Which value to take as a stopping criterion (tolerance)?
The problem is that this criterion is for local (nearfield) values. In theory, it is nothing to do with a far field. In theory, random small-scale oscillations, which radiate next to nothing, can hide smooth distributions that radiate well. In practice, I did a series of tests with model objects that showed that tolerance 1e-4 is enough – the error in directivity pattern is less that one dB. By default Puma-em uses 1e-3, which is good too.
Does anybody knows more theoretical approach to this question?
personnally I don't know any theoretical approach. I just know that since the fields are variational with respect to the currents, currents variations should not lead to outstanding fields variations, especially in the far field. Have you tried searching IEEE on this matter?
My search gave no results, but it was very brief. Computer scientists consider it as an input parameter, and they think that it is not their work to set it. They say just "small enough", or "take 10^-10 and ok".
Please, correct me, if I am wrong.
Sign up for the SourceForge newsletter:
You seem to have CSS turned off.
Please don't fill out this field.