Wrong simulation results when using simflags -nls="hybrid" or -nls="kinsol"

[bernhardbachmann]

Description

Please consider the attached test case GridBaseBreakers.mo.txt

Running this model with the default non-linear solver "hybrid" produces errors at time 83.093 and later as well as the simulation results are not as expected. . Running the same model with Kinsol produces even worse results. Running the same model with the non-linear solver "homotopy" or "newton" works just fine.

Steps to Reproduce

To reproduce the error please use the attached script file GridBaseBreakers.mos.txt

Screenshots

Wrong result using -nls="hybrid":

Wrong result using -nls="kinsol":

Correct result using -nls"homotopy":

Version and OS

• OpenModelica Version: OMCompiler v1.19.0-dev.603+g56e1f47568

[casella]

A similar behaviour took place with models from the PowerGrids library. In fact, all these models are best solved with daeMode for efficiency reasons (the jacobian in daeMode is sparse, while it becomes dense in odeMode).

In any case, there is no reason why odeMode simulation should produce wrong results, and this bug could also affect models where there is no advantage in using daeMode, so we need to figure out why this happens and fix it.

[bernhardbachmann]

I made further test with the basic model of the PowerGridDae library GridBase.mo model. See attached file GridBase.mo.txt

The solution of the problem highly depends on the tearing method used. Please look at the attached scripting file for the exact settings: GridBase.mos.txt

--tearingMethod=cellierTearing (default), -nls="hybrid"

--tearingMethod=cellierTearing (default), -nls="homotopy"

--tearingMethod=omcTearing, -nls="hybrid"

--tearingMethod=omcTearing, -nls="homotopy"

[casella]

Well, these plots are really awful. Apparently, tearing picks some iteration variables that have a low sensitivity on the solution.

However, all this noise has a visible influence on the state variables, and those have error control. I find it really weird that such solutions, with completely bogus oscillations on the omega and theta states, are accepted.