This is the opposite way of writing things than in the OpenFOAM example! If we want to couple to OpenFOAM we have to change that.
This is the opposite way of writing things than in the OpenFOAM example! If we want to couple to OpenFOAM we have to change that.
#### Code related questions
What is the best way to reload a checkpoint?
Saving `sol` and then
1. reloading `sol` via `GridVariables->update(sol)` followed by a `GridVariables->advanceTimeStep()`.
1. reloading `sol` via `GridVariables->init(sol)`;
Codes:
```c++
GridVariables->update(sol);
GridVariables->advanceTimeStep();
```
```c++
GridVariables->init(sol)
```
#### Problems
#### Problems
-[ ] We have to be careful with the initialization of the solvers.
-[x] We have to be careful with the initialization of the solvers.
-**Valid case** Heat solver starts: The heat solver computes a temperature to be send, the heat flux is zero. Thus, the temperature equals the temperature in the cell center.
-**Valid case** Heat solver starts: The heat solver computes a temperature to be send, the heat flux is zero. Thus, the temperature equals the temperature in the cell center.
-**Invalid case** Flow solver starts: The flow solver computes a heat flux from its temperatures. However, the temperature on the interface (stored in the precice adapter) is zero! Thus, a **very high** heat flux is initialized.
-**Invalid case** Flow solver starts: The flow solver computes a heat flux from its temperatures. However, the temperature on the interface (stored in the precice adapter) is zero! Thus, a **very high** heat flux is initialized.
-[ ] The implicit coupling creates a solution that develops too fast.
-**Note**: There were some errors in the preCICE configuration file.
-[x] How do we control the adaptive time stepping? It increases above the suggested time step size which makes it impossible for preCICE to catch the solvers.
- [x] The implicit coupling creates a solution that develops too fast.
-**Needs more testing** The checkpointing die only reload parts of the solution. In the end the Newton solver continued from its previous state, but with the initial guess being reset.
- [x] How do we control the adaptive time stepping? It increases above the suggested time step size which makes it impossible for preCICE to catch the solvers.
- I have adapted the adaptive solver. I hope it works properly now!
- I have adapted the adaptive solver. I hope it works properly now!