Speaker
Description
Warm inflation (WI) modifies the generation of primordial perturbations through dissipative dynamics and thermal fluctuations, making their accurate computation inherently challenging. Existing approaches typically rely on semi-analytical approximations in which the scalar power spectrum is expressed through a correction factor $G(Q)$ that depends only on the dissipation ratio $Q$. We present a C++ and Python based code, SWIM (Stochastic Warm Inflation Module), that is a numerical framework that computes the Warm Inflationary primordial power spectrum directly by solving the coupled stochastic perturbation equations, and yields the $G(Q)$ factor as well as the semi-analytical power spectrum in numerical form suitable for parameter inference. However, our analysis shows that the correction factor $G(Q)$ can depend non-trivially on additional model parameters, including the potential normalization and relativistic degrees of freedom of the radiation bath, even within the same class of WI models. This previously unaccounted parameter dependence leads to discrepancies between semi-analytical and numerical predictions in certain regimes, and introduces a source of systematic bias in parameter inference when semi-analytical approaches are used. These findings highlight a limitation of widely used approximation schemes in WI and motivate the use of full numerical power spectra in data-driven analyses for WI models. SWIM further interfaces with inference pipelines such as Cobaya and incorporates an on-the-fly surrogate emulator based on Random Forest Regression, trained dynamically in high-likelihood regions to enable efficient parameter inference while simultaneously constructing a surrogate model of the numerical solver. This trained model can subsequently be used for more detailed parameter inference and analysis. Thus, SWIM offers a complete numerical framework for analysis of any Warm Inflation model. While developed in the context of WI, this strategy is applicable more broadly to cosmological problems involving expensive numerical computations.
| Other topic / keywords: | Warm Inflation, Primordial Power Spectrum, Numerical Methods, Parameter Inference, Machine Learning |
|---|