Measuring the three-point correlation function of the cosmic microwave background

We present a new method to estimate three-point correlations in cosmic microwave background maps. Our fast Fourier transform-based implementation estimates three-point functions using all possible configurations (triangles) at a controlled resolution. The speed of the technique depends on both the resolution and the total number of pixels N. The resulting N log N scaling is substantially faster than naive methods with prohibitive N-3 scaling. As an initial application, we measure three-point correlation functions in the first-year data release of the Wilkinson Microwave Anisotropy Probe (WMAP). We estimate 336 cross-correlations of any triplet of maps from the eight differential assemblies, scanning altogether 2.6 million triangular configurations. Using Gaussian signal plus realistic noise simulations, we perform a null hypothesis testing with regards to the Gaussianity of the cosmic microwave background. Our main result is that at the three-point level, WMAP is fully consistent with Gaussianity. To quantify the level of possible deviations, we introduce false discovery rate analysis, a novel statistical technique to analyze three-point measurements. This confirms that the data are consistent with Gaussianity at better than the 1 sigma level when jointly considering all configurations. We constrain a specific non-Gaussian model using a quadratic expansion of the temperature field in terms of the f(NLT) parameter, Delta T/T (Delta T/T)(L) + f(NLT)[(Delta T/T)(L)(2) - <(Delta T/T)(L)(2)>], for which we construct an estimator from the three-point function. We find that using the skewness alone is more constraining than a heuristic suboptimal combination of all our results; our best estimate is f(NLT) = - 110 +/- 150, assuming a Lambda CDM concordance model.