Machine Learning applied to Reactivity (MaLeR)

EU flag

Dr. Matti Hellström will develop automated, high-accuracy reparameterization of ReaxFF potentials and DFTB parameters with machine learning techniques. It is an EU-funded Marie Skłodowska-Curie Individual Fellowship (H2020 Grant Agreement No 798129).

While ReaxFF potentials are fast, transferable and reasonably accurate, high-dimensional neural networks (HDNNs) are typically much more accurate at the cost of more training data and less generalizability.

In MaLeR Matti will combine the best of both worlds by enhancing ReaxFF with machine learning. The machine-learned potentials will be applied as a correction function to ReaxFF potentials, which reduces required training set sizes and increases transferability. As such an all-purpose computationally efficient method for large, reactive system sizes is within reach, with many promising applications in chemistry and materials science.

Machine Learning ReaxFF

Subscribe to Newsletters