Grid Skyrme functional obtained using a committee of multilayer neural networks

We present two new mass models based on a Skyrme energy density functional. The parameters of these models were adjusted using a coordinate representation on a three-dimensional grid, allowing for both axial and triaxial deformation during the fitting process. In order to manage its computational complexity, we have employed a committee of multilayer neural networks to accelerate the adjustment process. The resulting models, GSk1 and GSk2, achieve an rms error on the 2408 known masses of 661 and 731 keV respectively, with GSk2 achieving a better description of the pairing properties of finite nuclei. The 884 measured nuclear charge radii are well reproduced with rms errors of 0.025 and 0.024 fm, respectively, as are the properties of nuclear matter, as predicted by ab-initio calculations. Both models predict significant numbers of nuclei to exhibit triaxial deformation in their ground states.