Review

{v1,
, v p } is a basis of V if the vectors span V and are independent.
To obtain a basis for Nul(A), solve Ax = 0:

3 6 6 3
6 12 15 0

>
RREF

1 2 0 5
0 0 1 2


2x2 5x4 2 5
x2 1 0
x = = x2 + x4
2x4 0 2

x4 0 1
2 5
1 0
Hence, , form a basis for Nul(A).


0 2
0 1

To obtain a basis for Col(A), take the pivot columns of A.

>

1 2 0 4 1 2 0 4
2 4 1 3 0 0 1 5

3 6 2 22 0 0 0 0
4 8 0 16 0 0 0 0

1 0
2 1
Hence, 3 , 2
form a basis for Col(A).
4 0

Row operations do not preserve the column space.

For instance,

1
0

>
R1R2

0
1

.

On the other hand: row operations do preserve the null space.

Why? Recall why/that we can operate on rows to solve systems like Ax = 0!

Dimension of Col(A) and Nul(A)

Definition 1. The rank of a matrix A is the number of its pivots.

Theorem 2. Let A be an m n matrix of rank r. Then:

dim Col(A) = r
Why? A basis for Col(A) is given by the pivot columns of A.

dim Nul(A) = n r is the number of free variables of A

Why? In our recipe for a basis for Nul(A), each free variable corresponds to an element in the
basis.

dim Col(A) + dim Nul(A) = n

Why? Each of the n columns either contains a pivot or corresponds to a free variable.

Armin Straub 1
astraub@illinois.edu
 The four fundamental subspaces

Row space and left null space

Definition 3.
The row space of A is the column space of AT .
Col(AT ) is spanned by the columns of AT and these are the rows of A.

The left null space of A is the null space of AT .


Why left? A vector x is in Nul(AT ) if and only if ATx = 0.
Note that ATx = 0 (ATx)T = xTA = 0T .
Hence, x is in Nul(AT ) if and only if xTA = 0.

Example 4. Find a basis for Col(A) and Col(AT ) where

1 2 0 4
2 4 1 3
A =
3
.
6 2 22
4 8 0 16

Solution. We know what to do for Col(A) from an echelon form of A, and we could
likewise handle Col(AT ) from an echelon form of AT .
But wait!
Instead of doing twice the work, we only need an echelon form of A:

1 2 0 4 1 2 0 4
2 4 1 3 0 0 1 5

3 6 2 22 0 0 2 10
4 8 0 16 0 0 0 0

Armin Straub 2
astraub@illinois.edu

1 2 0 4
0 0 1 5
= B
0 0 0 0
0 0 0 0

Hence, the rank of A is 2.

1 0
2 1
A basis for Col(A) is 3 , 2 .


4 0

Recall that Col(A)  Col(B). Thats because we performed row operations.

However, the row spaces are the same! Col(AT ) = Col(B T )
The row space is preserved by elementary row operations.

1 0
2 0
In particular: a basis for Col(AT ) is given by 0 , 1 .


4 5

Theorem 5. (Fundamental Theorem of Linear Algebra, Part I)

Let A be an m n matrix of rank r.
dim Col(A) = r (subspace of R m)

dim Col(AT ) = r (subspace of R n)

dim Nul(A) = n r (subspace of R n) (# of free variables of A)

dim Nul(AT ) = m r (subspace of R m)

In particular:
The column and row space always have the same dimension!
In other words, A and AT have the same rank. [i.e. sam e num b er of pivots]

Easy to see for a matrix in echelon form

2 1 3 0
0 0 1 2 ,
0 0 0 7
but not obvious for a random matrix.

Armin Straub 3
astraub@illinois.edu
 Linear transformations
Throughout, V and W are vector spaces.

Definition 6. A map T : V W is a linear transformation if

T (cx + dy) = cT (x) + dT (y) for all x, y in V and all c, d in R.

In other words, a linear transformation respects addition and scaling:

T (x + y) = T (x) + T (y)
T (cx) = cT (x)
It also sends the zero vector in V to the zero vector in W :
T (0) = 0 [because T (0) = T (0 0) = 0 T (0) = 0]

Example 7. Let A be an m n matrix.

