Computational Mechanochemistry
Panosetti Group
Welcome to the webpage of the Computational Mechanochemistry group!
Mechanochemistry is the chemistry initiated by mechanical forces. Our interest in mechanochemistry is twofold.
On the one hand, mechanochemical synthesis is a popular route for the preparation of battery materials and catalysts, one prominent example being ball milling. The relationship between functional properties and the microstructure resulting from mechanochemical preparation can only be fully understood by means of atomistic modelling, in combination with experimental characterization.
On the other hand, approaches exist in catalysis where the role of the catalyst is played directly by, or enhanced by, some mechanical force. It has been shown, e.g., that significant increase in catalytic activity can be achieved upon ball milling at moderate temperature. The processes occurring inside the mechanical vessel are extremely difficult to follow, and only very recent experimental developments allow to monitor mechanochemistry in operando.
In both contexts, the determination of which active microstructures form, transient or not, and which products, and how, is mostly unexplored territory. My subgroup aims at developing computational strategies for direct simulations of mechanochemical phenomena, based on fast but accurate atomistic models such as ultrafast potentials, machine learning interatomic potentials, and semiempirical electronic structure methods.
Among the latter, Density-Functional Tight Binding (DFTB) offers a great compromise between speed and DFT-grade accuracy, while retaining access to electronic properties and explicit electrostatics – both of crucial importance in the modelling of active materials. However, this advantage comes at the cost of pairwise parametization: an N2 effort across the periodic table, limiting the availability of off-the-shelf parameters.
Over the years, we developed a robust divide-and-conquer parametrization workflow: while the electronic part of the DFTB interaction (largely reliant on explicit physics) is parametrized following a property-based approach via standard optimization ("white-box"), the so-called repulsive contribution (a complicated potential governing energies and forces) is parametrized with a machine-learning, data-based approach ("black-box"). The resulting "gray-box" model can be fine tuned by generating appropriate training sets while maintaining transferability. In parallel, we cooperate with software engineers to improve the computational efficiency of DFTB software, striving to painlessly access dynamics for simulation cells containing several thousands of atoms.
Selected Publications
Selected recent publications are listed below (a full list can be found under Publications):
Staged Training of Machine-Learning Potentials from Small to Large Surface Unit Cells: Efficient Global Structure Determination of the RuO2(100)-c(2 × 2) Reconstruction and (410) Vicinal, Y. Lee, J. Timmermann, C. Panosetti, C. Scheurer, and K. Reuter: J. Phys. Chem. C (2023)
Black box vs gray box: Comparing GAP and GPrep-DFTB for ruthenium and ruthenium oxide, C. Panosetti, Y. Lee, A. Samtsevych, and C. Scheurer, J. Phys. Chem. C (2023)
The intrinsic electrostatic dielectric behaviour of graphite anodes in Li-ion batteries-Across the entire functional range of charge, S. Anníes, C. Scheurer, and C. Panosetti, Electrochimica Acta (2023)
Systematic Comparison of Genetic Algorithm and Basin Hopping Approaches to the Global Optimization of Si(111) Surface Reconstructions, M.N. Bauer, M.I.J. Probert, and C. Panosetti, J. Phys. Chem. A (2022)
Accessing Structural, Electronic, Transport and Mesoscale Properties of Li-GICs via a Complete DFTB Model with Machine-Learned Repulsion Potential, S. Anníes, et al., Materials (2021)