(Reduced) Row Echelon Form

1.2 Review

A matrix is in row echelon form if

1. All zero rows are at the bottom.
2. Each leading nonzero entry of a row is to the right of the
leading entry of the row above.
3. Below a leading entry of a row, all entries are zero.
A matrix is in reduced row echelon form if it is in row echelon
form, and in addition,
4. The pivot in each nonzero row is equal to 1.
5. Each pivot is the only nonzero entry in its column.

Row echelon form: Reduced row echelon form:

   
? ? ? ? ? 1 0 ? 0 ?
 0 ? ? ? ? 0 1 ? 0 ?
   
 0 0 0 ? ? 0 0 0 1 ?
0 0 0 0 0 0 0 0 0 0
? = pivots
1.2: Row Reduction

Theorem
Every matrix is row equivalent to one and only one matrix in reduced row
echelon form.

We’ll give an algorithm, called row reduction or Gaussian elimination, which

demonstrates that every matrix is row equivalent to at least one matrix in
reduced row echelon form.

Note: Like echelon forms, the row reduction algorithm does not care if a
column is augmented: ignore the vertical line when row reducing.

The uniqueness statement is interesting—it means that, nomatter how you row
reduce, you always get the same matrix in reduced row echelon form.
(Assuming you only do the three legal row operations.) (And you don’t make
any arithmetic errors.)

Maybe you can figure out why it’s true!

Reinforcing 1.2: Row Reduction Algorithm

Step 1a Swap the 1st row with a lower one so a leftmost nonzero entry is in 1st
row (if necessary).
Step 1b Scale 1st row so that its leading entry is equal to 1.
Step 1c Use row replacement so all entries below this 1 are 0.
Step 2a Swap the 2nd row with a lower one so that the leftmost nonzero entry is
in 2nd row.
Step 2b Scale 2nd row so that its leading entry is equal to 1.
Step 2c Use row replacement so all entries below this 1 are 0.
Step 3a Swap the 3rd row with a lower one so that the leftmost nonzero entry is in
3rd row.
etc.
Last Step Use row replacement to clear all entries above the pivots, starting with the
last pivot (to make life easier).

Example  
0 −7 −4 2
2 4 6 12 
3 1 −1 −2
[animated]
Row Reduction
Example

   
0 −7 −4 2 R1 ←→ R2 2 4 6 12
 2 4 6 12   0 −7 −4 2
3 1 −1 −2 3 1 −1 −2
Step 1a: Row swap to make this nonzero. Step 1b: Scale to make this 1.
 
R1 = R1 ÷ 2 1 2 3 6
 0 −7 −4 2
3 1 −1 −2
Step 1c: Subtract a multiple of
the first row to clear this.
 
R3 = R3 − 3R1 1 2 3 6
 0 −7 −4 2
0 −5 −10 −20
 
R2 ←→ R3 1 2 3 6
Optional: swap rows 2 and 3 to  0 −5 −10 −20 
make Step 2b easier later on.
0 −7 −4 2
Row Reduction
Example, continued

   
1 2 3 6 R2 = R2 ÷ −5 1 2 3 6
0 −5 −10 −20  0 1 2 4
0 −7 −4 2 0 −7 −4 2
Step 2a: This is already nonzero. Step 2c: Add 7 times
Step 2b: Scale to make this 1. the second row to clear this.
 
(There are no fractions because R3 = R3 + 7R2 1 2 3 6
of the optional step before.) 0 1 2 4
0 0 10 30

Note: Step 2 never messes up the first (nonzero) column of the matrix,
because it looks like this:
 
1 ? ? ?
“Active” row  0 ? ? ? 
0 ? ? ?
Row Reduction
Example, continued

   
1 2 3 6 R3 = R3 ÷ 10 1 2 3 6
0 1 2 4 0 1 2 4
0 0 10 30 0 0 1 3
Step 3a: This is already nonzero.
Step 3b: Scale to make this 1.

Note: Step 3 never messes up the columns to the left.

Note: The matrix is now in row echelon form!
   
1 2 3 6 R2 = R2 − 2R3 1 2 3 6
0 1 2 4  0 1 0 −2 
0 0 1 3 0 0 1 3
 
Last step: Add multiples of R1 = R1 − 3R3 1 2 0 −3
the third row to clear these.  0 1 0 −2 
0 0 1 3
Last step: Add −2 times
the third row to clear this.
 
