Shows an optimization problem as (1) a feasible set X (hatched region from constraints) and (2) objective contours f(x). A point x moves into the feasible set, then performs projected descent to reduce f(x) while respecting constraints, and finally highlights first-order local optimality: the gradient ∇f is orthogonal to feasible (tangent) directions (no feasible descent direction).
Pure Canvas2D, green-on-black blocky grid hatch. Feasible set is intersection of a halfspace and a disk; objective is a convex quadratic with tilt. Motion uses time-phased segments and simple iterative projection + gradient steps (no external state). Contours are approximated by sampling and drawing near-level pixels for a retro look.