Fast Electron Density Estimation of Molecules, Liquids, and Solids Using Neural Networks

Watch a scientific seminar where Peter Bjørn Jørgensen from DTU Energy presents DeepDFT, a machine learning framework for predicting electron density in molecules, liquids, and solids. Learn about deep learning applications in physics and the use of graph neural networks as the model's foundation. Explore the model's performance across multiple datasets including QM9 molecules, liquid ethylene carbonate electrolyte, and lithium ion battery cathode material (NMC), demonstrating superior accuracy compared to existing methods while achieving significantly faster computation times than traditional DFT. Understand key concepts including different molecular representations, graph networks, edge states, directionality, and evaluation metrics through detailed explanations of the screening funnel, error analysis, and runtime comparisons. Hosted by Prof. Pablo Ordejón of ICN2's Theory and Simulation Group, this 46-minute presentation provides comprehensive insights into cutting-edge developments in computational chemistry and materials science.