Can we really measure fnl from the galaxy power spectrum?

TL;DR

This paper investigates how primordial non-Gaussianity parameters, especially fnl and gnl, influence galaxy bias measurements, revealing biases in current methods and advocating for model selection techniques to accurately constrain PNG parameters.

Contribution

It demonstrates that assuming only one PNG parameter leads to biased estimates and shows the importance of considering multiple parameters with Bayesian model selection.

Findings

Biases arise when multiple PNG parameters are ignored.

Opposite signs of fnl and gnl can cancel out signals, causing false null detections.

Bayesian evidence helps determine the necessity of multiple PNG parameters.

Abstract

The scale-dependent galaxy bias generated by primordial non-Gaussianity (PNG) can be used to detect and constrain deviations from standard single-field inflation. The strongest signal is expected in the local model for PNG, where the amplitude of non-Gaussianity can be expressed by a set of parameters (fnl, gnl, ...). Current observational constraints from galaxy clustering on fnl and gnl assume that the others PNG parameters are vanishing. Using two sets of cosmological N-body simulations where both fnl and gnl are non-zero, we show that this strong assumption generally leads to biased estimates and spurious redshift dependencies of the parameters. Additionally, if the signs of fnl and gnl are opposite, the amplitude of the scale-dependent bias is reduced, possibly leading to a false null detection. Finally we show that model selection techniques like the Bayesian evidence can (and should) be used to determine if more than one PNG parameter is required by the data.