Robustness of the primordial power spectrum in hybrid loop quantum cosmology to approximations near the bounce

The paper proves the robustness of the analytic approximation used in (hybrid) loop quantum cosmology to compute the primordial power spectrum of cosmological perturbations for a recently proposed vacuum state (the non-oscillatory state with asymptotic Hamiltonian diagonalization, NO-AHD). To investigate this, the authors study different approximations to the effective mass of these perturbations near the bounce, showing that the Pöschl-Teller (PT) potential employed in previous works leads to indistinguishable power spectra compared to other estimations of the mass, provided that they successfully capture the characteristic scale of the bounce, which translates into a scale of power suppression.

The paper considers three alternative approximations to the effective mass term in the quantum era: a double PT potential, a Rosen-Morse (RM) potential, and a de Sitter (dS) mass term. For all these models, the authors keep the same value for the transition time t0 as in the conventional PT case and obtain analytic mode solutions for the perturbations and the vacuum state selected by the NO-AHD criterion near the bounce.

For the double PT potential, the effective mass is approximated by a piecewise function composed of two PT potentials, covering the quantum era in two equally long intervals in cosmological time. The parameters are fixed by demanding that the approximated and exact mass coincide at the bounce and at the final point of the quantum era, and that the first PT potential matches the exact mass at the transition time t0/2. This leaves one free parameter P, chosen to minimize the relative error, which remains below 0.04 throughout the entire quantum period.

For the Rosen-Morse potential, the effective mass is given by the addition of a hyperbolic tangent term to a PT potential. The parameters are fixed by requiring that the approximated and exact mass coincide at the beginning and end of the quantum era, and that the second derivatives of both approximations coincide at the bounce. The relative error grows at most up to 0.08 for the whole quantum era. However, the characterization of the vacuum state breaks down for wavenumbers k ≤ √V1, where V1 is the parameter multiplying the tanh contribution, because the imaginary part of the complex frequency function vanishes. Nevertheless, this happens in the sector of wavenumbers where the PPS is already suppressed.

For the de Sitter case, the LQC evolution of the background in the Planck regime is replaced by a phase of exact de Sitter dynamics. The mass term is given by sdS = -2(a0H0)2 e 2H0(t-t0). This mass term provides a much worse approximation to the hybrid LQC mass than the other studied cases. In this case, the NO-AHD criterion naturally selects the Bunch-Davies state for the perturbative modes.

The PPS is computed using the parametrization PR = Ak - Bk2 × P R ΛCDM, where the preinflationary prefactor Ak - Bk2 accounts for the different choice of vacuum and the dynamics during the preinflationary era. The authors apply a Bogoliubov transformation to remove spurious power oscillations caused by the lack of smoothness in the approximations, replacing Ak → Ak and Bk → Bk, provided that the norms do not oscillate rapidly.

The main feature observed in the spectrum of the NO-AHD vacuum of all the considered cases is the power suppression in the infrared region, at wavenumber scales determined by the characteristic scale of the quantum preinflationary epoch. The specific scale at which this suppression occurs slightly depends on the particular model, but the sensitivity is certainly small. The steepness of the suppression is more clearly model dependent.

The PPS that most closely resembles that of the PT potential corresponds to the double PT approximation, with a tilt slightly less pronounced than in the PT case. The spectrum that most differs from its PT counterpart is provided by the dS case, with a considerably more inclined curve for 0.5 ≲ k ≲ 2. The RM curve is practically indistinguishable from the PT one until it reaches the surroundings of the critical scale (k = √V1), where the power is suppressed more sharply.

As an overall result, the three PPS obtained are very similar to the one computed with the PT potential. Given the current resolution in observational missions, it is highly unlikely that it will be possible to tell them apart. Therefore, the authors conclude that the predictions for the PPS of the NO-AHD vacuum in hybrid LQC are not significantly sensitive to the approximation employed for the effective mass of the scalar perturbations near the bounce.

