Feature to break the simulation when the steady state has been reached

[Adrian Pop]

Proposed here:
​http://stackoverflow.com/questions/42673420/global-measure-to-test-when-steady-state-of-system-is-reached

The first point sounds like a good to have feature and doesn't seem that hard to implement, basically a terminate when all the ders are 0 (within an epsilon) for a number of steps.

For the second point we would need a global operator (that returns true when all ders are within an epsilon interval) that one can use in equations to enable / disable parts of the system.

[Lennart Ochel]

I can take care of the first point.

[Adrian Pop]

For the second point we could propose an extension of the Modelica Specification with an operator called steadyState(epsilon) which would return true if the sums of ders are < epsilon.

The only issues to think about is what to return during initialization when the ders are zero, etc. I guess we can safely return false for any other phase than simulation.

[Adrian Pop]

This is back-end for the first point and front-end and back-end for the second point.

[Lennart Ochel]

See ​OMCompiler#1531 for the first point.

[anonymous]

Somehow this feature is invalidated by sources adding perturbations after derivatives are all zero.
Should this feature check also sources?

[Lennart Ochel]

Replying to anonymous:

Should this feature check also sources?

I don't think that this is possible (or necessary). Anyway, can you provide an example to emphasise your point?

[Adrian Pop]

Replying to lochel:

Replying to anonymous:

Should this feature check also sources?

I don't think that this is possible (or necessary). Anyway, can you provide an example to emphasise your point?

It is possible I guess if you have time events and the event comes after the steady state has been reached. Maybe we could check if there is a time event in the queue and give a warning about it when terminating the simulation.

[anonymous]

Replying to lochel:

Replying to anonymous:

Should this feature check also sources?

I don't think that this is possible (or necessary). Anyway, can you provide an example to emphasise your point?

Consider RL.mo: everything is constant between 4.5 and 5s; But starting from t=5s current starts to change again.
In this case there is a time event in the queue (inside step block)

[Christoph Buchner <buchner@…>]

When detecting the steady state, do you not also have to make sure that all derivatives *stay* below the threshold? Otherwise, if you have a simple oscillation, would the simple logic not interrupt the simulation as soon as the first maximum of the only state is reached (as there the der is 0)?

[Lennart Ochel]

Replying to Christoph Buchner <buchner@…>:

When detecting the steady state, do you not also have to make sure that all derivatives *stay* below the threshold? Otherwise, if you have a simple oscillation, would the simple logic not interrupt the simulation as soon as the first maximum of the only state is reached (as there the der is 0)?

This is a good point. So I could introduce a (user defined) time interval in which all derivatives have to be zero. Would that be a good idea?

[Lennart Ochel]

Replying to anonymous:

Replying to lochel:

Replying to anonymous:

Should this feature check also sources?

I don't think that this is possible (or necessary). Anyway, can you provide an example to emphasise your point?

Consider RL.mo: everything is constant between 4.5 and 5s; But starting from t=5s current starts to change again.
In this case there is a time event in the queue (inside step block)

I see. However, I'm not sure if we should include a check for time events, because it is not possible to do the same for state events, which would make things confusing.

[anonymous]

Furthermore, it is possible to create systems in which the simulation suddenly "fires-up", without events.
An example (a bit tricky, indeed) is in RL2.mo.

[anonymous]

AFAIK it is impossible to have oscillations system with a single state and non-oscillating source.
So if all derivatives are zero, they will stay as such, except for cases in which the sources change their values.

Replying to lochel:

Replying to Christoph Buchner <buchner@…>:

When detecting the steady state, do you not also have to make sure that all derivatives *stay* below the threshold? Otherwise, if you have a simple oscillation, would the simple logic not interrupt the simulation as soon as the first maximum of the only state is reached (as there the der is 0)?

This is a good point. So I could introduce a (user defined) time interval in which all derivatives have to be zero. Would that be a good idea?

[Martin Sjölund]

The testcase added for this is not working: 9bca22796