
2
A. Linking the Frenet-Serret equations with the equations of motion 31
B. Killing vectors of the hyperbolic space in two dimensions 32
C. Coordinate transformations and isometries 33
D. Eigenvalues of block matrices 34
E. Numerical solution for powerspectra 35
F. Initial condition search strategy 35
References 35
1. INTRODUCTION
The theory of cosmological inflation, originally introduced as a scenario that evades the fine-tunings of the classical
FLRW universe [1–7], is currently the leading paradigm for the origin of structure formation [8]. In its simplest form,
the evolution of a scalar field over flat regions of its potential causes the accelerated expansion of the universe. One
might think that inflation is also subject to fine-tuning problems of its own, because the onset of the accelerating
expansion is not guaranteed for all initial conditions. This turns out not to be the case due to the dissipative nature
of the cosmological equations in an expanding background which yields a notion of initial conditions independence
for models with a single scalar field. This fact makes single-field models of inflation both self-consistent and phe-
nomenologically successful in matching the CMB observations to a great accuracy. On a theoretical level this success
is eclipsed by obstacles in embedding the inflationary paradigm into high-energy physics.
More specifically, high-energy theories, such as string theory or supergravity, in general require consideration of the
dynamics of multiple scalar fields in the early universe, and through the swampland program, potentially limit their
allowed interactions (e.g. [9–17]). For a multi-field model to be consistent with the general philosophy of inflation,
observables should have a weak dependence on initial field configurations; this can only be achieved if any additional
degrees of freedom quickly become non-dynamical leaving just one field to drive the evolution.1The truncation to one
field is tricky and often times the dynamics of the effective description is quite distinct from what one obtains from
purely single-field models at the level of both the background and the perturbations (see e.g. [20–32]). In certain cases,
even though the background dynamics can be reduced to the dynamics of a single-field, the behaviour of perturbations
can be drastically different; the effect of isocurvature perturbations can be absorbed into the definition of a nontrivial
speed of sound in the evolution of the curvature perturbation, leading to deviations from the single-field predictions
(see for instance [33–39]).
Focusing on the background, it becomes apparent that consistent multi-field models should display a strong attractor
behaviour, akin to single-field models, that weakens their initial conditions dependence. The extra fields should remain
non-dynamical during the relevant evolution (at least 50-60 e-folds before the end of inflation) and, thus, the system
follows a specific trajectory in the field space that we call the attractor solution. In general, the attractor solution is
not equivalent to the solution of a single-field model because of the existence of turns in the trajectory; this happens
when the extra fields remain non-dynamical but are excited away from their respective minima of the potential. This
generic multi-field dynamics at the background level has been explored recently in Ref. [40, 41], showing the existence
of rapid-turn solutions based on a generic calculation of the turn rate, and in Ref. [42] where analogous expressions for
the late-time solution were provided using an effective potential that dictates dynamics at late-times. Moreover, this
class of models were shown to lie at the intersection of de-Sitter and scaling solutions where every “slow-roll”-type
approximation becomes exact [43]. It remains unclear whether two-field studies capture all essential features of the
multi- or many-field evolution, especially for concrete models derived from supergravity. Many studies of N>2
inflation exist in the literature, including [26, 44–55]. Many of these works have focused on specific types of slow-turn
inflationary models, the universality of the many-field limit, or the effective field theory of the curvature perturbation
for N>2 models, overall leaving generic multi-field background trajectories unexplored.
1For the simplest multi-field models without any hierarchies between the model’s parameters (such as the masses of the fields or parameters
that quantify the strength of the field space curvature) the non-uniqueness of observables is quite evident at the two-field level, (see
e.g. [18]), and, moreover, it persists even in the many-field limit [19].