Machine Learning Meets Quantum Mechanics for Materials

This talk will present our research on merging machine learning with quantum mechanics. Electron-nuclear interactions determine all materials' properties, and accurate simulations of electronic structure are essential to address critical scientific questions related to renewable energy, sustainable materials, and semiconductor devices. However, electronic structure simulations face an accuracy-size tradeoff.
I will first present our ongoing efforts on developing the Materials Learning Algorithms (MALA) package to solve the electronic structure problem faster. MALA leverages a combination of neural networks, physically constrained optimization algorithms, and efficient post-processing routines. Next, I will present our work on using physics-informed neural networks to solve the time-dependent Kohn-Sham equations, which describe electron dynamics in response to incident electromagnetic waves.