Set initial time step automatically for restarted simulations
@cbsilver opened this issue as a bug report ("Restart simulation under adaptive time stepping for THM"). A possible solution to this bug is probably to set the initial time step automatically for restarted simulations.
Description
Running a T(H)M restart simulation can lead to significant deviations if the time step size at the restart point is not identical.
Steps to Reproduce the Problem
- Take the test example
Tests/Data/ThermoHydroMechanics/MultiMaterialEhlers/TM/square_1e1_2_matIDs.prj
. - Change the time stepping parameters such that
<initial_dt>1.e-2</initial_dt>
<minimum_dt>1.e-3</minimum_dt>
<maximum_dt>1.e-1</maximum_dt>
- Renew the reference data, i.e. run
ogs square_1e1_2_matIDs.prj
inTests/Data/ThermoHydroMechanics/MultiMaterialEhlers/TM/square_1e1_2_matIDs.prj
. - Run
ctest -R MultiMaterialEhlers
in your OGS build directory.
Expected behavior: The results of the full and the restart simulation are equivalent up to machine precision.
Actual behavior: Due to time integration differences (different dt's), there are significant deviations and the TM restart test fails.
The deviations originate from the time integration of the temperature equation. By the TM coupling, the deviations spread to the mechanical quantities. The remedy is to use the time step size at the restart point and provide the exact value for it. In Tests/Data/ThermoHydroMechanics/MultiMaterialEhlers/TM/square_1e1_2_matIDs_restart.xml
the corresponding line must be re-commented again.
Ideally, the time step size at the restart point is saved and stored for the restart automatically.
Specifications
- Version: 6.5.0
- Platform: Linux Ubuntu