Experiment 01 · a flow of linear maps
Linear ODEs with time-dependent coefficients
Every particle follows ẋ(t) = A(t)x(t). The same fundamental matrix Φ(t) moves all initial points: x(t) = Φ(t)x(0). Compare this solution with the exponential of the integrated coefficient matrix.
Image of the initial conditions
t = 1.20Equal axis scales; the view stays fixed during playback.
A rotating strain field
A(t) = αI + ωJ + σR(νt)DR(−νt)
The instantaneous stretching axes turn in time. The matrices at different times generally do not commute, so the order of the small transformations matters.
I is the identity, J = [0, −1; 1, 0], H = [0, 1; 0, 0], V = [0, 0; 1, 0], D = diag(1, −1), and R(θ) is counterclockwise rotation by θ.
Smooth flow: fourth-order Runge–Kutta (RK4), maximum step Δt = 1/120. The displayed step-halving difference compares Φ at Δt and Δt/2; it is a convergence diagnostic, not a rigorous error bound.
The solid ellipse is the image of the unit circle, even when the particles start on a grid. The dashed amber ellipse uses exp(∫₀ᵗA(s) ds); it agrees for constant or pairwise-commuting A, but not in general. The matrix gap is ‖Φ − exp(∫A)‖F/‖Φ‖F. The determinant measures signed area, not the length of every vector.