Bayesian Optimization and Gaussian-Process Residual Correction for Neutron Skinthickness in Relativistic Mean Field Model Alpi Mahisha Nugraha (a,b), Rizal Kurniadi (c), and Anto Sulaksono (b)
(a) Teknik Informatika, FTIK, Universitas Indraprasta PGRI Jakarta, Jl Nangka, Jakarta, 12530, Indonesia
(b) Departemen Fisika, FMIPA Universitas Indonesia, Kampus UI, Depok, 16424, Indonesia
(c) Nuclear Physics and Biophysics Research Group, Institut Teknologi Bandung, Jl Ganesha 10, Bandung, 40116, Indonesia
Abstract
Neutron skinthickness is an important observable in nuclear physics because it is closely related to the isospin dependence of nuclear interactions and the symmetry energy of nuclear matter. Although relativistic mean field (RMF) models such as IUFSU* and NLVT1 successfully reproduce many bulk nuclear properties, their neutron skin predictions often show systematic deviations along isotopic chains. In this work, neutron skin thickness predictions are improved through a two-stage refinement approach. First, the RMF parameters are optimized using Bayesian optimization to achieve better global agreement with nuclear observables. The optimized RMF model is then combined with a Gaussian-process (GP) residual correction, where the GP model learns the remaining differences between calculated and experimental neutron skin thickness values as smooth functions of nuclear properties. The results show that Bayesian optimization improves the overall consistency of RMF calculations, while the additional GP correction significantly reduces systematic residual errors across medium and heavy nuclei. Compared with the original RMF parameter sets, the proposed hybrid approach provides a more accurate description of isotopic trends while preserving the physical foundation of the RMF framework.
Keywords: Neutron skinthickness- RMF- Bayesian Optimization- Gaussian process residual correction
