Signal-to-Noise Eigenmode Analysis of the Two-Year COBE Maps

Abstract

To test a theory of cosmic microwave background fluctuations, it is natural to expand an anisotropy map in an uncorrelated basis of linear combinations of pixel amplitudes --- statistically-independent for both the noise and the signal. These $S/N$-eigenmodes are indispensible for rapid Bayesian analyses of anisotropy experiments, applied here to the recently-released two-year COBE {\it dmr} maps and the {\it firs} map. A 2-parameter model with an overall band-power and a spectral tilt $\nu_{\Delta T}$ describes well inflation-based theories. The band-powers for {\it all} the {\it dmr} $53,90,31$ $a$+$b$ GHz and {\it firs} 170 GHz maps agree, $\{(1.1\pm 0.1)\times 10^{-5}\}^{1/2}$, and are largely independent of tilt and degree of (sharp) $S/N$-filtering. Further, after optimal $S/N$-filtering, the {\it dmr} maps reveal the same tilt-independent large scale features and correlation function. The unfiltered {\it dmr} $53$ $a$+$b$ index $\nu_{\Delta T}+1$ is $1.4\pm 0.4$; increasing the $S/N$-filtering gives a broad region at (1.0--1.2)$\pm$0.5, a jump to (1.4--1.6)$\pm$0.5, then a drop to 0.8, the higher values clearly seen to be driven by $S/N$-power spectrum data points that do not fit single-tilt models. These indices are nicely compatible with inflation values ($\sim$0.8--1.2), but not overwhelmingly so.