Row spaces Rank Change of basis

MATH 4A - Linear Algebra with Applications

Lecture 19: Rank, and change of basis

15 May 2019

Reading: §4.6-4.7, 5.1-5.2

Recommended problems from §4.6: 1, 3, 7, 9, 13, 15, 17, 18, 26,
27-29.
Recommended problems from §4.7: 1, 3, 7, 11, 12, 13
Row spaces Rank Change of basis

Lecture plan

1 Row spaces

2 Rank

3 Change of basis
Row spaces Rank Change of basis

Motivation

Given a m × n matrix A, the column space is very useful because it

describes the image of the linear transformation Rn → Rm
determined by A.

Is the “row space” similarly useful?

Row spaces Rank Change of basis

Row spaces

Let A be a m × n matrix. The row space of A, denoted Row A, is

the column space of the transpose matrix AT :

Row A = Col AT .

If we think of A as the standard matrix of a linear transformation

Rn → Rm , then Row A is a subspace of the domain Rn .

We want to determine if Row A is important or not. First, we

should figure out how to compute it.
Row spaces Rank Change of basis

Computing Row A

To be more precise, Row A is, by definition, just the span of the

columns of AT . But can we find a basis of Row A, not just a
spanning set?

Theorem
If two matrices A and B are row equivalent, then Row A = Row B.
If B is in echelon form, the nonzero rows of B form a basis for
both Row A and Row B.
Proof sketch: suppose a single row operation takes A to B. Then the rows of B are a linear combination of the
rows of A, hence Row B is a subspace of Row A. Since the row operation is invertible, we know Row A is a
subspace of Row B by similar reasoning. Thus Row B = Row A.

In particular, the theorem implies dim Row A is the number of pivot

columns of A. This is because every nonzero row of the echelon
form B has exactly one nonzero leading entry, i.e. exactly one
pivot, and all of the pivots are in distinct columns.
Row spaces Rank Change of basis

Example
Let’s find a basis for the row space of the matrix
 
2 4 −2 1
A = −2 −5 7
 3
3 7 −8 6
An echelon form of A is
 
1 0 9 0
0 1 −5 0
0 0 0 1
Thus, the three vectors
     
1 0 0
0 1 0
     
9 −5 0
0 0 1
form a basis of Row A.
Row spaces Rank Change of basis

Definition

The rank of a matrix A is the dimension of Col A. That is,

rank A = dim Col A.

To compute rank A, we should row reduce to an echelon form E .

The pivot columns of A (which are determined from the row
reduced form but are columns in A, not columns in E ) form a basis
of Col A. Thus, rank A is the number of pivot columns.

Sound familiar?!?
Row spaces Rank Change of basis

The rank theorem (sometimes called rank-nullity theorem)

Theorem
The dimensions of the column space and the row space of an
m × n matrix A are equal. In other words,

rank A = dim Col A = dim Row A = dim Col AT .

Moreover,
rank A + dim Nul A = n.
Proof: we already proved the first statement. To prove the second
statement, note that any one of the n columns of A is either a pivot
column or not a pivot column. Thus
n = # {pivot columns of A} + # {non-pivot columns of A}
= rank A + # {non-pivot columns of A} .
Recall that each of the non-pivot columns of A is a free variable for the
homogeneous equation Ax = 0. Thus
# {non-pivot columns of A} = dim Nul A.
Row spaces Rank Change of basis

iClicker 1

If A is a 4 × 12 matrix with rank 3, what is dim Nul A?

(a) 1
(b) 3
(c) 4
(d) 7
(e) 9
Row spaces Rank Change of basis

iClicker 2

If A is a 4 × 12 matrix with dim Nul A = 6, what is rank A?

(a) -2
(b) 4
(c) 6
(d) 10
(e) something else
Row spaces Rank Change of basis

Understanding the rank theorem geometrically

In a few weeks, we’ll develop a better, geometric understanding of

the rank theorem. Here’s the quick-and-dirty intro: we can
understand the linear transformation Rn → Rm with m × n
standard matrix A as first “projecting” on to the subspace Row A
of Rn , and then mapping Row A to the image of A (that is, Col A)
in a one-to-one way.
Row spaces Rank Change of basis

Motivation

We’ve shown, using bases and coordinate mappings, how to

convert abstract vector spaces and linear transformations between
them into copies of Rn and matrices. This is very useful.

However, this process depends on the choice of a basis, since that

determines a coordinate mapping. We should understand how to
modify things if we modify the basis.
Row spaces Rank Change of basis

Example

Consider two bases B = {b1 , b2 } and C = {c1 , c2 } for a two

dimensional vector space V such that

b1 = 4c1 + c2 and b2 = −6c1 + c2 .

Suppose x is a vector in V such that

 
3
[x]B = .
1

Can we find [x]C ?

Row spaces Rank Change of basis

Yes! Just use linearity of the coordinate mapping:

[x]C = [3b1 + 1b2 ]C = 3[b1 ]C + 1[b2 ]C .

In other words

[x]C = [b1 ]C [b2 ]C [x]B
Since b1 = 4c1 + c2 and b2 = −6c1 + c2 , we know
   
4 −6
[b1 ]C = and [b2 ]C = .
1 1

Thus,
    

4 −6 3 6
[x]C = [b1 ]C [b2 ]C [x]B = = .
1 1 1 4

(Look in the book for pictures)

Row spaces Rank Change of basis

This example generalizes

Theorem
Let B = {b1 , . . . , bn } and C = {c1 , . . . , cn } be two bases of a
P
vector space V . Then there is a unique n × n matrix C ← B such that
P
[x]C =C ← B [x]B .
P
The columns of C←B are the C-coordinate vectors for the basis
vectors in B:
P

C←B = [b1 ]C [b2 ]C · · · [bn ]C .
Row spaces Rank Change of basis

P
Note: the columns of C ← B are the linearly independent because
P
they are the coordinates of a linearly independent set. Thus C ← B is
P
invertible. The inverse matrix of C ← B changes coordinates from C
to C:  −1
P P
C←B =B ← C .
Row spaces Rank Change of basis

Changing coordinates in Rn

If B = {b1 , . . . , bn } and C = {c1 , . . . , cn } are two bases of Rn , we

P
compute C ← B as follows:
1 Form the “block-augmented” matrix


c1 · · · cn b1 · · · bn .

2 Put this matrix in reduced echelon form.

P
3 The right hand side will be C ← B. In fact, the reduced echelon
form of the block-augmented matrix is
 
P
In C ← B