A Measurement of Primordial Non-Gaussianity Using WMAP 5-Year Temperature Skewness Power Spectrum

TL;DR

This paper measures primordial non-Gaussianity in the CMB using WMAP 5-year data, employing a novel skewness power spectrum estimator that avoids collapsing information into a single number, resulting in constraints consistent with Gaussianity.

Contribution

It introduces a new estimator based on the skewness power spectrum to analyze primordial non-Gaussianity without losing information, improving upon previous methods.

Findings

Constraint on f_{NL} is 11.0 ± 23.7, consistent with zero.

No significant evidence for primordial non-Gaussianity was found.

Method effectively separates primordial signals from point sources and lensing effects.

Abstract

We constrain the primordial non-Gaussianity parameter of the local model f_{NL} using the skewness power spectrum associated with the two-to-one cumulant correlator of cosmic microwave background temperature anisotropies. This bispectrum-related power spectrum was constructed after weighting the temperature map with the appropriate window functions to form an estimator that probes the multipolar dependence of the underlying bispectrum associated with the primordial non-Gaussianity. We also estimate a separate skewness power spectrum sensitive more strongly to unresolved point sources. When compared to previous attempts at measuring the primordial non-Gaussianity with WMAP data, our estimators have the main advantage that we do not collapse information to a single number. When model fitting the two-to-one skewness power spectrum we make use of bispectra generated by the primordial non-Gaussianity, radio point sources, and lensing-secondary correlation. We analyze Q, V and W-band WMAP 5-year data using the KQ75 mask out to l_{max}=600. Using V and W-band data and marginalizing over model parameters related to point sources and lensing-secondary bispectrum, our overall and preferred constraint on f_{NL} is 11.0\pm23.7 at the 68% confidence level (-36.4 < f_{NL} < 58.4 at 95% confidence). We find no evidence for a non-zero value of f_{NL} even marginally at the 1\sigma level.