From the chemistry of solvated electrons to classical fields at the mesoscale

Machine learning and neural network representations facilitate the construction of bottom-up multi-scale coarse-grained models consistent with the underlying ab-initio molecular dynamics. Here, I will consider two recent advances in this line of research: a machine learning model capable of simulating the chemistry of a solvated electron in a polar medium, and a scheme to extract the Landau free energy surface from deep potential molecular dynamics simulations of ferroelectric materials. The first advance uses a coarse-grained quantum mechanical description of the electronic polaron and will be illustrated with a molecular dynamics study of the recombination reaction between an excess electron and an excess proton in water. The second advance is a first step toward the construction of microscopically consistent Ginzburg-Landau models for microstructure evolution in ferroelectric materials and will be illustrated with the derivation of the Landau free energy surface depending on polarization and strain in lead titanate.