Alexandr Sadovnikov <leos...@mail.ru> wrote: > I tried to download the article which Dr. Paul Kinsler refer > to from the website [...]
I suggest contacting the author directly -- Amiranashvili's email is visible at http://ito.wias-berlin.de/contact/staff/index.jsp?lang=1 In the other part of the thread, SGJ said: > It turns out that the types of convolutions that are most efficiently > implemented in the time domain correspond to a frequency dependence > that is a ratio of polynomials in the frequency (a "rational > function"), so the trick is to fit things to this form. I would add the caveat that (in the frequency domain) the polynomial in the denominator should be of higher order than that in the numerator, and (that unless you want gain) keep the poles below the real axis. If this is done, your response function will then satisfy the Kramers Kronig relations and remain properly causal. Of course the simulation (being time domain) must always be causal, but if the response function is mathematically incompatible with KK then I'd expect unwanted side effects (mostly nonphysical behaviour and/or subtle numerical errors, I suppose). -- ---------------------------------+--------------------------------- Dr. Paul Kinsler Blackett Laboratory (Photonics) (ph) +44-20-759-47734 (fax) 47714 Imperial College London, dr.paul.kins...@physics.org SW7 2AZ, United Kingdom. http://www.qols.ph.ic.ac.uk/~kinsle/ _______________________________________________ meep-discuss mailing list meep-discuss@ab-initio.mit.edu http://ab-initio.mit.edu/cgi-bin/mailman/listinfo/meep-discuss
