General-purpose machine-learned potential for 16 elemental metals and their alloys

Area of Science:

• Materials Science
• Computational Chemistry
• Condensed Matter Physics

Background:

• Machine-learned potentials (MLPs) offer high accuracy but lack general applicability due to limited scope.
• Developing universal MLPs for diverse elements and alloys is crucial for advancing materials simulations.

Purpose of the Study:

• To develop a unified, general-purpose machine-learned potential (MLP) for a wide range of elemental metals and their alloys.
• To demonstrate the efficacy of a novel approach for constructing comprehensive chemical space representation using minimal system types.

Main Methods:

• Utilized principal component analysis (PCA) to determine the optimal representation of chemical space.
• Developed the Unified Elemental Potential version 1 (UNEP-v1) model for 16 elemental metals and their alloys.
• Validated the model's performance against diverse physical properties and experimental observations.

Main Results:

• The UNEP-v1 model successfully represents the chemical space using one- and two-component systems.
• UNEP-v1 demonstrated superior accuracy and efficiency compared to the embedded-atom method (EAM) potential.
• The model accurately reproduced experimental data on chemical order, stable phases, plasticity, and radiation damage.

Conclusions:

• A unified general-purpose MLP approach is feasible and effective for diverse metallic systems.
• UNEP-v1 provides a significant advancement for large-scale materials simulations, enabling accurate prediction of properties and behaviors.
• This work paves the way for broader applications of MLPs in materials discovery and design.