What is Gaussian elimination used for?

Gaussian elimination solves systems of linear equations by reducing an augmented matrix to row-echelon form, then applying back-substitution. It also determines whether a system has a unique solution, infinitely many solutions, or no solution. In linear algebra, it forms the basis for computing matrix rank, finding null spaces, and inverting matrices.

What is the difference between row-echelon form and reduced row-echelon form?

Row-echelon form (REF) requires that each leading entry (pivot) sits strictly to the right of the pivot in the row above, and all entries below each pivot are zero. Reduced row-echelon form (RREF) adds two conditions: every pivot equals 1, and every entry above a pivot is also zero. RREF gives the solution directly without back-substitution.

How do I know when a system has no solution versus infinitely many solutions?

After row reduction, a row of the form [0 0 0 | c] where c is nonzero means the system is inconsistent: no solution exists. A row of the form [0 0 0 | 0] means at least one variable is free, so infinitely many solutions exist. The number of free variables equals the number of all-zero rows in row-echelon form.

What are the three elementary row operations?

The three allowed operations are: (1) swap two rows, written Ri ↔ Rj; (2) multiply a row by a nonzero scalar, written cRi → Ri; and (3) add a scalar multiple of one row to another, written Ri + cRj → Ri. Every operation is reversible, so the solution set of the system is unchanged.

Can Gaussian elimination fail due to numerical issues?

Yes. When the pivot entry is close to zero, dividing other rows by it amplifies rounding errors in floating-point arithmetic. Partial pivoting fixes this: before each elimination step, swap rows so the largest absolute value in the pivot column becomes the pivot. Most numerical linear algebra libraries use partial pivoting by default.

How does Gaussian elimination relate to Gauss-Jordan elimination?

Gaussian elimination stops at row-echelon form and uses back-substitution to read off the solution. Gauss-Jordan elimination continues to reduce to reduced row-echelon form, eliminating entries above each pivot as well as below. Gauss-Jordan is slightly more work upfront but avoids back-substitution entirely; both methods produce the same answer.

How long does Gaussian elimination take compared to other methods?

For an n x n system, Gaussian elimination requires approximately (2/3)n^3 floating-point operations, which grows cubically with system size. For small systems (2x2, 3x3) on an exam, it completes in a few minutes by hand. Cramer's rule grows factorially and is impractical for systems larger than 3x3, making Gaussian elimination the standard for exam and computational work alike.

Do I need to find a common denominator when working with fractions in row reduction?

No. You can work with fractions throughout the row operations. The scalar multiplier in the row operation Ri + cRj handles any coefficient, integer or fractional. Clearing fractions by multiplying a row by a common denominator is optional and sometimes makes arithmetic cleaner, but it is not required and can introduce larger integers that are harder to track.

Subject Mastery

Gaussian Elimination: Step-by-Step Worked Examples

By Jonas|20 September 2026|11 min read

Key Takeaways

Gaussian elimination converts a linear system into an augmented matrix, then reduces it row by row to row-echelon form via three reversible row operations.

Back-substitution reads the solution from the last nonzero row upward; three fully worked examples show every step explicitly.

A zero row with a nonzero constant signals no solution; a zero row with zero constant means a free variable and infinitely many solutions.

Partial pivoting (swap to the largest pivot before each elimination step) prevents division by zero and controls rounding error in floating-point work.

Gaussian elimination solves linear systems by reducing an augmented matrix to row-echelon form through three reversible row operations, then recovering each variable through back-substitution. It handles 2x2, 3x3, and n x n systems with the same procedure, making it the standard algorithm taught in every linear algebra course worldwide.

The three worked examples below progress from a straightforward 2x2 system to a 3x3 system, then to a case where the row-echelon form itself tells you the system has either no solution or infinitely many. Every row operation is written out explicitly so you can see exactly which arithmetic step produces each matrix entry. That step-level visibility is what these solving systems with Gaussian elimination examples are designed to provide.

What Is Gaussian Elimination?

Gaussian elimination is a systematic algorithm for solving systems of linear equations. It works by transforming the coefficient matrix, together with the constants, into a triangular form called row-echelon form, where all entries below the diagonal are zero. Once the matrix reaches that shape, back-substitution recovers each variable in reverse order.

The algorithm appears in every linear algebra course from introductory to graduate level, with full treatment in freely available open-access textbooks. Its O(n^3) complexity makes it practical for any system you will encounter in coursework by hand or computationally.

