Calulated net conductor joule heat not increasing with frequency as expected when using SOLID117 model

QUESTION:
A SOLID117 model (attached) was built for simulating the joule heating induced by both eddy current and electrical heat dissipation (I^2*R). The unit used in the model is umksv. It has been found that the joule heating is extremely insensitive to the applied AC frequency.
The joule heating calculated increases less than 1% when increasing the frequency from 250MHz to 400MHz. However, the measured temperature rise (over ambient) increases 50% when the frequency changes from 250MHz to 400MHz. I don't understand why the difference b/w measurement and modeling is so large. Can you please explain this behavior?
ANSWER:
Before a SOLID117 model will give accurate results for this type of analysis, the following conditions must be satisfied:
1) At release 9.0 and earlier, the conductor mesh must be hexahedral. Tetrahedral meshes will not give correct results, regardless of mesh size.
2) The current must be modeled in such a way that it is solenoidal (current cannot "appear from nowhere' and enter one end, nor can it "disappear into nowhere" and the other conductor end). This is not physically possible, and if conductor ends begin and terminate in the middle of an "air" mesh using SOLID117, calculated results will be completely wrong and there will be no warning to the user that the conductors have not been modeled correctly.
Satisfying the solenoidal condition can be accomplished in 2 ways:
a) It can be implied by symmetry. For example, a 1/4 symmetry model (90 degree sector) of a cylindrical air coil will correctly model the current as being solenoidal.
b) The conductor leads can be routed to and cut perpindicularly by a planar flux parallel model boundary. In this case, the conductor ends are modeled in the usual way: couple VOLT on one end and apply the total current
Calulated net conductor joule heat not increasing with frequency as expected when using SOLID117 model  QUESTION: A SOLID117 model (attached) was built for simulating the joule heating induced by both eddy current and electrical heat dissipation (I^2*R). The unit used in the model is umksv. It has been found that the joule heating is extremely insensitive to the applied AC frequency. The joule heating calculated increases less than 1% when increasing the frequency from 250MHz to 400MHz. However, the measured temperature rise (over ambient) increases 50% when the frequency changes from 250MHz to 400MHz. I don't understand why the difference b/w measurement and modeling is so large. Can you please explain this behavior? ANSWER: Before a SOLID117 model will give accurate results for this type of analysis, the following conditions must be satisfied: 1) At release 9.0 and earlier, the conductor mesh must be hexahedral. Tetrahedral meshes will not give correct results, regardless of mesh size. 2) The current must be modeled in such a way that it is solenoidal (current cannot "appear from nowhere` and enter one end, nor can it "disappear into nowhere" and the other conductor end). This is not physically possible, and if conductor ends begin and terminate in the middle of an "air" mesh using SOLID117, calculated results will be completely wrong and there will be no warning to the user that the conductors have not been modeled correctly. Satisfying the solenoidal condition can be accomplished in 2 ways: a) It can be implied by symmetry. For example, a 1/4 symmetry model (90 degree sector) of a cylindrical air coil will correctly model the current as being solenoidal. b) The conductor leads can be routed to and cut perpindicularly by a planar flux parallel model boundary. In this case, the conductor ends are modeled in the usual way: couple VOLT on one end and apply the total currentto any node in the coupled DOF set with the F command and set VOLT=0 on the other conductor end. Note that accuracy of results obtained with the 3D MVP (SOLID97) is not adversely affected by either condition. The conductor mesh may be tetrahedral (but keep in mind the mesh density must be sufficiently refined to spatially discretize skin depth) and solenoidal flow condition is not needed (a nonphysical "sudden appearance" and "sudden disappearance" of current at conductor ends terminated within an air mesh still yield reasonable results). The limitation to using SOLID97 is that the fields at material interfaces having widely varying magnetic permeabilities may be inaccurate. Attachments to this solution: 1) 411204c.zip: Input files for SOLID97 hex and tetmeshed spiral conductor and SOLID117 tet meshed spiral conductor. SOLID117 results differ substantially from both SOLID97 models. 2) 411204e.zip: Input files for SOLID97 hex and tet meshed spiral conductor, SOLID117 hex meshed spiral conductor, and quad meshed PLANE13 cylindrical conductor. Agreement good between SOLID97 nonsolenoidal spiral conductor (both hex and tet meshes) and PLANE13 solenoidal conductor. SOLID117 results differ substantially. 3) mdl05e.inp: input file: SOLID117 cylindrical (solenoidal) conductor w/hex mesh. This is the only SOLID117 model that agrees with all other MVP models. All other MVP model results are selfconsistent. 

