One Transition, Three Signatures: Tying Tilt, Cutoff and Ringing Together in Pre-Inflationary Spectra

The cosmic microwave background, the leftover glow of the early universe, shows less variation on the very largest angular scales than the standard model predicts. One popular explanation is that inflation did not start smoothly. Inflation is the brief burst of accelerated expansion thought to have seeded all cosmic structure. In this picture the universe began in a "kinetically dominated" state, where the inflaton field (the field that drives inflation) was moving fast rather than sitting on its potential. As the universe braked into slow inflation, it cut off power on the largest scales and left a ringing pattern just above the cutoff. The current analytic descriptions of this idea treat the braking as instantaneous. They also insert the overall slope of the spectrum, the "spectral tilt" $n_s \approx 0.965$, by hand. Cutoff, ringing and tilt are therefore three separate knobs, even though one physical event should produce all three.

This note proposes to replace the instantaneous step with a finite-duration transition. In that picture a single relaxation timescale sets three things together: how sharply the large-scale power is cut, how fast the ringing harmonics die off, and how the local tilt drifts back to its asymptotic value between the cutoff and the scales where the tilt is measured. The result is a template with fewer free parameters and an internal consistency relation that data can confirm or reject. That is a sharper test than adding yet another flexible feature to an already marginal low-multipole anomaly.

Where the gap is

The kinetic-dominance story is well developed. Case for kinetically dominated initial conditions for inflation argues that for plateau and hilltop potentials, a fast-moving start is generic rather than fine-tuned. Kinetic initial conditions for inflation gives universal analytic background solutions in that regime. Constraining the kinetically dominated universe fit Starobinsky inflation with kinetic-dominance initial conditions to Planck 2015 data. It found Bayesian evidence comparable to standard slow roll, despite the extra parameter $N_\dagger$, whose posterior peaks near 6 e-folds. Finite inflation in curved space pushed the cutoff into the observable window using spatial curvature, reporting $\Omega_{K,0} = -0.051 \pm 0.017$ for the curvature-extended standard model.

These works motivate the proposal. They establish that the cutoff and ringing come from the transition. They leave the tilt to the downstream slow-roll potential.

The work that supplies our starting method is Analytical approximations for curved primordial power spectra, together with the junction-matching approach of Analytic approximations for the primordial power spectrum with Israel junction conditions. These papers derive closed-form spectra by matching solutions across an instantaneous switch at conformal time $\eta_t$. Their stated limitation is that the switch is instantaneous and the tilt "requires manual inclusion", with smoothing and higher-order derivation of $n_s$ left to future work. That is exactly what we take up.

Two empirical results constrain how much we can hope to see. Bayesian inflationary reconstructions from Planck 2018 data finds a dip at $\ell \sim 20$–$30$ and a bump at $\ell \sim 40$–$50$. However, the spline evidence peaks at only three knots, and the wiggles do not survive the look-elsewhere penalty, which corrects for having searched many scales for a feature. The same paper shows a featureless power law is adequate across $50 \lesssim \ell \lesssim 2000$. Any template that adds unconstrained wiggles will be penalised. A template that ties the wiggles to quantities the data already measure has a better chance.

A note on the third direction. The research file's "spectral tilt modification" literature is largely acoustic: speech-intelligibility methods that flatten the spectral slope under a fixed power budget. This is a name collision, not a shared mechanism, and we do not pretend otherwise. We take the cosmological tilt as the relevant object. One acoustic result is a useful cautionary analogue. Mel cepstral coefficient modification based on the glimpse proportion measure found that modifying only the two lowest cepstral coefficients, which control broad tilt, beat modifying all 59. That matches the Planck reconstructions, where low-dimensional control of the broad shape beats many free features. It supports our choice of a template with very few parameters, and nothing more.

