The structure of correlation functions in single field inflation
Sarah Shandera
TL;DR
The paper investigates when single-field inflation with higher-derivative interactions yields perturbative, Gaussian-like fluctuations with a well-defined hierarchy of correlation functions. Using a general $P(X,\phi)$ framework and an EFT viewpoint, it shows that a perturbative regime characterized by $X/M^4\lesssim1$ and a small sound speed $c_s$ leads to a scalable, hierarchical structure of $n$-point functions, with higher-order cumulants suppressed by powers of $\mathcal{P}_\zeta^{1/2}/c_s^2$. The analysis extends to six-derivative terms, establishing that unless perturbativity fails, the dominant contributions follow from a finite set of operators, and that the DBI action provides a concrete, symmetry-protected example of this behavior. The findings clarify how derivative-driven non-Gaussianity can be large yet still amenable to perturbative, Gaussian-based expansions, and they outline the conditions under which this hierarchical picture holds in realistic models. This has implications for interpreting higher-order statistics in CMB and LSS data and for connecting observed non-Gaussianity to the UV structure of inflation.
Abstract
Many statistics available to constrain non-Gaussianity from inflation are simplest to use under the assumption that the curvature correlation functions are hierarchical. That is, if the n-point function is proportional to the (n-1) power of the two-point function amplitude and the fluctuations are small, the probability distribution can be approximated by expanding around a Gaussian in moments. However, single-field inflation with higher derivative interactions has a second small number, the sound speed, that appears in the problem when non-Gaussianity is significant and changes the scaling of correlation functions. Here we examine the structure of correlation functions in the most general single scalar field action with higher derivatives, formalizing the conditions under which the fluctuations can be expanded around a Gaussian distribution. We comment about the special case of the Dirac-Born-Infeld action.