Augmented Matrix Notation

Every linear system maps directly to an augmented matrix. Given the system:

2x + y = 5
x - 3y = -4

the augmented matrix is written with the coefficients on the left of a vertical divider and the constants on the right:

The augmented matrix encodes both the coefficients (left of the divider) and the constants (right). Row operations on this matrix never change the solution set.

The Three Allowed Row Operations

Only three operations are permitted on the rows of an augmented matrix. Each one is reversible, which guarantees the solution set stays the same before and after the operation.

Operation	Notation	What it does	Example
Row swap	Ri ↔ Rj	Exchange row i and row j	R1 ↔ R2 swaps the first and second rows
Row scale	cRi → Ri	Multiply every entry in row i by nonzero scalar c	(1/2)R1 → R1 halves all entries in row 1
Row replacement	Ri + cRj → Ri	Add c times row j to row i, storing the result in row i	R2 + (-1/2)R1 → R2 eliminates the x term in row 2

OperationRow swap

NotationRi ↔ Rj

What it doesExchange row i and row j

ExampleR1 ↔ R2 swaps the first and second rows

OperationRow scale

NotationcRi → Ri

What it doesMultiply every entry in row i by nonzero scalar c

Example(1/2)R1 → R1 halves all entries in row 1

OperationRow replacement

NotationRi + cRj → Ri

What it doesAdd c times row j to row i, storing the result in row i

ExampleR2 + (-1/2)R1 → R2 eliminates the x term in row 2

These three operations are the only moves in Gaussian elimination. Row replacement is the most frequent; it drives the elimination phase.

Why these operations preserve the solution set

Each operation corresponds to an algebraically valid step: swapping equations, multiplying an equation by a nonzero constant, or adding a multiple of one equation to another. Because every step is reversible, the augmented matrix at any stage represents a system with exactly the same solutions as the original.

Example 1: A 2x2 System (Unique Solution)

Solve the following system using solving systems with Gaussian elimination step by step:

2x + y = 5
x - 3y = -4

Setting Up the Augmented Matrix

Write the coefficients and constants as the augmented matrix:

[ 2 1 | 5 ]
[ 1 -3 | -4 ]

The goal of the forward elimination phase is to produce a zero below the leading entry in column 1.

Row Reduction to Row-Echelon Form

The pivot in position (1,1) is 2. To eliminate the entry 1 directly below it in row 2, compute the multiplier m = 1/2, then apply R2 + (-1/2)R1 to row 2.

Step 1: R2 + (-1/2)R1 → R2

New row 2: [1 - (1/2)(2), -3 - (1/2)(1), -4 - (1/2)(5)]
= [0, -7/2, -13/2]

The matrix is now in row-echelon form:

[ 2 1 | 5 ]
[ 0 -7/2 | -13/2 ]

Avoid arithmetic errors: track fractions immediately

The most common error at this stage is rounding -13/2 to -6 or -7. Keep the fraction exact. A single rounding error in the augmented column propagates into every back-substitution step that follows. Write -13/2 as a fraction rather than converting to a decimal mid-problem.

Back-Substitution

Starting from the last nonzero row: (-7/2)y = -13/2. Divide both sides by -7/2, which is the same as multiplying by -2/7:

y = (-13/2) x (-2/7) = 13/7

Substitute y = 13/7 into row 1: 2x + (13/7) = 5

2x = 5 - 13/7 = 35/7 - 13/7 = 22/7
x = 11/7

Solution: x = 11/7, y = 13/7. You can verify: 2(11/7) + 13/7 = 22/7 + 13/7 = 35/7 = 5. Check.

Example 2: A 3x3 System (Unique Solution)

This is the type of system that appears in linear algebra Gaussian elimination exam questions most frequently. Solve:

x + 2y - z = 8
-3x - y + 2z = -11
-2x + y + 2z = -3

Three row operations transform the initial 3x3 augmented matrix into row-echelon form. Each pivot (magenta box) generates the elimination multipliers for all rows below it.

Forward Elimination Phase

Write the augmented matrix:

[ 1 2 -1 | 8 ]
[ -3 -1 2 | -11 ]
[ -2 1 2 | -3 ]

Step 1: Pivot is 1 in position (1,1). Eliminate the entries in column 1 of rows 2 and 3.

