Regarding the support of multiple MicroSim models of potentially different fidelity in the model-adaptivity branch:

What should be done going forward?
Option A:

• Micro-Manager operates in phases, during which ROM and/or coarser models are automatically generated
• Subsequently, the generated MicroSimulation classes can have a well defined method to decide whether model fidelity should be in- or decreased using the following concepts:
  • for ROM: if input is far beyond scope of underlying dataset, then uncertainty is high => use FOM
  • for ANN: If input is far beyond scope of input training data, use other model. (similarly with output)
  • These can be expressed as a distance metric, which, if the provided thresholds are met, switching occurs

Option B:

• Models are fully provided by user
• Switching criterion is provided in additional script file, analogously to MicroSimulation using uniform interface
• Interface will likely contain input, model, location, time etc

Option C:

• Is combination of prior approaches
• Using Option A as base concept
• However, models are not created automatically.
• Instead, user will need to implement some switching criterion in their MicroSimulation classes