|
| 1 | +import firedrake as fd |
| 2 | + |
| 3 | +from gusto.time_discretisation.time_discretisation import TimeDiscretisation, wrapper_apply |
| 4 | +from gusto.core.labels import explicit |
| 5 | + |
| 6 | +from pySDC.implementations.controller_classes.controller_nonMPI import controller_nonMPI |
| 7 | +from pySDC.implementations.problem_classes.GenericGusto import GenericGusto, GenericGustoImex |
| 8 | +from pySDC.core.hooks import Hooks |
| 9 | +from pySDC.helpers.stats_helper import get_sorted |
| 10 | + |
| 11 | + |
| 12 | +class LogTime(Hooks): |
| 13 | + """ |
| 14 | + Utility hook for knowing how far we got when using adaptive step size selection. |
| 15 | + """ |
| 16 | + |
| 17 | + def post_step(self, step, level_number): |
| 18 | + L = step.levels[level_number] |
| 19 | + self.add_to_stats( |
| 20 | + process=step.status.slot, |
| 21 | + process_sweeper=L.sweep.rank, |
| 22 | + time=L.time, |
| 23 | + level=-1, |
| 24 | + iter=-1, |
| 25 | + sweep=-1, |
| 26 | + type='_time', |
| 27 | + value=L.time + L.dt, |
| 28 | + ) |
| 29 | + |
| 30 | + |
| 31 | +class pySDC_integrator(TimeDiscretisation): |
| 32 | + """ |
| 33 | + This class can be entered into Gusto as a time discretization scheme and will solve steps using pySDC. |
| 34 | + It will construct a pySDC controller which can be used by itself and will be used within the time step when called |
| 35 | + from Gusto. Access the controller via `pySDC_integrator.controller`. This class also has `pySDC_integrator.stats`, |
| 36 | + which gathers all of the pySDC stats recorded in the hooks during every time step when used within Gusto. |
| 37 | + """ |
| 38 | + |
| 39 | + def __init__( |
| 40 | + self, |
| 41 | + equation, |
| 42 | + description, |
| 43 | + controller_params, |
| 44 | + domain, |
| 45 | + field_name=None, |
| 46 | + solver_parameters=None, |
| 47 | + options=None, |
| 48 | + t0=0, |
| 49 | + imex=False, |
| 50 | + ): |
| 51 | + """ |
| 52 | + Initialization |
| 53 | +
|
| 54 | + Args: |
| 55 | + equation (:class:`PrognosticEquation`): the prognostic equation. |
| 56 | + description (dict): pySDC description |
| 57 | + controller_params (dict): pySDC controller params |
| 58 | + domain (:class:`Domain`): the model's domain object, containing the |
| 59 | + mesh and the compatible function spaces. |
| 60 | + field_name (str, optional): name of the field to be evolved. |
| 61 | + Defaults to None. |
| 62 | + solver_parameters (dict, optional): dictionary of parameters to |
| 63 | + pass to the underlying solver. Defaults to None. |
| 64 | + options (:class:`AdvectionOptions`, optional): an object containing |
| 65 | + options to either be passed to the spatial discretisation, or |
| 66 | + to control the "wrapper" methods, such as Embedded DG or a |
| 67 | + recovery method. Defaults to None. |
| 68 | + """ |
| 69 | + |
| 70 | + self._residual = None |
| 71 | + |
| 72 | + super().__init__( |
| 73 | + domain=domain, |
| 74 | + field_name=field_name, |
| 75 | + solver_parameters=solver_parameters, |
| 76 | + options=options, |
| 77 | + ) |
| 78 | + |
| 79 | + self.description = description |
| 80 | + self.controller_params = controller_params |
| 81 | + self.timestepper = None |
| 82 | + self.dt_next = None |
| 83 | + self.imex = imex |
| 84 | + |
| 85 | + def setup(self, equation, apply_bcs=True, *active_labels): |
| 86 | + super().setup(equation, apply_bcs, *active_labels) |
| 87 | + |
| 88 | + # Check if any terms are explicit |
| 89 | + imex = any(t.has_label(explicit) for t in equation.residual) or self.imex |
| 90 | + if imex: |
