MATH 231 Lecture Notes

Gaussian Elimination

In what follows we focus on solving systems of m linear equations in n unknown variables x1 , . . . ,

xn . In order for the system to be linear, each time a variable x j appears in some term, none of the
others appear in that same term, and x j appears to the first power. Essentially, this means if any x j
is in an equation, its appearance (when simplified) is in a term like ax j ; i.e., rescaled by a constant.

We set forward examples and solve them using the standard method discussed in high school
algebra courses: elimination.

Example 1:
#
x ´ 2y “ 1
Consider the system of equations:
3x ` 2y “ 11

As equations: A
« matrix storing
ff just the coefficients:
x ´ 2y “ 1 1 ´2 1
3x ` 2y “ 11 3 2 11

Replacing the 2nd equation: R2 ´ 3R1 Ñ R2 : « ff

x ´ 2y “ 1 1 ´2 1
8y “ 8 0 8 8

The first variable (x) remains in the first equation, but has been eliminated from the 2nd equation.
Note that we could achieve this working exclusively on the matrix of coefficients (right side)—
subtracting three multiples of the first row from the row. The resulting matrix is said to be in
row echelon form. (Follow the link https://en.wikipedia.org/wiki/Row_echelon_form to
learn what this term means.)

Pause to think through how, using only this last matrix, you would determine the values of
the variables x and y.

We could carry out a few more algebraic steps:

Divide 2nd equation by 8: p1{8qR2 Ñ R2 Associated

« ffmatrix:
x ´ 2y “ 1 1 ´2 1
y “ 1 0 1 1

Replacing the 1st equation: R1 ` 2R1 Ñ R1 : « ff

x “ 3 1 0 3
y “ 1 0 1 1

From the resulting matrix, said to be in reduced row echelon form, it is even easier to pick
o↵ the values of x and y. (The Wikipedia link explains this term, too.) Having begun with the
equations of two lines in the plane, we can now say those lines intersect at the unique point

2
MATH 231 Lecture Notes

px, yq “ p3, 1q.

It is crucial that you stop and affix in your mind the meanings of row echelon form (often called
simply echelon form) and reduced row echelon form (or RREF), as well as learn to identify when
a given matrix form has such a form. When you have read the Wikipedia descriptions (linked
above) and feel you are ready, try to identify which of the following matrices have echelon form,
RREF (which is just a special type of echelon form), or neither. You can learn whether your
answers are correct when you do the Socrative quiz at the end. » fi
» fi » fi 1 0 0 0
0 0 4 ´1 0 1 1 0 1 — ffi
— ffi — ffi —1 1 0 0ffi
(a) –0 0 0 0 0fl (c) –0 0 1 1fl (e) —— ffi
0 1 1 0ffi
– fl
0 0 0 0 3 0 0 0 0
0 0 1 1
» fi » fi
0 0 4 1 1 1 0 0 » fi
— ffi — ffi 0 1 1 1 1
(b) – 0 5 0 2fl (d) –0 1 1 0fl — ffi
—0 0 2 2 2ffi
´2 1 0 3 0 0 1 1 — ffi
(f) —
—0 0 0 0 3ffi
ffi
— ffi
–0 0 0 0 0fl
0 0 0 0 0

We will work toward solving systems of linear equations using only matrices. But let’s work
another example with equations and matrices side-by-side, one in which an oddity occurs.

Example 2:
#
3x ´ 6y “ 11
Consider the system of equations:
2x ´ 4y “ 8

As equations The
« corresponding
ff matrix:
3x ´ 6y “ 11 3 ´6 11
2x ´ 4y “ 8 2 ´4 8

Divide the 1st equation by 3: p1{3qR1 Ñ R1 : « ff

x ´ 2y “ 11{3 1 ´2 11{3
2x ´ 4y “ 8 2 ´4 8

Now R2 ´ 2R1 Ñ R2 : « ff
x ´ 2y “ 11{3 1 ´2 11{3
0 “ 2{3 0 0 2{3

Down the left column, our elimination steps resulted in the removal of both x and y. Or
focusing on how these same operations have a↵ected the original matrix, we see row 2 has
zeros in the columns corresponding to x and y. The final entry, the one that corresponds to the

3
MATH 231 Lecture Notes

right side of the equation, is not zero, however, giving us the nonsense statement 0 “ 2{3.

