1 − x3 − 2x4 = 0 x1 = x3 + 2x4
which is equivalent to .
x2 + 2x3 + 3x4 = 0 x2 = −2x3 − 3x4
Note that general element of Null(A) is in the form of:
       
x1 t + 2s 1 2
x2  −2t − 3s −2 −3
 =  = t   + s   where t, s ∈ R.
x3   t  1 0
x4 s 0 1
   

 1 2 
   
−2 −3

  ,   forms a basis for Null(A). Thus the nullity of the matrix A is 2.
 1
   0 

0 1
 

Example 4
     
 1 1 −1 
(a) Show B =  1  , −1 ,  1  is a basis for R3 .
−1 1 1
 
 
1
(b) If v =  1 , find [v]B .
−1
 
1
(c) If [v]B =  1 , find v.
−1
Answer:
   
1 1 −1 1 0 0
(a) Row reduce  1 −1 1  to get 0 1 0.
−1 1 1 0 0 1
Thus B is a basis for R3 .
 
1
(b) Since v is the first element in the basis B, we have [v]B = 0.
0
(c) Here v is the sum of the first two vectors in B minus the third, i.e.,
       
1 1 −1 3
v =  1  + −1 −  1  = −1 .
−1 1 1 −1

Cont.
T1A/T1B, hmcheungae Math 2121 Tutorial 6 Page 6 of 7

Example 5
True or False:
Let A ∈ Mm×n (R), B ∈ Mn×r (R),
1. Suppose C = AB. Then Col(C) ≤Col(A).
2. Suppose C = AB. Then rank(C) >rank(A).
3. Suppose C = AB. Then Nul(B) ≤ Nul(C).
4. One have Nul(AT A)=Nul A.
5. One have rank(AT A) = rank A.
6. One have rank A ≤ min{n, m}.
7. rank A = n if and only if T is one-to-one.
8. rank A = m if and only if T is onto.
Answer:

1. True: x ∈Col(C) then x = Ck for some k ∈ Rr . Then x = ABk = A(Bk) = Ab for some
b = Bk ∈ Rn . Hence x ∈Col(A).
2. False: rank(C) = dim Col(C) ≤ dim Col(A) =rank(A).
3. True: Suppose x ∈ N ul(B), then Bx = 0 and then Cx = ABx = 0. Hence x ∈ N ul(C).
4. True:
Clearly, Nul(A) ⊆ Nul(AT A) as Ax = 0 =⇒ AT Ax = 0.
 
y1
 y2 
First notice that y =  .  ∈ Rn and
 
 .. 
yn

yT y = 0 ⇐⇒ y12 + · · · + yn2 = 0 ⇐⇒ y1 = y2 = · · · = yn = 0 ⇐⇒ y = 0.

To show Nul(AT A) ⊆ Nul(A), suppose x ∈ Nul(AT A). Then

AT Ax = 0 =⇒ (Ax)T Ax = xT AT Ax = xT 0 = 0 =⇒ Ax = 0,

Hence x ∈ Nul(A).
5. True:
By Rank-Nullity Theorem, we get

dim Nul(A) + rank(A) = n.

dim Nul(AT A) + rank(AT A) = n.

Together with above, we get rank(AT A) = rank A,
6. True:
Note that rank(A) = n − dim Nul(A) ≤ n as dim Nul(A) ≥ 0.
Since rank(A) = dim Col(A) and Col(A) ≤ Rm by the subspace relation.
We get rank(A) = dim Col(A) ≤ dim Rm = m.
Therefore, rank(A) ≤ min{n, m}.

Cont.
T1A/T1B, hmcheungae Math 2121 Tutorial 6 Page 7 of 7

7. True:

rankA = n ⇐⇒ dim Nul(A) = 0

⇐⇒ Nul(A) = {0}
⇐⇒ T is one to one.

8. True:

rankA = m ⇐⇒ dim Col(A) = m

⇐⇒ Col(A) = Rm
⇐⇒ T is onto.

Example 6
Let T : Rn → Rm be a linear transformation.
Using Rank-Nullity Theorem to prove the following:
(a) If m < n, then T cannot be one-to-one.
(b) If m > n, then T cannot be onto.

[Compare with Tutorial 5, example 2(1), (2).]

Answer:
Let A be the standard matrix of A. Denote the nullity of A and the rank of A by n(A) and r(A)
respectively.

(a) If n > m, then since r(A) is the dimension of a subspace of Rm , then we get

r(A) ≤ m < n, n(A) = n − r(A) > 0.

This implies n(A) 6= 0 and then N ul(A) 6= {0}.

Therefore, T cannot be one-to-one.
(b) By Rank-Nullity Theorem,

n(A) + r(A) = n, where n(A), r(A), n ≥ 0.

Since n < m, then

r(A) = n − n(A) ≤ n < m.

And hence T cannot be onto due to the fact that r(A) 6= m, which implies Col(A) 6= Rm .

The End.