Pressure-Induced Phase Transition in the Ti3O5 Heat-Storage System: Insights from Machine-Learned Potential Driven Metadynamics Simulations

Heat-storage materials enable the conservation of thermal energy from intermittent energy generation, industrial processes, or in building applications. Trititanium pentoxide, Ti3O5, has emerged as a promising material being able to store low temperature thermal energy for prolonged periods of time and release the stored energy on demand by the application of an unusually small pressure. Our investigation of the pressure effect in Ti3O5 focuses on the (001) surface termination, where we discover a surface reconstruction during simulated annealing simulations. These simulations employ a moment tensor potential, which is tailored towards describing the whole phase space likely to be encountered during the phase transition. The validation of this interatomic potential shows excellent agreement with the underlying density functional method, while enabling molecular dynamics simulations of several hundreds of nanoseconds within a short timeframe. A layer-by-layer mechanism of the phase transition in direction normal to the reconstructed (001) surface is obtained from nudged elastic band simulations, corroborating recent theoretical predictions as well as experimental results. We employ repulsive potentials to simulate pressure during metadynamics simulations to study the pressure effect on the heat-storage material. Biasing the phase transition in the surface layer during molecular dynamics simulations reveals a lowering of the activation free energy for the phase transition, as well as a stabilization in terms of free energy of the transformed structure relative to the reconstructed surface. Translating the repulsive potential to a pressure value unveils very close agreement between predicted phase transition pressure and the unusually low experimentally observed pressure threshold. Our approach illustrates how to model the pressure effect in materials in contrast to the conventional approaches, which are unable to explain these experimental observations.