Linear Systems: REDUCED ROW ECHELON FORM

From both a conceptual and computational point of view, the trouble with using the echelon form to
describe properties of a matrixA is that A can be equivalent to several different echelon forms because
rescaling a row preserves the echelon form - in other words, there's no unique echelon form for A. This
leads us to introduce the next

Definition: a matrix is said to be in Reduced Row Echelon Form if

the leading entry in each non-zero row is 1,

each leading 1 is the only non-zero entry in its column.

A typical structure for a matrix in Reduced Row Echelon Form is thus

⎡1 ∗ 0 0 ∗ 0 ∗ ⋯ ∗⎤
⎢
⎢0
∗⎥
∗⎥
0 0 1 0 ∗ 0 ∗ ⋯
⎢
⎢
0 0 1 ∗ 0 ∗ ⋯ ⎥
⎥
⎢
⎢ ∗⎥
⎥
⎢0 0⎥
0 0 0 0 0 1 ∗ ⋯
⎣0 0⎦
0 0 0 0 0 0 ⋯
0 0 0 0 0 0 ⋯

Note that this matrix is still in echelon form but each pivot value is 1, and all the entries in a pivot column
are 0 except for the pivot itself. As the pivot values cannot now be rescaled, however, the next result
should come as no surprise:

Main Reduced Row Echelon Theorem: each matrix is row equivalent to

Uniqueness of the reduced row echelon form is a property we'll make fundamental use of as the
semester progresses because so many concepts and properties of a matrix A can then be described in
terms of rref(A). But first let's investigate how the presence of the 1 and 0's in the pivot column affects
the Gauss Elimination method for solving three particular systems of linear equations in 3 variables.
 But then by elementary row operations acting upwards:
I. Return to the system
1
⎡1 2
R3 → R3 −1 1⎤
R2 −R3 ⎡
1 2 −1 1⎤
y+z = 4,
⎣0 0 2⎦ ⎣0 0 2⎦
B −−−−−−→ 0 1
4
1 4 −−−−→ 0 1 0 2
3x + 6y − 3z = 3, 1 1
−2x − 3y + 7z = 10 ,
⎡1 2
R1 + R3
0 3⎤
R1 −2 R2 ⎡
1 0 0 −1 ⎤
⎣0 0 2⎦ ⎣0 0 2 ⎦
from Lecture 1. Its associated −−−−→ 0 1 0 2 −−−−→ 0 1 0 2 .
augmented matrix was 1 1

⎡ 0 4 ⎤ Consequently,
1 1
⎣ −2 −3 10 ⎦
A = 3 6 −3 3 ,
7 ⎡1 0 0 −1 ⎤
⎣0 0 2 ⎦
rref(A) = 0 1 0 2
and elementary row operations showed 1
that which is the augmented matrix associated with the system

⎡1 2 −1 1⎤ x = −1 , y = 2, z = 2.
⎣0 0 8⎦
A ∼ 0 1 1 4 ( = B, say, )
4 These solutions are exactly the same as before, of course,

associated with the equivalent system because elementary row operations produce equivalent
systems. The point is that the 1's and 0 in the pivot
x + 2y − z = 1, columns eliminate the need for back substitution.
y+z = 4,
4z = 8.

This refinement using the the Reduced Row Echelon Form of the Augmented matrix instead of the
Echelon Form in Gaussian Elimination is usually called Gauss-Jordan Elimination after the German
mathematician Wilhelm Jordan who used it extensively in his writings.
 II. Solve for x, y and z in III. Solve for x, y and z in

y+z = 4, y+z = 4,
3x + 6y − 3z = 3, 3x + 6y − 3z = 3,
−2x − 3y + 3z = 10 . −2x − 3y + 3z = 2.
The associated augmented The associated augmented matrix is
matrix is
⎡ 0 1 1 4⎤
⎡ 0 4 ⎤
⎣ −2 −3 2⎦
1 1 A = 3 6 −3 3 ,
3
⎣ −2 −3 10 ⎦
A = 3 6 −3 3 ,
3
and
and
⎡1 0 −3 7⎤
⎡1 0 0⎤
⎣0 0 0⎦
−3 rref(A) = 0 1 1 4 ,
0
⎣0 0 1⎦
rref(A) = 0 1 1 0 ,
0
associated with the equivalent system
associated with the equivalent
x − 3z = 7,
system
y+z = 4.
x − 3z = 0, There is no equation to specify z, so z is a free
y+z = 0,
variable:
0z = 1.
But there is no value of z such
x = 7 + 3t , y = 4 − t, z = t , (t arbitrary).
that 0z = 1, so the system is This can also be seen from rref(A): only the x- and y-
Inconsistent. We can see this columns are pivot columns.
from rref(A): the augmented
matrix A still has three pivot
columns, but the column of
right hand sides is one of
them!!

The procedure just gone through provides an algorithm for solving a general system of m linear
equations in n variables:

form the associated augmented matrix A and compute rref(A).

If the column of right hand sides is a pivot column of A, then the system is inconsistent, otherwise
the system is consistent.

If the system is consistent, then any variable corresponding to a pivot column is called a basic
variable, otherwise the variable is called a free variable.

Your Turn Now: consider the coefficient matrix for systems I, II, and III.

Compute the reduced row echelon form of each coefficient matrix.

How do these differ from the reduced row echelon matrix of the associated augmented matrix?

Does the number of pivots change? When can you use the reduced row echelon form of each
coefficient matrix to solve a system of linear equations?