The authors expect this conclusion to be valid beyond the restrictions adopted in the work, including for other inflaton potentials, tensor perturbations, or the dressed metric approach to LQC. Since the potential contribution to the inflaton energy density in the vicinity of the bounce is irrelevant, the suppression and slope of the PPS, which are primarily determined by the preinflationary dynamics of the quantum period in the NO-AHD prescription, will remain invariant under changes in the inflaton potential. The difference in the effective mass of scalar and tensor perturbations is practically negligible in regions in which the potential energy is, so the conclusion holds for tensor perturbations too. In the dressed-metric strategy, the effective mass for the perturbations can also be described by a PT potential in the quantum regime after the bounce, making it plausible that the results about the robustness of the adopted approximation apply to the dressed-metric approach too.

Improvements to AI Systems Based on This Paper:

1. Robustness-Aware Model Selection for Analytical Approximations

• Improvement: Implement a meta-learning module that evaluates the sensitivity of final predictions (e.g., power spectra) to variations in intermediate analytical approximations (e.g., effective mass potentials).

• Capability: The AI can automatically flag when a simplified model (e.g., single Pöschl-Teller) yields predictions within observational error bars compared to more complex alternatives (e.g., double PT, Rosen-Morse, de Sitter), thereby reducing computational cost without loss of accuracy.

2. Uncertainty Quantification via Approximation-Ensemble Averaging

• Improvement: Train an ensemble of surrogate models, each using a different analytical approximation for the same physical process (as done with PT, double PT, RM, dS).

• Capability: The AI can output a confidence interval for cosmological observables (e.g., primordial power spectrum) by measuring the spread across ensemble members, directly quantifying model-form uncertainty from the choice of approximation.

3. Automatic Detection of Vacuum-State Breakdown Regimes

• Improvement: Integrate a diagnostic layer that monitors the imaginary part of complex frequency functions (as in the Rosen-Morse case) to detect when a chosen vacuum state becomes ill-defined for certain wavenumbers.

• Capability: The AI can automatically restrict its predictive range to physically valid scales (e.g., k > √V1) or switch to an alternative vacuum criterion, preventing spurious results in suppressed regions.

4. Smoothness-Aware Post-Processing for Oscillation Removal

• Improvement: Implement an adaptive Bogoliubov transformation module that detects non-physical oscillations in outputs (e.g., power spectra) caused by non-smooth approximations, then replaces complex coefficients with their moduli only when the norm is stable.

• Capability: The AI can cleanly extract the underlying physical signal (e.g., power suppression) without manual intervention, improving the reliability of downstream parameter estimation.

5. Scale-Invariant Feature Extraction for Preinflationary Dynamics

• Improvement: Train a representation learner to identify the characteristic bounce scale (e.g., from the effective mass) and map it directly to the wavenumber where power suppression occurs.

• Capability: The AI can predict the suppression scale and tilt of the power spectrum for new inflaton potentials or perturbation types (scalar/tensor) without full numerical simulation, by leveraging the invariance shown in the paper.

6. Cross-Approach Generalization for LQC Frameworks

• Improvement: Develop a transfer-learning mechanism that takes conclusions from hybrid loop quantum cosmology (LQC) and applies them to the dressed-metric approach, using the paper’s proof of similar PT-like effective masses.

• Capability: The AI can generate reliable predictions for primordial spectra in alternative LQC formulations with minimal retraining, accelerating theoretical exploration.

7. Observational-Resolution-Aware Model Discriminator

• Improvement: Build a classifier that, given current or future observational error bars, determines whether two competing theoretical models (e.g., PT vs. RM) are distinguishable.

• Capability: The AI can prioritize which model refinements are worth pursuing (e.g., if spectra are indistinguishable, it recommends focusing on other observables), saving research effort.

8. Parameter-Free Error Bounding for Approximate Potentials

• Improvement: Use the paper’s relative-error metrics (e.g., <0.04 for double PT, <0.08 for RM) to train a neural network that predicts the maximum error of any new approximation based on its functional form and boundary conditions.

• Capability: The AI can propose new analytical approximations for other physical systems (e.g., black hole perturbations) with a guaranteed error bound, without exhaustive numerical testing.