Algebraic loop, stiff model > Matlab numerical problems with H-infinity controller design...

Hello group,

I do have some problems with MATLAB algorithms for controller
synthesis (H-infinity with Ricatti/LMI solver). I'm not sure what is
the reason for it, so I'm trying to figure it out. There are two
things that I could think of: the model is to stiff, so the algorithm
have numerical problems, or I do have a problem with a algebraic loop.

The nonlinear Simulink model contained a algebraic loop, so a PT1
function with a fast pole (tf([1000],[1 1000])) was added to avoid the
algebraic loop and get the model work. There also exist linear state
space models of the subsystems. When I connect the state space models
of the subsystems in Simulink (this was first done for better
validation in Simulink, later with append/connect), I get a warning
that there exist a algebraic loop, but the simulation works and the
results (outputs) are realistic.

But when I append+connect the various state spaces to one big state
space model and simulate it, there aren't any warnings. What happens
to the algebraic loop? Does it still exist within the state space? How
does Matlab handle theses algebraic calculations or are they in any
way ignored?

The other problem might be the stiffness of my model. I have a stable
system, but the left poles are at about -10^3 and the right poles are
near the imaginary axis at about -10^-4.

Could it be the algebraic loop that makes the model stiffer?

All algorithms that I use have problems my model. It is of order >40.
If I use the Ricatti algorithms with a reduced order model (balanced
realization) it works, but the results aren't very good, when I
simulate the controller in the original full order model. So I need to
make the controller design with the full order model in order to
account all dynamics.

What do you think? Any suggestions or hints? If you need further
information, please tell me.
Greetings, Clemens