Machine Learning-Guided Discovery of Ternary Compounds Containing La, P, and Group 14 Elements.

Area of Science:

• Computational Materials Science
• Machine Learning in Chemistry
• Solid State Physics

Background:

• Discovering novel ternary compounds with favorable energetic properties is crucial for materials innovation.
• Traditional methods for exploring ternary phase spaces are computationally intensive and time-consuming.
• Machine learning offers a promising avenue to accelerate computational materials discovery.

Purpose of the Study:

• To develop and demonstrate an efficient machine learning (ML)-guided workflow for identifying energetically favorable ternary compounds.
• To accelerate the discovery of new materials by integrating deep machine learning with first-principles calculations.
• To explore the La-Si-P system as a prototype for this accelerated materials discovery approach.

Main Methods:

• Integration of a deep machine learning (ML) model with first-principles calculations.
• Utilizing ML to efficiently search and predict energetically favorable ternary crystal structures.
• Employing first-principles calculations to verify the stability and formation energies of predicted compounds.

Main Results:

• Successfully identified several new La-Si-P ternary compounds with formation energies within 30 meV/atom of the convex hull.
• Discovered two dynamically stable La-Si-P phases, La5SiP3 and La2SiP, with formation energies of 2 and 10 meV/atom above the convex hull, respectively.
• Predicted numerous low-energy La-X-P phases (X = Ge, Sn, Pb) by substituting Si with heavier Group 14 elements in promising La-Si-P structures.

Conclusions:

• The ML-guided first-principles approach significantly accelerates the exploration of crystal structures and energetic stabilities.
• This methodology enables efficient discovery of novel, stable ternary compounds.
• The predicted La-X-P phases (X = Ge, Sn, Pb) represent promising new material candidates for further investigation.