R1 = R1 − 2R2 1 0 0 1
0 1 0 −2 
0 0 1 3
Row Reduction
Example, continued

Success! The reduced row echelon form is

  
1 0 0 1 x = 1
 0 1 0 −2  =⇒ y = −2
0 0 1 3

z = 3
Recap

Get 
a 1 here Clear down Get a 1 here
      Clear down 
? ? ? ? 1 ? ? ? 1 ? ? ? 1 ? ? ?
? ? ? ? ? ? ? ? 0 ? ? ? 0 1 ? ?
       
? ? ? ? ? ? ? ? 0 ? ? ? 0 ? ? ?
? ? ? ? ? ? ? ? 0 ? ? ? 0 ? ? ?

(maybe these are already zero) Get a 1 

here  Clear down Matrix is in REF
   
1 ? ? ? 1 ? ? ? 1 ? ? ? 1 ? ? ?
0 1 ? ? 0 1 ? ? 0 1 ? ? 0 1 ? ?
       
0 0 0 ? 0 0 0 ? 0 0 0 1 0 0 0 1
0 0 0 ? 0 0 0 ? 0 0 0 ? 0 0 0 0

Clear up Clear up Matrix is in RREF

     
1 ? ? ? 1 ? ? 0 1 0 ? 0
0
 1 ? ?
0
 1 ? 0
0
 1 ? 0 Profit?
0 0 0 1 0 0 0 1 0 0 0 1
0 0 0 0 0 0 0 0 0 0 0 0
Row Reduction
Another example

The linear system

 
2x + 10y = −1 2 10 −1
gives rise to the matrix .
3x + 15y = 2 3 15 2
Let’s row reduce it: [interactive row reducer]
R1 = R1 ÷ 2 − 12
   
2 10 −1 1 5
(Step 1b)
3 15 2 3 15 2
R2 = R2 − 3R1 − 12
 
1 5
7 (Step 1c)
0 0 2
R2 = R2 × 27
− 12
 
1 5
(Step 2b)
0 0 1
R1 = R1 + 1 R2  
2 1 5 0
(Step 2c)
0 0 1

The row reduced matrix

 
1 5 0 corresponds to the x + 5y = 0
0 0 1 inconsistent system 0 = 1.
Inconsistent Matrices

Question
What does an augmented matrix in reduced row echelon form look like, if its
system of linear equations is inconsistent?

Answer:  
1 0 ? ? 0
0 1 ? ? 0
0 0 0 0 1

An augmented matrix corresponds to an inconsistent system

of equations if and only if the last (i.e., the augmented)
column is a pivot column.
Section 1.3
Parametric Form
Example 1

How do we solve a system of linear equations if the row reduced matrix has a
column without a pivot? Let’s do an example.
 
2x + y + 12z = 1 2 1 12 1
gives rise to the matrix .
x + 2y + 9z = −1 1 2 9 −1
Let’s row reduce it: [interactive row reducer]
R1 ←→ R2
   
2 1 12 1 1 2 9 −1
(Optional)
1 2 9 −1 2 1 12 1
R2 = R2 − 2R1
 
1 2 9 −1
(Step 1c)
0 −3 −6 3
R2 = R2 ÷ −3
 
1 2 9 −1
(Step 2b)
0 1 2 −1
R1 = R1 − 2R2
 
1 0 5 1
(Step 2c)
0 1 2 −1

The row reduced matrix

  
1 0 5 1 corresponds to the x + 5z = 1
0 1 2 −1 linear system y + 2z = −1
Example 1
Continued

The system
x + 5z = 1
y + 2z = −1
comes from a matrix in reduced row echelon form. Are we done? Is the system
solved?
Yes! Rewrite:
x = 1 − 5z
y = −1 − 2z
For any value of z, there is exactly one value of x and y that makes the
equations true. But z can be anything we want! For this reason, we call z a
free variable, and it corresponds to a column with no pivot.
So we have found the solution set: it is all values x, y , z where
x = 1 − 5z
y = −1 − 2z for z any real number.
(z = z)
This is called the parametric form for the solution set. [interactive picture]
For instance, (1, −1, 0) and (−4, −3, 1) are solutions.
Free Variables

Definition
Consider a consistent linear system of equations in the variables x1 , . . . , xn . Let
A be a row echelon form of the matrix for this system.
We say that xi is a free variable if its corresponding column in A is not a pivot
column.

