One continuous room. You drag arrows on an honest grid that can stretch and shear but never tear, and feel each idea become necessary: a single arrow can’t chain journeys, so addition appears; tracking infinite points is hopeless, so the matrix appears; a flattened space can’t be undone, so the inverse — and its limits — appear. No sky-dropped definitions, no algebra prerequisites: just spatial intuition, in order, from the first arrow to the singular value decomposition. Built for a curious ten-year-old; honest enough for a maths degree.
Pointing is vague; record a precise "how far, which way" instruction instead — and the arrow is born.
One arrow reaches one spot; lay arrows tip-to-tail and stretch them by a number — and addition and scaling are born.
Keeping an arrow for every spot is hopeless — describe everywhere as a recipe of two master arrows, and the basis is born.
When two master arrows secretly do the same job, the whole world collapses to a line — so we ask which masters truly pull their weight.
Tracking infinite points is impossible — but the grid moves as one, so record only where î and ĵ land. That table is the matrix.
Given only where the master arrows landed, where does one point go? Re-run its recipe on the new columns — that is matrix × vector.
Doing move A then move B by hand each time is wasteful — bake them into a single recording, and discover the order can’t be swapped.
We can move space but not say how violently — so we measure the unit square’s area before and after. That single number is the determinant.
When the determinant hits zero, all of space is crushed onto a single line and different points pile onto the same spot — information lost forever.
Deform space with M, then press Undo — M⁻¹ is the move that perfectly reverses it. It exists only when det ≠ 0.
You only see where a point landed — so solving M·x = b means running the machine backward to find the exact recipe of columns that reaches the target.
A move can squish space — so which destinations stay reachable (the column space) and which arrows get crushed to nothing (the null space)? Their sizes are the rank.
Recipes place points but can’t say length or angle — so to ask "do these arrows agree?" we invent one number: the dot product.
We want the closest point on a line, so we split any arrow into a part that agrees and a part at a perfect right angle — orthogonal projection.
Two arrows in 3D conjure a brand-new direction square to both, whose length is the area of the flap they span — the cross product.
A point never moves — but in a tilted set of masters its address is a different recipe. Change of basis is the faithful translator between the two coordinate systems.
Most moves look like chaos — so we hunt the rare arrows a move refuses to knock off their own line, only stretching them by λ.
A tangled move becomes pure stretching seen along its eigen-arrows — so repeating it n times turns into multiplying by λⁿ.
Force a matrix symmetric and its two eigen-lines always cross at a perfect right angle — the spectral theorem, the clean skeleton the SVD will give every matrix next.
Most real moves have no clean eigen-arrows — so crack ANY matrix into three honest steps: rotate, stretch along clean perpendicular axes, rotate.