Abstract
New research on Machine Learning is the subject of a report. According to news reporting out of Davis, California, by NewsRx editors, research stated, “Machine learning potentials enable molecular dynamics simulations to exceed the size and time scales that can be accessed by first-principles methods like density functional theory, while still maintaining the accuracy of the underlying training data set. However, accurate machine learning potentials come with relatively high computational costs that limit their ability to predict properties requiring extensive sampling, large simulation cells, or long runs to converge.” Funders for this research include National Science Foundation (NSF), National Science Foundation (NSF). Our news journalists obtained a quote from the research from the University of California Davis, “Here, we have developed and tested a neuroevolution-potential model for water trained to hybrid-dispersioncorrected density functional calculations. This model exhibits accuracy and transferability comparable to state-of-the-art machine learning potentials but at a much lower computational cost. As a result, it enabled us to compute well-converged thermodynamic averages and fluctuations. This allowed us to assess the ability of our model to reproduce several thermodynamic properties of water and ice as well as the anomalous behavior of water density, heat capacity, and compressibility.”
NETL
NSTL
Amer Chemical Soc