|
| 1 | +import numpy as np |
| 2 | +from mpi4py import MPI |
| 3 | + |
| 4 | +from pySDC.implementations.controller_classes.controller_nonMPI import controller_nonMPI |
| 5 | +from pySDC.projects.DAE.misc.HookClass_DAE import ( |
| 6 | + LogGlobalErrorPostStepDifferentialVariable, |
| 7 | + LogGlobalErrorPostStepAlgebraicVariable, |
| 8 | +) |
| 9 | +from pySDC.helpers.stats_helper import get_sorted |
| 10 | + |
| 11 | + |
| 12 | +def run(dt, num_nodes, use_MPI, semi_implicit, residual_type, index_case, initial_guess='spread', comm=None): |
| 13 | + r""" |
| 14 | + Prepares the controller with all the description needed. Here, the function decides to choose the correct sweeper |
| 15 | + for the test. |
| 16 | +
|
| 17 | + Parameters |
| 18 | + ---------- |
| 19 | + dt : float |
| 20 | + Time step size chosen for simulation. |
| 21 | + num_nodes : int |
| 22 | + Number of collocation nodes. |
| 23 | + use_MPI : bool |
| 24 | + If True, the MPI sweeper classes are used. |
| 25 | + semi_implicit : bool |
| 26 | + Modules are loaded either for fully-implicit case or semi-implicit case. |
| 27 | + residual_type : str |
| 28 | + Choose how to compute the residual. |
| 29 | + index_case : int |
| 30 | + Denotes the index case of a DAE to be tested here, can be either ``1`` or ``2``. |
| 31 | + initial_guess : str, optional |
| 32 | + Type of initial guess for simulation. |
| 33 | + comm : mpi4py.MPI.COMM_WORLD |
| 34 | + Communicator. |
| 35 | + """ |
| 36 | + |
| 37 | + if not semi_implicit: |
| 38 | + if use_MPI: |
| 39 | + from pySDC.projects.DAE.sweepers.fully_implicit_DAE_MPI import fully_implicit_DAE_MPI as sweeper |
| 40 | + |
| 41 | + else: |
| 42 | + from pySDC.projects.DAE.sweepers.fully_implicit_DAE import fully_implicit_DAE as sweeper |
| 43 | + |
| 44 | + else: |
| 45 | + if use_MPI: |
| 46 | + from pySDC.projects.DAE.sweepers.SemiImplicitDAEMPI import SemiImplicitDAEMPI as sweeper |
| 47 | + |
| 48 | + else: |
| 49 | + from pySDC.projects.DAE.sweepers.SemiImplicitDAE import SemiImplicitDAE as sweeper |
| 50 | + |
| 51 | + if index_case == 1: |
| 52 | + from pySDC.projects.DAE.problems.DiscontinuousTestDAE import DiscontinuousTestDAE as problem |
| 53 | + |
| 54 | + t0 = 1.0 |
| 55 | + Tend = 1.5 |
| 56 | + |
| 57 | + elif index_case == 2: |
| 58 | + from pySDC.projects.DAE.problems.simple_DAE import simple_dae_1 as problem |
| 59 | + |
| 60 | + t0 = 0.0 |
| 61 | + Tend = 0.4 |
| 62 | + |
| 63 | + else: |
| 64 | + raise NotImplementedError(f"DAE case of index {index_case} is not implemented!") |
| 65 | + |
| 66 | + # initialize level parameters |
| 67 | + level_params = { |
| 68 | + 'restol': 1e-12, |
| 69 | + 'residual_type': residual_type, |
| 70 | + 'dt': dt, |
| 71 | + } |
| 72 | + |
| 73 | + # initialize problem parameters |
| 74 | + problem_params = { |
| 75 | + 'newton_tol': 1e-6, |
| 76 | + } |
| 77 | + |
| 78 | + # initialize sweeper parameters |
| 79 | + sweeper_params = { |
| 80 | + 'quad_type': 'RADAU-RIGHT', |
| 81 | + 'num_nodes': num_nodes, |
| 82 | + 'QI': 'MIN-SR-S', # use a diagonal Q_Delta here! |
| 83 | + 'initial_guess': initial_guess, |
| 84 | + } |
| 85 | + |
| 86 | + # check if number of processes requested matches with number of nodes |
| 87 | + if comm is not None: |
| 88 | + sweeper_params.update({'comm': comm}) |
| 89 | + assert ( |
| 90 | + sweeper_params['num_nodes'] == comm.Get_size() |
| 91 | + ), f"Number of nodes does not match with number of processes! Expected {sweeper_params['num_nodes']}, got {comm.Get_size()}!" |
| 92 | + |
