Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Watch a technical presentation exploring the development of Equivariant Graph Exchange Correlation (EG-XC), a groundbreaking non-local exchange-correlation functional that leverages equivariant graph neural networks to enhance density functional theory calculations. Discover how this novel approach combines semi-local functionals with non-local feature density using an equivariant nuclei-centered point cloud representation to effectively capture long-range interactions. Learn about the impressive performance improvements achieved by EG-XC, including a 35-50% reduction in relative MAE on out-of-distribution conformations of 3BPA and superior data efficiency in molecular size extrapolation compared to traditional force fields. Understand how the functional is trained through a self-consistent field solver using only energy targets, making it a more practical solution for computational chemistry applications. The presentation includes detailed discussion of empirical evaluations demonstrating EG-XC's ability to accurately reconstruct CCSD(T) energies on MD17 and its exceptional performance on the QM9 dataset.
