This posting has nothing to do with R (except maybe that I am using R very heavily in writing the paper to which the question pertains.) I simply wish to draw upon the impressive knowledge and wisdom of the R community. Since this question is way off topic, if anybody has the urge to reply, they should probably email me directly: rolf at math.unb.ca rather than via this list. My question is essentially about Fourier coefficients: Suppose pi / 2*pi*a_k = | f(omega)*exp(-i*k*omega) d omega / -pi and pi / 2*pi*b_k = | G(omega)*f(omega)*exp(-i*k*omega) d omega / -pi (The ``*''-s just mean multiplication here, not convolution; i is of course sqrt(-1).) The function f() is positive and symmetric about 0 (it's actually a spectral density function) and G() is the gain of a nice (ARMA) filter | p(exp(i*omega) |^2 G(omega) = | -------------- | | q(exp(i*omega) | where p() and q() are polynomials (with real coefficients); q() has no zeroes inside the unit disk. Suppose that the a_k satisfy an asymptotic condition: a_k * ln k ---> 0 as k ---> infinity. (The ``Berman condition''.) Can I say that the b_k satisfy this condition? If not, where would I look for a counter-example? And could I add some extra not-too-stringent restrictions on the spectrum f() so that I ***could*** say that the b_k satisfy the Berman condition? Any hints gratefully received. cheers, Rolf Turner rolf at math.unb.ca