Poster and Oral Presentation at Swedish e-Science Academy 2024

The Swedish e-Science Academy is organised by the Swedish national strategic programme in e-science titled eSSENCE and is open to anyone interested in e-Science, model and tool development as well as applications in a broad range of simulation and AI areas. We presented a poster as well as a short oral flash presentation about our work in the Phoenix-OoC project. For large cellulose systems, DFT becomes too computationally demanding. Additionally, the weak non-covalent interactions result in multiple meta-stable crystal polymorphs and hydrogen bonding patterns. Given the abundance of local minima, modelling requires consideration of dynamic behaviour and temperature effects through molecular dynamics (MD) simulations. In the poster we discussed two solutions to addressing these challenges: 1) DFTB models: We have employed the approximate semiempirical extended tight binding (xTB) approach, significantly accelerating calculations while preserving the electrons in the model, i.e. the ability to describe molecular polarisation and charge transfer. 2) ML (machine-learned) potential: To address the limitations of xTB at longer time scales, we train ML potentials. These potentials are trained to retain the accuracy of xTB while enabling more efficient exploration of dynamics over time and at different temperatures.