Training-free hyperparameter optimization of neural networks for electronic structures in matter

A myriad of phenomena in materials science and chemistry rely on quantum-level simulations the electronic structure matter. While moving to larger length time scales has been a pressing issue for decades, such large-scale calculations are still challenging despite modern software approaches advances high-performance computing. The silver lining this regard is use machine learning accelerate -- line research recently gained growing attention. grand challenge therein finding suitable machine-learning model during process called hyperparameter optimization. This, however, causes massive computational overhead addition that data generation. We construction neural network models by roughly two orders magnitude circumventing excessive training optimization phase. demonstrate our workflow Kohn-Sham density functional theory, most popular method chemistry.