This paper presents a forward-modeling framework for cosmic ray propagation, incorporating neutron-mediated escape mechanisms in magnetized astrophysical environments. It addresses limitations in traditional diffusion models by accounting for the stochastic conversion of CR protons to neutrons and their subsequent ballistic propagation. The study implements sample-efficient Bayesian inference methods to recover physical parameters from synthetic observations, demonstrating that active learning with Gaussian process surrogates achieves comparable accuracy to standard MCMC with significantly fewer model evaluations.
Key findings
Neutron-mediated escape can significantly alter cosmic ray transport, with escape efficiencies varying by orders of magnitude across different astrophysical environments.
Energy-dependent time delays of 10^4-10^5 seconds are predicted and should be observable with current and upcoming facilities.
Sample-efficient Bayesian inference using active learning with Gaussian process surrogates reduces computational cost by over two orders of magnitude while maintaining comparable accuracy to standard MCMC methods.
Cross-correlation analysis and damped random walk models provide complementary information about characteristic timescales but cannot fully constrain the underlying physical parameters without forward-modeling.
Limitations & open questions
The study focuses on four distinct environments, which may not cover all possible astrophysical scenarios.
The accuracy of the Bayesian inference methods is contingent upon the quality and representativeness of the synthetic data used for training.