IDA Fixed Time Step Support

[Steven-Roberts]

Summary

If this is something that we want as a proper GridKit option, I think this would require adding a setTimeStep member function for GridKit::Model::Evaluator and its subclasses. If the out parameter is 0, this will indicate to use adaptive steps, otherwise the given time step will be used. In IDA.c there would need to be some minor changes to the integrator and nonlinear solver configuration when using fixed steps as done in https://code.ornl.gov/scidac-oe/fixed-step-ida-demo. I can start a PR if this sounds reasonable.

Rationale

This will be useful for convergence studies and comparisons against adaptive time stepping

Description

No response

Additional information

No response

[nkoukpaizan]

I agree this is good to support. I wonder whether GridKit::Model::Evaluator needs a setTimeStep (and, I am guessing a member variable storing the value) or if this can be handled entirely through Ida.cpp.

[Steven-Roberts]

I wanted to follow the approach used with Evaluator ::setTolerances as that is a similar sort of property that needs to be passed to IDA. I'm not opposed to handling integrator options in Ida.cpp as long as we're consistent.

[pelesh]

I wanted to following the approach used with Evaluator ::setTolerances as that is a similar sort of property that needs to be passed to IDA. I'm not opposed to handling integrator options in Ida.cpp as long as we're consistent.

I was setting tolerances in Evaluator only because IDA offers option to set separate absolute tolerances for model variables. To do that, one needs to be familiar with the model.

Now I think it would be better if tolerances are set in DynamicSolver class instead. Component models can set nominal values for individual variables and the absolute tolerance for each variable could be set as min(y_nominal[I]*reltol, abstol).

I suggest we handle setting tolerances through DynamicSolver base class.