Publication details
Competition of volumetric and surface-related energy contributions in phase transformations in Sn: an ab-initio and machine-learned-potential study
| Authors | |
|---|---|
| Year of publication | 2023 |
| Type | Conference abstract |
| MU Faculty or unit | |
| Citation | |
| Description | The allotropic transformation from the higher-temperature tetragonal-body-centered beta-Sn to the lower-temperature diamond-lattice alpha-Sn upon cooling under the temperature of 13.2 °C is one of the most famous structural transformations known to our civilization. Interestingly, actual atomic-scale mechanisms of the transformation are much less studied and understood partly due to the 26% volumetric change accompanying this transformation and hindering a detailed examination by many experimental as well as theoretical methods. In order to shed new light on this centuries-long mystery, we have employed a combination of quantum-mechanical calculations and both machine-learned and classical atomistic potentials. In particular, a nanoparticle of undercooled beta-Sn was put into contact with a nanoparticle of the alpha-Sn surrounded by vacuum within quite a large computational cell. Our calculations were aimed at analyzing a competition of (i) volumetric energy contributions related to the thermodynamic energy difference between the phases, see also our recent paper [1], and (ii) surface-related and interface-related contributions to the free energies of both nanoparticles which are associated with their nano-scale size. Or preliminary results obtained for a simulation box containing a few hundreds of Sn atoms indicate that the surface-related contributions dominate for the studied nanoparticle sizes. In particular, both nanoparticles minimize their surface energies by re-shaping into close-to-spherical nanoparticles within a process accompanied by a partial loss of crystallinity. Qualitatively the same results were obtained by two approaches which we used. The first one was based on the machine-learned force fields obtained from the on-the-fly learning procedure during ab-initio molecular dynamics (MD) and the second one is characterized by the use of a classical MD potential. The VASP software [2,3] was employed for the former MD, while the LAMMPS package [4] for the latter. References [1] M. Friák et al., Computational Materials Science, 2022, 215, 111780. [2] G. Kresse, J. Hafner, Physical Review B, 1993, 47, 558. [3] G. Kresse, J. Furthmüller Physical Review B, 1996, 54, 11169. [4] A.P. Thompson et al. Comp. Phys. Comm., 2022, 271, 10817. |