Layer-by-layer phase transformation in Ti₃O₅ revealed by machine-learning molecular dynamics simulations

Area of Science:

• Materials Science
• Solid-State Physics
• Chemical Engineering

Background:

• Reconstructive phase transitions involve breaking and reforming chemical bonds, typically slow due to high energy barriers.
• The titanium oxide (Ti3O5) β- to λ- phase transformation is unusually fast and reversible, with its mechanism poorly understood.

Purpose of the Study:

• To elucidate the microscopic mechanism behind the ultrafast and reversible β- to λ-Ti3O5 phase transformation.
• To develop accurate computational methods for studying complex phase transitions.

Main Methods:

• Developed an efficient machine learning potential using on-the-fly active learning.
• Employed metadynamics and large-scale molecular dynamics simulations.
• Utilized advanced sampling techniques for accurate simulations.

Main Results:

• Discovered a kinetically favorable, in-plane nucleated, layer-by-layer transformation mechanism.
• Identified intermediate metastable phases facilitating the multistep barrier-lowering process.
• Confirmed the initiation of transformation via 2D nuclei formation in the ab-plane.

Conclusions:

• The β- to λ-Ti3O5 transition proceeds through a novel layer-by-layer mechanism, explaining its ultrafast and reversible nature.
• The developed computational strategies are applicable to studying other complex structural phase transitions.
• Provides fundamental insights into reconstructive phase transition dynamics.