<ul><li>Density Functional Theory (DFT) enables predicting chemical and physical properties of molecular systems by approximating the many-body Schrödinger equation.</li><li>Deep Learning models have shown promise in predicting DFT outputs for large datasets of molecular conformations, reducing computational costs.</li><li>A self-refining training method is proposed to decrease reliance on pre-collected datasets, allowing simultaneous deep-learning model training and molecular conformation sampling.</li><li>The method minimizes the discrepancy between generated samples and the target energy distribution, demonstrating its effectiveness in an empirical study and providing open-source implementation.</li></ul>

Self-Refining Training for Amortized Density Functional Theory

Discover more