Inflationary correlation functions are potentially affected by infrared divergences. For example, the two-point correlator of curvature perturbation at momentum k receives corrections ~ln(kL), where L is the size of the region in which the measurement is performed. We define infrared-safe correlation functions which have no sensitivity to the size L of the box used for the observation. The conventional correlators with their familiar log-enhanced corrections (both from scalar and tensor long-wavelength modes) are easily recovered from our IR-safe correlation functions. Among other examples, we illustrate this by calculating the corrections to the non-Gaussianity parameter f_NL coming from long-wavelength tensor modes. In our approach, the IR corrections automatically emerge in a resummed, all-orders form. For the scalar corrections, the resulting all-orders expression can be evaluated explicitly.

We investigate the signature of primordial non-Gaussianities in the weak lensing bispectrum, in particular the signals generated by local, orthogonal and equilateral non-Gaussianities. The questions we address include the signal-to-noise ratio generated in the Euclid weak lensing survey (we find the 1sigma-errors for fNL are 200, 575 and 1628 for local, orthogonal and equilateral non-Gaussianities, respectively), misestimations of fNL if one chooses the wrong non-Gaussianity model (misestimations by up to a factor of +/-3 in fNL are possible, depending on the choice of the model), the probability of noticing such a mistake (improbably large values for the chi^2-functional occur from fNL 200 on), degeneracies of the primordial bispectrum with other cosmological parameters (only the matter density Omega_m plays a significant role), and the subtraction of the much larger, structure-formation generated bispectrum. If a prior on a standard wCDM-parameter set is available from Euclid and Planck, the structure formation bispectrum can be predicted accurately enough for subtraction, and any residual structure formation bispectrum would influence the estimation of fNL to a minor degree. Configuration-space integrations which appear in the evaluation of chi^2-functionals and related quantities can be carried out very efficiently with Monte-Carlo techniques, which reduce the complexity by a factor of O(10^4) while delivering sub-percent accuracies. Weak lensing probes smaller scales than the CMB and hence provide an additional constraint on non-Gaussianities, even though they are not as sensitive to primordial non-Gaussianities as the CMB.

We calculate the scale dependence of the bispectrum and trispectrum in (quasi) local models of non-Gaussian primordial density perturbations, and characterize this scale dependence in terms of new observable parameters. They can help to discriminate between models of inflation, since they are sensitive to properties of the inflationary physics that are not probed by the standard observables. We find consistency relations between these parameters in certain classes of models. We apply our results to a scenario of modulated reheating, showing that the scale dependence of non-Gaussianity can be significant. We also discuss the scale dependence of the bispectrum and trispectrum, in cases where one varies the shape as well as the overall scale of the figure under consideration. We conclude providing a formulation of the curvature perturbation in real space, which generalises the standard local form by dropping the assumption that f_NL and g_NL are constants.

We describe a simple way of incorporating fluctuations of the Hubble scale during the horizon exit of scalar perturbations into the delta N formalism. The dominant effect comes from the dependence of the Hubble scale on low-frequency modes of the inflaton. This modifies the coefficient of the log-enhanced term appearing in the curvature spectrum at second order in field fluctuations. With this modification, the relevant coefficient turns out to be proportional to the second derivative of the tree-level spectrum with respect to the inflaton phi at horizon exit. A logarithm with precisely the same coefficient appears in a calculation of the log-enhancement of the curvature spectrum based purely on the geometry of the reheating surface. We take this agreement as strong support for the proposed implementation of the delta N formalism. Moreover, our analysis makes it apparent that the log-enhancement of the inflationary power-spectrum is indeed physical if this quantity is defined using a global coordinate system on the reheating surface (or any other post-inflationary surface of constant energy density). However, it can be avoided by defining the spectrum using invariant distances on this surface.