RemoveSimpleEquations module should remove simple equations with parametric coefficients

[Maksimov Doe]

RemoveSimpleEquations module cannot remove equations like a = k * b where a and b are some component references and k is a parameter expression. Because of that in the example below compiler cannot deduce that the value of z is zero.

model TestFactorSimpleRemoval
  Real x = time;
  Real y = 2 * time;
  Real z;
equation
  z = 2 * x - y;
end TestFactorSimpleRemoval;

Currently this model produce following equation system

########################################
dumpindxdae
######################################## 
unspecified partition
======================================== Variables (3)
========================================
1: z:VARIABLE() TestFactorSimpleRemoval type: Real 
2: y:VARIABLE() TestFactorSimpleRemoval type: Real 
3: x:VARIABLE() TestFactorSimpleRemoval type: Real  
Equations (3, 3)
========================================
1/1 (1): z = 2.0 * x - y   [dynamic |0|0|0|0|]
2/2 (1): x = time   [binding |0|0|0|0|]
3/3 (1): y = 2.0 * time   [binding |0|0|0|0|] 
Matching
========================================
3 variables and equations
var 1 is solved in eqn 1
var 2 is solved in eqn 3
var 3 is solved in eqn 2 
StrongComponents
========================================
{3:2}
{2:3}
{1:1} 

BackendDAEType: simulation 

Module RemoveSimpleEquations already handles negative sign and work with parametric coefficients could be done in the same manner.

[Martin Sjölund]

I believe the example is bad. time is never considered a simple equation. If it was some other expression it would show a valid enhancement. The thing is that the runtime system does not know about these things (and the simulation result file does not support these kinds of things). For a variable x = 5, we can simply put it in the simulation results as 5 (like a parameter), and for x = y we put it as part of the same column. We also have a special case to invert a variable x = -y.

​Tiller and Harman proposed not only this to have aliases:

• x = alias(y)
• x = inv(y)

But also the affine transformation:

• x = aff(y,s,o) (x = y*s + o)

You could of course add such a transformation and simply make the result-files output these in every step.

[Maksimov Doe]

I don't think that time variable has a matter, we can replace it with any other component reference.

Considering the affine transformations aliasing, arent that supported in OpenModelica?

[Francesco Casella]

Replying to gossen:

Considering the affine transformations aliasing, arent that supported in OpenModelica?

I'm not aware of that, only a = +/-b