The Probability Distribution for Non-Gaussianity Estimators

TL;DR

This paper investigates the probability distribution functions of non-Gaussianity estimators in CMB research, revealing non-Gaussian features when the true non-Gaussianity parameter is non-zero and providing analytic fits for these distributions.

Contribution

It introduces a Monte Carlo method to determine the PDFs of non-Gaussianity estimators, highlighting their non-Gaussian nature and offering improved estimators with nearly Gaussian PDFs.

Findings

The null-hypothesis estimator's PDF is nearly Gaussian when fnl=0.

The PDF becomes skewed with a long tail when |fnl| > 0.

The improved estimator's PDF is nearly Gaussian for allowed fnl values.

Abstract

One of the principle efforts in cosmic microwave background (CMB) research is measurement of the parameter fnl that quantifies the departure from Gaussianity in a large class of non-minimal inflationary (and other) models. Estimators for fnl are composed of a sum of products of the temperatures in three different pixels in the CMB map. Since the number ~Npix^2 of terms in this sum exceeds the number Npix of measurements, these ~Npix^2 terms cannot be statistically independent. Therefore, the central-limit theorem does not necessarily apply, and the probability distribution function (PDF) for the fnl estimator does not necessarily approach a Gaussian distribution for N_pix >> 1. Although the variance of the estimators is known, the significance of a measurement of fnl depends on knowledge of the full shape of its PDF. Here we use Monte Carlo realizations of CMB maps to determine the PDF for two minimum-variance estimators: the standard estimator, constructed under the null hypothesis (fnl=0), and an improved estimator with a smaller variance for |fnl| > 0. While the PDF for the null-hypothesis estimator is very nearly Gaussian when the true value of fnl is zero, the PDF becomes significantly non-Gaussian when |fnl| > 0. In this case we find that the PDF for the null-hypothesis estimator fnl_hat is skewed, with a long non-Gaussian tail at fnl_hat > |fnl| and less probability at fnl_hat < |fnl| than in the Gaussian case. We provide an analytic fit to these PDFs. On the other hand, we find that the PDF for the improved estimator is nearly Gaussian for observationally allowed values of fnl. We discuss briefly the implications for trispectrum (and other higher-order correlation) estimators.