Permanent magnet machine modelling!

I am simulating a permanent magnet generator, in 2D Comsol, 'emqa', 'ALE' mode. But there is a problem concerning magnetic circuit saturation as the followings:

- My programs are convergent with linear permeability, but converge in some cases
- In static mode, there are no problems; my program is convergent with any types of BH curves (using analytical function of BH curve or piecewise line or table).
- While generator is no load, my program is not convergent with BH curve using table interpolation; even BH curve is made so smoothly and corrects the natural of physical.
But my program converges with a few analytical functions, not all; if analytical function of BH is identical to measured BH curve, my program does not work.
- While generator operates with load, having electric current in winding, my program is not convergent. Even I operated my program in right way: first run with linear permeability, and then...However, if the remanent flux density of magnet is reduced to certain value, e.g (Br=0.8T), my program is convergent. It is so strange!

I also discussed with my colleagues using other FEM soft wares; it seems like having the same problem. Are there any limitations of FEM-softwares?

I would like to receive suggestion from Comsol experts.

Thanks in advance.