-
Notifications
You must be signed in to change notification settings - Fork 0
/
porepy_common.py
217 lines (188 loc) · 7.91 KB
/
porepy_common.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
import numpy as np
from porepy import SolutionStrategy
from solvers_common import LinearSolver, LinearSolverStatistics
from solver_selector.data_structures import (
NonlinearIterationStats,
NonlinearSolverStats,
SolverSelectionData,
)
from solver_selector.simulation_runner import SimulationModel, Solver
from solver_selector.utils import TimerContext
class PorepySimulation(SimulationModel):
porepy_setup: SolutionStrategy
def _compute_cfl(self):
setup = self.porepy_setup
subdomains = setup.mdg.subdomains()
velosity = setup.darcy_flux(subdomains) / setup.fluid_viscosity(subdomains)
velosity = velosity.value(setup.equation_system)
length = setup.mdg.subdomains()[0].face_areas
time_step = setup.time_manager.dt
CFL = velosity * time_step / length
CFL_max = abs(CFL).max()
CFL_mean = abs(CFL).mean()
return CFL_max, CFL_mean
def is_complete(self) -> bool:
time_manager = self.porepy_setup.time_manager
return time_manager.time >= time_manager.time_final
def after_time_step_success(self, solver_selection_data: SolverSelectionData):
model = self.porepy_setup
model.time_manager.increase_time()
model.time_manager.increase_time_index()
def after_simulation(self):
self.porepy_setup.after_simulation()
def before_time_step(self) -> None:
model = self.porepy_setup
print(
"\nTime step {} at time {:.1e} of {:.1e} with time step {:.1e}".format(
model.time_manager.time_index,
model.time_manager.time,
model.time_manager.time_final,
model.time_manager.dt,
)
)
model.before_nonlinear_loop()
class LinearSolverFailed(Exception):
pass
class PorepyNewtonSolver(Solver):
def __init__(self, linear_solver: LinearSolver) -> None:
self.linear_solver: LinearSolver = linear_solver
self.params = {"max_iterations": 15, "nl_convergence_tol": 1e-6}
self.error_norms = []
self.residual_norms = []
def make_time_step(self, simulation: PorepySimulation) -> NonlinearSolverStats:
return self.solve(simulation.porepy_setup)
def newton_iteration(
self,
model: SolutionStrategy,
init_sol: np.ndarray,
prev_sol: np.ndarray,
iteration_counter: int,
) -> dict:
try:
# Assemble the new Jacobian matrix
with TimerContext() as assembly_timer:
# Re-discretize the nonlinear term
model.before_nonlinear_iteration()
model.assemble_linear_system()
mat, rhs = model.linear_system
# If the matrix is very bad, the linear solver will anyway fail.
if np.any(np.isnan(mat.data)):
raise LinearSolverFailed("Jacobian contains nans.")
# Update the preconditioner with the new matrix
with TimerContext() as update_timer:
self.linear_solver.update(mat)
# Solve the Jacobian linear system
with TimerContext() as solve_timer:
sol, linear_stats = self.linear_solver.solve(rhs)
# Check convergence
model.after_nonlinear_iteration(sol)
if model._is_nonlinear_problem():
error_norm, is_converged, is_diverged = model.check_convergence(
sol, prev_sol, init_sol, self.params
)
# if not linear_stats.is_converged and not is_converged:
# is_diverged = True
# # This is kind of strict, but currently we do not consider
# # inexact Newton.
else: # Linear problem
if linear_stats.is_converged:
is_diverged = False
is_converged = True
error_norm = linear_stats.residual_decrease
else:
is_diverged = True
is_converged = False
error_norm = -1
self.residual_norms.append(np.linalg.norm(rhs))
if (
linear_stats.residual_decrease is not None
and linear_stats.residual_decrease > 1e16
):
is_diverged = True
if not linear_stats.is_converged:
print("Linear solver failed.")
if linear_stats.is_diverged:
print("Linear solver diverged.")
if is_diverged:
print("Nonlinear solver diverged.")
self.error_norms.append(error_norm)
print(
f"Newton iter: {iteration_counter}, error: {self.error_norms[-1]}, "
f"linear iters: {linear_stats.num_iters}"
)
except LinearSolverFailed as e:
print(e)
sol = prev_sol
is_diverged = True
is_converged = False
linear_stats = LinearSolverStatistics(
num_iters=-1, is_converged=False, is_diverged=True
)
solve_time = 0
update_time = 0
assembly_time = assembly_timer.elapsed_time
else:
solve_time = solve_timer.elapsed_time
assembly_time = assembly_timer.elapsed_time
update_time = update_timer.elapsed_time
return {
"sol": sol,
"is_converged": is_converged,
"is_diverged": is_diverged,
"linear_stats": linear_stats,
"solve_time": solve_time,
"assembly_time": assembly_time,
"update_time": update_time,
}
def solve(self, porepy_setup: SolutionStrategy) -> NonlinearSolverStats:
model = porepy_setup
sol = model.equation_system.get_variable_values(time_step_index=0)
init_sol = sol
self.error_norms = []
self.residual_norms = []
iteration_stats = []
iteration_counter = 0
for iteration_counter in range(self.params["max_iterations"]):
with TimerContext() as work_timer:
iteration = self.newton_iteration(
model, init_sol, sol, iteration_counter
)
is_converged = iteration["is_converged"]
is_diverged = iteration["is_diverged"]
sol = iteration["sol"]
linear_stats: LinearSolverStatistics = iteration["linear_stats"]
if (res_decrease := linear_stats.residual_decrease) is None:
res_decrease = -1
iteration_stats.append(
NonlinearIterationStats(
work_time=work_timer.elapsed_time,
solve_linear_system_time=iteration["solve_time"],
assembly_time=iteration["assembly_time"],
update_preconditioner_time=iteration["update_time"],
linear_solver_converged=linear_stats.is_converged,
num_linear_iterations=linear_stats.num_iters,
linear_residual_decrease=res_decrease,
)
)
if is_converged or is_diverged:
break
if len(self.residual_norms) > 1:
# We do not compute residual norm on the last iteration, so it must be less
# than on the one second from last iteration.
print(
"||F||/||F_0|| < "
f"{self.residual_norms[-1] / self.residual_norms[0]:.2e}"
)
if is_converged:
model.after_nonlinear_convergence(sol, self.error_norms, iteration_counter)
else:
try:
model.after_nonlinear_failure(sol, self.error_norms, iteration_counter)
except ValueError:
pass
return NonlinearSolverStats(
is_converged=is_converged,
is_diverged=is_diverged,
nonlinear_error=self.error_norms,
iterations=iteration_stats,
)