mrowell
09-04-2003, 10:19 AM
Hi
Whilst this is not strictly NR related, I wondered if anyone had had any experience in numerical inversion of chracteristic functions of continious distributions using FFT -- I have read this is possible but must admit am a trifle confused. I can get it to work for non-continuous distributions by writing the MGF as a z transform and substituting exp(i.2pi.u.t) for z... I guess it boilds down to aliasing/truncation errors to do with DFTs as my CF does not go to zero as "frequency" increases...
Any pointers would be gratefully appreciated..
Thanks
Whilst this is not strictly NR related, I wondered if anyone had had any experience in numerical inversion of chracteristic functions of continious distributions using FFT -- I have read this is possible but must admit am a trifle confused. I can get it to work for non-continuous distributions by writing the MGF as a z transform and substituting exp(i.2pi.u.t) for z... I guess it boilds down to aliasing/truncation errors to do with DFTs as my CF does not go to zero as "frequency" increases...
Any pointers would be gratefully appreciated..
Thanks