[hajott]

Hello,
what is the recommended way to repeat a time step in time dependent solvers, when a given accuracy criterion is not met at the end of the actual time step?

The existing solvers, which can work with "adjustableTimeStep yes;" in controlDict, set the FUTURE time step based on the courant number or other criteria and don't care if the courant number is too large AFTER the time step has been solved.

But what if my solver decides that the actual step has been calculated with a too large deltaT, and I have to dismiss the latest solution, and start over with a smaller deltaT? My idea is to implement something like this:

while(runTime.run()){
  runTime++;
  solve();
  while (!isAccurate()){
    undoSolve();
    runTime.setDeltaT(0.5*runTime.deltaTValue();
    solve();
  }
}

But how do I implement "undoSolve()"? All fields, their values stored from previous timesteps and all other conditions must be reset as if the last "solve()" didn't happen. Does OpenFOAM offer an easy solution for this, or do I have to backup and possibly restore all fields?

Regards

[alexeym]

Hi,

you can try doing something like:

while(runTime.run()){
  for (;;){
    solve();
    if (!isAccurate())
    {
      runTime.setDeltaT(0.5*runTime.deltaTValue();
      // Restoring old value of fields
        // Scalar fields
        {
            HashTable<volScalarField*> flds =
                lookupClass<volScalarField>();

            forAllIter(HashTable<volScalarField*>, flds, iter)
            {
                if (iter()->nOldTimes() < 1)
                {
                        continue;
                }

                iter()->internalField() = iter()->oldTime();
            }
        }

        // Vector fields
        {
            HashTable<volScalarField*> flds =
                lookupClass<volScalarField>();

            forAllIter(HashTable<volScalarField*>, flds, iter)
            {
                if (iter()->nOldTimes() < 1)
                {
                        continue;
                }

                iter()->internalField() = iter()->oldTime();
            }
        }
     }
     else
     {
        break;
     }
  }
  runTime++;
}

see http://foam.sourceforge.net/docs/cpp...c7d231aa6b5f3f for more thoughts.

[hajott]

Thank you, your suggestion looks very promising. I didn't know about lookupClass<T>, which is very helpful, because with this method I cannot forget to backup and restore those fields that I add to the solver during the development process.

One more question about GeometricField.oldTime(): I have a partial understanding about this method from reading the implementation. At which point within the runTime loop is the oldest (for example t(n-3)) history discarded if my solver needs the actual and too previous timesteps (t(n),t(n-1),t(n-2))?

I wonder if t(n-2) would be restored by your code at "iter()->internalField() = iter()->oldTime();" or if it has been discarded already at that point.

Finally, your example only resets the internalField to the old values. I suppose I have to reset the boundaryField, too, using iter()->correctBoundaryConditions(), don't I?

[alexeym]

The idea of lookupClass<T> was taken from src/dynamicFvMesh/dynamicRefineFvMesh/dynamicRefineFvMesh.C:305.

I wonder if t(n-2) would be restored by your code at "iter()->internalField() = iter()->oldTime();" or if it has been discarded already at that point.

Don't know, guess you have to use experimental verification.

Finally, your example only resets the internalField to the old values. I suppose I have to reset the boundaryField, too, using iter()->correctBoundaryConditions(), don't I?

Well, yes. Thought, partly it depends on what you are doing in solve method.