Sharp features in the primordial power spectrum are a powerful window into the inflationary epoch. To date, the cosmic microwave background (CMB) has offered the most sensitive avenue to search for these signatures. In this paper, we demonstrate the power of large-scale structure observations to surpass the CMB as a probe of primordial features. We show that the signatures in galaxy surveys can be separated from the broadband power spectrum and are as robust to the nonlinear evolution of matter as the standard baryon acoustic oscillations. As a result, analyses can exploit a significant range of scales beyond the linear regime available in the datasets. We develop a feature search for large-scale structure, apply it to BOSS DR12 data and find new bounds on oscillatory features that exceed the sensitivity of Planck for a significant range of frequencies. Moreover, we forecast that the next generation of galaxy surveys, such as DESI and Euclid, will be able to improve current constraints by up to an order of magnitude over an expanded frequency range. References 517 This result has also been independently derived in [43] using ω log 1 as an expansion parameter. 8 These findings are also confirmed by analyses of N -body simulations performed in [44].

Primordial black holes as dark matter may be generated in single-field models of inflation thanks to the enhancement at small scales of the comoving curvature perturbation. This mechanism requires leaving the slow-roll phase to enter a non-attractor phase during which the inflaton travels across a plateau and its velocity drops down exponentially. We argue that quantum diffusion has a significant impact on the primordial black hole mass fraction making the classical standard prediction not trustable.

We show that a large contribution to tensor modes during inflation can be generated by a spectator scalar field with a speed of sound lower than unity.

We measure the large-scale bias of dark matter halos in simulations with non-Gaussian initial conditions of the local type, and compare this bias to the response of the mass function to a change in the primordial amplitude of fluctuations. The two are found to be consistent, as expected from physical arguments, for three halo-finder algorithms which use different Spherical Overdensity (SO) and Friends-of-Friends (FoF) methods. On the other hand, we find that the commonly used prediction for universal mass functions, that the scale-dependent bias is proportional to the first-order Gaussian Lagrangian bias, does not yield a good agreement with the measurements. For all halo finders, high-mass halos show a non-Gaussian bias suppressed by 10-15% relative to the universal mass function prediction. For SO halos, this deviation changes sign at low masses, where the non-Gaussian bias becomes larger than the universal prediction.