Dear Simon, I think your conjectures are all correct. The pseudo-gaussian spectral profile (I guess the spectrum is more precisely ~ f*exp[-(f-fcen)**2 / fwidth**2] or something like that) requires a gaussian time evolution of the source. The cutoff defines sharp edges in time when the source starts and ends so a simple rectangular time window is thus used. Thanks to the fast convergence of the gaussian function, the default 10/fwidth timespan is usually sufficient to suppress the spectral leakage into high frequencies.
At http://fzu.cz/~dominecf/misc/meep/#bandsource I present a patch to meep (sources.cpp) that uses a band-source with lower spectral leakage and well-defined spectrum of frequencies. It however requires the source to be turned on for a longer time. Hope this helps. Filip 2014-02-06 21:41 GMT+01:00, Simon Bernard <simon.bern...@mail.mcgill.ca>: > Hi, I have a question about the way Meep treats Gaussian pulses. > > Suppose you specify the parameter fwidth to be df. Then, the bounds of the > pulse in frequency space will be -df/2 and +df/2. In the time domain, the > width of the pulse is characterized by 1/fwidth. And this width is not the > actual FWHM of the pulse in time, but is proportional to it. > > The thing that confuses me is the parameter cutoff. As described in Meep > Reference, cutoff determines the time t0 at which peak of the pulse occurs: > t0 = tstart + cutoff * width. > Therefore, increasing cutoff increases the duration of the pulse in time. > Shouldn't this affect fwidth? Why is fwidth independent of cutoff? > > My attempt at answering this: > > Cutoff only controls how sharply the pulse in time is turned on/off at its > tails, and does not affect its width. > The smaller the cutoff, the sharper the pulse's edges are in time. And I am > guessing that describing these sharp edges in frequency space requires > higher frequency Fourier components. > Meep Reference states that increasing cutoff decreases the amount of high > frequencies introduced by the start/stop of the source. "Introduced" is > still unclear to me as to how it concretely affects a simulation. Maybe > these high frequency components are responsible for the wiggles you would > usually see in a power spectrum near the edges of the bandwidth (the df)? > > Thank you, > -Simon > _______________________________________________ meep-discuss mailing list meep-discuss@ab-initio.mit.edu http://ab-initio.mit.edu/cgi-bin/mailman/listinfo/meep-discuss