Then the map T (x) = Ax is a linear transformation T : Rn Rm.
Why?
Because matrix multiplication is linear:
A(cx + dy) = cAx + dAy
The LHS is T (cx + dy) and the RHS is cT (x) + dT (y).

Example 8. Let Pn be the vector space of all polynomials of degree at most n. Consider
the map T : Pn Pn1 given by
d
T (p(t)) = p(t).
dt
This map is linear! Why?
Because differentiation is linear:
d d d
dt
[ap(t) + bq(t)] = a dt p(t) + b dt q(t)
The LHS is T (ap(t) + bq(t)) and the RHS is aT (p(t)) + bT (q(t)).

Representing linear maps by matrices

Let x1,
, xn be a basis for V .
A linear map T : V W is determined by the values T (x1),
, T (xn).

Armin Straub 4
astraub@illinois.edu
 Why?
Take any v in V .
It can be written as v = c1x1 +
+ cnxn because {x1,
, xn } is a basis and hence spans V .
Hence, by the linearity of T ,

T (v) = T (c1x1 +
+ cnx) = c1T (x1) +
+ cnT (xn).

Definition 9. (From linear maps to matrices)

Let x1,
, xn be a basis for V , and y1,
, ym a basis for W .
The matrix representing T with respect to these bases
has n columns (one for each of the x j ),
the j-th column has m entries a1,j ,
, am, j determined by

T (x j ) = a1, jy1 +
+ am,jym.

Example 10. Let V = R2 and W = R3. Let T be the linear map such that

  1   4
1 0
T = 2 , T = 0 .
0 1
3 7

What is the matrix A representing T with respect to the standard bases?

Solution. The standard bases are

    1 0 0
1 0
0
, 1
for R2, 0 , 1 , 0 for R3.
0 0 1
x1 x2
y1 y2 y3

1 1 0 0
T (x1) = 2 = 1 0 + 2 1 + 3 0
3 0 0 1
= 1y1 + 2y
2 + 3y


3
1
A = 2
3

4
T (x2) = 0 = 4y1 + 0y2 + 7y3
7



1 4
A = 2 0
3 7

Armin Straub 5
astraub@illinois.edu
(We did not have time yet to discuss the next example in class, but it will be helpful if
your discussion section already meets Tuesdays.)

Example 11. As in the previous example, let V = R2 and W = R3. Let T be the (same)
linear map such that

  1   4
1 0
T = 2 , T = 0 .
0 1
3 7

What is the matrix B representing T with respect to the following bases?

    1 0 0
1
1
, 1
2
for R2, 1 , 1 , 0 for R3.
1 0 1
x1 x2
y1 y2 y3

Solution. This time:

     
1 1 0
T (x1) =T =T +T
1 0 1

1 4 5
= 2 + 0 = 2
3 7 10

1 0 0
can you see it?
= 5 1 3 1 + 5 0 otherw ise: d o it!
1 0 1



5
B = 3
5
     
1 1 0
T (x2) =T = T + 2T
2 0 1

1 4 7
= 2 + 2 0 = 2
3 7 11

1 0 0
= 7 1 9 1 + 4 0
1 0 1



5 7
B = 3 9
5 4

Tedious, even in this simple example! (But we can certainly do it.)

A matrix representing T encodes in column j the coefficients of T (x j ) expressed as

a linear combination of y1,
, ym.

Armin Straub 6
astraub@illinois.edu
 Practice problems
 
1 2 3 4 1
Example 12. Suppose A = 2 4 7 8 1
. Find the dimensions and a basis for all four
fundamental subspaces of A.

Example 13. Suppose A is a 5 5 matrix, and that v is a vector in R5 which is not

a linear combination of the columns of A.
What can you say about the number of solutions to Ax = 0?

Solution. Stop reading, unless you have thought about the problem!

Existence of such a v means that the 5 columns of A do not span R 5.

Hence, the columns are not independent.
In other words, A has at most 4 pivots.
So, at least one free variable.
Which means that Ax = 0 has infinitely many solutions.

Armin Straub 7
astraub@illinois.edu