2008/7/29 Felix Richter email@example.com:
I learned a few things in the meantime:
In my installation, NumPy uses fftpack_lite while SciPy uses FFTW3. There are more test cases in SciPy which all pass. So I am confirmed my problem is a pure usage problem. One thing I was confused about is the fact that even if I calculate the function over a certain interval, I cannot tell FFT which interval this is, it will instead assume [0...n]. So actually I did not transform a Lorentz function centered at zero but rather centered at 500. Unfortunately, this solves only half of my problem, because I still cannot reproduce the exact FT. I'll ask for that on the SciPy list, this now seems more appropriate.
Do your answers differ from the theory by a constant factor, or are they completely unrelated?