IE3D simulation with more than 12000 elements

[mami_hacky]

Hi
My friend is simulating an antenna as his thesis.
Some times he wants to do his analysis with higher precision so he increases the total number of elements for simulation.
When the total number of elements is lower than 10000 every thing is ok and there is no problem however if he increase the total number of elements to above for example 14000 the simulation slows down and the simulation results get wrong.
He belives that the source of this problem is memory. When increasing the total number of elementes to above 12000 it seems that the total needed memory by IE3D increases up to 2GBytes. And it seems that zeland package can not manage this condition any more.
we are using a 32 bits version of zeland package.
we have added as much ram modules as possible to the computer. (4GBytes)
we have tested this on different version of windows, XP and 2003 server.

now my questions:
- what do you think about the source of this problem. is it true that the source of the problem can be memory?
- do we have to switch to 64bits version?
- Is it possible to solve the problem with 32 bits version?

Thanks.

 

[adel_48]

Dear mami_hacky,

I think there might be another source of problem, which is the following: in moment method there is no guarantee on convergence or accuracy as the number of mesh elements is increased. Also another sources of inaccuracies might appear such as resolution of matrix inversion, accuracy of the current elements, convergence of greens' functions series etc.

So, I think that 64 bit version should overcome the above limitations. Also, I do not think it is a problem of limited memory.

Best Regards,
Adel_48

 

[jian]

Hi:

Yes. The problem should be in the 2 GB RAM limitation on 32-bit OS. However, I have not seen any problem in incorrect results due to the RAM is approaching the 2 GB limit. IE3D is using double precision for the matrix. It is good enough for ten's of thousand unknowns for antenna problems. All MOM matrices have a singularity at DC. In fact, all full-wave EM simulation algorithms have such a singularity. For some low frequency applicaitons, all of them will diverge at some point when the frequency is low enough. However, such situation should not happen in antennas.

For us, we have solved some problems with 19000 unknknowns requiring 2.6 GB on a Win64 machine and the results are good. Some user solved some problem using 8 GB RAM and they got good results. Certainly, we also solve some antenna array problem with 65K unknowns using iterative matrix solvers. Iterative matrix solvers are supposed to be able to yield high accuracy results for very big strucutre as long as the iterative procedure converges.

If you want to verify the accuracy with increased meshing density on IE3D, you have to use bull-force and use the standard SMSi solver or use the GEMS or AIMS iterative matrix solvers. You should try to go for 64-bit with more RAM. If you want to solve very big structuers with possible reduction in accuracy, there are other ways. Regards.