Phase diagram of CO2-I/III from molecular dynamics simulation using a PBE0-accuracy machine learning potential

Area of Science:

• Computational materials science
• Chemical physics
• Solid-state chemistry

Background:

• Understanding the phase diagram of carbon dioxide (CO2) is crucial for various scientific and industrial applications.
• Previous studies have reported different CO2 crystal structures and transition pressures, leading to ambiguities in phase identification.
• Accurate theoretical modeling of CO2 phase transitions requires high-fidelity potentials and efficient simulation methods.

Purpose of the Study:

• To construct a highly accurate machine learning potential (MLP) for carbon dioxide (CO2) using the PBE0-D3(BJ)/aug-cc-pVDZ level of theory.
• To investigate the CO2 molecular crystal phase diagram across a wide range of temperatures and pressures.
• To elucidate the transition mechanisms and propose a definitive structure for the CO2-III phase.

Main Methods:

• Employed the PBC-GEBF-AL workflow, integrating active learning (AL) with the generalized energy-based fragmentation approach under periodic boundary conditions (PBC-GEBF).
• Utilized multiTPU-OPES enhanced sampling for molecular dynamics simulations, achieving up to 170 ns trajectory lengths.
• Generated a machine learning potential (MLP) with PBE0-D3(BJ)/aug-cc-pVDZ accuracy.

Main Results:

• Obtained the CO2 phase diagram from 250-700 K and 10.5-14.0 GPa, identifying a CO2-I/III coexistence line peaking at ~12.3 GPa and 525 K.
• Revealed a concerted CO2-I/III transition mechanism involving molecular rotation, lattice deformation, and non-monotonic volume changes.
• Proposed a new Cmca space group structure for CO2-III, characterized by tilted molecules and a > b lattice parameters, closely resembling CO2-VII.

Conclusions:

• The proposed tilted Cmca structure for CO2-III suggests it may be the same phase as CO2-VII, with experimental discrepancies attributed to thermal effects.
• The study provides deeper insights into CO2 phase transitions and the coexistence of different crystal structures.
• Established a generalizable strategy for developing high-precision MLPs for complex rare event systems.