R2 + 3R1 → R2: [-3+3(1), -1+3(2), 2+3(-1), -11+3(8)] = [0, 5, -1, 13]
R3 + 2R1 → R3: [-2+2(1), 1+2(2), 2+2(-1), -3+2(8)] = [0, 5, 0, 13]

Matrix after Step 1:

[ 1 2 -1 | 8 ]
[ 0 5 -1 | 13 ]
[ 0 5 0 | 13 ]

Step 2: Pivot is 5 in position (2,2). Eliminate the entry in column 2 of row 3.

R3 + (-1)R2 → R3: [0-0, 5-5, 0-(-1), 13-13] = [0, 0, 1, 0]

Row-echelon form:

[ 1 2 -1 | 8 ]
[ 0 5 -1 | 13 ]
[ 0 0 1 | 0 ]

Back-Substitution Phase

Start from row 3: z = 0.

Row 2: 5y - 1(0) = 13, so 5y = 13, y = 13/5.

Row 1: x + 2(13/5) - 1(0) = 8, so x + 26/5 = 8, x = 8 - 26/5 = 40/5 - 26/5 = 14/5.

Solution: x = 14/5, y = 13/5, z = 0. Verify in row 2: 0 + 5(13/5) - 1(0) = 13. Check.

3 pivots

confirm a unique solution

When a 3x3 system produces three nonzero pivots in row-echelon form, exactly one solution exists: the system has rank 3.

If you want to check your row operations or explore how changes in coefficients affect the solution, the matrix calculator on the subject calculators hub handles augmented matrices step by step.

Matrix Calculator

Enter your augmented matrix to check row reduction steps and verify solutions for solving systems with Gaussian elimination practice problems.

Open matrix calculator

Example 3: A 3x3 System with No Unique Solution

Solving systems with Gaussian elimination practice problems often include a case where the row-echelon form itself tells you the outcome before you attempt back-substitution. Consider:

x + y + z = 6
2x + 2y + 2z = 12
x - y + 3z = 4

Augmented matrix:

[ 1 1 1 | 6 ]
[ 2 2 2 | 12 ]
[ 1 -1 3 | 4 ]

Step 1: Pivot 1 in position (1,1).

R2 + (-2)R1 → R2: [0, 0, 0, 0]
R3 + (-1)R1 → R3: [0, -2, 2, -2]

After Step 1:

[ 1 1 1 | 6 ]
[ 0 0 0 | 0 ]
[ 0 -2 2 | -2 ]

Row 2 is all zeros. The entry in column 2 of row 2 is zero, so it cannot serve as the next pivot. Swap rows 2 and 3 to bring the nonzero row into position.

R2 ↔ R3:

[ 1 1 1 | 6 ]
[ 0 -2 2 | -2 ]
[ 0 0 0 | 0 ]

Row-echelon form shows only two pivots (positions (1,1) and (2,2)) for a 3x3 system. The third row is entirely zero. This means z is a free variable. Infinitely many solutions exist.

Inconsistent versus Dependent Systems

The bottom row of the row-echelon form tells you the solution type before you attempt back-substitution. Three nonzero pivots means one solution; a zero row with a zero constant means free variables; a zero row with a nonzero constant means no solution.

For the dependent system in Example 3, let z = t (a free parameter). Then from row 2: -2y + 2t = -2, so y = 1 + t. From row 1: x + (1 + t) + t = 6, so x = 5 - 2t. The full solution family is:

x = 5 - 2t, y = 1 + t, z = t, for any real number t.

Every value of t produces a valid solution. At t = 0: (5, 1, 0). At t = 1: (3, 2, 1). The MIT OCW column-space and null-space notes cover free variables and parametric solutions in detail, including the relationship between null space and the solution family.

Common Errors in Gaussian Elimination

Most arithmetic mistakes in these solving systems with Gaussian elimination practice problems come from three sources: incorrect multipliers, sign errors during row replacement, and skipping the row swap when a pivot entry is zero.