Core Claim

Model the transition from kinetic dominance to slow roll as lasting a finite number of e-folds $\Delta N$ rather than happening at an instant. Let $\epsilon(N)$ be the slow-roll parameter, which measures how fast the expansion rate is changing, as a function of e-folds $N$. During the approach to the slow-roll attractor, the excess kinetic energy decays geometrically. On the ultra-slow-roll-like approach the velocity excess falls roughly as $e^{-3N}$, which is the Hubble-friction decay of a field whose potential gradient is negligible. Since a mode of wavenumber $k$ exits the horizon roughly $\ln(k/k_t)$ e-folds after the transition scale $k_t$, the local tilt should relax to its asymptotic value as

$$ n_{\rm eff}(k) - n_s \;\approx\; \beta \left(\frac{k}{k_t}\right)^{-p}, \qquad p \approx 3, $$

where $n_{\rm eff}(k) = 1 + d\ln\mathcal{P}/d\ln k$ is the local tilt and $\beta$ is a drift amplitude.

The same finite duration acts as a smoothed step. Mode frequencies large compared with the inverse duration evolve almost adiabatically, so the $j$-th ringing harmonic should be damped roughly as $\exp(-c\, j\, k\,\Delta\eta)$, with $\Delta\eta$ the conformal duration of the transition and $c$ a number of order one fixed by the transition shape. The claim is that $\beta$, the damping and the cutoff position are all functions of $(k_t, \Delta N)$ and the potential's slow-roll parameters, not independent knobs.

The new deliverable is therefore a closed-form template $\mathcal{P}(k; A_s, n_s, k_t, \Delta N, K)$ that predicts cutoff, ringing and tilt drift together, along with the consistency relation linking them. The exponents above are expectations to be checked against exact numerics, not assumptions to be fitted.

Research Program

The analytic work extends the logolinear expansions of Logolinear series expansions with applications to primordial cosmology through a smooth transition region. We do not glue two phases with a single junction. We solve the Mukhanov–Sasaki equation, the standard wave equation for primordial curvature perturbations, perturbatively in the transition width, carrying the expansion one order further so that the time-dependence of $\epsilon$ and of the second slow-roll parameter appears in the mode functions. That higher-order term is where the tilt drift lives.

We do this for flat and closed geometries. In the closed case we keep the discrete wavenumber spectrum $k^2 = n(n+2)$ and the hard cutoff at $k = 3$ described in Primordial power spectra for curved inflating universes. Non-canonical kinetic terms, as in Primordial power spectra from k-inflation with curvature and the imaginary-wavenumber suppression of Zakhar Shumaylov et al., are a stretch extension, attempted only if the canonical case validates.

Validation uses exact numerics as ground truth: primpy for the background and perturbations from kinetic dominance onward, and oscode for the highly oscillatory modes. We compare template and exact spectra for Starobinsky and hilltop potentials across a grid of $N_\dagger$ and $\Omega_K$ values. The comparison is made in two places: in $\mathcal{P}(k)$, and after projection to $C_\ell$, the angular power spectrum of the sky, measured in units of cosmic variance. Cosmic variance is the irreducible scatter from having only one sky to observe. Projection matters because, as the curved-space papers note, sharp $k$-space ringing is heavily smoothed by the transfer to angular multipoles. The comparison that counts is therefore the one in $C_\ell$.

Inference uses PolyChord nested sampling through Cobaya. That pipeline produces both the Bayesian evidence and posteriors, so model comparison is consistent with the prior literature.

Baselines

The naive reference point is the standard power-law spectrum $A_s (k/k_*)^{n_s-1}$ in flat ΛCDM, fit to Planck 2018 with the same likelihoods. All other improvements are reported relative to it.

The strongest existing phenomenological competitor is the quantized spectrum of Improved cosmological fits with quantized primordial power spectra. It replaces the continuous wavenumber integral with a comb $k_n = k_0 + n\Delta k$. It reports $\Delta\chi^2 = -8.55$ against ΛCDM on Planck 2018 plik TTTEEE+lowl+lowE+lensing, with $k_0 = 3.225\times10^{-4}$ Mpc$^{-1}$ and $\Delta k = 2.257\times10^{-4}$ Mpc$^{-1}$. This is the number to beat in fit quality. It was obtained with Nelder–Mead optimisation, which finds a best fit but does not compute evidence. We will refit it inside the same PolyChord pipeline, with priors on $(k_0, \Delta k)$ of comparable width to ours, so that evidences are directly comparable.

