Change of input file / problem parameters during runtime

[terencelehmann]

In a recent project, I implemented the option to change the values of selected problem parameters in LMeX dynamically (i.e., during a simulation run). I’d like to know if there’s a relevant use case for this feature (or something similar) in the main LMeX version. The current setup is not ideal, but I’m open to improving it.

Overview of workflow

1. Add an updateProbParm function in pelelmex_prob.cpp, specifying which variables should be re-read.
2. Write these values to a separate input file, DynamicProbParms.inp.
3. During the simulation, edit the value(s) in DynamicProbParms.inp and trigger re-reading with touch update_prob_parm. This calls the checkMessage function (similar to plt_and_continue). When called, updateProbParm appends the content of DynamicProbParms.inp to the parameters and re-reads the specified values.

Why this could be useful

Unlike a time-dependent parameter change (e.g., in the bc_normal function), this approach does not require all changes to be known at the start of the simulation. For instance, in a flashback study, you can incrementally adjust the inflow velocity toward a limit, or drive the changes via an external controller. This workflow can be simpler than stopping and restarting the simulation with a modified input file.

What is not optimal with the current implementation

• While the goal is to provide flexibility at runtime, the set of controllable variables must still be known at compile time. A workaround is to include all problem parameters in the re-reading.
• Currently, the updated DynamicProbParms.inp is always appended to the existing parameters by calling pp.addfile("DynamicProbParms.inp"). A better approach might exist to avoid repeatedly stacking entries in ParmParse.
• Currently, the setup is quite manual. One could include a default version of the update function in PeleLMeX_ProblemSpecificFunctions.H.

I’d be happy to hear your opinions and suggestions on how to refine this feature before merging it into LMeX. If you think there’s a similar feature already, or that this isn’t needed, that’s also fine.

[marchdf]

@terencelehmann this is a neat feature. I am currently looking into whether AMReX has some machinery that can make it easier. Hang tight ;)

[dmontgomeryNREL]

I love this idea and always appreciated this capability in OpenFOAM. I found it especially helpful for updating simulation end times, dt, etc, but these are all AMReX-related parameters. Curious to see what @marchdf digs up.