CASUS Institute Seminar, Lenz Fiedler, CASUS
Location: CASUS Lecture Room
Density Functional Theory (DFT) is one of the most popular quantum mechanical simulation methods, since it balances sufficient accuracy with reasonable computational cost. It is often used in material science applications at ambient and extreme conditions. Nonetheless, DFT approaches its limits in terms of computational feasbility when faced with simulation problems at larger time and length scales, especially at temperatures >> 0K. Surrogate models based on neural networks can circumvent these limitations. By training a neural network to predict properties of interest (total energy, atomic forces) based on atomic configurations, predictions with DFT-like accuracy can be done at a fraction of the computational cost. At CASUS, the Matter under Extreme Conditons department is currently developing the Framework for Electronic Structure Learning (FESL), a modular open-source python package intended to serve as a toolbox for efficiently building these surrogate models. FESL enables users to preprocess DFT data, train networks and postprocess results using only a few lines of code.