def test_continuous_oscillation_optimal_control(
continuous_oscillation_operator, density, coordinate_tensor, num_dims,
periodic, control_weight):
if not continuous_oscillation_operator.has_analytic_solution:
pytest.skip()
# Define the inputs
t = np.sqrt(2)
ic = md.evaluate_gaussian_kernel(coordinate_tensor, .25, 1, 0.05**2)
ic = [ic, np.zeros_like(ic)]
setpoint = md.evaluate_gaussian_kernel(coordinate_tensor, .75, 1, 0.05**2)
setpoint = [setpoint, np.zeros_like(setpoint)]
z_nc = continuous_oscillation_operator.integrate_analytic(ic, t)
# Evaluate the control field
control = continuous_oscillation_operator.evaluate_control(
ic, setpoint, np.eye(2), control_weight * np.eye(2), t)
z_wc = continuous_oscillation_operator.integrate_analytic(ic,
t,
control=control)
# Check that the result with control is better than the one without
corrcoef_nc = np.corrcoef(np.ravel(setpoint), z_nc.ravel())[0, 1]
corrcoef_wc = np.corrcoef(np.ravel(setpoint), z_wc.ravel())[0, 1]
assert corrcoef_wc > corrcoef_nc
# Assert that we get the "right" result if the control weight is zero
if control_weight == 0:
md.assert_correlated(setpoint, z_wc)
def initial_conditions(coordinate_tensor, num_dims):
center = np.random.uniform(0, 1, num_dims)
ic = md.evaluate_gaussian_kernel(coordinate_tensor,
center,
1,
0.5**2,
domain=None)
return ic[None] # Add the first dimension
def test_continous_oscillation_operator_with_control(
continuous_oscillation_operator, density, coordinate_tensor, num_dims,
periodic):
center = np.random.uniform(0, 1, num_dims)
ic = md.evaluate_gaussian_kernel(coordinate_tensor, center, 1, 0.05**2)
ic = [ic, np.zeros_like(ic)]
_test_integration(continuous_oscillation_operator,
ic,
control=np.ones_like(ic))
def test_continuous_oscillation_supramatrix(continuous_oscillation_operator, density,
coordinate_tensor, num_dims, periodic):
center = np.random.uniform(0, 1, num_dims)
ic = md.evaluate_gaussian_kernel(coordinate_tensor, center, 1, 0.05 ** 2)
ic = np.asarray([ic, np.zeros_like(ic)])
grad1 = continuous_oscillation_operator.evaluate_gradient(ic)
grad2 = continuous_oscillation_operator.supramatrix.dot(ic.ravel()).reshape(ic.shape)
corrcoef = np.corrcoef(grad1.ravel(), grad2.ravel())
assert corrcoef[0, 1] > 0.95