When modelling inflaton fluctuations as a free quantum scalar field, the initial vacuum is conventionally imposed at the infinite past. This is called the Bunch-Davies (BD) vacuum. If however an asymptotically Minkowskian past does not exist, this requires modifications. We derive corrections to the scalar spectral index $n_s$ and the tensor tilt $n_t$ descending from arbitrary mixed states or from explicit non-BD initial conditions. The former may stem from some pre-inflationary background and can redshift away whereas the latter are induced by a timelike hypersurface parametrising a physical cut-off. In both cases, we find that corrections scale in parts or fully as $\mathcal O(\epsilon)$ where $\epsilon$ is the first slow-roll parameter. The precise observational footprint is hence dependent on the model driving inflation. Further, we show how the inflationary consistency relation is altered. We thus provide an analytic handle on possible high scale or pre-inflationary physics.

An appealing explanation for the Planck data is provided by inflationary models with a singular non-canonical kinetic term: a Laurent expansion of the kinetic function translates into a potential with a nearly shift-symmetric plateau in canonical fields. The shift symmetry can be broken at large field values by including higher-order poles, which need to be hierarchically suppressed in order not to spoil the inflationary plateau. The herefrom resulting corrections to the inflationary dynamics and predictions are shown to be universal at lowest order and possibly to induce power loss at large angular scales. At lowest order there are no corrections from a pole of just one order higher and we argue that this phenomenon is related to the well-known extended no-scale structure arising in string theory scenarios. Finally, we outline which other corrections may arise from string loop effects.

Working in the Large Volume Scenario (LVS) of IIB Calabi-Yau flux compactifications, we construct inflationary models from recently computed higher derivative $(\alpha')^3$-corrections. Inflation is driven by a Kaehler modulus whose potential arises from the aforementioned corrections, while we use the inclusion of string loop effects just to ensure the existence of a graceful exit when necessary. The effective inflaton potential takes a Starobinsky-type form $V=V_0(1-e^{-\nu\phi})^2$, where we obtain one set-up with $\nu=-1/\sqrt{3}$ and one with $\nu=2/\sqrt{3}$ corresponding to inflation occurring for increasing or decreasing $\phi$ respectively. The inflationary observables are thus in perfect agreement with PLANCK, while the two scenarios remain observationally distinguishable via slightly varying predictions for the tensor-to-scalar ratio $r$. Both set-ups yield $r\simeq (2\ldots 7)\,\times 10^{-3}$. They hence realise inflation with moderately large fields $\left(\Delta\phi\sim 6\thinspace M_{Pl}\right)$ without saturating the Lyth bound. Control over higher corrections relies in part on tuning underlying microscopic parameters, and in part on intrinsic suppressions. The intrinsic part of control arises as a leftover from an approximate effective shift symmetry at parametrically large volume.

Inflationary attractors predict the spectral index and tensor-to-scalar ratio to take specific values that are consistent with Planck. An example is the universal attractor for models with a generalised non-minimal coupling, leading to Starobinsky inflation. In this paper we demonstrate that it also predicts a specific relation between the amplitude of the power spectrum and the number of e-folds. The length and height of the inflationary plateau are related via the non-minimal coupling: in a wide variety of examples, the observed power normalisation leads to at least 55 flat e-foldings. Prior to this phase, the inflationary predictions vary and can account for the observational indications of power loss at large angular scales.