Home▸Mathematics▸Mathematics Optimization Simulation

🧪 Mathematics Optimization Simulation

Watch simulated annealing search a bumpy 3D loss landscape: a probe accepts uphill moves when the temperature is high, escaping local minima, then cools and settles into the global minimum. Adjust the cooling schedule and step size and watch it converge.

Mathematics2DModerate60 FPS📱 Mobile-adapted

Simulated annealing is an optimization heuristic that, unlike plain gradient descent, can climb out of local minima. This simulator renders a randomly generated, deliberately bumpy 3D loss landscape and runs a real Metropolis-criterion annealing search across it: a probe proposes random nearby moves, accepts every downhill step, and accepts uphill steps with probability exp(−Δf/T) — so a hot search explores freely while a cooled search commits to a basin. Adjust the starting temperature, the geometric cooling rate, and the step size to see the classic trade-off between exploration and convergence, with live readouts of temperature, iteration count, current and best objective value, and recent acceptance rate.

⚙ Under the hood

This simulation demonstrates optimization techniques such as gradient descent. Users can adjust parameters to minimize a function and observe the resulting changes in a dynamic environment.

OptimizationAlgorithmsModeling

2D · HTML5 Canvas 2D · 60 FPS target · runs fully client-side, no install