paper

Removing the ISW-lensing bias from the local-form primordial non-Gaussianity estimation

arXiv:1302.5799 · doi:10.1088/1475-7516/2013/04/021

Abstract

The Integrated Sachs-Wolfe (ISW) effect produces a secondary temperature anisotropy of CMB. The main contribution comes from z<2, where dark energy leads to a decay of potentials. As the same photons are gravitationally lensed by these decaying potentials, there exists a high degree of correlation between the ISW effect and CMB lensing, leading to a non-zero three-point correlation (bispectrum) of the observed temperature anisotropy. This ISW-lensing bispectrum, whose shape resembles that of the so-called "local-form" primordial bispectrum parametrized by fNL, is known to be the largest contamination of fNL. In order to avoid a spurious detection of primordial non-Gaussianity, we need to remove the ISW-lensing bias. In this work, we investigate three debiasing methods: (I) subtraction of an expected, ensemble average of the ISW-lensing bispectrum; (II) subtraction of a measured ISW-lensing bispectrum; and (III) direct subtraction of an estimated ISW signal from an observed temperature map. One may use an estimation of the ISW map from external non-CMB data or that from the CMB data themselves. As the methods II and III are based on fewer assumptions about the nature of dark energy, they are preferred over the method I. While the methods I and II yield unbiased estimates of fNL with comparable error bars, the method III yields a biased result when the underlying primordial fNL is non-zero and the ISW map is estimated from a lensing potential reconstructed from the observed temperature map. One of the sources of the bias is a lensing reconstruction noise bias which is independent of fNL and can be calculated precisely, but other fNL-dependent terms are difficult to compute reliably. We thus conclude that the method II is the best, model-independent way to remove the ISW-lensing bias of fNL, enabling us to test the physics of inflation with smaller systematic errors.

arXiv comment17 pages, comments are welcomed v2: references added, v3: references added

References in corpus (40)

Cited by in corpus (12)