An Estimate of the Primordial Non-Gaussianity Parameter f_NL Using the Needlet Bispectrum from WMAP

We use the full bispectrum of spherical needlets applied to the WMAP data of the cosmic microwave background as an estimator for the primordial non-Gaussianity parameter f_NL. We use needlet scales up to l_max=1000 and the KQ75 galactic cut and find f_NL=84 +/- 40 corrected for point source bias. We also introduce a set of consistency tests to validate our results against the possible influence of foreground residuals or systematic errors. In particular, fluctuations in the value of f_NL obtained from different frequency channels, different masks and different multipoles are tested against simulated maps. All variations in f_NL estimates are found statistically consistent with simulations.

💡 Research Summary

The paper presents a novel application of the full bispectrum of spherical needlets as an estimator for the primordial non‑Gaussianity parameter (f_{\mathrm{NL}}) using the Wilkinson Microwave Anisotropy Probe (WMAP) data. Needlets are a class of spherical wavelets that provide simultaneous localization in both harmonic (ℓ) space and real space, making them particularly well‑suited for analyses that must contend with Galactic masks and point‑source contamination. The authors first decompose the three WMAP frequency maps (Q, V, and W bands) into needlet coefficients up to a maximum multipole of (\ell_{\max}=1000). They then construct the needlet bispectrum, which is the three‑point correlation of needlet coefficients across all scale combinations ((j_1,j_2,j_3)).

To translate the measured bispectrum into an estimate of (f_{\mathrm{NL}}), the authors generate a large ensemble of Gaussian Monte‑Carlo simulations (5 000 realizations) to compute the covariance matrix of the needlet bispectrum and to verify that the Gaussian expectation value is zero. They also produce non‑Gaussian simulations with known (f_{\mathrm{NL}}) values to obtain the theoretical response (\langle B\rangle_{\mathrm{NG}}(f_{\mathrm{NL}})). Using a standard maximum‑likelihood estimator, \