Summarizing the matrix (right column) version of things, we began with a matrix that corre-
sponded to our system, reduced it to row echelon form, and the final version o↵ers us a row
with the untenable statement 0 “ 2{3. We conclude the system has no solution. When you
consider the original equations, it’s not difficult to see they expressed two parallel lines, so
there is no point of intersection to be found.

Example 3:
$
’
& 3x1 ´ x2 ` x3 “ ´2
Let’s look at a system with 3 variables: ´x1 ` x2 ` x3 “ 4
’
%
5x1 ` 2x2 ` 9x3 “ 15

This time I will form the corresponding matrix, and work exclusively with it, until (row)
echelon form is reached.
» fi » fi
3 ´1 1 ´2 ´1 1 1 4
— ffi — ffi
–´1 1 1 4fl R1 Ø R2 – 3 ´1 1 ´2fl
5 2 9 15 5 2 9 15
» fi
1 ´1 ´1 ´4
— ffi
p´1qR1 Ñ R1 –3 ´1 1 ´2fl
5 2 9 15
» fi
1 ´1 ´1 ´4
— ffi
p´3qR1 ` R2 Ñ R2 –0 2 4 10fl
5 2 9 15
» fi
1 ´1 ´1 ´4
— ffi
p´5qR1 ` R3 Ñ R3 –0 2 4 10fl
0 7 14 35
» fi
1 ´1 ´1 ´4
— ffi
p1{2qR2 Ñ R2 –0 1 2 5fl
0 7 14 35
» fi
1 ´1 ´1 ´4
— ffi
R3 ´ 7R2 Ñ R3 –0 1 2 5fl
0 0 0 0

We have arrived at (row) echelon form and, with one more step, namely adding row 2 to row

4
MATH 231 Lecture Notes

1, we arrive at RREF (reduced row echelon form):

» fi
1 0 1 1
— ffi
R1 ` R2 Ñ R1 –0 1 2 5fl .
0 0 0 0

Translating back to equations, this RREF says

x1 ` x3 “ 1
x2 ` 2x3 “ 5
0 “ 0

The last equation is vacuous (doesn’t add any information), but it is not false. A general
principle is that variables provide freedoms while equations act as constraints, taking freedoms
away. Applying this principle to the example at hand would predispose us to guess that the
three freedoms were balanced by three constraints, resulting in precisely one solution point
px1 , x2 , x3 q (or even that there might be no solution, as in Example 2, when the two variables
were subjected to two incompatable constraints). While there are many linear systems of 3
equations in 3 unknowns that go that way, the case here is that echelon form has revealed these
three equations actually place only two constraints on the variables (two meaningful ones,
along with the uninformative statement 0 “ 0). So, we have the situation that one variable
can have its values chosen freely (i.e., one degree of freedom remains), and once it is chosen,
the values of the other variables are fixed.

It is convenient to solve the above (informative equations) for x1 and x2 respectively:

+ #
x1 ` x3 “ 1 x1 “ 1 ´ x3
ñ
x2 ` 2x3 “ 5 x2 “ 5 ´ 2x3

and think of x3 as a free variable. Our original system of equations has infinitely many
solutions, one for each choice of x3 P R. We will generally express these solutions in vector
form: » fi » fi » fi » fi
x1 1 ´ x3 1 ´1
— ffi — ffi — ffi — ffi
–x2 fl “ –5 ´ 2x3 fl “ –5fl ` x3 –´2fl , x3 P R.
x3 x3 0 1
Note, finally, that had we used the notation of Calculus (first introduced in MATH 172, and
used further in MATH 271), our answer would have been

hx1 , x2 , x3 i “ h1, 5, 0i ` t h´1, ´2, 1i , t P R,

which, depending on your retention of Calculus concepts, you might recognize as expressing
(parametrically) a line in space. If you also recall that equations such as

ax ` by ` cz “ d

5
MATH 231 Lecture Notes

in variables x, y, and z have graphs which are planes, then the geometric interpretation of this
problem is that we were looking for points of intersection between three planes

