“Improving Machine Learning Models of Kinetic Energy Functionals Using Smoothed Density-Dependent Variables and the Fourth Order Gradient Expansion: A Study on Aluminum, Magnesium, and Silicon”

https://pubmed.ncbi.nlm.nih.gov/38112506

This study demonstrates that machine learning models of kinetic energy functionals can be made more accurate and efficient by using smoothed density-dependent variables and the fourth order gradient expansion, as shown by obtaining accurate kinetic energy models for aluminum, magnesium, and silicon using as few as 2000 samples.