Error	Where it happens	How to prevent it
Wrong multiplier sign	R2 + cR1 step: using +c instead of -c	Write the multiplier as m = -(entry below pivot / pivot) and check before computing
Partial row update	Updating only 2 of 4 entries in a row after replacement	Work left to right across every column including the augmented column
Forgetting the augmented column	Reducing A correctly but omitting the \| b column update	Treat the augmented column as a fourth column in all row operations
Skipping a row swap	Trying to divide by a zero pivot	Before each elimination step, check the pivot is nonzero; swap if needed
Premature rounding	Rounding -13/2 to -7 mid-problem	Keep fractions exact until the final back-substitution result

ErrorWrong multiplier sign

Where it happensR2 + cR1 step: using +c instead of -c

How to prevent itWrite the multiplier as m = -(entry below pivot / pivot) and check before computing

ErrorPartial row update

Where it happensUpdating only 2 of 4 entries in a row after replacement

How to prevent itWork left to right across every column including the augmented column

ErrorForgetting the augmented column

Where it happensReducing A correctly but omitting the | b column update

How to prevent itTreat the augmented column as a fourth column in all row operations

ErrorSkipping a row swap

Where it happensTrying to divide by a zero pivot

How to prevent itBefore each elimination step, check the pivot is nonzero; swap if needed

ErrorPremature rounding

Where it happensRounding -13/2 to -7 mid-problem

How to prevent itKeep fractions exact until the final back-substitution result

All five errors are preventable by writing out each row operation explicitly before entering the arithmetic.

When to Swap Rows: Partial Pivoting

Partial pivoting means scanning the current pivot column below the diagonal and swapping the row with the largest absolute value to the pivot position before eliminating. On exam questions by hand, it prevents division by zero (when the natural pivot is exactly zero). In numerical computation on a computer, it controls rounding error growth: dividing by a near-zero number amplifies any floating-point error in the subtracted row.

The rule is simple: before each elimination step, if the current diagonal entry is zero, find the first nonzero entry below it in the same column and swap. For how to do solving systems with Gaussian elimination on an exam, always check the pivot before computing the multiplier.

Swapping after you have already eliminated is too late

If you discover a zero pivot after performing row replacement on all lower rows, you cannot simply swap to fix it. You must redo the elimination from that stage. Check the pivot before each elimination step, not after. A 10-second check at the start of each column saves several minutes of redoing arithmetic.

For further practice with worked examples spanning elimination, eigenvalues, and determinants, the subject calculators hub links to interactive tools for linear algebra and related topics. The u-substitution worked examples and matrix multiplication guide from the subject-mastery cluster follow the same explicit step-level format used here.

If you want to run through Gaussian elimination problems interactively and get feedback on your row operations rather than just a final answer:

Try the AI tutor free

The university resources hub also links to the grade calculators hub and subject tools for every quantitative course.

Key Takeaways

Gaussian elimination converts a linear system into an augmented matrix, applies row operations to reach row-echelon form, then solves by back-substitution. The three row operations (swap, scale, replace) never change the solution set.
In a 2x2 system, one row replacement step eliminates the lower entry, and a single back-substitution step recovers both variables. Keep fractions exact throughout.
In a 3x3 system, two rounds of elimination target columns 1 and 2 in turn. Each round uses multipliers computed as -(entry below pivot) divided by the pivot. Writing out each operation before the arithmetic prevents sign errors.
The row-echelon form itself reveals the solution type: n pivots for n unknowns means exactly one solution; fewer pivots with all-zero constant entries means free variables and infinitely many solutions; an all-zero row with a nonzero constant means no solution.
Partial pivoting (swapping to place the largest absolute-value entry in the pivot position) avoids division by zero and limits rounding error in floating-point computation. Apply it before each elimination step when the current pivot is zero.
The five most common arithmetic errors are: wrong-sign multipliers, partial row updates, omitting the augmented column, skipping row swaps, and premature rounding. All five are eliminated by writing each operation symbol before computing the numbers.
For practice beyond these three examples, the game theory worked examples and other subject-mastery sibling posts use the same explicit step-level format to build problem-solving fluency.

Back-substitution works upward: solve the last nonzero row first (z), substitute into the row above (y), then substitute both into the first row (x). Each step uses only already-solved variables.

The three solving systems with Gaussian elimination examples above cover every outcome a linear algebra exam can produce from this method. Recognizing which outcome applies from the row-echelon form, before any back-substitution arithmetic, is the single most time-saving skill in this topic.

For open-textbook treatment of the same material, OpenStax College Algebra Chapter 9 and the MIT OCW 18.06 Linear Algebra materials provide a full formal derivation of why row operations preserve solution sets, including the connection to elementary matrices. The elimination with matrices lecture notes from the same course show how this scales to large n x n systems computationally.