Introduction to Linear Algebra
Four subspaces, infinite insight
Description
For most of the twentieth century, a first course in linear algebra opened the same way: with a definition. A vector space is a set closed under addition and scalar multiplication, satisfying eight axioms. Students copied the list, nodded, and waited for something to happen. Often it never quite did. The subject arrived as a tower of definitions stacked on definitions, and the objects that mattered most — matrices, the grids of numbers everyone actually computes with — showed up almost as an afterthought, a special case of something more general and less alive.
Gilbert Strang, who taught the course at MIT for decades and whose lectures reached millions online, wrote Introduction to Linear Algebra to do the opposite. He starts with the matrix. He starts with what a matrix does to a vector, what combinations of columns it can build, what it quietly throws away. And out of that concrete gesture he draws the spine of the whole book: four subspaces attached to every matrix, four sets of vectors that between them explain almost everything the matrix can and cannot do.
The move sounds modest. It is not. Choosing to organize the subject around four specific spaces, rather than around the abstract axioms, changed how the material lands. Equations that once looked like arithmetic puzzles turn into questions about geometry — which vectors are reachable, which are invisible, and how the two halves fit.
The question we’re asking : Why did reorganizing linear algebra around four particular subspaces make the subject click for so many people who had bounced off it before?What we’ll see : How Strang builds the whole edifice from what a matrix does to a vector, and why that concrete starting point turned an abstract discipline into something you can picture.
Table of contents
01Chapter 1 — A matrix is a machine that moves vectors
Strang's book refuses to treat a matrix as a static array of numbers. A matrix A is a machine. You feed it a vector x, and it hands back another vector, Ax. That single operation — multiply a matrix by a vector — is the beating heart of the whole subject, and Strang spends real time making sure we feel it before we manipulate it. Everything that follows is a question about this machine: what can it produce, and what does it lose along the way.
The crucial reframing is how to read the product Ax itself. The mechanical rule most of us learned in school treats it row by row, taking dot products one at a time. Strang insists on the other reading, the one that carries meaning: Ax is a combination of the columns of A. The entries of x are just the amounts. If A has columns that are vectors in space, then Ax is a weighted sum of those columns, mixed in whatever proportions x dictates. Solving Ax = b becomes a plain question — can we mix the columns of A to build b?
02Chapter 2 — The column space is the set of everything you can reach
Take the first question — what can the machine produce? Every output Ax is a combination of the columns of A. So the set of all possible outputs is exactly the set of all combinations of those columns. Strang calls it the column space, and the name is honest: it is the space that the columns span, everything reachable by mixing them. If A maps vectors into three-dimensional space, its column space is some flat piece of that space passing through the origin — a line, a plane, or the whole of it.
This single idea reorganizes the question of when Ax = b has a solution. The equation is solvable precisely when b lies in the column space of A — when the target is something the columns can actually build. If b sits off that plane, no combination of columns will ever reach it, and the honest answer is that there is no solution. What looked like a matter of grinding through elimination becomes a matter of position: is the target inside the reachable region or outside it? The algebra and the geometry finally say the same thing.
03Chapter 3 — The nullspace holds what the matrix forgets
The second question was: which inputs does the machine crush to zero? The nullspace of A is the collection of every vector x for which Ax equals the zero vector. These are the inputs that come out as nothing, the directions the matrix cannot detect. It always contains at least the zero vector itself, since a machine that respects addition and scaling must send zero to zero. The real interest is whether it contains anything more — whether there are nonzero inputs that vanish.
This is where solutions to Ax = b either become unique or multiply. Suppose we have found one solution. Now add to it any vector from the nullspace. Because that added piece contributes nothing to the output, the sum is still a solution. So if the nullspace holds anything beyond zero, there are infinitely many solutions, a whole family sliding along in the directions the matrix ignores. If the nullspace holds only zero, the solution, when it exists, stands alone. Uniqueness is not a separate fact to prove — it is a reading of how big the nullspace is.
04Chapter 4 — Four spaces, and the picture they draw together
Strang's real contribution is not the four subspaces themselves; mathematicians knew them long before he wrote. It is the decision to make them the organizing spine of a first course, and to draw them. The famous diagram in the book shows all four at once: the row space and nullspace living in the input world, the column space and left nullspace living in the output world, with the matrix carrying vectors from one side to the other. In a single figure the whole subject sits still long enough to be understood. Where a vector comes from, where it lands, what survives the trip and what disappears — it is all in the picture.
This is a deliberate reversal of how the subject had long been taught. The axiomatic approach starts general and hopes the examples will follow; Strang starts with the most concrete object, the matrix of numbers, and lets the general structure emerge from what that object does. The abstract vector space is still there, but it arrives late, as a summary of patterns already seen, rather than as a barrier to be cleared at the door. Understanding is built from a thing you can compute and draw, not from a definition you must accept on faith.
05Conclusion
The book that began by refusing to open with a definition ends by having earned one. By the time the abstract vector space appears in full generality, the reader has already watched columns span a plane, seen a nullspace hold the directions a matrix ignores, and traced a vector across Strang's four-part diagram. The axioms no longer arrive as an obstacle; they read as a tidy account of things already understood, which is exactly the order in which understanding tends to happen.