Lab 08 — When Information Is Lost

Explore rank, null space, column space, invertibility, collapse, and controlled information loss through interactive matrix machines.

1. Matrix machine with adjustable collapse

Use the sliders to define a matrix. Watch how a grid of input points transforms. When the transformed grid collapses toward a line, information is being lost.

2. Column space: reachable targets

The two columns of a matrix are output directions. Move the target point. The page checks whether the target is reachable for the rank-one matrix below.

A = [[1, 2], [2, 4]]

3. Null space: invisible directions

For the same rank-one matrix, moving along the null direction changes the input but keeps the output unchanged.

Here x₀ = (1,0) and v = (-2,1). Since Av = 0, all inputs x₀ + tv have the same output.

4. Rank-nullity calculator

Choose the input and output dimensions and an intended rank. The calculator shows how many input directions must disappear.

5. Near information loss

The matrix Aε = [[1,1],[1,1+ε]] is invertible when ε ≠ 0, but as ε becomes small its columns become almost dependent.

A matrix can have full rank and still be dangerous numerically. Near information loss creates large sensitivity.

6. Image compression preview

This toy image is a matrix. A rank-k approximation keeps only k directions. Smaller k means more compression and more information loss.