nominal values get not evaluated if they depend on parameters

[Lennart Ochel]

Not all nominal values (same for other attributes, except start values) get not evaluated if they depend on parameter expressions.

mos-script

loadString("
model m1
  parameter Real a=100;
  constant Real b=100;
  Real x(nominal=10*a);
  Real y(nominal=10*b);
equation
  der(x) = 0;
  der(y) = 0;
end m1;
"); getErrorString();

simulate(m1, simflags="-lv=LOG_DEBUG"); getErrorString();

It works for y but not for x:

LOG_DEBUG         | info    | read xml file for states
|                 | |       | | Real y({start=0}, fixed=false, nominal=1000, min=-1.79769e+308, max=1.79769e+308)
|                 | |       | | Real x({start=0}, fixed=false, {nominal=1}, min=-1.79769e+308, max=1.79769e+308)
LOG_DEBUG         | info    | read xml file for stateDerivatives
|                 | |       | | Real der(y)({start=0}, fixed=false, nominal=1000, min=-1.79769e+308, max=1.79769e+308)
|                 | |       | | Real der(x)({start=0}, fixed=false, {nominal=1}, min=-1.79769e+308, max=1.79769e+308)
[...]

[Adrian Pop]

Hi,

I think we should talk about this in the next OSMC dev. meeting as is not 100% clear
where these should be evaluated. We can evaluate them in the Front-End but you can
also evaluate them in the Back-End (similar to what EvaluateParameter does).

Cheers,
Adrian Pop/

[Adrian Pop]

Flattening this model:

model m1
  parameter Real a=100;
  constant Real b=100;
  Real x(start=10*a, nominal=10*a);
  Real y(nominal=10*b);
equation
  der(x) = 0;
  der(y) = 0;
end m1;

gets you:

adrpo@ida-liu050 ~/dev/OpenModelica/build/bin
$ ./omc m1.mo 
class m1
  parameter Real a = 100.0;
  constant Real b = 100.0;
  Real y(nominal = 1000.0);
  Real x(start = 10.0 * a, nominal = 10.0 * a);
equation
  der(x) = 0.0;
  der(y) = 0.0;
end m1;

so start values are not evaluated in the Front-End either.
They might be evaluated somewhere in the Back-End I guess
if they reach the code generation as values.

[Martin Sjölund]

They should be evaluated in the backend for sure, I think. But this is probably dependent on if they are structural parameters or not.

[Lennart Ochel]

Well, we calculate start values that depend on parameter expressions during simulation. This has the advantage that we can run a simulation without recompilation with several parameter sets.
I guess we should do it the same way for the other attributes, like min, max and nominal.

The following example shows that start values are supported in a proper way:

loadString("
model bug_2207
  parameter Real a=100;
  constant Real b=100;
  Real x(start=9*a, nominal=10*a);
  Real y(start=9*b, nominal=10*b);
equation
  der(x) = 0;
  der(y) = 0;
end bug_2207;
"); getErrorString();

simulate(bug_2207, simflags="-lv=LOG_DEBUG,LOG_INIT"); getErrorString();

Simulation output

LOG_DEBUG         | info    | read xml file for states
|                 | |       | | Real y(start=900, fixed=false, nominal=1000, min=-1.79769e+308, max=1.79769e+308)
|                 | |       | | Real x({start=0}, fixed=false, {nominal=1}, min=-1.79769e+308, max=1.79769e+308)
LOG_DEBUG         | info    | read xml file for stateDerivatives
|                 | |       | | Real der(y)({start=0}, fixed=false, nominal=1000, min=-1.79769e+308, max=1.79769e+308)
|                 | |       | | Real der(x)({start=0}, fixed=false, {nominal=1}, min=-1.79769e+308, max=1.79769e+308)
[…]
LOG_INIT          | info    | updating start-values
|                 | |       | | x(start=900)
[…]
|                 | |       | | states variables
|                 | |       | | | [1] Real y(start=900, nominal=1000) = 900 (pre: 900)
|                 | |       | | | [2] Real x(start=900, nominal=1) = 900 (pre: 900)
[…]

[Lennart Ochel]

I think it's not a question of how to do it. The only question is who has the time to do it. The implementation should be quite easy and straight forward.
I guess Adrian is right and we can discuss it in the next OMDev meeting.

[Martin Sjölund]

Postponed until 1.9.1

[Francesco Casella]

I understand this problem also affects the experimental optimal control problem solver Vitalij is working on; if one sets min and max attributes on variables, these should be considered as hard constraints, so they are an essential part of the problem and cannot be ignored at all. Most of the time they will be parameter depending, so this needs to be taken care of.

[Lennart Ochel]

I fixed it and will commit it soon. Before, I would like to clean up my changes a bit.

[Lennart Ochel]

fixed in r19076