4 December 2023
High Energy Physics - Phenomenology
José Luís Rodríguez-Sánchez,
Method optimizes model parameters to fit data, and quantifies parameter uncertainties
Uses Bayesian statistics - iterative expectation maximization and Gibbs sampling
Applied to nuclear reaction Monte Carlo model INCL/ABLA
Improved model agreement with data by factors of 100 for some reactions
Computationally intensive method requiring careful data selection
Optimizing Monte Carlo model parameters with Bayesian statistics
This paper presents a method to optimize the free parameters of models to better fit experimental data, and to estimate the uncertainties on those optimized parameters. The method is based on Bayesian statistics, using an iterative algorithm with expectation maximization and Gibbs sampling phases. It alternates between improving the parameter values and estimating the posterior distribution. The authors demonstrate its application to optimize four parameters of the INCL/ABLA Monte Carlo model for simulating nuclear reactions, significantly improving agreement with measured observables. They discuss requirements, optimizations, and limits when applying this general approach.
No comments yet, be the first to start the conversation...
Sign up to comment on this paper