Improve parallel network test.

The PID part of the timestep controller triggers and results in timestep
sizes depending on the state of the simulation. This can lead to different
timestepping for parallel and serial runs. Using only newton iteration count
for timestep calculations addresses this.

Also, tighten CNV tolerances to avoid solution differences that are accepted by
the simulator as within tolerances, but outside comparison tolerances.
This commit is contained in:
Atgeirr Flø Rasmussen 2024-07-31 17:14:26 +02:00
parent 125ee00336
commit e5fd6594d2

View File

@ -150,12 +150,12 @@ add_test_compare_parallel_simulation(CASENAME aquflux_02
TEST_ARGS --enable-tuning=true)
add_test_compare_parallel_simulation(CASENAME network_balance_01
FILENAME NETWORK-01
SIMULATOR flow
ABS_TOL 0.04
REL_TOL 0.02
DIR network
TEST_ARGS --enable-tuning=true)
FILENAME NETWORK-01
SIMULATOR flow
ABS_TOL ${abs_tol}
REL_TOL ${coarse_rel_tol_parallel}
DIR network
TEST_ARGS --enable-tuning=true --time-step-control=newtoniterationcount --time-step-control-growth-rate=3.0 --relaxed-max-pv-fraction=0.0 --tolerance-cnv=1e-3)
add_test_compare_parallel_simulation(CASENAME numerical_aquifer_3d_1aqu
FILENAME 3D_1AQU_3CELLS