Matthew - as you know, it's trivial to just let b=1/a and the write the distribution as (x-x)^s Exp[- (x-x0)/b] so that the gaussian limit is obtained as b->0. So infinite parameters are not a problem. Your suggestion and other observations are quite right. In fact, if convoluting distributions is ok, you can use whatever your favorites are, since cumulants just add under convolution. grant On Jul 16, 2008, at 11:08 AM, ifeffit- request@millenia.cars.aps.anl.gov wrote:
OK, here's my $0.02. I've used the convolution of an exponential tail function exp(-(r-r0)/w) (r-r0)*w >= 0 0 (r-r0)*r < 0
with a Gaussian. This avoids having to have parameters go to infinity to approach a gaussian. This function is a little unwieldy in real space but is simple in k-space. mam