Summary: Spans and basis

Give yourself a couple of vectors and the only two operations you have (adding and scaling), and ask which points you can reach. The answer is a set with a name, and the whole lesson hangs on three of them: span is what you can reach, a basis is the smallest kit that reaches everything, and dimension is the size of that kit. Get those three and the rest of the track has firm ground, because every later idea (transformations, rank, eigenvectors) is phrased in terms of what some set of vectors can and cannot reach. This is the scan-it-in-five-minutes version.

Core ideas

• A linear combination of vectors is any a · v + b · w for scalars you choose freely: scale each vector, add the results. It is just the previous lesson’s two operations used together.
• The span of a set of vectors is the set of all their linear combinations, every point reachable by adding and scaling. It is not the original arrows; a pair of short vectors and a pair of long ones can have the exact same span.
• For two vectors in the plane there are exactly three spans. The whole plane (when they point in different directions), a single line through the origin (when one is a scalar multiple of the other), or just the origin (when both are the zero vector). The line-versus-plane boundary is the whole game.
• A basis is a set that is linearly independent (no vector sits in the span of the others, so none is redundant) and spanning (it reaches every point). Independent means no waste; spanning means no gaps; a basis is the smallest set that reaches everything.
• The standard basis (i-hat [1, 0] and j-hat [0, 1]) is quietly what coordinates have meant all along: [3, 4] is shorthand for 3 · [1, 0] + 4 · [0, 1]. Every vector is a combination of the basis vectors in exactly one way, and that uniqueness is what lets coordinates be coordinates.
• Linear dependence is about reach, not value: a dependent vector is redundant (it adds no direction the others lacked), which is what collapses a span from a plane down to a line. Independence is exactly the property that fills the space.
• Dimension is the size of a basis: how many independent directions it takes to describe any point. Each genuinely independent vector you add lifts the span by one dimension, until you run out of new directions; past that, extra vectors are all dependent.
• This is why span and basis matter for AI. A model represents words or images in a space of fixed dimension (often hundreds), and that number is the size of a basis: how many independent directions describe any point. Every embedding is a linear combination of those directions, and a lot of practical work on shrinking models is, underneath, the search for a smaller basis that still spans almost everything.

What changes for you

Before this lesson, “span,” “basis,” and “dimension” were probably words you nodded at without a picture behind them. Now each one is a concrete question about reach: what can these vectors build, what is the smallest set that builds all of it, and how many independent directions is that. When you next see a paper talk about the dimension of an embedding or a low-rank approximation, you have the real question underneath it: how many independent directions does this representation actually need, and which ones are redundant. The next lesson turns these still pictures into motion, asking what happens when a single rule moves every vector in the space at once.