| 93 | + # initialize step parameters |
| 94 | + step_params = { |
| 95 | + 'maxiter': 20, |
| 96 | + } |
| 97 | + |
| 98 | + # initialize controller parameters |
| 99 | + controller_params = { |
| 100 | + 'logger_level': 30, |
| 101 | + 'hook_class': [LogGlobalErrorPostStepDifferentialVariable, LogGlobalErrorPostStepAlgebraicVariable], |
| 102 | + } |
| 103 | + |
| 104 | + # fill description dictionary for easy step instantiation |
| 105 | + description = { |
| 106 | + 'problem_class': problem, |
| 107 | + 'problem_params': problem_params, |
| 108 | + 'sweeper_class': sweeper, |
| 109 | + 'sweeper_params': sweeper_params, |
| 110 | + 'level_params': level_params, |
| 111 | + 'step_params': step_params, |
| 112 | + } |
| 113 | + |
| 114 | + # instantiate controller |
| 115 | + controller = controller_nonMPI(num_procs=1, controller_params=controller_params, description=description) |
| 116 | + P = controller.MS[0].levels[0].prob |
| 117 | + |
| 118 | + uinit = P.u_exact(t0) |
| 119 | + |
| 120 | + # call main function to get things done... |
| 121 | + uend, stats = controller.run(u0=uinit, t0=t0, Tend=Tend) |
| 122 | + controller.MS[0].levels[0].sweep.compute_end_point() |
| 123 | + |
| 124 | + residual = controller.MS[0].levels[0].status.residual |
| 125 | + |
| 126 | + return uend, residual, stats |
| 127 | + |
| 128 | + |
| 129 | +def check_order(comm): |
| 130 | + num_nodes = comm.Get_size() |
| 131 | + use_MPI = True |
| 132 | + residual_type = 'full_abs' |
| 133 | + for semi_implicit in [False, True]: |
| 134 | + for index_case in [1, 2]: |
| 135 | + dt_list = np.logspace(-1.7, -1.0, num=5) |
| 136 | + |
| 137 | + errorsDiff, errorsAlg = np.zeros(len(dt_list)), np.zeros(len(dt_list)) |
| 138 | + for i, dt in enumerate(dt_list): |
| 139 | + _, _, stats = run( |
| 140 | + dt=dt, |
| 141 | + num_nodes=num_nodes, |
| 142 | + use_MPI=use_MPI, |
| 143 | + semi_implicit=semi_implicit, |
| 144 | + residual_type=residual_type, |
| 145 | + index_case=index_case, |
| 146 | + comm=comm, |
| 147 | + ) |
| 148 | + |
| 149 | + errorsDiff[i] = max( |
| 150 | + np.array( |
| 151 | + get_sorted(stats, type='e_global_differential_post_step', sortby='time', recomputed=False) |
| 152 | + )[:, 1] |
| 153 | + ) |
| 154 | + errorsAlg[i] = max( |
| 155 | + np.array(get_sorted(stats, type='e_global_algebraic_post_step', sortby='time', recomputed=False))[ |
| 156 | + :, 1 |
| 157 | + ] |
| 158 | + ) |
| 159 | + |
| 160 | + # only process with index 0 should plot |
| 161 | + if comm.Get_rank() == 0: |
| 162 | + orderDiff = np.mean( |
| 163 | + [ |
| 164 | + np.log(errorsDiff[i] / errorsDiff[i - 1]) / np.log(dt_list[i] / dt_list[i - 1]) |
| 165 | + for i in range(1, len(dt_list)) |
| 166 | + ] |
| 167 | + ) |
| 168 | + orderAlg = np.mean( |
| 169 | + [ |
| 170 | + np.log(errorsAlg[i] / errorsAlg[i - 1]) / np.log(dt_list[i] / dt_list[i - 1]) |
| 171 | + for i in range(1, len(dt_list)) |
| 172 | + ] |
| 173 | + ) |
| 174 | + |
| 175 | + refOrderDiff = 2 * comm.Get_size() - 1 |
| 176 | + refOrderAlg = 2 * comm.Get_size() - 1 if index_case == 1 else comm.Get_size() |
| 177 | + assert np.isclose( |
| 178 | + orderDiff, refOrderDiff, atol=1e0 |
| 179 | + ), f"Expected order {refOrderDiff} in differential variable, got {orderDiff}" |
| 180 | + assert np.isclose( |
| 181 | + orderAlg, refOrderAlg, atol=1e0 |
| 182 | + ), f"Expected order {refOrderAlg} in algebraic variable, got {orderAlg}" |
| 183 | + |
| 184 | + |
| 185 | +if __name__ == "__main__": |
| 186 | + comm = MPI.COMM_WORLD |
| 187 | + check_order(comm) |
0 commit comments