Suggesting a new end criteria for step excitation

[ravid1323]

I understand from the documentation that setting end criteria is relevant only for decaying excitations which make sense.
For example, gaussian pulse is taking N timesteps to insert into the system, and then the energy in the line is decaying over time

On step excitation that's not the case ofcourse, step is converging to a constant, which where my suggestion is coming in: for step excitation, I think it would be good practice to consider an end criteria not for the total energy En(t), but for dEn(t)/dt, meaning that if En(t) is converging to a constant, dEn(t)/dt is converging to 0, also logarithmic-wise

ofcourse it is worth also defining when the solver will start checking the end criteria, like in gaussian the end criteria is effectively relevant only after timestep X+1, where the solver is determine that the gaussian pulse insertion is completed after X timesteps

[VolkerMuehlhaus]

I guess you already know this: you can also use the number of timesteps as the end criteria, which is useful for step & pulse excitation when looking at time domain response.

[ravid1323]

Indeed, the timesteps limitation exists and is useful for what it was purposed for, helping avoid the simulation from running infinitely in case of numeric instability and in this case as a good enough criteria for step and pulse excitation.
However, we might be able to add an improvement,, perhaps not good enough criteria in terms of physics for when the transient is ending. for that same reason, timestep limitation is not the only end criterita for other finite energy excitations like gaussian

for "infinite energy excitations" like step, looking ad the decaying of dEn(t)/dt, can give end criteria based on the physics of the scenario and tell us when the transient is close to end. So rather than saying "ok I decided it's running too long already", we can say "ok I decided that the transient close enough to converge", which is more similar to "ok I decided that most of the energy has left the system" that we determine in finite energy excitations

[VolkerMuehlhaus]

@ravid1323 Sure, maybe I didn't understand your use case, sorry for that!

For my RF/microwave stuff, I sometimes use step/pulse to look at wideband transitions in time domain, to identify undesired discontinuities (time domain reflectometry method). In that case, the time period of interest is given by time of flight considerations (get the first few reflections) and there is no useful energy convergence criteria involved.

But it seems you have very different use cases in mind!

[thliebig]

Well there could be had the discussion about how much sense a step excitation actually makes for FDTD which is really not ideal for DC or very low frequencies, but I guess that is a different topic.

I am not sure that looking at the diff for the energy will really always work as intended, mostly because the "energy" is only an estimate and if energy is still traveling around through different materials the value will not be steady...
But I assume it still would be much better as the energy itself never decaying.

Since this is all a very niche problem I am not sure that I am comfortable adding all kind of new and different end criteria for this.
But what could be very useful in many different ways could be implementing a callback function to be set by the python interface that gets called with every (energy/status) update, delivering all available data to python and letting that callback function decide would to do with it (e.g. stopping the sim or displaying some custom status message etc.)

I think this would be the most flexible and would solve the issue at hand nicely?