Important
1. You can choose any value for the free variables in a
(consistent) linear system.
2. Free variables come from columns without pivots in a
matrix in row echelon form.

In the previous example, z was free because the reduced row echelon form
matrix was  
1 0 5 4
.
0 1 2 −1
Example 2

Suppose the reduced row echelon form of the matrix for a linear system in
x1 , x2 , x3 , x4 is  
1 0 0 3 2
0 0 1 4 −1
The free variables are x2 and x4 : they are the ones whose columns are not pivot
columns.
This translates into the system of equations

x1 + 3x4 = 2 x1 = 2 − 3x4
=⇒
x3 + 4x4 = −1 x3 = −1 − 4x4.
What happened to x2 ? What is it allowed to be? Anything! The general
solution is
(x1 , x2 , x3 , x4 ) = (2 − 3x4 , x2 , −1 − 4x4 , x4 )
for any values of x2 and x4 .
For instance, (2, 0, −1, 0) is a solution (x2 = x4 = 0), and (5, 1, 3, −1) is a
solution (x2 = 1, x4 = −1).

The boxed equation is called the parametric form of the general

solution to the system of equations. It is obtained by moving all
free variables to the right-hand side of the =.
Example 3

Solve the system of linear equations in x1 , x2 , x3 , x4 :

x1 + 5x3 =0
x4 = 0

So the associated matrix is:

 
1 0 5 0 0
0 0 0 1 0

To solve, we move the free variable to the right. The parametric form for the
solution set is

x1 = −5x3
x2 = x2 (free)
x3 = x3 (free)
x4 = 0

Or: (−5x3 , x2 , x3 , 0). The solution set is a plane because it has two free
variables. Each solution has four entries, so the solution set is a plane in R4 .
Free variables
Geometry, looking ahead!

If we have a consistent system of linear equations, with n variables and k free

variables, then the set of solutions is a k-dimensional object in Rn . We will
make this precise later, but it is worth thinking about now.

Why does this make sense?

z
t = −1 t=0 (t, w ) = (−1, 1)
(t, w ) = (1, −1)
t=1
y
x

(t, w ) = (2, 2)
Poll

A linear system has 4 variables and 3 equations.

What are the possible solution sets?

(a) point
(b) two points
(c) line
(d) plane
(e) 3-dimensional object
(f) 4-dimensional object
Yet Another Example

The linear system

x +y +z =1 has matrix form 1 1 1 1 .

This is in reduced row echelon form. The free variables are y and z. The
parametric form of the general solution is

x = 1 − y − z.

Rearranging:
(x, y , z) = (1 − y − z, y , z),
where y and z are arbitrary real numbers.
[interactive]
Trichotomy

There are three possibilities for the reduced row echelon form of the augmented
matrix of a linear system.
1. The last column is a pivot column.
In this case, the system is inconsistent. There are zero solutions, i.e. the
solution set is empty. Picture:
 
1 0 0
0 1 0
0 0 1
2. Every column except the last column is a pivot column.
In this case, the system has a unique solution. Picture:
 
1 0 0 ?
0 1 0 ?
0 0 1 ?

3. The last column is not a pivot column, and some other column isn’t either.
In this case, the system has infinitely many solutions, corresponding to the
infinitely many possible values of the free variable(s). Picture:
 
1 ? 0 ? ?
0 0 1 ? ?
Summary

I Row reduction is an algorithm for solving a system of linear equations

represented by an augmented matrix.
I The goal of row reduction is to put a matrix into (reduced) row echelon
form, which is the “solved” version of the matrix.
I An augmented matrix corresponds to an inconsistent system if and only if
there is a pivot in the augmented column.
I Columns without pivots in the RREF of a matrix correspond to free
variables. You can assign any value you want to the free variables.
I We can tell whether a linear system has zero, one, or infinitely many
solutions using the RREF of the augmented matrix.