Three Elementary Row Operations

Why Row Operations Are Useful

A system of linear equations can be recorded in a matrix, so calculations on its rows provide a compact way to organize calculations on the equations. For a system with \(m\) equations and \(n\) unknowns, its coefficient matrix has \(m\) rows and \(n\) columns: each row contains the coefficients from one equation. The augmented matrix includes one additional column containing the right-hand-side constants. For example, the system \(x+2y=5\), \(3x-y=4\) has coefficient matrix \(\begin{bmatrix}1&2\\3&-1\end{bmatrix}\) and augmented matrix \(\begin{bmatrix}1&2&5\\3&-1&4\end{bmatrix}\).

An elementary row operation changes one or more rows according to a specific rule. When working with a coefficient matrix, the operation acts on its coefficient entries. When working with an augmented matrix, it acts on the entire selected row, including its right-hand-side entry. Keeping this distinction in view prevents a frequent error: treating the last column as though it were exempt from the operation.

The scalar field here is \(\mathbb{R}\), so the multiplier \(c\) may be any real number subject to the stated conditions. For a row replacement, \(c\) may be zero; in that case the row does not change. By contrast, a row scaling requires \(c\ne0\). The two row indices in a row replacement must be different: the operation uses row \(j\) to update row \(i\), leaving row \(j\) as it was.

Row Swaps

A row swap exchanges the complete contents of two rows. It can be useful when a desired row is not in the position where you need it. In an augmented matrix, the constant must move with its equation; swapping only the coefficients would mix up which constant belongs to which equation.

Nonzero Row Scalings

To scale a row, multiply every entry in it by the same nonzero number. This includes a right-hand-side constant when the matrix is augmented. Scaling by a fraction is allowed, and it can make entries easier to work with. The requirement \(c\ne0\) matters: if \(c=0\), the original row is erased, and the operation cannot be undone by multiplying by a reciprocal.

Row Replacements

In a row replacement, a multiple of one row is added to a different row. The source row remains unchanged, while every entry of the target row is updated. The same addition must be carried out across the whole row. Choosing the multiplier appropriately can produce a zero in a coefficient position, but the operation itself is defined whether or not it produces a zero.

Reversing Each Operation

Undoing a row operation provides a direct way to check the arithmetic. A swap undoes itself. A nonzero scaling by \(c\) is undone by scaling by \(1/c\). A replacement \(R_i\to R_i+cR_j\) is undone by \(R_i\to R_i-cR_j\), because row \(j\) was not changed. These inverse operations act on coefficient and augmented matrices in the same way as the original operations.

Proof. Let the original matrix be any matrix with the rows involved in the operation. If the operation is a swap \(R_i\leftrightarrow R_j\), swapping \(R_i\) and \(R_j\) again restores both rows to their original positions, and all other rows were unchanged. If the operation is a scaling \(R_i\to cR_i\), its hypothesis gives \(c\ne0\); scaling that row by \(1/c\) gives \((1/c)(cR_i)=R_i\), and leaves every other row unchanged. If the operation is a replacement \(R_i\to R_i+cR_j\) with \(i\ne j\), row \(j\) is unchanged. Replacing the updated row by \(R_i-cR_j\) therefore gives \((R_i+cR_j)-cR_j=R_i\), while all other rows remain unchanged. These cases cover all three operations and prove the claim. \(\square\)

This result gives a simple checking procedure for a sequence of row operations. Record each operation, then undo them in reverse order: reverse the last operation first, and continue until the first. If the original matrix is recovered, the sequence passes this arithmetic check. This checks that the matrix calculations were performed and recorded consistently; deciding what a row operation implies about a system’s solutions is a separate question.

Common Pitfalls

Leaving the constant unchanged when scaling an augmented row. In the scaling example, a mistaken calculation might change the first row of the augmented matrix from \((2,-1,3)\) to \((1,-\tfrac12,3)\), multiplying the coefficients by \(\tfrac12\) but leaving the constant at \(3\). That is not the result of \(R_1\to \tfrac12R_1\). The corrected row is \((1,-\tfrac12,\tfrac32)\), since \(\tfrac12(2,-1,3)=(1,-\tfrac12,\tfrac32)\).

Updating the source row during a replacement. In \(R_3\to R_3+2R_1\), row 1 is the source and must remain as it was. If a calculation changes row 1 as well, it has performed a different operation. The replacement example can be checked by confirming that the first two rows are unchanged and that subtracting twice row 1 from the new row 3 recovers the original row 3.

Scaling by zero is not an allowed row scaling. The expression \(R_i\to0R_i\) replaces row \(i\) by a zero row, and no scaling can recover the original nonzero row from it. This is why the definition requires a nonzero multiplier. Row replacement, in contrast, allows any real multiplier; when that multiplier is zero, the target row is simply unchanged.

A Reliable Calculation Routine

Before carrying out a row operation, identify the target row and, when needed, the source row. Then apply the operation to every entry in the target row; for an augmented matrix, include the right-hand-side column. Finally, check the result using the inverse operation. Keeping these steps separate makes it easier to locate an arithmetic error and makes the written row-operation notation an accurate record of the calculation.