Problems simulating simple nonlinear model

[Leonardo Laguna]

I have the following simple nonlinear model

  model SimpleNonlinear
    Real y2;
    Real y3;                              
  equation
    y2+y3=0;
    y2*y3=-3;
  end SimpleNonlinear;

This model cannot be simulated. It reports the following errors.

Translation	09:32:05		0:0-0:0	Internal error Transformation Module PFPlusExt index Reduction Method Pantelites failed!
Translation	09:32:05		0:0-0:0	Internal error Circular Equalities Detected
Translation	09:32:53		0:0-0:0	Error building simulator. Buildlog: gcc   -falign-functions -march=native -mfpmath=sse   -I"/usr/include/omc" -I.    -c -o ExtractionSetS_NL_Test.o ExtractionSetS_NL_Test.c
gcc   -falign-functions -march=native -mfpmath=sse   -I"/usr/include/omc" -I.    -c -o ExtractionSetS_NL_Test_records.o ExtractionSetS_NL_Test_records.c
g++ -I. -o ExtractionSetS_NL_Test ExtractionSetS_NL_Test.o ExtractionSetS_NL_Test_records.o -I"/usr/include/omc" -I.        -falign-functions -march=native -mfpmath=sse   -L"/usr/lib/omc" -Wl,-rpath,'/usr/lib/omc' -lSimulationRuntimeC   -ldl -linteractive  -lexpat -Wl,-Bsymbolic-functions -Wl,-z,relro  -lrt -lgc -fopenmp -lsundials_kinsol -lsundials_nvecserial -llapack  -lpthread -Wl,-Bstatic -lf2c -Wl,-Bdynamic
/usr/bin/ld: cannot find -lf2c
collect2: error: ld returned 1 exit status
make: *** [omc_main_target] Error 1

Translation	09:32:05		0:0-0:0	Internal error Optimisation Module removeSimpleEquations failed.

This model is a reduced version of one of the models in the uncertainties test bank.

[Martin Sjölund]

What OS / how did you install omc?

[Leonardo Laguna]

I have a nightly build version. I'm not sure the date but is less that a week old. I'm running on linux.

The main problem is that this model does not pass the "lower" part in the backend, therefore it cannot be used in the uncertainties algorithm.

The linking error is not an issue for me since the models for uncertainties are not compiled.

[Martin Sjölund]

It most certainly does pass the lowering part of the backend. I just tested it and there were no errors. Run dpkg -s omc to get me the version you are using.

[Leonardo Laguna]

Status: install ok installed
Priority: optional
Section: math
Installed-Size: 48129
Maintainer: OpenModelica Build System <build@…>
Architecture: i386
Source: openmodelica
Version: 15292-1

[Leonardo Laguna]

Actually, some of the error messages come from the full model (see below). I copied the all the error messages in the message window (my bad). This is the full model:

  model ExtractionSetS_NL_Test
    Real x1;
    Real x2;
    Real x3;
    Real y1;
    Real y2;
    Real y3;                      
  equation
    x1+x2 = 0;
    x1-x2 = 0;
    y1=x2+2*x3;
    x3-y1+y2=x2;
    y2+y3=0;
    y2*y3=-3;
  end ExtractionSetS_NL_Test;

Even with circular equalities, this model is valid and it should simulate.

[Martin Sjölund]

OK, the following is a smaller model that does still have the same error:

model M
  Real x1;
  Real x2;
equation
  x1+x2 = 0;
  x1-x2 = 0;
end M;

It's equivalent to:

     x1 = x2;
     0 = 0;

Which OpenModelica cannot solve, and I'm not sure it's supposed to either... Jens or Willi can tell you more.

[Martin Sjölund]

Or wait... Hmmm... Shouldn't it be

x1 = x2; // alias
2*x1 = 0; => x1 = 0.0

It should be possible to ignore the two potential aliases (x1=x2 or x1=-x2), and keep going using one of them.

[Jens Frenkel]

For me the models works as expected.

if you add both equations you get

x1+x2 = 0;
x1-x2 = 0;
 +
x1 = 0;

Hence the solution is x1=x2=0;

Known Variables (constants) (2)
========================================
1:  x2:VARIABLE(start = 0.0 )  = 0.0 .Wire.M, .Real type: Real 
2:  x1:VARIABLE(start = 0.0 )  = 0.0 .Wire.M, .Real type: Real 

[Martin Sjölund]

Works, yes. But not as expected. You get:

Error: Internal error Circular Equalities Detected
Error: Internal error Optimisation Module removeSimpleEquations failed.

Which I assume should not be there.

[Lennart Ochel]

Replying to sjoelund.se:

It's equivalent to:

     x1 = x2;
     0 = 0;

Not, it is not. It is equivalent to:

x1 = x2;
x1 = 0;

edit: not fast enough ...

[Jens Frenkel]

Ok, so I have fixed this on my local copy. I will push in my changes when testsuite was succesfull.

[Leonardo Laguna]

Is this fix already in the trunk code?

[Jens Frenkel]

• fixed in r15999