The strongest physical competitor is the instantaneous-transition template of Thavanesan et al. with $n_s$ inserted by hand. Alongside it we run full numerical kinetic-dominance spectra in the style of Constraining the kinetically dominated universe and Finite inflation in curved space. For the kinetic-dominance case the file reports only that evidence is comparable to slow roll: no reported result found on Lacuna for a Planck 2018 $\Delta\chi^2$ or $\Delta\ln Z$ of the kinetic-dominance model against ΛCDM. We will produce that number ourselves.

The model-agnostic competitor is the three-knot linear spline reconstruction from Bayesian inflationary reconstructions from Planck 2018 data. It represents what the data can support without physics.

The key ablation decouples the tilt drift. It is the same template with $\beta$ and the harmonic damping freed from their predicted dependence on $(k_t, \Delta N)$. This isolates whether the consistency relation, rather than the extra flexibility, is doing the work.

All models share data, likelihood choices, nuisance treatment and sampler settings. Beating a baseline means a higher evidence, not merely a lower $\chi^2$. Against the quantized comb, success means matching its fit improvement with comparable or higher evidence while being derived rather than imposed.

Why not just…?

Why not just run primpy inside the sampler and skip the analytic template? This is a real alternative, and it is effectively what the kinetic-dominance papers did. Exact numerics will remain our ground truth. A template earns its place for two reasons. First, it makes the consistency relation explicit: one sees which combination of parameters sets the drift. Second, it is cheap enough to sample the closed-geometry and non-canonical extensions densely. If the validation step shows the template errs by more than a small fraction of cosmic variance in $C_\ell$, we drop it for inference. We would then test the consistency relation directly on exact spectra, and the scientific claim would survive.

Why not just use the three-knot spline? The spline already captures the large-scale shape that Planck supports, and the evidence peak at $N = 3$ says the data cannot afford more freedom. But a spline cannot say whether the dip, the bump and the tilt are causally linked, because each knot is independent. Our ablation tests exactly this. If the coupled template does no better than three free knots, the physical link carries no information, and we report that.

Why not just adopt the quantized comb, since it already gives $\Delta\chi^2 = -8.55$? Its own authors report that their physical derivation from the future conformal boundary gives an infrared cutoff too high to match observations, so the comb is currently a fit without a mechanism. It also says nothing about tilt. If it wins on evidence in our common pipeline, that is a meaningful result in favour of discretisation over transition physics, and we would say so.

Why not just add a generic smooth cutoff times a power law? That captures suppression but neither ringing nor drift. It serves as a lower rung under the decoupled ablation and costs nothing to include.

Proposed Experiments

The first experiment is validation. Template against primpy/oscode spectra over Starobinsky and hilltop potentials, flat and closed geometries, and a range of $N_\dagger$. Metrics are the fractional error in $\mathcal{P}(k)$ and the $C_\ell$ residual in cosmic-variance units for $\ell \le 50$. We also measure the drift exponent $p$ and the harmonic damping rate directly from the exact spectra. If $p$ is far from 3, or the damping is not controlled by $\Delta\eta$, the core claim fails at this stage, before any data are touched.

The second experiment is detectability on mock skies. We inject template spectra into simulated low-multipole temperature and E-mode polarization data at Planck-like noise, with many independent cosmic-variance realisations. We ask whether the coupled and decoupled models can be distinguished at all. This step is necessary because cosmic variance at $\ell = 2$ is about 63%, and the relation may be invisible in temperature alone. We repeat the exercise with cosmic-variance-limited E-modes to say what future polarization data would need to deliver.

The third experiment is the real-data comparison on Planck 2018 plik TTTEEE+lowl+lowE+lensing, optionally with BK15, against every baseline above. Metrics are $\Delta\chi^2$, $\Delta\ln Z$, and posteriors on $(k_t, \Delta N, \Omega_K, \beta)$.

The decisive check is the decoupled ablation. If the freed drift amplitude $\beta$ lands within its predicted band from the fitted $(k_t, \Delta N)$, the single-transition picture gains support with no added freedom. If Planck pushes $\beta$ outside that band with the cutoff fixed, finite-duration kinetic dominance is disfavoured as the common origin of the low-$\ell$ deficit and the tilt. Either outcome is informative.

We will also report how $\Omega_K$ posteriors shift when the transition width is freed. This is relevant to the curvature preference reported in Finite inflation in curved space and to the known pull toward flatness from late-time baryon acoustic oscillation measurements, the imprint of early sound waves in the galaxy distribution.