What Row Reduction Remembers

Row reduction is usually introduced as a reliable sequence of instructions: exchange two rows, multiply a row by a nonzero scalar, or add a multiple of one row to another. The procedure is familiar, but its geometry can remain hidden. If two equations describe lines in the plane, a row operation may replace one of those lines by a visibly different line. Why does the solution remain unchanged, and what information survives after the original equations have disappeared?

The short answer is that row reduction changes the generating equations but preserves their row space. Reduced row echelon form is the unique, coordinate-adapted basis selected for that invariant space.

Two lines, one intersection

Consider the two systems

They have no equation in common, yet both reduce to the augmented matrix

The reduced equations are $x=3$ and $y=1$, so their common solution is $(3,1)$. Geometrically, these are the vertical and horizontal lines through that solution point. The original systems are different choices of two lines through the same point.

The object that does not move

What the two systems share is the two-dimensional space generated by their augmented rows. More generally, for matrices $M$ and $N$ of the same size, the guiding theorem is

An elementary row operation is an invertible change of generators for this space. It may alter every displayed equation, but it cannot alter the space of equations they generate. The change is invertible, so the new equations vanish simultaneously at exactly the same points as the old ones.

An augmented row $(\alpha,\beta,\gamma)$ may be read as the affine-linear function $f(x,y)=\alpha x+\beta y-\gamma$. For a fixed point $P=(a,b)$, all affine-linear functions vanishing at $P$ form the space

Any two independent line equations through $P$ form an ordered basis of $V_P$. Row reduction changes that basis while leaving $V_P$ fixed.

A coordinate-adapted basis

In the full-rank two-variable case, RREF selects the especially simple basis $x-a$ and $y-b$. It therefore presents the same equation space through the coordinate lines $x=a$ and $y=b$. This is why RREF remembers the solution point in this particular case. The more fundamental statement, which also makes sense for singular and inconsistent systems, is that it remembers the row space and forgets the ordered generating rows.

There is a visible motion behind elimination. If two lines with equations $f_1=0$ and $f_2=0$ meet at $P$, the operation $f_2\mapsto f_2+t f_1$ replaces the second line by another line through $P$. As $t$ varies, the line moves through the pencil centred at $P$. Cancelling one coefficient selects the horizontal member; cancelling the other selects the vertical member.

What the full note develops

The projective viewpoint places consistent and inconsistent rank-two systems in one larger picture. The two-dimensional row spaces of $2\times3$ matrices form the Grassmannian $\operatorname{Gr}(2,3)\cong\mathbb{RP}^2$. Systems with a finite solution occupy its affine part; distinct parallel lines appear at infinity through their common direction. This does not give an inconsistent system an affine solution—it records the direction its equations share.

For a general system $Ax=c$, the same principle survives. Invertible row operations preserve the augmented row space. A system is inconsistent exactly when its equation space contains a nonzero constant function, visible in RREF as a contradiction such as $0=1$. When the system is consistent, its row space consists of the affine-linear equations vanishing on the solution affine subspace.

The complete note begins with all seven possible RREF forms for a $2\times3$ matrix and proceeds from computation to structure: line pencils, the action of $\mathrm{GL}_2(\mathbb{R})$, degeneracies, projective completion, Grassmannians and arbitrary linear systems. Its worked examples, figures, proofs, geometric dictionary and references supply the details behind this short introduction.

Read the full note

@online{rajendran2026rowreduction,
  author       = {{Mohan R}},
  title        = {What Row Reduction Remembers: The geometry of intersecting lines, equation spaces, and the general-rank picture},
  date         = {2026-08-09},
  organization = {Math Nomad},
  url          = {https://mathnomad.in/notes/linear-algebra/what-row-reduction-remembers},
  langid       = {british}
}