Summary: Undoing a transformation

Last lesson warned that a zero determinant loses information. This lesson makes that precise and answers the practical question underneath every linear system: can a transformation be reversed, and if not, what did it destroy? The whole thing reduces to one line: the inverse undoes a transformation when nothing was lost, and the determinant, the rank, and the null space are three ways of checking whether anything was. This is the scan-it-in-five-minutes version.

Core ideas

• A linear system M · v = b is really the question “which input does the transformation M send to the target b?” That reframing is what makes the geometry useful for solving equations.
• The inverse M^-1 is the transformation that undoes M, defined by M^-1 · M = I (the identity, the do-nothing transformation). When it exists, the system has the unique solution v = M^-1 · b. It is not “one over the matrix”; there is no division.
• The inverse exists exactly when det(M) ≠ 0. A nonzero determinant means space was rearranged but not collapsed, so every output came from one input and can be traced back. A zero determinant means many inputs were crushed together, so the information needed to reverse it is gone.
• The column space is everything the transformation can output, which is the span of its columns. Its dimension is the rank: how many output dimensions the transformation actually uses (full plane = rank 2, collapse to a line = rank 1).
• The null space is everything the transformation crushes to the origin (M · v = 0). For an invertible matrix it is just the origin; for a collapsed one it is a whole line of inputs flattened to nothing, which is precisely the information destroyed. Its dimension is the nullity.
• These conserve: rank + nullity = input dimension. The dimensions kept plus the dimensions lost always equal the dimensions you started with.
• Three tests, one question. Nonzero determinant, full rank, and a trivial null space are the same statement about invertibility. Worked anchors: [[2,1],[1,1]] has det 1, rank 2, null {0} (invertible); [[2,4],[1,2]] has det 0, rank 1, null space the line through [-2,1] (not invertible). A system with a non-invertible matrix has no solution (target off the column space) or infinitely many (target on it), never exactly one.
• This is why the ideas matter for AI. Rank answers how many dimensions a matrix actually uses; large matrices are often low-rank, hence compressible, which underlies a lot of model-shrinking work. The null space is the set of input directions a transformation is blind to (it maps them to the same place), which is sometimes a defect and sometimes deliberate compression.

What changes for you

Before this lesson, “invertible,” “rank,” and “null space” were probably separate vocabulary items you met one at a time. Now they are three readings of a single event: whether a transformation kept space intact. When you next see “the matrix is singular,” “this layer is low-rank,” or “these inputs are in the null space,” you can translate each into the same geometric picture of what was kept and what was crushed. The next lesson breaks an assumption every matrix so far has quietly relied on, that input and output have the same number of dimensions, by allowing nonsquare matrices that move between dimensions.