Inductance with permanent magnet

[vecerkaj@xxxxxxxxx]

Hello
First, I would like to thank for the contribution to the discussion 
about ConText and other editors.
Now, I have met a problem, how to calculate self inductance, if 
magnetic circuit contains permanent magnets (NdFeB-linear 
characteristics) and iron (nonlinear).
My thinking: Calculation of L directly from the "A.J" integral is 
probably wrong, same as calculation of L from magnetic energy 
(because permanent magnets put additional energy to the magnetic 
circuit).
My approaches how to solve that:
1. To calculate inductance from model with magnet properties set on 0 
(Hc). But the magnetic circuit will not be saturated by magnets and 
result will not be accurate. 
2.To calculate model without current and with current. To make 
subtraction of energies obtained from the models. To use the result 
for calculation of inductance. I have realized, that this way is 
wrong.
3. To calculate model without current and with current. To make 
subtraction of A in every node in mesh (mesh should be the same) out 
of FEMM and to create new .ans file. After that to calculate L 
directly from the "A.J" integral. Till now I think, that this way is 
correct, because this one considers iron saturation. I guess, if 
the "test" current is small then the method is applicable even for 
nonlinear materials.
I have also a physical model, but the result obtained from FEMM is 
higher than reality I measured on the model ( about 100%). I don't 
understand why, because FEMM doesn't consider leakage inductance in 
winding heads, so the total inductance should be higher then 
calculated value.
Note all the calculations I did were for 0 Hz (magnetostatic).
So, I would like to ask anybody, what he thinks about the problem.
Thank you in advance.

Jiri Vecerka