Summary: Linear transformations as moves

A matrix looks like a grid of numbers with no obvious meaning. The whole lesson replaces that picture with one idea: a matrix is a record of where the two basis vectors land, and that record is enough to move all of space. Once you see it that way, matrix-vector multiplication stops being a rule to memorize and becomes something you can sketch, and every later operation in the track (composition, determinant, inverse, eigenvectors) is a question about what some matrix does to space. This is the scan-it-in-five-minutes version.

Core ideas

• A transformation is a function from vectors to vectors; picture every point in the plane moving to a new spot at once. It is a linear transformation when two things hold: the origin stays fixed, and straight lines stay straight (grid lines stay parallel and evenly spaced). Stretch, rotate, shear, and flip qualify; curving and shifting the whole plane do not.
• The basis-vector trick is the engine of the lesson. Because every vector is v = x · i-hat + y · j-hat and a linear transformation preserves that combination, you get L(v) = x · L(i-hat) + y · L(j-hat). Knowing where just two vectors land tells you where every vector lands.
• A matrix is exactly those two landing spots written as columns: the first column is where i-hat goes, the second is where j-hat goes. Every rule about matrices that looks arbitrary on first meeting falls out of this one fact.
• The matrix-vector product is then not a new rule: M · v = x · (first column) + y · (second column), the same scale-and-add from the span lesson, applied to the transformed basis. The columns are the ingredients; the product reassembles your vector from them.
• You can read a matrix by sketching the unit square. Plot the two columns and draw the parallelogram they span; that is the image of the unit square. A square stays a square (rotation), becomes a rectangle (axis stretch), leans into a parallelogram (shear), flips (reflection), or collapses to a line (columns parallel, a sign the plane is being squashed).
• Worked anchors: the stretch [[2,0],[0,1]] sends [3,4] to [6,4]; the quarter-turn [[0,-1],[1,0]] sends it to [-4,3]; the shear [[1,1],[0,1]] sends it to [7,4]. Same recipe each time, different columns.
• This is why matrices matter for AI. A linear layer, the most common building block of a neural network, holds a matrix of learned weights, and running data through it is exactly the M · v operation here. Most of a model’s billions of parameters live in such matrices, each one a linear transformation of its input space, so understanding what one matrix does to space is understanding the atom of what every layer does to data.

What changes for you

Before this lesson, a matrix was probably a grid of numbers you pushed through a multiplication rule without a mental picture. Now it is a geometric object: two arrows showing where the basis went, and a move applied to all of space at once. When you next see a weight matrix in a model diagram or a transformation in a graphics or robotics context, you have a concrete question to ask, “where does this send the basis, and what shape does the unit square become?”, instead of treating the numbers as opaque. The next lesson asks the obvious follow-up: if one matrix is one move, what does doing two moves in a row look like? That is matrix multiplication, and it turns out not to be arbitrary either.