An atomistic fingerprint algorithm for learning ab initio molecular force fields
Yu-Hang Tang, Dongkun Zhang, and George Em Karniadakis have published an article that has been featured on the front page of the Journal of Chemical Physics. The article is entitled, "An atomistic fingerprint algorithm for learning ab initio molecular force fields." This work turns out to be the Ph.D. thesis of Yuhang Tang, one of the graduate students in the Division of Applied Mathematics at Brown University. The article addresses that the DECAF molecular fingerprint allows machines to learn and predict interatomic forces and beyond. A molecular fingerprint is a feature vector that describes an atomistic neighborhood configuration. It is an important abstraction and a key ingredient for data-driven modeling of potential energy surface and interatomic force. The Density-Encoded Canonically Aligned Fingerprint (DECAF) fingerprint algorithm is robust and efficient for fitting per-atom scalar and vector quantities. We have demonstrated that the fingerprint algorithm can be used to implement highly accurate regressions of both scalar and vector properties of atomistic systems including energy, force and dipole moment, and could be a useful building block for constructing data-driven next generation force fields to accelerate molecular mechanics calculations with an accuracy comparable to those driven by quantum mechanical theories and calculators.