3x1 ´ x2 ` x3 “ ´2, ´x1 ` x2 ` x3 “ 4, and 5x2 ` 2x2 ` 9x3 “ 15,

and learned that these planes intersect not in just a single point, but along an entire line, as if
they were separate pages in a book with a common spine.

The subject of this handout is Gaussian elimination, which is what we call it when we work
with the matrix of a linear system of equations and take it to row echelon form (or even further,
to reduced row echelon form). I hope it becomes obvious that, once we have echelon form, we
can easily separate those systems that have no solution from those that do, and further distinguish
solvable (or consistent) systems that have precisely one solution from those that have infinitely
many solutions (i.e, where at least one degree of freedom remains). The real goal of Gaussian
elimination is, then, twofold:

• to distinguish between systems of linear equations having no solution, precisely one solution,
and infinitely many solutions, and

• for those systems that have solutions, identifying them, as we did in Example 1 where there
was the lone solution h3, 1i, and in Example 3 where there were infinitely many hx1 , x2 , x3 i “
h1, 5, 0i ` t h´1, ´2, 1i, t P R.

It is the process of morphing from the linear system’s original matrix to an echelon form which
comprises the real work of Gaussian elimination. There are just three types of steps/transformations
which are legal. These three “moves” are collectively known as elementary row operations (EROs),
and they are:

• swapping two rows, sometimes denoted by Ri Ø R j . You can see how, if a system led to the
matrix on the left, a row swap would be just the thing to quickly arrive at row echelon form.
« ff « ff
0 2 ´1 3 1 0 5 0
R1 Ø R2 .
1 0 5 0 0 2 ´1 3

• multiplying a row by a nonzero constant, sometimes denoted by cRi Ñ Ri . The first step of
Example 2 was an instance of this ERO:
« ff « ff
3 ´6 11 1 ´2 11{3
p1{3qR1 Ñ R1 .
2 ´4 8 2 ´4 8

• adding a multiple of one row to another row, sometimes denoted by cRi ` R j Ñ R j . We

have done this in several examples, with one instance being the second step of Example 2:
« ff « ff
1 ´2 11 1 ´2 11{3
p´2qR1 ` R2 Ñ R2 .
2 ´4 8 0 0 2{3

6
MATH 231 Lecture Notes

If you own a TI-8x calculator and think you would like to take advantage of its functionality in carry-
ing our EROs, I suggest you watch this video https://www.youtube.com/watch?v=ePtrvmMUMXU.
The man doing the demonstration shows you both how to take a matrix directly to RREF in one
step, as well how to carry out using individual EROs. While the RREF function is very useful, I
expect you to be able to do the individual EROs—both to be able to map out a sequence of EROs
that lead to echelon form (or to RREF), and to carry them out successfully.

Another good video using only the calculator’s RREF function, but doing so on at least one problem
with a free variable, is this one https://www.youtube.com/watch?v=_4aKp_ZbTEI.

Work through the examples in this handout and the videos to the point that you feel you have a
good handle on the material. When you are ready, proceed to take the quiz.

Socrative quiz

Some people will have installed a Socrative Student app on their phones. If you haven’t used Socra-
tive before and do not have the app, the easiest way to access the quiz is to point a web browser at
the website https://b.socrative.com/login/student/, enter as Room Name SCOFIELD3894,
hit ”Join”, and take the quiz called ”MATH 231: GE-Intro”. You will need to be able to view the
matrices (a)–(f) in the box at the top of the 2nd page of this handout while taking the quiz.