| 91 | + self.description['problem_class'] = GenericGustoImex |
| 92 | + else: |
| 93 | + self.description['problem_class'] = GenericGusto |
| 94 | + |
| 95 | + self.description['problem_params'] = { |
| 96 | + 'equation': equation, |
| 97 | + 'solver_parameters': self.solver_parameters, |
| 98 | + 'residual': self._residual, |
| 99 | + } |
| 100 | + self.description['level_params']['dt'] = float(self.domain.dt) |
| 101 | + |
| 102 | + # add utility hook required for step size adaptivity |
| 103 | + hook_class = self.controller_params.get('hook_class', []) |
| 104 | + if not type(hook_class) == list: |
| 105 | + hook_class = [hook_class] |
| 106 | + hook_class.append(LogTime) |
| 107 | + self.controller_params['hook_class'] = hook_class |
| 108 | + |
| 109 | + # prepare controller and variables |
| 110 | + self.controller = controller_nonMPI(1, description=self.description, controller_params=self.controller_params) |
| 111 | + self.prob = self.level.prob |
| 112 | + self.sweeper = self.level.sweep |
| 113 | + self.x0_pySDC = self.prob.dtype_u(self.prob.init) |
| 114 | + self.t = 0 |
| 115 | + self.stats = {} |
| 116 | + |
| 117 | + @property |
| 118 | + def residual(self): |
| 119 | + """Make sure the pySDC problem residual and this residual are the same""" |
| 120 | + if hasattr(self, 'prob'): |
| 121 | + return self.prob.residual |
| 122 | + else: |
| 123 | + return self._residual |
| 124 | + |
| 125 | + @residual.setter |
| 126 | + def residual(self, value): |
| 127 | + """Make sure the pySDC problem residual and this residual are the same""" |
| 128 | + if hasattr(self, 'prob'): |
| 129 | + self.prob.residual = value |
| 130 | + else: |
| 131 | + self._residual = value |
| 132 | + |
| 133 | + @property |
| 134 | + def level(self): |
| 135 | + """Get the finest pySDC level""" |
| 136 | + return self.controller.MS[0].levels[0] |
| 137 | + |
| 138 | + @wrapper_apply |
| 139 | + def apply(self, x_out, x_in): |
| 140 | + """ |
| 141 | + Apply the time discretization to advance one whole time step. |
| 142 | +
|
| 143 | + Args: |
| 144 | + x_out (:class:`Function`): the output field to be computed. |
| 145 | + x_in (:class:`Function`): the input field. |
| 146 | + """ |
| 147 | + self.x0_pySDC.functionspace.assign(x_in) |
| 148 | + assert self.level.params.dt == float(self.dt), 'Step sizes have diverged between pySDC and Gusto' |
| 149 | + |
| 150 | + if self.dt_next is not None: |
| 151 | + assert ( |
| 152 | + self.timestepper is not None |
| 153 | + ), 'You need to set self.timestepper to the timestepper in order to facilitate adaptive step size selection here!' |
| 154 | + self.timestepper.dt = fd.Constant(self.dt_next) |
| 155 | + self.t = self.timestepper.t |
| 156 | + |
| 157 | + uend, _stats = self.controller.run(u0=self.x0_pySDC, t0=float(self.t), Tend=float(self.t + self.dt)) |
| 158 | + |
| 159 | + # update time variables |
| 160 | + if self.level.params.dt != float(self.dt): |
| 161 | + self.dt_next = self.level.params.dt |
| 162 | + |
| 163 | + self.t = get_sorted(_stats, type='_time', recomputed=False)[-1][1] |
| 164 | + |
| 165 | + # update time of the Gusto stepper. |
| 166 | + # After this step, the Gusto stepper updates its time again to arrive at the correct time |
| 167 | + if self.timestepper is not None: |
| 168 | + self.timestepper.t = fd.Constant(self.t - self.dt) |
| 169 | + |
| 170 | + self.dt = self.level.params.dt |
| 171 | + |
| 172 | + # update stats and output |
| 173 | + self.stats = {**self.stats, **_stats} |
| 174 | + x_out.assign(uend.functionspace) |
0 commit comments