A Simple Test for Non-Gaussianity in CMBR Measurements

Abstract

We propose a set of statistics $S_q$ for detecting non-gaussianity in CMBR anisotropy data sets. These statistics are both simple and, according to calculations over a space of linear combinations of three-point functions, nearly optimal at detecting certain types of non-gaussian features. We apply $S_3$ to the UCSB SP91 experiment and find that the mean of the four frequency channels is by this criterion strongly non-gaussian. Such an observation would be highly unlikely in a gaussian theory with a small coherence angle, such as standard ($n=1$, $Ω_b = .05$, $h = .5$) inflation. We cannot conclude that the non-gaussianity is cosmological in origin, but if we assume it due instead to foreground contamination or instrumental effects, and remove the points which are clearly responsible for the non-gaussian behavior, the rms of the remaining fluctuations is too small for consistency with standard inflation. Further data are needed before definitive conclusions may be drawn. Finally, we generalize the ideas behind this statistic to other types of non-gaussian behavior that might be detected in other experimental schemes.