Constraining primordial oscillations and inflationary particle production with Planck, ACT DR6, and DESI DR2

Technical Summary: Constraining Primordial Oscillations and Inflationary Particle Production with Planck, ACT DR6, and DESI DR2

Problem Statement
Standard single-field slow-roll inflation predicts a nearly scale-invariant, featureless primordial power spectrum (PPS). However, various non-standard inflationary models—such as axion monodromy, unwinding inflation, trans-Planckian effects, and inflationary particle production—predict oscillatory features in the PPS. While previous analyses using Planck and Atacama Cosmology Telescope (ACT) data have hinted at mild oscillatory residuals relative to the $\Lambda$CDM best fit, the physics driving these potential departures remains poorly constrained. This paper aims to place robust constraints on general oscillatory templates and specific inflationary particle production models using the latest datasets: Planck 2018, ACT Data Release 6 (DR6), and Dark Energy Spectroscopic Instrument (DESI) Data Release 2 (DR2).

Methodology
The authors employ a "model-agnostic" approach to constrain general oscillatory features and a specific phenomenological model for particle production.

1. Models of Interest:

   • General Oscillations: The study utilizes templates where the primordial power spectrum $P_R(k)$ includes a cosine modulation term. Three variations are tested based on the spacing of oscillations: linear ($k/k_*$), logarithmic ($\ln(k/k_*)$), and logarithmic with a running frequency ($\ln(k/k_*)(1 + \alpha_{rf} \ln(k/k_*))$).
   • Particle Production: The analysis models inflationary particle production where an inflaton field ($\phi$) couples to an iso-inflaton field ($\chi$). This interaction triggers bursts of particle production, altering the PPS via rescattering (Bremsstrahlung radiation). The model assumes ten consecutive bursts ("multiburst") to cover the full wavenumber range, parameterized by a coupling constant $g$, an amplitude $A_I$, an initial burst location $k_1$, and a spacing parameter $\Delta$.

2. Datasets:

   • CMB: Planck 2018 low-$\ell$ and high-$\ell$ (plik lite) likelihoods, combined with ACT DR6 temperature and polarization power spectra (TT, TE, EE).
   • Lensing: Combined ACT DR6 and Planck PR4 NPIPE lensing likelihoods.
   • Large Scale Structure: DESI DR2 Baryon Acoustic Oscillation (BAO) likelihood.
   • The combined dataset is referred to as "P-ACT-LB."

3. Sampling and Analysis:

   • To address the multimodal and highly degenerate nature of the posteriors for these oscillatory models, the authors integrated pocoMC (a preconditioned Sequential Monte Carlo sampler utilizing normalizing flows) into the Cobaya framework. This replaces traditional MCMC methods which often struggle with convergence in such complex parameter spaces.
   • The analysis updates previous constraints (e.g., from C25) by incorporating DESI DR2 and extending the $k$-range considered for oscillatory features.
   • Numerical instabilities in the particle burst model (specifically in the $f_2(x_i)$ term) were mitigated using a Taylor expansion for small arguments.

Key Results

1. Constraints on General Oscillations:

   • The combined P-ACT-LB dataset tightens the 95% confidence level (CL) upper bounds on oscillation amplitudes relative to the scalar amplitude $A_s$:
     • Linear spacing: $A_{lin} < 0.021$
     • Logarithmic spacing: $A_{log} < 0.022$
     • Logarithmic with running frequency: $A_{log\,rf} < 0.023$
   • These bounds restrict the additional oscillatory power to approximately $\sim 2\%$ of the scalar non-oscillatory signal, representing a marginal improvement over previous constraints ($\sim 2.9\%$).
   • The spectral index $n_s$ shows a shift toward higher values when ACT data is included, consistent with trends observed in standard $\Lambda$CDM analyses, but the oscillation models do not significantly alter the $n_s$ constraints beyond increasing error bars slightly.
   • A phase shift is observed between Planck-only and P-ACT constraints, with Planck favoring a different phase ($\phi_{log} \approx 0.02$) compared to P-ACT ($\phi_{log} \approx 0.33$).

2. Constraints on Particle Production:

   • For the inflationary particle production model, the analysis places a maximum a posteriori (MAP) constraint on the dimensionless coupling constant of $g = 0.034$ (using P-ACT-LB).
   • This constraint limits the interaction strength between the inflaton and iso-inflaton, thereby constraining the number density and mass of the iso-inflaton.
   • The preferred initial burst location $k_1$ peaks around $\log_{10}(k_1) \simeq -1$, though the full prior range is allowed.

3. Model Comparison:

   • While all oscillatory models provide a better fit to the data (lower $\chi^2$) compared to the concordance $\Lambda$CDM model, the Bayesian evidence ($\ln B$) reveals a moderate preference for $\Lambda$CDM over models with general oscillations.
   • The evidence is inconclusive regarding the particle production model.
   • The authors conclude that the added complexity of these non-standard models is not statistically justified by the current data.

Significance and Claims
The paper claims to provide the most stringent constraints to date on general oscillatory templates and inflationary particle production by leveraging the high-resolution ACT DR6 data and the large-scale structure constraints from DESI DR2. A primary technical contribution is the successful integration of the pocoMC sampler into Cobaya, demonstrating its efficacy in efficiently exploring the multimodal posteriors characteristic of these inflationary models, which traditional MCMC methods struggle to navigate.

The authors modestly conclude that while hints of oscillations exist in the residuals, the current data does not provide sufficient statistical evidence to favor non-standard inflationary dynamics over the standard $\Lambda$CDM model. The results serve to constrain the parameter space for future theoretical developments and observational campaigns, specifically limiting the amplitude of any potential primordial features to the $\sim 2\%$ level.