elastic-tube-1d: Solvers do not support subcycling in implicit coupling

[MakisH]

Looking at the FluidSolver.py, I noticed that a few things are a bit uncommon in the time loop:

tutorials/elastic-tube-1d/fluid-python/FluidSolver.py

Lines 127 to 156 in 95eea58

	while interface.is_coupling_ongoing():
	# When an implicit coupling scheme is used, checkpointing is required
	if interface.is_action_required(action_write_iteration_checkpoint()):
	interface.mark_action_fulfilled(action_write_iteration_checkpoint())
	velocity, pressure, success = perform_partitioned_implicit_euler_step(
	velocity_old, pressure_old, crossSectionLength_old, crossSectionLength, dx, precice_dt, velocity_in(
	t + precice_dt), custom_coupling=True)
	interface.write_block_scalar_data(pressureID, vertexIDs, pressure)
	interface.advance(precice_dt)
	crossSectionLength = interface.read_block_scalar_data(
	crossSectionLengthID, vertexIDs)
	# i.e. not yet converged
	if interface.is_action_required(action_read_iteration_checkpoint()):
	interface.mark_action_fulfilled(action_read_iteration_checkpoint())
	else: # converged, timestep complete
	t += precice_dt
	if plotting_mode is config.PlottingModes.VIDEO:
	tubePlotting.doPlotting(
	ax, crossSectionLength_old, velocity_old, pressure_old, dx, t)
	if writeVideoToFile:
	writer.grab_frame()
	ax.cla()
	velocity_old = np.copy(velocity)
	pressure_old = np.copy(pressure)
	crossSectionLength_old = np.copy(crossSectionLength)
	writeOutputToVTK(time_it, "out_fluid_", dx, datanames=["velocity", "pressure", "diameter"], data=[
	velocity_old, pressure_old, crossSectionLength_old])
	time_it += 1

It looks like the checkpointing assumes that the solver time step is equal to the coupling time window. More specifically, whenever preCICE does not ask the solver to read a checkpoint, the solver writes a checkpoint of the current solution. However, this also means that the checkpoint will be updated at every solver time step and the solver will not return to the right state when subcycling.

Two more problems:

• There is no checkpointing of the time itself, we only have a time_it += 1.
• We are essentially tricking preCICE, telling it that we read/wrote checkpoints at different points that we actually do. This makes the code difficult to follow.

The SolidSolver.py, as well as the C++ solvers, also have the same flaw.