Some extentions and limitations of frequency domain experiments

Arnold H. Buss
DOI: https://doi.org/10.1145/318123.318254
1988-01-01
Abstract:We present three extensions of the frequency domain approach proposed by Schruben and Cogliano (1987). The first is the assignment of multiple frequencies to input factors. The frequency selection in this case is nearly identical to that of one frequency per factor. The second extension is the use of the time series of batch means to flatten the noise spectrum and make the identification of peaks in the output spectrum easier. Finally, we show that using common random numbers (as suggested in Schruben and Cogliano, 1987) for the signal and noise runs do indeed tend to decrease the peaks in the noise spectrum and also facilitate factor identification. We discuss several limitations to the use of frequency domain methodology as it currently exists. One limitation is the case in which a factor-frequency interaction is present. Although none of the simulation models currently being examined seems to have this property, it is one present in certain chaotic dynamical systems. These systems demonstrate qualitatively different behavior when oscillated. Another limitation involves the time index. For complicated systems it may not be feasible to define a single “clock” with which to oscillate parameters. Current work (Jacobson, et. al., 1988) shows that certain models may use the global time clock, but it is not clear that there is always a solution.
What problem does this paper attempt to address?