In Search of Lithium Intermetallics with Channel Structures: Prediction and Discovery of Ternary Silicides Li–RE–Si (RE = Pr, Nd, Tm, Lu)

Key findings from this study

• The study found that support vector classifier and SISSO models successfully predicted channel-structure adoption in ternary silicides with complementary strengths.
• The authors report that LiRESi and LiRESi2 phases with rare-earth elements Pr, Nd, Tm, and Lu adopt channel structures favorable for energy storage.
• The researchers demonstrate that machine learning combined with thermodynamic feasibility estimates accelerated discovery of new lithium intermetallics.

Overview

Researchers applied machine learning methods to predict structural properties of ternary lithium-metal-tetrelide intermetallics. The study aimed to identify candidates with channel structures suitable for energy storage applications. Two classification models—support vector classifier (SVC) and sure independence screening and sparsifying operator (SISSO)—systematized predictions of structural type.

Methods and approach

Machine learning classifiers trained on existing ternary tetrelide data differentiated channel from nonchannel structures. The authors combined SVC (conventional) and SISSO (interpretable) approaches to generate candidate structures. Formation energy estimates filtered predictions for synthetic feasibility.

Results

Four new ternary rare-earth silicides with channel structures were experimentally confirmed. The LiRESi series (RE = Pr, Nd, Tm, Lu) adopts a hexagonal ZrNiAl-type structure. The LiRESi2 series (RE = Pr, Nd) exhibits an orthorhombic LiCaSi2-type structure. Both series represent distinct structural families within lithium-containing rare-earth silicides.

Implications

Machine learning successfully identified experimentally viable intermetallics without exhaustive screening. The dual-model approach provided both predictive accuracy and interpretability for structure classification. Discovery of these channel-structured silicides expands the available materials library for ion-conduction applications.

Scope and limitations

This summary is based on the study abstract and available metadata. It does not include a full analysis of the complete paper, supplementary materials, or underlying datasets unless explicitly stated. Findings should be interpreted in the context of the original publication.