Skip to contents

Simulates samples from the double-well potential \(U(x) = (x^2 - 1)^2\) using Euler-Maruyama integration of the overdamped Langevin equation \(dX_t = -U'(X_t) dt + \sqrt{2\beta^{-1}} dW_t\). Samples are wrapped as a StateTransitionData with a PotentialGroundTruth recording the known well locations, barrier position, and barrier height.

Usage

synthetic_potential_control(
  n = 100L,
  beta = 2,
  n_steps = 5000L,
  dt = 0.01,
  seed = 42L
)

Arguments

n

integer number of samples to collect

beta

numeric inverse temperature (higher = sharper wells, default 2)

n_steps

integer Euler-Maruyama steps per sample (default 5000)

dt

numeric step size (default 0.01)

seed

integer RNG seed

Value

StateTransitionData with PotentialGroundTruth and metadata()$potential_control carrying generating parameters. The single experiment "coords" has one feature row: the simulated x values. colData carries x_coord and well ("left"/"right").

Details

The ground truth for this potential:

  • Wells at x = ±1 (minima of U)

  • Barrier at x = 0 (maximum of U between wells)

  • Barrier height = U(0) - U(1) = 1