initial() influences DAE

[Vitalij Ruge]

Consider this example

model nls
  Real x;
  Real y;
  Real z;
equation
  if initial() then
    x = y^2;
    y = x^2 + z;
    z = sqrt(x);
  else
    x = y;
    y = z;
    z = 2*x + time;
  end if;
end nls;

model ls
  Real x;
  Real y;
  Real z;
initial equation
  x = y^2;
  y = x^2 + z;
  z = sqrt(x);
equation
  x = y; // y = x
  y = z; // y = x = z
  z = 2*x + time; // x = 2*x - time // x = time
end ls;

In nls we creating a nonlinear system and in ls we creating a linear system for the DAE.
The models nls and ls are should be have the same DAE!

[Lennart Ochel]

First, there is no DAE at all ;-). The models are poorly designed. Probably there is a much better way to do whatever you want to do. What is the goal of this ticket? Just to do more simplifications on the dynamic problem?
The initial problems of both systems are different but weird anyway. Why should one formulate such a system?
The dynamic equations should be equal but are different due to the current back end design. Currently, there is no special optimization for the dynamic equations that evaluates initial() (same for #2751). That is the reason why there is a non-linear system in one case and a linear system in the other case.

[Vitalij Ruge]

First, there is no DAE at all ;-).

the dummy state say it's a DAE

The models are poorly designed.

this comment is still unhelpful

Probably there is a much better way to do whatever you want to do.

this comment is still unhelpful

What is the goal of this ticket?

Handling linear equation as linear equation, pls read description.

Just to do more simplifications on the dynamic problem?

Handling linear equation as linear equation, pls read description(it's a short one).

The initial problems of both systems are different but weird anyway.

Has not to do with the issue.

Why should one formulate such a system?

To show the issue

The dynamic equations should be equal but are different due to the current back end design.

The description tell, that the issue is spliting linear system and not lienear system on the relation with initial().
This issue can be expanded of general relations and not really binding on initial(). initial() is a good start, because is a known fixed realation and there must be are many groundwork for this case.

Currently, there is no special optimization for the dynamic equations that evaluates initial() (same for #2751).

I would prefer a solution with spliting ls and nls on realtion( for expanding it for all realtions/events).

[Vitalij Ruge]

So we can improve the evaluation of initial() or the spiting of linear system and not linear system depends on relation for solving this issues. I hope the second strategies is not really a design question of the BackEnd.

[Lennart Ochel]

I split up the equation systems for simulation and initialization in 56d186b60d7904ed253b34995e6ff4543a16e09d/OMCompiler. Now there are just some simplifications missing.

[Lennart Ochel]

Fixed in e7e6f99b7660a5213b710bca8cf679554de21f23/OMCompiler.

[Martin Sjölund]

Milestone renamed

[Martin Sjölund]

Milestone renamed