2.4.6 Custom noise

This example solves a diffusion equation with a custom noise.

  0%|          | 0/10.0 [00:00<?, ?it/s]
Initializing:   0%|          | 0/10.0 [00:00<?, ?it/s]
  0%|          | 0/10.0 [00:03<?, ?it/s]
  0%|          | 0.01/10.0 [00:04<1:17:25, 465.05s/it]
  0%|          | 0.02/10.0 [00:04<38:40, 232.54s/it]
  4%|▍         | 0.38/10.0 [00:04<01:57, 12.24s/it]
 89%|████████▉ | 8.94/10.0 [00:04<00:00,  1.91it/s]
 89%|████████▉ | 8.94/10.0 [00:04<00:00,  1.90it/s]
100%|██████████| 10.0/10.0 [00:04<00:00,  2.13it/s]
100%|██████████| 10.0/10.0 [00:04<00:00,  2.13it/s]

import numpy as np

from pde import DiffusionPDE, ScalarField, UnitGrid
from pde.tools.numba import jit


class DiffusionCustomNoisePDE(DiffusionPDE):
    """Diffusion PDE with custom noise implementations."""

    def noise_realization(self, state, t):
        """Numpy implementation of spatially-dependent noise."""
        noise_field = ScalarField.random_uniform(state.grid, -self.noise, self.noise)
        return state.grid.cell_coords[..., 0] * noise_field

    def _make_noise_realization_numba(self, state):
        """Numba implementation of spatially-dependent noise."""
        noise = float(self.noise)
        x_values = state.grid.cell_coords[..., 0]

        @jit
        def noise_realization(state_data, t):
            return x_values * np.random.uniform(-noise, noise, size=state_data.shape)

        return noise_realization


eq = DiffusionCustomNoisePDE(diffusivity=0.1, noise=0.1)  # define the pde
state = ScalarField.random_uniform(UnitGrid([64, 64]))  # generate initial condition
result = eq.solve(state, t_range=10, dt=0.01)
result.plot()