|
| 1 | +''' Script that demonstrates the functionality of the superstep in 1D |
| 2 | +"Wave on a string" |
| 3 | +''' |
| 4 | +import matplotlib.pyplot as plt |
| 5 | +import numpy as np |
| 6 | +from devito import Eq, Function, Grid, Operator, TimeFunction, solve |
| 7 | +from devito.timestepping.superstep import superstep_generator |
| 8 | + |
| 9 | +# Parameters |
| 10 | +## Spatial |
| 11 | +shape = (501, ) |
| 12 | +pad = (0, ) |
| 13 | +origin= (0, ) |
| 14 | +extent = (1, ) |
| 15 | +# Time |
| 16 | +t0 = 0 |
| 17 | +t1 = 0.15 |
| 18 | +critical_dt = 0.0014142 |
| 19 | +# Initial Condition |
| 20 | +mu = 0.5 |
| 21 | +sigma_sq = 0.005 |
| 22 | +ylim = np.ceil(1/np.sqrt(2*np.pi*sigma_sq)) |
| 23 | +xlim = (0, 1) |
| 24 | + |
| 25 | +def gaussian(x, mu=0, sigma_sq=1): |
| 26 | + ''' Generate a Gaussian initial condition |
| 27 | + ''' |
| 28 | + return np.exp(-((x - mu)**2)/(2*sigma_sq))/(np.sqrt(2*np.pi*sigma_sq)) |
| 29 | + |
| 30 | +def wave_on_string(step=1): |
| 31 | + grid = Grid(shape=shape, extent=extent) |
| 32 | + |
| 33 | + velocity = Function(name='velocity', grid=grid, space_order=(0, step, step)) |
| 34 | + velocity.data[:] = 1 |
| 35 | + |
| 36 | + u = TimeFunction(name='u', grid=grid, time_order=2, space_order=2) |
| 37 | + |
| 38 | + pde = (1/velocity**2)*u.dt2 - u.dx2 |
| 39 | + stencil = Eq(u.forward, solve(pde, u.forward)) |
| 40 | + |
| 41 | + # Initial condition |
| 42 | + x = np.linspace(0, 1, *shape) |
| 43 | + ic = gaussian(x, mu=mu, sigma_sq=sigma_sq) |
| 44 | + |
| 45 | + if step == 1: |
| 46 | + # Non-superstep case |
| 47 | + newu = u |
| 48 | + newu.data[0, :] = ic |
| 49 | + newu.data[1, :] = ic |
| 50 | + op = Operator(stencil) |
| 51 | + else: |
| 52 | + # Superstepping |
| 53 | + newu, newu_p, stencil1, stencil2 = superstep_generator(u, stencil.rhs, k=step) |
| 54 | + |
| 55 | + newu.data[0, :] = ic |
| 56 | + newu.data[1, :] = ic |
| 57 | + newu_p.data[0, :] = ic |
| 58 | + newu_p.data[1, :] = ic |
| 59 | + |
| 60 | + op = Operator([ |
| 61 | + stencil1, |
| 62 | + stencil2, |
| 63 | + ], opt='noop') |
| 64 | + |
| 65 | + tn = int(np.ceil(t1/critical_dt)) |
| 66 | + dt = t1/tn |
| 67 | + |
| 68 | + op(time=tn, dt=dt) |
| 69 | + |
| 70 | + idx = tn % 3 |
| 71 | + return newu.data[idx] |
| 72 | + |
| 73 | +if __name__ == '__main__': |
| 74 | + fig, ax = plt.subplots(1, 1) |
| 75 | + x = np.linspace(0, 1, *shape) |
| 76 | + ax.plot( |
| 77 | + x, gaussian(x, mu=mu, sigma_sq=sigma_sq), |
| 78 | + color='k', ls='--', label='Initial Condition' |
| 79 | + ) |
| 80 | + |
| 81 | + for ii in range(1, 6): |
| 82 | + label = 'Normal timestepping' if ii == 1 else f'Superstep size {ii}' |
| 83 | + ax.plot(x, wave_on_string(ii), label=label) |
| 84 | + |
| 85 | + ax.set_xlim(*xlim) |
| 86 | + ax.set_ylim(-ylim, ylim) |
| 87 | + ax.legend() |
| 88 | + plt.show() |
0 commit comments