Machine-learning force fields
Many-body gradient-domain machine learning#
Atomistic insight is fundamental for computational predictive studies of chemical and physical processes. Machine learning force fields provide a route to high-level ab initio calculations at a fraction of the cost. Gradient-domain machine learning (GDML), a kernel-based method, directly learns the relationship between atomic coordinates and interatomic forces. However, training in the gradient domain sacrifices generalized transferability to other species or number of atoms.
Many-body GDML (mbGDML), is a technique for GDML transferability to n-sized systems by using many-body machine learning models. Every aspect of the process from preparing quantum chemistry calculations, data set creation, training, and use of mbGDML force fields is taken care of in this user-friendly Python package.